Magnitude Limits of Subduction Zone Earthquakes

Bulletin of the Seismological Society of America, Vol. 104, No. 5, pp. 2359–2377, October 2014, doi: 10.
1785/0120130287
Magnitude Limits of Subduction Zone Earthquakes

by Yufang Rong, David D. Jackson, Harold Magistrale, and Chris Goldfinger
Abstract Maximum earthquake magnitude (mx ) is a critical parameter in seismic

hazard and risk analysis. However, some recent large earthquakes have shown that
most of the existing methods for estimating mx are inadequate. Moreover, mx itself
is ill-defined because its meaning largely depends on the context, and it usually cannot
be inferred using existing data without associating it with a time interval. In this study,
we use probable maximum earthquake magnitude within a time period of interest,
mp T, to replace mx . The term mp T contains not only the information of magnitude
limit but also the occurrence rate of the extreme events. We estimate mp T for circum-
Pacific subduction zones using tapered Gutenberg–Richter (TGR) distributions. The
estimation of the two TGR parameters, β-value and corner magnitude (mc ), is per-
formed using the maximum-likelihood method with the constraint from tectonic mo-
ment rate. To populate the TGR, the rates of smaller earthquakes are needed. We apply
the Whole Earth model, a high-resolution global estimate of the rate of m ≥ 5 earth-
quakes, to estimate these rates. The uncertainties of mp T are calculated using
Monte-Carlo simulation. Our results show that most of the circum-Pacific subduction
zones can generate m ≥ 8:5 earthquakes over a 250-year interval, m ≥ 8:8 earthquakes
over a 500-year interval, and m ≥ 9:0 earthquakes over a 10,000-year interval. For the
Cascadia subduction zone, we include the 10,000-year paleoseismic record based on
turbidite studies to supplement the limited instrumental earthquake data. Our results
show that over a 500-year period, m ≥ 8:8 earthquakes are expected in this zone; over
a 1000-year period, m ≥ 9:0 earthquakes are expected; and over a 10,000-year period,
m ≥ 9:3 earthquakes are expected.
Introduction
Almost all probabilistic seismic-hazard analyses use and from the available earthquake history, but magnitude
earthquake magnitude–frequency distributions to character- 9 events were unexpected. At least four factors contributed
ize earthquakes in a region. In the widely used truncated to the underestimation of mx : (1) no such large earthquakes
Gutenberg–Richter (GR) distribution, maximum earthquake had been recorded in the regions; (2) an ad hoc fault segmen-
magnitude, mx , is the hard cutoff magnitude: earthquakes tation model was overweighted in estimating mx ; (3) conver-
larger than mx will not be considered in the seismic-hazard gence is nearly fault parallel along the northern Sumatra
analysis. Unfortunately, due to the short time span of histori- subduction zone, that was thought to limit strain accumulation
cal earthquake catalogs, it is difficult to obtain a reliable es- on the megathrust; and (4) the proposed empirical relationship
timation of mx . between the largest earthquake magnitude, lithosphere age,
Both the devastating 2004 m 9.1 Sumatra–Andaman and and subduction rate gives only magnitude 8 earthquakes
the 2011 m 9.0 Tohoku earthquakes surprised geoscientists. (Ruff and Kanamori, 1980; Kanamori, 2006; Stein and Okal,
Most of the prior hazard estimates of the Tohoku region lim- 2011). These factors are key issues existing in some methods
ited the maximum magnitude to below 8.5 (Ruff and Kana- for estimating mx . Historical earthquake catalogs are too short,
mori, 1980; Earthquake Research Committee, 2005). The and paleoseismic data are far from complete. The common
largest known historical earthquake in the region before practice of adding a quarter- or half-unit of magnitude to
the 2011 earthquake was the A.D. 869 Jogan earthquake, es- the largest recorded magnitude (e.g., Gaull et al., 1990; Mus-
timated to have a magnitude of 8.3–8.4 based on the analysis son, 2000; Vilanova and Fonseca, 2007; Secanell et al., 2008)
of tsunami deposits (Minoura et al., 2001; Sugawara et al., is arbitrary and insufficient to capture mx . Another common
2012). In the Sumatra region, earthquakes of about magni- practice is to use fault length–magnitude scaling relationships.
tude 8 were not unexpected using the empirical relationship When this method is used to estimate mx , one of the assump-
between largest earthquake magnitude, oceanic lithosphere tions is that the two ends of a fault or fault segment are known,
age, and convergence rate by Ruff and Kanamori (1980) and earthquakes will not rupture through the ends and jump to
2359
Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

