09-09-13 - Chapter 1 Models

1
Global Climate Models and Their Limitations

Anthony Lupo (USA)
William Kininmonth (Australia)
Contributing: J. Scott Armstrong (USA), Kesten Green (Australia)
1. Global Climate Models and Their Limitations

Key Findings
Introduction
1.1 Model Simulation and Forecasting
1.2 Modeling Techniques
1.3 Elements of Climate
1.4 Large Scale Phenomena and Teleconnections
Key Findings Confidence in a model is further based on the

The IPCC places great confidence in the ability of careful evaluation of its performance, in which model
general circulation models (GCMs) to simulate future output is compared against actual observations. A
climate and attribute observed climate change to large portion of this chapter, therefore, is devoted to
anthropogenic emissions of greenhouse gases. They the evaluation of climate models against real-world
claim the “development of climate models has climate and other biospheric data. That evaluation,
resulted in more realism in the representation of many summarized in the findings of numerous peer-
quantities and aspects of the climate system,” adding, reviewed scientific papers described in the different
“it is extremely likely that human activities have subsections of this chapter, reveals the IPCC is
caused more than half of the observed increase in overestimating the ability of current state-of-the-art
global average surface temperature since the 1950s” GCMs to accurately simulate both past and future
(p. 9 and 10 of the Summary for Policy Makers, climate. The IPCC’s stated confidence in the models,
Second Order Draft of AR5, dated October 5, 2012). as presented at the beginning of this chapter, is likely
This chapter begins with a brief review of the exaggerated. The many and varied model deficiencies
inner workings and limitations of climate models. discussed in this chapter indicate much work remains
Climate models are important tools utilized to to be done before model simulations can be treated
advance our understanding of current and past with the level of confidence ascribed to them by the
climate. They also provide qualitative and IPCC.
quantitative information about potential future The following points summarize the main
climate. But in spite of all their sophistication, they findings of this chapter:
remain merely models. They represent simulations of
the real world, constrained by their ability to correctly • Properties inherent in models make dynamic
capture and portray each of the important processes predictability impossible. Without dynamic
that operate to affect climate. Notwithstanding their predictability, other techniques must be used to
complexities, the models remain deficient in many simulate climate. Such techniques introduce
aspects of their portrayal of the climate, which biases of varying magnitude into model
reduces their ability to provide reliable simulations of projections.
future climate.
9
Climate Change Reconsidered II
• To have any validity in terms of future Introduction

projections, GCMs must incorporate not only the Global Climate Models (GCMs) have evolved from
many physical processes involved in determining the Atmospheric General Circulation Models
climate, but also all important chemical and (AGCMs) widely used for daily weather prediction.
biological processes that influence climate over GCMs have been used for a range of applications,
long time periods. Several of these important including investigating interactions between
processes are either missing or inadequately processes of the climate system, simulating evolution
represented in today’s state-of-the-art climate of the climate system, and providing projections of
models. future climate states under scenarios that might alter
the evolution of the climate system. The most widely
• Limitations in computing power frequently result recognized application is the projection of future
in the inability of models to resolve important climate states under various scenarios of increasing
climate processes. Low-resolution models fail to atmospheric carbon dioxide (CO2).
capture many important phenomena of regional At the core of a GCM is an AGCM that
and lesser scales, such as clouds; downscaling to dynamically simulates the circulation of the
higher-resolution models introduces boundary atmosphere, including the many processes that
interactions that can contaminate the modelling regulate energy transport and exchange by and within
area and propagate error. the atmospheric flow. The basic atmospheric flow is
represented by fundamental equations that link the
• The magnitude of the range of projected mass distribution and the wind field. These equations
responses to a doubling of atmospheric CO2 by are represented on a spherically spatial grid field that
itself establishes that large errors and limitations has many levels representing the depth of the
in the models remain to be corrected. atmosphere. The flow equations are modified by the
representation of processes that occur on a scale
• Many GCMs fail to account properly for certain below that of the grid—including such processes as
“multiplier effects” that may significantly amplify turbulence, latent heat of condensation in cloud
the initial impacts of various biospheric formation, and dynamic heating as solar and infrared
processes. For example, although the absolute radiation interact with atmospheric gases, aerosols,
variations associated with some solar-related and clouds.
phenomena are rather small, Several multiplier The oceans are at least as important as the
effects may significantly amplify the initial atmosphere for the transport of energy. For that
perturbation. reason, the GCM also includes an Ocean General
Circulation Model (OGCM) that simulates the
• Major imperfections in the models prevent proper circulation of the oceans. The OGCM is vital for
simulation of important elements of the climate climate simulations because the oceans represent a
system, including pressure, wind, clouds, dynamic thermal reservoir that, through energy
temperature, precipitation, ocean currents, sea ice, exchange with the atmosphere, dominates the
permafrost, etc. Large differences between model evolution of the climate system. The specification of
predictions and observations frequently exist the processes that regulate heat, moisture, and
when comparing these elements or features. In momentum exchanges between the ocean and
some cases computer models fail to simulate even atmosphere is crucial to the integrity of a GCM.
the correct sign of the observed parameters. Land surface, and how soil moisture and
vegetation type regulate heat, moisture, and
• Although some improvements have been noted in momentum with the atmosphere, plays a lesser but
performance between the CMIP3 set of models nevertheless important role in the simulation of
used in AR4 and the newer CMIP5 models climate. Soil moisture and vegetation respond to local
utilized in AR5, many researchers report finding precipitation and affect the exchange of heat,
little or no improvement in the CMIP5 model moisture, and momentum with the atmosphere over
output for several important parameters and time. The soil moisture and vegetation (and their
features of Earth’s climate. regulation of land-atmosphere exchange processes)
respond to the climate on the shorter time-scale of
weather systems but, due to the varying accumulation
10
of soil moisture, the influence of land surface on on a limited wavelength band, will affect the local
climate is on seasonal and interannual time-scales. magnitude of infrared radiation loss to space. Apart
Surface ice sheets also have an important role in from water vapor concentration these variations are
the evolution of the climate system. Their formation not necessarily temperature-dependent. Thus a change
and expansion represent a lowering of the total energy to the internal structure of the climate system for
of the climate system as a whole because latent heat is whatever reason will—all else being equal—lead to
lost as water changes from the liquid to solid phase. change in Earth’s equilibrium temperature.
Likewise, contraction of surface ice sheets represents Within the AGCM there are many important
an increase in the total energy of the climate system. processes that operate on scales below the resolution
The representation of heat, moisture, and momentum of the computational grid (sub-grid scale processes)
exchanges between ice surfaces and the atmosphere and regulate local temperature, moisture, and
differs from that of land surfaces or ocean surfaces. momentum. Perhaps the most important of these is
Dominating the climate system and its evolution convection.
are the radiation processes that regulate the input and As described by Riehl and Malkus (1958), it is
output of energy. The shape of the rotating Earth, its the “hot towers” of deep tropical convection that
distance from the Sun, and the characteristics of its distribute the accumulating heat and latent energy of
orbit determine the nature of diurnal (daytime) and the tropical boundary layer through the troposphere.
seasonal solar heating, including its maximum over Correct specification of the mass flows within the
the tropics. The shedding of energy by infrared cloud mass and its surroundings, including the
radiation originates from the surface, from aerosols, updrafts and downdrafts, is essential for regulating
from clouds, and from greenhouse gases of the the Hadley Cell circulation and the availability of
atmosphere (CO2, H2O, O3, etc.). The latitudinal tropical energy for transport to higher latitudes.
spread of infrared loss radiation is less than for solar Correct specification of the mass flows is also
radiation and results in excess solar radiation being important if the local impact on temperature, water
absorbed over the tropics but excess radiation vapor, and momentum are to be quantified. Correctly
shedding over higher latitudes. specifying the mass flows remains a challenge to
A primary function of the climate system is to modelers.
transport energy from the tropics to higher latitudes; In general, clouds and their interaction with the
globally, there is an equilibrium between solar climate system are difficult to model. Clouds are an
radiation absorption and infrared radiation loss to outcome of vertical motion and saturation, but the
space. Of course, with such a complex system there is feedback to the circulation through radiation
rarely perfect balance. At times, especially during the processes is sensitive. Although cloud fields tend to
cycle of seasons, Earth is accumulating radiation be regulated by the larger scale circulation, the
energy and warming, whereas at other times it is processes leading to cloud formation and dissipation
losing energy and cooling. But the rate of radiation are operating on scales very much smaller than that of
loss varies with temperature and acts as a natural the computation grid, with individual clouds often
thermostat: when Earth warms, the infrared radiation occupying only a small part of a grid. Thus it is
loss to space increases such that it exceeds the solar necessary for models to specify the climate
input and warming ceases; when Earth cools, the interaction of a multitude of differing clouds across a
infrared radiation loss to space decreases such that grid space by a single process.
solar radiation exceeds the infrared radiation loss and AGCMs are very complex and their output should
cooling ceases. be examined carefully and cautiously. In the physical
The natural thermostat is more complex than this sciences, mathematical models are often used to
simple portrayal because different components of the formalize a theory or hypothesis. For example,
climate system interact with limited bands of the Newton’s famous law of gravity formalized a
infrared radiation spectrum. In particular, variation in statement of how objects fall or attract each other
surface characteristics, boundary layer aerosol under ideal conditions (without wind or friction, for
characteristics, cloud height and distribution, and example). Note that in Newton’s law “gravity” is
concentration of individual greenhouse gases can all undefined and remains undefined. Also, in this theory
affect the local infrared radiation loss to space across Newton was able to treat objects such as planets as
characteristic wavelengths with none affecting the full point masses, a successful but auxiliary assumption.
spectrum. Variations in each component, while acting Textbook physics is largely made up of laws based on
11
such idealized situations (point masses, frictionless processes. The equations represent four basic physical
planes, ideal geometric bodies), and in approx- principles; correct theories or models representing
imations to ideal situations the models of physics atmospheric motions will not violate these basic
work extremely well. principles: (1) conservation of mass (dry mass and
In the real world, however, inhomogeneity and water mass), (2) conservation of energy, (3)
complexity make the basic laws of physics less conservation of momentum, and (4) elemental kinetic
reliable. Whereas the breakage of simple rods under theory of gases. The equations are sequentially solved
strain is easy to model and predict, earthquakes, across the grid space for each time step such that the
which are also a breakage problem but occur in a time directions of quantities at each grid point are
complex setting of heterogeneous rock, are not affected by their neighbors according to the governing
predictable. Just because laws of physics are used equations.
does not mean a process is predictable; the climate Physical processes for which there is no precise
prediction problem is not “just physics” as some formulation, or where the formulation is on a scale
scientists like to claim. It is also helpful to remember below that able to be resolved in the model, are
the laws of physics were developed by many rounds represented within these equations by
of experimentation, but it is not possible to conduct parameterizations. Although generally based on
experiments at the scale of the whole Earth. observations and simple statistical relationships, the
This means models themselves are being tested, parameterizations often are no more than educated
in any study using them, to examine the behavior of a guesses. Representation of sub-grid scale processes is
phenomenon. The definition of an atmospheric just one of the problems with models, but more
(climate) model is: a hypothesis [frequently in the computer programming resources are devoted to it
form of mathematical statements] that describes the than to the basic equations referred to above.
processes physically important to describe the There are other problems with the models that
workings of the atmosphere (climate and/or climatic manifest themselves as “computational error,” which
change), that has physical consistency in the model with time will eventually cause the system evolution
formulation, and the agreement with the observations to depart from that of the prototype (e.g., Haltiner and
that serve to ‘test’ the hypothesis [i.e., the model]. Williams, 1980, Durran, 1999).
The model is typically approximated for testing the First, there simply are not enough data available
hypothesis, but the model should conform to reality to establish the initial conditions. For example,
(AMS Glossary, 2000). weather forecasts are made with data measured twice
Once formulated, any atmospheric or climate a day in the United States, but once a day in most
model is simply a “box” that represents our best other locations on the globe. Also, the highest density
estimate of the workings of the atmosphere or of weather information is garnered from stations over
climate. It is our best guess or approximation of the land, although data continue to be sparse over
main processes of the system being represented and uninhabited regions. There are vast areas where the
the mechanisms that link the processes. These models atmosphere is poorly represented or sampled by
can be as complex or as simple as the model creators conventional land-based balloon soundings.
make them. To some extent the data problem is overcome by
A model can be statistical or dynamic, and here the use of satellite information that has global
we focus mainly on dynamic models, or what are coverage, albeit the satellite data differ somewhat
called general circulation models. In a dynamic from traditional thermometer-based observations.
model, the system is represented in three dimensions, Atmospheric satellite sounding data differ from
the characteristics of the system are specified at an radiosonde data, and satellite-derived ocean skin
initial time, and the system is allowed to evolve with temperature differs from ship and buoy observations
time in accordance with the governing equations and of ocean surface temperature, to name just two.
boundary conditions that link essential processes. Differences between the satellite and traditional data
An Atmosphere-Ocean General Circulation need to be reconciled in the establishment of starting
Model (AOGCM) is composed of seven basic conditions and the evaluation of predictions.
mathematical equations with seven basic variables Second, many atmospheric processes, such as
that describe the instantaneous state of the atmosphere thunderstorms, occur on space scales much smaller
over time. This represents a closed and solvable set of than a model’s resolution. Energy exchange processes
equations that can describe atmospheric motions and occurring on scales below the resolution of the model
12
must therefore be approximated, or parameterized at to the models.

the larger scale, and are therefore no longer It also is critical to understand the difference
mechanistic. between weather forecasting and the generation of
The inability to make the required observations climate projections or scenarios. Weather forecasting,
with infinite precision means there is always some in principle, is referred to as an initial value problem.
degree of measurement error or uncertainty associated Observational data are gathered, quality-controlled,
with the initial conditions at the commencement of and rendered to the grid space. This is no small
the forecast period. This uncertainty and its impact on problem because data are gathered from several
the flow evolution can be measured by using sources, must be checked for consistency with other
differential equations and then making multiple runs data, and then are rendered to the grid space in a
of the model with slight variations in initial manner consistent with our basic understanding and
conditions. The error associated with the initial assumptions about atmospheric structure and
conditions amplifies into the flow structure and behavior. The forecasts that evolve from these initial
propagates through the system with time where it can conditions are then constrained by the governing
render a model’s prediction unreliable in as short a equations. To be useful, the evolution of the
time as four days (e.g., Lorenz, 1965). atmospheric flow must faithfully render the
This is not to imply that increased resolution in movement, development, and decay of the weather
the models will fix this problem. In fact, there is some systems specified in the initial analysis. For weather
indication that further increasing the resolution will forecasting models it is the structure and movement
lead to diminishing improvements when examined of the synoptic scale weather systems that is
against the cost, in computer time and budgets, of important. It is not as important to maintain global
increasing the resolution. energy equilibrium over the relatively short prediction
Third, there is also some difficulty in representing period of weather forecast.
the mathematical processes of the basic equations on How far out in time a useful forecast can be
the fixed grid space. The resolution of the model generated depends on the size and rotation rate of a
means the processes cannot be adequately specified planet, as well as the mixture of gases that make up
and errors in representing the local gradients are the atmosphere. Using Earth’s atmosphere and
amplified during the forecast period. Numerical finite dimensions, it is widely accepted that a useful
difference methods are used generally to solve the forecast can be made for no more than 10 to 15
GCM equations. For some operations there are days—referred to as the forecasting “wall.” Clearly,
several types of methods available (e.g., Wicker and another strategy is needed in making both long-range
Skamarock, 2002), making numerical modeling a weather and climate forecasts.
matter of choosing the right tool for the job. Climate modeling of the atmosphere also involves
GCMs have the added complexity of coupling an a boundary value problem. The climate system
ocean model, with its own difficulties of process responds very quickly (relatively speaking) to
specification, to an atmospheric model and regulating changes in the pattern of radiation exchange to and
the energy exchange processes through the system by from space and to changes in the pattern of heat and
specified energy constraints. As more processes are moisture exchange between the underlying surface
introduced into the climate model—such as the and the atmosphere. Tectonic movements of the
energetics of the cryosphere, water storage in soils, continents, on the other hand, are so slow that the
and the changing of vegetation patterns—the model land and ocean distribution is generally treated as a
becomes considerably more complex and the constant. But heat and moisture exchanges with the
potential for errors to be generated and to amplify in underlying surface vary significantly with ocean
the output is increased. surface temperature distribution and vegetation states.
All of these problems must be balanced against Thus the concept of atmospheric climate can be
the amount of data a computer can process and how thought of as a servant dependent on the underlying
long it takes to produce a model simulation. Thus surface. To complicate matters, scientists are still not
even this brief discussion and introduction of completely sure about how the exchange of heat and
computer models demonstrates the skepticism mass between the atmosphere and ocean (and land)
through which the validity of model simulations take place.
should be evaluated. Unfortunately, there is little The veracity of climate forecasts depends not
discussion by the IPCC about these problems inherent only on the ability of the general circulation equations
13
to reproduce the flow structures but also on the Niña), the Arctic Oscillation (AO)/North Atlantic
specification of the many sub-grid scale processes Oscillation (NAO), and, on longer time scales, the
that regulate energy, moisture, and momentum Pacific Decadal Oscillation (PDO) and Atlantic
exchanges at these scales. Errors in any of these Multidecadal Oscillation (AMO). The physical
specifications will bias local energy accumulations workings of these processes are not yet well
and become evident as a false trend in the evolution understood, thus it is not surprising there is difficulty
of local climate indices. Many of the sub-grid scale in modeling them. The effect of these cycles on
exchanges are functions of local temperature or regional and even global climate is not trivial, and
temperature gradient; locally developing biases will failure to understand a phenomenon does not excuse
propagate spatially with time. leaving it out of a model, which is often done.
Thus the main distinction between weather Omitting such phenomena subjects the models to
forecast models and climate models is that, in the additional errors and failure.
former, the objective is to specify the individual In evaluating model reliability, the standard
weather systems and reproduce their travel and decay, assumption has been to compare model output with
together with the development of new systems, over observations over a given period of Earth’s past
time. For climate models the interest is not in climate, e.g. since 1850. Doing so, however, requires
individual weather systems that cannot be replicated subjective adjustments to the model; for example, in
beyond about two weeks, but in the subtle changes in order to replicate estimated changing solar intensity
energy reservoirs over time, especially warming land and atmospheric aerosol loading. These adjustments
surfaces, the ocean surface mixed layer, and create additional sources of uncertainty, as limited
cryosphere extent. This difference in objective is very data exist in the past and knowledge about internal
large in principle. variability and the role of multidecadal-to-century-
There are two distinct types of climate models. scale climate cycles is also restricted.
Diagnostic, or equilibrium climate models (ECMs) Perhaps it is in light of such limitations that in
represent steady-state or unchanging processes with Chapter 9 of the Intergovernmental Panel on Climate
time. The ECM is most commonly solved for climatic Change’s (IPCC) Fifth Assessment Report (AR5)
means and variations and can employ statistical or (e.g., IPCC, 2013-I) it is conceded that “climate
dynamical methods, or some combination of these, to models provide realistic simulations of the large-scale
generate a future climate. features in the climate system and reproduce observed
In the second type of climate model, the historical change with only some fidelity. The climate
Prognostic, changes in variables with time are crucial. sensitivity of current models has not changed
The time variation for a given variable is the desired dramatically from that of models assessed in the AR4,
output (i.e., a time series). Thus climatic means and in spite of many improvements to the models’
variances are changing and can be calculated and representation of physical processes.”
compared over limited time intervals. Another strategy in creating model projections is
Both types of models are employed in the study to first allow the climate model to equilibrate; i.e., to
of climate. Generally, the diagnostic model is simpler find a steady climate state under control conditions
and produces numerical output faster. Prognostic and then again under conditions that may exist in the
models are more complex. Either modeling technique future (e.g., double the CO2 concentration). Then, one
can be utilized to create future climate projections or can analyze the differences in equilibrium climates in
scenarios. order to project how the climate of this future time
One way of making a climate model projection is will look. It should be kept in mind that a steady-state
to start with today’s conditions in a weather forecast climate may not exist in nature.
model retooled for climate simulation. Researchers What about projections that show global
“add” carbon dioxide to replicate the rise in temperature may rise by as much as 1o C to 6o C by
greenhouse gases and then run the model to see what 2100 (e.g., IPCC, 2007-I)? Such projections are based
output it produces. This approach would seem to be on a strategy in which a model is run many, many
ideal, except there are several atmosphere-ocean times from a particular set of initial conditions
interaction processes acting on multiyear to followed by slightly altered initial conditions. This is
multidecadal time scales that models have not yet called a model ensemble. Generally, due to many of
mastered. Examples of these phenomena are the El the problems discussed here and to the nature of the
Niño–Southern Oscillation (ENSO, or El Niño and La basic equations used in the model, the large range in
14
global temperature projections is a natural result of Haltiner, G.J. and Williams, R.T. 1980. Numerical
the ensemble technique. The more times a model is Prediction and Dynamic Meteorology, 2nd ed. Wiley and
run and the longer the time period of evolution used, Sons, Inc.
the greater the spread in the range of the predicted IPCC. 2013-I. Climate Change 2013: The Physical Science
variable, such as global temperature. This is referred Basis. Contribution of Working Group I to the Fourth
to as sensitivity to initial conditions (SDIC). Such Assessment Report of the Intergovernmental Panel on
behavior is inherent in any system that displays Climate Change.
chaotic characteristics, as does Earth’s climate
IPCC. 2007-I. Climate Change 2007: The Physical Science
system. Chaos theory is another name for the study of
Basis. Contribution of Working Group I to the Fourth
such nonlinear dynamics, which are represented in the Assessment Report of the Intergovernmental Panel on
raw forms of the basic equations. Climate Change. Solomon, S., Qin, D. Manning, M. Chen,
The IPCC places great confidence in the ability of Z., Marquis, M. Averyt, K.B., Tignor, M., and Miller, H.L..
GCMs to simulate future climate and attribute (Eds.) Cambridge University Press, Cambridge, UK.
observed climate change to anthropogenic emissions
of greenhouse gases. It says “climate models are Lorenz, E.N. 1965. A study of the predictability of a 28-
variable model. Tellus 17: 321–333.
based on well-established physical principles and
have been demonstrated to reproduce observed Riehl, H. and Malkus, J. 1958. On the heat balance in the
features of recent climate … and past climate equatorial trough zone. Geophysica 6: 3–4.
changes. … There is considerable confidence that
Wicker, L.J. and Skamarock, W.C. 2002. Time-splitting
Atmosphere-Ocean General Circulation Models
methods for elastic models using forward time schemes.
(AOGCMs) provide credible quantitative estimates of Monthly Weather Review 130: 2088–2097.
future climate change, particularly at continental and
larger scales” (IPCC, 2007-I, p. 591).
The IPCC’s confidence in the models, however, 1.1 Model Simulation and Forecasting
is likely considerably overstated. The magnitude of
the range of projected temperature responses to a 1.1.1 Methods and Principles
doubling of atmospheric CO2 itself suggests there are J. Scott Armstrong, a professor at The Wharton
large errors and limitations in the models that must be School of the University of Pennsylvania and a
overcome. To have any validity in terms of future leading figure in forecasting, has pointed out that
projections, GCMs must incorporate not only the forecasting is a scientific discipline built on more than
many physical processes described above but also the 70 years of empirical research, with its own institute
chemical and biological processes that influence (International Institute of Forecasters, founded in
climate over long time periods. In addition, current 1981), peer-reviewed journals (International Journal
computational errors resulting from finite grid of Forecasting and Journal of Forecasting), and
resolution must be overcome so as not to introduce annual International Symposium on Forecasting. The
growing biases. And as a final step, model output research on forecasting has been summarized as
must be compared with and evaluated against real- scientific principles, currently numbering 140, that
world observations. must be observed in order to make valid and useful
The remainder of this chapter delves further into forecasts (Principles of Forecasting: A Handbook for
the complexities and problems inherent to computer Researchers and Practitioners, edited by J. Scott
modeling of the climate system. Other chapters in this Armstrong, Kluwer Academic Publishers, 2001).
volume serve to evaluate the model projections using When physicists, biologists, and other scientists
real-world data observations. who are unaware of the rules of forecasting attempt to
make climate predictions, their forecasts are at risk of
References being no more reliable than those made by non-
experts, even when they are communicated through
American Meteorological Society (AMS). 2000. Glossary complex computer models (Green and Armstrong,
of Meteorology, Allen Press. 2007). In other words, when faced with forecasts by
Durran, D.R. 1999. Numerical Methods for Wave scientists, even large numbers of very distinguished
Equations in Geophysical Fluid Dynamics. Springer- scientists, one cannot assume the forecasts are
Verlag, Inc.. scientific. Green and Armstrong cite research by
Philip E. Tetlock (2005), a psychologist and now
15
professor at the University of Pennsylvania, who Scientists working in fields characterized by

“recruited 288 people whose professions included complexity and uncertainty are apt to confuse the
‘commenting or offering advice on political and output of models—which are nothing more than a
economic trends.’ He asked them to forecast the statement of how the modeler believes a part of the
probability that various situations would or would not world works—with real-world trends and forecasts
occur, picking areas (geographic and substantive) (Bryson, 1993). Computer climate modelers
within and outside their areas of expertise. By 2003, frequently fall into this trap and have been severely
he had accumulated more than 82,000 forecasts. The criticized for failing to notice their models fail to
experts barely, if at all, outperformed non-experts, replicate real-world phenomena by many scientists,
and neither group did well against simple rules” including Balling (2005), Christy (2005), Essex and
(Green and Armstrong, 2007). The failure of expert McKitrick (2007), Frauenfeld (2005), Michaels
opinion to provide reliable forecasts has been (2000, 2005, 2009), Pilkey and Pilkey-Jarvis (2007),
confirmed in scores of empirical studies (Armstrong, Posmentier and Soon (2005), and Spencer (2008).
2006; Craig et al., 2002; Cerf and Navasky, 1998; Canadian science writer Lawrence Solomon
Ascher, 1978) and illustrated in historical examples of (2008) asked many of the world’s leading scientists
wrong forecasts made by leading experts, including active in fields relevant to climate change for their
such luminaries as Ernest Rutherford and Albert views on the reliability of computer models used by
Einstein (Cerf and Navasky, 1998). the IPCC to detect and forecast global warming. Their
In 2007, Armstrong and Kesten C. Green of the answers showed a high level of skepticism:
Ehrenberg-Bass Institute at the University of South • Prof. Freeman Dyson, professor of physics at
Australia conducted a “forecasting audit” of the IPCC the Institute for Advanced Study at Princeton
Fourth Assessment Report (Green and Armstrong, University and one of the world’s most eminent
2007). The authors’ search of the contribution of physicists, said the models used to justify global
Working Group I to the IPCC “found no references warming alarmism are “full of fudge factors” and “do
… to the primary sources of information on not begin to describe the real world.”
forecasting methods” and “the forecasting procedures • Dr. Zbigniew Jaworowski, chairman of the
that were described [in sufficient detail to be Scientific Council of the Central Laboratory for
evaluated] violated 72 principles. Many of the Radiological Protection in Warsaw and former chair
violations were, by themselves, critical.” of the United Nations Scientific Committee on the
Green and Armstrong found the IPCC violated Effects of Atomic Radiation, a world-renowned
“Principle 1.3 Make sure forecasts are independent of expert on the use of ancient ice cores for climate
politics.” The two authors write, “this principle refers research, said the U.N. “based its global-warming
to keeping the forecasting process separate from the hypothesis on arbitrary assumptions and these
planning process. The term ‘politics’ is used in the assumptions, it is now clear, are false.”
broad sense of the exercise of power.” Citing David • Dr. Richard Lindzen, professor of
Henderson (2007), a former head of economics and meteorology at the Massachusetts Institute of
statistics at the Organization for Economic Technology and member of the National Research
Cooperation and Development (OECD), Green and Council Board on Atmospheric Sciences and Climate,
Armstrong state, “the IPCC process is directed by said the IPCC is “trumpeting catastrophes that
non-scientists who have policy objectives and who couldn’t happen even if the models were right.”
believe that anthropogenic global warming is real and • Prof. Hendrik Tennekes, director of research
dangerous.” They thus conclude: at the Royal Netherlands Meteorological Institute,
said “there exists no sound theoretical framework for
The forecasts in the Report were not the outcome climate predictability studies” used for global
of scientific procedures. In effect, they were the
warming forecasts.
opinions of scientists transformed by mathematics
and obscured by complex writing. Research on • Dr. Richard Tol, principal researcher at the
forecasting has shown that experts’ predictions are Institute for Environmental Studies at Vrije
not useful in situations involving uncertainty and Universiteit and adjunct professor at the Center for
complexity. We have been unable to identify any Integrated Study of the Human Dimensions of Global
scientific forecasts of global warming. Claims that Change at Carnegie Mellon University, said the
the Earth will get warmer have no more credence IPCC’s Fourth Assessment Report is “preposterous ...
than saying that it will get colder.
16
alarmist and incompetent.” Green and Armstrong subsequently collaborated

• Dr. Antonino Zichichi, emeritus professor of with Willie Soon in conducting validation tests of the
physics at the University of Bologna, former IPCC forecasts of global warming (Green,
president of the European Physical Society, and one Armstrong, and Soon 2009). To do so, they tested
of the world’s foremost physicists, said global whether the warming-trend forecasts used by the
warming models are “incoherent and invalid.” IPCC are more accurate than the standard benchmark
Princeton’s Freeman Dyson has written forecast that there will be no change. They tested the
elsewhere, “I have studied the climate models and I IPCC’s “business as usual” 0.03˚C p.a. forecast and
know what they can do. The models solve the the no-change forecast from one to 100 years ahead
equations of fluid dynamics, and they do a very good on a rolling basis over the period of exponentially
job of describing the fluid motions of the atmosphere increasing human CO2 emissions from 1851 to 1975.
and the oceans. They do a very poor job of describing The procedure generated 7,550 forecasts from each of
the clouds, the dust, the chemistry, and the biology of the forecasting procedures.
fields and farms and forests. They do not begin to The Green, Armstrong, and Soon validation test
describe the real world that we live in” (Dyson, was a weak one, in that the IPCC forecasts were
2007). tested against historical data (HadCRUt3) the IPCC
Many of the scientists cited above observe modelers knew exhibited a warming trend. Green,
computer models can be “tweaked” to reconstruct Armstrong, and Soon therefore were surprised to find
climate histories after the fact. But this provides no the errors from the IPCC warming trend forecasts
assurance that the new model will do a better job of were nearly eight times greater than the errors from
forecasting future climates, and it points to how the no-change forecasts. For the longer 91 to 100
unreliable the models are. Individual climate models years-ahead forecast horizons, the IPCC errors were
often have widely differing assumptions about basic nearly 13 times greater for the 305 forecasts. The no-
climate mechanisms but are then “tweaked” to change forecasts were so accurate in the validation
produce similar forecasts. This is nothing like how test that the authors forecast annual global mean
real scientific forecasting is done. temperatures will be within 0.5˚C of the 1988 figure
Kevin Trenberth, a lead author along with Philip for the next 100 years. For public policy and business
D. Jones of Chapter 3 of the Working Group I planning purposes, it is hard to see that any economic
contribution to the IPCC’s Fourth Assessment Report, benefit could be obtained from forecasts that were
replied to some of these scathing criticisms on the less accurate than forecasts from the no-change
blog of the science journal Nature. He argued “the model. The implication of the Green, Armstrong, and
IPCC does not make forecasts” but “instead proffers Soon forecasts is that the best policy is to do nothing
‘what if’ projections of future climate that correspond about global warming. Their findings did not,
to certain emissions scenarios” and then hopes these however, stop the claims that “we are at a turning
“projections” will “guide policy and decision makers” point” and “it is different this time.”
(Trenberth, 2007). He says “there are no such If public policy to address global warming is to
predictions [in the IPCC reports] although the be made rationally, it must be based on scientific
projections given by the Intergovernmental Panel on forecasts of (1) substantial global warming, the
Climate Change (IPCC) are often treated as such. The effects of which are (2) on balance seriously harmful,
distinction is important.” and for which (3) cost-effective policies can be
This defense is hardly satisfactory. As Green and implemented. Armstrong, Green, and Soon (2011)
Armstrong (2007) point out, “the word ‘forecast’ and refer to these logical requirements of policymaking as
its derivatives occurred 37 times, and ‘predict’ and its “the 3-legged stool” of global warming policy. A
derivatives occurred 90 times in the body of Chapter failure of any leg would invalidate policies. To date,
8 of the Working Group I report, and a survey of there are no evidence-based forecasts to support any
climate scientists conducted by those same authors of the legs.
found “most of our respondents (29 of whom were
IPCC authors or reviewers) nominated the IPCC References
report as the most credible source of forecasts (not
‘scenarios’ or ‘projections’) of global average Armstrong, J.S. 2001. Principles of Forecasting—A
temperature.” Green and Armstrong conclude, “the Handbook for Researchers and Practitioners. Kluwer
IPCC does provide forecasts.” Academic Publishers, Norwell, MA.
17
Armstrong, J.S. 2006. Findings from evidence-based Washington, DC.

forecasting: Methods for reducing forecast error.
International Journal of Forecasting 22: 583–598. Michaels, P.J. 2005. Meltdown: The Predictable Distortion
of Global Warming by Scientists, Politicians and the
Armstrong, J.S., Green, K.C., and Soon, W. 2011. Media. Cato Institute, Washington, DC.
Research on forecasting for the global warming alarm.
Energy and Environment 22: 1091–1104. Michaels, P.J. 2000. Satanic Gases: Clearing the Air About
Global Warming. Cato Institute. Washington, DC.
Ascher, W. 1978. Forecasting: An Appraisal for Policy
Makers and Planners. Johns Hopkins University Press. Pilkey, O.H. and Pilkey-Jarvis, L. 2007. Useless
Baltimore, MD. Arithmetic. Columbia University Press, New York, NY.
Balling, R.C. 2005. Observational surface temperature Posmentier, E.S. and Soon, W. 2005. Limitations of
records versus model predictions. In Michaels, P.J. (Ed.) computer predictions of the effects of carbon dioxide on
Shattered Consensus: The True State of Global Warming. global temperature. In Michaels, P.J. (Ed.) Shattered
Rowman & Littlefield. Lanham, MD. 50–71. Consensus: The True State of Global Warming. Rowman &
Littlefield. Lanham, MD. 241–281.
Bryson, R.A. 1993. Environment, environmentalists, and
global change: A skeptic’s evaluation. New Literary Solomon, L. 2008. The Deniers: The World Renowned
History 24: 783–795. Scientists Who Stood Up Against Global Warming
Hysteria, Political Persecution, and Fraud—And those
Cerf, C. and Navasky, V. 1998. The Experts Speak. Johns who are too fearful to do so. Richard Vigilante Books.
Hopkins University Press. Baltimore, MD. Minneapolis, MN.
Christy, J. 2005. Temperature changes in the bulk Spencer, R. 2008. Climate Confusion: How Global
atmosphere: Beyond the IPCC. In Michaels, P.J. (Ed.) Warming Hysteria Leads to Bad Science, Pandering
Shattered Consensus: The True State of Global Warming. Politicians and Misguided Policies that Hurt the Poor.
Rowman & Littlefield. Lanham, MD. 72–105. Encounter Books, New York, NY.
Craig, P.P., Gadgil, A., and Koomey, J.G. 2002. What can Tetlock, P.E. 2005. Expert Political Judgment—How Good
history teach us? A retrospective examination of long-term Is It? How Can We Know? Princeton University Press,
energy forecasts for the United States. Annual Review of Princeton, NJ.
Energy and the Environment 27: 83–118.
Trenberth, K.E. 2007. Global warming and forecasts of
Dyson, F. 2007. Heretical thoughts about science and climate change. Nature blog. http://blogs.nature.com/
society. Edge: The Third Culture. August. climatefeedback/2007/07/global_warming_and_forecasts_o
.html. Last accessed May 6, 2009.
Essex, C. and McKitrick, R. 2007. Taken by Storm. The
Troubled Science, Policy and Politics of Global Warming.
Key Porter Books. Toronto, Canada. 1.1.2 Computational Issues
Frauenfeld, O.W. 2005. Predictive skill of the El Niño- To commemorate the publication of the 100th volume
Southern Oscillation and related atmospheric of the journal Climatic Change, Norman Rosenberg
teleconnections. In Michaels, P.J. (Ed.) Shattered (Rosenberg, 2010) was asked to contribute an
Consensus: The True State of Global Warming. Rowman & overview paper on progress that had occurred since
Littlefield. Lanham, MD. 149–182. the journal’s inception in the interrelated areas of
climate change, agriculture, and water resources.
Green, K.C. and Armstrong, J.S. 2007. Global warming:
Rosenberg accepted and at the age of 80 conducted
forecasts by scientists versus scientific forecasts. Energy
Environ. 18: 997–1021. his review quite admirably.
He began by noting the “overarching concern” of
Green, K.C., Armstrong, J.S., and Soon, W. 2009. Validity the volumes he edited was “to gain understanding of
of climate change forecasting for public policy decision how climatic change affects agricultural production,
making. International Journal of Forecasting 25: 826–832. unmanaged ecosystems and water resources; how
Henderson, D. 2007. Governments and climate change farmers, foresters and water managers can strengthen
issues: The case for rethinking. World Economics 8: 183– these sectors against the negative impacts of climatic
228. change and capitalize on positive impacts if any; how
they can adapt to impacts that cannot be so modified
Michaels, P.J. 2009. Climate of Extremes: Global Warming or ameliorated and how they can contribute directly
Science They Don’t Want You to Know. Cato Institute.
or indirectly to mitigation of anthropogenic climatic
18
change—as, for example, through soil carbon claims to show “climate change that takes place due
sequestration and the production of biomass to to increases in carbon dioxide concentration is largely
substitute in part for the fossil fuels that are adding irreversible for 1,000 years after emissions stop”
CO2 to the atmosphere.” (Solomon et al., 2009). In the virtual world of
Rosenberg writes in his closing paragraph, “it computer-run climate models, that may be the case,
seems difficult to say with assurance that the ‘state- but that may not be true of the real world. Consider,
of-the-art’ in projecting climatic change impacts on for example, that the discernible climate signal from a
agriculture and water resources and unmanaged major volcanic eruption is lost after only a few years.
ecosystems is, today, that much better than it was 30 In their paper, Solomon et al. set forth three
years ago,” noting “the uncertainty and lack of criteria they say should be met by the modeled
agreement in GCMs is still too great.” He reported, climatic parameters they studied: “(i) observed
“much can and has been learned about possible changes are already occurring and there is evidence
outcomes,” but “for actual planning and policy for anthropogenic contributions to these changes, (ii)
purposes we are still unable to assure those who need the phenomen[a] [are] based upon physical principles
to know that we can forecast where, when and how thought to be well understood, and (iii) projections
much agriculture (as well as unmanaged ecosystems are available and are broadly robust across models.”
and water resources) will be affected by climatic Real-world data provide little or no support for
change.” the first criterion (as discussed in other chapters of
A similarly pessimistic commentary on the state this volume). The global warming of the past few
of climate modeling appeared in 2010 in Nature decades was part of a much longer warming trend that
Reports Climate Change. Kevin Trenberth, head of began in many places throughout the world a little
the Climate Analysis Section of the National Center more than three centuries ago (about 1680) with the
for Atmospheric Research in Boulder, Colorado dramatic “beginning of the end” of the Little Ice Age
(USA), writes that one of the major objectives of (LIA, see Figure 1.1.1), well before there was any
upcoming climate modeling efforts will be to develop significant increase in the air’s CO2 content. This
“new and better representations of important climate observation suggests a continuation of whatever
processes and their feedbacks.” The new work, phenomenon—or combination of phenomena—may
Trenberth wrote, should increase “our understanding have caused the greater initial warming may have
of factors we previously did not account for ... or even caused the lesser final warming, the total effect of
recognize.” which has been to transport Earth from the chilly
In expressing these sentiments, Rosenberg and depths of the Little Ice Age into the relative balminess
Trenberth gave voice to the concerns of many of the Current Warm Period.
scientists who are skeptical of the reliability of Climate history is discussed in greater detail in
GCMs. Such concerns should not be misinterpreted as Chapter 4, but it is useful to note here that Earth’s
“denial.” Trenberth, at least, would deny being a current temperature is no higher now (and may be
“skeptic” of the theory of anthropogenic global slightly less) than it was during the peak warmth of
warming. It is, rather, the humility of true scientists the Medieval Warm Period (MWP), when there was
who—attempting to comprehend the complexity of more than 100 ppm less CO2 in the air atmosphere
the world of nature and its innermost workings—are there is today. Consequently, since the great MWP-to-
well aware of their own limitations and those of all LIA cooling phase occurred without any significant
seekers of scientific truths. Although much has been change in the atmosphere’s CO2 concentration, the
learned, as Rosenberg and Trenberth outline in their opposite could occur just as easily. The planet could
respective essays, what is known pales in comparison warm, and by an equal amount, just as it actually did
to what is required “for actual planning and policy over the past three centuries without any help from an
purposes,” as Rosenberg describes it, or “certainty” as increase in the atmosphere’s CO2 content.
Trenberth puts it. Regarding the second criterion of Solomon et al.,
In contrast, consider a paper that fails to studies reported in this volume (see Chapter 2) also
recognize any such problems. Published in the show there are non-modeled chemical and biological
Proceedings of the National Academy of Sciences of processes that may be equally as important as the
the United States of America and written by Susan changes in radiation fluxes associated with carbon
Solomon (a co-chair of the IPCC’s 2007 Working dioxide employed in the models. The chemical and
Group 1 report for AR4) and three coauthors, it biological processes are simply not as “well
19
atmospheric processes affecting the climate of Europe

that models currently do not simulate well, noting (1)
the location of the jet stream over northern Europe in
most models diverges from reality, (2) zonal flow is
biased too far south in most models, (3) the models
can’t simulate or explain the North Atlantic
Oscillation with sufficient magnitude to match
historical data, and (4) heat waves and droughts, such
as the summer 2010 Moscow heat wave and fires, are
caused by blocking, a process the models are
currently unable to simulate reliably even in weather
forecast models (e.g., Matsueda, 2011), let alone
climate models.
Figure 1.1.1. The mean relative temperature history of the In addition, for several key processes the models
Earth (blue, cool; red, warm) over the past two millennia— produce widely varying predictions. The atmospheric
adapted from Loehle and McCulloch (2008)—highlighting circulation response to warming in climate models,
the Medieval Warm Period (MWP) and Little Ice Age (LIA), for example, is highly variable, as is the change in
together with a concomitant history of the atmosphere’s CO2 storm intensity, the projected change in the jet stream,
concentration (green). and changes in temperature. It is particularly
noteworthy that Europe is predicted to warm less than
understood” as Solomon et al. claim. A highly most Northern Hemisphere sites due to the slowing of
selective reading of the literature is required to miss the Gulf Stream providing reduced northward heat
the repeated admissions by leading researchers of the transport, a factor Woollings noted the models do not
uncertainty and outright ignorance of underlying simulate well.
processes that undermine the reliability of GCMs. It is thus easy to recognize that current climate
Regarding the third criterion of Solomon et al., models are unable to achieve the degree of accuracy
many computer model projections are “available and necessary in the details of atmospheric circulation that
are broadly robust across models.” To say such are critical to replicating current weather events, such
models are “robust” implies they are producing stable as droughts, heat waves, and major storms that
climate simulations that are similar to observations. contribute to the characteristic climate in Europe. Any
But these models often diverge so greatly in their assertion that frequency or intensity of these events
assumptions and in their specific spatial and temporal can be forecast 100 years in the future under a
projections that they cannot be said to validate each changed climate is simply false, and claims about
other. Additionally, there is no scientific basis for the negative impacts of climate change in Europe are
often-made claim that an average from such based upon no specific modeling skill.
discordant projections can be meaningful. Another problem with climate models is climate
Furthermore, many studies have identified real-world drift. Sen Gupta et al. (2012) write, “even in the
data that contradict what the models say should be absence of external forcing, climate models often
occurring. exhibit long-term trends that cannot be attributed to
A good example of an admission of the wide natural variability,” and they state “this so-called
range of uncertainty that undermines GCMs appears climate drift arises for various reasons,” such as
in Woollings (2010): “the spread between the “deficiencies in either the model representation of the
projections of different models is particularly large real world or the procedure used to initialize the
over Europe, leading to a low signal-to-noise ratio. model.” They note, however, that “significant efforts
This is the first of two general reasons why European by the climate modeling community have gone into
climate change must be considered especially reducing climate drift.” Nevertheless, they write,
uncertain. The other is the long list of physical “climate drift still persists.”
processes which are very important for defining Sen Gupta et al. “quantify the size of drift relative
European climate in particular, but which are to twentieth-century trends in climate models taking
represented poorly in most, if not all, current climate part in the Coupled Model Intercomparison Project
models.” phase 3 (CMIP3),” which they say “was used to
Woollings cited several examples of key inform the Intergovernmental Panel on Climate
20
Change (IPCC) Forth Assessment Report (AR4).” Climate drift in the CMIP3 models. Journal of Climate 25:
According to the seven Australian scientists, their 4621–4640.
analysis determined that below 1-2 km in the deep Solomon, S., Plattner, G.-K., Knutti, R., and Friedlingstein,
ocean, or for depth-integrated properties, drift P. 2009. Irreversible climate change due to carbon dioxide
generally dominates over any forced trend. They emissions. Proceedings of the National Academy of
report drift in sea level can be large enough to reverse Sciences USA 106: 1704–1709.
the sign of the forced change, “both regionally and in
some models for the global average.” In addition, Trenberth, K. 2010. More knowledge, less certainty.
Nature Reports Climate Change: 10.1038/climate.2010.06.
because surface drift is spatially heterogeneous, they
say “the regional importance of drift for individual Tziperman, E. 2000. Uncertainties in thermohaline
models can be much larger than the global figures circulation response to greenhouse warming. Geophysical
suggest.” As an example, they note “a typical error in Research Letters 27: 3077–3080.
calculating a regional forced sea surface temperature
Woollings, T. 2010. Dynamical influences on European
trend in the Bjerknes Center for Climate Research climate: An uncertain future. Philosophical Transactions of
Bergen Climate Model, version 2.0 (BCM2.0), the Royal Society A 368: 3733–3756.
CSIRO Mk3.0, and GISS-EH models without
accounting for drift would be 30% to 40%.” Because
this is an average value, still-larger errors would be 1.1.3 Dealing with Chaos
expected at some locations.
While providing some suggestions for addressing 1.1.3.1 Chaotic Systems
climate drift modeling problems, Sen Gupta et al. The ability of atmosphere-ocean GCMs to predict the
write, “in the absence of a clear direction forward to climatic effects of human alterations of greenhouse
alleviate climate drift in the near term, it seems gases and other factors cannot be tested directly with
important to keep open the question of flux respect to a point in time a hundred years in the
adjustment within climate models that suffer from future. However, it is still possible to determine
considerable drift.” They indicate “flux adjustments whether those models can in principle make such
are nonphysical and therefore inherently undesirable” predictions with a reasonable degree of accuracy.
and “may also fundamentally alter the evolution of a One way to evaluate this ability is to consider the
transient climate response,” citing the work of Neelin effects of errors in system initial values. If a system is
and Dijkstra (1995) and Tziperman (2000). well-behaved, small initial errors will lead to small
future errors or even damped responses. In a chaotic
References system, on the other hand, small initial errors will
cause trajectories to diverge over time; for such a
Idso, S.B. 1998. CO2-induced global warming: A skeptic’s system (or model), true predictability is low to
view of potential climate change. Climate Research 10: nonexistent. This does not mean realistic behavior in
69–82. the statistical sense cannot be simulated, only that
Loehle, C. and McCulloch, J.H. 2008. Correction to: A detailed predictability (will it rain 60 days from now,
2000-year global temperature reconstruction based on non- or how much will it rain this year) is impossible.
tree ring proxies. Energy & Environment 19: 93–100.
1.1.3.2 Sensitivity Dependence
Matsueda, M. 2011. Predictability of Euro-Russian
One of the characteristics of chaotic systems, such as
blocking in summer of 2010. Geophysical Research Letters
38: L06801, doi:10.1029/2010GL046557. the fluid we call our atmosphere, is sensitive
dependence on the initial conditions (SDIC). SDIC
Neelin, J.D. and Dijkstra, H.A. 1995. Ocean-atmosphere means one can take an initial state for our
interaction and the tropical climatology. Part I: The dangers atmosphere, including all the temperature, pressure,
of flux correction. Journal of Climate 8: 1325–1342. and other measurements, and put them into a
Rosenberg, N.J. 2010. Climate change, agriculture, water computer model and generate, for example, a 48-hour
resources: What do we tell those that need to know? weather forecast. If we use an identical model but
Climatic Change 100: 113–117. adjust these initial measurements by small amounts
representing error, it is possible to generate a 48-hour
Sen Gupta, A., Muir, L.C., Brown, J.N., Phipps, S.J., forecast much different from the first one.
Durack, P.J., Monselesan, D., and Wijffels, S.E. 2012.
In weather forecasting, for example, some 15 to
21
20 model runs are strategically generated in order to the-art AGCMs, as well as coupled models with
examine how much “spread” there is between the external forcings.” The authors also found Arctic sea
multiple runs. If the forecasts show little spread, a ice, CO2 changes, and stratospheric dynamics and
weather forecaster can be more confident in the chemistry also contributed to these trends separately
model projections; if there is a great deal of spread, and were each major contributors to the decadal
the forecaster must rely on other tools to develop a variations and trends. None of these forcings
forecast. separately or together was able to fully represent the
In a study addressing initial value errors, Collins observed trends during the 1958–1996 period. None
(2002) used the HadCM3 model, the output of which of the ensemble members could reproduce the
at a given date was used as the initial condition for amplitude of the trends reliably. As stated by the
multiple runs in which slight perturbations of the authors, something was missing: “another major
initial data were used to assess the effect of a lack of player in decadal climate variability is the ocean
perfect starting information, as can often occur in the circulation, which is not accounted for at all by the
real world. The results of the various experimental study here.”
runs were then compared to those of the initial control A frequent criticism of GCMs is their inability to
run, assuming the degree of correlation of the results effectively render past climate. The models used by
of each perturbed run with those of the initial run is a organizations such as the Intergovernmental Panel on
measure of predictability. Climate Change are similar to those used in the King
Collins found “annual mean global temperatures et al. study above. In King et al.’s paper the model
are potentially predictable one year in advance” and performed at its best only when tropical SSTs were
“longer time averages are also marginally predictable included. The authors also cite the need to include
five to ten years in advance.” In the case of ocean ocean dynamics. But even the use of ensemble
basin sea surface temperatures, coarse-scale techniques allowed for only limited success by the
predictability ranges from one year to several years models. Clearly, caution should be taken in
were found. But for land surface air temperature and interpreting future climate scenarios.
precipitation, and for the highly populated northern
land regions, Collin concludes, “there is very little References
sign of any average potential predictability beyond
seasonal lead times.” Collins, M. 2002. Climate predictability on interannual to
King et al. (2010) used an atmospheric GCM to decadal time scales: the initial value problem. Climate
gauge the ability of models to reproduce observed Dynamics 19: 671–692.
climate trends from the 1960s to the 1990s using King, M.P., Kucharski, F. and Molteni, F. 2010. The roles
model ensembles. They also attempted to quantify the of external forcings and internal variabilities in the
influence of driving factors both internal and external Northern Hemisphere atmospheric circulation change from
such as sea ice, stratospheric ozone, greenhouse the 1960’s to the 1990s. Journal of Climate 23: 6200–6220.
gases, and internal atmospheric variability. Their
research was performed using a 100-member
ensemble with climatological sea surface 1.1.4 Carbon Dioxide Forcing
temperatures (SSTs) over the globe from 1870 to The interaction between atmospheric carbon dioxide
2002. Three tests with ten members each were and Earth’s radiation field is at the heart of the
conducted, prescribing SSTs for tropical oceans, for anthropogenic climate change debate. In particular,
the Indian and Pacific Ocean, and for the tropical the effect of including in GCMs increasing
Pacific, respectively. concentrations of carbon dioxide has been to project
The authors found only when the tropical SSTs global temperature increases that have given rise to
were specified were the trends reproduced with a high additional climate-related concerns about potentially
degree of correlation (correlation = 0.80). The devastating impacts on the biosphere. The alleged
amplitude of these was only about 25 percent that of possibility of ecosystem extinctions is one example of
the observed amplitudes for the ensemble mean. such concerns that underlie calls to halt carbon
Individual ensemble members were at a maximum of dioxide emissions.
50 percent of the observed trends. The authors There is a long history of scientific debate linking
acknowledge “the underestimate of the trend carbon dioxide, through its interaction with Earth’s
amplitude is a common difficulty even for state-of- radiation field, to global climate and its variability.
22
The French mathematician Joseph Fourier (1824, human activity, especially the burning of fossil fuels.
1827) noted Earth should be colder than it is, given its The development of the first GCMs provided an
place in the solar system and the strength of solar opportunity to test the sensitivity of the climate
radiation it absorbs. Fourier’s explanation of the system to a CO2 forcing. The first GCMs were rather
apparently abnormal warmth was linked to the simple in construction, with the oceans represented as
insulating properties of Earth’s atmosphere. Earth’s a shallow swamp (Manabe et al., 1965; 1970).
greenhouse effect was claimed to be an outcome of Nevertheless, in steady-state the models were able to
absorption of radiation emitted from Earth’s surface represent the main features of Earth’s climate,
by gases in the atmosphere, which warmed the lower including the zonal gradients of temperature and
atmosphere and reduced infrared radiation emissions pressure and the main wind systems. Manabe et al.
to space. (1979) used a similar model to examine the effects of
Credence was given to Fourier’s hypothesis via a additional carbon dioxide. The GCM was run under
series of measurements carried out by the English conditions of 300 ppm carbon dioxide (1 X CO2) and
physicist John Tyndall beginning in the late 1850s. 1,200 ppm carbon dioxide (4 X CO2) and each came
Tyndall passed infrared (IR) radiation through to steady-state after about 12 years; the difference in
different atmospheric gases and measured the global average surface air temperature between the
absorption. He demonstrated water vapor is a strong two concentrations was 4.1°C.
absorber of infrared radiation, as is carbon dioxide In a general review of carbon dioxide and
and some other minor atmospheric constituents. climate, Manabe (1983) outlined the principles of a
The Swedish chemist Svante Arrhenius (1896) simple radiation-convection model that potentially
hypothesized that the shifts of Earth’s temperature yielded an indicative estimate of the sensitivity of
from glacial to interglacial conditions might be surface temperature to carbon dioxide forcing. First, it
explained by fluctuating changes in the atmospheric was noted earlier surface energy budget models
concentration of carbon dioxide. In simple terms, underestimated the sensitivity because of
when carbon dioxide concentrations are low, there is unrealistically large heat and moisture exchanges with
less absorption of infrared radiation in the atmosphere the boundary layer, a consequence of the assumption
and temperatures drop. But when concentrations are of constant temperature and specific humidity for that
high, the hypothesis suggests an increased radiative layer. An alternative approach was to drive the model
absorption of CO2 that keeps Earth’s temperature by changing the radiative divergence of the
warmer. The reason given for the hypothesized atmospheric layers as carbon dioxide concentration
fluctuating change in carbon dioxide concentration increased; net atmospheric radiation loss was offset
was varying volcanic activity: When volcanic activity by vertical convective mixing of heat and moisture
was low, less carbon dioxide was being emitted to the from the surface, the surface was assumed to have no
atmosphere than photosynthesis was removing, and heat capacity, and by convective adjustment the
atmospheric CO2 concentration fell; when activity tropospheric temperature lapse rate was constrained
was high, the atmospheric concentration increased. to no more than the saturated adiabatic lapse rate of
Arrhenius’ hypothesis linking glacial conditions 6.5°C/km. Such models consistently returned a
to low carbon dioxide concentrations fell from favour, surface temperature rise of between 1.5°C and 2.3°C
not because of the links to the greenhouse theory but for a doubling of the atmosphere’s carbon dioxide
because it became apparent that recurring glacial concentration.
events were regular and not explained by volcanic The Manabe review also outlined contemporary
activity. Nevertheless, through the twentieth century results from GCM forced by doubling of carbon
the notion that increasing carbon dioxide dioxide concentration to a new steady-state. Such
concentration in the atmosphere would increase models show consistent responses, including: (1)
global temperature remained an active hypothesis. stronger warming over polar regions due to positive
Systematic measurements of atmospheric carbon feedback as snow and sea ice melt to change surface
dioxide commenced at Mauna Loa Observatory, albedo, (2) amplification of tropical upper
Hawaii, during the 1957–58 International tropospheric warming due to the regulation of
Geophysical Year. Measurements at other sites have temperature by convective mixing, and (3) a
followed to give a global network of CO2 monitoring temperature increase in the Northern Hemisphere
and confirm a steadily increasing concentration of greater than that of the Southern Hemisphere. Results
atmospheric CO2. The increase has been attributed to from different models, each constructed on similar
23
principles, demonstrated a broader spread in the the range 12-18μm) vary the intensity of infrared
estimates of climate sensitivity, from 2°C to 3.9°C for radiation propagating both upwards and downwards
a doubling of carbon dioxide concentration. throughout the atmosphere, and hence the net
One characteristic of these early models was the radiation transfer from the surface to space. By
limited impact of the absence of ocean circulation. It increasing the concentration of carbon dioxide, the
was only over the North Atlantic region that emission to space in the active wavebands emanates
temperatures had a cold bias from the lack of ocean from higher in the atmosphere where temperatures are
heat transport. colder. As a consequence, the emission of radiation to
On the basis of such computer model findings, an space is reduced across the carbon dioxide
October 1985 U.N.-cosponsored conference in wavebands. To maintain balance in the overall
Villach, Austria (Bolin et. al., 1985) issued a infrared emission of energy to space it is therefore
statement asserting “many important economic and presumed that global temperatures would rise and
social decisions are being made today on long-term increase the emission intensity across non-carbon
projects … all based on the assumption that past dioxide wavebands.
climatic data, without modification, are a reliable As carbon dioxide concentrations increase so too
guide to the future. This is no longer a good does the intensity of back radiation at the surface
assumption since the increasing concentrations of across the active wavebands of CO2, and because this
greenhouse gases are expected to cause a significant radiation emanates from a lower and warmer layer of
warming of the global climate in the next century.” the atmosphere, the magnitude of the back radiation
The statement specifically claimed a doubling of increases. Consequently, the net infrared radiation
carbon dioxide concentration would lead to a global emanating from the surface is reduced, causing a rise
temperature rise between 1.5°C and 4.5°C. It also in temperature that generates increased heat exchange
asserted the global temperature rise of between 0.3°C and evaporation. This surface warming also
and 0.7°C during the twentieth century was consistent contributes to an increase in convective instability.
with the carbon dioxide increase, implying a cause- In addition to the reduction in infrared radiation
and-effect relationship, with CO2 as the cause. to space and the reduction in net infrared radiation
GCMs have evolved since those early days. loss from the surface, there is also a reduction in
Increased computing power has enabled higher spatial radiation flux divergence (cooling) over the
resolution both horizontally and vertically. The atmosphere, because the former is greater than the
models also better represent physical processes, latter. The reduction in radiative cooling is effectively
couple dynamic ocean and atmospheric circulations, a reduction in the rate of generation of convective
and include a range of bio-geo-chemical processes. instability necessary for distribution of heat and latent
Nevertheless, there remain fundamental problems energy from the surface and through the atmosphere.
with their representation and treatment of rising This is an additional, albeit indirect, factor leading to
carbon dioxide and its potential impact on climate. surface warming, which convectively influences
Early indicative measures of climate sensitivity tropospheric temperature.
were obtained via relatively simple models By convention, the sensitivity of surface
constrained by three essential assumptions: (1) there temperature to carbon dioxide forcing is expressed as
is an equilibrium balance between the solar radiation the relationship between the reduction in infrared
absorbed by the system and the infrared radiation radiation to space and the increase in surface
emitted to space, (2) the net radiation loss from the temperature. However, as described above, the
atmosphere (solar and infrared) is offset by heat and reduction in infrared radiation is confined to the
latent energy exchange from the surface, and (3) the carbon dioxide wavebands. As Earth’s climate
net radiation excess at the surface (solar and infrared) responds to increasing carbon dioxide, there is no
is offset by the surface-atmosphere heat and latent reduction in emission to space, only a shift in the
energy exchange plus heat that goes into surface distribution of energy across the infrared spectrum.
reservoirs (latent heat melting ice, warming of the The shift can be achieved by a general warming (as
land surface, and warming of the ocean). described in the sensitivity relationship) or by a
The concern over climate forcing by increasing change in circulation. An enhancement of convective
the carbon dioxide content of the atmosphere arises overturning will both expand the area occupied by
because changes in the absorption and emission of subtropical subsidence and increase the poleward
infrared radiation by CO2 in the active wavebands (in transport of heat. Enhanced subsidence will dry those
24
regions of the atmosphere, allowing emission to space temperature.

across the water vapor bands to emanate from a lower Nevertheless, after more than four decades of
and warmer region of the troposphere. Increased technological evolution in investigating the sensitivity
poleward transport of heat will warm the middle and of Earth’s climate to increasing carbon dioxide
high latitude troposphere. Both effects can increase concentration, there remains much uncertainty. The
the infrared radiation to space, but neither necessarily earliest simple surface energy balance models gave
leads to warming of the surface. very low sensitivity because the reduction in net
Held and Sodon (2006) analyzed the set of GCMs surface infrared radiation loss as carbon dioxide
used for the IPCC’s Fourth Assessment Report increased was offset largely by heat and moisture flux
(Solomon et al. 2007) and concluded there was a that damped surface temperature increase. The more
reduction in overturning as the model Earth warmed sophisticated radiation/convection models did not
with increasing atmospheric carbon dioxide. Their discriminate between sensible and latent heat loss and
analysis suggests the reason for the reduction in the approximately 2°C temperature rise for doubling
convective overturning was due to the differing rates of carbon dioxide concentration was considered
of increase in atmospheric water vapor increase and realistic. Contemporary high-resolution models with a
surface evaporation as temperature increased. In the complex representation of physics are even more
models the atmospheric water vapor mass increased sensitive to carbon dioxide forcing, but the clear
according to the Claussius Clapeyron relationship of suppression of surface evaporation increase with
about 7 percent per degree C (%/C), whereas the temperature possibly accounts for the heightened
surface evaporation increased at only about 4%/C. As sensitivity.
atmospheric water vapor increased, the convective This history in climate sensitivity makes clear
clouds processed the water vapor mass flow more there remain uncertainties with respect to how the
efficiently than the rate at which water vapor was interaction between increasing carbon dioxide
being delivered to the boundary layer by surface concentration and Earth’s infrared radiation fluxes
evaporation. As a consequence, a reduced rate of should be incorporated in complex GCMs. Also key
convective overturning could cope with the to determining climate sensitivity is the response of
marginally more active water cycle. the water cycle as the energy exchange processes
The convection overturning response identified in within the climate system respond. The early simple,
the GCMs, however, is in contrast to the tropical albeit flawed, surface energy balance models point to
observations of Chen et al. (2002), who identified a an enhanced water cycle damping the surface
decadal strengthening of the Hadley and Walker temperature sensitivity to carbon dioxide forcing.
Circulations during the warming period of the 1990s. This unresolved, elemental physics portrayal of the
The surface warming generated increased convective climate system does not lend confidence to current
instability as reflected in the responses of the two GCM projections under increasing carbon dioxide
major overturning circulations driven by buoyancy concentration.
forces.
Analyzing the same GCMs, Richter and Xie References
(2008) conclude there were three reasons for the
rather slow rate of increase in evaporation as Arrhenius, S., 1896. On the Influence of Carbonic Acid in
temperature increased: (1) an increase in boundary the Air upon the Temperature of the Ground. London,
layer stability, (2) a reduction in surface wind speed, Edinburgh, and Dublin Philosophical Magazine and
and (3) an increase in boundary layer relative Journal of Science (fifth series), April 1896. Volume 41:
237–275.
humidity. These are surprising findings that suggest
in the models, as carbon dioxide concentration Bolin, B., Döös, B.R., Jäger, J., and Warrick, R.A. 1986.
increases, the impact on convective instability from The greenhouse effect, climatic change, and ecosystems.
reduction in tropospheric radiation cooling Scientific Committee On Problems of the Environment
overwhelms the increase in convective instability (SCOPE).
from reduction in net surface infrared radiation loss. Chen, J., Carlson, B.E., and Del Genio, A.D. 2002.
The reduction in convective instability is realized as Evidence for strengthening of the tropical general
an increase in both boundary layer stability and circulation in the 1990s. Science 295: 838–841.
boundary layer relative humidity, each tending to
dampen the rate of increase in evaporation with Fourier, J. 1827. Mémoire sur les températures du globe
25
terrestre et des espaces planétaires. Mémoires de

l’Académie Royale des Sciences 7: 569–604.
Fourier, J. 1824. Remarques générales sur les températures
du globe terrestre et des espaces Planétaires. Annales de
Chimie et de Physique 27: 136–67.
Manabe, S., Smagorinsky, J., and Strickler, R.F. 1965.
Simulated climatology of a general circulation model with
a hydrologic cycle. Monthly Weather Review 93: 769–798.
Held, I.M. and Soden, B.J. 2006. Robust responses of the
hydrological cycle to global warming. Journal of Climate
19: 5686–5699.
Manabe, S., Smagorinsky, J., Holloway Jr., J.L., and Stone,
H. 1970. Simulated climatology of a general circulation
model with a hydrologic cycle. III. Effects of increased
horizontal computational resolution. Monthly Weather
Review 98: 175–212.
Manabe, S. and Stouffer, R.J. 1979. A CO2-climate
sensitivity study with a mathematical model of the global
climate. Nature 282: 491–493.
Manabe, S. 1985. Carbon dioxide and climate. Advances in
Geophysics 25: 39–82.
Richter, I. and Xie, S.-P. 2008. Muted precipitation Figure 1.1.5.1. Climate sensitivity estimates from new research
increase in global warming simulations: A surface published since 2010 (colored, compared with the range given
evaporation perspective. Journal of Geophysical Research, in the Intergovernmental Panel on Climate Change (IPCC)
Atmospheres 113: D24118, doi:10.1029/2008JD010561. Fourth Assessment Report (AR4) (gray) and the IPCC Fifth
Assessment Report (AR5; black). The arrows indicate the 5 to
Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., 95% confidence bounds for each estimate along with the best
Averyt, K.B., Tignor, M., and Miller, H.L. (eds.), 2007. estimate (median of each probability density function; or the
Contribution of Working Group I to the Fourth Assessment mean of multiple estimates; colored vertical line). Ring et al.
Report of the Intergovernmental Panel on Climate Change. (2012) present four estimates of the climate sensitivity and the
Cambridge University Press, Cambridge, United Kingdom, red box encompasses those estimates. The right-hand side of the
and New York, NY, USA. IPCC AR4 range is dotted to indicate that the IPCC does not
actually state the value for the upper 95% confidence bound of
their estimate and the left-hand arrow only extends to the 10%
1.1.5 Climate Sensitivity lower bound as the 5% lower bound is not given. The light grey
The “sensitivity” of temperature to carbon dioxide, vertical bar is the mean of the 16 estimates from the new
which is the amount of total warming for a nominal findings. The mean climate sensitivity (3.4°C) of the climate
doubling of atmospheric carbon dioxide, is the core models used in the IPCC AR5 is 13 percent greater than the
parameter that ultimately drives global warming IPCC’s “best estimate” of 3.0°C and 70% greater than the mean
policy, and its magnitude has been the subject of a of recent estimates (2.0°C).
vigorous debate between scientists. The current draft
of the United Nations’ Intergovernment-al Panel on NASA Senior Scientist David Rind of the
Climate Change AR5 report gives a mean model Goddard Institute for Space Studies subtitled a recent
sensitivity of 3.4°C, and calculations from standard paper, “We still can’t predict future climate responses
deviation tables given in the draft yield a 90% (5– at low and high latitudes, which constrains our ability
95%) range of 2.1–4.7°C. As can be seen in Figure to forecast changes in atmospheric dynamics and
1.1.5.1, these figures are quite high in comparison to a regional climate” (Rind, 2008). Rind began his review
number of prominent recent studies detailed below. and analysis of this important subject by noting
Charney et al. (1979) concluded global temperature
sensitivity to a doubling of the atmosphere’s CO2
26
concentration was “between 1.5° and 4.5°C,” while “substantial global warming impact” from a doubling
noting since then “we have not moved very far from or more of the air’s CO2 content.
that range.” In addition, Rind reported the uncertainty Lindzen and Choi (2009), two Massachusetts
in our assessment of high- and low-latitude climate Institute of Technology scientists, used the National
sensitivity “is also still as great as ever, with a factor Centers for Environmental Prediction’s 16-year
of 2 at both high and low latitudes.” (1985–1999) monthly record of sea surface
Rind identifiend several problems contributing to temperature (SST), together with corresponding
the uncertainty. Whether the water vapor response to radiation data from the Earth Radiation Budget
warming employed by climate models “is realistic is Experiment, to estimate the sign and magnitude of
hard to assess,” he noted, “because we have not had climate feedback over the oceanic portion of the
recent climate changes of the magnitude forecast for tropics and thus obtain an empirical evaluation of
the rest of this century” to test against. Closely Earth’s thermal sensitivity, as opposed to the model-
associated are low-latitude difficulties related to based evaluation employed by the IPCC.
modeling both low- and high-level clouds in the The scientists found all 11 models employed in
tropics and the physics and dynamics associated with the IPCC’s analysis “agree as to positive feedback,”
them, plus high-latitude difficulties associated with but they all disagree—and disagree “very sharply”—
cryosphere feedbacks related to sea ice and snow with the real-world observations Lindzen and Choi
cover. utilized, implying negative feedback actually prevails.
One way of dealing with these uncertainties has Moreover, the presence of that negative feedback
been to suggest, in Rind’s words, that “we can have reduced the CO2-induced propensity for warming to
greater confidence in the multi-model mean changes the extent that their analysis of the real-world
than in that of any individual model for climate observational data yielded only a mean SST increase
change assessments.” However, he writes, “it is “of ~0.5°C for a doubling of CO2.”
doubtful that averaging different formulations Lindzen and Choi (2009) were criticized for not
together will end up giving the ‘right’ result,” because using available tropical high-altitude radiation data
“model responses (e.g., tropical land precipitation) and for constraining feedback to within the tropics,
can often be of different signs, and there can be little which is highly unrealistic for a number of reasons,
confidence that averaging them together will produce including Hadley Cell-westerly interactions and the
a better result.” export of large amounts of energy from the tropics
Rind thus concludes, “at this point, we cannot into the temperate zone via tropical cyclones and
determine the low- and high-latitude sensitivities, and poleward currents. They also did not explicitly use the
we have no real way of obtaining them.” These IPCC-defined climate sensitivity.
unknowns, in his opinion, “affect the confidence we However, Lindzen and Choi (2011) addressed
can have in many of our projections of atmospheric these concerns by adding data from NASA’s Cloud
dynamic and hydrologic responses to global and Earth Radiant Energy System (CERES), rather
warming.” than simply using the older Earth Radiation Budget
Because of these and a host of other complexities Experiment (ERBE) data. They also shared radiation
he discusses, Rind states, “forecasting even the large- feedback fluxes with the extratropics and addressed
scale response to climate change is not easy given the the sensitivity issues. The derived mean sensitivity
current uncertainties” and “regional responses may be was 0.7°, with a 90% range of 0.6–1.0°C, indicating
the end result of varying influences in part due to virtually no positive amplification of warming beyond
warming in different tropical and high-latitude that from CO2.
regions.” Another empirically based analysis of climate
Rind concludes “real progress will be the result of sensitivity was published several years earlier in the
continued and newer observations along with review paper of Idso (1998), who described eight
modeling improvements based on these “natural experiments” he personally employed in
observations,” which observations must provide the prior studies designed to determine “how Earth’s
basis for evaluating all model implications. So near-surface air temperature responds to surface
difficult is this task, however, that he says “there is no radiative perturbations.” The eight naturally occurring
guarantee that these issues will be resolved before a phenomena employed by Idso were (1) the change in
substantial global warming impact is upon us.” There the air’s water vapor content that occurs at Phoenix,
is, of course, also no guarantee there even will be any Arizona, with the advent of the summer monsoon, (2)
27
the naturally occurring vertical redistribution of dust Investigating this subject further, a paper by
that occurs at Phoenix between summer and winter, Spencer and Braswell (2011) explored the sensitivity
(3) the annual cycle of surface air temperature caused of the surface temperature response to a forced
by the annual cycle of solar radiation absorption at radiative imbalance. They used observed shortwave
Earth’s surface, (4) the warming effect of the entire and longwave radiation gathered from satellite
atmosphere caused by its mean flux of thermal measurements and calculated the net atmospheric
radiation to the surface of Earth, (5) the annually radiation. They also used observed surface
averaged equator-to-pole air temperature gradient temperatures from the years 2000–2010 around the
sustained by the annually averaged equator-to-pole globe and calculated global monthly temperature
gradient of total surface-absorbed radiant energy, (6) anomalies relative to the average over the ten-year
the mean surface temperatures of Earth, Mars, and period. Using these two time series, they correlated
Venus relative to the amounts of CO2 contained in one versus the other using different time lags and
their respective atmospheres, (7) the paradox of the compared these observed values to the same variables
faint early Sun and its implications for Earth’s gathered from the twentieth century runs of the World
thermal history, and (8) the greenhouse effect of Climate Research Programme (WCRP) Coupled
water vapor over the tropical oceans and its impact on Model Intercomparison Project (CMIP) phase 3
sea surface temperatures. multi-model data set. The authors chose the three
These eight analyses, Idso writes, suggest “a 300 most- and least-sensitive model runs, rather than
to 600 ppm doubling of the atmosphere’s CO2 using all of them. They also used the observed and
concentration could raise the planet’s mean surface modeled data to investigate the mathematical
air temperature by only about 0.4°C,” in line with relationship for temperature change described above.
Lindzen and Choi’s deduced warming of ~0.5°C for a When Spencer and Braswell set a “feedback”
nominal doubling of the air’s CO2 content. There parameter as described above, they demonstrated that
would thus appear to be strong real-world data that only for pure non-radiative forcing and with no time
argue against the order of magnitude larger CO2 lag can the parameter be accurately diagnosed in the
sensitivity predicted by state-of-the-art climate model (see Figure 1.1.5.1). With radiative forcing and
models. a 70/30% mix of radiative versus non-radiative
Feedbacks are a natural part of the complexity of forcing, the response was quite different: “in this
our climate system. They represent nonlinear case, radiative gain precedes, and radiative loss
processes within a system that are the result of the follows a temperature maximum, as would be
interaction between two or more variables or the expected based upon conservation of energy
interaction of a variable with itself (or its changes— considerations.” They also found, more importantly,
“self” interaction) and influence overall climate that the pure radiative forcing curve in Figure 1.1.5.1
sensitivity. Generally, these processes can only be looks more like one produced using the data from the
parameterized, or represented in the models in an climate models, while the mixed curve looks more
empirical way. Because of the inability to precisely like that produced using the observed data.
represent feedbacks in a model, the model output may Spencer and Braswell then point out, “we are still
not be reasonable. faced with a rather large discrepancy in the time-
Temperature change in the climate system can be lagged regression coefficients between the radiative
represented simply as a “forced-dissipative” signatures displayed in the real climate system in
relationship, for example by non-radiative and satellite data versus the climate models.” Such
radiative processes as well as a net radiative restoring discrepancy indicates the climate system possesses
force, which is the feedback. As an example of this less sensitivity that the climate models project. That
feedback process, if the system warms radiatively or is, climate models may overestimate the temperature
non-radiatively, then the net radiative force may change forced by a certain process (e.g., increasing
become negative (longwave-out increases and atmospheric CO2).
becomes larger than shortwave-in). Thus, the net Further, it should be noted Earth’s climate is a
radiative force will act to counter the warming, and a complicated dynamic system in which each part has
larger restoring force would represent a less-sensitive characteristic time- and space-scales over which it
climate. The opposite can be argued for the system evolves, or as it reacts to a forcing that is external to
cooling. These principles can be represented by a that part of system or to the system as a whole. The
simple differential or mathematical equation. atmosphere, being less massive than the ocean or the
28
3.2°C, and a 90% range of from 1.4–2.8°C. What is

rather remarkable in this paper is how tightly
constrained the PDF becomes when “real world”
(interglacial/glacial) data are used, as the 66% range
is not much smaller than the 90%, at 1.7–2.6°C, a net
range change of only 0.5°C.
Aldrin et al. (2011) used a rather simple energy
balance model to determine hemispheric surface
temperature and global ocean heat content as a
function of the estimated changes in radiative forcing.
They derived a 90% range of 1.2–3.5°C, with a p=0.5
value of approximately 1.75°. Again using a single-
average Earth temperature as the predictand, van
Hatteren (2012) calculated an equilibrium sensitivity
90% confidence range of 1.7–2.3°C. Notably, the fit
to the observed temperature rise in the past two
centuries was significantly worse (and more warming
was predicted) when van Hatteren’s solar variability
Figure 1.1.5.1. The lag regression coefficients between parameter was reduced by an order of magnitude.
temperature (K) and radiative flux (W m-2 K-1) for the case of Ring et al. (2012) did not calculate a sensitivity
(a) non radiative forcing (dotted line), (b) pure radiative forcing
range, but rather estimated equilibrium warming with
(dashed line), and (c) a 70/30% mixture of radiative and non-
radiative forcing. Adapted from Spencer and Braswell (2011) a spectral decomposition model of the four main
their Figure 4. global temperature data sets, fit to long-lived
greenhouse gases, compensating aerosols, changes in
tropospheric ozone and land use, solar irradiance, and
cryosphere, tends to respond more rapidly than the volcanic activity in a model estimating land and
other parts of the system. Thus, when the climate of ocean temperatures, with a 40-layer oceanic model to
the atmosphere by itself is modeled, we consider the allow for latitudinal advection of heat. Their
atmosphere a boundary value problem; the equilibrium temperature change ranged from 1.5 to
atmosphere will be a servant to the underlying 2.0°C, and they noted
surface. It is often stated that the oceans, with their
greater mass and higher thermal capacity, are the …. These are on the low end of the estimates in
inertial and thermal flywheels of the climate system. the IPCC’s Fourth Assessment Report.[1] So,
A number of recent papers have eliminated the while we find that most of the observed warming
feared “fat tail” of a reasonably finite probability of is due to human emissions of [long-lived
very large warming and reduced or median estimates. greenhouse gases], future warming based on these
estimations will grow more slowly than that under
Annan and Hargreaves (2011) adjusted Bayesian
the IPCC’s “likely” range of climate sensitivity,
probability estimates of the range of sensitivity based from 2.0°C to 4.5°.
upon analysis of their sensitivity to the choice of prior
and recent ERBE data. They determined a sensitivity Notably, the fourth author of this paper,
of 1.2–3.6°C, with a most likely value of 2.2°, which University of Illinois climate modeler Michael
also eliminated the long “long fat tail” they note is Schlesinger, has been one of the most outspoken
“characteristic of all recent estimates of climate advocates of stringent and immediate controls of
sensitivity.” A Cauchy-derived probability gave 1.3– greenhouse emissions and, in earlier work, he had
4.2°C, with a quite low 1.9° at p=0.5. produced some of the largest estimates of equilibrium
Also using a Bayesian approach, Schmittner et al. warming from carbon dioxide.
(2011) created an empirical-dynamic combination of Using both posterior (regression) and prior
land and ocean temperature reconstructions during the (Bayesian) approaches, Hargreaves et al. (2012)
last glacial maximum combined with climate model calculated the sensitivity with models that either
simulations. Both eliminated the “fat tail,”with a included or excluded atmospheric dust, where these
probability density function (PDF) showing a approaches were coupled to seven common General
“vanishing” probability below 1.0°C and above
29
Circulation Models (GCMs). The dust-free Bayesian and then a gradual increase. The gradual increases
model had the greatest sensitivity, with a range of (1% per year) resulted in a doubling and quadrupling
1.0–4.2°, with a central estimate of 2.4°C. The dust- of CO2 in 70 and 140 years, respectively.
included regression model yielded the smallest Additionally, the authors initially used a shorter (10
warming, a range of 0.8–3.4°C, with a central years) and long-scale (100 years) CO2 forcing as well
estimate of 2.0°C. The Bayesian dust-included model as sensitivity values within the range of those given in
had a similar central estimate. Perhaps more the published literature for their a priori estimates.
important, Hargreaves et al. noted their collection of Then they solved the equations backward to get their
models yielded “a high probability of [equilibrium estimates of the short- and long-time-scale modes and
warming] lying below 4°C.” sensitivity.
Again coupling a Bayesian approach to a model, The results from Olivé et al. (2012) show the
this time the MIT two-dimensional model, Lewis short-time-scale temperature forcing for CO2 was on
(2013) generated a 90% range of 1.0–3.0°C, with a the order of three to four years, while it was 100 to
central value of 1.6°C. Masters (2013) is the only 300 years on the long-time-scale. The shorter mode of
recent sensitivity analysis based upon observed values temperature change was faster for the model
of oceanic heat uptake, and he found a 90% simulations where the CO2 forcing was sudden
sensitivity range of 1.2–5.1°, with a central value (Figure 1.1.5.2) rather than the gradual CO2 increase
again of 2.0°C, substantially below what is in the simulations. The uncertainty also was lower for the
suite of models employed in the existing draft of the sudden increase simulations. For the longer time-scale
IPCC’s Fifth Assessment Report. forcing, longer model integrations would be needed to
Scaling the relationship between observed estimate these and reduce the uncertainty. Concerning
temperature in the widely cited HadCRUT4 the model sensitivities, the authors write, “when
temperature history with calculated changes in the assuming two modes it varies between 0.49 and 0.83
total heat content of the Earth system, Otto et al. K W-1 m2 or between 0.56 and 1.01 K W-1 m2,” both
(2013) state their most reliable calculation yields a within the published range shown early in the paper.
range of 1.2–3.9°C, centered on 2.0°C. The analogous This kind of diagnostic work is an important use
range for the upcoming IPCC report is 2.2–4.7°C, of long-term computer model simulations. Even
centered on 3.4°C. This paper was authored by17 though there is no predictive work being done in this
very prominent climate scientists. study, it is remarkable that there is such variation in
Many of the senstitivity values are very similar to the estimate of the short- and long-term radiative
the twenty-first century warming Michaels et al. forcing as well as in the climate sensitivity. This is
(2002) had projected some ten years earlier based especially true between the two models used. Such
largely on observational data, which may be the first findings demonstrate the models are less than perfect,
of the low-sensitivity data-based papers in the as any of these internal differences would have to be
literature. the result of the model physics employed, especially
In total, there are 42 researchers describing 19 that pertaining to radiative forcing. Also, such long
separate experiments in this section. integrations should lead to the build-up of numerical
In another paper, Olivé et al. (2012) examined the errors. Lastly, the fact that the authors could not
changes in global temperature relative to changes in appreciably reduce the uncertainty in estimating these
CO2 concentration using two different models. This quantities indicates further improvement is not likely
study also endeavored to quantify the uncertainties in using the current technology.
the authors’ scenarios. The authors used output from
the Hadley Centre’s Atmosphere-Ocean (UKMO- References
HadCM3) and the CNRM-CM3 Centre National de
Recherches Meteorologiques Coupled Model, version Aldrin, M., et al. 2012. Bayesian estimation of climate
3 (CNRM-CM3) coupled Atmosphere Ocean General sensitivity based on a simple climate model fitted to
Circulation Models (AOGCMs). The simulations observations of hemispheric temperature and global ocean
were relatively short, on the time scale of 100 to 300 heat content. Environmetrics, doi: 10.1002/env.2140.
years, since the model spatial resolution was higher. Annan, J.D. and Hargreaves, J.D. 2011. On the generation
The authors also used CO2 scenarios where the and interpretation of probabilistic estimates of climate
increase was simulated to be sudden, corresponding sensitivity. Climatic Change 104, 324–436.
to a 6.5 times increase (10 W/m2), a sudden doubling,
30
Lindzen, R.S. and Choi, Y.-S. 2011. On the observational

determination of climate sensitivity and its implications.
Asia-Pacific Journal of Atmospheric Science 47: 377–390.
Masters, T. 2013. Observational estmates of climate
sensitivity from changes in the rate of ocean heat uptake
and comparison to CMIP5 models. Climate Dynamics,
doi:101007/s00382-013-1770-4.
Michaels, P.J., Knappenberger, P.C., Frauenfeld, O.W.,
and Davis, R.E. 2002. Revised 21st Century temperature
projections. Climate Research 22: 1–9.
Olivé, D.J.L., Peters, G.P., and Saint-Martin, D. 2012.
Atmosphere Response Time Scales Estimated from
AOGCM Experiments. Journal of Climate 25: 7956–7972.
Otto, A., Otto, F.E.L., Boucher, O., Church, J., Hegerl, G.,
Forster, P.M., Gillett, N.P., Gregory, J., Johnson, G.C.,
Knutti, R., Lewis, N., Lohmann, U., Marotzke, J., Myhre,
G., Shindell, D., Stevens, B., and Allen, M.R. 2013.
Energy budget constraints on climate response. Nature
Geoscience 6, 415–416.
Rind, D. 2008. The consequences of not knowing low- and
high-latitude climate sensitivity. Bulletin of the American
Figure 1.1.5.2. A time series of global- and annual-mean surface
Meteorological Society 89: 855–864.
air temperature obtained with a one(red line) and two (blue line)
mode approach for (top) UKMO-HadCM3 and (bottom) CNRM- Ring, M.J., et al., 2012. Causes of the global warming
CM3. The model data are represented by the symbols, and the observed since the 19th century. Atmospheric and Climate
modal results obtained using a priori (a posteriori) parameter Sciences 2: 401–415. doi: 10.4236/acs.2012.24035.
values are represented by the dashed (solid) lines. C20 and Cx2
represent scenarios with increased atmospheric CO2 in a stepwise Schmittner, A., et al. 2011. Climate sensitivity estimated
fashion (Cx2 is slower). The C2 scenario represents a sudden from temperature reconstructions of the Last Glacial
doubling of CO2. Adapted from Figure 1 of Olivé et al. (2012). Maximum. Science 334: 1385–1388. doi: 10.1126/science.
1203513.
Charney, J.G., Arakawa, A., Baker, D.J., Bolin, B., Spencer, R.W. and Braswell, W.D. 2011. On the
Dickinson, R.E., Goody, R.M., Leith, C.E., Stommel, misdiagnosis of surface temperature feedbacks from
H.M., and Wunsch, C.I. 1979. Carbon Dioxide and variations in Earth’s radiant energy balance. Remote
Climate: A Scientific Assessment. National Academy of Sensing 3: 1603–1613.
Sciences. Washington, DC (USA).
van Hateren, J.H. 2012. A fractal climate response function
Hargreaves, J.C., et al. 2012. Can the Last Glacial can simulate global average temperature trends of the
Maximum constrain climate sensitivity? Geophysical modern era and the past millennium. Climate Dynamics,
Research Letters 39: L24702, doi: 10.1029/ doi: 10.1007/s00382-012-1375-3.
2012GL053872.
Idso, S.B. 1998. CO2-induced global warming: a skeptic’s
view of potential climate change. Climate Research 10: 1.1.6 Climate Response and Projection
69–82. Climate change can occur due to internal variations in
Lewis, N. 2013. An objective Bayesian, improved the system or internal forcing, the interaction among
approach for applying optimal fingerprint techniques to properties or parts of the climate system with each
estimate climate sensitivity. Journal of Climate, doi: other. Climate also can be forced externally,
10.1175/JCLI-D-12-00473.1. influenced by phenomena considered to be outside the
climate system, such as volcanic action, solar forcing,
Lindzen, R.S. and Choi, Y.-S. 2009. On the determination
or an “artificial” disturbance such as that imposed by
of climate feedbacks from ERBE data. Geophysical
Research Letters 36: 10.1029/2009GL039628. human-emitted greenhouse gases.
The IPCC used GCMs to make two claims about
31
anthropogenic climate warming. The first is that most IPCC’s assumption that human forcing is the
or all of the warming occurring during the latter part dominant player in climate change. Even a further
of the twentieth century was human-induced. In its warming (of about 0.6°C or less) by 2100, as
Fourth Assessment Report (AR4), the IPCC asserts predicted by the IPCC, does not necessarily imply an
anthropogenic forcing dominates natural forcing. Yet anthropogenic origin.
even in AR4 (Chapter 2), the IPCC admits low
confidence in its knowledge of the size of solar
forcing since 1750. By contrast, the
Nongovernmental International Panel on Climate
Change (NIPCC, 2009) has demonstrated a strong
correlation between, for example, solar variations and
global temperature (see also Chapter 3 of this
volume). The second claim made by the IPCC in
AR4 is that by the end of the twenty-first century, a
warming of 1.0 °C to 3.6 °C or more will have
occurred due to human activity.
The IPCC’s assertion of the dominance of
anthropogenic forcing is merely that: an assertion.
Decades ago, Leith (1975) used the fluctuation-
dissipation (F-D) theorem from statistical mechanics Figure 1.1.6.1. The estimate of global temperature change for
the past 2,000 years (2 KY) in degrees centigrade (adapted from
to study “the way in which a climate mean would
Dr. R. Spencer, NASA).
return to its original value after it had been artificially
perturbed.” In his short paper he proved the theorem
mathematically and then applied it to the problem of
References
using climate models to create climate change
scenarios. In his proof of the F-D theorem, Leith Leith, C.E. 1975. Climate response and fluctuation
demonstrated that as long as the conditions for the dissipation. Journal of the Atmospheric Sciences 32: 2022–
theorem hold, the mean dissipation of a fluctuation 2026.
(or disturbance) is that projected by a linear
regression (statistical) model. It would not matter Idso, C. and Singer, S.F. 2009. Climate Change
whether the disturbance was artificial or natural, just Reconsidered: 2009 Report of the Nongovernmental Panel
that they be small and of similar size. Leith on Climate Change (NIPCC), The Heartland Institute,
Chicago, IL.
acknowledged the character of the climate system
may not be such that the F-D theorem holds exactly,
but studies in turbulence show it is a reasonable 1.1.7 Regional Projection
approximation for its behavior. Some observers of the climate change debate,
One implication is that it is difficult to ascribe scientists and non-scientists alike, have criticized the
temperature change to one process if two or more efforts to model future climate by asking how
processes causing the temperature change are of projections 100 years into the future can be made
similar magnitude. Moreover, if it is assumed that an when a weather model can’t get this weekend’s
artificial forcing is much greater than a natural weather right. Modeling is a complicated process, and
forcing, as does the IPCC, then the GCMs will weather forecasting and climate projection are two
generate climate responses much larger than those of very different problems that require different
natural variability. Lastly, the F-D theorem implies strategies.
the use of complex GCMs will not necessarily project Weather forecasting can be done dynamically, by
future conditions any better than a statistical model. (a) taking the mathematics and physics that represent
The fact that observed global temperature the atmosphere, (b) initializing with a clean set of
increases in the instrumental record since the late measurements, and (c) running a computer program
1800s do not show rates of change sufficiently larger, forward, generating a forecast that in theory
or even different from, those implied in the proxies represents the atmosphere at a certain prediction time.
for the last one to two millennia (Figure 1.1.6.1; see However, as noted earlier, such predictions can be
also Chapter 4) does not bode positively for the calculated for only about 10 to 14 days into the future.
32
Climate projection (as opposed to weather for all regions and seasons. The authors acknowledge
prediction) relies on the assumption that the statistics “accurate simulation of precipitation in all seasons
generated after lengthy simulations (years or decades) and regions remains a challenge for all of the tested
would adequately represent the climate of some future models.” Nonetheless, this initial test proved
period. The focus of climate projections is to replicate successful to the satisfaction of the authors, justifying
the energy flows within the climate system such that more testing and release of the model to the public.
any accumulation or decrease represents real changes The CWRF’s performance was not universally
due either to internal redistribution of energy or superior (see Figure 1.1.7.1). For example, used with
imbalances in the radiation exchange to space. different physics packages to project the great
The basis for confidence in the anthropogenic Mississippi River Basin flooding of 1993, the CWRF
global warming hypothesis is that changes in demonstrated limited skill. However, for seasonal
atmospheric carbon dioxide concentration introduce a rainfall rates, there was enhanced skill when the
bias in the exchange of radiation energy with space. projections were averaged. While this modeling
As the boundary energy exchanges slowly change, system represents an improvement over others in
there is a concomitant movement of internal indices, some respects and is another tool for modelers to use,
such as global surface temperature, that can be climate modeling remains an inexact science far from
modeled. Such projections require additional physics being able to show us the climate of some distant
and tighter constraints on the magnitudes of internal future.
energy exchanges than those utilized in weather
forecasting models. For example, changing aerosol Reference
forcing from natural or human sources would have a
negligible effect on the prediction of weekend Liang, X.Z., Xu, M., Yuan, X., Ling, T., Choi, H.I., Zhang,
weather, but over the course of many years can have a F., Chen, L., Liu, S., Su, S., Qiao, F., He, Y., Wang, J.X.L.,
large impact on regional or global climate. Kunkel, K.E., Gao, W., Joseph, E.E., Morris, V., Yu, T.-
In a recent paper, Liang et al. (2012) examined W., Dudhia, J., and Michalakes, J. 2012. Regional climate-
weather research and forecasting model. Bulletin of the
the feasibility of converting the Weather Research
American Meteorological Society 93: 1363–1380.
and Forecasting model (WRF) into a regional climate
model (CWRF). In doing so, they noted “there has
been some success using the WRF for regional 1.1.8 Seasonal Projection
climate downscaling,” adding that “such direct Climate models are tested and reworked with the goal
applications, however, also have encountered of developing ever-more-accurate representations of
numerous problems.” how the real world operates, so as to be able to make
In order to convert the WRF for climate confident projections of Earth’s future climate.
projections, the authors acknowledged the need for Kim et al. (2012) write, “the seasonal prediction
the use of cloud aerosol and radiation physics more skill for the Northern Hemisphere winter [was]
common to climate models. Also, atmospheric assessed using retrospective predictions (1982–2010)
“communication” with the oceans and land from the ECMWF System 4 (Sys4) and [the] National
(exchanges of heat, moisture, and momentum) needed Center for Environmental Prediction (NCEP) CFS
to be upgraded to deal with boundary value problems. version 2 (CFSv2) coupled atmosphere-ocean
Additionally, the horizontal and vertical resolutions seasonal climate prediction systems.” The analysis
used were the same as those for weather forecasting, revealed “for the Sys4, a cold bias is found across the
30 km in the horizontal and 36 vertical levels equatorial Pacific”; “the CFSv2 has [a] strong warm
(although weather forecasting can be done at finer bias from the cold tongue region of the Pacific to the
resolutions). The model focused on the United States equatorial central Pacific and [a] bias in broad areas
and adjacent regions, and the authors attempted to of the North Pacific and the North Atlantic.”
replicate the climate of the most recent decades and The researchers also found, “a cold bias over large
compared their results to observations and those regions of the Southern Hemisphere is a common
produced by another climate model. property of both reforecasts”; “with respect to
The CWRF generally improved the simulation of precipitation, the Sys4 produced excesses along the
the geographical distribution and seasonal/interannual Intertropical Convergence Zone, the equatorial Indian
variability of such quantities as precipitation, surface Ocean and the western Pacific”; “in the CFSv2, a
temperature, and downwelling radiation, although not strong wet bias is found along the South Pacific
33
Reference
Kim, H.-M., Webster, P.J., and Curry, J.A. 2012. Seasonal

prediction skill of ECMWF System 4 and NCEP CFSv2
retrospective forecast for the Northern Hemisphere winter.
Climate Dynamics 39: 2957–2973.
1.2 Modeling Techniques

Researchers are improving the ability of GCMs to
simulate the large-scale climate and its interannual
variability. However, the models still cannot
adequately simulate phenomena that arise from
smaller-scale processes such as precipitation, even in
areas where precipitation is quite frequent, such as the
East Asian Monsoon region. Efforts are continually
being made to improve the portrayal of regional
climates, model physics and parameterizations, and
even the representations of the equations themselves.
This section describes some of the studies exploring
the subject of model improvement.
1.2.1 Downscaling
One technique developed to improve model
performance is called downscaling, a process by
which a regional climate model is used over a limited
area but with a higher resolution, allowing smaller-
scale atmospheric processes to be represented more
faithfully, giving the regional climate scenario a finer
structure. Downscaling allows an examination of
model performance (validation) and a look at the
finer-scale details of the projections (e.g. Christensen
et al. 2007).
One limitation of this technique, however, is that
Figure 1.1.7.1. The 1982–2004 mean annual cycle of each of the lower-resolution errors and scale-
precipitation biases (mm day−1, left hand scale) simulated matching at the grid boundary travel to the interior of
by the models, and the observed precipitation (mm day−1, the finer-resolution portion of the grid and, to some
right hand scale) averaged over the four key regions. Also degree, confound the outcome. Given the short time
shown are the CWRF/ERI biases averaged during 1990– for these errors to propagate inwards and the long
2008. The zero line only pertains to the left axis. Adapted duration of climate simulations, this is an unresolved
from Fig 4 of Liang et al. (2012). problem that must be addressed.
Zou et al. (2010) sought to determine whether
downscaling can improve a GCM’s ability to replicate
Convergence Zone and the southern Indian Ocean, as the climatology of East Asian Monsoon precipitation.
well as in the western Pacific”; “a dry bias is found The authors used the LMDz4 model from the
for both modeling systems over South America and Laboratoire de Meteorologie Dynamique in France.
northern Australia and wet bias[es] in East Asia and This GCM can be run with a coarse grid (1.125
the equatorial Atlantic”; and “both models have degrees latitude and longitude, or roughly 125 km) or
difficulty in forecasting the North Atlantic Oscillation it can be run in “zoom” mode such that the regional
and the year-to-year winter temperature variability area has a grid resolution of about 50 km. The
over North America and northern Europe.” regional simulation has a grid resolution similar to
34
that of a weather forecast model.

Zou et al. ran the model for the East Asia Region
for the years 1958–2000 in order to compare the
modeled climate to monthly observations from this
region over the same time period. The raw output
revealed the model produced more (less) precipitation
when downscaling was (not) used. When the authors
extracted the larger-scale component of the
precipitation signal from the precipitation data
(Figure 1.2.1.1), they found the model results without
downscaling were unable to adequately represent the
monsoon pattern or its variability (compare Figure
1.2.1.1 frames a, b to frames e, f). The downscaling
results (Figure 1.2.1.1 frames c, d) compared more
favorably.
Even though the results were an improvement,
flaws remained. As Zou et al. point out, “it should be
acknowledged that our results of dynamic
downscaling are not perfect,” while adding “the main
weakness of the downscaling is the northward shift of
the monsoon rainbelt.” Although downscaling
improves some aspects of the performance of models
and their ability to represent observations, this
technique is still subject to the general limitations
suffered by all numerical models.
Indeed, numerous errors permeate numerical Figure 1.2.1.1. The strongest (left) and second strongest
models for both weather forecasts and climate (right) component of summer season precipitation shown
projections. These fall under three headings, (1) as a percentage from the rain gauge observations (top), the
observational error, (2) numerical error, and (3) GCM with downscaling (middle), and the GCM without
downscaling (bottom). Darker (lighter) colors represent
physical error.
more (less) precipitation than the actual measurements.
Observational error refers to the fact that Adapted from Figure 4 in Zou et al. (2010).
instrumentation cannot measure the state of the
atmosphere with infinite precision; it is important
both for establishing the initial conditions and
validation. Numerical error covers many accumulate energy at the smallest scales of resolution
shortcomings including “aliasing,” the tendency to or blow up rapidly due to computational error. In
misrepresent the sub-grid scale processes as larger- weather prediction, techniques have been developed
scale features. In the downscaling approach, to “smooth” and massage the data during a simulation
presumably errors in the large-scale boundary by removing as much grid-scale energy as possible
conditions also will propagate into the nested grid. while retaining the character of the larger scale.
Also, the numerical methods themselves are only Nonetheless, computational error does prevent
approximations to the solution of the mathematical models from forecasting perfectly, and this explains
equations, and this results in truncation error. why two different models (or even two different runs
Physical errors are manifest in parameterizations, with the same model) can give two different answers
which may be approximations, simplifications, or given the same or similar set of initial conditions.
educated guesses about how real processes work. An Otte et al. (2012) use a technique called
example of this type of error would be the “nudging” to improve the performance of a weather
representation of cloud formation and dissipation in a forecasting model (Weather Research and Forecasting
model, which is generally a crude approximation. model—WRF) for use in downscaling. Nudging is
Each of these error sources generates and used to keep the regional model results within the
propagates errors in model simulations. Without some bounds of the larger-scale model results.
“interference” from model designers, model solutions Generally applied to the boundaries of a model,
35
Figure 1.2.1.2. The annual area-averaged number of days with (a) precipitation greater than (a) 0.5 in, and (b) 1.0 in for
the Midwest region. Data are shown from North American Regional Reanalyses (O) and WRF runs for no nudging (N),
nudging toward analyses (A), and spectral nudging (S). Adapted from Figure 13 in Otte et al. (2012).
Otte et al. applied the nudging technique to the is meaningless if the GCM results that are being
interior of the grids. To test the procedure, they used a downscaled are themselves unrealistic.
coarse analysis data set from the National Centers for
Environmental Prediction to initialize the regional References
model. These analyses stood as a proxy for GCM
output. They then compared their results to two fine- Christensen, J.H., Hewitson, B., Busuioc, A., Chen, A.,
scale regional re-analyses for North America. Gao, X., Held, I., Jones, R., Kolli, R.K., Kwon, W.-T.,
This modeling technique generated a better Laprise, R., Magaña Rueda, V., Mearns, L., Menéndez,
representation of monthly means and improved C.G., Räisänen, J., Rinke, A., Sarr, A., and Whetton, P.
2007. Regional Climate Projections. In: Climate Change
interannual variability. The simulations of hot and
2007: The Physical Science Basis. Contribution of Working
cold extremes, as well as wet and dry extremes, were Group I to the Fourth Assessment Report of the
captured better in the nudged WRF. However, Otte et Intergovernmental Panel on Climate Change. Solomon, S.,
al. note, “all WRF runs overpredicted precipitation Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt,
totals through the multidecadal period … regardless K.B., Tignor, M., and H.L. Miller (eds.). Cambridge
of whether nudging was used” (see Figure 1.2.1.2). University Press, Cambridge, United Kingdom and New
Fewer false alarms of severe precipitation events were York, NY, USA.
produced with the nudging technique than when it
Otte, T.L., Nolte, C.G., Otte, M.J., and Bowden, J.H. 2012.
was not used, an important result given that most Does nudging squelch the extremes in regional climate
impacts on society result from the occurrence of modeling? Journal of Climate 25: 7046–7066.
extreme events. Reliable simulations that include the
realistic occurrence of extremes are valuable for Zou, L., Zhou, T., Li, L., and Zhang, J. 2010. East China
economic planners. summer rainfall variability of 1958–2000: Dynamical
In spite of the success of the nudging employed downscaling with a variable resolution AGCM. Journal of
Climate 23: 63946408.
by Otte et al. (2012), their results demonstrate
modeling today’s climate cannot be done either on the
largest scales or regionally without some way of 1.2.2 The Dynamic Core
“tuning” the model. This entails using parametric An atmospheric model, whether a forecast model or
means to force the model to adequately represent one used to diagnose which processes drive the
reality. Although downscaling can be a valuable tool climate, is built on a foundation of three main
for regional climate simulations, this newfound ability conservation principles: Within the Earth-atmosphere
36
system, mass, energy, and momentum must be Marshall and Molteni (1993) to show the evolution of
conserved. These basic physical principles are the wind fields. The model has a fairly coarse
represented by a closed set of equations termed the horizontal and vertical resolution by today’s
primitive equations. The equations form the “dynamic standards, but it has been shown to represent the
core” of climate models and represent the motions of largest scales of the atmosphere’s climate in a fairly
air within the model. The model output is generally realistic way. The experiment here examined the
some variable representing mass (pressure), tendencies of the 500 hPa mass field (streamfunction)
temperature, or both. due to linear and nonlinear processes. The nonlinear
The primitive equations are too complex in their processes were divided into those that are resolvable
raw form to be represented adequately in models and by the model and those that are of smaller scale than
are often simplified following the concept of Occam’s can be represented by the model.
Razor: They are simplified as much as possible, and The model output was slightly different
no more. One way these equations have been depending on which processes were included in the
simplified is to “linearize” them, thereby eliminating model run. A comparison of the full nonlinear
nonlinearity that can make a computer model processes or interactions (Figure 1.2.2.1a) gave
unstable. This nonlinearity is a process by which the slightly different results than when comparing only
atmosphere behaves in a chaotic fashion. It is also the nonlinear resolvable processes (Figure 1.2.2.1b). The
reason computational errors propagate and amplify. model output also differed from that of two similar
The falsely generated energy interacts across all studies because of the strategies used to calculate
scales of motion to distort the path of the model these model “climates.” The authors conclude the
evolution such that eventually the scales of motion of article by stating “the manner of defining interactions
interest (weather systems) have only limited forecast between the resolved and unresolved modes plays a
relationship to that of the atmosphere they are crucial role in the dynamical interpretation of the
attempting to reproduce. This is the predictability tendency-based statistical diagnostics.”
barrier. The basic lesson to be learned from the authors’
Kondrashov et al. (2011) used a simplified three- closing statement is that there can be output
level atmospheric circulation model designed by differences produced within the same model core
Figure 1.2.2.1. The mean tendencies (model output) in the phase space (mathematical space, not physical), for (a)
tendencies due to the full nonlinear processes,( b) nonlinear processes that are resolvable, (c) the linear inverse model
output, and (d) the linear part of the nonlinear tendencies. Adapted from Kondrashov et al. (2011).
37
even when using the same input data, depending on –WV). They limited their domain to the Eastern
how the nonlinear processes are represented. If the Pacific from the International Dateline to South
representation of climate can be different depending America, over the period 1995–2008.
on how these core processes are represented, then In the model, the presence of the ITCZ was
producing model scenarios on the time scale of a inferred based on the most likely location prior to
century must be done very cautiously and analysis. Then a point was labeled as ITCZ or non-
interpretation of the results done carefully. ITCZ by testing the value of the satellite data typical
for the ITCZ, as well as the status of the surrounding
References grid points in both time and space. A second pass of
the dataset blended the a priori information with the
Kondrashov, D., Kravtsov, S., and Ghil, M. 2011. satellite data and the likelihood that a particular point
Signatures of nonlinear dynamics in an idealized was ITCZ or non-ITCZ. The method was then
atmospheric model. Journal of the Atmospheric Sciences evaluated against manual observation and
68: 3–12. identification of the ITCZ (see Figure 1.2.3.1.1).
Marshall, J. and Molteni, F. 1993. Toward a dynamical
understanding of atmospheric weather regimes. Journal of
the Atmospheric Sciences 50: 1972–1818.
1.2.3 Statistical Models
1.2.3.1 Statistical Validation

The Intertropical Convergence Zone (ITCZ) is an
important feature in the atmosphere’s general
circulation. The ITCZ is a belt of low pressure near
the equator where the northeast and southeast trade
winds from the Northern and Southern Hemispheres
converge. The ITCZ is also characterized by light
winds and intermittent regions of convective activity.
In some parts of the globe it is easy to identify and in
other parts it is more difficult, as it is ill-defined.
This feature is important to the climate system
because convection in the ITCZ is the vehicle by
which the excess heat and momentum of the tropics
begins the journey poleward so as to keep the
energetics of the climate system in balance. The
ITCZ, which migrates northward in the tropics from
January to July (and back southward in July to
January), is also an important contributor to the
occurrence of the monsoons and often serves as the Figure 1.2.3.1.1. The ITCZ cloudiness for 2100 UTC 19 August
producer of seed disturbances that become tropical 2000 overlaid on the IR data for (a) the manual detection, (b) the
cyclones. Thus it would be important for any study of statistical model from the IR, VIS, and WV satellite data, and
climate to be able to identify and understand the (c) the threshold at 270 Kelvin without post-processing. The
dynamics of this feature. gray-scale bar below represents the temperatures in the image,
Bain et al. (2011) used satellite data to develop a and the bold contour in (a) and (b) the position of the ITCZ.
modeling tool that can be used to objectively and Adapted from Figure 3 in Bain et al. (2011).
reliably locate the ITCZ. They used images derived
from reflected light (visible—VIS) and radiated
energy (infrared images—IR) from Earth. They also The Bain et al. study revealed the statistical
used images generated by the intensity of microwave methodology tested very well in objectively
emission and measuring the return (water vapor depth identifying the ITCZ and was subsequently used to
38
examine the climatological characteristics of the identifying cyclic behavior in a time series.
ITCZ. The seasonal evolution of the ITCZ as Previous research from each author and their
determined by the methodology showed it migrated colleagues had found there are cyclic and quasi-cyclic
with the seasons and its passage followed the seasonal forcing processes that can be extracted from local and
location for the regions of warm sea surface global time series of temperature. Such cycles can be
temperatures (SSTs). In order to look at the related to solar and astronomical activity (for example
climatological trends and interannual variations, Bain 11, 22, 50–80 years; see Chapter 3 of this volume), or
et al. extended the satellite data set back to 1980 for terrestrial oscillations such as the Pacific Decadal
the IR data only. In doing so they found there was Oscillation (about 60 years). Their earlier research
enhanced (less) ITCZ convection in El Niño (La pointed to the presence and dominance of 20- and 60-
Niña) years, a finding reached in previous studies. year cycles in temperature time series. Their model
However, they found “no climatic shifts in the also identified a linear trend that may be related to
position of the ITCZ could be detected,” which solar and volcanic activity since the middle of the
suggests, they write, “any trends in the ITCZ location Little Ice Age. Additionally, they included an
over the 30 years (that are not due to ENSO) are quite anthropogenic trend emerging in the early 1940s
small or non-existent.” (Figure 1.2.3.2.1). These four processes combined to
Thus, statistical modeling can be used to produce a model that fit the global temperature series
objectively identify, in more detail, important features from 1850 to 2010 very well.
of the general circulation that we previously thought Loehle and Scafetta determined that more than 50
were well-known. The successful design of a model percent and as much as 60 percent of the climate
to objectively locate the ITCZ will be useful in the signal from 1850 to 2010 was the product of natural
future study of its dynamics as well as its interaction forcing. They projected future climate to the year
with other tropical features such as the monsoons or 2100 using the model, stating, “the result is a
the Madden Julian Oscillation (MJO), which continued warming with oscillations to a high in 2100
manifests as a 30- to 60-day oscillation in tropical of about 0.6°C above 2000 values.” This resulted
convection. Not all models are dynamic, and not all from the combined effect of natural and
models are used to project the future climate. But anthropogenic forcing.
using this particular model, Bain et al. showed no The 0.6°C temperature rise projected by the
significant change in the character of the ITCZ in the Loehle and Scafetta model is much less than the low-
past 30 years over the tropical Eastern Pacific. end estimates of 2.3°C projected by the GCMs relied
upon by the IPCC and is consistent with recent
Reference estimates of low climate sensitivity. Also, the Loehle
and Scafetta model shows the latest downturn in
Bain, C.L., J. DePaz, J. Kramer, G. Magnusdottir, P. temperatures since 2000. The IPCC (2007) models do
Smyth, H. Stern, and C.-C. Wang, 2011. Detecting the not show this downturn and assume 90 percent or
ITCZ in instantaneous satellite data using spatiotemporal more of the climate change since 1970 is
statistical modeling: ITCZ climatology in the East Pacific. anthropogenic. As Loehle and Scafetta point out, “we
Journal of Climate 24: 216–230.
have shown that the effect of natural oscillations is
critical for proper assessment of anthropogenic
1.2.3.2 Empirical Models impacts on the climate system.” Perhaps this is why
Models can be analytical, statistical, or empirically Loehle and Scafetta’s model fits the historic
derived and are then used to examine trends or cycles temperature data much better than do most of the
in a time series in an effort to determine the causes of IPCC models, and with far fewer parameters.
change or variability. Several studies have shown GCMs are improving
Loehle and Scafetta (2011) built a one- in their ability to reproduce the current climate,
dimensional model of global temperatures from 1850 including intraseasonal and interannual variations.
to 2010 using the inductive approach and used this The ability of the GCMs to capture internal nonlinear
model to project the change in climate from the processes, however, will always be suspect.
current time to the year 2100. The authors constructed But when examining climate, important external
their model using what they called “decomposition forcing processes need to be accounted for, including
analysis,” based on a method of cycles and used in such natural processes as volcanic activity and solar
variations. There is increasing evidence that
39
empirical model can be used to project climate into

the twenty-first century.
Scafetta also demonstrated the IPCC GCMs
cannot capture decadal and interdecadal variability.
As shown by the author (see Figure 1.2.3.2.2),
“although these GCM simulations present some kind
of red-noise variability supposed to simulate the
multi-annual, decadal, and multidecadal natural
variability, a simple visual comparison among the
simulations and the temperature record gives a clear
impression that the simulated variability has nothing
to do with the observed temperature dynamics.”
Scafetta then demonstrates natural variability is
not solely the result of internal variations; the external
forcing described above also plays a role. These
external forcings modulate solar output, which in turn
impacts electrical activity in the upper atmosphere.
This influences incoming cosmic ray fluxes, which
have been linked to variability in cloudiness. The
external cycles have periods similar to internal
climate variations. Natural variations then, likely
account for more than half the climate variability
since 1850.
Finally, Scafetta also demonstrated the IPCC is
erroneous in ascribing nearly all of the twentieth and
twenty-first century climate change to anthropogenic
forcing. When the anthropogenic effect is corrected, it
accounts for less than half the recent climate change.
Scafetta also showed the empirical model in his paper
Figure 1.2.3.2.1. (Top) A full reconstruction of the global
projected the cooling of the most recent decade,
temperature anomaly (°C) record of 1850 -2010 using the model
described by Loehle and Scafetta (2011). (Bottom) The whereas the IPCC GCMs produced a quasi-
individual components of the model, where the solid line is the monotonic warming of the climate from about the
long term trend and the dashed line is the anthropogenic year 2000 on. Scafetta’s model produces a warming
component. Adapted from Figure 3 in Loehle and Scafetta of only 0.8–1.5o C by the end of the twenty-first
(2011). century.
Scientists skeptical of the IPCC’s projections of
anthropogenic climate change have cautioned the
extraterrestrial forcing processes can have an impact public about the shortcomings of the IPCCs reliance
on solar variations and, thus, terrestrial climate. on GCMs in producing climate change scenarios for
Recent work by Scafetta (2011) explores several the next century. In addition to well-documented
such issues. First, the author expanded on his own problems with model numerics, the lack of data, and
work demonstrating a link between well-known chaos, the physics of the models fall short. This is
climate cycles of roughly 10, 20, and 60 years to true not only for such internal processes as cloud
celestial cycles. The celestial cycles result from the physics, but also for such external forcing as Sun-
solar-lunar tidal oscillation (9.1 years) and moon tidal forcing and other solar system
gravitational cycles related to the interaction between gravitational cycles that influence solar output.
the sun and the largest planets (Jupiter and Saturn). Scafetta’s work demonstrates there is increasing
These celestial cycles have periods of about 10, 20, evidence our solar system plays a significant role in
and 60 years. Loehle and Scafetta (2011) decadal and multidecadal climate variations. The
demonstrated these cycles can be added together and, climate projections produced by Scafetta’s empirical
with a realistic anthropogenic effect, reconstruct the harmonic model may be far more realistic and are
global climate record since 1950. They also show this certainly more optimistic.
40
At the time, meteorology was interested in

explaining why the mid-latitude flow (jet stream)
would transition from a higher amplitude state to a
more zonal state in an irregular fashion at a time scale
of roughly 10 to 14 days. There was also some
interest in being able to make monthly and seasonal
forecasts, which during the late 1950s to 1960s were
considered very long range. During this time,
numerical weather prediction was also in its infancy.
Lorenz (1963) was expressly interested in
examining the behavior of periodic and non-periodic
flows. Later Lorenz would describe the non-periodic
behavior as “chaos,” which he then termed “order
without periodicity.”
He used a technique called low order modeling to
Figure 1.2.3.2.2. The global surface temperature taken from
http://www.cru.uea.ac.uk/cru/data/temperature/ (black) and the
study the behavior of convection in a geophysical
GISS ModelE average simulation (blue), and a fit using an fluid. To do so he coupled the Equation of Motion
empirical harmonic model (red). Adapted from Scafetta (2011) and the First Law of Thermodynamics and then
Figure 1. represented the motion and temperature variables in
terms of waves, which were then characterized by the
lowest wave numbers that reasonably represented
these fields. He was able to demonstrate the solutions
References for the system in mathematical space and thus follow
the evolution of the system in time. The resultant
IPCC. 2007. Climate Change 2007 -I: The Science of graph is commonly known today as Lorenz’s butterfly
Basis, Contributions of Working Group I to the Fourth (Figure 1.2.4.1). He realized this model provided a
Assessment Report of the Intergovernmental Panel on
good analogue for the behavior of mid-latitude flow.
Climate Change. Edited by: Solomon, S., Qin, D.,
Manning, M., Marquis, M., Averyt, K, Tignor, M.M.B., In theory, then, if we know the precise initial
Miller, H.L. Jr., and Chen, Z.. Cambridge University Press, conditions of the system such that we lay on one of
Cambridge, UK. the trajectories in the “butterfly,” we could follow that
Loehle, C. and Scafetta, N. 2011. Climate change
attribution using empirical decomposition of climatic data.
The Open Atmospheric Science Journal 5: 74–86.
Scafetta, N. 2011. Testing an astronomically based
decadal-scale empirical harmonic climate model versus the
IPCC (2007) general circulation models. Journal of
Atmospheric and Solar-Terrestrial Physics 80: 124–137.
1.2.4 Low Order Models

In meteorology it is well-known that the limit of
dynamic predictability is approximately 10 to 14
days. This constraint is imposed by the composition
of our atmosphere and the size of the planet.
However, there has been some disagreement in the
climate community over whether there is a fair
amount of predictability in climate or whether it is too Figure 1.2.4.1. A representation of Lorenz’s butterfly
chaotic for meaningful prediction. It is appropriate developed using the model published in Lorenz (1963).
here to revisit a classic work written by E.N. Lorenz This picture was adapted from an Internet image.
nearly 50 years ago on the subject of predictability in trajectory forever and know what the state of the
atmospheric flows. weather would be into the future infinitely.
41
Computers could then make weather or climate Model runs of climate change scenarios exhibit this
forecasts, and meteorologists would be out of jobs. latter behavior. In these projections, the spread in the
However, even today’s best models are forecasts gets larger and larger as the projection time
comprised only of hypotheses about how scientists horizon lengthens. This is exactly the behavior
believe the atmosphere really works. As the past has expected from models attempting to simulate
made abundantly clear, state-of-the-art models of nonlinear flow—the energy of eddies below the grid
“today” quickly outlive their usefulness and are scale appears chaotically as eddies on scales of the
eventually replaced by the newer and improved motion resolved by the grid.
models of “tomorrow.” Today’s best representation of Thus there is ample evidence the climate behaves
the physics is inadequate and there are also numerical in a chaotic way and climate prediction is simply not
model errors, in that differential quantities can only possible; only the development of scenarios is
be estimated. possible. Although these scenarios might be
But the crux of predictability rests in the fact that statistically similar to the real climate (as opposed to
the initial conditions are not precisely known. There accurate in detail), this observation remains to be
is always a degree of uncertainty within them, and demonstrated and cannot be assumed.
thus it is not known on which trajectory the model
rests. According to Lorenz, “when our results Reference
concerning the instability of non-periodic flow are
applied to the atmosphere, which is ostensibly Lorenz, E.N. 1963. Deterministic nonperiodic flow.
nonperiodic, they indicate that prediction of the Journal of Climate 20: 130–139.
sufficiently distant future is impossible by any
method, unless the present conditions are known
1.2.5 Bias Correction
exactly. In view of the inevitable inaccuracy and
In a Hydrology and Earth System Sciences opinion
incompleteness of weather observations, precise very-
article, Ehret et al. (2012) write, “despite considerable
long-range forecasting would seem to be non-
progress in recent years, output of both global and
existent.” The problem of not being able to specify
regional circulation models is still afflicted with
the precise initial state—sensitive dependence on
biases to a degree that precludes its direct use,
initial conditions (SDIC)—is an integral characteristic
especially in climate change impact studies,” noting
of chaotic systems.
“this is well known, and to overcome this problem,
Lorenz’s statement makes it plain that beyond 10
bias correction (BC, i.e., the correction of model
to 14 days, the typical timescale for the evolution of
output towards observations in a post-processing step)
the mid-latitude flow, dynamic predictability is
has now become a standard procedure in climate
impossible. This is true in spite of the fact we can
change impact studies.” For example, GCMs often
specify the equations representing motions in the
produce a global climate that is too cool or too warm
flow. Such limits on predictability are thus scale-
by several degrees compared to the real world.
dependent. For example, because of their small spatial
Ehret et al. present “a brief overview of state-of-
scales, we cannot dynamically predict the occurrence
the-art bias correction methods, discuss the related
of individual thunderstorms beyond the 30- to 120-
assumptions and implications, draw conclusions on
minute time-scale it takes for them to evolve. Beyond
the validity of bias correction and propose ways to
the time limits, corresponding to the appropriate
cope with biased output of circulation models.”
space-scale, only statistical prediction is useful.
In discussing the findings of their review, the
One way modelers attempt to mitigate SDIC is
authors state: (1) “BC methods often impair the
via the use of ensemble modeling. The technique is to
advantages of circulation models by altering
take the initial conditions and several plausible
spatiotemporal field consistency, relations among
alternative initial conditions (but very close by to the
variables and by violating conservation principles,”
original) and run the model several times using each
(2) “currently used BC methods largely neglect
set. If the trajectories all evolve toward a common
feedback mechanisms, (3) “it is unclear whether they
trajectory, the modelers claim there is high
are time-invariant under climate change conditions,”
confidence that the predicted solution is highly
(4) “applying BC increases agreement of climate
probable. But when these trajectories diverge (another
model output with observations in hindcasts and
characteristic of chaotic systems), the probability that
hence narrows the uncertainty range of simulations
any one of them will be the actual prediction is small.
42
and predictions,” but (5) this is often done “without remains so today), noting “we could easily have
providing a satisfactory physical justification,” and uncertainties of many tens of W m-2 in our description
this sleight of hand “is in most cases not transparent of the radiative effect of such clouds, and how these
to the end user.” For example, the temperatures properties may change under climate forcing.”
produced by GCMs are presented as anomalies in the Potential errors of this magnitude are extremely
IPCC reports so their disagreement with each other disconcerting in light of the fact that the radiative
and with the real world are hidden. effect of a doubling of the air’s CO2 content is in the
Ehret et al. argue this set of negative lower single-digit range of Wm-2, and, to quote
consequences of bias correction “hides rather than Harries, “uncertainties as large as, or larger than, the
reduces uncertainty,” which they suggest may lead to doubled CO2 forcing could easily exist in our
avoidable forejudging of end users and decision modeling of future climate trends, due to uncertainties
makers. They conclude BC is often “not a valid in the feedback processes.”
procedure.” Furthermore, because of the vast complexity of
the subject, Harries declares, “even if [our]
Reference understanding were perfect, our ability to describe the
system sufficiently well in even the largest computer
Ehret, U., Zehe, E., Wulfmeyer, V., Warrach-Sagi, K., and models is a problem.”
Liebert, J. 2012. Should we apply bias correction to global Illustrative of a related problem is the work of
and regional climate model data? Hydrology and Earth Zender (1999), who characterized the spectral,
System Sciences 16: 3391–3404. vertical, regional, and seasonal atmospheric heating
caused by the oxygen collision pairs O2 . O2 and O2 .
1.3 Elements of Climate N2, which had earlier been discovered to absorb a
GCMs must incorporate all the many physical, small but significant fraction of the globally incident
chemical, and biological processes that influence solar radiation. This work revealed these molecular
climate over different spatial and temporal scales. collisions lead to the absorption of about 1 Wm-2 of
Although the models have evolved much in recent solar radiation, globally and annually averaged. This
years, limitations and deficiencies remain. discovery, in Zender’s words, “alters the long-
Many important processes are either missing or standing view that H2O, O3, O2, CO2 and NO2 are the
inadequately represented in today’s state-of-the-art only significant gaseous solar absorbers in Earth’s
climate models. Here we highlight many of the atmosphere,” and he suggests the phenomenon
insufficiencies researchers have found when “should therefore be included in ... large-scale
comparing model projections against real-world atmospheric models used to simulate climate and
observations, frequently using the researchers’ own climate change.” Zender’s work also raises the
words to report them. The list is long and varied, and possibility there are still other yet-to-be-discovered
it demonstrates much work remains to be done before processes that should be included in the models used
the model simulations can be treated with the level of to simulate Earth’s climate, and until we are confident
confidence ascribed to them by the IPCC. there is little likelihood of further such surprises, we
ought not rely too heavily on what the models of
1.3.1 Radiation today are telling us about the climate of tomorrow.
One of the most challenging and important problems In another revealing study, Wild (1999) compared
facing today’s general circulation models of the the observed amount of solar radiation absorbed in
atmosphere is how to accurately simulate the physics the atmosphere over equatorial Africa with what was
of Earth’s radiative energy balance. In commenting predicted by three general circulation models of the
on this task, Harries (2000) stated more than a decade atmosphere. Wild found the model predictions were
ago, “progress is excellent, on-going research is much too small. Regional and seasonal model
fascinating, but we have still a great deal to underestimation biases were as high as 30 Wm-2,
understand about the physics of climate.” He added, primarily because the models failed to properly
“we must exercise great caution over the true depth of account for spatial and temporal variations in
our understanding, and our ability to forecast future atmospheric aerosol concentrations. In addition, Wild
climate trends.” As an example, he points out our found the models likely underestimated the amount of
knowledge of high cirrus clouds is very poor (it solar radiation absorbed by water vapor and clouds.
Similar large model underestimations were
43
discovered by Wild and Ohmura (1999), who the observed climate record is to be reconciled with
analyzed a comprehensive observational data set model predictions and if estimates of future climate
consisting of solar radiation fluxes measured at 720 change are to be useful in formulating emission
sites across Earth’s surface and corresponding top-of- policies.”
the-atmosphere locations to assess the true amount of Pursuit of this goal, as Ghan et al. describe it,
solar radiation absorbed within the atmosphere. These requires achieving “profound reductions in the
results were compared with estimates of solar uncertainties of direct and indirect forcing by
radiation absorption derived from four atmospheric anthropogenic aerosols,” which is what they set out to
GCMs. They found “GCM atmospheres are generally do. This consisted of “a combination of process
too transparent for solar radiation,” as they produce a studies designed to improve understanding of the key
rather substantial mean error close to 20 percent processes involved in the forcing, closure experiments
below actual observations. designed to evaluate that understanding, and
Another solar-related deficiency of state-of-the- integrated models that treat all of the necessary
art GCMs is their failure to properly account for processes together and estimate the forcing.” At the
solar-driven variations in Earth-atmosphere processes conclusion of this laborious set of operations, Ghan et
that operate over a range of timescales extending al. arrived at numbers that considerably reduced the
from the 11-year solar cycle to century- and range of uncertainty in the “total global mean present-
millennial-scale cycles. Although the absolute solar day anthropogenic forcing,” but still implied a set of
flux variations associated with these phenomena are climate changes stretching from a small cooling to a
small, there are a number of “multiplier effects” that modest warming. Thus they provided a long list of
may significantly amplify their impacts. other things that must be done in order to obtain a
According to Chambers et al. (1999), most of the more definitive result, after which they acknowledged
many nonlinear responses to solar activity variability even this list “is hardly complete.” They conclude
are inadequately represented in the global climate their analysis by stating, “one could easily add the
models used by the IPCC to predict future greenhouse usual list of uncertainties in the representation of
gas-induced global warming. Other amplifier effects clouds, etc.” Consequently, they write, “much
are used to model past glacial/interglacial cycles and remains to be done before the estimates are reliable
even the hypothesized CO2-induced warming of the enough to base energy policy decisions upon.”
future, where CO2 is not the major cause of the Vogelmann et al. (2003) also studied the aerosol-
predicted temperature increase but rather an initial induced radiative forcing of climate, reporting,
perturber of the climate system that according to the “mineral aerosols have complex, highly varied optical
IPCC sets other, more-powerful forces in motion that properties that, for equal loadings, can cause
produce the bulk of the ultimate warming. So there differences in the surface IR flux between 7 and 25
appears to be a double standard within the climate Wm-2 (Sokolik et al., 1998),” but “only a few large-
modeling community that may best be described as an scale climate models currently consider aerosol IR
inherent reluctance to deal evenhandedly with effects (e.g., Tegen et al., 1996; Jacobson, 2001)
different aspects of climate change. When multiplier despite their potentially large forcing.” In an attempt
effects suit their purposes, they use them; but when to persuade climate modelers to rectify the situation,
they don’t suit their purposes, they don’t use them. Vogelmann et al. used high-resolution spectra to
In setting the stage for their study of climate calculate the surface IR radiative forcing created by
model inadequacies related to radiative forcing, Ghan aerosols encountered in the outflow of air from
et al. (2001) state, “present-day radiative forcing by northeastern Asia, based on measurements made by
anthropogenic greenhouse gases is estimated to be 2.1 the Marine-Atmospheric Emitted Radiance
to 2.8 Wm-2; the direct forcing by anthropogenic Interferometer aboard the NOAA Ship Ronald H.
aerosols is estimated to be -0.3 to -1.5 Wm-2, while Brown during the Aerosol Characterization
the indirect forcing by anthropogenic aerosols is Experiment-Asia. They determined “daytime surface
estimated to be 0 to -1.5 Wm-2,” so that “estimates of IR forcings are often a few Wm-2 and can reach
the total global mean present-day anthropogenic almost 10 Wm-2 for large aerosol loadings,” and these
forcing range from 3 Wm-2 to -1 Wm-2.” This implies values “are comparable to or larger than the 1 to 2
a climate change somewhere between a modest Wm-2 change in the globally averaged surface IR
warming and a slight cooling. They write, “the great forcing caused by greenhouse gas increases since pre-
uncertainty in the radiative forcing must be reduced if industrial times.” The researchers conclude their
44
results “highlight the importance of aerosol IR forcing cyclical seasonal change in tropical albedo. The
which should be included in climate model administrators of the test thus conclude “the missing
simulations.” variability in the models highlights the critical need to
Two papers published a year earlier (Chen et al., improve cloud modeling in the tropics so that
2002; Wielicki et al., 2002) revealed what Hartmann prediction of tropical climate on interannual and
(2002) called a pair of “tropical surprises.” The first decadal time scales can be improved.”
of the seminal discoveries was the common finding of Hartmann is considerably more candid in his
both groups of researchers that the amount of thermal scoring of the test, stating the results indicate “the
radiation emitted to space at the top of the tropical models are deficient.” Expanding on that assessment,
atmosphere increased by about 4 Wm-2 between the he further notes, “if the energy budget can vary
1980s and the 1990s. The second was that the amount substantially in the absence of obvious forcing,” as it
of reflected sunlight decreased by 1 to 2 Wm-2 over did over the past two decades, “then the climate of
the same period, with the net result that more total Earth has modes of variability that are not yet fully
radiant energy exited the tropics in the latter decade. understood and cannot yet be accurately represented
In addition, the measured thermal radiative energy in climate models.”
loss at the top of the tropical atmosphere was of the Also concentrating on the tropics, Bellon et al.
same magnitude as the thermal radiative energy gain (2003) note “observed tropical sea-surface
generally predicted to result from an instantaneous temperatures (SSTs) exhibit a maximum around
doubling of the air’s CO2 content. Yet as Hartman 30°C” and “this maximum appears to be robust on
noted, “only very small changes in average tropical various timescales, from intraseasonal to millennial.”
surface temperature were observed during this time.” They state, “identifying the stabilizing feedback(s)
The change in solar radiation reception was that help(s) maintain this threshold is essential in
driven by changes in cloud cover, which allowed order to understand how the tropical climate reacts to
more solar radiation to reach the surface of Earth’s an external perturbation,” which knowledge is needed
tropical region and warm it. These changes were for understanding how the global climate reacts to
produced by what Chen et al. determined to be “a perturbations such as those produced by solar
decadal-time-scale strengthening of the tropical variability and the ongoing rise in atmospheric CO2
Hadley and Walker circulations.” Another factor was levels. Pierrehumbert’s (1995) work confirms the
likely the past quarter-century’s slowdown in the importance of this matter, clearly demonstrating, in
meridional overturning circulation of the upper 100 to the words of Bellon et al., “that the tropical climate is
400 meters of the tropical Pacific Ocean (McPhaden not determined locally, but globally.” They also note
and Zhang, 2002); this circulation slowdown also Pierrehumbert’s work demonstrates interactions
promotes tropical sea surface warming by reducing between moist and dry regions are an essential part of
the rate of supply of relatively colder water to the tropical climate stability, harking back to the adaptive
region of equatorial upwelling. infrared iris concept of Lindzen et al. (2001).
What do these observations have to do with Noting previous box models of tropical climate
evaluating climate models? They provide several new have shown it to be sensitive to the relative areas of
phenomena for the models to replicate as a test of moist and dry regions of the tropics, Bellon et al.
their ability to properly represent the real world. analyzed various feedbacks associated with this
McPhaden and Zhang note the time-varying sensitivity in a four-box model of the tropical climate
meridional overturning circulation of the upper “to show how they modulate the response of the
Pacific Ocean provides “an important dynamical tropical temperature to a radiative perturbation.” In
constraint for model studies that attempt to simulate addition, they investigated the influence of the
recent observed decadal changes in the Pacific.” model’s surface-wind parameterization in an attempt
In an eye-opening application of this principle, to shed further light on the nature of the underlying
Wielicki et al. tested the ability of four state-of-the-art feedbacks that help define the global climate system
climate models and one weather assimilation model to responsible for the tropical climate observations of
reproduce the observed decadal changes in top-of-the- constrained maximum SSTs.
atmosphere thermal and solar radiative energy fluxes Bellon et al.’s work, as they describe it, “suggests
that occurred over the past two decades. No the presence of an important and as-yet-unexplored
significant decadal variability was exhibited by any of feedback in earth’s tropical climate, that could
the models and all failed to reproduce even the contribute to maintain the ‘lid’ on tropical SSTs,”
45
much like the adaptive infrared iris concept of for credible predictions of future Arctic sea ice.”
Lindzen et al. They also say the demonstrated Andronova et al. (2009) “used satellite-based
“dependence of the surface wind on the large-scale broadband radiation observations to construct a long-
circulation has an important effect on the sensitivity term continuous 1985–2005 record of the radiative
of the tropical system,” specifically stating “this budget components at the top of the atmosphere
dependence reduces significantly the SST sensitivity (TOA) for the tropical region (20°S-20°N).” They
to radiative perturbations by enhancing the then “derived the most conservative estimate of their
evaporation feedback,” which injects more heat into trends” and “compared the interannual variability of
the atmosphere and allows the atmospheric the net radiative fluxes at the top of the tropical
circulation to export more energy to the subtropical atmosphere with model simulations from the
free troposphere, where it can be radiated to space. Intergovernmental Panel on Climate Change Fourth
Clearly, therefore, the case is not closed on either Assessment Report (AR4) archive available up to
the source or the significance of the maximum 2000.” The three researchers report “the tropical
“allowable” SSTs of tropical regions; hence, neither system became both less reflective and more
is the case closed on the degree to which the planet absorbing at the TOA,” and “combined with a
may warm in response to continued increases in the reduction in total cloudiness (Norris, 2007), this
atmospheric concentrations of carbon dioxide and would mean that the tropical atmosphere had recently
other greenhouse gases. become more transparent to incoming solar radiation,
Eisenman et al. (2007) reported another problem which would allow more shortwave energy to reach
with model treatment of radiation. They used two Earth’s surface.” They also found “none of the
standard thermodynamic models of sea ice to models simulates the overall ‘net radiative heating’
calculate equilibrium Arctic ice thickness based on signature of the Earth’s radiative budget over the time
simulated Arctic cloud cover derived from 16 period from 1985–2000.”
different GCMs evaluated for the IPCC’s Fourth With respect to the first of their findings, and the
Assessment Report. Based on their analysis, they associated finding of Norris (2007), Andronova et al.
report there was a 40 Wm-2 spread among the 16 state these observations “are consistent with the
models in terms of their calculated downward observed near-surface temperature increase in recent
longwave radiation, for which both sea ice models years,” which provides an independent validation of
calculated an equilibrium ice thickness ranging from the TOA radiation measurements. With respect to
one to more than ten meters. They note the mean their second finding, the failure of all of the AR4
1980–1999 Arctic sea ice thickness simulated by the climate models to adequately simulate the TOA
16 GCMs ranged from only 1.0 to 3.9 meters, a far radiation measurements discredits the models. The
smaller inter-model spread. Hence, they say they were combination of these two conclusions suggests the
“forced to ask how the GCM simulations produce historical rise in the air’s CO2 content likely has
such similar present-day ice conditions in spite of the played a much lesser role in the post-Little Ice Age
differences in simulated downward longwave warming of the world than the IPCC has admitted.
radiative fluxes?” In another paper, Svensson and Karlsson (2012)
Answering their own question, the three use several GCMs of various horizontal and vertical
researchers state “a frequently used approach” to resolutions to examine the climate of the Arctic
resolving this problem “is to tune the parameters defined as the region north of the Arctic Circle (66.6o
associated with the ice surface albedo” to get a more N). They were concerned with modelling the winter
realistic answer. “In other words,” they continue, months during the period 1980–1999, a time of
“errors in parameter values are being introduced to limited observational data sets. The observations used
the GCM sea ice components to compensate were the European Centre for Long Range
simulation errors in the atmospheric components.” Forecasting (ECMWF) ERA reanalyses. The authors
The three researchers conclude “the thinning of Arctic write, “one should be cautious to interpret the data as
sea ice over the past half-century can be explained by ‘truth’ in this remote region. However, the abundance
minuscule changes of the radiative forcing that cannot of observations at lower latitudes [in the Arctic] and
be detected by current observing systems and require sea ice extent should at least constrain the properties
only exceedingly small adjustments of the model- of the air masses that enter and exit the Arctic.” The
generated radiation fields,” and, therefore, “the results observational data in this region were augmented with
of current GCMs cannot be relied upon at face value satellite observations.
46
Svensson and Karlsson found the winter sea ice