by gmontalva
2360 Y. Rong, D. D. Jackson, H. Magistrale, and C. Goldfinger
another fault or segment. The ends of a fault are usually de- Estimating Probable Maximum Magnitude
termined from fault traces on a map, subject to errors and
incompleteness. Moreover, multiple segments can be ruptured Magnitude–Frequency Distributions
by a single event. The 2011 m 9.0 Tohoku earthquake ruptured The GR relationship states that, in a given region and for
all the proposed subduction segments in the Tohoku area, a given period of time, the cumulative number of earthquakes
although evidence for the segmentation was weak. The can be represented by
2007 m 8.1 Solomon Islands earthquake ruptured through the
triple junction of the Australia, Pacific, and Woodlark plates, log10 N a − bm; 1
which was thought to be a substantial geologic boundary (Fur-
long et al., 2009). The 2002 m 7.9 Denali fault earthquake in which m is the magnitude, N is the number of events with
started with vertical motion on a previously unknown thrust magnitude equal to or larger than m, and a and b are con-
fault, the Susitna Glacier fault, continued on the Denali fault, stants. This equation describes the relative ratio of large-to-
and jumped onto the Totschunda fault. Stein and Okal (2011) small earthquakes. The GR relationship is one of the most
used additional and more accurate data of lithosphere age and important scaling relationships in seismology.
convergence rate to examine the relationship by Ruff and The seismic moment (M) of an earthquake is a measure
Kanamori (1980) and found that the seemingly compelling of its total displacement, integrated over area, and it is pro-
pattern between magnitude, oceanic lithosphere age, and con- portional to the energy released by the earthquake. There is a
vergence rate disappeared. McCaffrey (2008) and Goldfinger, close relationship between m and M (Hanks and Kanamori,
Ikeda, et al. (2013) both suggest for different reasons that 1979):
larger earthquakes than presently known may well character-
ize many subduction zones, but records are too short and in- M 101:5mC ; 2
complete to have revealed this conjecture.
The parameter mx is used not only in earthquake in which M is in newton meters (N·m). Here we take
magnitude–frequency distributions by seismologists but also C 9:0, although published values range from 9.0 to 9.1.
in engineering design by earthquake engineers. In practice, Thus, reported magnitudes may differ from one publication
different kinds of maximum earthquakes have been employed to another, but the seismic moments should be identical.
such as maximum possible earthquake, maximum credible Seismic moment can also be related to faulting parameters:
earthquake, maximum expectable, or maximum probable for a simple idealized earthquake with uniform slip d over a
earthquake (Reiter, 1990). Thus, the meaning of mx largely fault plane with area S in a uniform medium with rigidity μ,
depends on the context. In fact, without specifying a time the seismic moment is
interval over which it is valid, the mx itself is ill-defined (Pi-
sarenko et al., 2008). With an absolute mx , it is difficult to M μSd: 3
argue why an earthquake with a little bit larger magnitude is
not possible. In addition, mx is usually not inferable from the Seismic moment is a very useful concept because it is addi-
instrumental and historical earthquake data, because the rec- tive: a collection of earthquakes produces a total seismic
ord is too short for great earthquakes. Holschneider et al. moment equal to the sum of the individual moments. Further-
(2011) suggested using the maximum expected magnitude more, over a long time, the total moment on a fault or plate
in a given time interval T to replace the maximum possible boundary can be estimated from the product of its locked
magnitude, because confidence intervals of mx diverge for area and total slip. This moment, which we refer to as the
most of the cases. Zöller et al. (2013) presented a statistical tectonic moment, should be equal to the sum of seismic mo-
model for the estimation of the maximum earthquake mag- ments of the earthquakes accommodating the fault or plate
nitude to occur in a time interval T in the future. In line with motion over a long (generally very long) time. Application of
the above studies, we use the terminology mp T, probable this simple equality is of course complicated by several fac-
maximum magnitude within period of interest T, in this tors, including imperfect and poorly known coupling, poorly
study. The mp T contains not only the information of mag- characterized coupled area, and uncertain value of μ.
nitude limit but also the occurrence rate of the extreme Because each earthquake contributes a magnitude-
events. Instead of solving for the absolute mx , we estimate dependent seismic moment, the distribution of earthquake
mp T for each of the circum-Pacific subduction zones. The sizes implies a distribution of tectonic moment release. Mea-
largest expected earthquake in any time period is subject to sured tectonic moment might be explained by a few very
natural variability; we take mp T to represent the median of large earthquakes, or many more small ones, or more real-
a probability density distribution, which we estimate along istically by a distribution of large and small ones. For this
with its uncertainty. purpose, the GR relationship is inadequate: it implies an ex-
We use several different measures of large earthquake ponentially decreasing rate of large earthquakes, but their
magnitudes, distinguishable in the text only by some tiny seismic moments increase exponentially, with the result that
subscripts, but with their meanings differing significantly. the implied total seismic moment rate would be infinite. To
Table 1 summarizes the measures. overcome this paradox, novel models have been proposed to

by gmontalva
Magnitude Limits of Subduction Zone Earthquakes 2361
address the deviations from the GR relationship at the tail of Table 1