cover was similar among the models and accorded
with observations across much of the Arctic Ocean.
However, there were some differences at the margins,
and in the North Atlantic the overall result is that
most of the models produce too much sea ice. In
validating other quantities, it was found the models
tend to underestimate the longwave energy being
radiated into space, some by as much as 10 percent.
In terms of wintertime mean cloudiness, the models
generate winter season values between 35 and 95
percent, whereas observations show values ranging
from 68 percent to 82 percent, a smaller range. Near
the surface, the latent and sensible heat fluxes within
the Arctic were consistent with observations.
The vertical profiles (Figure 1.3.1.1) show the
models were generally cooler in the lower
troposphere and a little more humid. This led to wide
differences between models in the characteristics of
air masses. Also, all models showed stronger
gradients in temperature and humidity than were
observed in the lower troposphere, which resulted in
lower clear-sky modelled radiation than for
observations. One possible explanation was that many
of the models were less “active” in terms of synoptic
weather patterns in this region. From these findings,
the authors infer humidity was the most important
contributor to the radiation budgets in the Arctic
region. This leads to the conclusion that it is
important to know the models are simulating
temperature and moisture profiles correctly in this
region of the world.
Finally, Regional Climate Models (RCMs) are
often used to simulate the climate of more limited
spatial regions, especially if the focus is on
phenomena driven by smaller-scale processes or even
microphysical processes. Regional climate models are
similar to regional forecast models in much the same
way that their general circulation (GCM) counterparts
are similar to global forecast models. RCMs suffer
from the same deficiencies as all models, including
insufficient data and resolution, model physics, and
the numerical methods used. Figure 1.3.1.1. The median vertical profiles for
A study by Zubler et al. (2011) attempted to temperature (K), specific humidity (g kg−1), and relative
demonstrate changes in aerosol emissions over humidity over (a)–(c) entire Arctic, (d)–(f) open ocean, and
(g)–(i) sea ice from the GCMs and ERA-Interim over the
Europe lead to changes in the radiation budgets over
southernmost latitude band 66.6°–70°N. Adapted from
the region using the COSMO-CLM model (Doms and Svensson and Karlsson (2012) their Figure 8.
Schättler, 2002) RCM. The horizontal and vertical
resolution in the RCM was finer than that in a GCM.
The RCM was coupled with the aerosol model used and anthropogenic aerosols. The authors performed
by the European Centre for Medium Range Fore- two computer runs, one with climatologically
casting, Hamburg GCM and included both natural averaged aerosols and the other with aerosol
47
emissions that change with time (transient). On the would reflect the configuration of the storm track.
boundaries, the RCM was driven by the European Lastly, the authors found (Figure 1.3.1.2) “the use of
Centre re-analyses (ERA-40). transient emissions in TRANS does not improve the
Zubler et al. found under clear sky conditions temperature trends simulated with CLIM. In line with
there was a dimming over Europe due to transient the change in cloud fraction and thus all-sky SSR
Figure 1.3.1.2. The standardized anomalies for five year running averages for various regions in Europe (see paper) for the
all-sky downward surface shortwave radiation (solid black), cloud fraction multiplied by -1 (dashed black), ERA-40
(gray), transient aerosols (red), and climatological aerosols (blue). Adapted from Zubler et al. (2011) Figure 5.
emissions from the late 1950s to the late 1970s, then a (surface shortwave radiation), both RCM simulations
brightening over several locales. In comparison with show a similar dimming/brightening signal in
the ERA-40 re-analyses, they infer the RCM has temperature.” This points to the dominance of natural
underestimated the real trends. The authors also variations in driving the surface temperature changes
conclude processes occurring beyond the model within the European Region. And as stated above, the
domain were responsible for these changes, as RCM still over- and underestimated certain quantities.
inferred by the difference between the transient and But as the authors note, some processes may not be so
climatological runs. important that their actual representation is critical:
Zubler et al. also note the cloud fraction increased “due to the dominating dependence of all-sky SSR on
during the latter part of the twentieth century, but did clouds, our study implies that it may not be necessary
so more strongly in the RCM. This would impact the to use transient emissions in order to simulate
total sky brightening by countering the clear sky dimming/brightening in Europe with a RCM.
aerosol increases. However, it was then shown there The studies reviewed here suggest general
was a strong correlation between all sky changes and circulation models of the atmosphere are seriously
the North Atlantic Oscillation (NAO). The NAO inadequate in the way they treat several aspects of
48
Earth’s radiative energy balance—and fail entirely to (Eds.) Climate Variability and Extremes During the Past
address some pertinent phenomena. 100 Years. Springer, New York, NY, USA, pp. 169–178.
Pierrehumbert, R.T. 1995. Thermostats, radiator fins, and
References the local runaway greenhouse. Journal of the Atmospheric
Sciences 52: 1784–1806.
Andronova, N., Penner, J.E., and Wong, T. 2009. Observed
and modeled evolution of the tropical mean radiation Sokolik, I.N., Toon, O.B., and Bergstrom, R.W. 1998.
budget at the top of the atmosphere since 1985. Journal of Modeling the radiative characteristics of airborne mineral
Geophysical Research 114: 10.1029/2008JD011560. aerosols at infrared wavelengths. Journal of Geophysical
Research 103: 8813–8826.
Bellon, G., Le Treut, H., and Ghil, M. 2003. Large-scale
and evaporation-wind feedbacks in a box model of the Svensson, G. and J. Karlsson, 2012: On the arctic
tropical climate. Geophysical Research Letters 30: wintertime climate in global climate models. Journal of
10.1029/2003GL017895. Climate 24: 5757-5771. DOI:10.
Chambers, F.M., Ogle, M.I., and Blackford, J.J. 1999. Tegen, I., Lacis, A.A., and Fung, I. 1996. The influence on
Palaeoenvironmental evidence for solar forcing of climate forcing of mineral aerosols from disturbed soils.
Holocene climate: linkages to solar science. Progress in Nature 380: 419–422.
Physical Geography 23: 181–204.
Vogelmann, A.M., Flatau, P.J., Szczodrak, M., Markowicz,
Chen, J., Carlson, B.E., and Del Genio, A.D. 2002. K.M., and Minnett, P.J. 2003. Observations of large aerosol
Evidence for strengthening of the tropical general infrared forcing at the surface. Geophysical Research
circulation in the 1990s. Science 295: 838–841. Letters 30: 10.1029/2002GL016829.
Doms, G. and Schättler, U. 2002. A description of the Wielicki, B.A., Wong, T., Allan, R.P., Slingo, A., Kiehl,
nonhydrostatic regional model LM, Part I: Dynamics and J.T., Soden, B.J., Gordon, C.T., Miller, A.J., Yang, S.-K.,
numerics, technical report, Dtsch. Wetterdienst, Offenbach Randall, D.A., Robertson, F., Susskind, J., and Jacobowitz,
am Main, Germany. H. 2002. Evidence for large decadal variability in the
tropical mean radiative energy budget. Science 295: 841–
Eisenman, I., Untersteiner, N., and Wettlaufer, J.S. 2007. 844.
On the reliability of simulated Arctic sea ice in global
climate models. Geophysical Research Letters 34: Wild, M. 1999. Discrepancies between model-calculated
10.1029/2007GL029914. and observed shortwave atmospheric absorption in areas
with high aerosol loadings. Journal of Geophysical
Ghan, S.J., Easter, R.C., Chapman, E.G., Abdul-Razzak, Research 104: 27,361–27,371.
H., Zhang, Y., Leung, L.R., Laulainen, N.S., Saylor, R.D.,
and Zaveri, R.A. 2001. A physically based estimate of Wild, M. and Ohmura, A. 1999. The role of clouds and the
radiative forcing by anthropogenic sulfate aerosol. Journal cloud-free atmosphere in the problem of underestimated
of Geophysical Research 106: 5279–5293. absorption of solar radiation in GCM atmospheres. Physics
and Chemistry of the Earth 24B: 261–268.
Harries, J.E. 2000. Physics of the earth’s radiative energy
balance. Contemporary Physics 41: 309–322. Zender, C.S. 1999. Global climatology of abundance and
solar absorption of oxygen collision complexes. Journal of
Hartmann, D.L. 2002. Tropical surprises. Science 295: Geophysical Research 104: 24,471–24,484.
811–812.
Zubler, E.M, Folini, D., Lohmann, U., Lüthi, D., Schär, C.,
Jacobson, M.Z. 2001. Global direct radiative forcing due to and M. Wild, 2011: Simulation of dimming and
multicomponent anthropogenic and natural aerosols. brightening in Europe from 1958 to 2001 using a regional
Journal of Geophysical Research 106: 1551–1568. climate model. Journal of Geophysical Research 116:
D18205, doi:10.1029/2010JD015396.
Lindzen, R.S., Chou, M.-D., and Hou, A.Y. 2001. Does the
earth have an adaptive infrared iris? Bulletin of the
American Meteorological Society 82: 417–432. 1.3.2 Water Vapor
McPhaden, M.J. and Zhang, D. 2002. Slowdown of the “Water vapor feedback in climate models is large and
meridional overturning circulation in the upper Pacific positive,” note Paltridge et al. (2009), and “the
Ocean. Nature 415: 603–608. various model representations and parameterizations
of convection, turbulent transfer, and deposition of
Norris, J.R. 2007. Observed interdecadal changes in
latent heat generally maintain a more-or-less constant
cloudiness: Real or spurious? In: Broennimann, S. et al.
49
relative humidity (i.e., an increasing specific humidity References

q) at all levels in the troposphere as the planet
warms.” This “increasing q amplifies the response of Boehmer, S. 2012. Science debates must continue. Energy
surface temperature to increasing CO2 by a factor or 2 and the Environment 23: 1483–1487.
or more,” they write. They also note the behavior of Paltridge, G., Arking, A., and Pook, M. 2009. Trends in
water vapor in the middle and upper levels of the middle- and upper-level tropospheric humidity from NCEP
troposphere dominates overall water-vapor feedback. reanalysis data. Theoretical and Applied Climatology 98:
It is at these levels where long-term measurements of 351–359.
the trends in water vapor concentration are least
reliable.
Consequently, knowledge of how q (particularly 1.3.3 Aerosols
the q of the upper levels of the troposphere) responds
to atmospheric warming is of paramount importance 1.3.3.1 Aerosols
to the task of correctly predicting the water vapor Aerosols, whether natural or anthropogenic, can
feedback and how air temperatures respond to affect the weather and climate in many ways. The
increasing CO2 concentrations. Paltridge et al. explore most obvious is by increasing the planetary albedo,
this important subject by determining trends in which leads to a surface cooling. Aerosols also can
relative and specific humidity at various levels in the warm the middle and upper troposphere if they absorb
atmosphere based on reanalysis data of the National sunlight, possibly increasing atmospheric stability and
Centers for Environmental Prediction (NCEP) for the inhibiting cloud formation. They can have indirect
period 1973–2007. The three researchers report “the effects as well, by changing the nature of clouds and
face-value 35-year trend in zonal-average annual- the formation of precipitation. And like clouds, they
average specific humidity q is significantly negative can only be parameterized in weather and climate
at all altitudes above 850 hPa (roughly the top of the models. The inadequate treatment of aerosols by
convective boundary layer) in the tropics and GCMs represents a major limitation in the models’
southern midlatitudes and at altitudes above 600 hPa reliability.
in the northern midlatitudes.” They conclude Mishchenko et al. (2009) state “because of the
“negative trends in q as found in the NCEP data global nature of aerosol climate forcings, satellite
would imply that long-term water vapor feedback is observations have been and will be an indispensable
negative—that it would reduce rather than amplify the source of information about aerosol characteristics for
response of the climate system to external forcing use in various assessments of climate and climate
such as that from increasing atmospheric CO2.” change,” and they note “there have been parallel
As discussed by Boehmer (2012), there is a claims of unprecedented accuracy of aerosol
continuing argument on this topic that boils down to retrievals with the moderate-resolution imaging
two main issues. First, there is the question as to spectroradiometer (MODIS) and multi-angle imaging
whether in situ balloon measurements, which suggest spectroradiometer (MISR).”
a negative long-term trend in upper-level q and hence If both aerosol retrieval systems are as good as
a negative water vapor feedback, or remote sensing they have been claimed to be, they should agree on a
satellite measurements, which suggest the opposite, pixel by pixel basis as well as globally. Consequently,
are correct in their measurements of q. The second and noting “both instruments have been flown for
question concerns whether short-term correlations many years on the same Terra platform, which
between upper-level q and surface temperature can be provides a unique opportunity to compare fully
extrapolated to deduce the existence of such a collocated pixel-level MODIS and MISR aerosol
correlation over longer time scales. retrievals directly,” Mishchenko et al. decided to see
Boehmer’s work is critical, as the assumption that how they compare in this regard by analyzing eight
humidity will remain constant and act as an amplifier years of such data.
is based largely on models, which is circular The six scientists from NASA’s Goddard Institute
reasoning. If water vapor is not an amplifier but acts for Space Studies report finding what they describe as
as a negative feedback, the case for high climate “unexpected significant disagreements at the pixel
sensitivity and alarming rates of warming is based level as well as between long-term and spatially
entirely on models and not real-world data. averaged aerosol properties.” In fact, they note, “the
only point on which both datasets seem to fully agree
50
is that there may have been a weak increasing “the uncertainty due to combinations of these
tendency in the globally averaged aerosol optical parameters can reach more than 1 W/m2.” As for the
thickness (AOT) over the land and no long-term AOT significance of their findings, they write, “these
tendency over the oceans.” The NASA scientists state numbers should be compared with the sulfate aerosol
their conclusion quite succinctly: “[O]ur new results forcing of -1.9 W/m2 for the year 2000, obtained
suggest that the current knowledge of the global using the default values of the parameters.”
distribution of the AOT and, especially, aerosol The mean sulfate aerosol forcing component of
microphysical characteristics remains unsatisfactory.” Earth’s top-of-the-atmosphere radiative budget is thus
And since this knowledge is indispensable “for use in apparently not known to within anything better than ±
various assessments of climate and climate change,” 50 percent. In addition, Haerter et al. (2009) note
it would appear current assessments of greenhouse- structural uncertainties, such as “uncertainties in
gas forcing of climate made by the best models in use aerosol sources, representation of aerosols in models,
today may be of very little worth in describing the parameterizations that relate aerosols and cloud
real world of nature. droplets to simulate the indirect aerosol effect, and in
In a contemporaneous study, Haerter et al. (2009) cloud schemes” lead to an overall uncertainty in F of
note future projections of climate “have been—for a approximately ± 43 percent, as noted by IPCC. In
given climate model—derived using a ‘standard’ set reality, therefore, the current atmosphere’s aerosol
of cloud parameters that produce realistic present-day radiative forcing is probably not known to anything
climate.” However, they add, “there may exist better than ± 100%, which does not engender
another set of parameters that produces a similar confidence in the ability to simulate Earth’s climate
present-day climate but is more appropriate for the very far into the future.
description of climate change” and, “due to the high In another study, Booth et al. (2012) note “a
sensitivity of aerosol forcing (F) to cloud parameters, number of studies have provided evidence that
the climate projection with this set of parameters aerosols can influence long-term changes in sea
could be notably different from that obtained from the surface temperatures,” citing Mann and Emanuel
standard set of parameters, even though the present- (2006) and Evan et al. (2009), but “climate models
day climate is reproduced adequately.” This state of have so far failed to reproduce these interactions,”
affairs suggests replication of the present-day climate citing Knight (2009) and Ting et al. (2009). They
is no assurance that a climate model will accurately consequently note, as they phrase it, “the role of
portray Earth’s climate at some future time. It is also aerosols in decadal variability remains unclear.”
noted this study did not examine first principles, but Booth et al. used the Hadley Centre Global
rather the treatment of radiational forcing, which must Environmental Model version 2 (HadGEM2-ES)—a
be parameterized. next-generation Climate Model Intercomparison
To get a better idea of the magnitude of Project phase 5 (CMIP5) model—to determine
uncertainty associated with this conundrum, Haerter whether older CMIP3 models “contained the
et al. used the ECHAM5 atmospheric general complexity necessary to represent a forced Atlantic
circulation model, which includes parameterizations Multidecadal Oscillation.” The five researchers were
of direct and first indirect aerosol effects, to thus able to demonstrate that “aerosol emissions and
determine what degree of variability in F results from periods of volcanic activity explain 76% of the
reasonable uncertainties associated with seven simulated multidecadal variance in detrended 1860–
different cloud parameters: the entrainment rate (the 2005 North Atlantic sea surface temperatures,” and
rate at which environmental air and cloud air mix) for “after 1950, simulated variability is within
shallow convection, the entrainment rate for observational estimates,” while their estimates for
penetrative convection, the cloud mass flux above the 1910–1940 “capture twice the warming of previous
non-buoyancy level, the correction to asymmetry generation models,” although they still “do not
parameter for ice clouds, the inhomogeneity explain the entire observed trend.” Put another way,
parameter for liquid clouds, the inhomogeneity they state that “mechanistically, we find that inclusion
parameter for ice clouds, and the conversion of aerosol-cloud microphysical effects, which were
efficiency from cloud water to precipitation. included in few previous multimodel ensembles,
Upon completion of their analyses, the four dominates the magnitude (80%) and the spatial
researchers report “the uncertainty due to a single one pattern of the total surface aerosol forcing in the
of these parameters can be as large as 0.5 W/m2” and North Atlantic.”
51
Booth et al. conclude their paper by noting, “one Mishchenko, M.I., Geogdzhayev, I.V., Liu, L., Lacis, A.A.,
of the reasons why the role of aerosols in driving Cairns, B., and Travis, L.D. 2009. Toward unified satellite
multidecadal variability has not previously been climatology of aerosol properties: What do fully
identified” is that “although all the CMIP3 models compatible MODIS and MISR aerosol pixels tell us?
Journal of Quantitative Spectroscopy & Radiative Transfer
represented the direct effect of aerosols on shortwave
110: 402–408.
radiation, most omitted or only partly represented the
indirect aerosol effects,” citing Chang et al. (2011). Ting, M., Kushnir, Y., Seager, R., and Li, C. 2009. Forced
They conclude, “we need to reassess the current and internal twentieth-century SST trends in the North
attribution to natural ocean variability of a number of Atlantic. Journal of Climate 22: 1469–1481.
prominent past climate impacts linked to North
Atlantic sea surface temperatures.”
Similarly, it may be that climatologists need to
1.3.3.2 Aerosol Nuclei
Many physical processes must be approximated in
reassess the attribution of the post-Little Ice Age
GCMs using separate calculations at the grid scale,
warming of the world to anthropogenic CO2
called parameterizations. In many cases this involves
emissions, to account for the possible warming effects
averaging the effect of subgrid scale processes. These
of still other as-yet-unappreciated phenomena that are
processes include the existence and formation of
either “omitted or only partly represented” in current
clouds and also relate to the presence of aerosols or
state-of-the-art climate models. Some of these
particulates in the air that are not considered to be
phenomena may be associated with things transpiring
part of the basic makeup the atmosphere but which
on (or within) the Sun; other phenomena that may
strongly affect radiation fluxes. Aerosols are
thwart or significantly reduce the warming effect of
important because they can affect energy exchanges
rising atmospheric CO2 concentrations may be
in the atmosphere and serve as condensation nuclei
associated with a variety of biological responses of
for cloud formation. Clouds have an impact on
both marine and terrestrial vegetation to atmospheric
Earth’s energy budget through their ability to reflect
CO2 enrichment, as well as to warming itself.
and scatter light and to absorb and emit infrared
radiation. Also, cloud properties such as droplet size
References
and concentration can influence how effectively
Booth, B.B.B., Dunstone, N.J., Halloran, P.R., Andrews, clouds contribute to the planet’s albedo.
T., and Bellouin, N. 2012. Aerosols implicated as a prime Roesler and Penner (2010) used a microsphysical
driver of twentieth-century North Atlantic climate model to explore the effects of the chemical
variability. Nature 484: 228–232. composition and size of aerosols on the concentration
of cloud droplets over the United States. Using
Chang, C.Y., Chiang, J.C.H., Wehner, M.F. Friedman, A., aerosol composition measurements from 1988 to
and Ruedy, R. 2011. Sulfate aerosol control of tropical
2004, the authors found aerosols influenced the size
Atlantic climate over the 20th century. Journal of Climate
24: 2540–2555. and concentration of the cloud droplets that ultimately
formed in their experiments.
Evan, A.T., Vimont, D.J., Heidinger, A.K., Kossin, J.P., They also included tests varying the strength of
and Bennartz, R. 2009. The role of aerosols in the the atmospheric vertical motions lifting air parcels,
evolution of tropical North Atlantic Ocean temperature which are initially saturated, over a distance of about
anomalies. Science 324: 778–781. 300 meters in order to induce cloud formation. They
Haerter, J.O., Roeckner, E., Tomassini, L., and von Storch, found that as vertical motion increased in strength
J.-S. 2009. Parametric uncertainty effects on aerosol within their model, the number of cloud droplets
radiative forcing. Geophysical Research Letters 36: increased. They also found that larger aerosols,
10.1029/2009GL039050. though fewer in number, were more soluble as they
formed cloud droplets. Smaller aerosols were more
Knight, J.R. 2009. The Atlantic Multidecadal Oscillation
inferred from the forced climate response in coupled numerous but less soluble. Thus, the larger aerosols
general circulation models. Journal of Climate 22: 1610– were found to be better at producing cloud droplets.
1625. The size of the aerosols depended on their chemical
composition, which varied by region across the
Mann, M.E. and Emanuel, K.A. 2006. Atlantic hurricane United States and by season. They found the
trends linked to climate change. EOS, Transactions, concentrations of droplets were largest in the eastern
American Geophysical Union 87: 233–244.
52
U.S. and in the spring season. The exact formulation of model physics and
Roesler and Penner’s work makes clear that in assumptions used for variables such as cloud droplet
order to model cloud forcing in a GCM, which numbers can have a large impact on the predicted
ultimately affects the ability of the model to capture droplet numbers. These in turn can have a relatively
climate or climate change, the chemical composition large impact on the net radiation budgets. Golaz et al.
of the condensation nuclei that form these clouds showed that, in spite of these differences, there was
must be properly accounted. Roesler and Penner only a small impact on the present-day climate
pointed out, “a global model using an empirical overall. However, when the three formulations were
relationship based on regional measurements could applied between present-day and pre-industrial
over- or under-predict droplet concentrations when climate, there was a large difference in the net
applied to other regions depending on differences in radiation budgets, which can be attributed to the AEI
composition.” GCMs must not ignore the chemical (Figure 1.3.3.2.1). This likely would result in the
composition of aerosols—but they currently do. model yielding “an unrealistic temperature evolution
In another study, Golaz et al. (2011) used the [from preindustrial to current times] compared to
most recently released version of the Geophysical observations.”
Fluid Dynamics Laboratory’s (GFDL) Atmospheric
General Circulation Model (AGCM), called AM3
(Donner et al., 2011), to explore the “aerosol indirect
effect” (AIE) using a prognostic parameterization
scheme for cloud droplet numbers. The AIE is simply
the impact particulates and chemicals have on Earth’s
radiation budget by their influence on cloud
properties. Cloud properties such as composition (ice
versus water), droplet size, and droplet density are
critical determinants of how clouds influence Earth’s
radiation budget.
Past studies have shown cloud droplet numbers Figure 1.3.3.2.1. The zonal mean difference in the top of the
are a function of aerosol types, temperature, pressure, atmosphere shortwave absorption between model runs with
and vertical motion (which in stratiform clouds is present-day conditions and preindustrial conditions using the
related to turbulence). Golaz et al. provided a six-year control and three experimental parameterizations described
control simulation of climate after allowing one year here. The thicker lines indicate which results are statistically
for model equilibration. Then they changed the cloud significant at the 95% confidence level. The red line represents
droplet numbers in a predictive cloud scheme by the control run, and the blue, green, and purple represent the
reducing the turbulence in the model (experiment 1). model configured for differences in cloud formation. Adapted
from Golaz et al. (2011) Figure 5a.
In the second experiment an additional adjustment
was made by allowing droplet formation in new
clouds and preexisting clouds. The third and final This paper demonstrates that uncertainty in model
experiment adjusted the vertical motion profile and formulations, especially processes such as cloud
the turbulence used in experiment 2. parameterizations, can yield considerable uncertainty
When these experiments were run for one year, in climate projections and scenarios. The results also
the latter two experiments produced more droplets, show the expected magnitude of the output is not
making the clouds more reflective and resulting in known and model physics (parameterizations) must
less incoming solar radiation. Golaz et al. note the be adjusted within allowable ranges to conform with
differences among all the experiments were similar to our current understanding of a process, or back
the radiative forcing differences between today and towards basic conservation laws. These precautions
preindustrial times. But they also observe these short- are what many scientists have advised when looking
term experiments “could not be used for long-term at future climate scenarios.
coupled climate experiments, because the magnitude
of their net top-of-the-atmosphere (TOA) radiation References
fluxes is unrealistically large.” The model
configurations were readjusted to bring energy Donner, L.J., Wyman, B.L., Hemler, R.S., Horowitz, L.W.,
balance in line with the reference run. Ming, Y., Zhao, M., Golaz, J.C., Ginoux, P., Lin, S.-J.,
53
Schwarzkopf, M.D., Austin, J., Alaka, G., Cooke, W.F., providing suggestions for improving the model’s
Delworth, T.L., Freidenreich, S.M., Gordon, C.T., Griffies, capabilities. They used the National Aeronautics and
S.M., Held, I.M., Hurlin, W.J., Klein, S.A., Knutson, T.R., Space Administration (NASA) Goddard Institute for
Langenhorst, A.R., Lee, H.-C., Lin, Y., Magi, B.I., Space Studies Model-E, which employs a coupled
Malyshev, S.L., Milly, P.C.D., Naik, V., Nath, M.J.,
atmosphere-ocean GCM with fairly coarse resolution
Pincus, R., Ploshay, J.J., Ramaswamy, V., Seman, C.J.,
Shevliakova, E., Sirutis, J.J., Stern, W.F., Stouffer, R.J., in the horizontal (4° by 5° lat/lon) and vertical (23
Wilson, R.J., Winton, M., Wittenberg, A.T., and Zenga, F. layers). The model contained levels up to 80 km,
2011. The dynamical core, physical parameterizations, and necessarily including the stratosphere.
basic simulations characteristics of the atmospheric The control model run consisted of a 20-member
component of the GFDL Global Coupled Model CM3. suite globally from 2007 to 2010. In the experiment,
Journal of Climate 24: 3484–3519. 1.5 Tg of volcanic material was injected into the
atmosphere at a point near Sarychev in 2008 of the
Golaz, J.-C., Salzmann, M., Donner, L.J., Horowitz, L.W.,
Ming, Y., and Zhao, M. 2011. Sensitivity of the aerosol model year. The observed aerosol measurements
indirect effect to subgrid variability in the cloud came from ground-based LIDARS at six locations
parameterization of the GFDL atmosphere general around the world as well as satellite-based
circulation model AM3. Journal of Climate 24: 3145– measurements that profile the aerosol concentration
3160. using scattered sunlight (Optical Spectrograph and
Infrared Imaging System (OSIRIS)).
Roesler, E.L. and Penner, J.E. 2010. Can global models
The authors found the model did a reasonably
ignore the chemical composition of aerosols? Geophysical
Research Letters 37: L24809, doi:10.1029/2010GL044282. good job of spreading the volcanic material around
the Northern Hemisphere, but there were some
important discrepancies between the model and
1.3.3.3 Volcanic Aerosols observations (e.g. Figure 1.3.3.3.1). For example, the
The effects of volcanic eruptions on climate have model transported the material too quickly into the
been studied and are well-known. Volcanic eruptions tropics and too slowly into the high latitudes. The
eject particulate and aerosol materials into the authors speculate this error may indicate a need to
stratosphere. These aerosols, primarily sulfate type, improve the model’s depiction of stratospheric
increase the albedo of the planet, resulting in less circulation. Also, the model tended to remove
incoming solar radiation and a cooling of the lower aerosols too quickly from the atmosphere, especially
troposphere and surface. Generally, the greater the in the high latitudes, which may have been an
eruption, the stronger this effect will be and the indication of model overestimation of particulate size.
longer it will last. Volcanic eruptions whose Note that in Figure 1.3.3.3.1, the modeled peak
emissions are confined to the troposphere generally aerosol values occurred one month earlier than
have little effect on climate, as the troposphere is observed and then decreased in concentration too
more efficient in scavenging out the relatively large quickly.
particulates. The sensitivity of GCMs to aerosol forcing also
Volcanism is an “external” forcing to Earth’s can be assessed based on volcanic activity. The
atmosphere and its occurrence is considered to be Pinatubo eruption of 1991 was very large and
unpredictable and irregular. However, once the explosive, injecting a huge amount of sulfate aerosols
particulate matter and aerosols are injected into the into the atmosphere. The climate cooled in response,
atmosphere, it is possible to project the spread of the but the climate models cooled much more than the
material using a GCM. In June 2009, the Sarychev actual atmosphere, showing the models assumed
volcano in Russia’s Kamchatka Peninsula erupted strong radiative forcing (cooling) by sulfate aerosols,
explosively for approximately five days. At the time it a result also noted by Landrum et al. (2013). This is
was the second such eruption within a year. It injected important because one of the key assumptions of the
1.2 teragrams (Tg) of material into the atmosphere to models is that human pollution, particularly in the
an estimated height of as much as 16 km—nearly 10 post-1970 period, strongly dampens out the warming
miles. that greenhouse gas emissions otherwise would have
Kravitz et al. (2011) studied measurements of the caused. For most of the period simulated, however, no
optical depth of the aerosol sulfates from this eruption data (or only fragmentary data) are available to
and compared these with the projected output from a estimate sulfate aerosol pollution. Furthermore, the
20-member ensemble using a GCM, with the goal of forcing by any given level of aerosols is not easily
54
seemingly “random” forcing processes. The likely

impact on surface temperatures from the Kravitz et al.
(2011) experiment would be a bias toward warm
temperatures on the time scale of months.
Driscoll et al. (2012) report that Stenchikov et al.
(2006) analyzed seven models used in the Fourth
Assessment Report of the Intergovernmental Panel on
Climate Change (IPCC, 2007) that “included all the
models that specifically represented volcanic
eruptions.” The scientists found the strength and
spatial pattern of the surface temperature anomalies
predicted by them were not “well reproduced.”
Hoping to find some improvement in more recent
versions of the models, Driscoll et al. repeated the
Figure 1.3.3.3.1. The LIDAR retrievals from Hefei, China
analysis of Stenchikov et al. (2006), using 13 model
as compared to ModelE output and OSIRIS retrievals (left).
The monthly averages of backscatter as a function of
simulations from the Coupled Model Intercomparison
altitude maximizing in September 2009 (right). The Project phase 5 (CMIP5)—an overview of which is
integrated (15–25 km) optical depth through the stratosphere given by Taylor et al. (2011)—while focusing their
for the LIDAR data (black), zonally averaged stratospheric analysis on the regional impacts of the largest
aerosol optical depth calculated by the model in the grid volcanic eruptions on the Northern Hemisphere (NH)
latitude containing the Hefei LIDAR (28°–32°N) (red), and large-scale circulation during the winter season.
OSIRIS retrievals zonally averaged over the latitude band According to the five researchers, “the models
30°–35°N (blue). Adapted from Figure 11 from Kravitz et generally fail to capture the NH dynamical response
al. (2011). following eruptions.” More specifically, they state the
models “do not sufficiently simulate the observed
measured and is usually estimated from model post-volcanic strengthened NH polar vortex, positive
responses to things like volcanic activity, which is not North Atlantic Oscillation, or NH Eurasian warming
a direct parallel to human pollution. Thus the aerosol pattern, and they tend to overestimate the cooling in
forcing history of the twentieth century used by the the tropical troposphere.” They also note “none of the
IPCC is very approximate, and in fact different models simulate a sufficiently strong reduction in the
histories are often used by different modeling groups. geopotential height at high latitudes,” and
Kiehl (2007) found GCMs with a larger correspondingly, “the mean sea level pressure fields
anthropogenic forcing (more warming, which varied and temperature fields show major differences with
by a factor of 2) used a larger aerosol forcing respect to the observed anomalies.” In addition, they
(compensating cooling, which varied by a factor of find “all models show considerably less variability in
3). Modeling groups thus appear to have chosen an high-latitude stratospheric winds than observed,” and
aerosol history that makes their model fit the “none of the models tested have a Quasi-Biennial
historical record or tuned their model to accommodate Oscillation in them.”
the aerosol history they utilized. Given such “substantially different dynamics
Allen et al. (2013) found models do not capture between the models,” Driscoll et al. indicate they had
the global dimming from the 1950s to 1980s and “hoped to find at least one model simulation that was
brightening from the 1990s, which implies “model dynamically consistent with observations, showing
underestimation of the observed trends is related to improvement since Stenchikov et al. (2006).” But
underestimation of aerosol radiative forcing and/or “disappointingly,” as they put it, they found “despite
deficient aerosol emission inventories.” This relatively consistent post volcanic radiative changes,
undermines the claim that the models are simply none of the models manage to simulate a sufficiently
deduced from basic physics, and it also explains why strong dynamical response.” Thus they state their
they currently seem to be running hotter than the study “confirms previous similar evaluations and
actual climate. raises concern for the ability of current climate
Volcanic aerosols represent yet another models to simulate the response of a major mode of
demonstration that climate models have a difficult global circulation variability to external forcings,”
time representing the impact of external and indicating “this is also of concern for the accuracy of
55
geoengineering modeling studies that assess the interactions. Several studies suggest older model
atmospheric response to stratosphere-injected parameterizations did not succeed in this regard
particles.” (Groisman et al., 2000), and subsequent studies, as
discussed in this section, suggest they still are not
References succeeding. The lack of success may not be due
entirely to inadequate model resolution: If cloud
Allen, R.J., Norris, J.R., and Wild, M. 2013. Evaluation of processes have not been adequately and properly
multidecadal variability in CMIP5 surface solar radiation quantified, not even the highest resolution will bring
and inferred underestimation of aerosol direct effects over success.
Europe, China, Japan and India. Journal of Geophysical Lane et al. (2000) evaluated the sensitivities of
Research 118: 6311–6336.
the cloud-radiation parameterizations utilized in
Driscoll, S., Bozzo, A., Gray, L.J., Robock, A., and contemporary GCMs to changes in vertical model
Stenchikov, G. 2012. Coupled Model Intercomparison resolution, varying the latter from 16 to 60 layers in
Project 5 (CMIP5) simulations of climate following increments of four and comparing the results to
volcanic eruptions. Journal of Geophysical Research 117: observed values. This effort revealed cloud fraction
D17105, 10.1029/JD017607. varied by approximately 10 percent over the range of
Kiehl, J.T. 2007. Twentieth century climate model resolutions tested, which corresponded to about 20
response and climate sensitivity. Geophysical Research percent of the observed cloud cover fraction.
Letters 34: L22710, doi:10.1029/2007GL031383. Similarly, outgoing longwave radiation varied by 10
to 20 Wm-2 as model vertical resolution was varied,
Kravitz, B., Robock, A., Bourassa, A., Deshler, T., Wu, D., amounting to approximately 5 to 10 percent of
Mattis, I., Finger, F., Hoffmann, A., Ritter, C., Bitar, L.,
observed values, and incoming solar radiation
Duck, T.J., and Barnes J.E. 2011. Simulation and
observations of stratospheric aerosols from the 2009 experienced similar significant variations across the
Sarychev volcanic eruption. Journal of Geophysical range of resolutions tested. The model results did not
Research-Atmospheres 116: D18211, doi:10.1029/ converge, even at a resolution of 60 layers.
2010JD015501. In an analysis of the multiple roles played by
cloud microphysical processes in determining tropical
Landrum, L., Otto-Bliesner, B.L., Wahl, E.R., Conley, A., climate, Grabowski (2000) found much the same
Lawrence, P.J., Rosenbloom, N., and Teng, H. 2013. Last
thing, noting there were serious problems of computer
millennium climate and its variability in CCSM4. Journal
of Climate 26: 1085–1111. models failing to correctly incorporate cloud
microphysics. These observations led him to conclude
Stenchikov, G., Hamilton, K., Stouffer, R.J., Robock, A., “it is unlikely that traditional convection
Ramaswamy, V., Santer, B., and Graf, H.-F. 2006. Arctic parameterizations can be used to address this
Oscillation response to volcanic eruptions in the IPCC AR4 fundamental question in an effective way.” He also
climate models. Journal of Geophysical Research 111: became convinced that “classical convection
D07107, doi:10.1029/2005JD006286.
parameterizations do not include realistic elements of
Taylor, K.E., Stouffer, R.J., and Meehl, G.A. 2011. An cloud physics and they represent interactions among
overview of CMIP5 and the experiment design. Bulletin of cloud physics, radiative processes, and surface
the American Meteorological Society 93: 485–498. processes within a very limited scope.” Consequently,
he added, “model results must be treated as
qualitative rather than quantitative.”
1.3.4 Clouds
Reaching similar conclusions were Gordon et al.
Correctly parameterizing the influence of clouds on
(2000), who determined many GCMs of the late
climate is an elusive goal the creators of atmospheric
1990s tended to under-predict the presence of
GCMs have yet to achieve. One major reason for their
subtropical marine stratocumulus clouds and failed to
lack of success has to do with inadequate model
simulate the seasonal cycle of clouds. These
resolution on vertical and horizontal space scales.
deficiencies are important because these particular
Lack of resolution forces modelers to parameterize
clouds exert a major cooling influence on the surface
the ensemble large-scale effects of processes that
temperatures of the sea below them. In the situation
occur on smaller scales than the models are capable of
investigated by Gordon and his colleagues, the
handling. This is particularly true of physical
removal of the low clouds, as occurred in the normal
processes such as cloud formation and cloud-radiation
application of their model, led to sea surface
56
temperature increases on the order of 5.5°C. tropics and ... low boundary clouds are optically
Further condemnation of turn-of-the-century thick, the cloud albedo calculated by [Fu et al.] is too
model treatments of clouds came from Harries large for cirrus clouds and too small for boundary
(2000), who write our knowledge of high cirrus layer clouds,” they write, so “the near-zero contrast in
clouds is very poor and “we could easily have cloud albedos derived by [Fu et al.] has the effect of
uncertainties of many tens of Wm-2 in our description underestimating the iris effect.” In the end, Chou et
of the radiative effect of such clouds, and how these al. agreed Lindzen et al. “may indeed have
properties may change under climate forcing.” overestimated the iris effect somewhat, though hardly
Lindzen et al. (2001) analyzed cloud cover and by as much as that suggested by [Fu et al.].”
sea surface temperature (SST) data over a large Grassl (2000), in a review of the then-current
portion of the Pacific Ocean, finding a strong inverse status of the climate-modeling enterprise two years
relationship between upper-level cloud area and mean before the infrared iris effect debate emerged, noted
SST, such that the area of cirrus cloud coverage changes in many climate-related phenomena,
normalized by a measure of the area of cumulus including cloud optical and precipitation properties
coverage decreased by about 22 percent for each caused by changes in the spectrum of cloud
degree C increase in cloudy region SST. Essentially, condensation nuclei, were insufficiently well known
as the researchers describe it, “the cloudy-moist to provide useful insights into future conditions. In
region appears to act as an infrared adaptive iris that light of this knowledge gap, he recommended “we
opens up and closes down the regions free of upper- must continuously evaluate and improve the GCMs
level clouds, which more effectively permit infrared we use,” although he acknowledges contemporary
cooling, in such a manner as to resist changes in climate model results were already being “used by
tropical surface temperature.” The sensitivity of this many decision-makers, including governments.”
negative feedback was calculated by Lindzen et al. to Some may consider what is currently known
be substantial. They estimate it would “more than about clouds to be sufficient for predictive purposes,
cancel all the positive feedbacks in the more sensitive but the host of questions posed by Grassl—for which
current climate models” being used to predict the definitive answers are still lacking—demonstrates this
consequences of projected increases in atmospheric assumption is erroneous. As but a single example,
CO2 concentration. Charlson et al. (1987) describe a negative feedback
Lindzen’s conclusions did not go uncontested, process that links biologically produced dimethyl
and Hartmann and Michelsen (2002) quickly claimed sulfide (DMS) in the oceans with climate. (See
the correlation noted by Lindzen et al. resulted from Chapter 2 of this volume for a more complete
variations in subtropical clouds not physically discussion of this topic.) This hypothesis holds that
connected to deep convection near the equator, and it the global radiation balance is significantly influenced
was thus “unreasonable to interpret these changes as by the albedo of marine stratus clouds and that the
evidence that deep tropical convective anvils contract albedo of these clouds is a function of cloud droplet
in response to SST increases.” Fu et al. (2002) also concentration, which is dependent upon the
chipped away at the adaptive infrared iris concept, availability of condensation nuclei that have their
arguing “the contribution of tropical high clouds to origin in the flux of DMS from the world’s oceans to
the feedback process would be small since the the atmosphere.
radiative forcing over the tropical high cloud region is Acknowledging that the roles played by DMS
near zero and not strongly positive,” while also oxidation products in the context described above are
claiming to show water vapor and low cloud effects “diverse and complex” and in many instances “not
were overestimated by Lindzen et al. by at least 60 well understood,” Ayers and Gillett (2000)
percent and 33 percent, respectively. As a result, Fu et summarized empirical evidence, derived from data
al. obtained a feedback factor in the range of -0.15 to collected at Cape Grim, Tasmania, and from reports
-0.51, compared to Lindzen et al.’s much larger of other pertinent studies in the peer-reviewed
negative feedback factor of -0.45 to -1.03. scientific literature, supporting Charlson et al.’s
In a contemporaneously published reply to this (1987) hypothesis. Ayers and Gillett found the “major
critique, Chou et al. (2002) state Fu et al.’s approach links in the feedback chain proposed by Charlson et
of specifying longwave emission and cloud albedos al. (1987) have a sound physical basis” and there is
“appears to be inappropriate for studying the iris “compelling observational evidence to suggest that
effect.” Since “thin cirrus are widespread in the DMS and its atmospheric products participate
57
significantly in processes of climate regulation and community … the problem remains largely unsolved”
reactive atmospheric chemistry in the remote marine and note, “at the current rate of progress, cloud
boundary layer of the Southern Hemisphere.” parameterization deficiencies will continue to plague
The empirical evidence analyzed by Ayers and us for many more decades into the future.”
Gillett highlights an important suite of negative Randall et al. declare “clouds are complicated,”
feedback processes that act in opposition to model- highlighting what they call the “appalling
predicted CO2-induced global warming over the complexity” of the cloud parameterization situation.
world’s oceans. These processes are not fully They also state “our understanding of the interactions
incorporated into even the best of the current climate of the hot towers [of cumulus convection] with the
models, nor are analogous phenomena that occur over global circulation is still in a fairly primitive state,”
land included in them, such as those discussed by and not knowing all that much about what goes up,
Idso (1990). it’s not surprising we also don’t know much about
Further to this point, O’Dowd et al. (2004) what comes down, as they report “downdrafts are
measured size-resolved physical and chemical either not parameterized or crudely parameterized in
properties of aerosols found in northeast Atlantic large-scale models.” It also should be noted Riehl and
marine air arriving at the Mace Head Atmospheric Malkus (1958), in an analysis of the equatorial trough
Research station on the west coast of Ireland during region, could not achieve closure of their large-scale
phytoplanktonic blooms at various times of the year. energy budgets without resorting to significant
They found in the winter, when biological activity vertical exchanges by way of the cumulonimbus
was at its lowest, the organic fraction of the sub- downdrafts.
micrometer aerosol mass was about 15 percent. With respect to stratiform clouds, the situation is
During the spring through autumn, however, when no better, as their parameterizations are described by
biological activity was high, “the organic fraction Randall et al. as “very rough caricatures of reality.”
dominates and contributes 63 percent to the sub- As for interactions between convective and stratiform
micrometer aerosol mass (about 45 percent is water- clouds, during the 1970s and ‘80s, Randall et al.
insoluble and about 18 percent water-soluble),” they report “cumulus parameterizations were extensively
write. Based on these findings, they performed model tested against observations without even accounting
simulations that indicated the marine-derived organic for the effects of the attendant stratiform clouds.”
matter “can enhance the cloud droplet concentration They reported the concept of detrainment was
by 15 percent to more than 100 percent and is “somewhat murky” and the conditions that trigger
therefore an important component of the aerosol- detrainment were “imperfectly understood.” “At this
cloud-climate feedback system involving marine time,” as they put it, “no existing GCM includes a
biota.” satisfactory parameterization of the effects of
O’Dowd et al. (2004) state their findings mesoscale cloud circulations.”
“completely change the picture of what influences Randall et al. additionally state “the large-scale
marine cloud condensation nuclei given that water- effects of microphysics, turbulence, and radiation
soluble organic carbon, water-insoluble organic should be parameterized as closely coupled processes
carbon and surface-active properties, all of which acting in concert,” but they report only a few GCMs
influence the cloud condensation nuclei activation have even attempted to do so. Why? Because, as they
potential, are typically not parameterized in current continue, “the cloud parameterization problem is
climate models,” or as they note elsewhere in their overwhelmingly complicated” and “cloud
paper, “an important source of organic matter from parameterization developers,” as they call them, are
the ocean is omitted from current climate-modeling still “struggling to identify the most important
predictions and should be taken into account.” processes on the basis of woefully incomplete
Another perspective on the cloud-climate observations.” They add, “there is little question why
conundrum is provided by Randall et al. (2003), who the cloud parameterization problem is taking a long
state at the outset of their review of the subject that time to solve: It is very, very hard.” The four
“the representation of cloud processes in global scientists conclude, “a sober assessment suggests that
atmospheric models has been recognized for decades with current approaches the cloud parameterization
as the source of much of the uncertainty surrounding problem will not be ‘solved’ in any of our lifetimes.”
predictions of climate variability.” They report that To show the basis for this conclusion is robust
“despite the best efforts of [the climate modeling] and cannot be said to rest on the less-than-enthusiastic
58
remarks of a handful of exasperated climate modelers, climatologies derived from ten atmospheric GCMs
additional studies on the subject have been published with satellite measurements obtained from the
subsequent to Randall et al., any of which could have International Satellite Cloud Climatology Project
readily refuted their assessment of the situation if they (ISCCP) and the Clouds and Earth’s Radiant Energy
thought it was appropriate. System (CERES) program. ISCCP data were
Siebesma et al. (2004), for example, report available for 1983–2001, and data from the CERES
“simulations with nine large-scale models [were] program were available for the winter months of 2001
carried out for June/July/August 1998 and the quality and 2002 and for the summer months of 2000 and
of the results [was] assessed along a cross-section in 2001. The purpose of their analysis was twofold: to
the subtropical and tropical North Pacific ranging assess the current status of climate models in
from (235°E, 35°N) to (187.5°E, 1°S),” in order to simulating clouds so that future progress can be
“document the performance quality of state-of-the-art measured more objectively, and to reveal serious
GCMs in modeling the first-order characteristics of deficiencies in the models so as to improve them.
subtropical and tropical cloud systems.” The main The work of 20 climate modelers involved in this
conclusions of this study, according to the authors exercise revealed a long list of major model
are: “(1) almost all models strongly under predicted imperfections. First, Zhang et al. report a fourfold
both cloud cover and cloud amount in the difference in high clouds among the models, and that
stratocumulus regions while (2) the situation is the majority of the models simulated only 30 to 40
opposite in the trade-wind region and the tropics percent of the observed middle clouds, with some
where cloud cover and cloud amount are over models simulating less than a quarter of observed
predicted by most models.” They report “these middle clouds. For low clouds, they report half the
deficiencies result in an over prediction of the models underestimated them, such that the grand
downwelling surface short-wave radiation of typically mean of low clouds from all models was only 70 to
60 Wm-2 in the stratocumulus regimes and a similar 80 percent of what was observed. Furthermore, when
under prediction of 60 Wm-2 in the trade-wind regions stratified in optical thickness ranges, the majority of
and in the intertropical convergence zone (ITCZ),” the models simulated optically thick clouds more than
and these discrepancies are to be compared with a twice as frequently as was found to be the case in the
radiative forcing of only a couple of Wm-2 for a 300 satellite observations, and the grand mean of all
ppm increase in the atmosphere’s CO2 concentration. models simulated about 80 percent of optically
In addition, they note “similar biases for the short- intermediate clouds and 60 percent of optically thin
wave radiation were found at the top of the clouds. In the case of individual cloud types, the
atmosphere, while discrepancies in the outgoing long- group of researchers note “differences of seasonal
wave radiation are most pronounced in the ITCZ.” amplitudes among the models and satellite
The 17 scientists hailing from nine different measurements can reach several hundred percent.”
countries who comprised Siebesma et al. also found Zhang et al. (2005) conclude “much more needs to be
“the representation of clouds in general-circulation done to fully understand the physical causes of model
models remains one of the most important as yet cloud biases presented here and to improve the
unresolved issues in atmospheric modeling.” This is models.”
partially due, they continue, “to the overwhelming L’Ecuyer and Stephens (2007) used multi-sensor
variety of clouds observed in the atmosphere, but observations of visible, infrared, and microwave
even more so due to the large number of physical radiance obtained from the Tropical Rainfall
processes governing cloud formation and evolution as Measuring Mission satellite for the period from
well as the great complexity of their interactions.” January 1998 through December 1999 in order to
Hence, they conclude that through repeated critical evaluate the sensitivity of atmospheric heating—and
evaluations of the type they conducted, “the scientific the factors that modify it—to changes in east-west sea
community will be forced to develop further surface temperature gradients associated with the
physically sound parameterizations that ultimately strong 1998 El Niño event in the tropical Pacific, as
result in models that are capable of simulating our expressed by the simulations of nine general
climate system with increasing realism.” circulation models of the atmosphere that were
In their effort to assess the status of state-of-the- utilized in the IPCC’s most recent Fourth Assessment
art climate models in simulating cloud-related Report. This protocol, in their words, “provides a
processes, Zhang et al. (2005) compared basic cloud natural example of a short-term climate change
59
scenario in which clouds, precipitation, and regional June 1998.

energy budgets in the east and west Pacific are Zhou et al. report: (1) “the model has much
observed to respond to the eastward migration of higher domain-averaged OLR (outgoing longwave
warm sea surface temperatures.” radiation) due to smaller total cloud fraction”; (2) “the
Results indicate “a majority of the models model has a more skewed distribution of OLR and
examined do not reproduce the apparent westward effective cloud top than CERES observations,
transport of energy in the equatorial Pacific during the indicating that the model’s cloud field is insufficient
1998 El Niño event.” They also found “the inter- in area extent”; (3) “the GCE is ... not very efficient
model variability in the responses of precipitation, in stratiform rain conditions because of the large
total heating, and vertical motion is often larger than amounts of slowly falling snow and graupel that are
the intrinsic ENSO signal itself, implying an inherent simulated”; and (4) “large differences between model
lack of predictive capability in the ensemble with and observations exist in the rain spectrum and the
regard to the response of the mean zonal atmospheric vertical hydrometeor profiles that contribute to the
circulation in the tropical Pacific to ENSO.” In associated cloud field.”
addition, they report “many models also misrepresent One year later, a study by Spencer and Braswell
the radiative impacts of clouds in both regions [the (2008) observed “our understanding of how sensitive
east and west Pacific], implying errors in total the climate system is to radiative perturbations has
cloudiness, cloud thickness, and the relative been limited by large uncertainties regarding how
frequency of occurrence of high and low clouds.” As clouds and other elements of the climate system
a result of these much-less-than-adequate findings, feedback to surface temperature change (e.g., Webster
the two researchers from Colorado State University’s and Stephens, 1984; Cess et al., 1990; Senior and
Department of Atmospheric Science conclude Mitchell, 1993; Stephens, 2005; Soden and Held,
“deficiencies remain in the representation of 2006; Spencer et al., 2007).” The two scientists from
relationships between radiation, clouds, and the Earth System Science Center at the University of
precipitation in current climate models,” and these Alabama in Huntsville, Alabama then point out
deficiencies “cannot be ignored when interpreting computer models typically assume that if the causes
their predictions of future climate.” of internal sources of variability (X terms) are
In a contemporaneous publication, this one in the uncorrelated to surface temperature changes, then
Journal of the Atmospheric Sciences, Zhou et al. they will not affect the accuracy of regressions used
(2007) state “clouds and precipitation play key roles to estimate the relationship between radiative flux
in linking the Earth’s energy cycle and water cycles,” changes and surface temperature (T). But “while it is
noting “the sensitivity of deep convective cloud true that the processes that cause the X terms are, by
systems and their associated precipitation efficiency [Forster and Gregory (2006)] definition, uncorrelated
in response to climate change are key factors in to T, the response of T to those forcings cannot be
predicting the future climate.” They also report cloud- uncorrelated to T—for the simple reason that it is a
resolving models, or CRMs, “have become one of the radiative forcing that causes changes in T.” They then
primary tools to develop the physical ask, “to what degree could nonfeedback sources of
parameterizations of moist and other subgrid-scale radiative flux variability contaminate feedback
processes in global circulation and climate models,” estimates?”
and CRMs could someday be used in place of In an attempt to answer this question, Spencer
traditional cloud parameterizations in such models. and Braswell used a “very simple time-dependent
In this regard, the authors note “CRMs still need model of temperature deviations away from an
parameterizations on scales smaller than their grid equilibrium state” to estimate the effects of “daily
resolutions and have many known and unknown random fluctuations in an unknown nonfeedback
deficiencies.” To help stimulate progress in these radiative source term N, such as those one might
areas, the nine scientists compared the cloud and expect from stochastic variations in low cloud cover.”
precipitation properties observed from CERES and Repeated runs of the model found the diagnosed
Tropical Rainfall Measuring Mission (TRMM) feedback departed from the true, expected feedback
instruments against simulations obtained from the value of the radiative forcing, with the difference
three-dimensional Goddard Cumulus Ensemble increasing as the amount of nonfeedback radiative
(GCE) model during the South China Sea Monsoon flux noise was increased. “It is significant,” the
Experiment (SCSMEX) field campaign of 18 May–18 authors write, “that all model errors for runs
60
consistent with satellite-observed variability are in the radiative physics they employ, as parameterized at the
direction of positive feedback, raising the possibility grid scale, is probably faulty.
that current observational estimates of cloud feedback Shifting to a different aspect of the topic, climate
are biased in the positive direction.” In other words, modelers have long struggled to adequately represent
as the authors report in their abstract, “current the sensitivity of convective cloud systems to
observational diagnoses of cloud feedback—and tropospheric humidity in their mathematical
possibly other feedbacks—could be significantly representations of Earth’s climate system. Del Genio
biased in the positive direction.” (2012) reviewed the rate of progress in this important
Writing as background for their work, Zhang et endeavor in a paper published in the journal Surveys
al. (2010) state different representations of clouds and in Geophysics. The U.S. National Aeronautics and
their feedback processes in GCMs have been Space Administration scientist—stationed at the
identified as major sources of differences in model Goddard Institute for Space Studies in New York—
climate sensitivities, noting “contemporary GCMs found a number of important problems that have yet
cannot resolve clouds and highly simplified to be adequately resolved. He notes, for example, that
parameterizations are used to represent the many parameterizations of convective cloud
interactions between clouds and radiation.” In variability “are not sufficiently sensitive to variations
conducting their own study of the subject, therefore, in tropospheric humidity.” That “lack of sensitivity,”
they combine cloud profiling radar data from the as he describes it, “can be traced in part to
CloudSat satellite with lidar data from the CALIPSO underestimated entrainment of environmental air into
satellite to obtain 3D profiles of clouds and rising convective clouds and insufficient evaporation
precipitation regimes across the tropics. Some of of rain into the environment.” As a result of these
these profiles corresponded to well-known weather deficiencies, he notes, “the parameterizations produce
features, such as low clouds, thin cirrus, cirrus anvils, deep convection too easily while stabilizing the
etc., and they were compared to output obtained from environment too quickly to allow the effects of
the Community Atmosphere Model version 3 convective mesoscale organization to occur.”
(CAM3.1). Del Genio does note “recent versions of some
This analysis revealed the model “overestimates models have increased their sensitivity to
the area coverage of high clouds and underestimates tropospheric humidity and improved some aspects of
the area coverage of low clouds in subsidence their variability,” but he says “a parameterization of
regions.” Zhang et al. found particularly striking “the mesoscale organization is still absent from most
model overestimate of the occurrence frequency of models,” and “adequately portraying convection in all
deep convection and the complete absence of cirrus its realizations remains a difficult problem.”
anvils,” plus the fact that “the modeled clouds are too On another note, Del Genio writes, “to date,
reflective in all regimes.” metrics for model evaluation have focused almost
Since incoming and outgoing radiation are exclusively on time mean two-dimensional spatial
strongly affected by the 3D spatial pattern of clouds distributions of easily observed parameters,” and he
of various types, a model that gets the “right” current indicates “it has become clear that such metrics have
global temperature with the wrong pattern of clouds no predictive value for climate feedbacks and climate
must have errors in its radiation and/or heat transfer sensitivity (e.g., Collins et al., 2011),” while adding
parameterizations. In addition, the manner in which those metrics “are also probably not helpful for
future climate scenarios achieve amplification of the assessing most other important features of future
direct radiative effect of increased greenhouse gases climate projections, because temporal variability
(the assumed positive feedback) is also not likely to gives greater insight into the physical processes at
be correct if the 3D pattern of simulated clouds is as work.”
far off as shown in this study. What is more, the Del Genio concludes, “given the insensitivity of
pattern of clouds also reflects convective processes these models to tropospheric humidity and their
that distribute heat and water vapor in the atmosphere, failure to simulate the Madden-Julian Oscillation and
and the results of Zhang et al. point to deficiencies in diurnal cycle, ... it seems unlikely that it will ever be
the handling of this aspect of atmospheric dynamics possible to establish a general set of metrics that can
as well. Climate modelers’ claims of physical realism be used to anoint one subset of models as our most
in their models are not supported by detailed reliable indicators of all aspects of climate change.”
comparisons with the real world, and the basic In another study, Cesana and Chepfer (2012)
61
compare the most recent cloud representations of five large biases,” “despite the availability of relevant
of the climate models involved in the Coupled Model observations.” Even in cases where “the models may
Intercomparison Project Phase 5 (CMIP5) effort be providing roughly the correct radiative energy
described by Taylor et al. (2012) with real-world budget,” they state “many are accomplishing it by
satellite-derived observations obtained from the means of unrealistic cloud characteristics of cloud ice
GCM-Oriented CALIPSO Cloud Product (GOCCP), mass at a minimum, which in turn likely indicates
described by Chepfer et al. (2010). According to unrealistic cloud particle sizes and cloud cover.” Li et
Cesana and Chepfer, the results indicated: (1) “low- al. conclude “cloud feedback will undoubtedly still
and mid-level altitude clouds are underestimated by represent a key uncertainty in [even] the next IPCC
all the models (except in the Arctic),” (2) “high assessment report.”
altitude cloud cover is overestimated by some Cesana et al. (2012) state “low-level clouds
models,” (3) “some models shift the altitude of the frequently occur in the Arctic and exert a large
clouds along the ITCZ by 2 km (higher or lower) influence on Arctic surface radiative fluxes and Arctic
compared to observations,” (4) “the models hardly climate feedbacks,” noting that during winter, in
reproduce the cloud free subsidence branch of the particular, surface net longwave radiation (FLW,NET)
Hadley cells,” (5) “the high-level cloud cover is often has a bimodal distribution, with extremes that have
too large,” (6) “in the tropics, the low-level cloud been termed “radiatively clear” and “radiatively
cover (29% in CALIPSO-GOCCP) is underestimated opaque.” They note Arctic ice clouds “tend to have
by all models in subsidence regions (16% to 25%),” small optical depths and a weak influence on
and (7) “the pronounced seasonal cycle observed in FLW,NET,” which explains the “radiatively clear”
low-level Arctic clouds is hardly simulated by some condition, whereas Arctic liquid-containing clouds
models.” “generally have large optical depths and a dominant
Also writing in 2012, Li et al. (2012) state influence on FLW,NET (Shupe and Intrieri, 2004),”
“representing clouds and cloud climate feedback in which explains the “radiatively opaque” condition, as
global climate models (GCMs) remains a pressing discussed by Doyle et al. (2011).
challenge,” but one that must be overcome in order Against this backdrop, Cesana et al. employed
“to reduce and quantify uncertainties associated with real-world Cloud-Aerosol Lidar and Infrared
climate change projections.” Two of the primary Pathfinder Satellite Observation (CALIPSO) data to
parameters that must be accurately modeled in this document cloud phases over the Arctic basin (60–
regard are cloud ice water content (CIWC) and cloud 82°N) during the five-year period 2006–2011, after
ice water path (CIWP). which they used the results they obtained “to evaluate
Li et al. performed, in their words, “an the influence of Arctic cloud phase on Arctic cloud
observationally based evaluation of the cloud ice radiative flux biases in climate models.” The five
water content and path of present-day GCMs, notably researchers report their evaluation of climate models
20th century CMIP5 simulations,” after which they participating in the most recent Coupled Model Inter-
compared the results to two recent reanalyses. They comparison Project (Taylor et al., 2012) revealed
used “three different CloudSat + CALIPSO ice water “most climate models are not accurately representing
products and two methods to remove the contribution the bimodality of FLW,NET in non-summer seasons.”
from the convective core ice mass and/or precipitating Even when advanced microphysical schemes that
cloud hydrometeors with variable sizes and falling predict cloud phase have been used, such as those
speeds so that a robust observational estimate can be currently employed in the fifth version of the
obtained for model evaluations.” Community Atmosphere Model (CAM5, Neale et al.,
The 11 U.S. scientists report, “for annual mean 2010), “insufficient liquid water was predicted.”
CIWP, there are factors of 2–10 in the differences Cesana et al. conclude “the simple prescribed
between observations and models for a majority of relationships between cloud phase and temperature
the GCMs and for a number of regions,” and that have historically been used in climate models are
“systematic biases in CIWC vertical structure occur incapable of reproducing the Arctic cloud phase
below the mid-troposphere where the models observations described here,” which must inevitably
overestimate CIWC.” They ultimately conclude lead to similarly inaccurate values of “Arctic surface
“neither the CMIP5 ensemble mean nor any radiative fluxes and Arctic climate feedbacks” when
individual model performs particularly well,” adding, employed in current climate models.
“there are still a number of models that exhibit very Also focusing on low-level clouds were Nam et
62
al. (2012), who write the response of low-level clouds climatology of a model, Stratton and Stirling used a
has long been identified as “a key source of GCM with a horizontal resolution of less than two
uncertainty for model cloud feedbacks under climate degrees in latitude and longitude and ran two ten-year
change,” citing the work of Bony and Dufresne simulations. The first was a control run with the
(2005), Webb et al. (2006), Wyant et al. (2006), and model as it was provided to them. Then a ten-year run
Medeiros et al. (2008). They state “the ability of was performed with the new convective
climate models to simulate low-clouds and their parameterizations added. Results were compared with
radiative properties” plays a large role in assessing observed precipitation, as derived from output
“our confidence in climate projections.” Nam et al. provided by the Tropical Rainfall Measurement
analyzed “outputs from multiple climate models Mission (TRMM) satellite.
participating in the Fifth phase of the Coupled Model According to the authors, the control run (Figure
Intercomparison Project (CMIP5) using the Cloud 1.3.4.1) “tends to lack precipitation over India and
Feedback Model Intercomparison Project have too much precipitation over tropical land in
Observations Simulator Package (COSP), and Africa and in South America. The new run has
compared them with different satellite data sets,” generally reduced the precipitation over tropical land,
including “CALIPSO lidar observations, PARASOL tending to improve agreement with CMAP” (CMAP
mono-directional reflectances, and CERES radiative is an acronym for the observations). In the mid-
fluxes at the top of the atmosphere.” latitudes though, there were regions (e.g., western
The comparison revealed “the current generation Russia) where the new parameterization improved the
of climate models still experiences difficulties in model results, but other places where the new
predicting the low-cloud cover and its radiative parameterization was worse (e.g., Europe). The
effects.” In particular, they report the models: (1) results also did not noticeably alter the general
“under-estimate low-cloud cover in the tropics,” (2) circulation.
“over-estimate optical thickness of low-clouds, Stratton and Stirling improved the attendant
particularly in shallow cumulus regimes,” (3) “poorly physics associated with convection and in general
represent the dependence of the low-cloud vertical improved the climatological representation of
structure on large-scale environmental conditions,” precipitation. But there were still differences in
and (4) “predict stratocumulus-type of clouds in comparison with observed precipitation. In some
regimes where shallow cumulus cloud-types should regions, the new convective scheme performance was
prevail.” However, they say “the impact of these not as good. Improvement was not uniform, even
biases on the Earth’s radiation budget ... is reduced by though the model was overall closer to reality.
compensating errors,” including “the tendency of The new parameterizations may have done a
models to under-estimate the low-cloud cover and to good job of representing precipitation over short time-
over-estimate the occurrence of mid- and high-clouds scales, but over longer time-scales it still did not
above low-clouds.” represent the climatology of precipitation very well.
Convective-type clouds and precipitation, Ahlgrimm and Forbes (2012) investigated an
especially when occurring over land areas, in the irradiance bias in the European Centre for Medium
tropics, or during the warm season, have a strong Range Forecasting (ECMWF) GCM, focusing on the
diurnal cycle in phase with solar heating. People Southern Great Plains of the USA. This GCM was
living in coastal regions are familiar with the diurnal built to generate daily weather predictions. In one
cycle via the sea-breeze phenomenon. Unfortunately, experiment, the authors compared measured radiation
models have difficulty representing the diurnal daily in 2004–2009 from the Atmospheric Radiation
component of convection. Stratton and Stirling (2012) Measurement Site (ARM) in the Southern Great
devised a better way to represent this phenomenon in Plains (SGP) (Figure 1.3.4.2). In the second, they
an atmospheric general circulation model. In compared the ECMWF model output to observations
particular, they improved the rate at which of clouds, radiation, and the state of the atmosphere
environmental air and cloud air mix (entrainment). archived as the Climate Modelling Best Estimate
They also revised the mass flux over land in order to product from 1997–2009. Ahlgrimm and Forbes
improve the timing and strength of convection selected 146 days when fair weather cumulus clouds
influenced by diurnal heating. dominated the Southern Great Plains. They compared
In order to test the impact of the changes made to these to the same days run using the ECMWF model
the diurnal convection scheme on the precipitation initialized the day before and using the 18–42 hour
63
Figure 1.3.4.1. The mean summer season precipitation for (a) new convective climatology, (b) new run minus the control,
(c) control minus CMAP (observations), (d) new run minus CMAP. Adapted from Figure 3 from Stratton and Sterling
(2012).
forecasts. examining the formulation of the shallow convection