the magnitude–frequency distributions and the issue of infin- A Glossary of Large Magnitude Parameters
ite seismic energy. Unfortunately, all the models suffer from Measure Meaning
large statistical uncertainty due to the insufficient number of
mx Absolute maximum; assumed maximum possible
observed large earthquakes. earthquake size for a given region.
A simple modification to the GR relationship is to trun- mh Historic maximum; magnitude of largest earthquake
cate it, by setting either the density (earthquakes per magni- observed or inferred to have occurred in the region.
tude unit) or cumulative rate abruptly to zero. The truncated mc Corner magnitude; a parameter in the tapered Gutenberg–
GR has been widely used in probabilistic seismic-hazard Richter distribution, indicating the magnitude at which
the rate is 1=e times that implied by a simple
analyses such as the national seismic-hazard maps of the Gutenberg–Richter relationship.
United States (Frankel et al., 1996, 2002; Petersen et al., 2008). mp T Probable maximum magnitude expected in time interval T.
Another modification of the GR relationship is the ta-
pered Gutenberg–Richter (TGR) distribution (Kagan, 1997;
Kagan and Schoenberg, 2001; Bird and Kagan, 2004). The
3. estimate the threshold rate, that is, the rate of magnitude
TGR relation is best expressed in seismic moment, M, instead
mt and larger earthquakes within each polygon; and
of magnitude. It has an exponential taper applied to the num-
4. construct earthquake magnitude distributions and esti-
ber of events of large seismic moment. The TGR magnitude
mate the probability of great earthquakes within each
distribution is
polygon from the threshold rate and the appropriate mag-
β nitude distribution.
Mt Mt − M
FM exp for Mt ≤ M < ∞; 4
M Mc For the first step, we adopt polygons (F-E zones) defined
by Flinn and Engdahl (1965) and Flinn et al. (1974) (Fig. 1).
(Kagan, 2002a) in which β is the index parameter of the dis- We chose them because they define the subduction zones
tribution, and β 2=3b. M c is called corner moment (the reasonably well, and most important, they were defined be-
corresponding magnitude is called corner magnitude, mc ; fore the start of the Global Centroid Moment Tensor (CMT)
Table 1), which controls the distribution in the upper ranges (Ekström et al., 2005) earthquake catalog that we use for
of M. M t is a threshold moment (the corresponding magni- many of our statistical tests. Thus, they are independent of
tude is threshold magnitude, mt ) above which the catalog is the test earthquakes, avoiding the potential selection bias that
assumed to be complete. FM is the rate of earthquake could occur if the zones were drawn to include or exclude
moment larger than M, normalized to a cumulative rate individual earthquakes. F-E zone 3 covers only the southern
of 1 at the threshold magnitude mt . When M c → ∞, the Cascadia subduction zone (CSZ); we will deal with this
TGR relationship is equivalent to the GR relationship.
subduction zone separately.
Both truncated GR and TGR have two parameters: a
For the earthquake magnitude completeness and thresh-
b-value and a parameter describing a modification of the sim-
old rate estimates of steps 2 and 3, we rely on the Whole
ple GR law for very large earthquakes, and both distributions
Earth model study by Kagan and Jackson (2012), which is
ensure a finite seismic moment flux for a region. Bell et al.
a smoothed seismicity model for the world. Kagan and Jack-
(2013) showed that a preferred model of global earthquakes
son (2012) used the U.S. Geological Survey (USGS) Prelimi-
converges to a TGR distribution. However, Zöller (2013)
demonstrated that, statistically, the truncated and TGR distri- nary Determinations of Epicenters (PDE) earthquake catalog
butions performed equally well for the global seismicity. (Sipkin et al., 2000), which is complete at m ≥ 5 since 1969,
Because a sharp cutoff at the tail of the truncated GR distri- for earthquake locations. For magnitudes, they generally
bution contradicts the known behavior of dissipative physical used the moment magnitude. Moment magnitudes were first
dynamic systems that require a smooth transition, the TGR reported consistently in the Global CMT catalog beginning
distribution is more physically defensible using observed in about 1977, but now the PDE catalog includes moment
seismicity and seismic moment balances (Kagan, 2002a; Bird magnitudes for most earthquakes over magnitude 5.0. In
and Kagan, 2004; Pisarenko and Sornette, 2004). some cases, locations were reported in the PDE catalog, but
moment magnitudes were reported only in the Global CMT
Approach catalog. In those cases, Kagan and Jackson used the more
accurate locations from the PDE and the more accurate mo-
To solve for the probable maximum magnitude, mp T, ments from the Global CMT.
the basic steps in our analysis are to: For step 4, we construct an earthquake magnitude–
1. identify polygons describing major subduction zones frequency distribution for each F-E zone. We choose the TGR
where we will estimate the probabilities of large earth- distribution because it provides the needed limit on the long-
quakes; term moment rate, it fits several catalogs well with many
2. choose the lower threshold magnitude mt , at or above sufficiently large earthquakes, and it affords relatively easy
which the catalog is complete; computation. After completing the four steps, mp T can be

by gmontalva
1
19
20
21 18
22 5 7
46 6
23
16
24 15
14 8
12
Sources: Esri, USGS, NOAA
Slab 1.0, Depth (km)

Flinn-Engdahl zones
0
0
60
25
75
15
30
50
Figure 1. >
Subduction regions around the Pacific Ocean defined by Flinn–Engdahl (F-E) (Flinn and Engdahl, 1965; Flinn et al., 1974) and
Slab 1.0 (Hayes et al., 2012). F-E zones (black polygons) are: 1, Alaska-Aleutian arc; 3, Oregon, California, and Nevada; 5, Mexico-
Guatemala; 6, Central America; 7, Caribbean Loop; 8, Andean South America; 12, Kermadec-Tonga-Samoa; 14, New Hebrides Islands;
15, Bismarck-Solomon Islands; 16, New Guinea; 18, Guam-Japan; 19, Japan-Kamchatka; 20, Japan-Ryukyu Islands; 21, Taiwan; 22,
Philippines; 23, Borneo-Celebes; 24, Sunda arc; 46, Andaman Island. The Slab 1.0 model is represented by light (shallow) to dark (deep)
depth-to-slab contour lines. The color version of this figure is available only in the electronic edition.
read from the TGR curve: mp T is the magnitude corre- typical of most of our subduction zone catalogs, the values
sponding to 1=T occurrence rate on the curve. are (mc 0:09) 0:25 and (b0:006) 0:076. Given the un-
Theoretically, the TGR parameters β and Mc (or mc ) can certainties, we interpret results using the 95% confidence
be estimated by the maximum-likelihood method from earth- limits derived from contours of the likelihood function. For
quake data alone, if the catalog has a sufficient number of large N, the quantity w ≡ 2lmax − l is distributed approx-
large earthquakes. The maximum-likelihood method re- imately as a chi-square variable with two degrees of freedom,
quires computation of a log-likelihood function l, which implying elliptical contours as generally observed. Then the
is the natural logarithm of the likelihood that the given cata- 95% confidence limit for w is about 6 (Bird and Kagan,
log would be observed if the current trial values of β and Mc 2004), and lmax − 3 serves as a useful measure of the 95%
were correct, under the constraint that the number of earth- confidence region. Assuming a uniform constant prior distri-
quakes in the catalog with Mi > M t , N, is fixed. For the TGR bution and that the likelihood function can be normalized,
distribution, the posterior joint probability density function of the two
parameters is by Bayes’ rule a normalized version of the
XN XN
1 likelihood function (Zöller et al., 2013). Jackson and Mat-
l Nβ lnM t NM t − Mi − β lnM i su’ura (1985) showed by simulation that confidence limits
Mc i1 i1
estimated from the likelihood function are close to those
X N
β 1 from the posterior distribution if the prior distribution is
ln 5
i1
Mi Mc nearly uniform within the inferred confidence region, and the
product of likelihood function and prior distribution can be
(Kagan, 2002a; Bird and Kagan, 2004). The maximum-like- normalized. If the likelihood function is indeed nearly chi-
lihood estimates of β and Mc are associated with lmax , the square, then the 95% confidence limits are approximated
highest value of l. The maximum-likelihood estimates of well by lmax − 3. However, that contour may not close at
β and Mc (or b and mc ) have a slight negative bias. From a reasonable value of the corner moment if the catalog is too
1000 simulations, we estimate that, with 100 events above deficient in large earthquakes. As shown below, inclusion of
the magnitude mc -2, the estimated mc and b-values are additional data such as tectonic moment rate may be valuable
(mc 0:16) 0:32 and (b0:02) 0:10. With 200 events, for placing a meaningful upper confidence bound on mc .