The results indicate the mean model radiative bias trigger, mass transport, and cloud microphysical
compared to the observed product was approximately properties.” If there is a net positive bias in shortwave
23 W m-2. When the authors looked at the fair reaching the surface, it will result in the
weather cumulus regime and studied the bias over the overestimation of temperatures in the two situations
course of a day, they found the bias in this regime described above and overall. If these
was weak. This means the parameterizations for fair parameterizations, or others that produce warm
weather convective cloudiness were largely biases, are used in climate models, the impact on
successful. When the authors attempted to classify climate scenarios would be a net surface warming.
clouds, they found the deep clouds (convective), thick Grodsky et al. (2012) point out “the seasonal
mid-level clouds, and low clouds as a group climate of the tropical Atlantic Ocean is notoriously
accounted for more than 50 percent of the bias. As difficult to simulate accurately in coupled models,”
there was no way to account for high clouds and other noting a long history of studies, including those of
phenomena, the remaining bias was simply referred to Zeng et al. (1996), Davey et al. (2002), Deser et al.
as “residual.” They also demonstrated refinements in (2006), Chang et al. (2007), and Richter and Xie
the cloud liquid water content and distributions could (2008), “have linked the ultimate causes of the
improve these parameterizations. persistent model biases to problems in simulating
Lastly, Alhgrimm and Forbes showed the biases winds and clouds by the atmospheric model
are largest when the observations were cloudy and the component.” In an effort designed to “revisit” this
model clear, or when the observations were overcast unsolved problem, Grodsky et al. utilized the
and the model produced broken skies. All other Community Climate System Model, version 4
categories showed small biases that largely cancelled (CCSM4; Gent et al., 2011), a coupled climate model
each other out. The two researchers conclude, “it will that simultaneously simulates Earth’s atmosphere,
be possible to carry out targeted sensitivity studies to ocean, land surface, and sea ice processes. They did
improve cloud occurrence and radiative properties by so by comparing twentieth century runs forced by
64
Writing in the Journal of Geophysical Research

(Atmospheres), Wang and Su (2013) note “coupled
general circulation models (GCMs) are the major tool
to predict future climate change, yet cloud-climate
feedback constitutes the largest source of uncertainty
in these modeled future climate projections.” Thus
they state “our confidence in the future climate
change projections by the coupled GCMs to a large
extent depends on how well these models simulate the
observed present-day distribution of clouds and their
associated radiative fluxes.” About their own work,
they write, “the annual mean climatology of top of the
atmosphere (TOA) shortwave and longwave cloud
radiative effects in 12 Atmospheric Model
Intercomparison Project (AMIP)-type simulations
participating in the Coupled Model Intercomparison
Project Phase 5 (CMIP5) [was] evaluated and
investigated using satellite-based observations, with a
Figure 1.3.4.2. The multiyear (2004–09) all-sky diurnal focus on the tropics.”
composite of surface irradiance. The black line is the
The two researchers report (1) the CMIP5 AMIPs
ECMWF model, and the grey line is the observations from the
CMBE product. Only daytime samples (modelled short wave “produce considerably less cloud amount [than what
down exceeding 1 W m−2) with good-quality coincident is observed], particularly in the middle and lower
observations are included. Adapted from Figure 1 in troposphere,” (2) there are “good model simulations
Ahlgrimm and Forbes (2012). in tropical means,” but they are “a result of
compensating errors over different dynamical
regimes,” (3) “over the Maritime Continent, most of
time-varying solar output, greenhouse gas, and the models simulate moderately less high-cloud
volcanic and other aerosol concentrations for the fraction, leading to weaker shortwave cooling and
period 1850–2005 with observed (real world) longwave warming and a larger net cooling,” (4)
monthly variability computed from observational “over subtropical strong subsidence regimes, most of
analyses during the 26-year period 1980–2005. the CMIP5 models strongly underestimate
The four researchers report finding (1) the stratocumulus cloud amount and show considerably
“atmospheric component of CCSM4 has abnormally weaker local shortwave cloud radiative forcings,” (5)
intense surface subtropical high pressure systems and “over the transitional trade cumulus regimes, a
abnormally low polar low pressure systems,” (2) “in notable feature is that while at varying amplitudes,
the tropics and subtropics, the trade wind winds are 1- most of the CMIP5 models consistently simulate a
2 m/sec too strong [and] latent heat loss is too large,” deeper and drier boundary layer, more moist free
(3) “sea surface temperature in the southeast has a troposphere, and more high clouds and consequently
warm bias [due in part to] erroneously weak overestimate shortwave cooling and longwave
equatorial winds,” (4) “the warm bias evident along warming effects there,” such that, in the final
the coast of southern Africa is also partly a result of analysis, (6) “representing clouds and their TOA
insufficient local upwelling,” (5) “excess radiation is radiative effects remains a challenge in the CMIP5
evident in the south stratocumulus region of up to 60 models.”
W/m2,” (6) there is “excess precipitation in the In a revealing paper published in the American
Southern Hemisphere,” and (7) “errors in cloud Meteorological Society’s Journal of Climate, Lauer
parameterization lead to “massively excess solar and Hamilton (2013) report numerous previous
radiation in austral winter and spring in CCSM4.” studies from the Coupled Model Intercomparison
It is clear that many important aspects of clouds Project phase 3 (CMIP3) showed large biases in the
and cloud cover remain inadequately modeled. The simulated cloud climatology affecting all GCMs, as
cloud parameterization problem is extremely complex well as “a remarkable degree of variation among the
and will likely remain that way for the foreseeable models that represented the state of the art circa
future. 2005.” The two researchers set out to provide an
65
update by describing the progress that has been made Ayers, G.P. and Gillett, R.W. 2000. DMS and its oxidation
in recent years by comparing mean cloud properties, products in the remote marine atmosphere: implications for
interannual variability, and the climatological climate and atmospheric chemistry. Journal of Sea
seasonal cycle from the CMIP5 models with results Research 43: 275–286.
from comparable CMIP3 experiments, as well as with Bony, S. and Dufresne, J. 2005. Marine boundary layer
actual satellite observations. clouds at the heart of tropical cloud feedback uncertainties
Lauer and Hamilton conclude “the simulated in climate models. Geophysical Research Letters 32:
cloud climate feedbacks activated in global warming 10.1029/2005GL023851.
projections differ enormously among state-of-the-art
Cesana, G. and Chepfer, H. 2012. How well do climate
models,” while noting “this large degree of models simulate cloud vertical structure? A comparison
disagreement has been a constant feature documented between CALIPSO-GOCCP satellite observations and
for successive generations of GCMs from the time of CMIP5 models. Geophysical Research Letters 39:
the first Intergovernmental Panel on Climate Change 10.1029/2012GL053153.
assessment through the CMIP3 generation models
used in the fourth IPCC assessment.” They add, “even Cesana, G., Kay, J.E., Chepfer, H., English, J.M., and de
Boer, G. 2012. Ubiquitous low-level liquid-containing
the model-simulated cloud climatologies for present-
Arctic clouds: New observations and climate model
day conditions are known to depart significantly from constraints from CALIPSO-GOCCP. Geophysical
observations and, once again, the variation among Research Letters 39: 10.1029/2012GL053385.
models is quite remarkable (e.g., Weare, 2004; Zhang
et al., 2005; Waliser et al., 2007, 2009; Lauer et al., Cess, R.D. et al. 1990. Intercomparison and interpretation
2010; Chen et al., 2011).” of climate feedback processes in 19 atmospheric general
The two researchers determined (1) “long-term circulation models. Journal of Geophysical Research 95:
16601–16615.
mean vertically integrated cloud fields have quite
significant deficiencies in all the CMIP5 model Chang, C.-Y., Carton, J.A., Grodsky, S.A., and Nigam, S.
simulations,” (2) “both the CMIP5 and CMIP3 2007. Seasonal climate of the tropical Atlantic sector in the
models display a clear bias in simulating too high NCAR Community Climate System Model 3: Error
LWP [liquid water path] in mid-latitudes,” (3) “this structure and probable causes of errors. Journal of Climate
bias is not reduced in the CMIP5 models,” (4) there 20: 1053–1070.
have been “little to no changes in the skill of Charlson, R.J., Lovelock, J.E., Andrea, M.O., and Warren,
reproducing the observed LWP and CA [cloud S.G. 1987. Oceanic phytoplankton, atmospheric sulfur,
amount],” (5) “inter-model differences are still large cloud albedo and climate. Nature 326: 655–661.
in the CMIP5 simulations,” and (6) “there is very
little to no improvement apparent in the tropical and Chen, W.-T., Woods, C.P., Li, J.-L.F., Waliser, D.E.,
Chern, J.-D., Tao, W.-K., Jiang, J.H., and Tompkins, A.M.
subtropical regions in CMIP5.”
2011. Partitioning CloudSat ice water content for
Lauer and Hamilton indicate there is “only very comparison with upper tropospheric ice in global
modest improvement in the simulated cloud atmospheric models. Journal of Geophysical Research 116:
climatology in CMIP5 compared with CMIP3,” 10.1029/2010JD015179.
adding even this slightest of improvements “is mainly
a result of careful model tuning rather than an Chou, M.-D., Lindzen, R.S., and Hou, A.Y. 2002. Reply to:
accurate fundamental representation of cloud “Tropical cirrus and water vapor: an effective Earth
infrared iris feedback?” Atmospheric Chemistry and
processes in the models.”
Physics 2: 99–101.
It would therefore appear, given the findings
described in this section, that the outlook for Collins, M., Booth, B.B.B., Bhaskaran, B., Harris, G.R.,
adequately modeling clouds and cloud processes must Murphy, J.M., Sexton, D.M.H., and Webb, M.J. 2011.
still be characterized as cloudy. Climate model errors, feedbacks and forcings: a
comparison of perturbed physics and multi-model
References ensembles. Climate Dynamics 36: 1737–1766.
Davey, M.K., Huddleston, M., Sperber, K. R., Braconnot,
Ahlgrimm, M. and Forbes, R. 2012. The Impact of Low P., Bryan, F., Chen, D., Colman, R., Cooper, C., Cubasch,
Clouds on Surface Shortwave Radiation in the ECMWF U., Delecluse, P., DeWitt, D., Fairhead, L., Flato, G.,
Model. Monthly Weather Review 140: 3783–3794. Gordon, C., Hogan, T., Ji, M., Kimoto, M., Kitoh, A.,
Knutson, T. R., Latif, M., Le Treut, H., Li, T., Manabe, S.,
66
Mechoso, C. R., Meehl, G. A., Power, S. B., Roeckner, E., balance. Contemporary Physics 41: 309–322.
Terray, L., Vintzileos, A., Voss, R., Wang, B., Washington,
W. M., Yoshikawa, I., Yu, J. Y., Yukimoto, S., and Zebiak, Hartmann, D.L. and Michelsen, M.L. 2002. No evidence
S. E. 2002. STOIC: A study of coupled model climatology for IRIS. Bulletin of the American Meteorological Society
and variability in tropical ocean regions. Climate Dynamics 83: 249–254.
18: 403–420. Idso, S.B. 1990. A role for soil microbes in moderating the
Del Genio, A.D. 2012. Representing the sensitivity of carbon dioxide greenhouse effect? Soil Science 149: 179–
convective cloud systems to tropospheric humidity in 180.
general circulation models. Surveys in Geophysics 33: 637– Lane, D.E., Somerville, R.C.J., and Iacobellis, S.F. 2000.
656. Sensitivity of cloud and radiation parameterizations to
Deser, C., Capotondi, A., Saravanan, R., and Phillips, A. changes in vertical resolution. Journal of Climate 13: 915–
2006. Tropical Pacific and Atlantic climate variability in 922.
CCSM3. Journal of Climate 19: 403–420. Lauer, A. and Hamilton, K. 2013. Simulating clouds with
Doyle, J.G., Lesins, G., Thakray, C.P., Perro, C., Nott, G.J., global climate models: A comparison of CMIP5 results
Duck, T.J., Damoah, R., and Drummond, J.R. 2011. Water with CMIP3 and satellite data. Journal of Climate 26:
vapor intrusions into the High Arctic during winter. 3823–3845.
Geophysical Research Letters 38: 10.1029/2011GL047493. Lauer, A., Hamilton, K., Wang, Y., Phillips, V.T.J., and
Dufresne, J.-L. and Bony, S. 2008. An assessment of the Bennartz, R. 2010. The impact of global warming on
primary sources of spread of global warming estimates marine boundary layer clouds over the eastern Pacific—A
from coupled atmosphere-ocean models. Journal of regional model study. Journal of Climate 23: 5844–5863.
Climate 21: 5135–5144. L’Ecuyer, T.S. and Stephens, G.L. 2007. The tropical
Forster, P.M., and Taylor, K.E. 2006. Climate forcings and atmospheric energy budget from the TRMM perspective.
climate sensitivities diagnosed from coupled climate model Part II: Evaluating GCM representations of the sensitivity
integrations, Journal of Climate 19: 6181–6194. of regional energy and water cycles to the 1998-99 ENSO
cycle. Journal of Climate 20: 4548–4571.
Fu, Q., Baker, M. and Hartmann, D.L. 2002. Tropical
cirrus and water vapor: an effective Earth infrared iris Li, J.-L.F., Waliser, D.E., Chen, W.-T., Guan, B., Kubar,
feedback? Atmospheric Chemistry and Physics 2: 31–37. T., Stephens, G., Ma, H.-Y., Deng, M., Donner, L., Seman,
C., and Horowitz, L. 2012. An observationally based
Gent, P.R., Danabasoglu, G., Donner, L.M., Holland, evaluation of cloud ice water in CMIP3 and CMIP5 GCMs
M.M., Hunke, E.C., Jayne, S.R., Lawrence, D.M., Neale, and contemporary reanalyses using contemporary satellite
R.B., Rasch, P.J., Vertenstein, M., Worley, P.H., Yang, Z.- data. Journal of Geophysical Research 117:
L., and Zhang M. 2011. The Community Climate System 10.1029/2012JD017640.
Model version 4. Journal of Climate 24: 4973–4991.
Lindzen, R.S., Chou, M.-D., and Hou, A.Y. 2001. Does the
Gordon, C.T., Rosati, A., and Gudgel, R. 2000. Tropical earth have an adaptive infrared iris? Bulletin of the
sensitivity of a coupled model to specified ISCCP low American Meteorological Society 82: 417–432.
clouds. Journal of Climate 13: 2239–2260.
Medeiros, B., Stevens, B., Held, I., Zhao, M., Williamson,
Grabowski, W.W. 2000. Cloud microphysics and the D., Olson, J., and Bretherton, C. 2008. Aquaplanets,
tropical climate: Cloud-resolving model perspective. climate sensitivity, and low clouds. Journal of Climate 21:
Journal of Climate 13: 2306–2322. 4974–4991.
Grassl, H. 2000. Status and improvements of coupled Nam, C., Bony, S., Dufresne, J.-L., and Chepfer, H. 2012.
general circulation models. Science 288: 1991–1997. The ‘too few, too bright’ tropical low-cloud problem in
CMIP5 models. Geophysical Research Letters 39:
Grodsky, S.A., Carton, J.A., Nigam, S., and Okumura, 10.1029/2012GL053421.
Y.M. 2012. Tropical Atlantic biases in CCSM4. Journal of
Climate 25: 3684–3701. Neale, R.B., et al. 2010. Description of the NCAR
Community Atmosphere Model (CAM 5.0). Technical Note
Groisman, P.Ya., Bradley, R.S., and Sun, B. 2000. The 486+STR. National Center for Atmospheric Research,
relationship of cloud cover to near-surface temperature and Boulder, Colorado, USA.
humidity: Comparison of GCM simulations with empirical
data. Journal of Climate 13: 1858–1878. O’Dowd, C.D., Facchini, M.C., Cavalli, F., Ceburnis, D.,
Mircea, M., Decesari, S., Fuzzi, S., Yoon, Y.J., and Putaud,
Harries, J.E. 2000. Physics of the Earth’s radiative energy J.-P. 2004. Biogenically driven organic contribution to
67
marine aerosol. Nature 431: 676–680. Taylor, K.E., Stouffer, R.J., and Meehl, G.A. 2012. An
overview of CMIP5 and the experimental design. Bulletin
Randall, D., Khairoutdinov, M., Arakawa, A., and of the American Meteorological Society 93: 485–498.
Grabowski, W. 2003. Breaking the cloud parameterization
deadlock. Bulletin of the American Meteorological Society Waliser, D.E., Seo, K.-W., Schubert, S., and Njoku, E.
84: 1547–1564. 2007. Global water cycle agreement in the climate models
assessed in the IPCC AR4. Geophysical Research Letters
Randall, D.A., Wood, R.A., Bony, S., Coleman, R., 34: 10.1029/2007GL030675.
Fichefet, T., Fyfe, J., Kattsov, V., Pitman, A., Shukla, J.,
Srinivasan, J., Stouffer, R.J., Sumi, A., and Taylor, K.E. Waliser, D.E., Li, J.-L.F., Woods, C.P., Austin, R.T.,
2007. Chapter 8: Climate Models and Their Evaluation. In: Bacmeister, J., Chern, J., Del Genio, A., Jiang, J.H.,
Climate Change 2007: The Physical Science Basis. Kuang, Z., Meng, H., Minnis, P., Platnick, S., Rossow,
Contribution of Working Group I to the Fourth Assessment W.B., Stephens, G.L., Sun-Mack, S., Tao, W.-K.,
Report of the Intergovernmental Panel on Climate Change. Tompkins, A.M., Vane, D.G., Walker, C., and Wu, D.
Cambridge University Press, Cambridge, United Kingdom. 2009. Cloud ice: A climate model challenge with signs and
expectations of progress. Journal of Geophysical Research
Richter, I. and Xie, S.-P. 2008. On the origin of equatorial 114: 10.1029/2008JD010015.
Atlantic biases in coupled general circulation models.
Climate Dynamics 31: 587–598. Wang, H. and Su, W. 2013. Evaluating and understanding
top of the atmosphere cloud radiative effects in
Riehl, H. and Malkus, J. 1958. On the heat balance in the Intergovernmental Panel on Climate Change (IPCC) Fifth
equatorial trough zone. Geophysica 6: 503–538. Assessment Report (AR5) Coupled Model Intercomparison
Senior, C.A. and Mitchell, J.F.B. 1993. CO2 and climate: Project Phase 5 (CMIP5) models using satellite
The impact of cloud parameterization. Journal of Climate observations. Journal of Geophysical Research:
6: 393–418. Atmospheres 118: 683–699.
Shupe, M.D. and Intrieri, J.M. 2004. Cloud radiative Weare, B.C. 2004. A comparison of AMIP II model cloud
forcing of the Arctic surface: The influence of cloud layer properties with ISCCP D2 estimates. Climate
properties, surface albedo, and solar zenith angle. Journal Dynamics 22: 281–292.
of Climate 17: 616–628. Webb, M.J., Senior, C.A., Sexton, D.M.H., Ingram, W.J.,
Siebesma, A.P., Jakob, C., Lenderink, G., Neggers, R.A.J., Williams, K.D., Ringer, M.A., McAvaney, B.J., Colman,
Teixeira, J., van Meijgaard, E., Calvo, J., Chlond, A., R., Soden, B.J., Gudgel, R., Knutson, T., Emori, S., Ogura,
Grenier, H., Jones, C., Kohler, M., Kitagawa, H., Marquet, T., Tsushima, Y., Andronova, N., Li, B., Musat, I., Bony,
P., Lock, A.P., Muller, F., Olmeda, D., and Severijns, C. S., and Taylor, K.E. 2006. On the contribution of local
2004. Cloud representation in general-circulation models feedback mechanisms to the range of climate sensitivity in
over the northern Pacific Ocean: A EUROCS two GCM ensembles. Climate Dynamics 27: 17–38.
intercomparison study. Quarterly Journal of the Royal Webster, P.J. and Stephens, G.L. 1984. Cloud-radiation
Meteorological Society 130: 3245–3267. feedback and the climate problem. In Houghton, J. (Ed.)
Soden, B.J. and Held, I.M. 2006. An assessment of climate The Global Climate. Cambridge University Press. 63–78.
feedbacks in coupled ocean-atmosphere models. Journal of Wyant, M., Khairoutdinov, M., and Bretherton, C. 2006.
Climate 19: 3354–3360. Climate sensitivity and cloud response of a GCM with a
Spencer, R.W. and Braswell, W.D. 2008.Potential biases in superparameterization. Geophysical Research Letters 33:
feedback diagnosis from observational data: A simple 10.1029/2005GL025464.
model demonstration. Journal of Climate 21: 5624–5628. Zeng, N., Dickinson, R.E., and Zeng, X. 1996. Climate
Spencer, R.W., Braswell, W.D., Christy, J.R., and Hnilo, J. impact of Amazon deforestation—A mechanistic model
2007. Cloud and radiation budget changes associated with study. Journal of Climate 9: 859–883.
tropical intraseasonal oscillations. Geophysical Research Zhang, M.H., Lin, W.Y., Klein, S.A., Bacmeister, J.T.,
Letters 34: L15707, doi:10.1029/2007GLO296998. Bony, S., Cederwall, R.T., Del Genio, A.D., Hack, J.J.,
Stephens, G.L. 2005. Clouds feedbacks in the climate Loeb, N.G., Lohmann, U., Minnis, P., Musat, I., Pincus, R.,
system: A critical review. Journal of Climate 18: 237–273. Stier, P., Suarez, M.J., Webb, M.J., Wu, J.B., Xie, S.C.,
Yao, M.-S., and Yang, J.H. 2005. Comparing clouds and
Stratton, R.A. and Stirling, A.J. 2012. Improving the their seasonal variations in 10 atmospheric general
diurnal cycle of convection in GCMs. Quarterly Journal of circulation models with satellite measurements. Journal of
the Royal Meteorological Society 138: 1121–1134. Geophysical Research 110: D15S02, doi:10.1029/2004
JD005021.
68
Zhang, Y., Klein, S.A., Boyle, J., and Mace, G.G. 2010. poorly in these several respects, Lebel et al. gave two
Evaluation of tropical cloud and precipitation statistics of main reasons: parameterization of rainfall processes
Community Atmosphere Model version 3 using CloudSat in the GCM was much too simple, and spatial
and CALIPSO data. Journal of Geophysical Research 115: resolution was much too coarse.
doi:10.1029/2009JD012006.
Lau et al. (2006) considered the Sahel drought of
Zhou, Y.P., Tao, W.-K., Hou, A.Y., Olson, W.S., Shie, C.- the 1970s–1990s to provide “an ideal test bed for
L., Lau, K.-M., Chou, M.-D., Lin, X., and Grecu, M. 2007. evaluating the capability of CGCMs [coupled general
Use of high-resolution satellite observations to evaluate circulation models] in simulating long-term drought,
cloud and precipitation statistics from cloud-resolving and the veracity of the models’ representation of
model simulations. Part I: South China Sea monsoon coupled atmosphere-ocean-land processes and their
experiment. Journal of the Atmospheric Sciences 64: 4309– interactions.” They chose to “explore the roles of sea
4329.
surface temperature coupling and land surface
processes in producing the Sahel drought in CGCMs
1.3.5 Precipitation that participated in the twentieth-century coupled
climate simulations of the Intergovernmental Panel on
1.3.5.1 Precipitation Climate Change [IPCC] Assessment Report 4,” in
Correctly simulating future precipitation has proven which the 19 CGCMs “are driven by combinations of
to be an extremely difficult task for modelers. One realistic prescribed external forcing, including
reason for the lack of success in this area is anthropogenic increase in greenhouse gases and
inadequate model resolution on both vertical and sulfate aerosols, long-term variation in solar radiation,
horizontal spatial scales, a limitation that forces and volcanic eruptions.”
climate modelers to parameterize the large-scale The climate scientists found “only eight models
effects of processes (such as deep convection, which produce a reasonable Sahel drought signal, seven
is the source of most precipitation) that occur on models produce excessive rainfall over [the] Sahel
smaller scales than the models are capable of during the observed drought period, and four models
simulating. But there are other problems as well that show no significant deviation from normal.” In
result in vast differences between model projections addition, they report, “even the model with the
and real-world observations. This section documents highest skill for the Sahel drought could only simulate
such problems and variances as they relate to model the increasing trend of severe drought events but not
projections of precipitation. the magnitude, nor the beginning time and duration of
In an early study of the subject, Lebel et al. the events.” All 19 of the CGCMs employed in the
(2000) compared rainfall simulations produced by a IPCC’s Fourth Assessment Report, in other words,
GCM with real-world observations from West Africa failed to adequately simulate the basic characteristics
for the period 1960–1990. Their analysis revealed the of “one of the most pronounced signals of climate
model output was affected by a number of temporal change” of the past century—as defined by its start
and spatial biases that led to significant differences date, severity and duration.”
between observed and modeled data. The simulated Writing in Science, Wentz et al. (2007) noted the
rainfall totals, for example, were significantly greater Coupled Model Intercomparison Project, as well as
than what was typically observed, exceeding real- various climate modeling analyses, predicted an
world values by 25 percent during the dry season and increase in precipitation on the order of 1 to 3 percent
75 percent during the rainy season. In addition, the per °C of surface global warming. They decided to
seasonal cycle of precipitation was not well see what had happened in the real world in this regard
simulated, as the researchers found the simulated over the prior 19 years (1987–2006) of supposedly
rainy season began too early and the increase in unprecedented global warming, when data from the
precipitation was not rapid enough. Shortcomings Global Historical Climatology Network and satellite
were also evident in the GCM’s inability to accurately measurements of the lower troposphere indicated
simulate convective rainfall events, as it typically there had been a global temperature rise on the order
predicted too much precipitation. Furthermore, it was of 0.20°C per decade.
found “inter-annual variability [was] seriously Using satellite observations obtained from the
disturbed in the GCM as compared to what it [was] in Special Sensor Microwave Imager (SSM/I), the four
the observations.” As for why the GCM performed so Remote Sensing Systems scientists derived
precipitation trends for the world’s oceans over this
69
period. Using data obtained from the Global trends are substantially larger in magnitude than
Precipitation Climatology Project acquired from both present-day simulations and projections into the 21st
satellite and rain gauge measurements, they derived century,” especially in the case of the descending
precipitation trends for Earth’s continents. Combining regions. More specifically, for the tropics “the GPCP
the results of these two endeavors, they derived a trend is about 2–3 times larger than the model
real-world increase in precipitation on the order of 7 ensemble mean trend, consistent with previous
percent per °C of surface global warming, somewhere findings (Wentz et al., 2007) and also supported by
between 2.3 and 7.0 times larger than what is the analysis of Yu and Weller (2007),” who
predicted by state-of-the-art climate models. additionally contended “observed increases of
How was this discrepancy to be resolved? Wentz evaporation over the ocean are substantially greater
et al. conclude the only way to bring the two results than those simulated by climate models.” In addition,
into harmony was for there to have been a 19-year Allan and Soden note “observed precipitation changes
decline in global wind speeds. But when looking at over land also appear larger than model simulations
the past 19 years of SSM/I wind retrievals, they found over the 20th century (Zhang et al., 2007).”
just the opposite, an increase in global wind speeds. Noting the difference between the models and
In quantitative terms, the two results were about as real-world measurements “has important implications
opposite as they could possibly be: “when averaged for future predictions of climate change,” Allan and
over the tropics from 30°S to 30°N, the winds Soden state “the discrepancy cannot be explained by
increased by 0.04 m s-1 (0.6 percent) decade-1, and changes in the reanalysis fields used to subsample the
over all oceans the increase was 0.08 m s-1 (1.0 observations but instead must relate to errors in the
percent) decade-1,” while global coupled ocean- satellite data or in the model parameterizations.” This
atmosphere models or GCMs, in their words, “predict dilemma also was faced by Wentz et al. (2007), and
that the 1987-to-2006 warming should have been they too state the resolution of the issue “has
accompanied by a decrease in winds on the order of enormous impact” and likewise conclude the
0.8 percent decade-1.” questions raised by the discrepancy “are far from
Wentz et al. state, “the reason for the discrepancy being settled.”
between the observational data and the GCMs is not According to Lin (2007), “a good simulation of
clear.” They also observe this dramatic difference tropical mean climate by the climate models is a
between the real world of nature and the virtual world prerequisite for their good simulations/predictions of
of climate modeling “has enormous impact” and the tropical variabilities and global teleconnections,” but
questions raised by the discrepancy “are far from “unfortunately, the tropical mean climate has not been
being settled.” well simulated by the coupled general circulation
In a separate paper published that year, Allan and models (CGCMs) used for climate predictions and
Soden (2007) quantified trends in precipitation within projections.” They note “most of the CGCMs produce
ascending and descending branches of the planet’s a double-intertropical convergence zone (ITCZ)
tropical circulation and compared their results with pattern” and they acknowledge “a synthetic view of
simulations of the present day and projections of the double-ITCZ problem is still elusive.”
future changes provided by up to 16 state-of-the-art Lin analyzed tropical mean climate simulations of
climate models. The precipitation data for this the 20-year period 1979–1999 provided by 22 IPCC
analysis came from the Global Precipitation Fourth Assessment Report CGCMs, together with
Climatology Project (GPCP) of Adler et al. (2003) concurrent Atmospheric Model Intercomparison
and the Climate Prediction Center Merged Analysis Project (AMIP) runs from 12 of them. This revealed,
of Precipitation (CMAP) data of Xie and Arkin in Lin’s words, that “most of the current state-of-the-
(1998) for the period 1979–2006, while for the period art CGCMs have some degree of the double-ITCZ
1987–2006 the data came from the monthly mean problem, which is characterized by excessive
intercalibrated Version 6 Special Sensor Microwave precipitation over much of the Tropics (e.g., Northern
Imager (SSM/I) precipitation data described by Hemisphere ITCZ, Southern Hemisphere SPCZ
Wentz et al. (2007). [South Pacific Convergence Zone], Maritime
The researchers report “an emerging signal of Continent, and equatorial Indian Ocean), and often
rising precipitation trends in the ascending regions associated with insufficient precipitation over the
and decreasing trends in the descending regions are equatorial Pacific,” as well as “overly strong trade
detected in the observational datasets,” but “these winds, excessive LHF [latent heat flux], and
70
insufficient SWF [shortwave flux], leading to forecasts be provided that would be useful for
significant cold SST (sea surface temperature) bias in decision makers,” a quest they frankly state “may
much of the tropical oceans.” prove to be elusive.”
Lin further notes “most of the models also Anagnostopoulos et al. (2010) compared
simulate insufficient latitudinal asymmetry in observed versus modeled precipitation values over the
precipitation and SST over the eastern Pacific and twentieth century for 55 locations across the globe.
Atlantic Oceans,” and “the AMIP runs also produce Their results indicate the six models investigated
excessive precipitation over much of the Tropics (three from the IPCC’s Third Assessment and three
including the equatorial Pacific, which also leads to from its Fourth Assessment) reproduced only poorly
overly strong trade winds, excessive LHF, and the observed precipitation values over the period of
insufficient SWF.” All of this suggests “the excessive study. In far too many instances the models showed a
tropical precipitation is an intrinsic error of the rise in precipitation when observed values actually
atmospheric models,” Lin writes, adding, “over the fell, or vice versa. The models fared worse when a
eastern Pacific stratus region, most of the models similar analysis was conducted in the aggregate for
produce insufficient stratus-SST feedback associated the entire conterminous United States. Model output
with insufficient sensitivity of stratus cloud amount to differed “substantially” from the observed time series,
SST.” with annual precipitation values overestimating
With the solutions to all of these longstanding observed values by up to 300 mm, or 40 percent. The
problems continuing to remain “elusive,” and with authors also state the results from the three models
Lin suggesting the sought-for solutions are in fact used in the IPCC’s Fourth Assessment Report were
prerequisites for “good simulations/predictions” of “no better” than the three models used in the IPCC’s
future climate, there is significant reason to conclude Third Assessment Report.
that current state-of-the-art CGCM predictions of Ault et al. (2012) write “the last generation of
CO2-induced global warming should not be models, those comprising [the] Climate Model
considered reliable. Intercomparison Project III (CMIP3) archive, was
Lavers et al. (2009) examined the predictive skill unable to capture key statistics characterizing decadal
of eight seasonal climate forecast models developed to multidecadal (D2M) precipitation fluctuations,”
at various European climate centers. Specifically, noting specifically “CMIP3 simulations
they assessed the predictability of monthly overestimated the magnitude of high frequency
precipitation “retrospective forecasts” or hindcasts, fluctuations and consequently underestimated the risk
composed of multiple nine-month projections of future decadal-scale droughts.” Since “a new
initialized during each month of the year over the generation of coupled general circulation models
period 1981–2001. They compared the projections (GCMs) has been developed and made publicly
against real-world precipitation values obtained from available as part of the Climate Model
Global Precipitation Climatology Center data. In Intercomparison Project 5 (CMIP5) effort,” it is
addition, they conducted a virtual-world analysis, critical, the note, “to evaluate the ability of these
where the output of one of the models was arbitrarily models to simulate realistic 20th century variability
assumed to be the truth and the average of the rest of regionally and across a variety of timescales.”
the models was assumed to be the predictor. Using gridded (2.5 x 2.5) version 4 reanalysis
These analyses indicate that in the virtual world product data made available to them by the Global
of the climate models, there was quite good skill over Precipitation Climatology Centre (Rudolf et al.,
the first two weeks of the forecast, when the spread of 2005)—which spanned the period January 1901
ensemble model members was small, but there was a through December 2007—Ault et al. assessed the
large drop off in predictive skill in the second 15-day magnitude of D2M variability in new CMIP5
period. Things were even worse in the real world, simulations. The three U.S. researchers report their
where the models had negligible skill over land at a results suggest “CMIP5 simulations of the historical
31-day lead time, which the researchers describe as era (1850–2005) underestimate the importance [of]
being “a relatively short lead time in terms of D2M variability in several regions where such
seasonal climate prediction.” The three researchers behavior is prominent and linked to drought,” namely,
conclude that given the lack of real-world skill “northern Africa (e.g., Giannini et al., 2008),
demonstrated by models, “it appears that only through Australia (Cai et al., 2009; Leblanc et al., 2012),
significant model improvements can useful long-lead western North America (Seager, 2007; Overpeck and
71
Udall, 2010), and the Amazon (Marengo et al., radiative heating of the atmosphere but would be
2011).” expected to impact the circulation, and possibly even
Ault et al. further state “the mismatch between the manner the model adjusts to external forcings
20th century observations and simulations suggests such as increasing greenhouse gases.” In addition,
that model projections of the future may not fully they note, since the “models are tuned to get the right
represent all sources of D2M variations,” noting, “if TOA radiation balance, the implications here are that
observed estimates of decadal variance are accurate, without considering the ice in precipitating
then the current generation of models depict D2M hydrometeors explicitly, the models will be getting
precipitation fluctuations that are too weak, implying the right result (i.e., TOA balance) for the wrong
that model hindcasts and predictions may be unable to reasons,” and “in doing so, there are likely to be
capture the full magnitude of realizable D2M compensating errors in other quantities such as cloud
fluctuations in hydroclimate.” As a result, “the risk of cover, cloud particle effective radius and/or cloud
prolonged droughts and pluvials in the future may be mass.”
greater than portrayed by these models.” In another paper, Soncini and Bocchiola (2011)
Aaliser et al. (2011) write, “key to the proper use note “General Circulation Models (GCMs) are widely
of satellite retrievals in the evaluation of modeled adopted tools to achieve future climate projections.”
cloud ice and liquid is that many global climate model However, they write, “one needs to assess their
representations ignore or diagnostically treat the accuracy, which is only possible by comparison of
falling hydrometeor components (e.g., rain and snow) GCMs’ control runs against past observed data,”
and only consider—for the purposes of radiation which they proceeded to do in the case of snowfall
calculations—the ‘suspended’ component of water regimes within the Italian Alps. Specifically, the two
that the model deems ‘clouds.’” They state “the Italian researchers investigated the accuracy of
variations in the annual mean integrated ice water simulations of snowfall throughout the Italian Alps
path and liquid water path between global climate provided by two GCMs (HadCM3, CCSM3), which
models contributing to the IPCC AR4 range over two are included within the family of models employed by
orders of magnitude,” citing Li et al. (2008) and the IPCC. This was done by comparing the models’
Waliser et al. (2009). Employing estimates of cloud output with a set of comprehensive ground data
and precipitating ice mass and characterizations of its obtained from some 400 snow-gauging stations
vertical structure supplied by CloudSat retrievals, located within the region of interest for the period
Waliser et al. set out to perform radiative transfer 1990–2009.
calculations “to examine the impact of excluding In examining the model versus observation
precipitating ice on atmospheric radiative fluxes and comparison, Soncini and Bocchiola determined “the
heating rates.” investigated GCMs provide poor depiction of the
According to the four researchers, the exclusion snowfall timing and amount upon the Italian Alps,”
of precipitating ice “can result in underestimates of noting, in fact, the HadCM3 model actually “displays
the reflective shortwave flux at the top of the considerable snowfall during summer,” which they
atmosphere (TOA) and overestimates of the down- indicate “is clearly not supported by ground data.” In
welling surface shortwave and emitted TOA addition, they report obtaining “contrasting results
longwave flux, with the differences being about 5–10 between the two models,” with HadCM3 providing
Wm-2 in the most convective and rainfall intensive substantially constant volumes of snow received over
areas.” In addition, they report, “there are also time and CCSM3 projecting decreasing snowfall
considerable differences (~10-25%) in the vertical volumes. “Overall,” in the words of the two
profiles of shortwave and longwave heating, resulting researchers, “given the poor depiction of snowfall by
in an overestimation (~up to 10%) of the integrated the GCMs here tested, we suggest that care should be
column cooling.” And they state “the magnitude of taken when using their outputs for predictive
these potential errors is on the order of the radiative purposes.”
heating changes associated with a doubling of Also investigating snowfall were Salzmann and
atmospheric carbon dioxide.” Mearns (2012), who note “climate impact
With respect to the implications of their findings, assessments require primarily regional- to local-scale
Waliser et al. state “when the above results are climate data for the past and the present and scenarios
considered in the context of a climate model for the future,” stating, with respect to the future, that
simulation, the changes would not only impact the “regional climate models (RCMs) are among the most
72
promising tools to simulate climate on the regional Alps was not enhanced during warmer (or drier)
scale.” However, they also note “the effective benefit periods,” Stewart et al. acknowledge “evidence could
of each of these RCMs and their ensembles for not be found that summer-autumn floods would
specific climate impact assessments remains to be increase in the Eastern Swiss Alps in a warmer
proven for individual impact studies.” Salzmann and climate of the 21st century,” pretty much debunking
Mearns explore this issue within the context of the the projections of regional climate models that have
North American Regional Climate Change Program suggested otherwise.
(NARCCAP; Mearns et al., 2009) with regard to the Soares et al. (2012) note “Regional Climate
seasonal snow regime in the Upper Colorado River Models (RCMs) are increasingly used to assess the
Basin. They compared NARCCAP results with in situ impact of climate change at regional and local scales
observations and data obtained from various (Giorgi and Mearns, 1999; Wang et al., 2004;
reanalysis projects. Christensen and Christensen, 2007),” because “in
The two researchers at the National Center for regions where local features affecting the atmospheric
Atmospheric Research in Boulder, Colorado (USA) flow, such as topography and coastal processes, are
report—quite bluntly and to the point: “[T]he RCMs prevalent, finer resolution simulations with state-of-
are generally too dry, too warm, simulate too little the-art mesoscale models are required to reproduce
snow water equivalent, and have a too-short snow observed weather and climate (Mass et al., 2002;
cover duration with a too-late start and a too-early end Salathe et al., 2008).” They utilized “a new data set of
of a significant snow cover.” To these problems they daily gridded observations of precipitation, computed
add, “attributing the found biases to specific features from over 400 stations in Portugal, to assess the
of the RCMs remains difficult or even impossible performance of 12 regional climate models at 25-km
without detailed knowledge of the physical and resolution, from the ENSEMBLES set, all forced by
technical specification of the models.” ERA-40 boundary conditions, for the 1961–2000
Stewart et al. (2011) write, “regional climate period,” while “standard point error statistics,
models project that future climate warming in Central calculated from grid point and basin aggregated data,
Europe will bring more intense summer-autumn and precipitation related climate indices are used to
heavy precipitation and floods as the atmospheric analyze the performance of the different models in
concentration of water vapor increases and cyclones representing the main spatial and temporal features of
intensify,” citing the studies of Arnell and Liu (2001), the regional climate, and its extreme events.”
Christensen and Christensen (2003), and Kundzewicz Although the five Portuguese researchers say the
et al. (2005). In an exercise designed to assess the models achieved what they called a “good
reasonableness of these projections, Stewart et al. representation” of the features listed above, they also
derived “a complete record of paleofloods, regional list a number of less-than-hoped-for results: (1) “10 of
glacier length changes (and associated climate the 12 analyzed models under-predict Portuguese
phases) and regional glacier advances and retreats precipitation,” (2) “half of the models under-represent
(and associated climate transitions) … from the observed variability of daily precipitation,” (3)
varved sediments of Lake Silvaplana (ca. 1450 BC– “models were found to underestimate the number of
AD 420; Upper Engadine, Switzerland),” while wet days,” (4) “grid point percentiles of precipitation
indicating “these records provide insight into the are generally under-predicted,” (5) “in all cases, there
behavior of floods (i.e. frequency) under a wide range is a significant model spread,” (6) “the 95th percentile
of climate conditions.” is under-predicted by all models in most of the
The five researchers report uncovering pertinent country,” and (7) “there is an important model spread
data from the period they investigated that suggests in all analyzed variables.” Such findings led Soares et
“an increase in the frequency of paleofloods during al. to state in their concluding paragraph, “the present
cool and/or wet climates and windows of cooler June- results suggest that there is still some way to go in
July-August temperatures,” which further suggests— this research.”
as they also note—the frequency of flooding “was Focusing on a nearby region, Kelley et al. (2012)
reduced during warm and/or dry climates.” In state “winter and summer Mediterranean precipitation
reiterating that “the findings of this study suggest that climatology and trends since 1950 as simulated by the
the frequency of extreme summer-autumn newest generation of global climate models, the
precipitation events (i.e. flood events) and the Coupled Model Intercomparison Project phase 5
associated atmospheric pattern in the Eastern Swiss (CMIP5), [were] evaluated with respect to
73
observations and the previous generation of models al. therefore state “the detection of an outright sign
(CMIP3) used in the Intergovernmental Panel on mismatch of observed and projected trends in autumn
Climate Change Fourth Assessment Report,” with the and late summer, leads us to conclude that the
objective of determining “to what extent we can trust recently observed trends cannot be used as an
the multi-model mean (MMM) trends as representing illustration of plausible future expected change in the
the externally forced trends.” Upon analysis, Kelly et Mediterranean region,” once again illustrating the
al. determined “the Mediterranean precipitation many problems besetting even the best of climate
trends of the last half century in the CMIP5 MMMs models.
and the observations differ significantly, particularly In another paper, Miao et al. (2012) assessed the
in winter and over the northern Mediterranean performance of the AR4 GCMs (CMIP3 models) in
region.” The CMIP5 MMM trend, for example, simulating precipitation and temperature in China
“indicates a modest drying throughout the seasonal from 1960 to 1999 by comparing the model
cycle, with the strongest drying in the March, April simulations with observed data, using “system bias
and May spring season.” The observed trend, on the (B), root-mean-square error (RMSE), Pearson
other hand, “shows a predominantly winter drying,” correlation coefficient (R) and Nash-Sutcliffe model
and they state “it is not entirely clear what causes this efficiency (E)” as evaluation metrics. The four
discrepancy.” researchers conclude certain of the CMIP3 models
Although the four researchers report “there is a “are unsuitable for application to China, with little
modest improvement of the CMIP5 climatology over capacity to simulate the spatial variations in climate
CMIP3,” it would appear “modest” is too generous a across the country,” adding all of them “give
word to describe what was accomplished between the unsatisfactory simulations of the inter-annual
development of the CMIP3 and CMIP5 models, temporal variability.” In addition, they found “each
particularly given their conclusion that the slight AR4 GCM performs differently in different regions of
improvement they detected may have been due China.” In light of these findings, Miao et al.
merely to the “improved horizontal resolution” of the conclude “the inter-annual simulations (temperature
CMIP5 models. The ultimate implication of Kelly et and precipitation) by AR4 GCMs are not suitable for
al.’s findings is presented in the concluding paragraph direct application,” and “caution should be applied
of their paper, where they state their findings when using outputs from the AR4 GCMs in
“reinforce the need for further research and better hydrological and ecological assessments” because of
understanding of the mechanisms driving the region’s their “poor performance.”
hydroclimate.” Kataoka et al. (2012) note the Indian Ocean
Also working in the Mediterranean region, Subtropical Dipole (IOSD; Behera et al., 2000;
Barkhordarian et al. (2013) assessed the role of Behera and Yamagata, 2001) is “one of the climate
anthropogenic forcing due to greenhouse gases and modes that generate climate variations in the Southern
sulphate aerosols (GS) in recently observed Hemisphere,” having “a great impact on the
precipitation trends over the Mediterranean region in surrounding countries through its influence on the
order to determine whether the observed trends over rainfall (Behera and Yamagata, 2001; Reason, 2001;
the period 1966–2005 (over land) and 1979–2008 Washington and Preston, 2006).” This mode is
(over land and sea) “are consistent with what 22 characterized by “a dipole pattern in the sea surface
models project as response of precipitation to GS temperature anomaly in the southern Indian Ocean
forcing,” where “significance is estimated using with a warm (cold) southwestern pole and cold
9,000-year control runs derived from the CMIP3 (warm) northeastern pole during austral summer.”
archive.” “[S]ince southern Africa is one of the most vulnerable
The three researchers discovered “the observed regions to abnormal weather events,” they note, “an
trends are markedly inconsistent with expected accurate prediction of the IOSD together with its
changes due to GS forcing,” as “observed changes are influence on rainfall is necessary to mitigate the
several times larger than the projected response to GS impacts.” Using observational data and mathematical
forcing in the models.” But they state “the most outputs from 22 coupled general circulation models
striking inconsistency” was “the contradiction (CGCMs) submitted to the World Climate Research
between projected drying and the observed increase in Programme’s Coupled Model Intercomparison
precipitation in late summer and autumn.” Coming to Project phase 3 (CMIP3), Kataoka et al. proceeded to
a conclusion that cannot be avoided, Barkhordarian et assess each model’s ability to simulate the IOSD and
74
its influence on rainfall anomalies over southern precipitation tends to be above average during the
Africa. PDO warm phase, while precipitation during the PDO
In discussing their findings, the four Japanese cold phase is below average,” and [6] “most of the
researchers report the location and orientation of sea considered GCMs failed to reproduce similar
surface temperature anomaly poles “differ variability.” Their wavelet analysis revealed [7] “even
considerably” from one model to another, owing the successful GCMs on reproducing the low-
primarily to model biases in sea level pressure frequency variability associated with ENSO and
anomalies. This finding, as they describe it, supports PDO, showed inconsistency in the occurrence or
“the earlier results of Morioka et al. (2010) based on timing of 2–8-year bands.” Jiang et al. conclude their
an ocean general circulation model.” This problem, in “comparative analyses suggest that current
their words, “may partly explain the poor skills of GCMs/RCMs do not adequately capture multi-scale
CGCMs in simulating the influence of the IOSD on temporal variability of precipitation,” and, therefore,
the rainfall anomalies.” In addition, they state “some “using GCM/RCM output to conduct future flood
models fail to simulate the statistical relation between projections is not creditable.”
the positive (negative) rainfall anomaly and La Niña
(El Niño).” The authors conclude “more accurate References
simulation of the IOSD as well as the influence of the
ENSO is necessary to improve the seasonal prediction Adler, R.F., Huffman, G.J., Chang, A., Ferraro, R., Xie, P.,
of southern African rainfall.” Janowiak, J., Rudolf, B., Schneider, U. Curtis, S., Bolvin,
According to Jiang et al. (2013), “multi-scale D., Gruber, A., Susskind, J. Arkin, P., and Nelkin, E. 2003.
temporal variability of precipitation has an The version-2 Global Precipitation Climatology Project
(GPCP) monthly precipitation analysis (1979-present).
established relationship with floods and droughts,”
Journal of Hydrometeorology 4: 1147–1167.
and GCMs can provide “important avenues to climate
change impact assessment and adaptation planning,” Allan, R.P. and Soden, B.J. 2007. Large discrepancy
but only if they possess an “ability to capture the between observed and simulated precipitation trends in the
climatic variability at appropriate scales.” Jiang et al. ascending and descending branches of the tropical
assessed “the ability of 16 GCMs from the Bias circulation. Geophysical Research Letters 34:
Corrected and Downscaled (BCSD) World Climate 10.1029/2007GL031460.
Research Program’s (WCRP’s) Coupled Model Anagnostopoulos, G.G., Koutsoyiannis, D., Christofides,
Intercomparison Project Phase 3 (CMIP3) projections A., Efstradiadis, A., and Mamassis, N. 2010. A comparison
and 10 Regional Climate Models (RCMs) that of local and aggregated climate model outputs with
participated in the North American Regional Climate observed data. Hydrological Sciences Journal 55: 1094–
Change Assessment Program (NARCCAP) to 1110.
represent multi-scale temporal variability determined Arnell, N. and Liu, C. 2001. Hydrology and water
from observed station data.” They focused on four resources. In: McCarthy, J.J., Canziani, O.F., Leary, N.A.,
regions in the Southwest United States—Los Angeles, Dokken, D.J., and White, K.S. (Eds.), Climate Change
Las Vegas, Tucson, and Cimarron—because these 2001: Impacts, Adaptation and Vulnerability. Contribution
places “represent four different precipitation regions of Working Group II to the Third Assessment Report of the
classified by clustering method.” They specifically Intergovernmental Panel on Climate Change. Cambridge
investigated “how storm properties and seasonal, University Press, Cambridge, United Kingdom.
inter-annual, and decadal precipitation variabilities Ault, T.R., Cole, J.E., and St. George, S. 2012. The
differed between GCMs/RCMs and observed records amplitude of decadal to multidecadal variability in
in these regions.” precipitation simulated by state-of-the-art climate models.
The four U.S. researchers report “RCMs tend to Geophysical Research Letters 39: 10.1929/2012GL053424.
simulate [1] longer duration, [2] shorter inter-storm
periods, and [3] lower storm intensity than observed.” Barkhordarian, A., von Storch, H., and Bhend, J. 2013. The
expectation of future precipitation change over the
Moreover, they state [4] “RCMs fail to simulate high
Mediterranean region is different from what we observe.
average storm intensity during the summer period as Climate Dynamics 40: 225–244.
seen in observed precipitation records.” They also say
[5] bias-corrected and downscaled GCMs “lack the Behera, S.K. and Yamagata, T. 2001. Subtropical SST
ability to reproduce observed monthly precipitation dipole events in the southern Indian Ocean. Geophysical
patterns.” In addition, they note “observed Research Letters 28: 327–330.
75
Behera, S.K., Salvekar, P.S., and Yamagata, T. 2000. Leblanc, M., Tweed, S., Van Dijk, A., and Timbal, B.
Simulation of interannual SST variability in the tropical 2012. A review of historic and future hydrological changes
Indian Ocean. Journal of Climate 13: 3487–3499. in the Murray-Darling Basin. Global and Planetary
Change 80-81: 226–246.
Cai, W., Cowan, T., Briggs, P., and Raupach, M. 2009.
Rising temperature depletes soil moisture and exacerbates Li, F.-F., Waliser, D.E., Woods, C., Teixeira, J.,
severe drought conditions across southeast Australia. Bacmeister, J., Chern, J., Shen, B.W., Tompkins, A., and
Geophysical Research Letters 36: 10.1029/2009GL040334. Kohler, M. 2008. Comparisons of satellites liquid water
estimates with ECMWF and GMAO analyses, 20th century
Christensen, J.H. and Christensen, O.B. 2003. Climate IPCC AR4 climate simulations, and GCM simulations.
modeling: severe summertime flooding in Europe. Nature Geophysical Research Letters 35: 10.1029/2008GL035427.
421: 805–806.
Lin, J.-L. 2007. The double-ITCZ problem in IPCC AR4
Christensen, J.H. and Christensen, O.B. 2007. A summary coupled GCMs: Ocean-atmosphere feedback analysis.
of the PRUDENCE model projections of changes in Journal of Climate 20: 4497–4525.
European climate by the end of this century. Climatic
Change 81: 7–30. Marengo, J.A., Tomasella, J., Alves, L.M., Soares, W.R.
and Rodriguez, D.A. 2011. The drought of 2010 in the
Giannini, A., Biasutti, M., Held, I.M., and Sobel, A.H. context of historical droughts in the Amazon region.
2008. A global perspective on African climate. Climatic Geophysical Research Letters 38: 10.1029/2011GL047436.
Change 90: 359–383.
Mass, C.F., Ovens, D., Westrick, K., and Colle, B.A. 2002.
Giorgi, F. and Mearns, L.O. 1999. Introduction to special Does increasing horizontal resolution produce more skillful
section: Regional climate modeling revisited. Journal of forecasts? Bulletin of the American Meteorological Society
Geophysical Research 104: 6335–6352. 83: 407–430.
Jiang, P., Gautam, M.R., Zhu, J., and Yu, Z. 2013. How Mearns, L.O., Gutowski, W., Jones, R., Leung, R.,
well do the GCMs/RCMs capture the multi-scale temporal McGinnis, S., Nunes, A., and Qian, Y. 2009. A regional
variability of precipitation in the southwestern United climate change assessment program for North America.
States. Journal of Hydrology 479: 75–85. EOS, Transactions of the American Geophysical Union 90:
Kataoka, T., Tozuka, T., Masumoto, Y., and Yamagata, T. 311.
2012. The Indian Ocean subtropical dipole mode simulated Miao, C., Duan, Q., Yang, L., and Borthwick, A.G.L. 2012.
in the CMIP3 models. Climate Dynamics 39: 1385–1399. On the applicability of temperature and precipitation data
Kelley, C., Ting, M., Seager, R., and Kushnir, Y. 2012. from CMIP3 for China. PLoS ONE 7: e44659.
Mediterranean precipitation climatology, seasonal cycle, Morioka, Y., Tozuka, T., and Yamagata, T. 2010. Climate
and trend as simulated by CMIP5. Geophysical Research variability in the southern Indian Ocean as revealed by self-
Letters 39: 10.1029/2012GL053416. organizing maps. Climate Dynamics 35: 1059–1072.
Kundzewicz, Z.W., Ulbrich, U., Brucher, T., Graczyk, D., Overpeck, J. and Udall, B. 2010. Dry times ahead. Science
Kruger, A., Leckebusch, G.C., Menzel, L., Pinskwar, I., 328: 1642–1643.
Radziejewski, M., and Szwed, M. 2005. Summer floods in
Central Europe—climate change track? Natural Hazards Reason, C.J.C. 2001. Subtropical Indian Ocean SST dipole
36: 165–189. events and southern African rainfall. Geophysical Research
Letters 28: 2225–2227.
Lau, K.M., Shen, S.S.P., Kim, K.-M., and Wang, H. 2006.
A multimodel study of the twentieth-century simulations of Rudolf, B., Beck, C., Grieser, J., and Schneider, U. 2005.
Sahel drought from the 1970s to 1990s. Journal of Global Precipitation Analysis Products of Global
Geophysical Research 111: 10.1029/2005JD006281. Precipitation Climatology Centre (GPCC). Technical
Report. Dtsch. Wetterdienst, Offenbach, Germany.
Lavers, D., Luo, L., and Wood, E.F. 2009. A multiple
model assessment of seasonal climate forecast skill for Salathe, E.P., Steed, R., Mass, C.F., and Zahn, P.H. 2008.
applications. Geophysical Research Letters 36: A high-resolution climate model for the United States
10.1029/2009GL041365. Pacific Northwest: Mesoscale feedbacks and local
responses to climate change. Journal of Climate 21: 5708–
Lebel, T., Delclaux, F., Le Barbé, L., and Polcher, J. 2000. 5726.
From GCM scales to hydrological scales: rainfall
variability in West Africa. Stochastic Environmental Salzmann, N. and Mearns, L.O. 2012. Assessing the
Research and Risk Assessment 14: 275–295. performance of multiple regional climate model
simulations for seasonal mountain snow in the Upper
76
Colorado River Basin. Journal of Hydrometeorology 13: estimates from satellite-observed outgoing longwave
539–556. radiation. Journal of Climate 11: 137–164.
Seager, R. 2007. The turn of the century North American Yu, L. and Weller, R.A. 2007. Objectively analyzed air-sea
drought: Global context, dynamics, and past analogs. heat fluxes for the global ice-free oceans (1981–2005).
Journal of Climate 20: 5527–5552. Bulletin of the American Meteorological Society 88: 527–
539.
Soares, P.M.M., Cardoso, R.M., Miranda, P.M.A., Viterbo,
P., and Belo-Pereira, M. 2012. Assessment of the Zhang, X., Zwiers, F.W., Hegerl, G.C., Lambert, F.H.,
ENSEMBLES regional climate models in the Gillett, N.P., Solomon, S., Stott, P.A., and Nozawa, T.
representation of precipitation variability and extremes 2007. Detection of human influence on twentieth-century
over Portugal. Journal of Geophysical Research 117: precipitation trends. Nature 448: 461–465.
10.1029/2011JD016768.
Solomon, S., Qin, D., Manning, M., Marquis, M., Averyt, 1.4.5.2 Monsoons
K., Tignor, M.B., Miller Jr., H.L., and Chen, Z. (Eds.). Climate modelers have long struggled to simulate
2007. Climate Change 2007: The Physical Science Basis.
seasonal monsoons despite their significance in global
Cambridge University Press, Cambridge, United Kingdom.
climate. As mentioned in a prior section of this
Soncini, A. and Bocchiola, D. 2011. Assessment of future chapter, when the 2004 summer monsoon season of
snowfall regimes within the Italian Alps using general India experienced a 13 percent deficit not predicted
circulation models. Cold Regions Science and Technology by either empirical or dynamical models used in
68: 113–123. making rainfall forecasts, Gadgil et al. (2005) decided
Stewart, M.M., Grosjean, M., Kuglitsch, F.G., to perform a historical analysis of the models’ skills
Nussbaumer, S.U., and von Gunten, L. 2011. over the period 1932 to 2004. They found despite
Reconstructions of late Holocene paleofloods and glacier numerous model changes and an ever-improving
length changes in the Upper Engadine, Switzerland (ca. understanding of monsoon variability, Indian
1450 BC-AD 420). Palaeogeography, Palaeoclimatology, monsoon model forecasting skill had not improved
Palaeoecology 311: 215–223. since 1932. Large differences often were observed
Waliser, D.E., Li, J.-L.F., L’Ecuyer, T.S., and Chen, W.-T. when comparing monsoon rainfall measurements with
2011. The impact of precipitating ice and snow on the empirical model predictions; and the models often
radiation balance in global climate models. Geophysical failed to correctly predict even the sign of the
Research Letters 38: 10.1029/2010GL046478. precipitation anomaly.
Dynamical models fared even worse. In
Waliser, D.E., Li, J.-L.F., Woods, C.P., Austin, R.T., comparing observed versus predicted monsoon
Bacmeister, J., Chern, J., Del Genio, A., Jiang, J.H.,
rainfall from 20 atmospheric general circulation
Kuang, Z., Meng, H., Minnis, P., Platnick, S., Rossow,
W.B., Stephens, G.L., Sun-Mack, S., Tao, W.-K., models and one supposedly superior coupled
Tompkins, A.M., Vane, D.G., Walker, C., and Wu, D. atmosphere-ocean model, Gadgil et al. report none
2009. Cloud ice: A climate model challenge with signs and was able “to simulate correctly the interannual
expectations of progress. Journal of Geophysical Research variation of the summer monsoon rainfall over the
114: 10.1029/2008JD010015. Indian region.” Like the empirical models, they
frequently failed to simulate not only the magnitude
Wang, Y., Leung, L.R., McGregor, J.L., Lee, D.K., Wang,
but also the sign of the real-world rainfall anomalies.
W.C., Ding, Y., and Kimura, F. 2004. Regional climate
modeling: Progress, challenges and prospects. Journal of Few improvements seem to have occurred since that
the Meteorological Society of Japan 82: 1599–1628. time.
Zhang et al. (2012) report “the Asian monsoon is
Washington, R. and Preston, A. 2006. Extreme wet years a major component in the global climate system,
over southern Africa: Role of Indian Ocean sea surface characterized by remarkable seasonal and inter-annual
temperatures. Journal of Geophysical Research 111: rainfall variations which have significant social and
10.1029/2005JD006724.
economic influences on large populations in the
Wentz, F.J., Ricciardulli, L., Hilburn, K., and Mears, C. region,” the complexity of which, in their words, “is
2007. How much more rain will global warming bring? never overstated.” Using daily perceptible water and
Science 317: 233–235. 850 hPa monsoon wind data, which in the words of
Zhang et al. “represent large-scale moisture and
Xie, P. and Arkin, P.A. 1998. Global monthly precipitation
dynamic conditions for monsoon development,” the
77
four scientists analyzed “potential changes in Asian reanalysis.” With both of the world’s most advanced
monsoon onset, retreat and duration simulated by 13 climate modeling systems “performing poorly,” in the
IPCC AR4 models.” estimation of Kim et al., in simulating monsoon
The Chinese-Australian research team report precipitation that affects almost half of the world’s
there “is no single outstanding model out of the 13 population, it would appear the climate modeling
models used in the analysis,” noting “some of the enterprise has a long way to go before confidence can
models have shown significant biases in mean be placed in its projections, whether under CO2
onset/retreat dates and some failed to produce the forcing or otherwise.
broad features of how [the] monsoon evolves.” Over Wu and Zhou (2013) sought to evaluate the
East Asian land, for example, they found “the models performance of the Flexible Global Ocean-
are nearly equally divided about the sign of potential Atmosphere-Land System model, Spectral Version 2
changes of onset/retreat.” And, sounding rather (FGOALS-s2)—a CGCM developed by the National
frustrated, they lament they “do not know why the Key Laboratory of Numerical Modeling for
models are different in simulating these dominant Atmospheric Sciences and Geophysical Fluid
processes and why in some models the ENSO Dynamics, Institute of Atmospheric Physics of the
influence is more significant than others,” adding “it Chinese Academy of Sciences—with respect to its
is unclear what are the key parameterizations leading ability to simulate the relationship between ENSO
to the differences in simulating ENSO and its and the East Asian-western North Pacific (EA-WNP)
responses to global warming,” citing Solomon et al. monsoon. The two authors write, “after nearly five
(2007) and Wang et al. (2009). They conclude, “there years of effort,” the original model (Wu et al., 2009)
is a long way ahead before we can make skillful and “has been improved in various facets” and it was
reliable prediction of monsoon onset, duration, therefore necessary “to carefully assess the ENSO-
intensity and evolution in [a] warmed climate.” monsoon relationship in the current version of the
In another study, Kim et al. (2012) note “the model.”
Asian monsoon influences almost half of the world’s Wu and Zhou noted several problems with the
population with their agriculture, life and society model: it “fails to simulate the asymmetry of the
depending on monsoon climate” and, therefore, wintertime circulation anomalies over the WNP
“understanding the physical processes that determine between El Niño and La Niña”; “the simulated
the character of the monsoon systems and also anomalous cyclone over the WNP (WNPAC)
providing accurate extended range predictions on a associated with La Niña is generally symmetric about
seasonal timescale is crucial for the economy and the WNPAC associated with El Niño, rather than
policy planning in the monsoon regions.” shifted westward as that in the observation”;
Kim et al. assessed the seasonal prediction skill “simulated La Niña events decay much faster than
regarding the Asian summer monsoon via the use of observed,” and “the precipitation anomalies over East
“retrospective predictions (1982–2009) from the Asia, especially those of the Meiyu rain belt, are
ECMWF System 4 (SYS4) and NCEP CFS version 2 much weaker than that in the observation.”
(CFSv2) seasonal prediction systems.” The four According to Chaudhari et al. (2013), “despite the
researchers state, “in both SYS4 and CFSv2, a cold potential for tropical climate predictability, and the
bias of sea-surface temperature (SST) is found over advances made in the development of climate models,
the Equatorial Pacific, North Atlantic [and] Indian the seasonal dynamical forecast of [the] Indian
Oceans,” as well as “over a broad region in the summer monsoon remains a challenging problem,”
Southern Hemisphere relative to observations,” and which explored via a study of model biases and how
“a warm bias is found over the northern part of the they create further biases as they wend their way
North Pacific and North Atlantic.” In addition, they through multiple stages of both simultaneous and
state “excessive precipitation is found along the sequential processes. Chaudhari et al. examined the
Intertropical Convergence Zone, equatorial Atlantic, performance of the National Centers for
equatorial Indian Ocean and the maritime continent.” Environmental Prediction (NCEP) Climate Forecast
And they find “the southwest monsoon flow and the System (CFS) over the Indian monsoon region in a
Somali Jet are stronger in SYS4, while the south- 100-year-long coupled run, framed “in terms of biases
easterly trade winds over the tropical Indian Ocean, of sea surface temperature (SST), rainfall and
the Somali Jet and the Subtropical northwestern circulation,” while also exploring “the role of
Pacific high are weaker in CFSv2 relative to the feedback processes in maintaining these biases.”
78
According to the nine researchers, the model affecting the monsoon evolution (e.g., delaying its
shows dry (wet) rainfall bias concomitant with cold onset).” They find “the ability of local anomalies over
(warm) SST bias over the east (west) equatorial the SWEIO to force a large-scale remote response to
Indian Ocean. They say these biases of SST and the north is further supported by numerical
rainfall affect both lower- and upper-level circulations experiments with the GFDL spectral dry dynamical
in a feedback process, which in turn regulates the SST core model.”
and rainfall biases by maintaining a coupled feedback Bollasina and Ming say their study “makes the
process. Subsequently, a dry (wet) rainfall bias over case that the precipitation bias over the SWEIO is
the east (west) Indian Ocean induces anomalous low forced by the model excess response to the local
level easterlies over the tropical Indian Ocean and meridional sea surface temperature gradient through
causes cold SST bias over the east Indian Ocean by enhanced near-surface meridional wind
triggering evaporation and warm SST bias over the convergence.” They thus conclude “a detailed
west Indian Ocean through advection of warm waters. investigation into the model physics to identify
They find the persistent SST bias then retains the possible parameters which may alleviate the model
zonal asymmetric heating and meridional temperature bias would be the natural extension of this work.”
gradient, resulting in a circum-global subtropical Shifting to the African monsoon, writing as
westerly jet core, which in turn magnifies the mid- background for their work, authors Zheng and
latitude disturbances and decreases the Mascarene Braconnot (2013) report “despite recent progress in
high, which in its turn diminishes the strength of the monitoring and understanding of the WAM [West
monsoon cross-equatorial flow and results in less African Monsoon] within the framework of the
upwelling as compared to that in the observations. African Monsoon Multidisciplinary Analysis
The latter phenomenon, they say, increases the SST (AMMA), there are still large uncertainties in
bias over the West Indian Ocean. In conclusion, projections of future climate in this region, such that
Chaudhari et al. state, “the coupled interaction among even the sign of future precipitation change is
SST, rainfall and circulation works in tandem through uncertain,” citing Solomon et al. (2007). The authors
a closed feedback loop to maintain the model biases “revisit the results of PMIP2 simulations over Africa
over the tropical Indian Ocean.” using two approaches.” The first “considers the
In one additional study on the Asian monsoon, ensemble of simulations in order to determine how
Bollasina and Ming (2013) note most current general well the PMIP2 models [of today] reproduce some of
circulation models “show a remarkable positive the basic features of the summer monsoon
precipitation bias over the southwestern equatorial precipitation,” while the objective of the second is “to
Indian Ocean (SWEIO), which can be thought of as a understand model differences by considering model
westward expansion of the simulated IO convergence characteristics for present-day climate and their
zone toward the coast of Africa.” They further note sensitivities to insolation change.”
“the bias is common to both coupled and uncoupled The scientists learned several things from the
models, suggesting that its origin does not stem from simulations. First, they report, the “meridional
the way boundary conditions are specified.” temperature gradient is underestimated between 0°
To further explore this issue, Bollasina and Ming and 20°N by the PMIP2 model median, resulting in a
“comprehensively characterized” “the spatio-temporal smaller gradient of sea level pressure between the
evolution of the precipitation and associated three- Gulf of Guinea and [the] Sahel,” which helps to
dimensional atmospheric circulation biases … by explain “a lower than observed low-level moisture
comparing the GFDL [Geophysical Fluid Dynamics flux and an underestimate of rainfall intensity when
Laboratory] AM3 atmospheric model to compared with observations.” Second, “the northward
observations.” extent of the rain belt and the intensity of
In the words of the two researchers, “the oceanic precipitation change are underestimated.” Third, “the
bias, which develops in spring and reduces during the models overestimate the solar radiation.” Fourth, the
monsoon season, is associated [with] a consistent models “underestimate the cloud radiative forcing in
precipitation and circulation anomalous pattern over deep and moderate convective regimes.” And fifth,
the whole Indian region,” where “in the vertical, the “some of the models have too strong a coupling
areas are linked by an anomalous Hadley-type between the latent heat and convection in deep
meridional circulation, whose northern branch convective regimes.”
subsides over northeastern India significantly Moving across the Atlantic to North America,
79
according to Cerezo-Mota et al. (2011), “the North out-of-phase annual cycles simulated near the equator
American monsoon (NAM) is the regional-scale by many GCMs” and “poor representation of the total
atmospheric circulation system (Stensrud et al., 1997) monsoonal precipitation over the Amazon and
responsible for the dramatic increase in precipitation northeast Brazil is observed in a large majority of the
during the summer in northwestern Mexico and the models.” As a consequence, they note, “simulations
southwest United States (Grantz et al., 2007).” They of the total seasonal precipitation, onset and end of
state “understanding the mechanisms that govern the the rainy season diverge among models and are
timing and intensity, as well as the impacts of climate notoriously unrealistic over [the] north and northwest
change on the NAM, is a priority for the scientific Amazon for most models.”
community, watershed managers and farmers in the
NAM area” because “the impacts of droughts/floods References
are devastating.”
Cerezo-Mota et al. investigated the degree of Bollasina, M.A. and Ming, Y. 2013. The general
realism in its simulation by a major regional climate circulation model precipitation bias over the southwestern
model (RCM)—the Hadley Centre Regional Model equatorial Indian Ocean and its implications for simulating
version 3P (HadRM3P)—analyzing the moisture the South Asian monsoon. Climate Dynamics 40: 823–838.
sources of the NAM by employing two different Bombardi, R.J. and Carvalho, L.M.V. 2009. IPCC global
boundary-condition data sets used to drive the model, coupled model simulations of the South America monsoon
which allowed them to assess the ability of the RCM system. Climate Dynamics 33: 893–916.
to reproduce rainfall under climate-change conditions
Cerezo-Mota, R., Allen, M., and Jones, R. 2011.
in the NAM region as predicted by GCMs.
Mechanisms controlling precipitation in the northern
As a result of their tests, the three U.K. portion of the North American monsoon. Journal of
researchers determined “two of the most commonly Climate 24: 2771–2783.
used GCMs that simulate well the NAM precipitation
(HadCM3 and MIROC) do not reproduce correctly Chaudhari, H.S., Pokhrel, S., Saha, S.K., Dhakate, A.,
the Great Plains low-level jet nor the moisture in the Yadav, R.K., Salunke, K., Mahapatra, S., Sabeerali, C.T.,
Gulf of Mexico,” both of which play major roles in and Rao, S.A. 2013. Model biases in long coupled runs of
NCEP CFS in the context of Indian summer monsoon.
the northern portion of the NAM. The implication of
International Journal of Climatology 33: 1057–1069.
their results, in the words of Cerezo-Mota et al., is
that “precipitation in Arizona-New Mexico would not Gadgil, S., Rajeevan, M., and Nanjundiah, R. 2005.
be correctly represented by a regional model driven Monsoon prediction—Why yet another failure? Current
by these GCMs.” Thus, they write RCMs driven by Science 88: 1389–1400.
the “most commonly used” GCMs “would not give Grantz, K., Rajagoopalan, B., Clark, M., and Zagona, E.
realistic simulations of the current climate of the 2007. Seasonal shifts in the North American monsoon.
region and therefore would not offer a realistic Journal of Climate 20: 1923–1935.
projection of climate change of the NAM.”
Moving further south and using real-world data Kim, H.-M., Webster, P.J., Curry, J.A., and Toma, V.E.
pertaining to the onset, end, and total rainfall of the 2012. Asian summer monsoon prediction in ECMWF
System 4 and NCEP CFSv2 retrospective seasonal
South American Monsoon System (SAMS)—as
forecasts. Climate Dynamics 39: 2975–2991.
characterized by precipitation data for the period
1979–2006, derived from the Global Precipitation Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M.,
Climatology Project—Bombardi and Carvalho (2009) Averyt, K.B., Tignor, M., and Miller, H.L. (Eds.) 2007.
evaluated the ability of ten IPCC global coupled Climate Change 2007: The Physical Science Basis.
climate models with distinct physics and resolutions Contribution of Working Group I to the Fourth Assessment
to simulate real-world SAMS characteristics. Over Report of the Intergovernmental Panel on Climate Change.
Cambridge University Press, Cambridge, United Kingdom.
northern South America, they find the annual
precipitation cycle “is poorly represented by most Stensrud, D.J., Gall, R.L., and Nordquist, M.K. 1997.
models” and “most models tend to underestimate Surges over the Gulf of California during the Mexican
precipitation during the peak of the rainy season.” In monsoon. Monthly Weather Review 125: 417–437.
addition, they write, “the misrepresentation of the
Wang, X., Wang, D., and Zhou, W. 2009. Decadal
Intertropical Convergence Zone and its seasonal cycle variability of twentieth-century El Niño and La Niña
seems to be one of the main reasons for the unrealistic
80
occurrence from observations and IPCC AR4 coupled land, and at the other it was mostly over the oceans;
models. Geophysical Research Letters 36: the other models fell across the range between the
10.1029/2009GL037929. extremes.
Wu, B. and Zhou, T. 2013. Relationships between the East A detailed analysis of how the models compared
Asian-Western North Pacific Monsoon and ENSO over Australia demonstrated even more variability.
simulated by FGOALS-s2. Advances in Atmospheric Lim and Roderick found the 1970–1999 annual
Sciences 30: 713–725. rainfall of the models ranged from 190.6 mm to
1059.1 mm, whereas the observed rainfall is in the
Wu, B., Zhou, T., Li, T., and Bao, Q. 2009. Inter-annual
range between 400 and 500 mm. Across the twenty-
variability of the Asian-Australian monsoon and ENSO
simulated by an ocean-atmosphere coupled model. Chinese first century under the A1B scenario, 24 models
Journal of Atmospheric Science 33: 285–299. returned an increase in precipitation over Australia,
while 15 showed decreases. Some models had the
Zhang, H., Liang, P., Moise, A., and Hanson, L. 2012. evaporation over Australia (and the Middle East)
Diagnosing potential changes in Asian summer monsoon exceeding precipitation.
onset and duration in IPCC AR4 model simulations using Stephens et al. (2010) write that in prior studies
moisture and wind indices. Climate Dynamics 39: 2465–
“land surface observations of the daily-accumulated
2486.
rainfall intensities of rates >1 mm/day were compiled
Zheng, W. and Braconnot, P. 2013. Characterization of from the Global Historical Climatology Network by
model spread in PMIP2 Mid-Holocene simulations of the Sun et al. (2006) and compared to analogous model
African Monsoon. Journal of Climate 26: 1192–1210. accumulated precipitation,” and they report “as in
other studies (e.g., Dai and Trenberth, 2004), the Sun
et al. comparison revealed a general overestimate in
1.4.5.3 Extreme Precipitation
the frequency of modeled precipitation and an
Atmospheric general circulation models (GCMs)
associated underestimate of intensity,” while noting
predict the planet’s hydrologic cycle will intensify as
“Wilcox and Donner (2007) reached a similar
the world warms, leading to an increase in the
conclusion.” To further examine the issue Stephens et
frequency and intensity of extreme precipitation
al. focused on the much larger portion of the planet
events. In an early review of the subject, Walsh and
that is occupied by oceans, using “new and definitive
Pittock (1998) reported “there is some evidence from
measures of precipitation frequency provided by
climate model studies that, in a warmer climate,
CloudSat [e.g., Haynes et al., 2009]” to assess the
rainfall events will be more intense” and “there is
realism of global model precipitation. Their analysis
considerable evidence that the frequency of extreme
employed five different computational techniques
rainfall events may increase in the tropics.” Upon
representing “state-of-the-art weather prediction
further study, however, they concluded, “because of
models, state-of-the-art climate models, and the
the insufficient resolution of climate models and their
emerging high-resolution global cloud ‘resolving’
generally crude representation of sub-grid scale and
models.”
convective processes, little confidence can be placed
Stephens et al. determined “the character of liquid
in any definite predictions of such effects.”
precipitation (defined as a combination of
More than a decade later, Lim and Roderick
accumulation, frequency, and intensity) over the
(2009) compared the water cycle characteristics of the
global oceans is significantly different from the
39 model runs used in the IPCC AR4 (2007)
character of liquid precipitation produced by global
assessment. The range of annual average global
weather and climate models,” noting “the differences
precipitation for the period 1970–1999 was 916.5 to
between observed and modeled precipitation are
1,187.2 mm/yr. For the A1B scenario of CO2 increase
larger than can be explained by observational retrieval
over the twenty-first century the increase in model
errors or by the inherent sampling differences
precipitation ranged between 22.9 and 69.1 mm/yr
between observations and models.” More specifically,
with those that warmed the most generally showing
they report for the oceans as a whole, “the mean
greater increase in precipitation. The rate of increase
model intensity lies between 1.3 and 1.9 times less
in precipitation with temperature over the period,
than the averaged observations” and occurrences “are
averaged about 2%/°C. A surprising result was the
approximately twice the frequency of observations.”
marked difference in the distribution of rainfall
They also note the models “produce too much
increase: At one extreme most rainfall was over the
precipitation over the tropical oceans” and “too little
81
mid-latitude precipitation.” And they indicate the extremes based on the large-scale balances or by a
large model errors “are not merely a consequence of simple scaling.” And they say “these results draw
inadequate upscaling of observations but indicative of attention to the need to understand why different deep
a systemic problem of models more generally.” convective storm types exist, how they interact with
In concluding their study, the nine U.S., U.K., each other and with the larger-scale circulation, and
and Australian researchers say their results imply what role they each play in the atmospheric general
state-of-the-art weather and climate models have circulation.” With respect to the ultimate consequence
“little skill in precipitation calculated at individual of these model deficiencies, in the concluding
grid points” and “applications involving downscaling sentence of their paper the four researchers state:
of grid point precipitation to yet even finer-scale “Until the full range of deep convective processes in
resolution has little foundation and relevance to the the tropics is more realistically represented in climate
real Earth system.” models, they cannot be used to predict the changes of
In a study published in the Journal of Climate, extreme precipitation events in a changing (warming)
Rossow et al. (2013) write “some of the concern climate.”
about possible negative impacts of a warming climate Schleip et al. (2010) compared the results of six
is focused on possible increases of precipitation regional climate models (RCMs) that were forced
extremes.” They sought to “exploit more than a with a common set of reanalysis data created by
decade of independent cloud and precipitation data running a climate model fed real-world data for a 20-
products covering the whole tropics (15°S-15°N) to year simulation period. The area analyzed was North
more clearly separate the contributions to average America, where winter precipitation was the response
precipitation intensity and daily average accumulation variable and the 100-year extremum of daily winter
rate made by the different types of deep convective precipitation was the test statistic, extreme values of
systems,” focusing on the period 1998–2008. which were estimated by fitting a tailed distribution to
The four researchers determined “the whole the data, taking into account their spatial aspects.
distribution of instantaneous precipitation intensity The six RCMs maintained similar general spatial
and daily average accumulation rate is composed of patterns of extrema across North America, with the
(at least) two separate distributions representing highest extremes in the Southeast and along the West
distinctly different types of deep convection Coast. However, when comparing absolute levels,
associated with different meteorological conditions.” which are most relevant to risk forecasts, the models
In particular, they found the extreme portion of the exhibited strong disagreement. The lowest-predicting
tropical precipitation intensity distribution “is model was low almost everywhere in North America
produced by 40% of the larger, longer-lived compared to the mean of the six models, and the
mesoscale-organized type of convection with only highest-predicting model was above the mean almost
about 10% of the ordinary convection occurrences everywhere. The difference between the two models
producing such intensities.” When accumulation rates was almost 60mm of daily precipitation for the 100-
were considered, they found “essentially all of the year extreme event over much of the United States.
values above 2 mm/hour are produced by the The other four models showed greatly differing
mesoscale systems.” spatial patterns of extremes from each other, and
Rossow et al. note “all of the climate GCMs these differences were found to be statistically
currently parameterize tropical deep convection as a significant by an F-test. The researchers speculate that
single process, localized to individual grid cells (on when driven by multiple GCMs rather than reanalysis
the order of 25-200 km in size) with short lifetimes data, the range of extreme outcomes would only
(on the order of minutes to a few hours) that most increase.
resembles ordinary cumulonimbus.” Thus, “today’s Other studies have further demonstrated the
atmospheric models do not represent mesoscale- difficulties models have in simulating extreme
organized deep convective systems that are generally precipitation properties and trends. Kiktev et al.
larger than current-day circulation model grid cell (2007), for example, analyzed the abilities of five
sizes but smaller than the resolved dynamical scales.” global coupled climate models that played important
On the basis of these observations, the four roles in the IPCC’s Fourth Assessment Report to
researchers contend “the observed distinctive simulate temporal trends over the second half of the
behavior of the different deep convective storm types twentieth century for five annual indices of
undercuts the simple projection of changes of precipitation extremes. Their results revealed “low
82
skill” or an “absence” of model skill. and a modest increase over coastal British Columbia.”
More of the same was reported by O’Gorman and However, they note, “most of these trends are not
Schneider (2009), who assessed “how precipitation significantly different from zero at the 95% level.” In
extremes change in simulations with 11 different addition, they found “trends in maximum daily
climate models in the World Climate Research discharge of unregulated rivers are consistent with the
Program’s (WCRP’s) Coupled Model above pattern, with increasing discharges over the
Intercomparison Project phase 3 (CMIP3) archive.” past three decades over Washington and northern
Based on their findings, as well as those of others, Oregon and declines over the remainder of Oregon
O’Gorman and Schneider report, “in simulations with and northern California.”
comprehensive climate models, the rate of increase in With respect to how consistent these results are
precipitation extremes varies widely among models, with what climate models suggest should occur in
especially in the tropics (Kharin et al., 2007).” They response to rising temperatures, Mass et al. report the
also note, “the variations among models in the tropics results of the two climate models analyzed by Chen et
indicate that simulated precipitation extremes may al. (2003) suggest “a pattern quite different from the
depend sensitively on the parameterization of one described above,” and they state the model
unresolved and poorly understood processes,” citing employed by Kim (2005) also produced “a pattern
the work of Wilcox and Donner (2007). They find, quite distinct from that observed since 1950.” In
“climate models do not correctly reproduce the addition, they note the models studied by Tebaldi et
interannual variability of precipitation extremes in the al. (2006) produced “a pattern closer, but not identical
tropics (Allan and Soden, 2008), or the frequency and to, that observed over the past 60 years,” and Duffy et
intensity distribution of precipitation generally al. (2006) “analyzed the precipitation produced over
(Wilcox and Donner, 2007; Dai, 2006; Sun et al., the western United States by four regional climate
2006).” Thus the two researchers conclude, “current models,” finding the spatial distributions of
climate models cannot reliably predict changes in precipitation they produced to “vary substantially,”
tropical precipitation extremes,” noting “inaccurate even among themselves.
simulation of the upward velocities may explain not Mass et al. conclude, “considering the large
only the intermodal scatter in changes in tropical variability in precipitation trends among the various
precipitation extremes but also the inability of models general circulation models in the above studies and
to reproduce observed interannual variability.” their associated regional climate models, and the
Noting “a number of articles in the media and differences between the simulated trend distributions
reports by some non-governmental organizations have and the observed trend patterns found in this study
suggested an increasing number of heavy and others, it is unclear whether anthropogenic global
precipitation events over portions of the western warming is the source of past spatial patterns of
United States and have proposed that anthropogenic extreme precipitation trends along the west coast of
global warming could be the cause,” Mass et al. North America.”
(2011) analyzed “trends in heavy precipitation for the Mishra et al. (2012) point out “about half of the
period 1950–2009 by examining the decadal human population lives in urban areas (Martine et al.,
distributions of the top 60, 40 and 20 two-day 2007), in contrast with only about 10 percent a
precipitation events for a collection of stations along century ago (Grimm et al., 2008).” Therefore, they
the coastal zone of the United States and British note, “changes in extreme precipitation as the climate
Columbia [Canada], as well as the decadal warms may pose challenges for stormwater
distribution of maximum daily discharge for management in urban areas, because most stormwater
unregulated rivers from northern California to infrastructure was designed under the assumption of
Washington State.” stationary climate that is ‘dead’ as argued by Milly et
The three researchers from the University of al. (2008).”
Washington’s Department of Atmospheric Sciences In an effort to better assess both the likelihood
report their findings showed “during the past 60 years and significance of this potential problem, Mishra et
there has been a modest increase in heavy al. compared precipitation output from all regional
precipitation events over southern and central coastal climate models that participated in the North
California, a decline in heavy events from northern American Regional Climate Change Assessment
California through the central Oregon coast, a Program (NARCCAP), described by Mearns et al.
substantial increase in major events over Washington, (2009), with observations made at 100 urban U.S.
83
weather stations with data for the period 1950–2009. in global mean temperature ranging from 0.5 to 6°C.
This work involved two distinct RCM simulations: The two Vietnamese researchers report finding
one forced by output from the National Center for “the greatest source of uncertainty in impact of
Environmental Prediction/Department of Energy climate change on streamflow is GCM structure
(NCEP/DOE) reanalysis (Kanamitsu et al., 2002) for (choice of GCM).” They say this result “is in
the period 1979–2000, and one forced by selected accordance with findings of other authors who also
global circulation models that provided RCM suggest that the choice of the GCM is the largest
boundary conditions for the period 1968–2000. source of uncertainty in hydrological projection,”
The analysis showed, “for most urban areas in the citing Kingston and Taylor (2010), Kingston et al.
western and southeastern U.S.,” in the words of the (2011), Nobrega et al. (2011), Thorne (2011), and Xu
three scientists, “the seasonality of 3-hour et al. (2011), adding the range of uncertainty could
precipitation extremes was not successfully increase even further if the analysis employed a larger
reproduced by the RCMs with either reanalysis or number of GCMs.
GCM boundary conditions,” because “the RCMs Khoi and Suetsugi say their findings indicate
tended to predict 3-hour precipitation maxima in “single GCM or GCMs ensemble mean evaluations of
winter, whereas the observations indicated summer.” climate change impact are unlikely to provide a
The authors also report the RCMs “largely representative depiction of possible future changes in
underestimated 3-hour precipitation maxima means streamflow.”
and 100-year return period magnitudes at most In one other streamflow-based study, Lloyd
locations across the United States for both reanalysis (2010) notes the Breede River “is the largest in South
and GCM boundary conditions.” For both 3- and 24- Africa’s Western Province, and plays a significant
hour annual precipitation maxima, they write, “RCMs part in the province’s economy,” and “models predict
with reanalysis boundary conditions underestimated that flows into it could be seriously affected by
interannual variability” while they “overestimated climate change.” More specifically, he reports
interannual variability with GCM boundary Steynor et al. (2009) used “a form of neural network”
conditions.” “trained on historical climate data” that were “linked
With respect to the ultimate utility of the RCM to historical flow data at five stations in the Breede
projections, Mishra et al. state performance deemed River valley” in order to “downscale from a global
acceptable for stormwater infrastructure design was climate model to the typical area of a catchment” and
adequate at only about 25 percent of the urban areas. thereby determine the consequences of predicted
Regardless of boundary conditions, they note, “RCM- future global warming for Breede River flows. That
simulated 3-hour precipitation maxima at a 100-year analysis projected Breede River flows would decrease
return period could be considered acceptable for if temperatures rise as predicted by climate models
stormwater infrastructure design at less than 12% of over the next 60 years.
the 100 urban areas.” As a check upon this approach to projecting the
Khoi and Suetsugi (2012) write “many general region’s hydrologic future, Lloyd, a researcher at the
circulation models (GCMs) consistently predict Energy Institute of the Cape Peninsula University of
increases in frequency and magnitudes of extreme Technology located in Cape Town, used flow data for
climate events and variability of precipitation (IPCC, five sites in the Breede Valley that had been
2007),” noting “this will affect terrestrial water maintained by the Department of Water Affairs to
resources in the future, perhaps severely (Srikanthan compute historical flow-rate trends over prior periods
and McMahon, 2001; Xu and Singh, 2004; Chen et of warming ranging from 29 to 43 years in length.
al., 2011).” Therefore, they conducted a study to see All of the future flow-rates calculated by Steynor
what aspect of the climate modeling enterprise led to et al. exhibited double-digit negative percentage
the greatest degree of uncertainty in predicting rates changes that averaged -25% for one global climate
of streamflow in Vietnam’s Be River Catchment. The model and -50% for another global climate model.
climate scenarios investigated by Khoi and Suetsugi Similarly, the mean past trend of four of Lloyd’s five
were generated from seven different CMIP3 GCMs— stations was also negative (-13%). But the other
CCCMA CGCM3.1, CSIRO Mk30, IPSL CM4, MPI station exhibited a positive trend (+14.6%). In
ECHAM5, NCAR CCSM3.0, UKMO HadGEM1, addition, by “examination of river flows over the past
UKMO Had CM3—using SRES emission scenarios 43 years in the Breede River basin,” Lloyd was able
A1B, A2, B1, and B2, along with prescribed increases to demonstrate that “changes in land use, creation of
84
impoundments, and increasing abstraction have Van Haren et al. conclude their report by
primarily been responsible for changes in the declaring “it is important that we improve our
observed flows” of the negative-trend stations. understanding of circulation changes, in particular
Interestingly, Steynor et al. had presumed related to the cause of the apparent mismatch between
warming would lead to decreased flow rates, as their observed and modeled circulation trends over the past
projections suggested, and they thus assumed their century,” citing Haarsma et al. (2013).
projections were correct. Lloyd was able to
demonstrate those results were driven primarily by References
unaccounted-for land use changes in the five
catchments, and that in his newer study the one site Allan, R.P. and Soden, B.J. 2008. Atmospheric warming
with “a pristine watershed” was the one that had the and the amplification of precipitation extremes. Science
“14% increase in flow over the study period,” which 321: 1481–1484.
was “contrary to the climate change predictions” and Chen, M., Pollard, D., and Barron, E.J. 2003. Comparison
indicative of the fact that “climate change models of future climate change over North America simulated by
cannot yet account for local climate change effects.” two regional climate models. Journal of Geophysical
Lloyd concludes “predictions of possible adverse Research 108: 4348–4367.
local impacts from global climate change should
Dai, A. 2006. Precipitation characteristics in eighteen
therefore be treated with the greatest caution” and,
coupled climate models. Journal of Climate 19: 4605–
“above all, they must not form the basis for any 4630.
policy decisions until such time as they can reproduce
known climatic effects satisfactorily.” Dai, A. and Trenberth, K.E. 2004. The diurnal cycle and its
Van Haren et al. (2013) write “estimates of future depiction in the Community Climate System Model.
changes in extremes of multi-day precipitation sums Journal of Climate 17: 930–951.
are critical for estimates of future discharge extremes Duffy, P.B., Arritt, R.W., Coquard, J., Gutowski, W., Han,
of large river basins and changes in [the] frequency of J., Iorio, J., Kim, J., Leung, L.-R., Roads, J., and Zeledon,
major flooding events,” citing Kew et al. (2010). E. 2006. Simulations of present and future climates in the
They indicate “a correct representation of past western United States with four nested regional climate
changes is an important condition to have confidence models. Journal of Climate 19: 873–895.
in projections for the future.”
Grimm, N.B., Faeth, S.H., Golubiewski, N.E., Redman,
In an attempt to achieve some of that all- C.L., Wu, J., Bai, X., and Briggs, J.M. 2008. Global change
important confidence, van Haren et al. investigated and the ecology of cities. Science 319: 756–760.
changes in multi-day precipitation extremes in late
winter in Europe and the Rhine river basin over the Haarsma, R., Selten, F., and van Oldenborgh, G. 2013.
past 60 years using daily precipitation data and “the Anthropogenic changes of the thermal and zonal flow
state-of-the-art gridded high resolution (0.5°) structure over Western Europe and Eastern North Atlantic
in CMIP3 and CMIP5 models. Climate Dynamics
precipitation fields of the European ENSEMBLES
10.1007/s00382-013-1734-8.
project version 7.0 (Haylock et al. 2008),” where
“observations [were] averaged to the same regular Haynes, J.M., L’Ecuyer, T.S., Stephens, G.L., Miller, S.D.,
1.5° grid when compared directly with the model Mitrescu, C., Wood, N.B., and Tanelli, S. 2009. Rainfall
results.” retrieval over the ocean with spaceborne W-band radar.
The four researchers report the climate models Journal of Geophysical Research 114: 10.1029/
“underestimate the trend in extreme precipitation in 2008JD009973.
the northern half of Europe” because they Haylock, M.R., Hofstra, N., Tank, A.M.G.K., Klok, E.J.,
“underestimate the change in circulation over the past Jones, P.D., and New, M. 2008. A European daily high-
century and as a result have a much smaller (extreme) resolution gridded data set of surface temperature and
precipitation response.” More specifically, they state precipitation for 1950–2006. Journal of Geophysical
“a dipole in the sea-level pressure trend over Research 113: 10.1029/2008JD010201.
continental Europe causes positive trends in extremes Kanamitsu, M., Ebisuzaki, W., Woollen, J., Yang, S.K.,
in northern Europe and negative trends in the Iberian Hnilo, J., Fiorino, M., and Potter, G. 2002. NCEP-DOE
Peninsula,” while “climate models have a much AMIP-II reanalysis (R-2). Bulletin of the American
weaker pressure trend dipole and as a result a much Meteorological Society 83: 1631–1643.
weaker (extreme) precipitation response.”
85
Kew, S.F., Selten, F.M., Lenderink, G., and Hazeleger, W. Urban precipitation extremes: How reliable are regional
2010. Robust assessment of future changes in extreme climate models? Geophysical Research Letters 39:
precipitation over the Rhine basin using a GCM. 10.1029/2011GL050658.
Hydrology and Earth Systems Science Discussions 7:
9043–9066. Nobrega, M.T., Collischonn, W., Tucci, C.E.M., and Paz,
A.R. 2011. Uncertainty in climate change impacts on water
Kharin, W., Zwiers, F.W., Zhang, X., and Hegerl, G.C. resources in the Rio Grande Basin, Brazil. Hydrology and
2007. Changes in temperature and precipitation extremes in Earth System Sciences 15: 585–595.
the IPCC ensemble of global coupled model simulations.
Journal of Climate 20: 1419–1444. O’Gorman, P.A. and Schneider, T. 2009. The physical
basis for increases in precipitation extremes in simulations
Khoi, D.N. and Suetsugi, T. 2012. Uncertainty in climate of 21st-century climate change. Proceedings of the
change impacts on streamflow in Be River Catchment, National Academy of Sciences, USA 106: 14,773–14,777.
Vietnam. Water and Environment Journal 26: 530–539.
Rossow, W.B., Mekonnen, A., Pearl, C., and Goncalves,
Kiktev, D., Caesar, J., Alexander, L.V., Shiogama, H., and W. 2013. Tropical precipitation extremes. Journal of
Collier, M. 2007. Comparison of observed and Climate 26: 1457–1466.
multimodeled trends in annual extremes of temperature and
precipitation. Geophysical Research Letters 34: 10.1029/ Schliep, E.M., Cooley, D., Sain, S.R., and Hoeting, J.A.
2007GL029539. 2010. A comparison study of extreme precipitation from
six different regional climate models via spatial
Kim, J. 2005. A projection of the effects of the climate hierarchical modeling. Extremes 13: 219–239.
change induced by increased CO2 on extreme hydrologic
events in the western U.S. Climatic Change 68: 153–168. Srikanthan, R. and McMahon, T.A. 2001. Stochastic
generation of annual, monthly and daily climate data: A
Kingston, D.G. and Taylor, R.G. 2010. Sources of review. Hydrology and Earth System Sciences 5: 653–670.
uncertainty in climate change impacts on river discharge
and groundwater in a headwater catchment of the Upper Steynor, A.C., Hewitson, B.C., and Tadross, M.A. 2009.
Nile Basin, Uganda. Hydrology and Earth System Sciences Projected future runoff of the Breede River under climate
14: 1297–1308. change. Water SA 35: 433–440.
Kingston, D.G., Thompson, J.R., and Kite, G. 2011. Stephens, G.L., L’Ecuyer, T., Forbes, R., Gettlemen, A.,
Uncertainty in climate change projections of discharge for Golaz, J.-C., Bodas-Salcedo, A., Suzuki, K., Gabriel, P.,
Mekong River Basin. Hydrology and Earth System and Haynes, J. 2010. Dreary state of precipitation in global
Sciences 15: 1459–1471. models. Journal of Geophysical Research 115: 10.1029/
2010JD014532.
Lim, W.H. and Roderick, M.L. 2009. An atlas of the global
water cycle. http://epress.anu.edu.au?p=127241 Sun, Y., Solomon, S., Dai, A., and Portmann, R.W. 2006.
How often does it rain? Journal of Climate 19: 916–934.
Lloyd, P. 2010. Historical trends in the flows of the Breede
River. Water SA 36: 329–333. Tebaldi, C., Hayhoe, K., Arblaster, J.M., and Meehl, G.A.
2006. Going to the extremes: An intercomparison of
Martine, G. et al. 2007. State of World Population 2007: model-simulated historical and future changes in extreme
Unleashing the Potential of Urban Growth. United Nations events. Climatic Change 79: 185–211.
Population Fund, New York, NY, USA.
Thorne, R. 2011. Uncertainty in the impacts of projected
Mass, C., Skalenakis, A., and Warner, M. 2011. Extreme climate change on the hydrology of a subarctic
precipitation over the west coast of North America: Is there environment: Laird River Basin. Hydrology and Earth
a trend? Journal of Hydrometeorology 12: 310–318. System Sciences 15: 1483–1492.
Mearns, L.O., Gutowski, W., Jones, R., Leung, R., Van Haren, R., van Oldenborgh, G.J., Lenderink, G., and
McGinnis, S., Nunes, A., and Qian, Y. 2009. A regional Hazeleger, W. 2013. Evaluation of modeled changes in
climate change assessment program for North America. extreme precipitation in Europe and the Rhine basin.
EOS: Transactions, American Geophysical Union 90: 311. Environmental Research Letters 8: 10.1088/1748–
9326/8/1/014053.
Milly, P.C.D., Betancourt, J., Falkenmark, M., Hirsch,
R.M., Kundzewicz, Z.W., Lettenmaier, D.P., and Stouffer, Walsh, K. and Pittock, A.B. 1998. Potential changes in
R.J. 2008. Stationarity is dead: Whither water tropical storms, hurricanes, and extreme rainfall events as a
management? Science 319: 573–574. result of climate change. Climatic Change 39: 199–213.
Mishra, V., Dominguez, F., and Lettenmaier, D.P. 2012. Wilcox, E.M. and Donner, L.J. 2007. The frequency of
86
extreme rain events in satellite rain-rate estimates and an