by gmontalva
Otherwise, only the assumed prior distribution controls the They also calculated tectonic moment rates for the zones based
confidence limits of mc . on the geodetic strain rate map by Kreemer et al. (2003) using
In another method, we use the moment rate conservation the method described in Bird and Liu (2007) and Bird et al.
principle to constrain M c (or mc ). That principle states that a (2010). They used W 104 km, μ 49 GPa, and χ 0:5.
certain part of tectonic deformation is released by the cumu- Alternative values of W, μ, and χ with W 100 km,
lative effects of earthquakes. The seismic moment rate, M_ s , can μ 30 GPa, and χ 0:85 (e.g., Scholz and Campos,
be estimated either by summing up moments of known earth- 2012) produce the same value of M _ T (see equation 6) as
quakes or by integrating theoretical moment–frequency rela- those of Kagan and Jackson (2013). The coupling coeffi-
tions. The former method may yield very unstable estimates cient may be time dependent. Here, we use a constant value
because of the relatively short instrumental earthquake catalog and consider it a long-term average. Kagan and Jackson
and relatively long recurrence of great earthquakes. The latter (2013) estimated the corner magnitude for a theoretical mag-
method depends on the estimation of the parameters in the nitude distribution referred to as the gamma magnitude distri-
theoretical relations. Using a TGR relation, the estimation of bution in Kagan (2002a,b). The gamma magnitude distribution
M_ s depends on three variables: rt , the rate of earthquakes in is closely related to the TGR distribution we use here, but it is
the region larger than the threshold magnitude, the β-value, and slightly different and the meaning of corner magnitude is differ-
Mc . For an active subduction zone, quite stable estimates of rt ent for the two distributions.
and β can be obtained from catalog statistics. Using the seismic moment conservation principle, the
The tectonic moment rate can be estimated by TGR M c can be estimated as (Kagan, 2002b)
1=1−β
_ T χμWLu;
M 6 χM
_ T 1 − β
Mc ≃ ; 7
αt M βt Γ2 − β
in which χ is the seismic coupling coefficient, μ is the
rigidity, W is the down-dip width of the seismogenic zone, in which αt is the seismic activity level (occurrence rate)
L is the length of a fault, and u is the plate velocity. The for earthquakes with moment M t and greater, and Γ is the
parameters χ, W, and μ are usually not well determined gamma function.
and are subject to large uncertainties. In this study, for all the subduction zones except Casca-
The seismic coupling coefficient represents the fraction dia, we adopted the β-values and tectonic moment rates in
of plate tectonic motion within the lock zone released by Kagan and Jackson (2013), and derived mc and its uncer-
earthquakes. Other strain release processes include creep tainty for the TGR distribution, constrained by the observed
(as afterslip or in the velocity strengthening zone below the tectonic moment rate. The results are listed in Table 2. We
locked zone) and slow-slip events, with or without seismic present two sets of mc and β-values: one mc set is derived
tremors (Schwartz and Rokosky, 2007). Slow-slip events using zone-specific β-value for each region, and the other is
usually occur at the transition between the velocity weakening derived using the generic β-value of 0.65 for all the F-E
and strengthening zones (Cascadia, Mexico, New Zealand, zones. The specific β-value for an F-E zone is the value de-
Alaska) but may also occur within the velocity weakening zone rived using the earthquake data within that zone. The generic
(Japan, Costa Rica; Schwartz and Rokosky, 2007). In Casca- β-value is the value derived using the earthquake data in all
dia, 45%–65% of the plate convergence rate is accommodated the zones. The generic β-value is close to a b-value of 1.0.
by slow-slip events below the locked zone, however, any po- When the specific β-values are used, mc varies significantly
tential slow-slip strain release in the locked zone is too small to between zones. When the generic β-value is used, the mc
be detected by Global Positioning System (GPS; Wech et al., values are similar for all the zones. The uncertainties of
2009; Gomberg et al., 2010). It is difficult to quantify the frac- TGR mc are very similar to those for the gamma distribution
tion of tectonic moment released by aseismic processes in the used in Kagan and Jackson (2013). The distinction between
locked zone; for subduction zones, the coupling coefficient is this study and Kagan and Jackson (2013) is that (1) we de-
usually between 0.5 and 1.0 (Scholz and Campos, 2012). rived the mc values for TGR distribution instead of gamma
The seismogenic zone down-dip width is defined as the distribution; (2) we integrated paleoseismic data into the
distance between the up-dip limit near the trench and the Cascadia analysis; and (3) we estimated mp T values, pre-
down-dip limit (i.e., the locked part of a subduction zone) sented in the Results section.
along the subduction slab, and typical values are about
60–170 km (Hayes et al., 2012). Typical rigidity values used
in the calculation are 30 (McCaffrey, 2008) and 49 GPa (Ka- Cascadia Subduction Zone
gan and Jackson, 2013). The tectonic moment rate is usually
Magnitude of Cascadia Earthquakes
estimated from geodetic strain rate maps (Bird and Kagan,
2004; Bird and Liu, 2007; Bird et al., 2010). The CSZ is formed by the subduction of the oceanic Juan
Kagan and Jackson (2013) calculated the β-values for de Fuca and Gorda Plates beneath the North America Plate
the F-E subduction zones using the maximum-likelihood off the coast of northern California, Oregon, Washington,
method and the Global CMT catalog from 1977 to 2010. and Vancouver Island. The earthquake potential of the

by gmontalva
Table 2
Corner Magnitude (mc ), β-value, and Standard Deviations (σ) of Each of the F-E Subduction
Zones Based on TGR Distribution
Zone-Specific β-Value Generic β-Value
F-E Zone Number Name N* β σβ mc σ mc β mc
1 Alaska-Aleutian arc 280 0.65 0.04 9.01 0.28 9.01

5 Mexico-Guatemala 164 0.60 0.06 8.55 0.29 8.82
6 Central America 161 0.68 0.06 9.08 0.29 8.88
7 Caribbean Loop 59 0.62 0.09 8.81 0.31 8.99
8 Andean South America 286 0.57 0.04 8.89 0.28 9.42
12 Kermadec-Tonga-Samoa 439 0.80 0.04 10.14 0.28 8.77
14 New Hebrides Islands 424 0.59 0.03 8.32 0.28 8.62
15 Bismarck-Solomon Islands 448 0.60 0.03 8.35 0.28 8.59
16 New Guinea 266 0.66 0.05 9.56 0.28 0.65 9.48
18 Guam-Japan 88 0.86 0.10 13.05 0.30 9.50
19 Japan-Kamchatka 425 0.62 0.04 8.89 0.28 9.09
20 S.E. Japan-Ryukyu Islands 57 0.62 0.10 9.25 0.31 9.47
21 Taiwan 110 0.64 0.07 8.73 0.29 8.79
22 Philippines 244 0.68 0.05 9.02 0.28 8.82
23 Borneo-Celebes 266 0.68 0.05 9.09 0.28 8.89
24 Sunda arc 278 0.65 0.04 9.22 0.28 9.22
46 Andaman Island Sumatra 143 0.71 0.07 9.50 0.29 9.03
All (1977–2010) 4217 0.65 0.01 9.02 0.27 9.02
*Number of m ≥ 5:8 earthquakes in the Global CMT catalog from 1977 to 2010.
CSZ has been discussed by Atwater (1987), Savage and Li- correlation length) to represent the approximate length of
sowski (1991), Mitchell et al. (1994), Atwater et al. (1995), the ruptures. Goldfinger, Morey, et al. (2013) reiterate and
Hyndman and Wang (1995), Satake et al. (1996, 2003), slightly revise the southern rupture segment conclusions using
McCaffrey et al. (2000), Mazzotti et al. (2003), and Nelson additional cores and high-resolution seismic reflection data.
et al. (2006). Despite the presence of abundant paleoseismic The turbidite data do not contain direct indicators of
evidence for rapid coastal subsidence and tsunamis, the Cas- earthquake magnitude. However, some indirect indicators
cadia plate boundary remains seismically the quietest of all can be used for that estimate. First, Goldfinger et al. (2012)
circum-Pacific subduction zones, with only one significant noted strong relations between the correlated rupture length
interplate thrust event (the 1992 m 7.1 Cape Mendocino and turbidite thickness and turbidite mass, and interpreted
earthquake) ever recorded instrumentally (Oppenheimer et al., those to indicate that longer ruptures are associated with
1993). Fortunately, strong shaking caused by the great subduc- greater shaking. Second, there is a correlation between tur-
tion zone earthquakes can trigger turbidity currents, which leave bidite mass per event at several widely separated sites. This
deposits in marine sediments. As a result, an ∼10;000-year indicates that, at least for the Holocene and the selected sites,
turbidite record has been developed during the investigation of there is a consistent relationship between the evidence for
the turbidite system along the CSZ (Goldfinger et al., 2012). shaking along strike for many of the events. Goldfinger et al.
The turbidite and onshore paleoseismic records can be used to (2012) estimated the magnitude of the events based on these
reconstruct the large earthquake history in the region. observations. They used interevent time following each event
Goldfinger et al. (2012) interpreted the turbidite data to calculate average slip, an assumed coupling coefficient of
and suggested four types of earthquake rupture along the 1.0 as strongly suggested by the geodetic data (Hyndman and
CSZ during the past 10,000 years (Fig. 2 and Table 3): Wang, 1995; McCaffrey et al., 2000, 2013; Wang et al.,
(1) 19–20 full-margin or nearly full-margin ruptures (sec- 2003; Scholz and Campos, 2012), and rupture lengths based
tion A); (2) three or four ruptures of the southern 50%–70% on the distances between correlated sites. The rupture widths
of the margin (section B); (3) 10–12 southern ruptures from were based on geophysical and geodetic estimates, and were
central Oregon southward (section C); and (4) 7–8 southern assumed to be 83, 60, 50, and 40 km for four types of rup-
Oregon/northern California events with the possibility of un- ture, respectively, assuming reduced width for smaller rup-
resolved events (section D). Regarding the accuracy of the tures, except that 55 km is used for events T2, T12, T17, and
rupture extents shown in Figure 2, Goldfinger et al. (2007) T17a to account for their relatively thin turbidite beds and/or
concluded the maximum distance from an earthquake rup- short interevent time. Table 3 lists the parameters and the
ture to a triggered turbidity current is likely less than 90 km estimated moment magnitudes.
for full-margin Cascadia events and less for smaller events. The estimated magnitudes by Goldfinger et al. (2012)
Hence, it is reasonable to use the distance between offshore are uncertain considering that the strain accumulation and
core sites containing turbidites from the same event (or slip for each event were unknown (Goldfinger et al., 2012).