atmospheric General Circulation Model. Journal of
Climate 20: 53–69.
Xu, C.Y. and Singh, V.P. 2004. Review on regional water
resources assessment models under stationary and changing
climate. Water Resources Management 18: 591–612.
Xu, H., Taylor, R.G. and Xu, Y. 2011. Quantifying
uncertainty in the impacts of climate change on river
discharge in sub-catchments of the Yangtze and Yellow
River Basins, China. Hydrology and Earth System Sciences
15: 333–344.
1.3.6 Temperature
How much of the warming of the past 100 years is
due to human activity? When multiple forcings are
varying and poorly characterized, and there is also
internal variation, this question is more difficult, if
not impossible, to answer. Nevertheless, several
studies have attempted to do so
Figure 1.3.6.1.1. Twentieth-century Central United States
and mean global temperature anomalies, as described in the
1.3.6.1 Surface and Near-Surface text above. Adapted from Kunkel et al. (2006).
Citing the work of Folland et al. (2001), Robinson et
al. (2002), and Pan et al. (2004), Kunkel et al. (2006)
note there was a lack of warming throughout the contemporary CGCMs to the estimated external
central and southeastern United States over the course forcings.”
of the twentieth century, dubbed a “warming hole” by
the latter set of investigators. For an area they denote Kiktev et al. (2007) introduce their study by
the Central United States (CUS), which they stating the importance of “comparing climate
described as “one of the most agriculturally modeling results with historical observations … to
productive regions of the world and roughly defined further develop climate models and to understand the
around what is known as the ‘Corn Belt’,” Kunkel et capabilities and limitations of climate change
al. used a data set of 252 surface climate stations with projections.” They analyzed the abilities of five
less than 10 percent missing temperature data over the global coupled climate models that played important
period 1901–1999 to construct the CUS temperature roles in the IPCC’s Fourth Assessment Report to
time series plotted in Figure 1.3.6.1.1, where mean simulate temporal trends over the second half of the
global temperature as determined by Hansen et al. twentieth century of five annual indices of extremes
(2001) is also plotted. Then, for comparative in surface temperature—annual percentage of days
purposes, they examined 55 coupled general with Tmin < 10th percentile, with Tmax < 10th
circulation model (CGCM) simulations driven by percentile, with Tmin > 90th percentile, with Tmax >
“modern estimates of time-varying forcing” plus 19 90th percentile, and annual number of frost days, i.e.,
preindustrial unforced simulations, all derived from Tmin < 0°C—as well as five annual indices of
18 CGCMs. extremes in precipitation, the observational data for
It is obvious, as shown in Figure 1.3.6.1.1, that which they obtained from the HadEX global data set
the Central U.S. twentieth century temperature series that contains gridded annual and seasonal values of
is vastly different from that of the globe as a whole, at the ten extreme indices calculated from series of in
least as the latter is represented by Hansen et al. situ daily measurements (Alexander et al., 2006). The
Rather than the final temperature of the twentieth international research team, hailing from Australia,
century being warmer than the rest of the century, for Japan, Russia, and the United Kingdom, state “the
this region it was more than 0.7°C cooler than it was a results mostly show moderate skill for temperature
mere 65 years earlier. In addition, Kunkel et al. report indices.”
“the warming hole is not a robust response of In an effort “to distinguish between simultaneous
87
natural and anthropogenic impacts on surface cooling: 9.0 for the low Arctic and 12.5 for the high
temperature, regionally as well as globally,” Lean and Arctic. Such findings constitute another important
Rind (2008) performed “a robust multivariate analysis example of the principle described by Reifen and
using the best available estimates of each together Toumi (2009): that a model that performs well for one
with the observed surface temperature record from time period will not necessarily perform well for
1889 to 2006.” They report, “contrary to recent another.
assessments based on theoretical models (IPCC, Expounding on this principle, Reifen and Toumi
2007) the anthropogenic warming estimated directly (2009) note, “with the ever increasing number of
from the historical observations is more pronounced models, the question arises of how to make a best
between 45°S and 50°N than at higher latitudes,” estimate prediction of future temperature change.”
which, in their words, “is the approximate inverse of That is to say, which model should one use? With
the model-simulated anthropogenic plus natural respect to this question, they note, “one key
temperature trends ... which have minimum values in assumption, on which the principle of performance-
the tropics and increase steadily from 30 to 70°N.” based selection rests, is that a model which performs
Furthermore, they continue, “the empirically-derived better in one time period will continue to perform
zonal mean anthropogenic changes have approximate better in the future.” In other words, if a model
hemispheric symmetry whereas the mid-to-high predicts past climate fairly well, it should predict
latitude modeled changes are larger in the Northern future climate fairly well. The principle sounds
hemisphere.” The two researchers conclude “climate reasonable enough, but does it hold true?
models may therefore lack—or incorrectly Reifen and Toumi examined this question “in an
parameterize—fundamental processes by which observational context” for what they describe as “the
surface temperatures respond to radiative forcings.” first time.” Working with the 17 climate models
Chylek et al. (2009) point out “one of the robust employed by the IPCC in its Fourth Assessment
features of the AOGCMs [Atmosphere-Ocean Report, they determined how accurately individual
General Circulation Models] is the finding that the models, as well as subsets of the 17 models, simulated
temperature increase in the Arctic is larger than the the temperature history of Europe, Siberia, and the
global average, which is attributed in part to the entire globe over a selection period (such as 1900–
ice/snow-albedo temperature feedback.” More 1919) and a subsequent test period (such as 1920–
specifically, they say, “the surface air temperature 1939), asking whether the results for the test period
change in the Arctic is predicted to be about two to are as good as those of the selection period. They
three times the global mean,” citing the IPCC (2007). followed this procedure while working their way
In conducting their own study, the authors utilized through the entire twentieth century at one-year time-
Arctic surface air temperature data from 37 steps for not only 20-year selection and test intervals
meteorological stations north of 64°N in an effort to but also for 10- and 30-year intervals.
explore the latitudinal variability in Arctic The two researchers could find “no evidence of
temperatures within two belts—the low Arctic (64°N- future prediction skill delivered by past performance-
70°N) and the high Arctic (70°N-90°N)—comparing based model selection,” noting, “there seems to be
them with mean global air temperatures over three little persistence in relative model skill.” They
sequential periods: 1910–1940 (warming), 1940–1970 speculate “the cause of this behavior is the non-
(cooling), and 1970–2008 (warming). stationarity of climate feedback strengths,” which
The five researchers report, “the Arctic has they explain by stating, “models that respond
indeed warmed during the 1970–2008 period by a accurately in one period are likely to have the correct
factor of two to three faster than the global mean.” feedback strength at that time,” but “the feedback
More precisely, the Arctic amplification factor was strength and forcing is not stationary, favoring no
2.0 for the low Arctic and 2.9 for the high Arctic. But particular model or groups of models consistently.”
that is the end of the real world’s climate-change Given such findings, the U.K. physicists conclude
agreement with theory. During the 1910–1940 their analysis of the subject by stating “the common
warming, for example, the low Arctic warmed 5.4 investment advice that ‘past performance is no
times faster than the global mean, while the high guarantee of future returns’ and to ‘own a portfolio’
Arctic warmed 6.9 times faster. Even more out of line appears also to be relevant to climate projections.”
with climate model simulations were the real-world Also studying the Arctic, Liu et al. (2008)
Arctic amplification factors for the 1940–1970 “assessed how well the current day state-of-the-art
88
reanalyses and CGCMs [coupled global climate

models] are reproducing the annual mean, seasonal
cycle, variability and trend of the observed SAT
[surface air temperature] over the Arctic Ocean for
the late twentieth century (where sea ice changes are
largest).” The results indicate “large uncertainties are
still found in simulating the climate of the twentieth
century,” they write, and on an annual basis, “almost
two thirds of the IPCC AR4 models have biases that
[are] greater than the standard deviation of the
observed SAT variability.” Liu et al. further note the
models “cannot capture the observed dominant SAT
mode variability in winter and seasonality of SAT
trends.” The majority of the models “show an out-of-
phase relationship between the sea ice area and SAT
biases,” they write, and “there is no obvious
improvement since the IPCC Third Assessment
Report.”
Anagnostopoulos et al. (2010) evaluate the ability
Figure 1.3.6.1.2. The observed and model temperature time
of models used by the IPCC to generate regional
series integrated over the USA (annual means and maximum
climates to reproduce observed climate and climate and minimum monthly) for the annual and 30-year time
change, and they take issue with the IPCC’s scales. Adapted from Figure 12 in Anagnostopoulos et al.
assessment of its models. They compare the output of (2010).
these models to temperature and precipitation
observations at 55 points worldwide. They then do a
similar comparison for 70 points in the USA on As Anagnostopoulos et al. state, “It is claimed
several time scales. They selected the USA for the that GCMs provide credible quantitative estimates of
refined study because it has a dense network of future climate change, particularly at continental
surface stations. Thus, they can evaluate the model times scales and above. Examining the local
performance at the large scale and the regional scale. performance of the models at 55 points, we found that
In order to compare observed records to model local projections do not correlate well with observed
projections, they use a statistical technique called measurements.” They found the model performance
“best linear unbiased estimation,” comparing was actually worse on larger scales than on regional
observations at a particular station and nearby stations ones. They suggest the central issue is not about the
to a linear combination of the model outputs and performance of the GCMs, but whether climate can
comparing the time series via traditional statistical be predicted deterministically at all given the
measures. uncertainties inherent in the atmosphere.
The authors demonstrate, “At the monthly time Anagnostopoulos et al. use statistics to make the
scale the models generally reproduce the sequence of point others have made regarding models using
cold-warm and wet-dry periods at all stations dynamics. Anagnostopoulos et al. say there simply
examined.” However, the results were much worse at may not be a predictable “core” to the climate that is
the annual time scale and the variability in the general only obscured by layers of complexity or uncertainty.
circulation is underestimated. For the climate time- They also propose a shift in our modeling approach,
scales, some of the grid points will show 30-year suggesting statistical methods, in addition to
temperature rises in the model simulation, but the dynamical methods and models, be used to generate
actual data shows temperature falls. This is similar for future scenarios for climate as has been done
both the large and regional scales. Additionally, the successfully in other fields such as thermophysics or
models could not capture the long-term climate turbulence.
changes from 1890 to the end of the twentieth century Lo and Hsu (2010) point out “widespread abrupt
at many of the points (Figure 1.3.6.2). warming in the extratropical Northern Hemisphere
occurred in the late 1980s” and say the warming was
associated with a change in the relative influence of a
89
Pacific Decadal Oscillation (PDO)-like pattern and an latter period of time. They also note “the trend due to
Arctic Oscillation (AO)-like pattern. Utilizing land only the forced component is statistically the same in
surface temperature data obtained from the University the two 32-year periods and in the 63-year period.”
of East Anglia’s Climatic Research Unit (Mitchell et That is to say, the forced part was not accelerating.
al., 2004), plus sea surface temperature data obtained Taken together, these results imply the observed trend
from the U.K.’s Meteorological Office (Rayner et al., differs between the periods 1946–1977 and 1977–
2003), the authors explored the nature of this 2008 not because the forced response accelerated but
temperature increase and tested the ability of because “internal variability led to relative cooling in
IPCC/CMIP3 models to simulate it. the earlier period and relative warming in the later
The two Taiwanese researchers report the period.” Their results suggest simple extrapolations of
“accelerated warming in the Northern Hemisphere” rates of warming from 1980 onward overestimate the
was related to the emergence of an AO-like pattern in forced component of warming, and therefore using
the late 1980s and the concomitant weakening of the this period without factoring out internal variability
previously prevailing PDO-like pattern occurring in will likely lead to unrealistic values of climate
tandem. These resultsthey say, , together with results sensitivity.
obtained from current IPCC/CMIP3 models, “do not Lavers et al. (2009), in a study described
support the scenario that the emerging influence of previously in Section 1.4.5.1, assessed the
the AO-like pattern in the 1980s can be attributed to predictability of monthly “retrospective forecasts,” or
the anthropogenic greenhouse effect.” Lo and Hsu hindcasts, composed of multiple nine-month
also conclude, “this study indicates the importance of projections initialized during each month of the year
the changing behavior of the decadal fluctuations in over the period 1981–2001, comparing the projections
the recent climate regime shift,” and they highlight against real-world air temperatures obtained from
what they call “the insufficient capability of the ERA-40 reanalysis data. In addition, they conducted a
present state-of-the-art IPCC/CMIP3 models in virtual-world analysis where the output of one of the
simulating this change.” models was arbitrarily assumed to be the truth and the
DelSole et al. (2011) used a set of climate models average of the rest of the models was assumed to be
run in “control” or unforced mode to develop a 300- the predictor.
year data set of spatial ocean temperature data, where The researchers report that in the virtual world of
it was found that an internal pattern, detectable using the climate models, there was quite good skill over
a spatial fingerprinting technique, could be identified the first two weeks of the forecast, when the spread of
in the simulated data. This spatial pattern of ocean ensemble model members was small, but there was a
temperature anomalies was labeled the Internal large drop-off in predictive skill in the second 15-day
Multidecadal Pattern (IMP). It was found to be highly period. Things were worse in the real world, where
coherent with the Atlantic Multidecadal Oscillation they say the models had negligible skill over land at a
(AMO), suggesting the models were able to match the 31-day lead time, “a relatively short lead time in
internal dynamics of the real-Earth system reasonably terms of seasonal climate prediction.” The three
well. The researchers next extracted, also by means of researchers conclude, “it appears that only through
discriminant fingerprinting, the forced component of significant model improvements can useful long-lead
the spatial patterns produced in the absence of the forecasts be provided that would be useful for
IMP as an orthogonal function, which they decision makers,” a quest they state “may prove to be
demonstrated has only a minor effect (less than 1/7 elusive.”
the amplitude) on the IMP. They then used historical Crook and Forster (2011) note “predicting our
sea surface temperature data to evaluate the relative future influence on climate requires us to have
importance of the forced vs. IMP components of confidence in the climate models used to make
change since 1850. predictions, and in particular that the models’ climate
In considering the latter portion of the record sensitivity and ocean heat storage characteristics are
(1946–2008), the results indicate the internal realistic.” They go on to say that confidence may be
variability component of climate change (the IMP) gained “by assessing how well climate models
operated in a cooling mode between 1946 and 1977, reproduce current climatology and climate variability,
but switched thereafter to a warming mode between and how their feedback parameters compare with
1977 and 2008, suggesting the IMP is strong enough estimates from observations.”
to overwhelm any anthropogenic signal during this Using the World Climate Research Programme’s
90
Coupled Model Intercomparison Project phase 3 The four researchers find certain of the CMIP3
(CMIP3) and real-world data from the HadCRUT3 models “are unsuitable for application to China, with
database, Crook and Forster first determined “the little capacity to simulate the spatial variations in
surface temperature response contributions due to climate across the country,” adding that all of them
long-term radiative feedbacks, atmosphere-adjusted “give unsatisfactory simulations of the inter-annual
forcing, and heat storage and transport for a number temporal variability.” In addition, they find “each
of coupled ocean-atmosphere climate models,” after AR4 GCM performs differently in different regions of
which they compared “the linear trends of global China.” Miao et al. conclude “the inter-annual
mean, Arctic mean and tropical mean surface simulations (temperature and precipitation) by AR4
temperature responses of these models with GCMs are not suitable for direct application” and
observations over several time periods.” They also “caution should be applied when using outputs from
investigated “why models do or do not reproduce the the AR4 GCMs in hydrological and ecological
observed temperature response patterns” and assessments” due to their “poor performance.”
performed “optimal fingerprinting analyses on the According to Morak et al. (2013), studies of
components of surface temperature response to test observational temperature records over the past 50 to
their forcing, feedback and heat storage responses.” 100 years have found evidence for increases in both
The two University of Leeds (U.K.) researchers mean and extreme (maximum and minimum) near-
found tropical twentieth century warming was too surface air temperatures, but they note the increase in
large and Arctic amplification too low in the maximum temperature has been of smaller magnitude
Geophysical Fluid Dynamics Laboratory CM2.1 than the increase in minimum temperature. This state
model, the Meteorological Research Institute of affairs has led to a decrease in the diurnal
CGCM232a model, and the MIROC3(hires) model temperature range. They compared “observed and
“because of unrealistic forcing distributions.” They climate model-simulated trends in mean values of
also determined “the Arctic amplification in both temperature extreme indices, splitting the year into
National Center for Atmospheric Research models is the dynamically active boreal cold (ONDJFM) and
unrealistically high because of high feedback warm (AMJJAS) seasons.” To do so, they used
contributions in the Arctic compared to the tropics.” “modeled daily minimum and maximum surface
In addition, they report, “few models reproduce the temperature data derived from simulations with the
strong observed warming trend from 1918 to 1940,” Hadley Centre Global Environmental Model, version
noting “the simulated trend is too low, particularly in 1 (HadGEM1).”
the tropics, even allowing for internal variability, The three U.K. researchers report, among other
suggesting there is too little positive forcing or too findings, that the model “significantly underestimates
much negative forcing in the models at this time.” changes in some regions, particularly in winter across
According to Miao et al. (2012), the accuracy of large parts of Asia” and “has a tendency to
any GCM “should be established through validation overestimate changes in the frequency of hot days in
studies before using it to predict future climate both the [a] winter and [b] summer seasons over [d]
scenarios.” “Although accurate simulation of the most regions, and in the [e] global and [f]
present climate does not guarantee that forecasts of hemispheric mean.” The model “also overestimates
future climate will be reliable,” they write, “it is changes in the frequency of warm winter days on
generally accepted that the agreement of model larger scales.” With respect to changes in cold
predictions with present observations is a necessary extremes the model “does underestimate them in
prerequisite in order to have confidence in the quality some regions” and “there are some regions with
of a model.” trends of the opposite sign.” In addition, Morak et al.
Working within this conceptual framework, Miao “the particular regional trend pattern, often also
et al. assessed the performance of the AR4 GCMs, referred to as the ‘warming hole,’ is not evident in the
otherwise known as the CMIP3 models, in simulating simulated trend pattern,” citing Pan et al. (2004),
precipitation and temperature in China from 1960 to Kunkel et al. (2006), Portmann et al. (2009), and
1999, comparing the model simulations with observed Meehl et al. (2012). And they say “the model shows a
data using “system bias (B), root-mean-square error tendency to significantly overestimate changes in
(RMSE), Pearson correlation coefficient (R) and warm daytime extremes, particularly in summer.”
Nash-Sutcliffe model efficiency (E)” as evaluation Although the HadGEM1 does some things well, there
metrics. are a number of other things it has yet to accomplish
91
satisfactorily. impossible to present reliable future projections from

As the observed rate of rise in the global average a collection of climate models which generally cannot
temperature has slowed during recent decades, the simulate observed change.”
gap between observations and climate models results In the same vein, Fyfe et al. (2013) examined the
has widened. The discrepancy is now so large as to performance of the newer CMIP5 climate models and
indicate a statistically significant difference between found a similar result. The researchers “considered
the climate modeled trends and the observed trend. trends in global mean surface temperature computed
The existence of such a large discrepancy is a strong from 117 simulations of the climate by 37 CMIP5
indication that climate models are failing in their models.” They note the “models generally simulate
ability to accurately capture known changes in this natural variability—including that associated with the
key parameter of Earth’s climate system. El Niño–Southern Oscillation and explosive volcanic
A clear demonstration the models are failing to eruptions—as well as estimate the combined response
contain observations was made by Knappenberger of climate to changes in greenhouse gas
and Michaels (2013), who compiled the range of concentrations, aerosol abundance (of sulphate, black
trends in surface temperatures during the first several carbon and organic carbon, for example), ozone
decades of the twenty-first century projected by concentrations (tropospheric and stratospheric), land
CMIP3 climate models running emissions scenario use (for example, deforestation) and solar variability.”
A1B. Knappenberger and Michaels placed the But despite the models’ claimed ability to simulate
observed temperatures within this modeled both natural variability and human influences, when
distribution (see Figure 1.3.6.1.3). The observed compared against actual observations, their
trends lie very close to, and in some cases below, the simulation of the evolution of the global average
lower bound of the 95% certainty range derived from temperature is flawed. Over the past 20 years (1993–
climate model projections. Knappenberger and 2012), Fyfe et al. find the “rate of warming is
Michaels report “at the global scale, climate models significantly slower than that simulated by the climate
are on the verge of failing to adequately capture models participating in Phase 5 of the Coupled Model
observed changes in the average temperature over the Intercomparison Project (CMIP5)” and “the
past 10 to 30 years—the period of the greatest human inconsistency between observed and simulated global
influence on the atmosphere.” They conclude, “It is warming is even more striking for temperature trends
Figure 1.3.6.1.3. Trends (through 2012) in three observed global surface temperature records of length 5 to 15 years
(colored lines) set against the probability (gray lines) derived from the complete collection of climate model runs used in
the IPCC Fourth Assessment Report under the SRES A1B emissions scenario. Adapted from Knappenberger and Michaels
(2013).
92
computed over the past fifteen years (1998–2012).” Hansen, J., Ruedy, R., Sato, M., Imhoff, M., Lawrence, W.,
They add, “It is worth noting that the observed trend Easterling, D., Peterson, T., and Karl, T. 2001. A closer
over this period—not significantly different from look at United States and global surface temperature
zero—suggests a temporary ‘hiatus’ in global change. Journal of Geophysical Research 106: 23,947–
23,963.
warming.” Fyfe et al. conclude, “Ultimately the
causes of this inconsistency will only be understood Intergovernmental Panel on Climate Change. 2007.
after careful comparison of simulated internal climate Climate Change 2007: The Physical Science Basis,
variability and climate model forcings with Contribution of Working Group I to the Fourth Assessment
observations from the past two decades, and by Report of the Intergovernmental Panel on Climate Change.
waiting to see how global temperature responds over Solomon, S. et al. (Eds.) Cambridge University Press,
the coming decades.” Cambridge, UK.
The IPCC is in a very difficult position. In order Kiktev, D., Caesar, J., Alexander, L.V., Shiogama, H., and
to defend the CMIP5 suite of models, it must Collier, M. 2007. Comparison of observed and
somehow argue they did not underpredict warming as multimodeled trends in annual extremes of temperature and
the greatest increases in atmospheric carbon dioxide precipitation. Geophysical Research Letters 34: 10.1029/
content occurred. Further, they must invalidate at 2007GL029539.
least 19 separate experiments authored by 42 Knappenberger, P.C. and Michaels, P.J., 2013. Policy
researchers in the citations noted earlier in this Implications of Climate Models on the Verge of Failure.
section. American Geophysical Union Science Policy Conference.
Washington, DC, June 24–26, 2013.
References
Kunkel, K.E., Liang, X.-Z., Zhu, J., and Lin, Y. 2006. Can
Alexander, L.V. et al. 2006. Global observed changes in CGCMs simulate the twentieth-century “warming hole” in
daily climate extremes of temperature and precipitation. the central United States? Journal of Climate 19: 4137–
Journal of Geophysical Research 111: 10.1029/ 4153.
2005JD006290. Lavers, D., Luo, L., and Wood, E.F. 2009. A multiple
Anagnostopoulos, G.G., Koutsoyiannis, D., Christofides, model assessment of seasonal climate forecast skill for
A., Efstradiadis, A., and Mamassis, N. 2010. A comparison applications. Geophysical Research Letters 36: 10.1029/
of local and aggregated climate model outputs with 2009GL041365.
observed data. Hydrological Sciences Journal 55: 1094– Lean, J.L. and Rind, D.H. 2008. How natural and
1110. anthropogenic influences alter global and regional surface
Chylek, P., Folland, C.K., Lesins, G., Dubey, M.K., and temperatures: 1889 to 2006. Geophysical Research Letters
Wang, M. 2009. Arctic air temperature change 35: 10.1029/2008GL034864.
amplification and the Atlantic Multidecadal Oscillation. Liu, J., Zhang, Z., Hu, Y., Chen, L., Dai, Y., and Ren, X.
Geophysical Research Letters 36: 10.1029/2009GL038777. 2008. Assessment of surface air temperature over the
Crook, J.A. and Forster, P.M. 2011. A balance between Arctic Ocean in reanalysis and IPCC AR4 model
radiative forcing and climate feedback in the modeled 20th simulations with IABP/POLES observations. Journal of
century temperature response. Journal of Geophysical Geophysical Research 113: 10.1029/2007JD009380.
Research 116: 10.1029/2011JD015924. Lo, T.-T. and Hsu, H.-H. 2010. Change in the dominant
DelSole, T., Tippett, M.K., and Shukla, J. 2011. A decadal patterns and the late 1980s abrupt warming in the
significant component of unforced multidecadal variability extratropical Northern Hemisphere. Atmospheric Science
in the recent acceleration of global warming. Journal of Letters 11: 210–215.
Climate 24: 909–926. Meehl, G.A., Arblaster, J.M., and Branstator, G. 2012.
Folland, C.K. et al. 2001. Observed climate variability and Mechanisms contributing to the warming hole and the
change. In: Houghton, J.T. et al. (Eds.). Climate Change consequent U.S. east-west differential of heat extremes.
2001: The Scientific Basis, Cambridge University Press, Journal of Climate 25: 6394–6408.
Cambridge, UK, pp. 99–181. Miao, C., Duan, Q., Yang, L., and Borthwick, A.G.L. 2012.
Fyfe, J.C., Gillett, N.P., and Zwiers, F.W. 2013. On the applicability of temperature and precipitation data
Overestimated global warming over the past 20 years. from CMIP3 for China. PLoS ONE 7: e44659.
Nature Climate Change 3: 767–769. Mitchell, T.D., Carter, T.R., Jones, P.D., Hulme, M., and
93
New, M. 2004. A comprehensive set of high-resolution data for this type of calculation. For example,
grids of monthly climate for Europe and the globe: the particularly for older radiosonde observations, on
observed record (1901-2000) and 16 scenarios (2001– days when the upper wind was stronger, the balloons
2100). Tyndall Centre Working Paper 55, Norwich, United would tend to blow out of receiver range. This created
Kingdom.
a bias by causing missing data for high winds for
Pan, Z., Arritt, R.W., Takle, E.S., Gutowski Jr., W.J., older observations, leading to a spurious warm trend
Anderson, C.J., and Segal, M. 2004. Altered hydrologic over time. Overall, the TWE-based trends were three
feedback in a warming climate introduces a “warming times greater than trends derived from all other types
hole.” Geophysical Research Letters 31: 10.1029/2004GL of data. In addition, they did not agree with other
020528. wind data and also were based on much sparser data.
Portmann, R.W., Solomon, S., and Hegerl, G.C. 2009. Radiosonde data were therefore not used in the
Spatial and seasonal patterns in climate change, Christy et al. analysis, which also identified a small
temperatures, and precipitation across the United States. warm bias in the RSS satellite data that was explained
Proceedings of the National Academy of Sciences 106: by Christy and his colleagues.
7324–7329. The next innovation was to use the Scaling Ratio
(SR), the ratio of atmospheric temperature trend to
Rayner, N.A., Parker, D.E., Horton, E.B., Folland, C.K.,
Alexander, L.V., Rowell, D.P., Kent, E.C., and Kaplan, A. surface temperature trend. The SR attempts to factor
2003. Global analyses of sea surface temperature, sea ice, out the effect of the lack of actual (historic) El Niños
and night marine air temperature since the late nineteenth or other oscillations in climate model runs, and such
century. Journal of Geophysical Research 35: 10.1029/ simulated events in different computer runs. Christy
2002JD002670. and his eight colleagues found the SR for real-world
data was 0.8 ± 0.3, whereas the model simulations
Reifen, C. and Toumi, R. 2009. Climate projections: Past
had a SR of 1.38 ± 0.08 (a significant difference).
performance no guarantee of future skill? Geophysical
Research Letters 36: 10.1029/2009GL038082. That is, the data show a lower rate of warming for the
lower troposphere than for the surface (though not
Robinson, W.A., Reudy, R., and Hansen, J.E. 2002. On the statistically different), whereas the models show
recent cooling in the east-central United States. Journal of amplification. The SR value for the middle
Geophysical Research 107: 10.1029/2001JD001577. troposphere data was 0.4, even more different from
the model predictions. Only the SR for RSS data,
1.4.6.2 Mid- and Upper-Troposphere which has a documented warming bias, overlaps with
Several studies have examined model treatment of any model SR results. Given these findings, the work
processes in the troposphere. The testing of climate of Christy et al. suggests current state-of-the-art
model results is an important but difficult problem. climate models are fundamentally wrong in how they
One of the key model results is the presence of a represent this portion of Earth’s atmosphere.
tropical troposphere “hotspot” in which the The tropical troposphere also was the focus of
troposphere warms faster than the surface under study for Fu et al. (2011), who note the IPCC Fourth
conditions of enhanced greenhouse gas forcing. Assessment Report (AR4) general circulation models
Previous studies have produced disagreement over “predict a tropical tropospheric warming that
whether data were consistent with models on this increases with height, reaches its maximum at ~200
question. However, Christy et al. (2010) made several hPa, and decreases to zero near the tropical
advances by doing enhancing the data for surface tropopause.” They add, “this feature has important
trends, extending the data to a 31-year length, implications to the climate sensitivity because of its
evaluating the wind-based temperature estimates, and impact on water vapor, lapse rate and cloud feedbacks
clarifying the meaning of “best estimate” multi-data and to the change of atmospheric circulations.”
warming trends from data and models. Therefore, they write, it is “critically important to
Two prior studies had derived tropospheric observationally test the GCM-simulated maximum
temperature trends from the Thermal Wind Equation warming in the tropical upper troposphere.”
(TWE)—which uses radiosonde measurements of Fu et al. thus examined trends in the temperature
wind speed to calculate temperature—on the difference (ΔT) between the tropical upper- and
theoretical basis that warmer air should move faster lower-middle-troposphere based on satellite
than cooler air. They found there were biases in the microwave sounding unit (MSU) data, as interpreted
by University of Alabama at Huntsville (UAH) and
94
Remote Sensing System (RSS) teams, comparing evaluation of climate models requires the evaluation
both sets of results with AR4 GCM ΔT simulations of the simulated climate variability in terms of
for the period 1979–2010. teleconnection patterns.” Handorf and Dethloff
The three researchers report the RSS and UAH evaluated “the ability of state-of-the-art climate
ΔT time series “agree well with each other” and models to reproduce the low-frequency variability of
showed little trend over the period of record. By the mid-tropospheric winter flow of the Northern
contrast, they note there is “a steady positive trend” in Hemisphere in terms of atmospheric teleconnection
the model-simulated ΔT results, concluding the patterns.” To do so, they used the CMIP3 multimodel
significantly smaller ΔT trends from both the RSS ensemble for the period 1958–1999, for which
and UAH teams “indicate possible common errors reliable reanalysis data were available for
among AR4 GCMs.” In addition, they note the comparison.
tropical surface temperature trend of the multi-model The two researchers conclude, “current state-of-
ensemble mean is more than 60% larger than that the-art climate models are not able to reproduce the
derived from observations, “indicating that AR4 temporal behavior, in particular the exact phasing of
GCMs overestimate the warming in the tropics for the dominant patterns due to internally generated
1979–2010.” model variability.” In addition, they write, “the state-
In addition to greatly overestimating the tropical of-the-art climate models are not able to capture the
surface temperature trend, Fu et al. note, “it is evident observed frequency behavior and characteristic time
that the AR4 GCMs exaggerate the increase in static scales for the coupled runs satisfactorily ... in
stability between [the] tropical middle and upper accordance with Stoner et al. (2009) and Casado and
troposphere during the last three decades.” Pastor (2012),” both of which studies conclude, in
One year later, Po-Chedley and Fu (2012) write, their words, that “the models are not able to
“recent studies of temperature trend amplification in reproduce the temporal characteristics of atmospheric
the tropical upper troposphere relative to the lower- teleconnection time-series.”
middle troposphere found that coupled atmosphere- “In light of the strong need to provide reliable
ocean models from CMIP3 exaggerated this assessments of decadal predictability,” Handorf and
amplification compared to satellite microwave Dethloff additionally state “the potential of
sounding unit (Fu et al., 2011) and radiosonde (Seidel atmospheric teleconnections for decadal predictability
et al., 2012) observations.” The two authors “revisit needs further investigations” that “require a better
this issue using atmospheric GCMs with prescribed understanding of [1] the underlying mechanisms of
historical sea surface temperatures (SSTs) and variability patterns and flow regimes, [2] an
coupled atmosphere-ocean GCMs that participated in improvement of the skill of state-of-the-art climate
the latest model intercomparison project, CMIP5.” and Earth system models in reproducing atmospheric
The pair of researchers report their work teleconnections and [3] the identification of sources
demonstrated “even with historical SSTs as a for long-range predictive skill of teleconnections.”
boundary condition, most atmospheric models exhibit In other research, Santer et al. (2011) examined
excessive tropical upper tropospheric warming the consistency between satellite observations of the
relative to the lower-middle troposphere as compared lower troposphere and climate model simulations of
with satellite-borne microwave sounding unit that atmospheric region. The 17 researchers set the
measurements.” In addition, they note, “the results average of the observed temperature trends of lengths
from CMIP5 coupled atmosphere-ocean GCMs are varying from 10 to 32 years against the distribution of
similar to findings from CMIP3 coupled GCMs.” the same trend as simulated by a collection of CMIP3
Focusing more on the mid-troposphere, Handorf climate models (see Figure 1.4.6.2.1). While the
et al. (2012) write, “atmospheric teleconnections lower troposphere temperature trends (TLT) “in the
describe important aspects of the low-frequency observational satellite datasets are not statistically
atmospheric variability on time-scales of months and unusual relative to model-based distributions of
longer,” adding “in light of the increased need to externally forced TLT trends” they found “it should
provide reliable statements about seasonal to decadal be qualified by noting that: 1) The multi-model
predictability, it is necessary that state-of-the-art average TLT trend is always larger than the average
climate models simulate the spatial and temporal observed TLT trend” and “2) As the trend fitting
behavior of atmospheric teleconnections period increases … average observed trends are
satisfactorily.” Therefore, they continue, “an increasingly more unusual with respect to the multi-
95
References
Casado, M. and Pastor, M. 2012. Use of variability modes