by gmontalva
130° W 125° W 120° W 115° W

characteristics proposed by Kagan (2002c), the CSZ rup-
ture lengths of 1000, 660, 444, and 222 km correspond to
rupture widths of 126, 84, 56, and 28 km, respectively.
50° N By replacing the rupture widths in Table 3 with these values,
A
we obtain a set of alternative magnitudes (the column malt1 in
B
Table 3).
The columns m and malt1 in Table 3 were estimated based
C
45° N on a simple time-predicable model, which is essentially the
D seismic gap theory, and assumes that the seismic moment
in a given megaquake is accumulated only since the previous
one. Rong et al. (2003) showed that the seismic gap model
40° N does not work well globally. Goldfinger, Ikeda, et al. (2013)
estimated the energy state of the 19 full-margin rupture events
using the recurrence time between events to represent energy
gain and the turbidite mass per event to represent energy loss
35° N (mass per event was estimated at several core sites based on
PAGER-Cat core density data, averaged and given separately for each
m 5.0 - 5.5
event). They concluded the Cascadia earthquakes followed
m 5.5 - 6.0
m 6.0 - 6.5 neither a time-predicable nor a slip-predicable model: instead,
30° N m 6.5 - 7.0 they propose a long-term cycling of energy storage and re-
m > 7.0 Sources: Esri, USGS, NOAA
lease. For the 19 full-margin rupture events, the turbidite mass
and thickness vary considerably from event to event at each
Figure 2. Original Flinn–Engdahl zone 3 (the polygon defined site but are relatively consistent for the same event from site to
by black dashed lines) and the redefined Cascadia region (the rec- site (Goldfinger et al., 2012, Goldfinger, Ikeda, et al., 2013).
tangular box defined by thick solid black line). Thin black lines
show the plate boundaries in the Cascadia region. Circles are his-
Because of this consistency for individual events along the
torical earthquakes in the PAGER-CAT catalog (Allen et al., 2009) margin, and despite the obvious simplification involved, Gold-
since 1900. The four gray lines labeled from A to D represent the finger, Ikeda, et al. (2013) inferred that turbidite mass can be
four rupture sections listed in Table 3. Note that the four sections are considered a crude proxy for seismic moment or intensity of
intentionally shifted from the plate boundary so that they can be ground shaking at offshore sites for at least the 19 full-margin
seen on the map. The color version of this figure is available only
in the electronic edition.
rupture events.
Accordingly, we propose another way to estimate mag-
nitudes of the full-margin rupture events. We use the 1700
We consider three alternate sets of magnitudes based on dif- earthquake as a reference and assume that it has a magnitude
ferent assumptions. In the first, we use the method of Gold- of 9.0 (Satake et al., 2003), then the seismic moment of the
finger et al. (2012) but with different rupture widths. The ith event, Mi, can be estimated by
rupture widths used by Goldfinger et al. (2012) were inferred
from published thermal models (Hyndman and Wang, 1995;
Flück et al., 1997), GPS-based models (McCaffrey et al.,
Mi M1 × massi=mass1 x ; 8
2007), and the structural transition from contraction to exten-
sion (Priest et al., 2009). However, episodic tremor and slip in which M(1) and mass(1) are the seismic moment and tur-
(ETS) data have been interpreted to suggest that future co- bidite mass (the average of four sites) of the 1700 earthquake,
seismic rupture might extend to 25 km depth, or ∼60 km and massi is the turbidite mass (the average of five sites) of
inland of the Pacific coast, rather than stopping offshore at the ith event. We vary x and find that the total seismic mo-
15 km depth (Chapman and Melbourne, 2009). In fact, al- ment rate is about in the range of the tectonic moment rate
most all of the m ≥ 8:8 earthquakes along other subduction estimated using the Global Strain Rate Model (Kreemer et al.,
zones ruptured more than a 100 km down-dip width (Strasser 2003) when x is between 0.5 and 1.0. Thus, we obtain an-
et al., 2010). The rupture width of the 2011 m 9.0 Tohoku other two sets of magnitudes by assuming x 0:5 and x
earthquake was about 200 km, although the rupture length 1:0 (the columns malt2 and malt3 in Table 3, respectively). For
was only about 450 km (Yoshida et al., 2011). The rupture the partial-margin rupture events, we cannot use this method
width of the 2011 Chile m 8.8 earthquake was estimated to to infer magnitudes because a reliable correlation of turbidite
be larger than 140 km (Wang et al., 2012, and references mass between different sites has not been demonstrated.
therein). Kagan (2002c) investigated the distribution of after- Therefore, for those partial margin rupture events, we use
shock zones for large earthquakes. He found that all earth- the magnitudes determined by the first alternative method.
quakes show the same magnitude-rupture scaling and that The estimated alternate magnitudes are subject to large un-
earthquake geometrical focal zone parameters are self- certainties; however, they represent some reasonable range of
similar. Using the scaling relationship and the self-similarity the magnitudes of great CSZ events.