to evaluate AR4 climate models over the Euro-Atlantic
region. Climate Dynamics 38: 225–237.
Christy, J.R., Herman, B., Pielke Sr., R., Klotzbach, P.,
McNider, R.T., Hnilo, J.J., Spencer, R.W., Chase, T., and
Douglass, D. 2010. What do observational datasets say
about modeled troposphere temperature trends since 1979?
Remote Sensing 2: 2148–2169.
Fu, Q., Manabe, S., and Johanson, C.M. 2011. On the
warming in the tropical upper troposphere: models versus
Figure 1.4.6.2.1. A comparison between modeled and observations. Geophysical Research Letters 38: 10.1029/
observed trends in the average temperature of the lower 2011GL048101.
atmosphere, for periods ranging from 10 to 32 years
(during the period 1979 through 2010). The yellow is the 5- Handorf, D. and Dethloff, K. 2012. How well do state-of-
95 percentile range of individual model projections, the the-art atmosphere-ocean general circulation models
green is the model average, the red and blue are the average reproduce atmospheric teleconnection patterns? Tellus A
of the observations, as compiled by Remote Sensing 64: org/10.3402/tellusa.v64i0.19777.
Systems and University of Alabama in Huntsville Hansen, J.E., Sato M., Ruedy, R., Nazarenko, L., Lacis, A.,
respectively (adapted from Santer et al., 2011). Schmidt, G.A., Russell, G., Aleinov, I., Bauer, S., Bell, N.,
Cairns, B., Canuto, V., Chandler, M., Cheng, Y., Del
Genio, A., Faluvegi, G., Fleming, E., Friend, A., Hall, T.,
model distribution of forced trends.” Put another way, Jackman, C., Kelley, M., Kiang, N., Koch, D., Lean, J.,
over longer periods of time that include a more robust Lerner, J., Lo, K., Menon, S., Miller, R., Minnis, P.,
signal, the discrepancy between model simulations Novakov, T., Oinas, V., Perlwitz, Ja., Perlwitz, Ju., Rind,
and observed trends in the temperature of the lower D., Romanou, A., Shindell, D., Stone, P., Sun, S., Tausnev,
troposphere increases, with the model overestimates N., Thresher, D., Wielicki, B., Wong, T., Yao, M., and
being on the verge of statistical significance. Zhang, S. 2005. Efficacy of climate forcings. Journal of
Santer et al. (2013) performed “a multimodel Geophysical Research 110: 10.1029/2005JD005776.
detection and attribution study with climate model Po-Chedley, S. and Fu, Q. 2012. Discrepancies in tropical
simulation output and satellite-based measurements of upper tropospheric warming between atmospheric
tropospheric and stratospheric temperature change,” circulation models and satellites. Environmental Research
using “simulation output from 20 climate models Letters 7: 10.1088/1748-9326/7/4/044018.
participating in phase 5 of the Coupled Model
Santer, B.D., Mears, C., Doutriaux, C., Caldwell, P.,
Intercomparison Project,” which “provides estimates
Gleckler, P.J., Wigley, T.M.L., Solomon, S., Gillett, N.P.,
of the signal pattern in response to combined Ivanova, D., Karl, T.R., Lanzante, J.R., Meehl, G.A., Stott,
anthropogenic and natural external forcing and the P.A., Taylor, K.E., Thorne, P.W., Wehner, M.F., and
noise of internally generated variability.” Among Wentz, F.J. 2011. Separating signal and noise in
other things, the 21 researchers report “most models atmospheric temperature changes: the importance of
do not replicate the size of the observed changes,” in timescale. Journal of Geophysical Research 116: 10.1029/
that “the models analyzed underestimate the observed 2011JD016263.
cooling of the lower stratosphere and overestimate the
Santer, B.D., Painter, J.F., Mears, C.A., Doutriaux, C.,
warming of the troposphere,” where warming, in their Caldwell, P., Arblaster, J.M., Cameron-Smith, P.J., Gillett,
opinion (Santer et al., 2003; Hansen et al., 2005), “is N.P., Gleckler, P.J., Lanzante, J., Perlwitz, J., Solomon, S.,
mainly driven by human-caused increases in well- Stott, P.A., Taylor, K.E., Terray, L., Thorne, P.W.,
mixed greenhouse gases,” and where “CMIP5 Wehner, M.F., Wentz, F.J., Wigley, T.M.L., Wilcox, L.J.,
estimates of variability on 5- to 20-year timescales are and Zou, C.-Z. 2013. Identifying human influences on
(on average) 55-69% larger than in observations.” atmospheric temperature. Proceedings of the National
Academy of Sciences USA 110: 26–33.
Santer, B.D., Wehner, M.F., Wigley, T.M.L., Sausen, R.,
Meehl, G.A., Taylor, K.E., Ammann, C., Arblaster, J.,
96
Washington, W.M., Boyle, J.S., and Bruggemann, W. Atlantic Current around the Grand Banks play
2003. Contributions of anthropogenic and natural forcing to important roles in affecting the atmospheric
recent tropopause height changes. Science 301: 479–483. circulation”; they further note “reducing these
Seidel, D.J., Free, M., and Wang, J.S. 2012. Reexamining coupled model errors could improve significantly the
the warming in the tropical upper troposphere: Models representation of the large-scale atmospheric
versus radiosonde observations. Geophysical Research circulation of the North Atlantic and European
Letters 39: 10.1029/2012GL053850 region.”
Stoner, A.M.K., Hayhoe, K., and Wuebbles, D.J. 2009.
References
Assessing general circulation model simulations of
atmospheric teleconnection patterns. Journal of Climate
22: 4348–4372. Keeley, S.P.E., Sutton, R.T., and Shaffrey, L.C. 2012. The
impact of North Atlantic sea surface temperature errors on
the simulation of North Atlantic European region climate.
1.3.7 Oceans Quarterly Journal of the Royal Meteorological Society
The vast majority of the surface of the planet consists 138: 1774–1783.
of oceans, which play a significant role in the Shaffrey, L.C., Stevens, I., Norton, W.A., Roberts, M.J.,
modulation and control of Earth’s climate. As is true Vidale, P.L., Harle, J.D., Jrrar, A., Stevens, D.P., Woodage,
of other factors we have previously discussed, M.J., Demory, M.E., Donners, J., Clark, D.B., Clayton, A.,
modeling of the oceans has proved to be a difficult Cole, J.W., Wilson, S.S., Connolley, W.M., Davies, T.M.,
task. Iwi, A.M., Johns, T.C., King, J.C., New, A.L., Slingo, J.M.,
Slingo, A., Steenman-Clark, L., and Martin, G.M. 2009.
U.K. HiGEM: The new U.K. High-Resolution Global
1.3.7.1 Atlantic Ocean
Environment Model: model description and basic
Keeley et al. (2012) note, “current state-of-the-art evaluation. Journal of Climate 22: 1861–1896.
climate models fail to capture accurately the path of
the Gulf Stream and North Atlantic Current,” and this
model failure “leads to a warm bias near the North 1.3.7.2 Pacific Ocean
American coast, where the modeled Gulf Stream Thompson and Kwon (2010) used a state-of-the-art
separates from the coast further north, and a cold coupled-atmosphere-ocean general circulation model
anomaly to the east of the Grand Banks of known as the Community Climate System Model
Newfoundland, where the North Atlantic Current version 3 (CCSM3) to demonstrate that climate
remains too zonal.” models have difficulty simulating interdecadal
Using a high-resolution coupled atmosphere- variations in the Pacific Ocean Circulation. The
ocean model (HiGEM), described in detail by authors used the model with atmospheric radiative
Shaffrey et al. (2009), Keeley et al. analyzed the forcing fixed to 1990 levels and with medium-scale
impacts of the sea surface temperature (SST) biases grid resolution. The model output was run for 700
created by the model in the North Atlantic in winter— years and a 100-year slice was selected starting with
approximately 8°C too cold to the east of the Grand year 500.
Banks of Newfoundland and 6°C too warm near the The model was unable to capture the natural
east coast of North America—on the mean climatic characteristics of the Kuroshio Current and its
state of the North Atlantic European region, along extension (KE), resulting in a poor representation of
with the variability associated with those model- climate variability in the North Pacific region. In
induced SST biases. nature, the mean position of the KE and the strongest
The three UK researchers say their results sea surface temperature (SST) gradients are typically
indicate the model-induced SST errors produce a separated by 500 km. In the model control run, the
mean sea-level pressure response “similar in two phenomena were coincident, resulting in
magnitude and pattern to the atmospheric circulation excessively strong SST variations (Figure 1.3.7.2.1).
errors in the coupled climate model.” They also note The Kuroshio Current, like its Atlantic
“errors in the coupled model storm tracks and North counterpart, the Gulf Stream, is a strong, narrow
Atlantic Oscillation, compared to reanalysis data, can current in the Northwest Pacific that runs along Japan
also be explained partly by these SST errors.” Keeley and continues into the North Pacific. The Oyashio
et al. conclude, “both [1] the error in the Gulf Stream Extension (OE) is a distinct current running along the
separation location and [2] the path of the North Kurile Islands and Hokkaido. This current is
97
Figure 1.3.7.2.1. The gradients of sea surface height (SSH) and SST for the (a) and (b) control run—CCSM3, (c)
generated by Maximenko and Niiler (2004), and (d) National Oceanic and Atmospheric Administration (NOAA). Adapted
from Thompson and Kwan (2010), their Figure 1.
analogous to the Labrador Current along Northeast circulation. Thus the reliability and accuracy of
North America. In the model, there was only one regional climate change model scenarios far from the
broad current, weaker than what is observed in nature. Pacific (e.g., agriculturally important continental
As a result, the SSTs are too warm near the Japan regions) will be strongly influenced by the model’s
coast and too cold into the Pacific. The long-term ability to capture Pacific region atmosphere and ocean
SST variations were also too strong in the model. circulations.
Interdecadal climate variations are internal As policymakers use climate simulations for the
forcings to Earth’s atmosphere system that, in concert creation of policy and for planning purposes, it is
with external forcing, will be important in important that biases in the climate models be
determining what our future climate may look like. In acknowledged and understood. As Thompson and
order to create reasonable scenarios of climate Kwon (2010) write, “it is important to note that
change, these phenomena must be modeled CCSM3 is not unique in its poor representation of the
accurately. Their strength and intensity also must be KOE. The diffuse front and lack of distinction
put into proper context with that of the total natural between the OE and KE is typical of low-resolution
and greenhouse forcing in order to attribute the source climate models.”
of climate change. In a paper published in the Journal of
Many studies have shown the importance of SST Geophysical Research, Guemas et al. (2012) also
gradients in contributing to the occurrence, strength, focused on this region of the Pacific Ocean, writing
and location of atmospheric phenomena such as jet “the North Pacific region has a strong influence on
streams, storm tracks, and blocking (e.g., Kung et al. North American and Asian climate.” They also state it
1990; 1992; 1993). These phenomena, as well as the is “the area with the worst performance in several
ocean circulations, are important in transporting heat state-of-the-art decadal climate predictions” and add
and momentum poleward and maintaining the general that the failure of essentially all climate models “to
98
represent two major warm sea surface temperature which data sets contain monthly mean values from
events occurring around 1963 and 1968 largely 1950–2008 gridded onto a global 2.5° x 2.5° latitude-
contributes to this poor skill,” noting “understanding longitude grid for SLP and a 2° x 2° grid for SST.
the causes of these major warm events is thus of The four U.S. scientists report model-derived
primary concern to improve prediction of North “temporal and spatial statistics of the North Pacific
Pacific, North American and Asian climate.” Ocean modes exhibit significant discrepancies from
The five researchers investigated “the reasons for observations in their twentieth-century climate, most
this particularly low skill,” identifying and describing visibly for the second mode, which has significantly
the two major warm events they say have been more low-frequency power and higher variance than
“consistently missed by every climate forecast in observations.” They also note the two dominant
system.” Based on their study of 11 observational modes of North Pacific oceanic variability “do not
data sets, Guemas et al. suggest the 1963 warm event exhibit significant changes in their spatial and
“stemmed from the propagation of a warm anomaly temporal characteristics under greenhouse warming,”
along the Kuroshio-Oyashio extension” and the 1968 stating “the ability of the models to capture the
warm event “originated from the upward transfer of a dynamics associated with the leading North Pacific
warm water mass centered at 200 meters depth.” They oceanic modes, including their link to corresponding
conclude, “biases in ocean mixing processes present atmospheric forcing patterns and to tropical
in many climate prediction models seem to explain variability, is questionable.”
the inability to predict these two major events.” There were even more “issues with the models,”
Although they acknowledge “reducing systematic Furtado et al. note. “In contrast with observations,”
biases in ocean stratification and improving the they report, “the atmospheric teleconnection excited
representation of ocean mixing processes has been a by the El Niño-Southern Oscillation in the models
long-standing effort,” Guemas et al. say their findings does not project strongly on the AL [Aleutian low]-
suggest allocating still more resources to “improving PDO coupled mode because of the displacement of
simulation of ocean mixing has the potential to the center of action of the AL in most models.” In
significantly improve decadal climate prediction.” addition, they note, “most models fail to show the
Furtado et al. (2011) note the North Pacific observational connection between El Niño Modoki-
Decadal Variability (NPDV) “is a key component in central Pacific warming and NPO [North Pacific
predictability studies of both regional and global Oscillation] variability in the North Pacific.” They
climate change,” adding that “two patterns of climate write, “the atmospheric teleconnections associated
variability in the North Pacific generally characterize with El Niño Modoki in some models have a
NPDV.” These two “dominant modes,” as they refer significant projection on, and excite the AL-PDO
to them, are the Pacific Decadal Oscillation (PDO; coupled mode instead.”
Mantua et al., 1997) and the recently identified North Furtado et al. conclude “for implications on
Pacific Gyre Oscillation (NPGO; Di Lorenzo et al., future climate change, the coupled climate models
2008). The scholars emphasize that given the links show no consensus on projected future changes in
between the PDO and the NPGO and global climate, frequency of either the first or second leading pattern
the accurate characterization and the degree of of North Pacific SST anomalies” and “the lack of a
predictability of these two modes in coupled climate consensus in changes in either mode also affects
models is an important “open question in climate confidence in projected changes in the overlying
dynamics.” atmospheric circulation.” In addition, they note the
Furtado et al. investigate this situation by lack of consensus they found “mirrors parallel
comparing the output from the 24 coupled climate findings in changes in ENSO behavior conducted by
models used in the IPCC AR4 with observational van Oldenborgh et al. (2005), Guilyardi (2006) and
analyses of sea level pressure (SLP) and sea surface Merryfield (2006),” and they state these significant
temperature (SST), based on SLP data from the issues “most certainly impact global climate change
National Centers for Environmental Prediction predictions.”
(NCEP)–National Center for Atmospheric Research Climate in the Southeast Pacific (SEP) near the
(NCAR) Reanalysis Project (Kistler et al., 2001), and coast of Peru and Chile, in the words of Zheng et al.
SST data from the National Oceanic and Atmospheric (2011), “is controlled by complex upper-ocean,
Administration (NOAA) Extended Reconstruction marine boundary layer and land processes and their
SST dataset, version 3 (Smith et al., 2008), both of interactions,” and they say variations in this system
99
have “significant impacts on global climate,” citing Furtado, J.C., Di Lorenzo, E., Schneider, N., and Bond,
Ma et al. (1996), Miller (1977), Gordon et al. (2000), N.A. 2011. North Pacific decadal variability and climate
and Xie (2004). However, they write, “it is well change in the IPCC AR4 models. Journal of Climate 24:
known that coupled atmosphere-ocean general 3049–3067.
circulation models tend to have systematic errors in Gordon, C.T., Rosati, A., and Gudgel, R. 2000. Tropical
the SEP region, including a warm bias in sea surface sensitivity of a coupled model to specified ISCCP low
temperature and too little cloud cover,” as clouds. Journal of Climate 13: 2239–2260.
demonstrated by Mechoso et al. (1995), Ma et al.
Guemas, V., Doblas-Reyes, F.J., Lienert, F., Soufflet, Y.,
(1996), Gordon et al. (2000), McAvaney et al. (2001),
and Du, H. 2012. Identifying the causes of the poor decadal
Kiehl and Gent (2004), Large and Danabasoglu climate prediction skill over the North Pacific. Journal of
(2006), Wittenberg et al. (2006), and Lin (2007). Geophysical Research 117: 10.1029/2012JD018004.
Even though these biases have what Zheng et al. call
“important impacts” on the simulation of Earth’s Guilyardi, E. 2006. El Niño mean state-seasonal cycle
radiation budget and climate sensitivity, they note “it interactions in a multi-model ensemble. Climate Dynamics
is still uncertain whether a similar bias is evident in 26: 329–348.
most state-of-the-art coupled general circulation Kiehl, J.T. and Gent, P.R. 2004. The Community Climate
models [CGCMs] and to what extent the sea surface system Model, version 2. Journal of Climate 17: 3666–
temperature [SST] biases are model dependent.” 3682.
Using 20-year (1980–1999) model runs of the
Kistler, R., Kalnay, E., Collins, W., Saha, S., White, G.,
Climate of the Twentieth Century simulations of the
Woollen, J., Chelliah, M., Ebisuzaki, W., Kanamitsu, M.,
19 CGCMs that figured prominently in the (IPCC) Kousky, V., van den Dool, H., Jenne, R., and Fiorino M.
Fourth Assessment Report (AR4), Zheng et al. 2001. The NCEP-NCAR 50-year reanalysis: Monthly
examined systematic biases in SSTs under the stratus means CD-ROM and documentation. Bulletin of the
cloud deck in the SEP and upper-ocean processes American Meteorological Society 82: 247–267.
relevant to those biases, attempting to isolate their
causes. Kung, E.C., Susskind, J., and DaCamara, C.C. 1993.
Prominent northern hemisphere winter blocking episodes
The four U.S. researchers discovered
and associated anomaly fields of sea surface temperatures.
“pronounced warm SST biases in a large portion of Terrestrial Atmospheric and Oceanic Sciences 4: 273–291.
the southeast Pacific stratus region … in all models”
and “negative biases in net surface heat fluxes … in Kung, E.C., Min, W., Susskind, J., and Park, C.K. 1992.
most of the models.” They found “biases in heat An analysis of simulated summer blocking episodes.
transport by Ekman currents largely contribute to Quarterly Journal of the Royal Meteorological Society
warm SST biases both near the coast and the open 118: 351–363.
ocean” and “in the coastal area, southwestward Kung, E.C., C.C. DaCamara, W.E. Baker, J. Susskind, and
Ekman currents and upwelling in most models are C.K. Park, 1990: Simulations of winter blocking episodes
much weaker than observed.” “In the open ocean,” using observed sea surface temperatures. Quarterly Journal
they observed, “warm advection due to Ekman of the Royal Meteorological Society 116: 1053–1070.
currents is overestimated.” They write, “negative Large, W.G. and Danabasoglu, G. 2006. Attribution and
biases (cooling the ocean) in net surface heat flux” impacts of upper-ocean biases in CCSM3. Journal of
and “positive biases in shortwave radiation” are found Climate 19: 2325–2346.
in most models, because most models “do not
generate sufficient stratus clouds” and “underestimate Lin, J.-L. 2007. The double-ITCZ problem in IPCC AR4
alongshore winds and coastal upwelling.” coupled GCMs: Ocean-atmosphere feedback analysis.
Journal of Climate 20: 4497–4525.
References Ma, C.-C., Mechoso, C.R., Robertson, A.W., and Arakawa,
A. 1996. Peruvian stratus clouds and the tropical Pacific
Di Lorenzo, E., Schneider, N., Cobb, K.M., Franks, P.J.S., circulation: A coupled ocean-atmosphere GCM study.
Chhak, K., Miller, A.J., McWilliams, J.C., Bograd, S.J., Journal of Climate 9: 1635–1645.
Arango, H., S.J., Curchitser, E., Powell, T.M., and Rivière,
P. 2008. North Pacific Gyre Oscillation links ocean climate Mantua, N.J., Hare, S.R., Zhang, Y., Wallace, J.M., and
and ecosystem change. Geophysical Research Letters 35: Francis, R. 1997. A Pacific interdecadal climate oscillation
10.1029/2007GL032838. with impacts on salmon production. Bulletin of the
American Meteorological Society 78: 1069–1079.
100
Maximenko, N.A., and P.P. Niiler, 2004: Hybrid decade- Wittenberg, A.T., Rosati, A., Lau, N.-C., and Ploshay, J.J.
mean global sea level with mesoscale resolution. Recent 2006. GFDL’s CM2 global coupled climate models. Part
advances in Marine Science and Technology, N. Saxena, III: Tropical Pacific climate and ENSO. Journal of Climate
Ed., PACON International, 55–59. 19: 698–722.
McAvaney, B., Covey, C., Joussaume, S., Kattsov, V., Xie, S.-P. 2004. The shape of continents, air-sea
Kitoh, A., Ogana, W., Pitman, A.J., Weaver, A.J., Wood, interaction, and the rising branch of the Hadley circulation.
R.A., Zhao, Z.-C., AchutaRao, K., Arking, A., Barnston, In: Diaz, H.F. and Bradley, R.S. (Eds.). The Hadley
A., Betts, R., Bitz, C., Boer, G., Braconnot, P., Broccoli, Circulation: Past, Present and Future. Advances in Global
A., Bryan, F., Claussen, M., Colman, R., Delecluse, P., Del Change Research 25: 121–152.
Genio, A., Dixon, K., Duffy, P., Dümenil, L., England, M.,
Fichefet, T., Flato, G., Fyfe, J.C., Gedney, N., Gent, P., Zheng, Y., Shinoda, T., Lin, J.-L., and Kiladis, G.N. 2011.
Genthon, C., Gregory, J., Guilyardi, E., Harrison, S., Sea surface temperature biases under the stratus cloud deck
Hasegawa, N., Holland, G., Holland, M., Jia, Y., Jones, in the Southeast Pacific Ocean in 19 IPCC AR4 coupled
P.D., Kageyama, M., Keith, D., Kodera K., Kutzbach, J., general circulation models. Journal of Climate 24: 4139–
Lambert, S., Legutke, S., Madec, G., Maeda, S., Mann, 4164.
M.E., Meehl, G., Mokhov, I., Motoi, T., Phillips, T.,
Polcher, J., Potter, G.L., Pope, V., Prentice, C., Roff, G.,
Semazzi, F., Sellers, P., Stensrud, D.J., Stockdale, T., 1.3.7.3 Indian Ocean
Stouffer, R., Taylor, K.E., Trenberth, K., Tol, R., Walsh, J., Nagura et al. (2013) explain the term “Seychelles
Wild, M., Williamson, D., Xie, S.-P., Zhang, X.-H., and Dome” refers to the shallow climatological
Zwiers, F. 2001. Model evaluation. In: Houghton, J.T., thermocline in the southwestern Indian Ocean,
Ding, Y., Griggs, D.J., Noguer, M., van der Linden, P.J., “where ocean wave dynamics efficiently affect sea
Dai, X., Maskell, K., and Johnson, C.A. (Eds.). Climate surface temperature [SST], allowing SST anomalies
Change 2001: The Scientific Basis. Contribution of to be predicted up to 1–2 years in advance.” They also
Working Group I of the Third Assessment Report of the indicate this ability is significant: “The anomalous
Intergovernmental Panel on Climate Change. Cambridge
SSTs in the dome region subsequently impact various
University Press, Cambridge, United Kingdom, pp. 471–
523. atmospheric phenomena, such as tropical cyclones
(Xie et al., 2002), the onset of the Indian summer
Mechoso, C.R., Robertson, A.W., Barth, N., Davey, M.K., monsoon (Joseph et al., 1994; Annamalai et al., 2005)
Delecluse, P., Gent, P.R., Ineson, S., Kirtman, B., Latif, and precipitation over India and Africa (Annamalai et
M., Le Treut, H., Nagai, T., Neelin, J.D., S.G.H., Polcher, al., 2007; Izumo et al., 2008).”
J., Stockdale, T., Terray, L., Thual, O., and Tribbia, J.J. They note “Yokoi et al. (2009) examined the
1995. The seasonal cycle over the tropical Pacific in
outputs from models used in phase three of the
coupled ocean-atmosphere general circulation models.
Monthly Weather Review 123: 2825–2838. Coupled Model Intercomparison Project (CMIP3) and
found that many CMIP3 models have serious biases
Merryfield, W.J. 2006. Changes to ENSO under CO2 in this region.” Hoping to find some improvement in
doubling in a multimodel ensemble. Journal of Climate 19: the four years since Yokoi et al. conducted their
4009–4027. research, Nagura et al. examined model biases
Miller, R.L. 1997. Tropical thermostats and low cloud associated with the Seychelles Dome using state-of-
cover. Journal of Climate 10: 409–440. the-art ocean-atmosphere coupled general circulation
models (CGCMs), “including those from phase five
Smith, T.M., Reynolds, R.W., Peterson, T.C., and of the Coupled Model Intercomparison Project
Lawrimore, J. 2008. Improvements to NOAA’s historical (CMIP5).”
merged land-ocean surface temperature analysis (1880–
Nagura et al. report several of the tested models
2006). Journal of Climate 21: 2283–2296.
“erroneously produce an upwelling dome in the
Thompson, L. and Y.O. Kwon, 2010: An enhancement of eastern half of the basin, whereas the observed
Low-Frequency Variability in the Kuroshio-Oyashio Seychelles Dome is located in the southwestern
Extension in CCM3 owing to Ocean Model Biases. Journal tropical Indian Ocean.” They also note “the annual
of Climate 23, 6221-6233. DOI:10.1175/2010JCLI3402.1 mean Ekman pumping velocity in these models is
van Oldenborgh, G.J., Philip, S.Y., and Collins, M. 2005. found to be almost zero in the southern off-equatorial
El Niño in a changing climate: A multi-model study. Ocean region,” which “is inconsistent with observations, in
Science 1: 81–95. which Ekman upwelling acts as the main cause of the
Seychelles Dome.” Moreover, “in the models
101
reproducing an eastward-displaced dome, easterly Cai, W. and Cowan, T. 2013. Why is the amplitude of the
biases are prominent along the equator in boreal Indian Ocean Dipole overly large in CMIP3 and CMIP5
summer and fall, which result in shallow thermocline climate models? Geophysical Research Letters 40: 1200–
biases along the Java and Sumatra coasts via Kelvin 1205.
wave dynamics and a spurious upwelling dome in the Izumo, T., Montegut, C.D.B., Luo, J.-J., Behera, S.K.,
region.” In a revealing final assessment of their Masson, S., and Yamagata, T. 2008. The role of the
findings, Nagura et al. conclude “compared to the western Arabian Sea upwelling in Indian Monsoon rainfall
CMIP3 models, the CMIP5 models are even worse in variability. Journal of Climate 21: 5603–5623.
simulating the dome longitudes.”
Joseph, P.V., Eischeid, J.K., and Pyle, R.J. 1994.
The Indian Ocean Dipole (IOD) is an irregular Interannual variability of the onset of the Indian summer
oscillation of sea-surface temperatures in which the monsoon and its association with atmospheric features, El
western Indian Ocean becomes alternately warmer Niño, and sea surface temperature anomalies. Journal of
and then colder than the eastern part of the ocean. Cai Climate 7: 81–105.
and Cowan (2013) note “in most models, IOD peak-
season amplitudes are systematically larger than that Nagura, M., Sasaki, W., Tozuka, T., Luo, J.-J., Behera, S.,
and Yamagata, T. 2013. Longitudinal biases in the
of the observed,” and they say “understanding the
Seychelles Dome simulated by 35 ocean-atmosphere
cause of this bias is … essential for alleviating model coupled general circulation models. Journal of Geophysical
errors and reducing uncertainty in climate Research: Oceans 118: 1–16.
projections.”
The two Australian researchers analyzed sea Xie, S.-P., Annamalai, H., Schott, F.A., and McCreary Jr.,
surface temperatures (SSTs), thermocline J.P. 2002. Structure and mechanisms of south Indian Ocean
characteristics (20°C isotherm depth), and zonal wind climate variability. Journal of Climate 15: 864–878.
and precipitation outputs from 23 CMIP3 models and Yokoi, T., Tozuka, T., and Yamagata, T. 2009. Seasonal
33 CMIP5 models that attempted to simulate these variations of the Seychelles Dome simulated in the CMIP3
climatic features over the last half of the twentieth models. Journal of Climate 39: 449–457.
century, after which they compared the model
simulations with real-world observations. Cai and
Cowan report “most models generate an overly deep 1.3.7.4 Equatorial/Tropical Regions
western Indian Ocean thermocline that results in an Wan et al. (2011) state “the notorious tropical bias
unrealistic upward slope toward the eastern tropical problem in climate simulations of global coupled
Indian Ocean” and “the unrealistic thermocline general circulation models (e.g., Mechoso et al.,
structure is associated with too strong a mean easterly 1995; Latif et al., 2001; Davey et al., 2002; Meehl et
wind over the equatorial Indian Ocean, which is in al., 2005) manifests itself particularly strongly in the
turn supported by a slightly stronger mean west minus tropical Atlantic,” and “while progress towards
east SST gradient, reinforced by the unrealistic reducing tropical climate biases has been made in the
thermocline slope.” tropical Pacific over the past decades (e.g., Deser et
They conclude, “these biases/errors have al., 2006), little or no progress has been made in the
persisted in several generations of models,” such that tropical Atlantic (Breugem et al., 2006; Richter and
“there is no clear improvement from CMIP3 to Xie, 2008; Wahl et al., 2009).” They write, “the
CMIP5.” climate bias problem is still so severe that one of the
most basic features of the equatorial Atlantic Ocean—
References the eastward shoaling thermocline—cannot be
reproduced by most of the Intergovernmental Panel
Annamalai, H., Liu, P., and Xie, S.-P. 2005. Southwest on Climate Change (IPCC) assessment report (AR4)
Indian Ocean SST variability: Its local effect and remote models,” citing Richter and Xie (2008).
influence on Asian monsoons. Journal of Climate 18: In their own investigation of the subject, Wan et
4150–4167. al. “show that the bias in the eastern equatorial
Atlantic has a major effect on sea-surface temperature
Annamalai, H., Okajima, H., and Watanabe, M. 2007.
Possible impact of the Indian Ocean SST on the Northern
(SST) response to a rapid change in the Atlantic
Hemisphere circulation during El Niño. Journal of Climate Meridional Overturning Circulation (AMOC).” They
20: 3164–3189. do so by exemplifying the problem “through an inter-
model comparison study of tropical Atlantic response
102
to an abrupt change in [the] AMOC using the atmosphere-ocean coupling, climate models with
Geophysical Fluid Dynamics Laboratory (GFDL) prescribed observed radiative forcing changes do not
Coupled Climate Model (CM2.1) and the National capture the pattern of the observed tropical oceanic
Center for Atmospheric Research (NCAR) warming suggests that either the radiatively forced
Community Climate System Model (CCSM3)” and component of this warming pattern was sufficiently
dissecting the oceanic mechanisms responsible for the small in recent decades to be dwarfed by natural
difference in the models’ SST responses. tropical SST variability, or that the coupled models
Their analysis demonstrates the different SST are misrepresenting some important tropical physics.”
responses of the two models are “plausibly attributed Since the greenhouse-gas forcing of climate “in recent
to systematic differences in the simulated tropical decades” is claimed by the IPCC to have been
Atlantic ocean circulation.” The ultimate implication unprecedented over the past millennium or more, it
of Wan et al.’s findings is, in their words, that “in would appear the models are indeed “misrepresenting
order to accurately simulate past abrupt climate some important tropical physics.”
changes and project future changes, the bias in In the introduction to their study of the origins of
climate models must be reduced.” tropical-wide sea surface temperature (SST) biases in
Shin and Sardeshmukh (2011) note there is CMIP multi-model ensembles, Li and Xie (2012)
“increased interest in the ability of climate models to write, “state-of-the-art coupled ocean-atmosphere
simulate and predict surface temperature and general circulation models (CGCMs) suffer from
precipitation changes on sub-continental scales,” and large errors in simulating tropical climate, limiting
they state these regional trend patterns “have been their utility in climate prediction and projection.”
strongly influenced by the warming pattern of the Describing the size of the several errors, they say they
tropical oceans,” which suggests correctly simulating “are comparable or larger in magnitude than observed
the warming pattern of the tropical oceans is a interannual variability and projected change in the
prerequisite for correctly simulating sub-continental- 21st century.”
scale warming patterns. The two researchers say the “well-known errors”
In exploring this subject further, Shin and include “too weak a zonal SST gradient along the
Sardeshmukh compared multi-model ensemble equatorial Atlantic,” citing Davey et al. (2002) and
simulations of the last half-century with Richter and Xie (2008); “an equatorial cold tongue
corresponding observations, focusing on the world’s that penetrates too far westward in the Pacific,” citing
tropical oceans, as well as the land masses Mechoso et al. (1995) and de Szoeke and Xie (2008);
surrounding the North Atlantic Ocean, including and “too warm SSTs over the tropical Southeast
North America, Greenland, Europe, and North Africa. Pacific and Atlantic, and a spurious double
This was done, as they describe it, using “all available intertropical convergence zone,” citing Lin (2007).
coupled [atmosphere-ocean] model simulations of the They point out “these errors have persisted in several
period 1951–1999 from 18 international modeling generations of models for more than a decade.”
centers, generated as part of the IPCC’s 20th century Closer inspection of zonal SST profiles along the
climate simulations with prescribed time-varying equator is found by Li and Xie to reveal “basin-wide
radiative forcings associated with greenhouse gases, offsets, most obvious in the Indian (Saji et al., 2006)
aerosols, and solar variations.” but visible in the Pacific (de Szoeke and Xie, 2008),
The two researchers report “the tropical oceanic and Atlantic (Richter and Xie, 2008) Oceans.” They
warming pattern is poorly represented in the coupled add “it is unclear whether such basin-wide offset
simulations,” and they state their analysis “points to errors are limited to individual tropical basins or
model error rather than unpredictable climate noise as shared among all the basins.”
a major cause of this discrepancy with respect to the Li and Xie then “examine the Climate of the
observed trends.” They found “the patterns of recent Twentieth Century (20C3M) simulations (also termed
climate trends over North America, Greenland, as ‘historical’ runs) from 22 CMIP3 and 21 CMIP5
Europe, and North Africa are generally not well CGCMs for a 30-year period of 1970–99, together
captured by state-of-the-art coupled atmosphere- with their available Atmospheric Model
ocean models with prescribed observed radiative Intercomparison Project (AMIP) runs that are forced
forcing changes.” with the observed SST.” They find two types of
Commenting on their work, Shin and tropical-wide offset biases. One of them “reflects the
Sardeshmukh write, “the fact that even with full tropical mean SST differences from observations and
103
among models, with a broad meridional structure and In looking at equatorial SSTs, Vanniere et al.
of the same sign across all basins of up to 2°C in (2013) state “the cold equatorial SST bias in the
magnitude,” while the other “is linked to inter-model tropical Pacific that is persistent in many coupled
variability in the cold tongue temperatures in the OAGCMs [Ocean-Atmosphere Global Climate
equatorial Pacific and Atlantic.” Models] severely impacts the fidelity of the simulated
In further describing their findings, the two climate and variability in this key region, such as the
researchers write “the first-type offset error is ENSO [El Niño-Southern Oscillation] phenomenon.”
ascribed to atmospheric model representation of cloud More specifically, they note “the seasonal equatorial
cover, with cloudy models biasing low in tropical- cold tongue extends too far west, is too cold in the
wide SST,” while “the second type originates from east Pacific and is associated with too strong trade
the diversity among models in representing the winds,” citing Davey et al. (2001), AchutaRao and
thermocline depth, with deep thermocline models Sperber (2006), and Lin (2007). In addition, they
biasing warm in the equatorial cold tongues.” write, “a warm SST bias is observed near the coast of
Also addressing the ability of CGCMs to South America,” which is “associated with a lack of
accurately model the equatorial and tropical regions, low clouds and deficient winds.” And they note
Zheng et al. (2012) note “the equatorial Pacific is “mean state biases relevant to ENSO also include too
observed to have a minimum sea surface temperature strong easterlies in the west Pacific and the double
(SST) that extends from the west coasts of the ITCZ [Intertropical Convergence Zone] syndrome,”
Americas into the central Pacific,” and this “extension citing Lin (2007) and de Szoeke and Xie (2008).
of cool water is commonly referred to as the cold In attempting to unscramble and resolve these
tongue (Wyrtki, 1981).” They state “it is generally many problems, Vanniere et al. (2013) used seasonal
argued that the Pacific cold tongue is maintained by re-forecasts or hindcasts to “track back” the origin of
horizontal advection of cold water from the east and the major cold bias, so that “a time sequence of
by upwelling of cold water from the subsurface.” The processes involved in the advent of the final mean
three researchers proceed to examine “the state errors can then be proposed,” applying this
contribution of ocean dynamics to sea surface strategy to the ENSEMBLES-FP6 project multi-
temperature biases in the eastern Pacific cold tongue model hindcasts of the last decades. The researchers
region in fifteen coupled general circulation models discovered “the models are able to reproduce either El
(CGCMs) participating in the Intergovernmental Niño or La Niña close to observations, but not both.”
Panel on Climate Change (IPCC) Fourth Assessment Therefore, they conclude, “more work is needed to
Report (AR4),” analyzing “twenty years (1980–1999) understand the origin of the zonal wind bias in
of the twentieth-century climate simulations from models,” and, in this regard, “understanding the
each model.” dynamical and thermodynamical mechanisms that
Zheng et al. find “errors in both net surface heat drive the tropical atmosphere is required both to
flux and total upper ocean heat advection significantly alleviate OAGCM errors and to describe the full
contribute to the excessive cold tongue in the extent of the atmosphere’s role in tropical variability,
equatorial Pacific” and conclude “the stronger heat such as ENSO.”
advection in the models is caused by overly strong
horizontal heat advection associated with too strong References
zonal currents, and overly strong vertical heat
advection due to excessive upwelling and the vertical AchutaRao, K. and Sperber, K. 2006. ENSO simulations in
gradient of temperature.” They note “the Bjerknes coupled ocean-atmosphere models: are the current models
feedback in the coupled models is shown to be better? Climate Dynamics 27: 1–16.
weaker than in observations, which may be related to Breugem, W.P., Hazeleger, W. and Haarsma, R.J. 2006.
the insufficient response of surface zonal winds to Multimodel study of tropical Atlantic variability and
SST in the models and an erroneous subsurface change. Geophysical Research Letters 33: 10.1029/
temperature structure,” such that “the cold tongue 2006GL027831.
becomes colder than the cold tongue in the
Davey, M.K., Huddleston, M., Sperber, K., Braconnot, P.,
observations.” Zheng et al. conclude “more work is
Bryan, F., Chen, D., Colman, R., Cooper, C., Cubasch, U.,
needed on the role of the ocean model and ocean- Delecluse, P., DeWitt, D., Fairhead, L., Flato, G., Gordon,
atmosphere feedback in the growth of the double- C., Hogan, T., Ji, M., Kimoto, M., Kitoh, A., Knutson, T.,
ITCZ pattern.” Latif, M., LeTreut, H., Li, T., Manabe, S., Mechoso, C.,
104
Meehl, G., Power, S., Roeckner, E., Terray, L., Vintzileos, and Woolnough, S. 2013. Using seasonal hindcasts to
A., Voss, R., Wang, B., Washington, W., Yoshikawa, I., understand the origin of the equatorial cold tongue bias in
Yu, J., Yukimoto, S., and Zebiak, S. 2002. STOIC: a study CGCMs and its impact on ENSO. Climate Dynamics 40:
of coupled model climatology and variability in tropical 963–981.
ocean regions. Climate Dynamics 18: 403–420.
Wahl, S., Latif, M., Park, W., and Keenlyside, N. 2009. On
de Szoeke, S.P. and Xie, S.P. 2008. The tropical Eastern the tropical Atlantic SST warm bias in the Kiel Climate
Pacific seasonal cycle: assessment of errors and Model. Climate Dynamics 33: 10.1007/s00382-009-0690-
mechanisms in IPCC AR4 coupled ocean atmosphere 9.
general circulation models. Journal of Climate 21: 2573–
2590. Wan, X., Chang, P., Jackson, C.S., Ji, L., and Li, M. 2011.
Plausible effect of climate model bias on abrupt climate
Deser, C., Capotondi, A., Saravanan, R., and Phillips, A.S. change simulations in Atlantic sector. Deep-Sea Research
2006. Tropical Pacific and Atlantic climate variability in II 58: 1904–1913.
CCSM3. Journal of Climate 19: 2451–2481.
Wyrtki, K. 1981. An estimate of equatorial upwelling in the
Latif, M., Sperber, K., Arblaster, J., Braconnot, P., Chen, Pacific. Journal of Physical Oceanography 11: 1205–1214.
D., Colman, A., Cubasch, U., Cooper, C., Delecluse, P.,
Dewitt, D., Fairhead, L., Flato, G., Hogan, T., Ji, M., Zheng, Y., Lin, J.-L., and Shinoda, T. 2012. The equatorial
Kimoto, M., Kitoh, A., Knutson, T., Le Treut, H., Li, T., Pacific cold tongue simulated by IPCC AR4 coupled
Manabe, S., Marti, O., Mechoso, C., Meehl, G., Power, S., GCMs: Upper ocean heat budget and feedback analysis.
Roeckner, E., Sirven, J., Terray, L., Vintzileos, A., Voss, Journal of Geophysical Research 117: 10.1029/
R., Wang, B., Washington, W., Yoshikawa, I., Yu, J., and 2011JC007746.
Zebiak, S. 2001. ENSIP: the El Niño simulation
intercomparison project. Climate Dynamics 18: 255–276.
1.3.7.5 Southern Ocean
Li, G. and Xie, S.-P. 2012. Origins of tropical-wide SST Weijer et al. (2012) note “the Southern Ocean is a
biases in CMIP multi-model ensembles. Geophysical region of extremes: it is exposed to the most severe
Research Letters 39: 10.1029/2012GL053777. winds on the Earth (Wunsch, 1998), the largest ice
Lin, J.L. 2007. The double-ITCZ problem in IPCC AR4 shelves (Scambos et al., 2007), and the most
coupled GCMs: ocean-atmosphere feedback analysis. extensive seasonal sea ice cover (Thomas and
Journal of Climate 20: 4497–4525. Dieckmann, 2003).” They indicate various
interactions among the atmosphere, ocean, and
Mechoso, C.R., Roberston, A.W., Barth, N., Davey, M.K., cryosphere in this region “greatly influence the
Delecluse, P., Gent, P.R., Ineson, S., Kirtman, B., Latif,
dynamics of the entire climate system through the
M., Le Treut, H., Nagai, T., Neelin, J.D., Philander, S.G.H.,
Polcher, J., Schopf, P.S., Stockdale, T., Suarez, M.J., formation of water masses and the sequestration of
Terray, L., Thual, O., and Tribbia, J.J. 1995. The seasonal heat, freshwater, carbon, and other properties (Rintoul
cycle over the tropical Pacific in general circulation et al., 2001).”
models. Monthly Weather Review 123: 2825–2838. Against this backdrop, Weijer et al. “explored
several key aspects of the Southern Ocean and its
Meehl, G.A., Covey, C., McAvaney, B., Latif, M., and climate in the new Community Climate System
Stouffer, R.J. 2005. Overview of the coupled model
Model, version 4 (CCSM4),” including “the surface
intercomparison project. Bulletin of the American
Meteorological Society 86: 89–93. climatology and inter-annual variability, simulation of
key climate water masses (Antarctic Bottom Water
Richter, I. and Xie, S.-P. 2008. On the origin of equatorial [AABW], Subantarctic Mode Water [SAMW], and
Atlantic biases in coupled general circulation models. Antarctic Intermediate Water [AAIW]), the transport
Climate Dynamics 31: 587–595. and structure of the Antarctic Circumpolar Current
Saji, N.H., Xie, S.-P., and Yamagata, T. 2006. Tropical [ACC], and inter-basin exchange via the Agulhas and
Indian Ocean variability in the IPCC twentieth century Tasman leakages and at the Brazil-Malvinas
climate simulations. Journal of Climate 19: 4397–4417. Confluence [BMC].”
The nine researchers find “the CCSM4 has
Shin, S.-I. and Sardeshmukh, P.D. 2011. Critical influence varying degrees of accuracy in the simulation of the
of the pattern of Tropical Ocean warming on remote
climate of the Southern Ocean when compared with
climate trends. Climate Dynamics 36: 1577–1591.
observations.” Results of this comparison include: (1)
Vanniere, B., Guilyardi, E., Madec, G., Doblas-Reyes, F.J., “the seasonally ice-covered regions are mildly colder
105
(ΔSST > -2°C) than observations,” (2) “sea ice extent (2004). They add “one of the most pressing issues in
is significantly larger than observed,” (3) “north of oceanography is to understand the rate, the structure
the seasonal ice edge, there is a strong (-4°C < ΔSST and the controls of the water mass overturning
< -1°C) cold bias in the entire Pacific sector south of circulation in the Southern Ocean and to accurately
50°S and in the western Australian-Antarctic Basin,” represent these aspects in climate models.” Focusing
(4) “positive biases (1° < ΔSST < 4°C) are found in on five water masses crucial for the Southern Ocean
the Indian and Atlantic sectors of the Southern overturning circulation—surface subTropical Water
Ocean,” (5) “significant differences are found in the (TW), Mode Water (MW), Intermediate Water (IW),
Indian and Pacific sectors north of the ACC, with the Circumpolar Deep Water (CDW), and Antarctic
CCSM4 model being too cold (< -2°C) and fresh (<- Bottom Water (AABW)—Sallee et al. studied the
0.3 psu),” (6) “AABW adjacent to the Antarctic ability of 21 of the CMIP5 models to simulate what
continent is too dense,” (7) “North Atlantic Deep they describe as the most basic properties of each of
Water is too salty (>0.2 psu),” (8) “in the Indian and these water masses: temperature, salinity, volume,
Pacific sectors of the Southern Ocean, north of 50°S and outcrop area.
and below 3000 meters, the too-salty AABW The authors describe several important findings.
penetrates northward, resulting in a denser-than- They note, “the water masses of the Southern Ocean
observed abyssal ocean in CCSM4,” (9) “the model in the CMIP5 models are too warm and light over the
underestimates the depth of the deep winter mixed entire water column,” with the largest biases being
layers in the Indian and eastern Pacific sectors of the found in the ventilated layers, and “the mode water
Southern Ocean north of the ACC,” (10) “in the layer is poorly represented in the models and both
southern Tasman Sea and along the eastern Indian mode and intermediate water have a significant fresh
Ocean boundary ... the model mixed layer depth is bias.” They further find, “in contrast to observations
deeper than observed by more than 400 meters,” (11) (e.g., Rintoul, 2007), bottom water is simulated to
“in all sectors of the Southern Ocean, Model CFC-11 become slightly saltier” and “when compared to
concentrations in the lower thermocline and observation-based reconstructions,” the models
intermediate waters are lower than observed,” (12) “exhibit a slightly larger rate of overturning at
“model CFC-11 concentrations in the deep ocean shallow to intermediate depths, and a slower rate of
(below 2000 meters) are lower than observed in the overturning deeper in the water column.” Given such
basins adjacent to the Antarctic continent,” (13) discrepancies, the seven scientists conclude “many of
“model surface CFC-11 concentrations are higher the biases and future changes identified in this study
than observed,” (14) “the production of overflow are expected to have significant impacts on the
waters in the Ross Sea is too low by about a factor of marine carbon cycle.” These biases and the changes
2 relative to the limited observations,” (15) “the depth they spawn are not trivial and must be corrected
at which the product water settles was also shown to before they are used to forecast the future of the
be too shallow by about a factor of 2,” (16) “the overturning circulation of the Southern Ocean and its
subtropical gyre of the South Atlantic is too strong by impact on global climate.
almost a factor of 2, associated with a strong bias in Heuze et al. (2013) point out “the ability of a
the wind stress,” (17) the mean position of the BMC model to adequately depict bottom water formation is
is too far south in the CCSM4,” and (18) “the model crucial for accurate prediction of changes in the
variability in the position of the BMC is significantly thermohaline circulation,” citing Hay (1993). They
less than observations.” note, however, “this process is particularly
Weijer et al. conclude that as the CCSM4 challenging to represent in the current generation of
currently stands, it “may underestimate the climate models” and “the last generation of models in
sequestration of heat, carbon, and other properties to CMIP3 poorly represented Southern Ocean transport
the interior ocean,” such that its parameterizations and heat fluxes,” citing Russell et al. (2006).
may “lead to significant biases in the representation Heuze et al. assessed “Southern Ocean potential
of the Southern Ocean and its climate.” temperature, salinity, density and sea ice
According to Sallee et al. (2013), “the Southern concentration in fifteen CMIP5 historical simulations
Ocean is the dominant anthropogenic carbon sink of (means of the twenty August monthly mean fields
the world’s oceans and plays a central role in the from 1986 to 2005),” after which they compared the
redistribution of physical and biogeochemical 20-year mean model fields with historical
properties around the globe,” citing Sarmiento et al. hydrographic data and Hadley Centre sea ice
106
climatologies. They note no model reproduces the Wunsch, C. 1998. The work done by the wind on the
process of Antarctic bottom water formation oceanic general circulation. Journal of Physical
accurately, for “instead of forming dense water on the Oceanography 28: 2332–2340.
continental shelf and allowing it to spill off,” the
models “present extensive areas of deep convection,
1.3.7.6 Sea Ice
thus leading to an unrealistic unstratified open
Writing about how well the models of more than a
ocean.”
decade ago simulated changes in sea ice, Holland
(2001) states “the present situation with respect to the
References
state-of-the-art global climate models is that some
physical processes are absent from the models and,
Hay, W.W. 1993. The role of polar deep water formation in
global climate change. Annual Reviews of Earth and with the rather coarse-resolution grids used, some
Planetary Sciences 21: 227–254. physical processes are ill resolved ... and therefore in
practical terms missing from the simulations.”
Heuze, C., Heywood, K.J., Stevens, D.P., and Ridley, J.K. Holland thus questions “whether the simulations
2013. Southern Ocean bottom water characteristics in obtained from such models are in fact physically
CMIP5 models. Geophysical Research Letters 40: 1409– meaningful” and conducted an analysis to determine
1414.
the difference in model evolution of sea ice cover
Rintoul, S.R. 2007. Rapid freshening of Antarctic Bottom using a relatively coarse-resolution grid versus a fine-
Water formed in the Indian and Pacific oceans. resolution grid, with specific emphasis on the
Geophysical Research Letters 34: 10.1029/2006GL028550. presence and treatment of a mesoscale ocean eddy
and its influence on sea ice cover.
Rintoul, S.R. and Sokolov, S. 2001. Baroclinic transport
variability of the Antarctic Circumpolar Current south of Resolving the ocean eddy field using the fine-
Australia (WOCE repeat section SR3). Journal of resolution model was found to have a measurable
Geophysical Research 106: 2815–2832. impact on sea ice concentration, implying a “fine-
resolution grid may have a more efficient atmosphere-
Russell, J.L., Stouffer, R.J., and Dixon, K.W. 2006. sea ice-ocean thermodynamic exchange than a coarse
Intercomparison of the southern ocean circulations in IPCC one.” Holland concludes his study demonstrated “yet
coupled model control simulations. Journal of Climate 19:
again that coarse-resolution coupled climate models
4060–4075.
are not reaching fine enough resolution in the polar
Sallee, J.-B., Shuckburgh, E., Bruneau, N., Meijers, A.J.S., regions of the world ocean to claim that their
Bracegirdle, T.J., Wang, Z., and Roy, T. 2013. Assessment numerical solutions have reached convergence.”
of Southern Ocean water mass circulation and Two years later, the situation had not improved
characteristics in CMIP5 models: Historical bias and much. Laxon et al. (2003) used an eight-year time
forcing response. Journal of Geophysical Research series (1993–2001) of Arctic sea-ice thickness
(Oceans) 118: 1830–1844.
derived from measurements of ice freeboard made by
Sarmiento, J.L., Gruber, N., Brzezinski, M.A., and Dunne, 13.8-GHz radar altimeters carried aboard ERS-1 and
J.P. 2004. High-latitude controls of thermocline nutrients 2 satellites to determine the mean thickness and
and low latitude biological productivity. Nature 427: 56– variability of Arctic sea ice between latitudes 65 and
60. 81.5°N, a region covering the entire circumference of
Scambos, T.A., Haran, T.M., Fahnestock, M.A., Painter, the Arctic Ocean, including the Beaufort, Chukchi,
T.H., and Bohlander, J. 2007. MODIS-based Mosaic of East Siberian, Kara, Laptev, Barents, and Greenland
Antarctica (MOA) data sets: Continent-wide surface Seas. Mean winter sea-ice thickness over the region
morphology and snow grain size. Remote Sensing of was found to be 2.73 meters with a standard deviation
Environment 111: 242–257. of ± 9% of the average, a variability 50 percent
greater than predicted by climate models “and
Thomas, D.N. and Dieckmann, G. 2003. Sea Ice: An
probably more,” the authors state. They report their
Introduction to Its Physics, Chemistry, Biology, and
Geology. Wiley-Blackwell, Hoboken, New Jersey, USA. analysis “excludes variability that occurs over
timescales of longer than a decade.”
Weijer, W., Sloyan, B.M., Maltrud, M.E., Jeffery, N., Further comparing their observations with model
Hecht, M.W., Hartin, C.A., van Sebille, E., Wainer, I., and projections, the authors noted several discrepancies.
Landrum, L. 2012. The Southern Ocean and its climate in First, Laxon et al. specifically note their observations
CCSM4. Journal of Climate 25: 2652–2675.
107
“show an interannual variability in ice thickness at simulated downward longwave radiative fluxes?”
higher frequency, and of greater amplitude, than The three researchers say “a frequently used
simulated by regional Arctic models,” clearly approach” to resolving this problem “is to tune the
indicating the models do not reproduce reality very parameters associated with the ice surface albedo” to
well in this regard. Second, they state “the interannual get a more realistic answer. “In other words,” as they
variability in thickness [9%] compares with a continue, “errors in parameter values are being
variability in mean annual ice extent of 1.7% during introduced to the GCM sea ice components to
the same period,” which, in the words of the authors, compensate simulation errors in the atmospheric
“undermines the conclusion from numerical models components.” They conclude “the thinning of Arctic
that changes in ice thickness occur on much longer sea ice over the past half-century can be explained by
timescales than changes in ice extent.” Third, minuscule changes of the radiative forcing that cannot
concerning the origin of Arctic sea-ice thickness be detected by current observing systems and require
variability, the authors discovered “a significant (R2 = only exceedingly small adjustments of the model-
0.924) correlation between the change in the generated radiation fields” and, therefore, “the results
altimeter-derived thickness between consecutive of current GCMs cannot be relied upon at face value
winters, and the melt season length during the for credible predictions of future Arctic sea ice.”
intervening summer.” This “observed dominant Kwok (2011) notes near the midpoint of the last
control of summer melt on the interannual variability decade, simulations of Arctic Ocean sea ice
of mean ice thickness,” according to the researchers, characteristics produced by the climate models
“is in sharp contrast with the majority of models, included in the World Climate Research Programme’s
which suggest that ice thickness variability in the Coupled Model Intercomparison Project phase 3
Arctic Ocean is controlled mainly by wind and ocean (CMIP3) were far from what might have been hoped.
forcing.” Fourth, the authors’ data demonstrate “sea Specifically, he writes “Zhang and Walsh (2006) note
ice mass can change by up to 16% within one year,” that even though the CMIP3 models capture the
which “contrasts with the concept of a slowly negative trend in sea ice area, the inter-model scatter
dwindling ice pack, produced by greenhouse is large,” “Stroeve et al. (2007) show that few models
warming,” representing still another failure of the exhibit negative trends that are comparable to
models. observations” and “Eisenman et al. (2007) conclude
Laxon et al. close their analysis by stating their that the results of current CMIP3 models cannot be
results “show that errors are present in current relied upon for credible projections of sea ice
simulations of Arctic sea ice,” concluding, “until behavior.”
models properly reproduce the observed high- In his more recent analysis of the subject, based
frequency, and thermodynamically driven, variability on the multi-model data set of Meehl et al. (2007),
in sea ice thickness, simulations of both recent, and Kwok, a researcher at the Jet Propulsion Laboratory,
future, changes in Arctic ice cover will be open to compares CMIP3 model simulations “with
question.” observations of sea ice motion, export, extent, and
Eisenman et al. (2007) used two standard thickness and analyzes fields of sea level pressure and
thermodynamic models of sea ice to calculate geostrophic wind of the Arctic Ocean.” Kwok’s
equilibrium Arctic ice thickness based on simulated analysis demonstrated “the skill of the CMIP3 models
Arctic cloud cover derived from 16 different global (as a group) in simulation of observed Arctic sea ice
climate models evaluated for the IPCC’s Fourth motion, Fram Strait export, extent, and thickness
Assessment Report. Results indicate there was a 40 between 1979 and 2008 seems rather poor.” He notes
Wm-2 spread among the models in terms of their “model-data differences and inter-model scatter of the
calculated downward longwave radiation, for which sea ice parameters in the summarizing statistics are
both sea ice models calculated an equilibrium ice high” and “the spatial pattern of Arctic sea ice
thickness ranging from one to more than ten meters. thickness, a large-scale slowly varying climatic
They note the mean 1980–1999 Arctic sea ice feature of the ice cover, is not reproduced in a
thickness simulated by the 16 GCMs ranged from majority of the models.” Consequently, he writes,
only 1.0 to 3.9 meters, a far smaller inter-model “the models will not get the main features of natural
spread. Hence, they say they were “forced to ask how sea ice variability that may be dominating recent sea
the GCM simulations produce such similar present- ice extent declines as well as the long-term
day ice conditions in spite of the differences in greenhouse response.”
108
“Because the model simulations have difficulties

reproducing the mean patterns of Arctic circulation
and thickness,” Kwok writes in his concluding
paragraph, his analysis suggests there are
“considerable uncertainties in the projected rates of
sea ice decline even though the CMIP3 data set agrees
that increased greenhouse gas concentrations will
result in a reduction of Arctic sea ice area and
volume.”
The most recent investigation into the topic was
conducted by Turner et al. (2013). The authors state
“Phase 5 of CMIP (CMIP5) will provide the model
output that will form the basis of the Fifth Assessment
Report (AR5) of the IPCC,” and they therefore
thought it important to determine how well these
models represent reality. They set out to examine “the
annual cycle and trends in Antarctic sea ice extent
(SIE) for 18 models used in phase 5 of the Coupled
Model Intercomparison Project that were run with
historical forcing for the 1850s to 2005.”
According to the five researchers, their
examination indicated “the majority of models have
too small of an SIE at the minimum in February” and
“several of the models have less than two-thirds of the
observed SIE at the September maximum. They
further note, “in contrast to the satellite data, which
exhibit a slight increase in SIE, the mean SIE of the
models over 1979–2005 shows a decrease in each
month”; “the models have very large differences in
SIE over 1860–2005”; and “the negative SIE trends in Figure 1.3.7.6.1. (a) Annual mean Antarctic sea ice area
most of the model runs over 1979–2005 are a anomaly relative to the 1979–1989 base period, for satellite
continuation of an earlier decline, suggesting that the observations using the NASATEAM algorithm, the
processes responsible for the observed increase over ensemble mean of 38 CMlP5 models (with a total of 135
the last 30 years are not being simulated correctly.” realizations, listed in the supporting information), with the
envelope indicating the 95% confidence interval, and the
Turner et al. conclude “many of the SIE biases in
University of Victoria (UVic) model. The thick green curve
the CMIP3 runs remain in CMIP5.” More is the linear least squares fit to the observed anomalies. (b)
particularly, for example, they state, “as with CMIP3, Distribution of linear trends in annual mean sea ice area for
the models do not simulate the recent increase in the CMlP5 models and the observed trend. (Figure 1 of
Antarctic SIE observed in the satellite data.” Swart and Fyfe, 2013)
Confirming the results of Turner et al. (2013),
Swart and Fyfe (2013) examined the consistency
between satellite observations and CMIP5 climate References
model projections of the evolution of Antarctic sea
ice extent. They note, “The CMIP5 multimodel Eisenman, I., Untersteiner, N., and Wettlaufer, J.S. 2007.
ensemble mean shows a large negative trend in On the reliability of simulated Arctic sea ice in global
annual mean sea ice area of -3.0 x 1011 m2/decade climate models. Geophysical Research Letters 34: 10.1029/
over the historical period” and “By contrast, the 2007GL029914.
observations show a statistically significant positive Holland, D.M. 2001. An impact of subgrid-scale ice-ocean
trend of 1.39 ± 0.82 x 1011 m2/decade (95% dynamics on sea-ice cover. Journal of Climate 14: 1585–
confidence interval accounting for serial 1601.
correlation).” (See Figure 1.3.7.6.1.)
Kwok, R. 2011. Observational assessment of Arctic Ocean
109
sea ice motion, export, and thickness in CMIP3 climate surface representations producing better soil moisture
simulations. Journal of Geophysical Research 116: simulations in the revised climate models.” They
10.1029/2011JC007004. report a “tendency toward unrealistic summer drying
Laxon, S., Peacock, N., and Smith, D. 2003. High in several models,” which they note was “particularly
interannual variability of sea ice thickness in the Arctic relevant in view of the summer desiccation projected
region. Nature 425: 947–950. by GCMs considered in future assessments of climate
change.”
Meehl, G.A., Covey, C., Delworth, T., Latif, M., Although Srinivasan et al. report “simpler land-
McAvaney, B., Mitchell, J.F.B., Stouffer, R.J., and Taylor,
surface parameterization schemes are being replaced
K.E. 2007. The WCRP CMIP3 multi-model dataset: A new
era in climate change research. Bulletin of the American by conceptually realistic treatments,” as the climate
Meteorological Society 88: 1383–1394. modeling enterprise evolves, they note
“improvements gained by such changes are ... not
Stroeve, J., Holland, M.M., Meier, W., Scambos, T., and very apparent.”
Serreze, M. 2007. Arctic sea ice decline: Faster than A similar assessment was supplied that year by
forecast. Geophysical Research Letters 34: 10.1029/ Robock et al. (2000), who developed a massive
2007GL029703.
collection of soil moisture data for more than 600
Swart, N.C. and Fyfe, J.C., 2013. The influence of recent stations from a wide variety of climatic regimes found
Antarctic ice sheet retreat on simulated sea iec area trends. within the former Soviet Union, China, Mongolia,
Geophysical Research Letters 40: 10.1002/grl.50820. India, and the United States. In describing these data
sets they also state an important ground rule.
Turner, J., Bracegirdle, T.J., Phillips, T., Marshall, G.J.,
and Hosking, J.S. 2013. An initial assessment of Antarctic Sometimes, they note, “the word ‘data’ is used to
sea ice extent in the CMIP5 models. Journal of Climate 26: describe output from theoretical model calculations,
1473–1484. doi: http://dx.doi.org/10.1175/JCLI-D-12- or values derived from theoretical analysis of
00068.1 radiances from remote sensing.” However, they state,
“we prefer to reserve this word for actual physical
Zhang, X. and Walsh, J.E. 206. Toward a seasonally ice- observations,” noting “all the data in our data bank
covered Arctic Ocean: Scenarios from the IPCC AR4
are actual in situ observations.”
model simulations. Journal of Climate 19: 1730–1747.
This distinction is important, for one of the
illuminating analyses Robock et al. performed with
1.3.8 Soil Moisture their data was to check summer soil moisture trends
Climate models have long indicated that CO2-induced simulated by the Geophysical Fluid Dynamics
global warming will increase evapotranspiration, Laboratory’s general circulation model of the
causing decreases in soil moisture content that may atmosphere as forced by transient CO2 and
offset modest increases in continental precipitation tropospheric sulfate aerosols for specific periods and
and lead to greater aridity in both water-limited regions for which they had actual soil moisture data.
natural ecosystems and lands devoted to agriculture They found, “although this model predicts summer
(Manabe and Wetherald, 1986; Rind, 1988; Gleick, desiccation in the next century, it does not in general
1989; Vlades et al., 1994; Gregory et al., 1997; reproduce the observed upward trends in soil moisture
Komescu et al., 1998). This section examines very well,” a mammoth understatement considering
pertinent scientific literature to assess this claim. the predictions and observations go in opposite
In a turn-of-the century evaluation of how climate directions. As noted elsewhere in their paper, “in
modelers had progressed in their efforts to improve contrast to predictions of summer desiccation with
their simulations of soil moisture content over the increasing temperatures, for the stations with the
prior few years, Srinivasan et al. (2000) examined longest records, summer soil moisture in the top 1 m
“the impacts of model revisions, particularly the land has increased while temperatures have risen.”
surface representations, on soil moisture simulations, Another important report on the subject is
by comparing the simulations to actual soil moisture presented five years later, again by Robock et al.
observations.” They write, “the revised models do not (2005), who note “most global climate model
show any systematic improvement in their ability to simulations of the future, when forced with increasing
simulate observed seasonal variations of soil moisture greenhouse gases and anthropogenic aerosols, predict
over the regions studied.” They also note, “there are summer desiccation in the midlatitudes of the
no indications of conceptually more realistic land Northern Hemisphere (e.g., Gregory et al., 1997;
110
Wetherald and Manabe, 1999; Cubasch et al., 2001),” enrichment and soil warming were of similar
adding “this predicted soil moisture reduction, the magnitude, and acting together they enhanced mean
product of increased evaporative demand with higher spring soil moisture content by about 15 percent over
temperatures overwhelming any increased that of the control treatment. The effect of CO2 was
precipitation, is one of the gravest threats of global produced primarily as a consequence of its ability to
warming, potentially having large impacts on our cause partial stomatal closure and thereby reduce
food supply.” season-long plant water loss via transpiration. In the
With the explicit intent “to evaluate these model case of warming, there was an acceleration of canopy
simulations,” the three American and two Ukrainian senescence that further increased soil moisture by
scientists present “the longest data set of observed reducing the period of time over which transpiration
soil moisture available in the world, 45 years of losses occur, all without any decrease in total plant
gravimetrically-observed plant available soil moisture production.
for the top 1 m of soil, observed every 10 days for Zaveleta et al. note their findings “illustrate the
April-October for 141 stations from fields with either potential for organism-environment interactions to
winter or spring cereals from the Ukraine for 1958- modify the direction as well as the magnitude of
2002.” As they describe it, “the observations show a global change effects on ecosystem functioning.”
positive soil moisture trend for the entire period of Whereas model projections suggest vast reaches of
observation, with the trend leveling off in the last two agricultural land will dry up and be lost to profitable
decades,” noting “even though for the entire period production in a CO2-enriched world of the future, this
there is a small upward trend in temperature and a study suggests just the opposite could occur. As the
downward trend in summer precipitation, the soil six researchers describe it, “we suggest that in at least
moisture still has an upward trend for both winter and some ecosystems, declines in plant transpiration
summer cereals.” mediated by changes in phenology can offset direct
In light of these real-world observations, Robock increases in evaporative water losses under future
et al. note “although models of global warming warming.”
predict summer desiccation in a greenhouse-warmed Guo and Dirmeyer (2006) compared soil moisture
world, there is no evidence for this in the observations simulations made by 11 models in the Second Global
yet, even though the region has been warming for the Soil Wetness Project, a multi-institutional modeling
entire period.” In attempting to explain this research activity intended to produce a complete
dichotomy, they state the real-world increase in soil multi-model set of land surface state variables and
moisture content may have been driven by a fluxes by using land surface models driven by the 10-
downward trend in evaporation caused by the year period of data provided by the International
controversial “global dimming” hypothesis (Liepert et Satellite Land Surface Climatology Project Initiative
al., 2004). Alternatively, we offer it may have been II, against real-world observations made on the top
driven by the well-known anti-transpirant effect of meter of grassland and agricultural soils located
atmospheric CO2 enrichment, which tends to conserve within parts of the former Soviet Union, the United
water in the soils beneath crops and thereby leads to States (Illinois), China, and Mongolia that are
enhanced soil moisture contents, as has been archived in the Global Soil Moisture Data Bank.
demonstrated in a host of experiments conducted in According to the authors, “simulating the actual
real-world field situations. values of observed soil moisture is still a challenging
One especially outstanding study in this regard is task for all models,” as they note “both the root mean
that of Zaveleta et al. (2003), who tested the square of errors (RMSE) and the spread of RMSE
hypothesis that soil moisture contents may decline in across models are large” and “the absolute values of
a CO2-enriched and warmer world, in a two-year soil moisture are poorly simulated by most models.”
study of an annual-dominated California grassland at In addition, they report “within regions there can be
the Jasper Ridge Biological Preserve, Stanford, tremendous variations of any model to simulate the
California, USA, where they delivered extra heating time series of soil moisture at different stations.”
to a number of free-air CO2 enrichment (FACE) plots Guo and Dirmeyer suggest the errors and
(enriched with an extra 300 ppm of CO2) via IR heat variations are serious problems for the models. First,
lamps suspended over the plots that warmed the the two researchers write “the land surface plays a
surface of the soil beneath them by 0.8-1.0°C. vital role in the global climate system through
The individual effects of atmospheric CO2 interactions with the atmosphere.”They also note
111
“accurate simulation of land surface states is critical existence of “a warm period positive temperature
to the skill of weather and climate forecasts” and soil dependent bias” for “many of the models with many
moisture “is the definitive land surface state variable; of the chosen climate regions”; the magnitude of this
key for model initial conditions from which the global temperature dependence varied considerably among
weather and climate forecasts begin integrations, and the models.
a vital factor affecting surface heat fluxes and land Analyzing the role of this difference as “a
surface temperature.” contributing factor for some models to project
In their own study of the subject, Li et al. (2007) stronger regional warming than others,” the two
compared soil moisture simulations derived from the scientists found “models with a positive temperature
IPCC’s AR4 climate models, which were driven by dependent bias tend to have a large projected
observed climate forcings, for the period 1958–1999 temperature change” and “these tendencies increase
with actual measurements of soil moisture made at with increasing global warming level.” In addition,
more than 140 stations or districts in the mid-latitudes they state this situation “appears to be linked with the
of the Northern Hemisphere. The latter were averaged ability of models to capture complex feedbacks
in such a way as to yield six regional results: one each accurately,” noting in particular that land-
for the Ukraine, Russia, Mongolia, Northern China, surface/atmosphere interactions are treated differently
Central China, and Illinois (USA). According to the and with different degrees of realism among the
three researchers, the models showed realistic various models they investigated and “soil moisture-
seasonal cycles for the Ukraine, Russia, and Illinois temperature feedbacks are relevant for temperature
but “generally poor seasonal cycles for Mongolia and extremes in a large fraction of the globe.”
China.” In addition, they report the Ukraine and Christensen and Boberg conclude, “accepting
Russia experienced soil moisture increases in summer model spread as a way to portray uncertainty of the
“that were larger than most trends in the model projection estimate may result in an overestimation of
simulations.” The researchers found “only two out of the projected warming and at the same time indicate
25 model realizations show trends comparable to little model agreement on the mean value.” They note
those observations,” and the two realistic model- “a non-negligible part” of this overestimation “is due
derived trends were “due to internal model variability to model deficiencies” that have yet to be overcome.
rather than a result of external forcing,” which means In light of the observations discussed above, it
the two reasonable matches were in fact accidental. would appear almost all climate models employed to
Noting further “changes in precipitation and date have greatly erred with respect to what Robock
temperature cannot fully explain soil moisture et al. (2005) describe as “one of the gravest threats of
increases for [the] Ukraine and Russia,” Li et al. global warming”—soil moisture content. Not only has
write, “other factors might have played a dominant the model-predicted decline in Northern Hemispheric
role on the observed patterns for soil moisture.” They midlatitude soil moisture contents failed to
mention solar dimming, plus the fact that in response materialize under the combined influence of many
to elevated atmospheric CO2 concentrations, “many decades of rising atmospheric CO2 concentrations and
plant species reduce their stomatal openings, leading temperatures, it has become less of a threat, possibly
to a reduction in evaporation to the atmosphere,” so as a consequence of biological impacts of the ongoing
“more water is likely to be stored in the soil or rise in the air’s CO2 content.
[diverted to] runoff,” reporting this phenomenon was
detected in continental river runoff data by Gedney et References
al. (2006).
Publishing in Geophysical Research Letters, Christensen, J.H. and Boberg, F. 2012. Temperature
Christensen and Boberg (2012) describe how they dependent climate projection deficiencies in CMIP5
compared monthly mean temperatures projected by models. Geophysical Research Letters 39: 10.1029/
34 global climate models included in phase 5 of the 2012GL053650.
Coupled Model Intercomparison Project (CMIP5) Cubasch, U., Meehl, G.A., Boer, G.J., Stouffer, R.J., Dix,
with observations from the University of East M., Noda, A., Senior, C.A., Raper, S., and Yap, K.S. 2001.
Anglia’s Climatic Research Unit for 26 different Projections of future climate change. In: Houghton, J.T. et
regions covering all major land areas of the world for al. (Eds.) Climate Change 2001: The Scientific Basis:
the period 1961–2000, for which they employed Contribution of Working Group I to the Third Assessment
quantile-quantile (q-q) diagrams. This revealed the Report of the Intergovernmental Panel on Climate Change,
112
Cambridge University Press, Cambridge, New York, USA, Vlades, J.B., Seoane, R.S., and North, G.R. 1994. A
pp. 525–582. methodology for the evaluation of global warming impact
on soil moisture and runoff. Journal of Hydrology 161:
Gedney, N., Cox, P.M., Betts, R.A., Boucher, O., 389–413.
Huntingford, C., and Stott, P.A. 2006. Detection of a direct
carbon dioxide effect in continental river runoff records. Wetherald, R.T. and Manabe, S. 1999. Detectability of
Nature 439: 835–838. summer dryness caused by greenhouse warming. Climatic
Change 43: 495–511.
Gleick, P.H. 1989. Climate change, hydrology and water
resources. Reviews of Geophysics 27: 329–344. Zavaleta, E.S., Thomas, B.D., Chiariello, N.R., Asner,
G.P., Shaw, M.R., and Field, C.B. 2003. Plants reverse
Gregory, J.M., Mitchell, J.F.B., and Brady, A.J. 1997. warming effect on ecosystem water balance. Proceedings
Summer drought in northern midlatitudes in a time- of the National Academy of Science USA 100: 9892–9893.
dependent CO2 climate experiment. Journal of Climate 10:
662–686.
1.3.9 Biological Processes
Guo, Z. and Dirmeyer, P.A. 2006. Evaluation of the
Second Global Soil Wetness Project soil moisture In a landmark paper published in Global Change
simulations: 1. Intermodel comparison. Journal of Biology, Eastman et al. (2001) described the first
Geophysical Research 111: 10.1029/2006JD007233. comprehensive study of CO2-induced regional climate
change based on a hybrid atmosphere/vegetation
Komescu, A.U., Eikan, A., and Oz, S. 1998. Possible model composed of linked meteorological and plant
impacts of climate change on soil moisture availability in growth sub-models.
the Southeast Anatolia Development Project Region
The authors of the groundbreaking study began
(GAP): An analysis from an agricultural drought
perspective. Climatic Change 40: 519–545. by citing a number of peer-reviewed scientific
research papers that demonstrated the likelihood of
Li, H., Robock, A., and Wild, M. 2007. Evaluation of what they called “a crucial role for biospheric
Intergovernmental Panel on Climate Change Fourth feedbacks on climate,” including processes driven by
Assessment soil moisture simulations for the second half of CO2-induced changes in land surface albedo, leaf
the twentieth century. Journal of Geophysical Research stomatal conductance, plant rooting profile, fractional
112: 10.1029/2006JD007455.
coverage of the land by vegetation, plant roughness
Liepert, B.G., Feichter, J., Lohmann, U., and Roeckner, E. length and displacement height, vegetation
2004. Can aerosols spin down the water cycle in a warmer phenology, time of planting and harvesting (in the
and moister world? Geophysical Research Letters 31: case of agricultural crops), and plant growth. Next,
10.1029/2003GL019060. they validated the model against real-world
Manabe, S. and Wetherald, R.T. 1986. Reduction in meteorological and plant growth data obtained for the
summer soil wetness induced by an increase in atmospheric 1989 growing season for the area located between
carbon dioxide. Science 232: 626–628. approximately 35° and 48° N latitude and 96° and
110° W longitude. Last, they investigated how the
Rind, D. 1988. The doubled CO2 climate and the sensitivity climate of the region changes when (1) only the
of the modeled hydrologic cycle. Journal of Geophysical radiative effects of a doubling of the air’s CO2
Research 93: 5385–5412.
concentration are considered, (2) only the biological
Robock, A., Mu, M., Vinnikov, K., Trofimova, I.V., and effects of a doubling of the air’s CO2 concentration
Adamenko, T.I. 2005. Forty-five years of observed soil are considered, and (3) the radiative and biological
moisture in the Ukraine: No summer desiccation (yet). effects of a doubling of the air’s CO2 concentration
Geophysical Research Letters 32: 10.1029/2004GL021914. occur simultaneously.
Robock, A., Vinnikov, K.Y., Srinivasan, G., Entin, J.K., With respect to the area-averaged and seasonally
Hollinger, S.E., Speranskaya, N.A., Liu, S., and Namkhai, averaged daily maximum air temperature, the
A. 2000. The global soil moisture data bank. Bulletin of the radiative effects of a doubling of the atmospheric CO2
American Meteorological Society 81: 1281–1299. concentration lead to a warming of only 0.014°C, and
the biological effects of the extra CO2 produced a
Srinivasan, G., Robock, A., Entin, J.K., Luo, L., Vinnikov, cooling of fully 0.747°C. Considered together and
K.Y., Viterbo, P., and Participating AMIP Modeling
Groups. 2000. Soil moisture simulations in revised AMIP
including a nonlinear interaction term, the
models. Journal of Geophysical Research 105: 26,635– simultaneous radiative and biological effects of a
26,644. doubling of the air’s CO2 content thus produced a net
113
cooling of 0.715°C. al. 1996) and the ORCHIDEE biosphere model