by gmontalva
Table 3
Estimated Earthquake Magnitudes of the Turbidite Events
Northern Margin Southern Margin
Turbidite Mean Following Interval Following Interval Section Rupture Rupture
Number Age* (years)* (years)* Name* Length (km)* Width (km)* m* malt1 † malt2 † malt3 †
1 250 A 1000 83 9.00 9.15 9.00 9.00

2 482 232 232 A 1000 55 8.70 8.97 8.95 8.89
2a 550 57 D 222 40 8.19 8.56 7.70 7.70
3 798 305 248 A 1000 83 8.87 9.02 9.05 9.10
3a 1077 279 C 444 50 8.34 8.68 8.55 8.55
4 1243 446 167 A 1000 83 8.90 9.05 9.03 9.06
4a 1429 186 C 444 50 8.25 8.33 8.43 8.43
5 1554 311 125 A 1000 83 8.80 8.95 9.05 9.11
5a 1820 266 C 444 50 8.41 8.61 8.54 8.54
5b 2036 216 B 660 60 8.66 7.96 8.71 8.71
5c 2323 286 C 444 50 8.41 8.28 8.56 8.56
6 2536 982 213 A 1000 83 9.09 9.24 9.07 9.14
6a 2730 194 D 222 40 8.24 8.19 8.04 8.04
7 3028 492 298 A 1000 83 8.97 9.12 9.09 9.18
7a 3157 129 D 222 40 8.23 8.16 7.92 7.92
8 3443 415 286 A 1000 83 8.94 9.09 9.08 9.16
8a 3599 442 B 660 60 8.67 8.00 8.61 8.61
8b 3890 447 D 222 40 8.15 7.70 8.16 8.16
9 4108 665 218 A 1000 83 9.01 9.16 9.04 9.08
9a 4438 548 B 660 60 8.35 8.55 8.83 8.83
9b 4535 426 D 222 40 8.17 8.43 7.84 7.84
10 4770 661 235 A 1000 83 9.01 9.09 9.05 9.10
10a 5062 292 C 444 50 8.39 8.54 8.56 8.56
10b 5260 198 B 660 60 8.43 8.71 8.68 8.68
10c 5390 130 C 444 50 8.55 8.56 8.33 8.33
10d 5735 344 C 444 50 7.9 8.04 8.61 8.61
10f 5772 37 C 444 50 8.37 7.92 7.96 7.96
11 5959 1189 187 A 1000 83 9.13 9.28 9.19 9.38
12 6466 508 508 A 1000 55 8.93 9.19 8.83 8.67
12a 6903 437 D 222 40 8.22 8.61 8.28 8.28
13 7182 715 278 A 1000 83 9.04 9.19 9.07 9.14
14 7625 443 443 A 1000 83 9.01 9.16 9.01 9.02
14a 7943 318 D 222 40 8.17 8.16 8.19 8.19
15 8173 548 230 A 1000 83 8.97 9.12 9.02 9.04
15a 8459 286 D 222 40 8.36 8.83 8.16 8.16
16 8906 733 447 A 1000 83 9.09 9.24 9.21 9.41
16a 9074 169 D 222 40 7.54 7.84 8.00 8.00
17 9101 195 27 A 1000 55 8.49 8.76 9.01 9.02
17a 9218 117 117 A 1000 55 8.5 8.77 8.99 8.99
18 9795 577 577 A 1000 83 9.08 9.23 9.08 9.15
*From Goldfinger et al. (2012).

†Magnitudeestimated by this study.
Magnitude–Frequency Distributions for the earthquake occurrence above a lower-threshold magnitude in

Cascadia Region the zone. We first apply the maximum-likelihood method pre-
sented by equation (5) to estimate β-value and corner magni-
The Cascadia subduction region was not analyzed by
Kagan and Jackson (2013) because it was not well defined tude for the region. We use the earthquakes in the Global CMT
in the F-E zones. The F-E zone 3 (Figs. 1 and 2) covers Cal- catalog from 1976 to 2012 to compute the log-likelihood func-
ifornia, Nevada, and the southern CSZ but does not include tion (equation 5). We determine the Global CMT catalog is
the northern CSZ. Here, we define a new Cascadia region (the complete at magnitude 5.5 since 1976 for the region. Figure 3a
rectangular box in Fig. 2) to cover the entire CSZ. The newly illustrates the likelihood contour map using m ≥ 5:5 earth-
defined Cascadia region also covers the Explorer, Juan de quakes in the Global CMT catalog. The maximum-likelihood
Fuca, and Gorda plates; the coordinates of its four corners β-value is about 0.62 with a very wide 95% confidence range.
are (132°, 52°; 123°, 52°; 123°, 39°; 132°, 39°). The maximum-likelihood estimate of the corner magnitude
As for other subduction zones, we used the Whole Earth is unreasonably small due to the lack of large earthquakes
model by Kagan and Jackson (2012) to estimate the rate of in the catalog, but the 95% confidence region is essentially

by gmontalva
(a) 9 10,000-year medium-to-large event set (with magnitudes
−6
−3
smaller than turbidite events). Figure 3b displays the likeli-
0
−6
hood contour plot using the combined catalog. In this case,
8.5
both the maximum-likelihood β-value and corner magnitude
are well constrained. The maximum-likelihood β-value is
Corner magnitude
−3
8
about 0.65, same as the generic value for all the subduction
zones. The narrow 95% confidence limit of the β-value re-
−6
−3
sults from the combination of 270 samples. The maximum-
0
−6
7.5 likelihood mc is 8.84 with a 95% confidence range of 8.6 and