With respect to the area-averaged and seasonally (Krinner et al. 2005), respectively. Each is a land
averaged daily minimum air temperature, the surface model that includes plant characteristics such
radiative effects of a doubling of the atmospheric CO2 as the physiology of plant cover, plant phenology,
concentration led to a warming of 0.097°C, while the carbon cycling, plant type competition,
biological effects of the extra CO2 produced a photosynthesis, and respiration. Both include daily
warming of 0.261°C. Considered together and again and annual vegetation cycles and distinguish between
including the nonlinear interaction term, the trees and other types of vegetation (grasses, shrubs).
simultaneous radiative and biological effects of a Delire et al. ran each model with full capabilities and
doubling of the air’s CO2 content thus produced a net then by keeping the vegetation constant (fixed).
warming of 0.354°C. The GCM modeling strategy used by Delire et al.
During the day, then, when high air temperatures was to simulate the climate using observed sea
can be detrimental to both plant and animal life, the surface temperatures from 1870–1899 available from
combined effect of the simultaneous radiative and the Hadley Centre in England. Their goal was to
biological impacts of an increase in the air’s CO2 remove the impact of the ocean in order to highlight
content acts to decrease daily maximum air land-surface process differences in the coupled
temperature, alleviating potential heat stress. During systems. The model was run for 400 years; the last
the night, when low temperatures can be detrimental 300 were used for analysis.
to plant and animal life, the combined effect of the The interannual variability in plant cover
radiative and biological impacts of an increase in the generated by both models was similar to that of
air’s CO2 content acts to increase daily minimum air observed plant cover variability derived using satellite
temperature, alleviating potential cold stress. When observations for 1982–1994. The LMDz-ORCHIDEE
considering the day and night air temperature changes model showed stronger interannual variability, but in
together, the mean daily air temperature range is both the variability in tree cover was less than 5
reduced by approximately 1.069°C, leading to a more percent, whereas for grasses it could be as much as 10
thermally stable environment, which in this case was to 20 percent. Vegetation affected climate, as inferred
also about 0.180°C cooler in the mean. by comparing the dynamic to fixed vegetation. The
The authors also found the CO2-induced change model strategy precluded comparison with observed
in area-averaged and seasonally averaged leaf area climate.
index was increased (+21.8%) with the simultaneous The dynamic vegetation runs showed stronger
expression of the radiative and biological effects of a low frequency variability than the fixed runs for both
doubling of the air’s CO2 content. temperature and precipitation for each model system,
In summarizing their findings, the authors report but the disparity between the fixed and dynamic run
“it is clear” the radiative effects of a doubling of the was stronger for the CCMv3-IBIS system (Figure
air’s CO2 content have “little effect on anything” and 1.3.9.1). The strength of the variability (0.05-2.0°C)
play but a “minor role in the coupled biological and compared favorably to that inferred by previous
atmospheric system.” The authors acknowledge their studies. The feedback between temperature and plant
analysis is “a regional-scale sensitivity study,” the growth was generally positive (warmer temperatures,
results of which “cannot be linearly scaled up to increased vegetation) in the mid-latitudes and
global scales.” poleward. The feedback was generally negative and
Nevertheless the authors follow that caveat by weaker in semi-arid regions (increased vegetation,
noting their results “suggest that the regional response more evaporation, cooler temperatures). There was
could be on the order of global climate sensitivities.” only a weak positive feedback in the precipitation
Thus they conclude, “climate change that results from over most areas of the globe.
anthropogenic increases of CO2 must consider the As we begin to understand more about the climate
biological effects of enriched CO2 as well as its system, including the complexity of heat and mass
radiative effect.” Most models exclude this important exchange between various portions of it, we can more
interaction. effectively model the past climate, including its
In a paper published a decade later, Delire et al. variability. However, many of the climate simulations
(2011) used the CCMv3 and LMDz atmospheric used to project future climate are fairly simple models
GCMs and coupled these models to the latest versions that do not include a dynamic ocean or realistic
of the Integrated Biosphere Simulator (IBIS) (Foley et representations of biological processes on land. These
114
Figure 1.3.9.1. Power spectra of precipitation (top) and temperature (bottom) for the CCMv3-IBIS (left) and LMDz-
ORCHIDEE (right) model systems. The solid line represents data from the Haldey Centre (observed), and the dashed
(dotted) line represents dynamic [DYN] (fixed [FIX]) vegetation. Adapted from Delire et al. (2011).
are problems that cannot continue to be ignored or consistent with empirical estimates at the global and
overlooked. As Delire et al. indicate, “terrestrial biome scales,” but all of the models “had difficulty
ecosystems provide ‘memory’ to the climate system, representing soil carbon at the 1° scale.” They note,
causing important variations in the climate and “despite similar overall structures, the models do not
ecological conditions on long-time-scales.” agree well among themselves or with empirical data
Todd-Brown et al. (2013) point out, “because on the global distribution of soil carbon.” Todd-
future climate projections depend on the carbon cycle, Brown et al. conclude “all model structures may have
ESMs [Earth System Models] must be capable of serious shortcomings, since net primary productivity
accurately representing the pools and fluxes of carbon and temperature strongly influenced soil carbon
in the biosphere, particularly in soils that store a large stocks in ESMs but not in observational data.”
fraction of terrestrial organic carbon,” but they note Todd-Brown et al. outline what may need to be
“there have been few quantitative assessments of done in order to resolve the failure of ESMs to
ESM skill in predicting soil carbon stocks, adequately replicate the real world, including “better
contributing to uncertainty in model simulations.” prediction of soil carbon drivers, more accurate model
“To help reduce this uncertainty,” as Todd-Brown et parameterization, and more comprehensive
al. describe it, they “analyzed simulated soil carbon representation of critical biological and geochemical
from ESMs participating in the Fifth Climate Model mechanisms in soil carbon sub-models.”
Intercomparison Project (CMIP5),” comparing the
results from 11 model centers to empirical data References
obtained from the Harmonized World Soil Database
(HWSD) and the Northern Circumpolar Soil Carbon Delire, C., De Noblet-Ducoudre, N., Sima, A., and
Database (NCSCD). According to the seven Gouriand, I. 2011. Vegetation dynamics enhancing long-
scientists, some ESMs “simulated soil carbon stocks term climate variability confirmed by two models. Journal
of Climate 24: 2238–2257.
115
Eastman, J.L., Coughenour, M.B., and Pielke Sr., R.A. then current permafrost trends are likely to be
2001. The regional effects of CO2 and landscape change manifestations of past climatic trends, some of which
using a coupled plant and meteorological model. Global may have taken place several decades ago or more.
Change Biology 7: 797–815. Koven et al. (2013) note “permafrost is a critical
Foley, J.A., Prentice, C.I., Ramankutty, N., Levis, S., component of high-latitude land and determines the
Pollard, D., Sitch, S., and Haxeltine, A. 1996. An character of the hydrology, ecology, and
integrated biosphere model of land surface processes, biogeochemistry of the region.” Therefore, they write,
terrestrial carbon balance, and vegetation dynamics. Global there is “widespread interest in the use of coupled
Biogeochemical Cycles 10: 603–628. atmosphere-ocean-land surface models to predict the
Krinner, G., Viovy, N., de Noblet-Ducoudré, N., Ogée, J., fate of permafrost over the next centuries because (1)
Polcher, J., Friedlingstein, P., Ciais, P., Sitch, S., and permafrost contains the largest organic carbon (C)
Prentice, I.C. 2005. A dynamic global vegetation model for reservoir in the terrestrial system (Tarnocai et al.,
studies of the coupled atmosphere-biosphere system. 2009), (2) permafrost stability is primarily dependent
Global Biogeochemical Cycles 19: 1–33. on temperature, and (3) global warming is expected to
be relatively larger over the permafrost domain
Todd-Brown, K.E.O., Randerson, J.T., Post, W.M.,
because of arctic amplification processes (Holland
Hoffman, F.M., Tarnocai, C., Schuur, E.A.G., and Allison,
S.D. 2013. Causes of variation in soil carbon simulations and Bitz, 2003).”
from CMIP5 Earth system models and comparison with Koven et al. analyzed “output from a set of Earth
observations. Biogeosciences 10: 1717–1736. system models (ESMs) that participated in phase 5 of
the Coupled Model Intercomparison Project (CMIP5)
(Taylor et al., 2009) to evaluate the permafrost model
1.3.10 Permafrost predictions against observations and theoretical
Almost all assessments of the potential impacts of expectations and to compare the predicted fate of
climate change on the world’s permafrost are based permafrost under warming scenarios.” The three U.S.
on a two-layer model that incorporates a seasonally researchers revealed “the models show a wide range
frozen active layer and an underlying perennially of behaviors under the current climate, with many
frozen soil. Shur et al. (2005) examined the virtues of failing to agree with fundamental aspects of the
adding a transition zone layer to produce a more observed soil thermal regime at high latitudes.”
realistic three-layer model. Koven et al. report, “under future climate
Through a review of the literature and theoretical change, the models differ in their degree of warming,
and data analyses, Shur et al. showed, among other both globally and at high latitudes, and also in the
things, that the transition zone alternates between response of permafrost to this warming.” They note
seasonally frozen ground and permafrost over sub- “there is a wide range of possible magnitudes in their
decadal to centennial time scales, functioning as a responses, from 6% to 29% permafrost loss per 1°C
buffer between the active layer and the underlying high-latitude warming.” Several of the models, they
perennial permafrost by increasing the latent heat report, “predict substantial permafrost degradation
required for thaw. Consequently, in the words of Shur has already occurred (ranging from 3% gain to 49%
et al., use of a two-layer conceptual model in loss relative to 1850 conditions)” and “the majority of
permafrost studies “obscures effective understanding models at the high end of relative twentieth-century
of the formation and properties of the upper permafrost loss also show unrealistically small
permafrost and syngenetic permafrost, and makes a preindustrial permafrost extent.” They also note
realistic determination of the stability of arctic “there is wide model disagreement on the value of the
geosystems under climatic fluctuations virtually difference in mean temperature across the air-soil
impossible.” They conclude “the impacts of possible interface, with several of the models [even] predicting
global warming in permafrost regions cannot be the wrong sign for this statistic” and “there is wide
understood fully without consideration of a more model disagreement in the changes of [the] mean and
realistic three-layer model.” [the] amplitude of soil temperatures with depth.”
In light of the authors’ findings, it would appear Koven et al. conclude by stating, “with this
two-layer model forecasts of future permafrost trends analysis, we show that widespread disagreement
under various global warming scenarios are exists among this generation of ESMs,” once again
inadequate. And if the transition zone does indeed act suggesting current Earth system models are not yet
as a buffer at sub-decadal to centennial time scales, accurate enough for real-world application.
116
References University of Kiel—first describe various

mechanisms responsible for internal variability,
Holland, M.M. and Bitz, C.M. 2003. Polar amplification of giving special attention to the variability of the
climate change in coupled models. Climate Dynamics 21: Atlantic Meridional Overturning Circulation
221–232. (AMOC), which they suggest is likely the origin of a
Koven, C.D., Riley, W.J., and Stern, A. 2013. Analysis of considerable part of the decadal variability within the
permafrost thermal dynamics and response to climate Atlantic Sector, after which they discuss the challenge
change in the CMIP5 earth system models. Journal of of decadal SAT predictability and various factors
Climate 26: 1877–1900. limiting its realization.
The two researchers identify numerous problems
Shur, Y., Hinkel, K.M., and Nelson, F.E. 2005. The
that hamper decadal climate predictability, including
transient layer: implications for geocryology and climate-
change science. Permafrost and Periglacial Processes 16: that “the models suffer from large biases.” In the
5–17. cases of annual mean sea surface temperature (SST)
and SAT over land, for example, they state “typical
Tarnocai, C., Canadell, J.G., Schuur, E.A.G., Kuhry, P., errors can amount up to 10°C in certain regions,” as
Mazhitova, G., and Zimov, S. 2009. Soil organic carbon Randall et al. (2007) found to be the case for many of
pools in the northern circumpolar permafrost region. the IPCC-AR4 models. Latif and Keenlyside also
Global Biogeochemical Cycles 23: 10.1029/
note several models “fail to simulate a realistic El
2008GB003327.
Niño/Southern Oscillation.” In addition, the
Taylor, K.E., Stouffer, R.J., and Meehl, G.A. 2009. A researchers point out “several assumptions have
Summary of the CMIP5 Experiment Design. Technical generally to be made about the process under
Report: Program for Climate Model Diagnosis and consideration that cannot be rigorously justified, and
Intercomparison. Lawrence Livermore National this is a major source of uncertainty.”
Laboratory, Livermore, California, USA. Another problem they discuss is that “some
components of the climate system are not well
1.3.11 Miscellaneous represented or not at all part of standard climate
Several other studies have documented inadequacies models,” one example being the models’ neglect of
in climate model projections. This subsection the stratosphere. This omission is serious; Latif and
highlights those that do not quite fit in other Keenlyside say “recent studies indicate that the mid-
subsections of Section 1.4 or that deal with multiple latitudinal response to both tropical and extra-tropical
climatic elements and, as such, are best suited for this SST anomalies over the North Atlantic Sector may
miscellaneous category. critically depend on stratospheric feedbacks,” noting
“Climate variability,” in the words of Latif and Ineson and Scaife (2009) present evidence for “an
Keenlyside (2011), “can be either generated internally active stratospheric role in the transition to cold
by interactions within or between the individual conditions in northern Europe and mild conditions in
climate subcomponents (e.g., atmosphere, ocean and southern Europe in late winter during El Niño years.”
sea ice) or externally by e.g., volcanic eruptions, An additional common model shortcoming, even
variations in the solar insolation at the top of the in standalone integrations with models forced by
atmosphere, or changed atmospheric greenhouse gas observed SSTs, is that model simulations of rainfall in
concentrations in response to anthropogenic the Sahel “fail to reproduce the correct magnitude of
emissions.” Some examples of these internal the decadal precipitation anomalies.” Still another
variations are “the North Atlantic Oscillation (NAO), failure, as shown by Stroeve et al. (2007), is that
the El Niño/Southern Oscillation (ENSO), the Pacific “virtually all climate models considerably
Decadal Variability (PDV), and the Atlantic underestimate the observed Arctic sea ice decline
Multidecadal Variability (AMV),” all of which during the recent decades in the so-called 20th
“project on global or hemispheric surface air century integrations with prescribed (known natural
temperature (SAT), thereby masking anthropogenic and anthropogenic) observed forcing.” In addition,
climate change.” “atmospheric chemistry and aerosol processes are still
In a review of this complex subject, Latif and not well incorporated into current climate models.”
Keenlyside—who hold positions at Germany’s In summing up their findings, which include
Leibniz-Institute for Meerewissenschaften at the those noted above and many more, Latif and
Keenlyside state, “a sufficient understanding of the
117
mechanisms of decadal-to-multidecadal variability is time scales (e.g., daily, month-to-month, or

lacking.” They note “state-of-the-art climate models interannual, etc.) is crucial and probably provides an
suffer from large biases” and “are incomplete and do even better test as to whether [real-world] physical
not incorporate potentially important physics.” processes are well represented in the models.”
Various mechanisms “differ strongly from model to Working with twentieth century climate model runs
model,” they point out; “the poor observational prepared within the context of the IPCC AR4
database does not allow a distinction between assessment (now called the CMIP3 data set), Scherrer
‘realistic’ and ‘unrealistic’ simulations”; and many set out to compare model simulations of the
models “still fail to simulate a realistic El interannual variability (IAV) of 2-m-height air
Niño/Southern Oscillation.” Therefore, they conclude, temperature (T), sea level pressure (SLP), and
“it cannot be assumed that current climate models are precipitation (P) over the twentieth century with
well suited to realize the full decadal predictability observational and reanalysis data sets for the same
potential”—a somewhat obscure way of stating time period using standard deviation-based variability
current state-of-the-art climate models are not good indices.
enough to make reasonably accurate simulations of The Swiss scientist describes a number of
climate change over a period of time (either in the problems he encountered with the CMIP3 models.
past or the future) that is measured in mere decades. With respect to SLP, the situation was pretty good:
In another paper, Lucarini et al. (2007) compared “only minor IAV problems are found.” With respect
for the period 1962–2000 “the estimate of the to temperature, however, differences between
northern hemisphere mid-latitude winter atmospheric observations and models are, in general, “larger than
variability within the available 20th century those for SLP.” And for precipitation, “IAV is ‘all
simulations of 19 global climate models included in over the place’ and no clear relations with T and SLP
the Intergovernmental Panel on Climate Change IAV problems can be established.”
[IPCC] 4th Assessment Report” with “the NCEP- Concentrating thereafter mostly on temperature,
NCAR and ECMWF reanalyses,” compilations of Scherrer notes “a few models represent T IAV much
real-world observations produced by the National worse than others and create spurious relations of
Center for Environmental Prediction (NCEP), in IAV representation and the climate change signal.”
collaboration with the National Center for Among the “better” IAV models, he finds “the ‘good’
Atmospheric Research (NCAR), and by the European IAV models in the tropics are in general not also the
Center for Mid-Range Weather Forecast (ECMWF). ‘good’ IAV models in the extra-tropics,” and “the
The five Italian researchers report “large biases, in ‘good’ IAV models over the sea are in general not the
several cases larger than 20%, are found in all the ‘good’ IAV models over land.” He further notes
considered metrics between the wave climatologies of “similar results are found for the relation between T
most IPCC models and the reanalyses, while the span IAV representation and the amplitude of projected
of the climatologies of the various models is, in all changes in temperature.”
cases, around 50%.” They also report “the traveling “In general,” Scherrer writes, “it is concluded
baroclinic waves are typically overestimated by the that, aggregated over very large regions, hardly any
climate models, while the planetary waves are usually robust relations exist between the models’ ability to
underestimated,” and “the model results do not cluster correctly represent IAV and the projected temperature
around their ensemble mean.” The authors conclude change.” He says these results represent “a plea to
by stating, “this study suggests caveats with respect to remove the ‘obviously wrong’ models (e.g., like those
the ability of most of the presently available climate that have sea ice extending to below 50°N in the
models in representing the statistical properties of the Atlantic and DJF temperature biases of ~40°C in
global scale atmospheric dynamics of the present Iceland, cf. Raisanen, 2007) before doing climate
climate and, a fortiori [“all the more,” as per analyses.”
Webster’s Dictionary], in the perspective of modeling de Boer et al. (2012) point out “observed and
climate change.” projected changes in the Arctic region are some of the
According to Scherrer (2011), “climate model most striking concerns surrounding climate trends,”
verification primarily focused [in the past] on the noting the latter “likely have important consequences
representation of climatological means.” But “on the both within the Arctic and globally.” They further
other hand,” he continues, “a good representation of note “a new generation of Earth system models has
second-order moments (i.e., variability) on different been utilized to prepare climate projections for the
118
fifth phase of the Coupled Model Intercomparison climate,” which “allows determining the model
Project (CMIP5),” the results of which are planned to deficiencies.” The two authors set out to “[evaluate] a
be used “in the Intergovernmental Panel on Climate present climate simulation over southern South
Change (IPCC) Fifth Assessment Report (AR5).” America performed with the Meteorological Research
They set out to determine how well these models Institute/Japanese Meteorological Agency (MRI/
perform, but the closest they could come to JMA) high resolution global model.”
conducting such a test was to interrogate the models Comparing their simulated wind results with data
used in the AR4 report of the IPCC. Thus, de Boer et from the European Centre Medium Range Weather
al. simulated key features of the Arctic atmosphere in Forecasts (ECMWF) 40-year Reanalysis (ERA 40),
the Community Climate System Model, version 4 and their temperature and precipitation simulations
(CCSM4) and compared the results of those with data from various meteorological stations, the
simulations “against observational and reanalysis two Argentinian researchers report discovering
datasets for the present-day (1981–2005).” significant model deficiencies: Speeds of the low
Describing problems they encountered in this level jet and the westerlies “are generally
endeavor, the seven scientists report “simulated underestimated” and at upper levels “the westerlies
surface air temperatures are found to be slightly too are overestimated over central Argentina.” During
cold,” “evaluation of the sea level pressure [SLP] December-February, March-May, and September-
demonstrates some large biases, most noticeably an November, they report, “the MRI/JMA global model
under simulation of the Beaufort High during spring underestimates the temperature over east of
and autumn,” “monthly Arctic-wide [SLP] biases of Argentina, west of Uruguay, south of Chile and over
up to 13 mb are reported,” “cloud cover is under- tropical latitudes”and “overestimates are observed
predicted for all but summer months,” and “cloud over central Argentina,” while “in June-August the
phase is demonstrated to be different from model underestimates the temperature over most of
observations.” They also found “simulated all-sky Argentina, south of Chile and to the north of 20°S.”
liquid water paths are too high,” “ice water path was They also found “the model overestimates
generally too low,” and “precipitation is found to be temperature interannual variability in all regions and
excessive over much of the Arctic compared to ERA- all seasons, except in [June-July-August].”
40 and the Global Precipitation Climatology Project With respect to precipitation, the authors found in
estimates.” They report “biases of 40%-150% are all seasons the model yields “an underestimation of
calculated over northern North America, northern the precipitation in the southeast of Brazil and south
Greenland, and the Arctic Ocean,” while “over the of Peru and an overestimation in Bolivia, Uruguay,
Norwegian Sea ... evaporation is over-simulated by north and central Chile and north of Peru,” and during
up to 3.5 mm/day,” such that “P-E is generally too “the dry season (JJA) the model greatly overestimates
high over much of the Arctic, particularly over coastal the precipitation over northeastern and central
Greenland.” Argentina.” They note “in regions located over
de Boer et al. also found “CCSM4 over-predicts mountainous areas the model presents a poor
surface energy fluxes during summer months” and reproduction of the annual cycle” and “observed
“under-predicts it during winter.” They also report precipitation trends are generally positive whereas
“the strengths of surface inversions were found to be simulated ones are negative.”
too great in CCSM4 when compared to ERA-40, with Landrum et al. (2013) compared Last Millennium
distributions showing a near-doubling of strength,” (LM) simulations of the Community Climate System
and (15) “CCSM4 is found to have more inversions Model, version 4 (CCSM4) to real-world “data
than ERA-40 for all months.” reconstructions of temperature, the hydrologic cycle,
Oddly, de Boer et al. conclude “CCSM4 provides and modes of climate variability.” In addition to some
a consistent representation of present-day Arctic successes of the CCSM4, the seven scientists report a
climate” and “in doing so it represents individual number of failures. They note “the LM simulation
components of the Arctic atmosphere with does not reproduce La Niña-like cooling in the
respectable accuracy.” This statement seems to us to eastern Pacific Ocean during the MCA [Medieval
be an egregious misuse of the word “respectable.” Climate Anomaly] relative to the LIA [Little Ice
Blazquez and Nuñez (2013) point out “the first Age], as has been suggested by proxy
step to understand climate changes that are likely to reconstructions,” and “patterns of simulated
occur in the future is the assessment of the present precipitation change for the Asian monsoon to large
119
volcanic eruptions have nearly opposite anomalies on local scales.” They note “a new generation of more
from those reconstructed from tree-ring complex models running scenarios for the upcoming
chronologies.” They report CCSM4 “does not Intergovernmental Panel on Climate Change Fifth
simulate a persistent positive NAO [North Atlantic Assessment Report (IPCC AR5) is widely, and
Oscillation] or a prolonged period of negative PDO perhaps naively, expected to provide more detailed
[Pacific Decadal Oscillation] during the MCA, as and more certain projections.”
suggested by some proxy reconstructions” and “the Exploring whether these expectations are being
model simulates cooling of ~1.0°-1.5°C after the large met, the two researchers performed “a first
eruptions of the late thirteenth, mid-fifteenth, late comparison between projections from CMIP3 and
eighteenth, and early nineteenth centuries, 2-3 times CMIP5,” in order to see to what extent real progress
larger than the NH [Northern Hemisphere] summer in the modeling of Earth’s global climate may have
anomalies estimated from tree-ring or multiproxy been made. Knutti and Sedlacek report “projected
reconstructions.” They further report “twentieth- global temperature change from the new models is
century simulations indicate that the CCSM4 remarkably similar to that from those used in IPCC
hemispheric response to volcanic eruptions is stronger AR4 after accounting for the different underlying
than observed (Meehl et al., 2012)” and they “do not scenarios” and “the local model spread has not
find a persistent positive NAO or a prolonged period changed much despite substantial model development
of negative PDO during the MCA suggested by the and a massive increase in computational capacity.”
proxy reconstructions (MacDonald and Case, 2005; They write “there is … little evidence from CMIP5
Trouet et al., 2009).” that our ability to constrain the large-scale climate
Su et al. (2013) evaluated “the performance of 24 feedbacks has improved significantly”; “model mean
GCMs available in the fifth phase of the Coupled patterns of temperature and precipitation change …
Model Intercomparison Project (CMIP5) … over the are remarkably similar in CMIP3 and CMIP5”; and
eastern Tibetan Plateau (TP) by comparing the model “robustness over land is slightly higher but also
outputs with ground observations for the period similar in CMIP3 and CMIP5,” which they describe
1961–2005,” focusing their attention on temperature as “troublesome.”
and precipitation. The five researchers report that In light of these findings, and “if the past is a
with respect to temperature, “most GCMs reasonably guide to the future,” as the two researchers put it,
capture the climatological patterns and spatial “then uncertainties in climate change are unlikely to
variations of the observed climate,” but “the majority decrease quickly, and may even grow temporarily.”
of the models have cold biases, with a mean The scientists say they “have illustrated this for
underestimation of 1.1°-2.5°C for the months seasonal temperature and precipitation” and “it is
December-May, and less than 1°C for June-October.” likely that impact-relevant predictions, for example of
As for precipitation, they state “the simulations of all extreme weather events, may be even harder to
models overestimate the observations in improve.”
climatological annual means by 62.0%-183.0%,”
while noting “only half of the 24 GCMs are able to References
reproduce the observed seasonal pattern,” including
“the sharp contrast between dry winters and wet Blazquez, J. and Nuñez, M.N. 2013. Performance of a high
summers.” The last of these observations clearly resolution global model over southern South America.
suggests, as Su et al. note, that there is “a critical need International Journal of Climatology 33: 904–919.
to improve precipitation-related processes in these de Boer, G., Chapman, W., Kay, J.E., Medeiros, B., Shupe,
models.” They found 90-year forward projections of M.D., Vavrus, S., and Walsh, J. 2012. A characterization of
both precipitation and temperature “differ much more the present-day Arctic atmosphere in CCSM4. Journal of
among various models than among emissions Climate 25: 2676–2695.
scenarios,” suggesting temperature-related processes
Ineson, S. and Scaife, A.A. 2009. The role of the
in the models must be improved upon as well.
stratosphere in the European climate response to El Niño.
In a paper published in Nature Climate Change, Nature Geoscience 2: 32–36.
Knutti and Sedlacek (2013) write “estimates of
impacts from anthropogenic climate change rely on Knutti, R. and Sedlacek, J. 2013. Robustness and
projections from climate models,” but “uncertainties uncertainties in the new CMIP5 climate model projections.
in those have often been a limiting factor, particularly Nature Climate Change 3: 369–373.
120
Landrum, L., Otto-Bliesner, B.L., Wahl, E.R., Conley, A., 1.4 Large Scale Phenomena and
Lawrence, P.J., Rosenbloom, N., and Teng, H. 2013. Last Teleconnections
millennium climate and its variability in CCSM4. Journal
of Climate 26: 1085–1111.
The role of irregular interannual and interdecadal
Latif, M. and Keenlyside, N.S. 2011. A perspective on climate variations in forcing climate change is now
decadal climate variability and predictability. Deep-Sea being quantified by climate science. Some specific
Research II 58: 1880–1894. variations—the Pacific Decadal Oscillation (PDO),
Lucarini, V., Calmanti, S., Dell’Aquila, A., Ruti, P.M., and for example—have been identified and evaluated, and
Speranza A. 2007. Intercomparison of the northern they correlate well with decadal changes in global
hemisphere winter mid-latitude atmospheric variability of temperatures throughout the twentieth century. Other
the IPCC models. Climate Dynamics 28: 829–848. variations—such as the El Niño and Southern
Oscillation (ENSO)—have been recognized for
MacDonald, G.M. and Case, R.A. 2005. Variations in the decades.
Pacific decadal oscillation over the past millennium.
Although scientists can explain the role of these
Geophysical Research Letters 32: 10.1029/2005GL022478.
circulations in local climate variability and
Meehl, G. A., Washington, W.M., Arblaster, J.M., Hu, A., understand their dynamic evolution, the trigger
Teng, H., Tebaldi, C., Sanderson, B.N., Lamarque, J.-F., mechanisms for initiating changes in these
Conley, A., Strand, W.G., and White III, J.B. 2012. oscillations are not well understood. We use climate
Climate system response to external forcings and climate models to improve our understanding of them. A
change projections in CCSM4. Journal of Climate 25: model, however, cannot replicate the basic flow
3661–3683.
adequately. A by-product of this failure is that model
Raisanen, J. 2007. How reliable are climate models? Tellus output can have significant biases, a problem weather
59A: 2–29. forecasters have recognized for decades.
This section examines several of these features of
Randall, D.A. and Wood, R.A. et al. 2007. Chapter 8:
Earth’s climate and how well they are simulated by
Climate Models and Their Evaluation. In: Climate Change
2007: The Physical Science Basis. Contribution of Working the models. Many such studies are highlighted in the
Group I to the Fourth Assessment Report of the subsections below; many more were addressed in
Intergovernmental Panel on Climate Change. Cambridge earlier sections of this chapter, including the sections
University Press, Cambridge, United Kingdom. on oceans, temperature, and precipitation.
Scherrer, S.C. 2011. Present-day interannual variability of 1.4.1 El Niño/Southern Oscillation
surface climate in CMIP3 models and its relation to future
Computer model simulations have given rise to three
warming. International Journal of Climatology 31: 1518–
1529. claims regarding the influence of global warming on
El Niño/Southern Oscillation (ENSO) events: (1)
Stroeve, J., Holland, M.M., Meier, W., Scambos, T., and global warming will increase the frequency of ENSO
Serreze, M. 2007. Arctic sea ice decline: faster than events, (2) global warming will increase the intensity
forecast. Geophysical Research Letters 34: 10.1029/ of ENSO events, and (3) weather-related disasters
2007GL029703. will be exacerbated under El Niño conditions. This
Su, F., Duan, X., Chen, D., Hao, Z., and Cuo, L. 2013. section highlights findings that suggest the virtual
Evaluation of the Global Climate Models in the CMIP5 world of ENSO, as simulated by state-of-the-art
over the Tibetan Plateau. Journal of Climate 26: 3187– climate models, is at variance with reality, beginning
3208. with several studies that described the status of the
Trouet, V., Esper, J., Graham, N.E., Baker, A., Scourse,
problem a decade ago.
J.D., and Frank, D.C. 2009. Persistent positive North In a comparison of 24 coupled ocean-atmosphere
Atlantic oscillation mode dominated the medieval climate climate models, Latif et al. (2001) report, “almost all
anomaly. Science 324: 78–80. models (even those employing flux corrections) still
have problems in simulating the SST [sea surface
temperature] climatology.” They also note, “only a
few of the coupled models simulate the El
Niño/Southern Oscillation in terms of gross equatorial
SST anomalies realistically.” And they state, “no
model has been found that simulates realistically all
121
aspects of the interannual SST variability.” situation has improved over the past decade.
Consequently, because “changes in sea surface Huber and Caballero (2003) introduce their
temperature are both the cause and consequence of contribution to the subject by stating, “studies of
wind fluctuations,” according to Fedorov and future transient global warming with coupled ocean-
Philander (2000), and because these phenomena atmosphere models find a shift to a more El Niño-like
figure prominently in the El Niño-La Niña oscillation, state,” although they also report the “permanent El
it is not surprising the researchers conclude climate Niño state”—which has been hypothesized by some
models near the turn of the century did not do a good in the climate community—“is by no means
job of determining the potential effects of global uniformly predicted by a majority of models.” To
warming on ENSO. help resolve this battle of the models, they worked
Human ignorance likely also played a role in with still another model plus real-world data
those models’ failure to simulate ENSO. According to pertaining to the Eocene, the past geologic epoch—
Overpeck and Webb (2000), there was evidence that much warmer than the recent past—which provided,
“ENSO may change in ways that we do not yet in their words, “a particularly exacting test of the
understand,” and these “ways” clearly had not yet robustness of ENSO.” They used the Community
been modeled. White et al. (2001), for example, Climate System Model of the National Center for
found “global warming and cooling during Earth’s Atmospheric Research, which they state yielded “a
internal mode of interannual climate variability [the faithful reproduction of modern-day ENSO
ENSO cycle] arise from fluctuations in the global variability,” to “simulate the Eocene climate and
hydrological balance, not the global radiation determine whether the model predicts significant
balance”; they note these fluctuations are the result of ENSO variability.” In addition, they compared the
no known forcing of either anthropogenic or model results against middle Eocene lake-sediment
extraterrestrial origin, although Cerveny and Shaffer records from the Lake Gosiute complex in Wyoming
(2001) made a case for a lunar forcing of ENSO and Eckfield Maar in Germany.
activity, a factor not included in any climate model of Huber and Caballero report the model simulations
that time. showed “little change in ... ENSO, in agreement with
Another example of the inability of the most proxies.” They also note other studies “indicate an
sophisticated of late twentieth century climate models ENSO shutdown as recently as ~6000 years ago, a
to properly describe El Niño events was provided by period only slightly warmer than the present.” They
Landsea and Knaff (2000), who employed a simple conclude “this result contrasts with theories linking
statistical tool to evaluate the skill of 12 state-of-the- past and future ‘hothouse’ climates with a shift
art climate models in real-time predictions of the toward a permanent El Niño-like state.”
development of the 1997–98 El Niño. They found the Three years later, Joseph and Nigam (2006)
models exhibited essentially no skill in forecasting evaluated several climate models “by examining the
this very strong event at lead times ranging from zero extent to which they simulated key features of the
to eight months. They also determined no models leading mode of interannual climate variability: El
were able to anticipate even one-half of the actual Niño-Southern Oscillation (ENSO), which they
amplitude of the El Niño’s peak at a medium-range describe as “a dominant pattern of ocean-atmosphere
lead time of six to 11 months. Hence, they state, variability with substantial global climate impact,”
“since no models were able to provide useful based on “the Intergovernmental Panel on Climate
predictions at the medium and long ranges, there were Change’s (IPCC) Fourth Assessment Report (AR4)
no models that provided both useful and skillful simulations of twentieth-century climate.” Different
forecasts for the entirety of the 1997–98 El Niño.” models were found to do well in some respects but
It is little wonder several scientists criticized not so well in others. For example, they found climate
model simulations of ENSO behavior at the turn of models “are still unable to simulate many features of
the century, including Walsh and Pittock (1998), who ENSO variability and its circulation and hydroclimate
conclude, “there is insufficient confidence in the teleconnections.” They found the models had only
predictions of current models regarding any changes “begun to make inroads in simulating key features of
in ENSO” and Fedorov and Philander (2000), who ENSO variability.”
wrote, “at this time, it is impossible to decide which, The two scientists say their study suggests
if any, are correct.” “climate system models are not quite ready for
The rest of this section considers whether the making projections of regional-to-continental scale
122
hydroclimate variability and change” and “predicting stating these deficiencies “cannot be ignored when
regional climate variability/change remains an interpreting their predictions of future climate.”
onerous burden on models.” Paeth et al. (2008) compared 79 coupled ocean-
One year later, L’Ecuyer and Stephens (2007) atmosphere climate simulations derived from 12
asked how well state-of-the-art climate models different state-of-the-art climate models forced by six
reproduced the workings of real-world energy and IPCC emission scenarios with observational data in
water cycles, noting “our ability to model the climate order to evaluate how well they reproduced the
system and its response to natural and anthropogenic spatio-temporal characteristics of ENSO over the
forcings requires a faithful representation of the twentieth century, after which they compared the
complex interactions that exist between radiation, various models’ twenty-first century simulations of
clouds, and precipitation and their influence on the ENSO and the Indian and West African monsoons to
large-scale energy balance and heat transport in the one another.
atmosphere,” further stating “it is also critical to With respect to the twentieth century, this work
assess [model] response to shorter-term natural revealed “all considered climate models draw a
variability in environmental forcings using reasonable picture of the key features of ENSO.”
observations.” With respect to the twenty-first century, on the other
The two researchers used multi-sensor hand, they note “the differences between the models
observations of visible, infrared, and microwave are stronger than between the emission scenarios,”
radiance obtained from the Tropical Rainfall while “the atmospheric component of ENSO and the
Measuring Mission satellite for the period January West African monsoon are barely affected.” Their
1998 through December 1999 in order to evaluate the “overall conclusion” is that “we still cannot say much
sensitivity of atmospheric heating (and the factors that about the future behavior of tropical climate.” They
modify it) to changes in east-west SST gradients consider their study to be merely “a benchmark for
associated with the strong 1998 El Niño event in the further investigations with more recent models in
tropical Pacific, as expressed by the simulations of order to document a gain in knowledge or stagnation
nine general circulation models of the atmosphere over the past five years.”
utilized in the IPCC’s AR4. This protocol, in their Jin et al. (2008) investigated the overall skill of
words, “provides a natural example of a short-term ENSO prediction in retrospective forecasts made with
climate change scenario in which clouds, ten different state-of-the-art ocean-atmosphere
precipitation, and regional energy budgets in the east coupled general circulation models with respect to
and west Pacific are observed to respond to the their ability to hindcast real-world observations for
eastward migration of warm sea surface the 22 years from 1980 to 2001. They found almost
temperatures.” all models have problems simulating the mean
L’Ecuyer and Stephens report “a majority of the equatorial SST and its annual cycle. They write,
models examined do not reproduce the apparent “none of the models we examined attain good
westward transport of energy in the equatorial Pacific performance in simulating the mean annual cycle of
during the 1998 El Niño event.” They also discovered SST, even with the advantage of starting from
“the intermodel variability in the responses of realistic initial conditions.” They also note, “with
precipitation, total heating, and vertical motion [was] increasing lead time, this discrepancy gets worse” and
often larger than the intrinsic ENSO signal itself, “the phase and peak amplitude of westward
implying an inherent lack of predictive capability in propagation of the annual cycle in the eastern and
the ensemble with regard to the response of the mean central equatorial Pacific are different from those
zonal atmospheric circulation in the tropical Pacific to observed.” They also found “ENSO-neutral years are
ENSO.” In addition, they found “many models also far worse predicted than growing warm and cold
misrepresent the radiative impacts of clouds in both events” and “the skill of forecasts that start in
regions [the east and west Pacific], implying errors in February or May drops faster than that of forecasts
total cloudiness, cloud thickness, and the relative that start in August or November.” They and others
frequency of occurrence of high and low clouds.” In call this behavior “the spring predictability barrier.”
light of these much-less-than-adequate findings, they Jin et al. conclude, “accurately predicting the strength
conclude, “deficiencies remain in the representation and timing of ENSO events continues to be a critical
of relationships between radiation, clouds, and challenge for dynamical models of all levels of
precipitation in current climate models,” further complexity.”
123
McLean et al. (2009) quantified “the effect of The El Niño (warm East Pacific tropical water
possible ENSO forcing on mean global temperature, temperatures)/La Niña (cold East Pacific tropical
both short-term and long-term,” using Southern water temperatures) phenomenon is known to occur
Oscillation Index (SOI) data provided by the in a quasi-cyclic fashion repeating every two to seven
Australian government’s Bureau of Meteorology. years. It is the leading reason for the global variation
This parameter was defined as “the standardized in temperatures and precipitation on an interannual
anomaly of the seasonal mean sea level pressure time scale. Conceptual models and general circulation
difference between Tahiti and Darwin, divided by the models (GCMs) have been used to hypothesize that
standard deviation of the difference and multiplied by El Niño may arise as a result of internal nonlinear
10.” The temperature data employed in this endeavor processes.
were “the University of Alabama in Huntsville lower- White and Liu (2008) found El Niño/La Niña
tropospheric (LT) temperature data based on pairs may be “phase locked” to the quasi-decadal
measurements from selected view angles of oscillation (QDO), which is linked to the 11-year
Microwave Sounding Unit (MSU) channel LT 2” for solar cycle. Phase locking means two different
the period December 1979 to June 2008, harmonics vary in the same way with respect to each
supplemented by “balloon-based instrumentation other. The simplest example of this is pure
(radiosondes).” For the latter data, dating back to “constructive” or “destructive” interference. The
1958, they employed the Radiosonde Atmospheric authors performed harmonic analysis on a time series
Temperature Products for Assessing Climate of Pacific region sea surface temperatures (SSTs) for
(RATPAC) product (A) of the U.S. National Climatic 1895–2005. They also gathered 110 years of data
Data Center, which represents the atmospheric layer from a multi-century run of a coupled atmosphere-
between approximately 1,500 and 9,000 meters ocean GCM corresponding to the observations.
altitude. The authors found an 11-year QDO cycle in the
McLean et al. found “change in SOI accounts for observed record as well as strong peaks in the years
72% of the variance in GTTA [Global Tropospheric associated with El Niño, especially at 3.6 and 2.2
Temperature Anomalies] for the 29-year-long MSU years. When the authors ran the GCM without the 11-
record and 68% of the variance in GTTA for the year solar forcing, the computer model could not
longer 50-year RATPAC record,” as well as “81% of reproduce the QDO in its SST record. When the
the variance in tropospheric temperature anomalies in GCM included the forcing, the model not only
the tropics,” where they note ENSO “is known to reproduced the QDO but also the strong peaks in the
exercise a particularly strong influence.” In addition, 3.6 and 2.2 year period similar to observations.
they determined “shifts in temperature are consistent White and Liu also found a “phase locking” of
with shifts in the SOI that occur about 7 months the 11-year cycle with the 3.6 and 2.2 year cycles in
earlier.” The three researchers conclude, “natural both the model and the observations. This suggests
climate forcing associated with ENSO is a major the higher frequency oscillations had higher
contributor to variability and perhaps recent trends in amplitudes in step with the lower frequency one and
global temperature, a relationship that is not included are “hitting” a maximum (minimum) roughly in
in current global climate models.” correspondence with the low frequency cycle.
Harmonic analysis is useful in this discussion White and Liu went further with their analysis,
because it can be used to construct models of time taking the nine 11 year cycles found in each record
series or spatial patterns. It also can be used as a and “compositing” them (Figure 1.4.1.1). In both the
diagnostic tool on the same type of data sets. model and observations, similar behavior was
Harmonics are natural solutions to differential observed. When the 11, 3.6, and 2.2 (ridge—warm—
equations that represent the motion in oscillating El Niño / trough—cold—La Niña) year cycles were
systems and take the form of waves, usually added together and superimposed on the 11-year
represented as trigonometric functions (for example cycle as a visual aid, it was apparent El Niño/La Niña
sine or cosine). Harmonics simply represent the couplets occurred together on the ascending and
transformation of data from Cartesian coordinates descending side of the QDO, but that a strong El Niño
(space and/or time) to a wave coordinate. Thus, (La Niña) also can occur at the peak (in the valley) of
“wave-like” phenomena not readily apparent to the the QDO.
eye, or which appear as random noise, can be Finally, White and Liu used the previously
identified in a data set. derived QDO model of Jin (1997) and incorporated
124
their findings, finding a pattern similar to that shown “the amplitude and character of ENSO have been
in Figure 1.4.1.1. They also used the 3.6 and 2.2 year observed to exhibit substantial variations on
harmonics to compare to the observed record with the timescales of decades to centuries” and “many of
11-year cycle filtered out. They found this these changes over the past millennium resemble
combination reliably identified 26 of 32 El Niño those that arise from internally generated climate
events of 1895–2005. variations in an unforced climate model.” In addition,
they report “ENSO activity and characteristics have
been found to depend on the state of the tropical
Pacific climate system, which is expected to change
in the 21st century in response to changes in radiative
forcing and internal climate variability.” The two
scientists also note “the extent and character of the
response of ENSO to increases in greenhouse gases
are still a topic of considerable research” and “given
the results published to date, we cannot yet rule out
possibilities of an increase, decrease, or no change in
ENSO activity arising from increases in CO2.”
Vecchi and Wittenberg conclude their review of
the subject by stating “we expect the climate system
to keep exhibiting large-scale internal variations,” but
they add, “the ENSO variations we see in decades to
Figure 1.4.1.1. The nine member composites of each 11-year come may be different than those seen in recent
QDO cycle for the observations (obs) and the GCM (FOAM): decades.” They admit “we are not currently at a state
(a) is the QDO (thin) and 2.2 year El Niño (thick), (b) is the to confidently project what those changes will be.”
same as (a) except for the 3.6 year cycle, and (c) is the sum of Catto et al. (2012a) write “the El Niño-Southern
the three cycles (thick) shown against the 11 year cycle (thin). Oscillation (ENSO) is linked to the interannual
For (d) and (e) the figure shows the sum of the 2.2 and 3.6 year climate variability of Australia, in part through its
cycle against the entire record and the 11 year cycle for the (d) effect on the sea surface temperatures (SSTs) around
model, and (e) observations. Adapted from Figure 1 White and northern Australia,” as has been documented by
Liu (2008). Hendon (2003) and Catto et al. (2012b). They explain
“it is important that global coupled climate models
The authors therefore provide convincing are able to represent this link between ENSO and
evidence the 11-year solar cycle may be the trigger north Australian SSTs so that we can have more
for El Niño/La Niña events. By using harmonic confidence in the projections of future climate change
analysis on observed and model data, they found for the Australian region.” For the authors’
similar El Niño-related behavior in each, meaning contribution to the topic, “the link between ENSO and
“the solar forced QDO forces the ~3.6 year ENSO north Australian SSTs has been evaluated in the
signal; which in turn forces the ~2.2 year ENSO models participating in CMIP5 with a view to
signal, and so on.” There are two important results to comparing them with the CMIP3 models evaluated in
take away from this work: Models that include solar Catto et al. (2012b).”
forcing have become more proficient at capturing The three Australian researchers’ study revealed
interannual variability, and El Niño and La Niña “the CMIP5 models still show a wide range in their
onsets may be somewhat predictable even 10 years in ability to represent both ENSO events themselves,
advance. Such developments would be a boon for and their relationship to north Australian SST”; “most
long-range forecasting. of the models fail to capture the strong seasonal cycle
In a review paper two years later, Vecchi and of correlation between the Niño-3.4 and north
Wittenberg (2010) “[explored] our current Australian SSTs”; and “the models in general are still
understanding of these issues,” stating it is “of great missing some underlying process or mechanism.”
interest to understand the character of past and future Catto et al. conclude, “gaining a deeper understanding
ENSO variations.” The two researchers at the U.S. of the physical mechanism behind the strong link
National Oceanic and Atmospheric Administration’s between the SSTs in the Niño-3.4 region and to the
Geophysical Fluid Dynamics Laboratory point out north of Australia using these models” is “a vital next
125
step” for this work, which they state is required “to A1B, … to study whether these models can reproduce
elucidate the processes missing from the models that the ENSO and ENSO Modoki patterns” and “their
cannot capture the link.” teleconnections with the Indian summer monsoon,
Zhang and Jin (2012) indicate “ENSO behaviors and also the implications for the future.”
in coupled models have been widely evaluated,” The four researchers from India report “only
citing Neelin et al. (1992), Delecluse et al. (1998), ~1/4th of the models from 20C3M capture either
Latif et al. (2001), Davey et al. (2002), AchuataRao ENSO or ENSO Modoki patterns in June, July,
and Sperber (2002, 2006), Capotondi et al. (2006), August and September.” They note “of this 1/4th,
Guilyardi (2006), and Zhang et al. (2010). However, only two models simulate both ENSO and ENSO
they write, “coupled models still exhibit large biases Modoki patterns as important modes” and “of these
in modeling the basic features of ENSO,” citing two, only one model simulates both ENSO and ENSO
Guilyardi et al. (2009). Among these biases is “a sea Modoki as important modes during both summer and
surface temperature (SST) anomaly (SSTA) too winter.” In addition, they note the two models that
tightly confined to the equator (e.g., Stockdale et al., demonstrate ENSO Modoki, as well as ENSO
1998; Kang et al., 2001).” More specifically, and associated variance in both 20C3M and SRES A1B,
recently, they say, “it was shown that the ENSO project the opposite types of impacts of SRES A1B.
meridional width in the models participating in Phase Roxy et al. say their findings are indicative of
3 of the Coupled Model Inter-comparison Project “the challenges associated with the limitations of the
(CMIP3) is only about two thirds of what is models in reproducing the variability of the monsoons
observed,” citing Zhang et al. (2012). and ENSO flavors, not to speak of failing in capturing
Zhang and Jin ask the obvious question: “Does the potential impacts of global warming as they are
the systematical narrow bias in ENSO width still exist expected to.”
in current models developed for Phase 5 of the CMIP Finally, Koumoutsaris (2013) reports, “currently,
(CMIP5)?” They answer the question by assessing the global climate models disagree in their estimates of
ENSO meridional widths simulated by 15 CMIP5 feedbacks, and this is one of the main reasons for
models and 15 CMIP3 models for the period 1900– uncertainty in future climate projections,” citing Bony
1999, comparing the results of both groups against et al. (2006). He further notes “in order to unveil the
observation-based monthly SST data from the Hadley origin of these inter-model differences, model
Center Sea Ice and Sea Surface Temperature simulations need to be evaluated against observations
(HadISST) data of Rayner et al. (2003). of present climate.”
The analysis indicated “a systematic narrow bias Koumoutsaris estimated “the feedbacks from
in ENSO meridional width remains in the CMIP5 water vapor, lapse-rate, Planck, surface albedo and
models,” although they state the newest results clouds, using models and observations based on the
represent “a modest improvement over previous climate response over the last 30 years,” short-term
models.” Is a modest improvement good enough? feedbacks that “result both from external changes in
That question remains to be answered. the forcing (due to greenhouse gas increases, volcanic
Roxy et al. (2013) point out recent studies have and industrial aerosol emissions) and internal climate
highlighted the existence of a new phenomenon, variations (mostly due to ENSO variability).” The
referred to as the El Niño Modoki, characterized by a Swiss scientist reports “the CMIP3 models show a
warm sea surface temperature (SST) anomaly in the much larger interdecile range for all short-term
central equatorial Pacific and a cold SST anomaly in feedbacks in comparison to the long-term ones,”
the western and eastern Pacific. Some observers of which he states “is also the case for the three models
this phenomenon have argued “the increasing with the most realistic ENSO representation,” citing
frequency of the El Niño Modoki in recent decades is van Oldenborgh et al. (2005). He also indicates the
due to global warming.” Roxy et al., considering it models have difficulty capturing “the position and
imperative to examine the changing teleconnection magnitude of ENSO teleconnection patterns.” In
between ENSO/Modoki and the Indian summer addition, he reports “the uncertainty in the cloud
monsoon, revisited “climate change experiments feedback, using a combination of reanalysis and
under the fourth Assessment Report (AR4) of the satellite data, is still very large.”
Intergovernmental Panel on Climate Change (IPCC), Koumoutsaris concludes his several analyses
namely the twentieth century simulations (20C3M) indicate “important aspects of the ENSO variability
and Special Report on Emissions Scenarios (SRES) are still poorly understood and/or simulated.”
126
Regarding cloud feedback, he says it is difficult to general circulation modeling of the tropical Pacific.
come to “any firm conclusion,” even as to the sign of Journal of Geophysical Research 103: 14,357–14,373.
the feedback. Fedorov, A.V. and Philander, S.G. 2000. Is El Niño
Clearly there remain multiple problems in the changing? Science 288: 1997–2002.
ability of models to reliably simulate various aspects
of climate associated with ENSO events, casting Guilyardi, E. 2006. El Niño mean state-seasonal cycle
further doubt on the overall ability of models to interactions in a multi-model ensemble. Climate Dynamics
simulate the future climate of the planet in general. 26: 329–348.
Guilyardi, E., Wittenberg, A., Fedorov, A., Collins, M.,
References Wang, C., Capotondi, A., Jan, G., Oldenborgh, V., and
Stockdale, T. 2009. Understanding El Niño in ocean-
AchutaRao, K. and Sperber, K.R. 2002. Simulation of the atmosphere general circulation models: Progress and
El Niño Southern Oscillation: Results from the Coupled challenges. Bulletin of the American Meteorological
Model Intercomparison Project. Climate Dynamics 19: Society 90: 325–340.
191–209.
Huber, M. and Caballero, R. 2003. Eocene El Niño:
AchutaRao, K. and Sperber, K.R. 2006. ENSO simulation Evidence for robust tropical dynamics in the “Hothouse.”
in coupled ocean-atmosphere models: are the current Science 299: 877–881.
models better? Climate Dynamics 27: 1–15.
Jin, E.K., Kinter III, J.L., Wang, B., Park, C.-K., Kang, I.-
Bony, S., Colman, R., Kattsov, V.M., Allan, R.P., S., Kirtman, B.P., Kug, J.-S., Kumar, A., Luo, J.-J.,
Bretherton, C.S., Dufresne, J., Hall, A., Hallegatte, S., Schemm, J., Shukla, J., and Yamagata, T. 2008. Current
Ingram, W., Randall, D.A., Soden, B.J., Tselioudis, G., and status of ENSO prediction skill in coupled ocean-
Webb, M.J. 2006. How well do we understand and evaluate atmosphere models. Climate Dynamics 31: 647–664.
climate change feedback processes? Journal of Climate 19:
3445–3482. Jin, F. F. 1997. An equatorial ocean recharge paradigm for
ENSO. Part I: Conceptual Model. Journal of Atmospheric
Capotondi, A., Wittenberg, A. and Masina, S. 2006. Spatial Science 54: 811–829.
and temporal structure of tropical Pacific interannual
variability in 20th century coupled simulations. Ocean Joseph, R. and Nigam, S. 2006. ENSO evolution and
Modeling 15: 274–298. teleconnections in IPCC’s twentieth-century climate
simulations: Realistic representation? Journal of Climate
Catto, J.L., Nicholls, N., and Jakob, C. 2012a. North 19: 4360–4377.
Australian sea surface temperatures and the El Niño-
Southern Oscillation in the CMIP5 models. Journal of Kang, I.-S., An, S.-I. and Jin, F.-F. 2001. A systematic
Climate 25: 6375–6382. approximation of the SST anomaly equation for ENSO.
Journal of the Meteorological Society of Japan 79: 1–10.
Catto, J.L., Nicholls, N., and Jakob, C. 2012b. North
Australian sea surface temperatures and the El Niño- Koumoutsaris, S. 2013. What can we learn about climate
Southern Oscillation in observations and models. Journal feedbacks from short-term climate variations? Tellus A 65:
of Climate 25: 5011–5029. 10.3402/tellusa.v65i0.18887.
Cerveny, R.S. and Shaffer, J.A. 2001. The moon and El Landsea, C.W. and Knaff, J.A. 2000. How much skill was
Niño. Geophysical Research Letters 28: 25–28. there in forecasting the very strong 1997–98 El Niño?
Bulletin of the American Meteorological Society 81: 2107–
Davey, M.K., Huddleston, M., Sperber, K., Braconnot, P., 2119.
Bryan, F., Chen, D., Colman, R., Cooper, C., Cubasch, U.,
Delecluse, P., DeWitt, D., Fairhead, L., Flato, G., Gordon, Latif, M., Sperber, K., Arblaster, J., Braconnot, P., Chen,
C., Hogan, T., Ji, M., Kimoto, M., Kitoh, A., Knutson, T., D., Colman, A., Cubasch, U., Cooper, C., Delecluse, P.,
Latif, M., LeTreut, H., Li, T., Manabe, S., Mechoso, C., DeWitt, D., Fairhead, L., Flato, G., Hogan, T., Ji, M.,
Meehl, G., Power, S., Roeckner, E., Terray, L., Vintzileos, Kimoto, M., Kitoh, A., Knutson, T., Le Treut, H., Li, T.,
A., Voss, R., Wang, B., Washington, W., Yoshikawa, I., Manabe, S., Marti, O., Mechoso, C., Meehl, G., Power, S.,
Yu, J., Yukimoto, S., and Zebiak, S. 2002. STOIC: a study Roeckner, E., Sirven, J., Terray, L., Vintzileos, A., Voss,
of coupled model climatology and variability in tropical R., Wang, B., Washington, W., Yoshikawa, I., Yu, J., and
ocean regions. Climate Dynamics 18: 403–420. Zebiak, S. 2001. ENSIP: the El Niño simulation
intercomparison project. Climate Dynamics 18: 255–276.
Delecluse, P., Davey, M.K., Kitamura, Y., Philander,
S.G.H., Suarez, M., and Bengtsson, L. 1998. Coupled L’Ecuyer, T.S. and Stephens, G.L. 2007. The tropical
atmospheric energy budget from the TRMM perspective.
127
Part II: Evaluating GCM representations of the sensitivity Research 106: 4349–4367.
of regional energy and water cycles to the 1998–99 ENSO
cycle. Journal of Climate 20: 4548–4571. Zhang, W., Li, J. and Zhao, X. 2010. Sea surface
temperature cooling mode in the Pacific cold tongue.
McLean, J.D., de Freitas, C.R., and Carter, R.M. 2009. Journal of Geophysical Research 115: 10.1029/
Influence of the Southern Oscillation on tropospheric 2010JC006501.
temperature. Journal of Geophysical Research 114:
10.1029/2008JD011637. Zhang, W. and Jin, F.-F. 2012. Improvements in the
CMIP5 simulations of ENSO-SSTA Meridional width.
Neelin, J.D., Latif, M., Allaart, M.A.F., Cane, M.A., Geophysical Research Letters 39: 10.1029/2012GL053588.
Cubasch, U., Gates, W.L., Gent, P.R., Ghil, M., Gordon,
C., Lau, N.C., Mechoso, C.R., Meehl, G.A., Oberhuber, Zhang, W., Jin, F.-F., Li, J., and Zhao, J.-X. 2012. On the
J.M., Philander, S.G.H., Schopf, P.S., Sperber, K.R., Sterl, bias in simulated ENSO SSTA meridional widths of
K.R., Tokioka, T., Tribbia, J., and Zebiak, S.E. 1992. CMIP3 models. Journal of Climate: org/10.1175/JCLI-D-
Tropical air-sea interaction in general circulation models. 12-00347.1.
Climate Dynamics 7: 73–104.
Overpeck, J. and Webb, R. 2000. Nonglacial rapid climate 1.4.2 Atlantic and Pacific Ocean Multidecadal
events: Past and future. Proceedings of the National Variability (AMV)
Academy of Sciences USA 97: 1335–1338. The study of Semenov et al. (2010) shows natural
variability contributed significantly to the warming
Paeth, H., Scholten, A., Friederichs, P., and Hense, A.
2008. Uncertainties in climate change prediction: El Niño- period in the last few decades of the twentieth century
Southern Oscillation and monsoons. Global and Planetary and even may be responsible for the majority of the
Change 60: 265–288. warming. The work of Petoukhov and Semenov
(2010) has been cited as evidence that cold winters
Rayner, N.A., Parker, D.E., Horton, E.B., Folland, C.K., would be consistent with anthropogenic global
Alexander, L.V., Rowell, D.P., Kent, E.C., and Kaplan, A. warming. The reasoning follows this path: global
2003. Global analyses of sea surface temperature, sea ice,
warming in the Arctic would leave less sea ice, and
and night marine air temperature since the late nineteenth
century. Journal of Geophysical Research 108: 10.1029/ with less sea ice, more heat is released to the air. As a
2002JD002670. result, continents could be colder since more
moisture-laden air can mean more clouds and more
Roxy, M., Patil, N., Ashok, K., and Aparna, K. 2013. snow, and thus colder temperatures over land. This is
Revisiting the Indian summer monsoon-ENSO links in the called the “Warm Arctic-Cold Continents”
IPCC AR4 projections: A cautionary outlook. Global and conjecture.
Planetary Change: 10.1016/j.gloplacha.2013.02.003.
Semenov et al. (2010) first examined the results
Stockdale, T.N., Busalacchi, A.J., Harrison, D.E., and of several different model experiments performed by
Seager, R. 1998. Ocean modeling for ENSO. Journal of others, finding “North Atlantic sea surface
Geophysical Research 103: 14,325–14,355. temperature changes do project onto Northern
Hemisphere and global surface air temperatures in
Van Oldenborgh, G.J., Philip, S.Y., and Collins, M. 2005.
El Niño in a changing climate: a multi-model study. Ocean these models,” especially as these relate to the
Science 1: 81–95. warming of the early twentieth century. They used the
ECHAM5 model (European Centre for Medium
Vecchi, G.A. and Wittenberg, A.T. 2010. El Niño and our Range Forecasting general circulation model, at the
future climate: where do we stand? WIREs Climate Change Max Planck Institute, Hamburg) and ran the model
1: 10.1002/wcc.33. for 80 to 100 years using heat flux patterns (heat
Walsh, K. and Pittock, A.B. 1998. Potential changes in transport from the surface into the atmosphere or vice
tropical storms, hurricanes, and extreme rainfall events as a versa) related to the recent positive and negative
result of climate change. Climatic Change 39: 199–213. extremes of the Atlantic Multidecadal Variability
(AMV) for three differing regions in the North
White, W.B. and Liu, Z. 2008. Non-linear alignment of El
Atlantic and Arctic.
Niño to the 11-yr solar cycle. Geophysical Research
Letters, 35, L19607, doi:10.1029/2008GL034831. The authors’ work demonstrated the AMV forced
heat fluxes project statistically significant temperature
White, W.B., Cayan, D.R., Dettinger, M.D., and Auad, G. and pressure changes onto the Northern Hemisphere
2001. Sources of global warming in upper ocean climate, including influencing the NAO. The authors
temperature during El Niño. Journal of Geophysical
128
conclude, “the results show that such an internal and delayed relative to the observed response,” a tendency
regionally confined climate variation can drive they say “is consistent with model biases toward
relatively large surface climate anomalies on regional, deeper oceanic mixed layers and weaker air-sea
hemispheric, and even global scales. The Arctic plays feedbacks.” They found “the simulated amplitude of
an important role in this, explaining about 60% of the the ENSO-related signal in the North Pacific is
total Northern Hemisphere surface air temperature overestimated by about 30%” and “model power
response.” They also point out these results are spectra of the PDO signal and its ENSO-forced
applicable to recent decades. However, in their study component are redder than observed because of errors
the forcing of the Arctic on lower latitudes such as originating in the tropics and extratropics.”
Europe is always in the same direction (warmer Lienert et al. describe three implications of their
Arctic means warmer Europe), contradicting the findings. First, “because the simulated North Pacific
“Warm Arctic-Cold Continents” conjecture. response lags ENSO unrealistically, seasonal
The results of Semenov et al. and Petoukhov and forecasts may tend to exhibit insufficient North
Semenov are not surprising, as it has been theorized Pacific responses to developing El Niño and La Niña
for some time that multidecadal oscillations, or events in the first few forecast months.” Second, “at
natural variations, can imprint on regional, longer forecast lead times, North Pacific SST
hemispheric, or even global climate change. anomalies driven by ENSO may tend to be
However, in both studies, the modeling strategy is overestimated in models having an overly strong
meant to test the sensitivity of the models to forced ENSO, as the models drift away from observation-
interdecadal variability. In neither study was the based initial conditions and this bias sets in.” And
model able to capture interdecadal variability as it third, “the relative preponderance of low-frequency
occurs in nature. Additionally, neither paper explicitly variability in the models suggests that climate
endorsed its results as being supportive of forecasts may overestimate decadal to multidecadal
anthropogenic global warming as some have claimed. variability in the North Pacific.”
Although Semenov et al. are careful to add statements Kim et al. (2012) state “the Coupled Model
of assurance that their results do not contradict the Intercomparison Project Phase 5 (CMIP5) has devised
global warming orthodoxy, they offer no evidence in an innovative experimental design to assess the
support of it. predictability and prediction skill on decadal time
Lienert et al. (2011) state “climate models are scales of state-of-the-art climate models, in support of
increasingly being used to forecast future climate on the Intergovernmental Panel on Climate Change
time scales of seasons to decades,” and “since the (IPCC) 5th Assessment Report,” citing Taylor et al.
quality of such predictions of the future evolution of (2012). To date, however, they note decadal
the PDO [Pacific Decadal Oscillation] likely depends predictions from different CMIP5 models “have not
on the models’ ability to represent observed PDO been evaluated and compared using the same
characteristics, it is important that the PDO in climate evaluation matrix,” a problem they resolve with their
models be evaluated.” study for some of the models. Kim et al. assessed the
Working with observed monthly-mean SST (sea CMIP5 decadal hindcast-forecast simulations of
surface temperature) anomalies they obtained from seven state-of-the-art ocean-atmosphere coupled
the Met Office Hadley Centre Sea Ice and Sea models for situations where “each decadal prediction
Surface Temperature version-1 (Rayner et al., 2003) consists of simulations over a 10-year period each of
data set for 1871–1999, as well as the extended which are initialized every five years from climate
reconstructed SST version-3b data set (Smith et al., states of 1960/1961 to 2005/2006.”
2008), Lienert et al. assessed the ability of 13 The three U.S. researchers report “most of the
atmosphere-ocean global climate models from the models overestimate trends,” in that they “predict less
third phase of the Coupled Model Intercomparison warming or even cooling in the earlier decades
Project (CMIP3), conducted in support of the IPCC compared to observations and too much warming in
Fourth Assessment Report (Solomon et al., 2007), to recent decades.” They also report “low prediction
“reproduce the observed relationship between tropical skill is found over the equatorial and North Pacific
Pacific forcing associated with ENSO and North Ocean” and “the predictive skill for the Pacific
Pacific SST variability associated with the PDO.” Decadal Oscillation index is relatively low for the
The three Canadian researchers found “the entire period.” They conclude “the prediction of
simulated response to ENSO forcing is generally decadal climate variability against a background of
129
global warming is one of the most important and the Northern Hemisphere spring season only, and not
challenging tasks in climate science.” during the rest of the year.
Flaws in the construction of the models, in their
References numerics and in their physics, are discussed often.
But analyses using even erroneous models can
Kim, H.-M., Webster, P.J., and Curry, J.A. 2012. sometimes be useful not only in learning where
Evaluation of short-term climate change prediction in models are deficient but also in identifying where or
multi-model CMIP5 decadal hindcasts. Geophysical how the observations need to be improved.
Research Letters 39: 10.1029/2012GL051644. Liu et al. (2012) used the Community Climate
Lienert, F., Fyfe, J.C., and Merryfield, W.J. 2011. Do System Model version 3 (CCSM3), which includes an
climate models capture the tropical influences on North atmospheric model, a land surface model, and a
Pacific sea surface temperature variability? Journal of dynamic ocean model. The model was of intermediate
Climate 24: 6203–6209. resolution in the horizontal, the equivalent of 2.8
degrees latitude/longitude. There are 26 model layers
Petoukhov, V. and Semenov, V.A. 2010. A link between
reduced Barents-Kara sea ice and cold winter extremes in the atmosphere. The authors focused on the first
over northern continents. Journal of Geophysical one to two years of the model integration in order to
Research—Atmospheres. 115: D21111, doi:10.1029 locate the problems causing the double ITCZ over the
/2009JD013568. tropics between 170° E and 150°W. The observed
data used as input were archived at the Massachusetts
Rayner, N.A., Parker, D.E., Horton, E.B., Folland, C.K., Institute of Technology, and the cloud data were
Alexander, L.V., Rowell, D.P., Kent, E.C., and Kaplan, A.
derived from the International Satellite Cloud
2003. Global analyses of sea surface temperature, sea ice,
and night marine air temperature since the late nineteenth Climatology Project.
century. Journal of Geophysical Research 108: 10.1029/ Liu et al. learned warm biases in the sea surface
2002JD002670. temperatures developed within two years of the initial
start-up, and these became substantial and similar to
Semenov, V.A., Latif, B., Dommenget, D., Keenlyside, those of other studies after five simulation years. The
N.S., Strehz, A., Martin, T., and Park, W. 2010. The Impact CCSM model also produced a cold bias south of
of North Atlantic–Arctic Multidecadal Variability on
10°S. The warm bias was present in all seasons. The
Northern Hemisphere surface air temperature. Journal of
Climate 23: 5668–5677. effect of the warm and cold bias was to establish
unrealistic temperature gradients, which in turn
Smith, T.M., Reynolds, R.W., Peterson, T.C., and affected the winds and the latent heat flux (moisture)
Lawrimore, J. 2008. Improvements to NOAA’s historical into the atmosphere.
merged land-ocean surface temperature analysis (1880– The authors additionally found the warm bias was
2006). Journal of Climate 21: 2283–2296. due to excessive solar radiation, at least at first,
Solomon, S., Qin, D., Manning, M., Marquis, M., Averyt, augmented by excess latent heating later. The root
K., Tignor, M.B., Miller Jr., H.L., and Chen, Z. (Eds.). cause was less cloudiness in the model than was
2007. Climate Change 2007: The Physical Science Basis. observed in the region. These biases induced errors in
Cambridge University Press, Cambridge, United Kingdom. the ocean structure of the region studied, resulting in
a modeled eastward current where a westward current
Taylor, K.E., Stouffer, R.J., and Meehl, G.A. 2012. An
overview of CMIP5 and the experiment design. Bulletin of exists in the observations. An experiment was then
the American Meteorological Society 93: 485–498. run by artificially increasing the cloudiness within the
region and the problems described above were
alleviated.
1.4.3 Intertropical Convergence Zone Liu et al. eventually traced the problem of
It is well-known among scientists in the climate insufficient cloudiness in the study region to low
modeling community that GCMs erroneously produce cloud liquid water path in the model (Figure 1.4.3.1),
a double Intertropical Convergence Zone (ITCZ) in but “whether this is due to insufficient absorption of
the Central Pacific much more often than is typically solar radiation in clouds, or insufficient cloud amount
observed. The ITCZ is a “belt” of low pressure that or aerosols, is not clear.”
girdles the equatorial tropics and identifies the The goal of Liu et al.’s work was to determine
convergence of Earth’s trade wind belts. Usually, a how the CCSM model developed a double ITCZ in
double ITCZ is observed in the Central Pacific during the Central Pacific. This double Central Pacific ITCZ
130
Figure 1.4.3.1. The differences in the total cloud amount between the model and observed cloudiness (%) for the first
month of the (a) atmospheric model only, and (b) the full CCSM ensemble mean. The green box bounded by 170oE and
150oW and 5° to 10°S is the domain over which the authors studied the heat budgets. Adapted from Liu et al. (2012).
appears to be a problem for many models. The In a letter to Nature, Cai et al. (2012) compared
appearance of a double ITCZ means a model fails to the larger-scale components of the SPCZ rainfall
simulate tropical climate well and, given the anomalies derived from observations during the
interactions of this region with the mid-latitudes, period 1979–2008 to rainfall anomalies generated by
these simulation problems would be felt in the model 17 CMIP3 general circulation model simulations
far from the study region. covering the period 1891–2000. These simulations
included known natural and anthropogenic forcings.
Reference The team of researchers then generated 90-year
simulations using the A2 greenhouse emission
Liu, H., Zhang, M., and Lin, W. 2012. An investigation of scenarios, which assume business as usual or
the initial development of the double-ITCZ warm SST accelerating greenhouse gas emissions. Cai et al. also
biases in the CCSM. Journal of Climate 25: 140–155. applied the same strategy with CMIP5 models under a
scenario that levels out at 850 ppm during 200 years
1.4.4 South Pacific Convergence Zone assuming 1 percent increases per year until
The South Pacific Convergence Zone (SPCZ) is a stabilization. This is similar to the A2 scenario.
permanent feature in the general circulation that The work of Cai et al. revealed stronger SPCZ
stretches southeastward from the equatorial region of events will occur with greater frequency in the future
the west Pacific into the central mid-latitude Pacific under the increased greenhouse gas scenarios in both
near Polynesia. This band of precipitation has drawn models (doubled in CMIP3 and tripled in CMIP5).
the interest of the climatological community because This implies increased frequency of drought for the
it behaves like the ITCZ in the tropics but is driven by regions of the southwest Pacific. Both models also
jet-stream dynamics in the mid-latitudes. Like the implied stronger El Niño events occur more
ITCZ, it is a locus for the occurrence of precipitation frequently in the future scenario, but not as frequently
in the regions it impacts. as the increased SPCZ occurrences. Curiously, these
Previous research has demonstrated the observed results also imply SPCZ behavior decouples from that
behavior and location of the SPCZ varies in relation of El Niño.
to the phase of El Niño. During an El Niño year, the The conclusion that stronger SPCZ events will
SPCZ is located closer to the equator and can be occur in the future was based on a greenhouse gas
oriented in a more zonal fashion, and these changes emission scenario considered quite extreme.
are more robust during stronger El Niño events. At Additionally, although there may be a physical reason
such times, Indonesia, Australia, and southwest for future decoupling of the relationship between
Pacific nations are subjected to reduced rainfall, SPCZ strength and location and the strength of El
which of course affects the regional ecology. Niño, it should be noted the modeled SPCZ
131
orientation seems to be more zonal than observed The authors admit to some uncertainty in the
even before the future scenarios are generated observed SPCZ position: “Although the simulated
(compare Figure 1.4.4.1 and Figure 1.4.4.2). frequency in the control period is comparable to the
Figure 1.4.4.1. The variability of observed rainfall for the (a) spatial pattern of the
largest-scale component of the observed precipitation anomalies that was filtered out,
and (b) second-largest component of the observed anomalies extracted from the
satellite-era rainfall anomaly data (mm d-1) (Global Precipitation Climatology Project
version 225) focusing on the western South Pacific during the austral summer
(December to February). The first (second) principal patterns account for 47% (16%) of
the total variance. The SPCZ position (max rainfall greater than 6mm d-1) for El Niño
(green line), La Niña (blue line), and neutral (black line) states is superimposed in a,
and the position for zonal SPCZ events (red line) in b. Cold (warm) contours indicate
increased (decreased) rainfall per one standard deviation (s.d.) change. Adapted from
Figure 1 in Cai et al. (2012).
Figure 1.4.4.2. As in Figure 1.5.4.1, except for the CMIP model runs.
132
observed frequency, the latter is based on model was based on one used at the Geophysical
observations over some three decades and may carry Fluid Dynamics Laboratory, a model with relatively
a large uncertainty.” Like many modeling studies, the coarse resolution (about 2.8° latitude/longitude) and
outcome of Cai et al. is properly understood as 30 levels in the vertical. The planet “surface” was
nothing more than a scenario based on a specific set water-covered only. The oceans included simple
of assumptions, with limitations on the ability to dynamics for heat transport (which was turned “on”
capture the phenomenon in question precisely. Their and “off”), and the model was driven with the annual
results have only limited scientific value. mean insolation, thus there was no diurnal or annual
cycle. In this way the authors could isolate two
Reference variables, the strength of the HC versus global
temperature (controlled by the absorption of
Cai, W., Langaigne, M., Borlace, S., Collins, M., Cowan, longwave radiation only).
T., McPhadden, M.J., Timmermann, A., Power, S., Brown, The authors established a baseline climate for
J., Menkes, C., Ngari, A., Vincent, E.M. and Widlansky, their aqua-planet and ran the model with and without
M.J. 2012. More extreme swings of the South Pacific ocean dynamics. The results were more realistic with
Convergence Zone due to greenhouse warming. Nature
the ocean dynamics. Then, the longwave absorption, a
488: 365-369.
function of latitude and pressure (height), was varied
by a constant amount between 0.2 and 6.0 times the
1.4.5 Hadley Circulation control value. This produced a global climate with
The Hadley Circulation, or Hadley Cell, is an planetary temperatures varying between 260 K and
important general circulation of Earth’s climate. It is 315 K (roughly -13°C to 42°C; today it is roughly
named for Sir George Hadley, who first attempted to 15°C). The equator-to-pole temperature differences
describe Earth’s circulation in a general way. Warm decreased linearly with increasing temperature.
equatorial air rises, while cold polar air sinks, and a In their simulations, the strength of the Hadley
hemisphere-wide circulation loop is formed. Hadley Cell did not change linearly (Figure 1.4.5.1) but rather
was partly correct but failed to account for the was more parabolic in shape, which makes the
Coriolis Effect, which was discovered by later situation more complicated. The Hadley Circulation
researchers. in their model was weaker in both very cold and very
The Hadley Cell emerges from an analysis of the warm climates. Also, Figure 1.4.5.1 implies ocean
tropics when atmospheric motions are averaged on dynamics become less important in determining the
the time scale of a month or more. In a classical view strength of the Hadley Circulation as climate warms.
of this circulation, there are upward motions and low This is true because the equator-to-pole temperature
pressure at the equator (mainly associated with deep difference becomes negligible. In colder climates, the
convection clouds), poleward moving air aloft, weakening is the result of a more geostrophic, or
downward motion around 30o latitude, associated zonal, atmospheric flow. A more zonal flow would
with the subtropical highs, and equatorward moving imply less wave action in the jet stream, which means
air (the trade winds) in the lower atmosphere. Studies less energy exchange. Today’s climate is roughly at
have suggested the latitudinal extent of the Hadley the maximum in the ocean dynamics curve.
Cell may have widened in the past few decades. The behavior of atmospheric systems can be
Levine and Schneider (2011) used a crude complicated even when there are few variables. This
atmospheric general circulation model to examine the has led to studies whose findings initially seem to be
strength and span of the HC over a wide range of contradictory, for example, in a warmer climate. The
climates. They performed this study because some behavior of changes in the strength of the Hadley
studies show the HC strengthens, while others show it Circulation is nonlinear even in the simple model of
weakens, in response to global warming. As Levine Levine and Schneider (2011). Adding complexity to
and Schneider report in describing the rationale the model would make interpretation of the output
behind their study, “despite a large body of more difficult. This demonstrates the importance of
observations and numerous studies with GCMs, it understanding the fundamental behavior of
remains unclear how the width and strength of the phenomena like the Hadley Circulation.
Hadley Circulation are controlled.”
Levine and Schneider used an idealized radiative
transfer scheme and moist thermodynamics. The
133
convection. This is but one factor that causes GCMs

fail, to a certain degree, in representing the large scale
correctly.
Subramanian et al. (2011) used the National
Center for Atmospheric Research (NCAR)
Community Climate System Model version 4
(CCSM4) model to examine its ability to capture the
MJO, which included an upgrade to the convective
parameterization scheme. This upgrade “improves the
correlation between intraseasonal convective heating
and intraseasonal temperature, which is critical for the
buildup of available potential energy.”
The authors used a 500-year simulation of the
CCSM4 using pre-1850 conditions as a control run.
They then extracted two ten-year periods from these
data, each representing the strongest and weakest
Figure 1.4.5.1. The strength of HC in simulations with ENSO variability in the 500-year period. They
(solid, circles) and without (dashed, squares) ocean
compared this data set to observed Outgoing
dynamics. Shown is the absolute value of the mass flux at
the latitude of its maximum and at the level σc = 0.7, Longwave Radiation (OLR), a commonly used proxy
averaged for both hemispheres. The filled symbols are the for convection, as well as tropical precipitation and
reference simulations. Adapted from Levine and Schneider zonal winds at 850 and 200 hPa taken from the
(2011), their Figure 4. NCAR/NCEP reanalyses. Lastly, the authors
examined the relationship between the MJO and such
general circulation features as ENSO and the
monsoons in both the model and observations.
Reference
Subramanian et al. found the CCSM4 produced a
feature that propagated eastward in the “intraseasonal
Levine, X.J. and Schneider, T. 2011. Response of the
Hadley Circulation to Climate Change in an Aquaplanet zonal winds and OLR in the tropical Indian and
GCM Coupled to a Simple Representation of Ocean Heat Pacific Oceans that are generally consistent with MJO
Transport. Journal of the Atmospheric Sciences 68: 769– characteristics.” Thus the model performed well
783. overall in capturing the MJO (Figure 1.4.6.1), but
there were still some differences. Whereas the
observations produced a strong (weaker) wave
1.4.6 The Madden-Julian Oscillation number one (two and three) in the data, the modeled
The Madden-Julian Oscillation (MJO) is a tropical MJO showed more coherency among the wave
phenomenon discovered in the early 1970s; it is a 30- numbers one-three. This suggests stronger Kelvin
to 70-day oscillation in the west-to-east component of wave activity at the higher wave numbers in the
the tropical winds between 20°N and 20°S. The MJO model.
has been detected in tropical convection and When examining the relationship of the MJO to
precipitation, both of which are enhanced in the other phenomenon, the authors found MJO activity
“trough” phase. Although the dynamics that drive the was enhanced (weaker) during El Niño (La Niña)
MJO are not fully understood, it is known to interact years. Also, the MJO occurred more often in the
with larger-scale phenomena such as tropical Indian Ocean monsoon region during negative shear
cyclones, monsoons, and El Niño and the Southern regimes (shear defined by the zonal wind at 850 hPa
Oscillation (ENSO), as well as with the general minus the meridional [north-south] wind at 200 hPa).
circulation itself. The MJO has been linked to mid- The MJO occurred more often when the Hadley
latitude circulations, especially during the warm Circulation in the tropics was weaker as well. These
season and over North America. phenomena are interrelated, and thus the authors state,
General circulation models have had difficulty in the “MJO could thereby be simultaneously affected in
representing the MJO and its impact on larger-scale multiple ways when these type of large-scale climate
phenomena could not be captured, due primarily to mode interactions occur and possibly feed back onto
the inadequate representation of the associated the entire coupled system.”
134
Figure 1.4.6.1. A Hovemöller plot of the zonal winds at 850 hPa filtered to retain the 20-100 day signal for the (a)
NCAR/NCEP reanalyses from 1997 and (b) year 3 of the CCSM4 run. The arrow is added to demonstrate the eastward
propagation of weaker 850 hPa winds which represents the MJO. Adapted from Subramanian et al. (2011).
The outcome of this work demonstrates how there is a stationary ridge of high pressure in the mid-
difficult it is to represent the current state of the latitude jet stream. This phenomenon is typically
climate, as well as the interannual variability in the associated with unusually warm and dry weather in
climate system. Models continue to improve through areas where these high-pressure ridges form, and
the increase in resolution and improvement in sub- cooler or wetter conditions upstream and downstream
grid-scale physical processes. Even though this model of where they occur. The Western European heat
showed a distinct improvement in representing the wave of 2003, the extreme heat in Russia in 2010 and
MJO, there were still some differences between the the downstream flooding in Pakistan, and the cold
model and the observed MJO. The MJO represents temperatures over most of North America and Europe
only one critical link between the small-scale during December 2010 are recent examples of
atmospheric motions and the general circulation. blocking and its impact on regional weather.
The first investigation into blocking
characteristics in an increased CO2 and warmer
Reference
environment was performed by Lupo et al. (1997)
using the Community Climate Model version 1
Subramanian, A.C., Jochum, M., Miller, A.J., Murtugudde,
R., Neale, R.B., and Waliser, D.E. 2011. The Madden- (CCM1). The results of their blocking climatology
Julian Oscillation in CCSM4. Journal of Climate 24: 6261– using CCM1 were comparable to those of Bates and
6282. Meehl (1986), who used CCM0B. Both of these
efforts found significant differences from
observations: in the Atlantic there were fewer events,
1.4.7 Atmospheric Blocking and in the Northern Hemisphere there were more
A phenomenon not often discussed in climate change summer season events. Model blocking also was
studies is atmospheric blocking, which develops when weaker and less persistent than observations.
135
When present-day CO2 concentrations were projections.

doubled, the researchers generally found blocking With respect to the re-analysis data, Kreienkamp
anticyclones were more persistent but weaker than et al. found little evidence of a statistically significant
their counterparts in the control experiment. All other trend over the period 1951–2007 apart from a weak
characteristics of the overall sample—such as decrease in the European region; this decrease
frequency of occurrence, size, preferred genesis suggests extreme weather events caused by blocking
locations, and annual variations in size and events probably have also declined. With respect to
intensity—were generally similar to the control model simulations, they found the models showed
experiment. The regional and seasonal analysis little change in the frequency, seasonality, or
identified significant differences between the two interannual variability of blocking for the Atlantic/
climatologies. Perhaps the most striking difference European region as a whole but a significant decrease
was the threefold or more increase in continental in Central European region frequency.
region block occurrences and total block days, due to Another study of blocking by Scaife et al. (2010)
the appearance of the observed western Asian set out to determine whether model error in the large-
Continental maximum in the double CO2 run. There or small-scale processes is responsible for the
also was an increase in blocking frequency over the underestimate of blocking. They first produced a
North American continent compared to the control blocking climatology using the Hadley Centre Global
and observations. Environmental Model (HadGEM), an atmospheric
A newer investigation into the blocking general circulation model. They used a version of the
phenomenon by Kreienkamp et al. (2010) used zonal index to diagnose blocking (Berrisford et al.,
National Centers for Atmospheric Research re- 2007) and observed data archived at the European
analyses to examine the occurrence of blocking Centre for Medium Range Forecasting (ERA-40 re-
events over Europe since the 1950s, using a well- analyses) for comparison (Figure 1.4.7.1). Then the
known blocking index (Tibaldi and Molteni, 1990). authors extended the study to 18 other simulations
Kreienkamp et al. employed the atmospheric general produced using various GCMs for the
circulation model (ECHAM) used by the IPCC in an Intergovernmental Panel on Climate Change (IPCC)
effort to determine how well the model simulated Fourth Assessment Report.
such blocking. They also examined two climate Since the zonal index is generally used to
warming scenarios (A1B and B1) for the twenty-first represent blocking, the authors separated the zonal
century in order to infer whether blocking will mass gradient produced by the HadGEM model and
become more or less common in based on model in the observed data into a climate mean (large-scale)
Figure 1.4.7.1. The climatological frequency of the winter season blocking using the Berrisford et al (2007) index (not
blocking frequency). Shown are the (left) ERA-40 reanalyses, (middle) HadGEM Model, and (right) the model error. The
hemispheric nature of the model error shows that it is present at all longitudes. Units are K-1. Adapted from Figure 3 in
Scaife et al. (2010).
136
component versus that of the time-varying portion the lack of trends in Northern Hemisphere blocking
(small-scale). When they replaced the large-scale patterns. Screen and Simmonds (2013) studied
model data with the observed (keeping the small-scale “observed changes (1979–2011) in atmospheric
model part), they showed the blocking frequency was planetary-wave amplitude over northern mid-
reproduced more faithfully. The same result was latitudes, which have been proposed as a possible
found in the IPCC simulations and by Lupo et al. mechanism linking Arctic Amplification (AA) and
(1997). mid-latitude weather extremes.” However, they found
Scaife et al. state, “commonly applied statistics few if any large or statistically significant changes,
based on absolute measures (such as the reversal of instead noting “Statistically significant changes in
the geopotential height or potential vorticity either metric are limited to few seasons, wavelengths
gradients) show large errors in current climate and longitudinal sectors” and “Even ignoring the lack
models, but these are directly attributable to a large of significance, there is no clear tendency towards
degree to errors in the main state.” Scaife et al. also larger or smaller [wave amplitude] (60% negative
show improvements in the large-scale model trends and 40% positive trends). Nor is there a
performance can greatly improve the capability of the general tendency towards longer or shorter
model to reproduce blocking characteristics. wavelengths (i.e., longer wavelengths are not
The latest study by Mokhov et al. (2013) for the increasing in zonal amplitude at the expense of
projection of blocking in the twenty-first century shorter wavelengths), or vice versa.”
using the Institut Pierre Simon Laplace Climate Most recently, Barnes (2013) investigated “trends
Model versions 4 (IPSL-CM4) GCM and the SRES- in the meridional extent of atmospheric waves over
A1B and SRES-A2 scenarios showed a slight increase North America and the North Atlantic … in three
in the number and persistence of blocking events over reanalyses.” Barnes’s work “demonstrated that
the Atlantic-Eurasian region and no change in the previously reported positive trends are an artifact of
interannual or interdecadal variability. These results the methodology” and further “the mechanism put
are similar to the other studies examined here. forth by previous studies (e.g. Francis and Vavrus
Although caution should be emphasized in [2012]; Liu et al. [2012]), that amplified polar
interpreting the model projections, the findings of warming has led to the increased occurrence of slow-
these studies are good news, for they suggest the moving weather patterns and blocking episodes, is
number of heat waves and/or cold waves that can be unsupported by the observations.”
attributed to atmospheric blocking will not increase One of the criticisms of model projections used to
for the Atlantic/European region in the twenty-first produce climate change scenarios for the twenty-first
century. In fact, the model output suggests fewer of century is that the models still have difficulty
these occurrences and/or a shorter duration of such reproducing various aspects of the observed climate,
events (e.g., Kreiencamp et al., 2010). including even the large-scale features (Scaife et al.,
Studies by Liu et al. (2012) and Jaiser et al. 2010). Blocking is one of these phenomena, and it is
(2012) find an increase in observed blocking since often used as a surrogate for the occurrence of
1995 consistent with that found by Mokhov et al. extreme conditions (heat waves, cold waves). If the
(2013), but rather than linking this to internal models cannot adequately represent blocking in the
variability, including changes in the phases of current climate, results regarding the occurrence of
multidecadal oscillations, they attribute the change to extreme events in future climate change scenarios
anthropogenic climate warming. A paper by must be examined with caution.
Hakkinen et al. (2011) takes a more measured
approach, stating, “the warm-ocean/cold-land References
anomaly pattern has been linked to a dynamical
environment favorable for blocking” and “the Barnes, E. 2013. Revisiting the evidence linking Arctic
possibility of coupled interaction of atmosphere with Amplification to extreme weather in the midlatitude.
[Atlantic Multidecadal Variability] seems likely, Geophysical Research Letters, in press: 10.1002/grl.50880.
given the long-period variability of blocking reported Bates, G.T. and Meehl, G.A. 1986. The effect of CO2
here and in the even longer paleoclimate time series.” concentration on the frequency of blocking in a general
The latter position is more consistent with the studies circulation model coupled to a single mixed layer ocean
reviewed here. model. Monthly Weather Review 114: 687–701.
Two studies provide additional evidence for the
137
Berrisford, P., Hoskins, B.J., and Tyrlis, E. 2007. Blocking atmospheric conditions (e.g., low wind shear) are just
and Rossby wave breaking on the dynamical tropopause in as critical. These atmospheric conditions can be
the Southern Hemisphere. Journal of Atmospheric Science correlated to warm tropical SSTs. But warmer SSTs
64: 288–-2898. by themselves do not always portend more TCs.
Francis, J.A. and Vavrus, S.J. 2012. Evidence linking Some general circulation model studies have
Arctic amplification to extreme weather in mid-latitudes. projected there will be as much as a 200 percent
Geophysical Research Letters 39: 10.1029/ increase in TC occurrence in a warmer world, while
2012GL051000. other GCM studies suggest such a world will bring
fewer TCs.
Hakkinen, S., Rhines, P.B., and Worthen, D.L. 2011. Villarini et al. (2011) attempted to explain why
Atmospheric blocking and Atlantic Multidecadal Ocean
there is a large spread in the projection of TC
Variability. Science 334: 655–659.
occurrences that may occur in the future. They used a
Jaiser, R., Dethloff, K., Handorf, D., Rinke, A., and Cohen, statistical model trained using a 131-year record of
J. 2012. Impact of sea ice cover changes on the Northern North Atlantic TC occurrences, a Poisson regression
Hemisphere atmospheric winter circulation. Tellus A 64: type model that generated land-falling or Atlantic TC
11595, DOI: 10.3402/tellusa.v64i0.11595. frequencies as a function of Atlantic region and/or
Kreienkamp, F., Spekat, A., and Enke, W. 2010. tropical mean SSTs. The SSTs included variability
Stationarity of atmospheric waves and blocking over related to the North Atlantic Oscillation (interdecadal)
Europe—based on a reanalysis dataset and two climate and the Southern Oscillation Index (interannual). The
scenarios. Theory of Applied Climatology 102: 205–212. authors compared the statistical model results with
those produced by dynamic GCMs and statistical-
Liu, J., Curry, J.A., Wang, H., Song, M., and Horton, R.M.
dynamic models for the Intergovernmental Panel on
2012. Impact of declining Arctic sea ice on winter
snowfall. Proceedings of the National Academy of Sciences Climate Change (IPCC). In order for the comparisons
109: 6781–6783. to be more faithful, the authors used the SST time
series in the statistical model that each particular
Lupo, A.R., Oglesby, R.J., and Mokhov, I.I. 1997. IPCC model scenario used.
Climatological features of blocking anticyclones: A study When both the Atlantic region and tropical mean
of Northern Hemisphere CCM1 model blocking events in SSTs were used, there was good agreement between
present-day and double CO2 atmospheres. Climate
the results of the statistical model and that of the
Dynamics 13: 181–195.
dynamic models (Figure 1.4.8.1). Tropical SSTs
Mokhov, I.I., Akperov, M.G., Prokofyeva, M.A., elsewhere can affect Atlantic TC occurrence via
Timazhev, A.V., Lupo, A.R., and Le Treut, H. 2013. variations in the jet stream. The authors note, “the
Blockings in the Northern Hemisphere and Euro-Atlantic agreement between the statistical and dynamical
Region: Estimates of changes from reanalysis data and models is rather remarkable, considering the
model simulations. Doklady Earth Sciences 449: 430–433. simplicity of the statistical model.” They also state “it
Scaife, A.A., Wollings, T., Knight, J., Martin G., and appears that the differences among the published
Hinton, T. 2010. Atmospheric blocking and mean biases in results can be largely reduced to differences in the
climate models. Journal of Climate 23: 6143–6152. climate model projections of tropical Atlantic SST
changes relative to the global tropics used by the
Screen, J. and Simmonds, I. 2013. Exploring links between
studies.”
Arctic amplification and midlatitude weather. Geophysical
Research Letters 40: 10.1002/GRL.50174. When Villarini et al. used Atlantic SST alone,
there was little agreement between the statistical and
Tibaldi, S. and Molteni, F. 1990. On the operational dynamic models, with the statistical models showing
predictability of blocking. Tellus 42A: 343–365. large biases in TC frequency. The dynamic models
did not show this bias, supporting the notion that
1.4.8 Tropical Cyclones SSTs alone cannot explain variations in TC
Tropical cyclones (TCs) form and are maintained by occurrence. There were also mixed signals produced
the cooperative forcing between the atmosphere and by the IPCC models with regard to landfalling TCs,
oceans. Warm tropical waters (sea surface half showing a statistically significant increase and
temperatures or SSTs) are the energy source needed the other half showing a significant decrease. The
to generate and maintain them, but favorable authors suggest that in order to reduce the uncertainty
in future TC occurrence, the uncertainty in tropical
138
Figure 1.4.8.1. A comparison of the fractional TC count changes between the statistical model and IPCC dynamic models.
The top (bottom) figures shows the use of Atlantic region and tropical (Atlantic region only) SSTs. The left (right) side
panels were formulated using the NOAA ERSSTv3b (HadISSTv1) dataset. The gray lines define the 90% prediction
interval for the linear regression model. Adapted from Figure 2 of Villarini et al. (2011).
Atlantic and tropical mean SSTs and their difference dominated by deep convection, which is on a scale
should be reduced. below the resolution of the grid. Another contributing
With respect to claims that increases in TC factor is that the models’ resolution is insufficient to
frequency and strength will occur as a result of define the relatively small scale circulation, especially
anthropogenic global warming, and coastal areas will the eye wall where the circulation is most intense. In
be more vulnerable to destruction, the authors state, addition, the paucity of observations means the
“the results do not support the notion of large factors triggering tropical cyclone development are
(~200%) increases in tropical storm frequency in the poorly known. Early atmospheric general circulation
North Atlantic Basin over the twenty-first century in models (AGCMs) had horizontal resolutions the
response to increasing greenhouse gasses (GHGs).” equivalent of approximately 450 km. In such a model,
Instead they found it is more likely TC frequency may resolving individual storm events was impossible, and
change by +/- 40% by the late twenty-first century these could only be inferred from the aggregate
and this “is consistent with both the observed record statistical properties of quantities such as momentum
and with the range of projections of SST patterns.” or temperature flux.
Models also have had difficulty replicating the Manganello et al. (2012) employed the European
dynamic structure of tropical cyclones. This is Centre for Medium-Range Weather Forecasts
partially due to the vertical exchanges being (ECMWF) IFS model to examine the climatology of
139
TCs by identifying individual events. This model is

the recent version of the ECMWF GCM and includes
the latest physical packages for cloud formation (i.e.,
prognostic equations for cloud water), land surface
processes and hydrology, and improved convective
adjustment schemes. The latter provided for a better
representation of tropical weather and year-to-year
variations in tropical climate.
The model was run for “Project Athena” (Jung et
al., 2012) with resolutions at 125 km, 39 km, 16 km
(the current resolution of a weather forecast model),
and 10 km; for all there were 91 levels in the vertical.
The model also was run for each year from 1960 to
2008 for the larger resolutions, and for 1989–2008 for
the 10 km run. To keep the data storage manageable,
the authors used data for May through November in
the Northern Hemisphere (NH) for 1990–2008. The
authors also controlled the data to make sure storms
identified were indeed TCs, such as requiring a warm
core, meaning the center is warmer than the
surrounding environment.
Manganello et al. found over the entire NH, the
39 km run produced the most realistic count for TCs;
the higher resolution model produced too many TCs.
Within each basin (e.g., Atlantic) or sub-basin
Figure 1.4.8.2. Life cycle composite of the (a) maximum 10-
(Northwest Pacific), however, the higher resolutions
m wind speed and (b) minimum SLP for the 25 most intense
occasionally produced the best results. One problem typhoons, in terms of the maximum 10-m wind speed, over
for models is tropical cyclogenesis tends to be too the northwest Pacific for observed data (black), the 10-km
weak, and it was a problem here in spite of better (purple), the 16-km (red), and 39-km (green) runs during
resolution. In a related matter, the model produced MJJASON of 1990–2008. The time step is in 6-h increments.
TCs that were weaker than observed in terms of wind Adapted from Figure 9, Manganello et al. (2012).
speed but comparable in terms of central pressure
(Figure 1.4.8.2). Finally, the higher model resolution
runs did better in capturing ENSO variability of TCs. their wakes for periods of several days to weeks.
ENSO is well-known to have a large influence on the Manucharyan et al. (2011) write, “TC-induced ocean
interannual variation of TC occurrence and intensity. mixing can have global climate impacts as well,
Manganello et al. note, “as computing power including changes in poleward heat transport, ocean
continues to grow, it is becoming increasingly circulation, and thermal structure.” They note,
possible to model the global climate at horizontal however, that “in several previous modeling studies
resolutions presently used in short-term weather devoted to this problem, the TC mixing was treated as
prediction.” Although hindcasts with improved a permanent (constant in time) source of additional
resolution and physics still have difficulty vertical diffusion in the upper ocean.” They thus
representing observations to a high degree, the explore the “highly intermittent character of the
models are able to reasonably reproduce the number mixing” and what it portends for global climate, using
of events, and even produced structures that look like a global ocean-atmosphere coupled model and a
observed TCs (Figure 1.4.8.3), thus, passing the test. simple heat transfer model of the upper ocean.
These kinds of advances represent progress and The three Yale University researchers mimicked
should make future computer scenarios for climate the effects of TCs using several representative cases
more reasonable and useful. of time-dependent mixing that yield the same annual
The studies of Dare and McBride (2011) and Park mean values of vertical diffusivity, conforming with
et al. (2011) have demonstrated tropical cyclones the studies of Jansen and Ferrari (2009) and Fedorov
(TCs) can significantly cool the surface waters in et al. (2010), wherein spatially uniform (but varying
140
Figure 1.4.8.3. (Left panel) 850-hPa tangential velocity (m s−1), (middle) 700-hPa omega (Pa s−1), and (right) TCLIW (kg
m−2) for 30 and 36 hours after onset as shown on the left for an intense TC (10-km resolution). Radius is 3.5° from the
storm center. Blue shading on the left roughly delineates the regions where the local wind maxima occur. Adapted from
Figure 12 of Manganello et al. (2012).
in time) mixing is imposed on zonal bands in the epoch. Nature 463: 1066–1070.
upper ocean.
Gualdi, S., Scoccimarro, E., and Navarra, A. 2008.
Manucharyan et al. observed “a weak surface Changes in tropical cyclone activity due to global warming:
cooling at the location of the mixing (~0.3°C), a Results from a high-resolution coupled general circulation
strong warming of the equatorial cold tongue (~2°C), model. Journal of Climate 21: 5204–5228.
and a moderate warming in middle to high latitudes
(0.5°C-1°C),” together with “a deepening of the Jansen, M. and Ferrari, R. 2009. Impact of the latitudinal
tropical thermocline with subsurface temperature distribution of tropical cyclones on ocean heat transport.
Geophysical Research Letters 36: 10.1029/2008GL036796.
anomalies extending to 500 m [depth].” They say
“additional mixing leads to an enhanced oceanic heat Jung, T., Miller, M.J., Palmer, T.N., Towers, P., Wedi, N.,
transport from the regions of increased mixing toward Achuthavarier, D., Adams, J.M., Altshuler, E.L., Cash,
high latitudes and the equatorial region.” B.A.,Kinter III, J.L., Marx, L., Stan, C., and Hodges, K.I.
“Ultimately,” the researchers state, “simulations 2012. High-resolution global climate simulations with the
with TC-resolving climate models will be necessary ECMWF model in Project Athena: Experimental design,
to fully understand the role of tropical cyclones in model climate, and seasonal forecast skill. Journal of
Climate 25: 3155–3172.
climate,” for “the current generation of GCMs are
only slowly approaching this limit and are still unable Manganello, J.V., Hodges, K.I., Kinter III, J.L., Cash,
to reproduce many characteristics of the observed B.A., Marx, L., Jung, T., Achuthavarier, D., Adams, J.M.,
hurricanes, especially of the strongest storms critical Altshuller, E.L., Huang, B., Jin, E.K., Stan, C., Towers, P.,
for the ocean mixing (e.g., Gualdi et al., 2008; and Wedi, N. 2012. Tropical cyclone climatology in a 10-
Scoccimarro et al., 2011).” km global atmospheric GCM: Toward weather-resolving
climate modeling. Journal of Climate 25: 3867–3892.
References Manucharyan, G.E., Brierley, C.M., and Fedorov, A.V.
2011. Climate impacts of intermittent upper ocean mixing
Dare, R.A. and McBride, J.L. 2011. Sea surface induced by tropical cyclones. Journal of Geophysical
temperature response to tropical cyclones. Monthly Research 116: 10.1029/2011JC007295.
Weather Review 139: 3798–3808.
Park, J.J., Kwon, Y.-O., and Price, J.F. 2011. Argo array
Fedorov, A., Brierley, C., and Emanuel, K. 2010. Tropical observation of ocean heat content changes induced by
cyclones and permanent El Niño in the early Pliocene tropical cyclones in the north Pacific. Journal of
141
Geophysical Research 116: 10.1029/2011JC007165. climate change, can induce changes in the storm
tracks. As noted by Wilson et al. (2010), “as the
Soccimarro, E., Gualdi, S., Bellucci, A., Sanna, A., Fogli,
P.G., Manzini, E., Vichi, M., Oddo, P., and Navarra, A. climate changes, the ocean dynamics will change, but
2011. Effects of tropical cyclones on ocean heat transport the land and mountains remain unchanged over
in a high resolution coupled general circulation model. millennia. Therefore in order to understand how the
Journal of Climate 24: 4368–4384. shape of the storm tracks will evolve, it is crucial to
understand the relative impact of ocean dynamics and
Villarini, G., Vecci, G.A., Knutson, T.R., Zhao, M., and orography, as well as their interactions.”
Smith, J.A. 2011. North Atlantic tropical storm frequency
Four model worlds were constructed by Wilson et
response to anthropogenic forcing: Projections and sources
of uncertainty. Journal of Climate 24: 3224–3238. al. using a coupled ocean-atmosphere general
circulation model. A model control run was
performed with realistic topography and a fully
1.4.9 Storm Tracks and Jet Streams dynamic ocean. Then, a typical modeling strategy was
The climate of Earth’s atmosphere is largely driven used to produce three other runs. This strategy
by forcing from the underlying surface, of which 71 involved using extreme and unrealistic
percent is covered by the oceans. Thus the oceans and parameterizations in order to isolate the impact of
mountains are important in simulating storm tracks, certain factors. A scenario was developed in which all
and any simulations of present or future climate must land topography was leveled to 1 meter in height and
be able to represent the land surface topography and the ocean was a 30-meter “slab” that does not move; a
the world oceans in a realistic way if the models are second scenario added mountains; and a third
to be useful. scenario was created without mountains but
Storm tracks (Figure 1.4.9.1) are the containing a fully dynamic ocean.
climatological “signature” representing the frequency From the results of the model runs, Wilson et al.
and strength of the regular passage of cyclonic showed storm tracks are inherent in our atmosphere,
disturbances. These disturbances are generally as these features were found in the model run even
responsible for bringing weather that can be a minor without mountains and topography. In the model runs
nuisance on a typical day, such as rain or snow that included a dynamic ocean, the effect was to push
showers, or as severe weather that can cause loss of the storm tracks poleward. The impact of the
life and property. These low pressure systems are also mountains was to reduce storm activity on their
a significant mechanism for the poleward transport of leeward side. The two combined features resulted in
heat, momentum, and water vapor out of the tropics. distinct Atlantic and Pacific storm tracks rather than
These actions are necessary for maintaining the just one long track in the Northern Hemisphere.
structure of Earth’s current general circulation and Changes in climate may also alter the storm
climate. tracks through changes in the frequency and intensity
Natural variations, as well as global or regional of the cyclones that comprise them. This would have
Figure 1.4.9.1. The mean winter Northern Hemisphere storm tracks from the European Center’s observational reanalyses
(colors), and the atmospheric general circulation model used in the Wilson et al. (2010) study (contours). The storm tracks
were derived from the cyclone scale filtered upper air (250 hPa) kinetic energy. Adapted from Wilson et al. (2010).
142
an impact on the frequency and intensity of related order to compare the impacts of temperature gradient
phenomena such as blocking anticyclones. Blocking and atmospheric stability in changing the strength of
anticyclones are implicated in contributing to the the general circulation.
European and Russian heat waves of 2003 and 2010, The authors tested the proposition that AGW will
respectively. Such events also are implicated in warm the poles faster than the tropics near Earth’s
severe winter cold across all continents. Thus, in surface (SFC, first trial run). This argues for a
order to discuss possible future scenarios for the weakening of the general circulation as horizontal
climate or even the occurrence and severity of temperature gradients decrease. However, warming
extreme events, any models used to generate these the SFC polar region decreased the stability of the
features must include interactive ocean dynamics. model atmosphere. Others have argued AGW will
And as Wilson et al. pointed out, “increased warm the upper troposphere relative to the surface,
resolution is needed in coupled climate models to especially in the tropics (UP, second trial run). This
adequately address the role of ocean dynamics.” increases tropical atmospheric stability but increases
In examining the strength of storms, the upper air temperature gradients. HDvS12 then
anthropogenic global warming (AGW) supporters added the two effects together in a third trial
have argued for decades that an increase in CO2 will (UP+SFC).
bring warmer temperatures and stronger and more HDvS12 found the general circulation “weakens
destructive storm, including more hurricanes, by almost 10% in the UP experiment whereas it
tornadoes, and mid-latitude cyclones. Recently, they strengthens by almost 4% in the SFC experiment. In
have argued the warmer temperatures will strengthen the FULL experiment, it weakens by about 5%.” The
the water cycle, making these storms even stronger. FULL and the UP+SFC experiments were similar in
Models have not been helpful in settling this outcome. They also note “the expected effects due to
argument, as some models imply storminess will mean static [atmospheric] stability and meridional
decrease while others imply the opposite. In addition, temperature gradient change are opposite of each
empirical observations, which will be discussed in other.” The results demonstrate the stability impact
detail in Chapter 7 of this volume, do not support seems to be dominant (Figure 1.4.9.2).
such projections. Some scientists have argued storminess should
The strength of the general circulation is decrease under AGW scenarios due to the weakening
controlled by two main factors. One is the equator-to- of the equator-to-pole temperature differences. That is
pole temperature difference (gradient) and the other is generally correct but overlooks the impact of
atmospheric stability. These ultimately drive the atmospheric stability. Models have not sorted out the
general circulation and give rise to the jet streams. In storminess issue because different models employ
the mid-latitudes, where the gradient is strong, storms different physics for processes that affect the
arise and are driven by these gradients. The ultimate temperature gradient. Models also handle the heating
role of storms is to decrease the equator-to-pole effect of CO2 differently.
gradient. The impact of the horizontal temperature These differences have resulted in surface
gradient is well-known through knowledge of temperature gradients and upper tropospheric
baroclinic instability. A stronger gradient will warming of various strengths. But as noted in Section
generally lead to more storminess. 1.4.6.2, there is no strong observational evidence that
Hernandez-Deckers and von Storch (2012) the upper troposphere is warming as AGW scenarios
(hereafter HDvS12) show the second factor, suggest. HDvS12 demonstrate the utility of models in
atmospheric stability, is often overlooked in the breaking down complicated climate problems. More
climate change debate. Stability is simply the importantly, however, model construction clearly can
atmosphere’s resistance to overturning where warm make a significant difference as to how the outcome
air rises and cold air sinks. For example, to make the is interpreted.
atmosphere less (more) stable, one could warm (cool) Lang and Waugh (2011) note “understanding the
the surface relative to the upper air. characteristics and trends in summer cyclones is
HDvS12 used the European Centre for Medium important not only for understanding mid-latitude
Range Forecasting general circulation model weather systems and extreme events, but it is also
(ECHAM5) coupled with an ocean model from the important for understanding the Arctic hydrological
Max Planck Institute. They calculated the generation, cycle and radiation budget (e.g., Orsolini and
flow, and dissipation of energy in the atmosphere in Sortberg, 2009).” In addition, they note “the surface
143
large spread in the magnitude of the trend from

individual models, and, hence, a large uncertainty in
the trends from a single model.” They conclude, “the
general lack of consistency among models indicates
that care is required when interpreting projected
changes in summer weather systems.”
In another study, Zhang et al. (2012) examined
the behavior of a jet stream in a channel model of
similar length to Earth’s diameter and about 10,000
km in the “north-south” (meridional) direction, with
17 vertical levels centered on a jet maximum. The
simplified model used equations representing the
basic conservation laws (mass, momentum, and
energy) and allowed the “rotation speed” to vary
linearly (in reality, it gets stronger from equator to
pole). This model also allowed the variation of
atmospheric stability in the meridional direction as
well as a meridional temperature difference of 43° C,
comparable to terrestrial equator-to-pole temperature
differences. The model contained surface friction,
varied in order to examine the behavior of the jet.
When the surface friction was increased, the
Figure 1.4.9.2. Diagrams of atmospheric energy
production, reservoirs, conversions, and dissipation. The researchers found discernible differences in the
numbers given are differences from the control run of the zonally and time-averaged wind (Figure 1.4.9.3) and
model (today’s climate). The FULL, UP, SFC, and temperature gradient (Figure 1.4.9.4) profiles. The
UP+SFC, represent an experiment with double CO2, differences were more discernible in the temperature
warmer upper troposphere only, warmer polar surface fields, and as the friction was increased, the
temperatures only, and the combined effects, respectively. prominent wave number in the model increased from
Units are W m−2. Arrows indicate the direction of energy four to six. Values smaller (greater) than wave
flow. Each square contains a simple two box and a more numbers four (six) are associated with the larger
complex four box energy model. Adapted from Figure 5 in (smaller) scale that is considered the planetary
HDvS12.
(synoptic) scale. Annular mode behavior could be
found only in the larger-scale “climate” for the model
concentrations of ozone and aerosols, and as a result at weak frictional values.
surface air quality, depend on a range of Zhang et al. thus demonstrated the importance of
meteorological factors [that] are closely connected larger waves in maintaining large-scale jet stream
with cyclones (e.g., Jacobs and Winner, 2009).” baroclinicity (density differences) and a baroclinic
Lang and Waugh examined “the robustness of mechanism for the cooperation between synoptic and
trends in Northern Hemisphere (NH) summer large-scale eddies in maintaining the persistence of
cyclones in the World Climate Research annular mode behavior. The authors note, “as an
Programme’s Coupled Model Intercomparison internal mode of variability, understanding the
Project phase 3 (CMIP3) multi-model data set that mechanism that sustains the zonal wind anomalies is
was used in the Fourth Assessment (AR4) of the useful not only to predict the intra-seasonal variability
Intergovernmental Panel on Climate Change (IPCC, in the extra-tropics but also for climate change
2007).” The two Johns Hopkins University projections.”
researchers report they could find “little consistency” These internal variations in the jet stream can
among the 16 models they studied. They write, “there represent the maintenance of atmospheric blocking,
is no consistency among the models as to whether the and both are important to account for in seasonal
frequency of hemispheric-averaged summer cyclones range forecasting. Also, as the climate changes there
will increase or decrease.” For some sub-regions, the may be a change in jet stream behavior, but the
sign of the trend was consistent among most of the internal variations will remain. Conversely, if more
models, but even then, as they report, “there is still a complicated models cannot replicate annular mode
144
behavior noted in the observations and the simpler Also exploring the subject were Chang et al.
model here in future climate change scenarios, the (2013), who write that “midlatitude storm tracks are
model projections will be of limited value. marked by regions frequented by baroclinic waves
and their associated surface cyclones,” which bring
with them “strong winds and heavy precipitation,
seriously affecting regional weather and climate.”
They also note that such storms “transport large
amounts of heat, momentum and moisture poleward,”
making up “an important part of the global
circulation.” And they state that how these storm
tracks may change as a result of global warming “is
thus of huge societal interest.”
Chang et al. used “storm-track activity derived
from ERA-Interim [European Centre for Medium-
Range Weather Forecasts (ECMWF) Re-Analysis
(ERA-40; Uppala et al. 2005)] data as the current best
estimate to assess how well models that participated
in phase 3 of the Coupled Model Intercomparison
Project (CMIP3; Meehl et al. 2007) that were
considered in the Intergovernmental Panel on Climate
Figure 1.4.9.3. The latitudinal distribution of the zonal and Change Fourth Assessment Report (Solomon et al.
time-averaged zonal winds at (a) 437.5 and (b) 875 hPa for 2007) do in simulating storm-track activity.”
a range of surface friction values. The latitude distance on The four researchers report “only 2 of the 17
the abscissa is in kilometers (0 km the channel center). models have both the Northern Hemisphere [NH] and
Positive (negative) values are the distance poleward Southern Hemisphere [SH] storm-track activity
(equatorward). Adapted from Figure 3 in Zhang et al. within 10% of that based on ERA-Interim” and “four
(2012). models simulate storm tracks that are either both
significantly (>20%) too strong or too weak.” They
also note “the SH to NH ratio of storm-track activity
... is biased in some model simulations due to biases
in midtropospheric temperature gradients” and “storm
tracks in most CMIP3 models exhibit an equatorward
bias in both hemispheres.” Further, “some models
exhibit biases in the amplitude of the seasonal cycle”;
“models having a strong (weak) bias in storm-track
activity also have a strong (weak) bias in poleward
eddy momentum and heat fluxes, suggesting that
wave-mean flow interactions may not be accurately
simulated by these models”; and “preliminary
analyses of Fifth Assessment Report (AR5)/CMIP5
model data suggest that CMIP5 model simulations
also exhibit somewhat similar behaviors.”
References
Chang, E.K.M., Guo, Y., Xia, X., and Zheng, M. 2013.

Storm-track activity in IPCC AR4/CMIP3 model
simulations. Journal of Climate 26: 246–260.
Figure 1.4.9.4. As in Figure 1.4.9.3, except for temperature Hernandez-Deckers, D. and von Storch, J.S. 2012. Impact
gradients, and adapted from Figure 4 in Zhang et al. of the warming pattern on global energetics. Journal of
(2012). Climate 25: 5223–5240.
145
Intergovernmental Panel on Climate Change. 2007. on models that consider both human and natural
Climate Change 2007: The Physical Science Basis. factors provide “credible quantitative estimates of
Contribution of Working Group 1 to the Fourth Assessment future climate change.” However, they continue,
Report of the Intergovernmental Panel on Climate Change. mismatches between IPCC AR4 model ensembles
Cambridge University Press, New York, New York, USA.
and observations, especially the multidecadal
Jacob, D.J. and Winner, D.A. 2009. Effect of climate variability (MDV), “have cast shadows on the
change on air quality. Atmospheric Environment 43: 51–63. confidence of the model-based decadal projections of
future climate,” as also has been noted by Meehl et al.
Lang, C. and Waugh, D.W. 2011. Impact of climate change
(2009), who indicate considerably more work needs
on the frequency of Northern Hemisphere summer
cyclones. Journal of Geophysical Research 116: 10.1029/ to be done in this important area.
2010JD014300. In an exercise designed to illustrate the extent of
this model failure, Fu et al. evaluated “many
Meehl, G.A., Covey, C., Delworth, T., Latif, M., individual runs of AR4 models in the simulation of
McAvaney, B., Mitchell, J.F.B., Stouffer, R.J., and Taylor, past global mean temperature,” focusing on the
K.E. 2007. The WCRP CMIP3 multi-model dataset: A new performance of individual runs of models included in
era in climate change research. Bulletin of the American
the Coupled Model Intercomparison Project phase 3
Meteorological Society 88: 1383–1394.
(CMIP3) in simulating the multidecadal variability of
Orsolini, Y.J. and Sorteberg, A. 2009. Projected changes in the past global mean temperature.
Eurasian and Arctic summer cyclones under global The three researchers determined “most of the
warming in the Bergen Climate Model. Atmospheric and individual model runs fail to reproduce the MDV of
Oceanic Science Letters 2: 62–67. past climate, which may have led to the
Solomon, S., Qin, D., Manning, M., Marquis, M., Averyt, overestimation of the projection of global warming
K., Tignor, M.B., Miller Jr., H.L., and Chen, Z. (Eds.). for the next 40 years or so.” More specifically, they
2007. Climate Change 2007: The Physical Science Basis. note simply taking into account the impact of the
Cambridge University Press, Cambridge, United Kingdom. Atlantic Multi-decadal Oscillation shows “the global
average temperature could level off during the 2020s–
Uppala, S.M., Kallberg, P.W., Simmons, A.J., Andrae, U.,
2040s,” such that the true temperature change
Da Costa Bechtold, V., Fiorino, M., Gibson, J.K, Haseler,
J., Hernandez, A., Kelly, G.A., Li, X., Onogi, K., Saarinen, between 2011 and 2050 “could be much smaller than
S., Sokka, N., Allan, R.P., Andersson, E., Arpe, K., the AR4 projection.”
Balmaseda, M.A., Beljaars, A.C.M., Van De Berg, L., Jeong et al. (2011) state “the Siberian High (SH)
Bidlot, J., Bormann, N., Caires, S., Chevallier, F., Dethof, is the most conspicuous pressure system found in the
A., Dragosavac, M., Fisher, M., Fuentes, M., Hagemann, Northern Hemisphere during wintertime,” when
S., Holm, E., Hoskins, B.J., Isaksen, L., Janssen, P.A.E.M., “strong radiative cooling over the snow covered
Jenne, R., McNally, A.P., Mahfouf, J.-F., Morcrette, J.-J., Eurasian continent forms a cold-core high-pressure
Rayner, N.A., Saunders, R.W., Simon, P., Sterl, A., system in the lower troposphere over northern
Trenberth, K.E., Untch, A., Vasiljevic, D., Viterbo, P., and Mongolia” that exerts “tremendous influences on
Woollen, J. 2005. The ERA-40 reanalysis. Quarterly
weather and climate in Northern Eurasia, East Asia,
Journal of the Royal Meteorological Society 131: 2961–
3012. and even further into South Asia (e.g., Cohen et al.,
2001; Panagiotopoulos et al., 2005; Wang, 2006).”
Wilson, C., Sinha, B., and Williams, R.G. 2010. The The authors further state SH intensity variations—as
shaping of storm tracks by mountains and ocean dynamics. simulated by 22 global climate models under 20C3M
Weather 65: 320–323. and A1B scenarios in the CMIP3—show “a steady
Zhang, Y., Yang, X.Q., Nie, Y., and Chen, G. 2012. decreasing trend in the SH intensity from the late 20th
Annular mode-like variation in a multilayer century throughout the 21st century, leading to a
quasigeostrophic model. Journal of the Atmospheric decrease of about 22% in SH intensity at the end of
Sciences 69: 2940–2958. the 21st century compared to the 1958-1980 average.”
In a study designed to determine to what degree
the temporal SH intensity simulations of these models
1.4.10 Miscellaneous mimic reality, Jeong et al. employed two
Fu et al. (2011) note the Fourth Assessment Report observational gridded sea level pressure (SLP) data
(AR4) of the Intergovernmental Panel on Climate sets, that of the Hadley Centre and the National
Change (IPCC) concluded climate projections based Centre for Atmospheric Research, plus two reanalysis
146
data sets (NCEP and ERA40) and in situ SLP References

observations from 20 stations located in the central
SH region to create a history of SH intensity over the Cohen, J., Saito, K., and Entekhabi, D. 2001. The role of
past several decades. the Siberian High in Northern Hemisphere climate
The climatic reconstructive work of the seven variability. Geophysical Research Letters 28: 299–302.
scientists revealed “a pronounced declining trend of Fu, C.-B., Qian, C., and Wu, Z.-H. 2011. Projection of
the SH intensity from the late 1960s to the early global mean surface air temperature changes in next 40
1990s,” which would appear to mesh well with GCM years: Uncertainties of climate models and an alternative
simulations presented in the IPCC AR4 that indicate a approach. Science China Earth Sciences 54: 1400–1406.
“steady weakening of the SH intensity for the entire
Jeong, J.-H., Ou, T., Linderholm, H.W., Kim, B.-M., Kim,
21st century.” The authors report, however, that in the
S.-J., Kug, J.-S., and Chen, D. 2011. Recent recovery of the
real world the declining SH intensity trend “was Siberian High intensity. Journal of Geophysical Research
sharply replaced by a fast recovery over the last two 116: 10.1029/2011JD015904.
decades.” The thus note “this feature has not been
successfully captured by the GCM simulations used Meehl, G.A., Goddard, L., Murphy, J., Stouffer, R.J., Boer,
for the IPCC AR4,” all of which predict “a steady G., Danabasoglu, G., Dixon, K., Giorgetta, M.A., Greene,
decreasing trend in the SH intensity from the late 20th A.M., Hawkins, E., Hegerl, G., Karoly, D., Keenlyside, N.,
Kimoto, M., Kirtman, B., Navarra, A., Pulwarty, R.S.,
century throughout the 21st century.”
Smith, D., Stammer, D., and Stockdale, T. 2009. Decadal
Jeong et al. conclude, “an improvement in prediction: Can it be skillful? Bulletin of the American
predicting the future climate change in regional scale Meteorological Society 90: 1467–1485.
is desirable.”
Panagiotopoulos, F., Shahgedanova, M., Hannachi, A., and
Stephenson, D.B. 2005. Observed trends and
teleconnections of the Siberian High: A recently declining
center of action. Journal of Climate 18: 1411–1422.
Wang, B. 2006. The Asian Monsoon. Springer, Berlin,
Germany.
147

09-09-13 - Chapter 1 Models

Uploaded by

Copyright:

Available Formats

09-09-13 - Chapter 1 Models

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

09-09-13 - Chapter 1 Models

Uploaded by

Copyright:

Available Formats

1

Global Climate Models and Their Limitations

1. Global Climate Models and Their Limitations

Key Findings Confidence in a model is further based on the

• To have any validity in terms of future Introduction

must therefore be approximated, or parameterized at to the models.

professor at the University of Pennsylvania, who Scientists working in fields characterized by

alarmist and incompetent.” Green and Armstrong subsequently collaborated

Armstrong, J.S. 2006. Findings from evidence-based Washington, DC.

atmospheric processes affecting the climate of Europe

regions of the atmosphere, allowing emission to space temperature.

terrestre et des espaces planétaires. Mémoires de

3.2°C, and a 90% range of from 1.4–2.8°C. What is

Lindzen, R.S. and Choi, Y.-S. 2011. On the observational

Kim, H.-M., Webster, P.J., and Curry, J.A. 2012. Seasonal

1.2 Modeling Techniques

that of a weather forecast model.

1.2.3 Statistical Models

1.2.3.1 Statistical Validation

empirical model can be used to project climate into

At the time, meteorology was interested in

1.2.4 Low Order Models

Svensson and Karlsson found the winter sea ice

relative humidity (i.e., an increasing specific humidity References

seemingly “random” forcing processes. The likely

scenario in which clouds, precipitation, and regional June 1998.

forecasts. examining the formulation of the shallow convection

Writing in the Journal of Geophysical Research

extreme rain events in satellite rain-rate estimates and an

reanalyses and CGCMs [coupled global climate

satisfactorily. impossible to present reliable future projections from

Casado, M. and Pastor, M. 2012. Use of variability modes

“Because the model simulations have difficulties

cooling of 0.715°C. al. 1996) and the ORCHIDEE biosphere model

References University of Kiel—first describe various

mechanisms of decadal-to-multidecadal variability is time scales (e.g., daily, month-to-month, or

convection. This is but one factor that causes GCMs

When present-day CO2 concentrations were projections.

TCs by identifying individual events. This model is

large spread in the magnitude of the trend from

Chang, E.K.M., Guo, Y., Xia, X., and Zheng, M. 2013.

data sets (NCEP and ERA40) and in situ SLP References

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.