9.5. This exercise demonstrates that combining an instru-
+3
mental catalog with paleoseismic data is an effective way
−3
7 0 −3 −6 to solve for magnitude limits for a region.

0
−6 −3
−6
Because the instrumentally recorded earthquake data are
very limited for the CSZ, joint estimation of M c and β intro-
6.5 duces very large uncertainties for both. Thus, we also estimate
0.4 0.5 0.6 0.7 0.8 0.9 1
β−value β separately by a widely used maximum-likelihood technique
(Aki, 1965; Bender, 1983; Utsu, 1999; Knopoff, 2000):
(b)
−3
−6
9.8
log10 e
b 9
−3
9.6 hmi − mt − Δm2bin

−6
9.4 and
b
0
10
Corner magnitude
9.2 σ b p ;
N−1
−3
−6
9
0
−3
+3 in which Δmbin is the magnitude binning width of the catalog,

8.8
mt is the lower threshold magnitude of the catalog, hmi is the
−6
8.6 0 mean magnitude of the earthquakes equal to or larger than mt

−3
−6 in the catalog, and N is the total number of earthquakes with
8.4
magnitude equal to or larger than mt . Using equations (9) and
8.2 (10), we obtain a β-value of 0:59 0:05. As verification, we
estimate the β-value using the m 5:5 earthquakes after 1990
8
0.6 0.62 0.64 0.66 0.68 0.7 and obtain the same value but with a slightly larger uncertainty
β−value (0:06) because the total number of earthquakes is reduced
when a shorter time period is used.
Figure 3. Contour plots of relative log likelihood (l) for tapered The TGR mc can be estimated using the maximum-
Gutenberg–Richter (TGR) distributions for the Cascadia region. For likelihood estimate of equation (5) from the combination
convenience, a constant is added to all values of l to adjust lmax to
the value of +3, so that the zero contour is approximately the 95% of instrumental and paleoseismic data, as illustrated in
confidence limit. (a) Fit to m ≥ 5:5 earthquakes in Global Centroid Figure 3b. To be consistent with other zones, we also em-
Moment Tensor (CMT) catalog; (b) fit to the combination of Global ployed equation (7) to calculate mc . There are four param-
CMT catalog and turbidite data. The color version of this figure is eters in the equation:
available only in the electronic edition.
1. The seismic coupling coefficient, χ. We use 1.0 for this
region because geodetic data show that this zone is
unbounded. That is, the catalog data alone provide no useful strongly coupled (Hyndman and Wang, 1995; McCaffrey
upper limit to the corner magnitude. et al., 2000, 2013; Wang et al., 2003; Scholz and Cam-
Because the instrumental catalog is too short to include pos, 2012), and tsunami data require full coupling to fit
great CSZ earthquakes, we supplement it with the turbidite- runup data where present (Witter et al., 2012). Aseismic
based events (Table 3) to compute the log-likelihood func- processes apparently operate only below the coupled
tion. We present the results using the magnitudes estimated zone (Gomberg et al., 2010).
by Goldfinger et al. (2012), because the results from the 2. The TGR β-value. We use 0.59 as the zone-specific β-value
other alternate sets of magnitudes are similar. To combine and 0.65 as the generic β-value, like for the other zones.
the 37-year long (1976–2012) Global CMT catalog with the 3. The seismicity rate αt , which we estimate from the Whole
10,000-year long turbidite data, we apply a bootstrap tech- Earth model by Kagan and Jackson (2012).
nique to the 37-year Global CMT catalog to draw 270 4. The tectonic moment rate M _ T of the region, which we
(10;000=37 ≈ 270) bootstrap samples. We assume that the estimate using the Global Strain Rate Model v2.0, an
combination of the 270 bootstrap samples represents a update of the model by Kreemer et al. (2003).

by gmontalva
10
Cumulative Annual Rate of Occurence

1
0.1
0.01
GCMT
PAGER-Cat
PDE
0.001
TGR (beta=0.59,mc=9.02)
TGR (beta=0.65,mc=9.58)
0.0001
5 6 7 8 9 10
Magnitude
Figure 4. Earthquake magnitude–frequency distributions for the Cascadia region. The dashed curve represents the TGR model with mc
9:02 and β 0:59. The solid curve represents the TGR model with mc 9:58 and β 0:65.
The seismogenic width in Cascadia is uncertain without curve in Fig. 4). We also plot the empirical distributions
a large instrumental event. Thermal, geodetic, structural, and based on the Global CMT (1976–2012), Prompt Assessment
paleoseismic sources of evidence are relatively consistent, of Global Earthquakes for Response (PAGER-CAT) (1900–
suggesting 20 km is a reasonable average value for the var- 2008), and PDE (1973–2012) catalogs. For the PAGER-CAT
iable depth of the down-dip edge of plate coupling. Frankel catalog, the rate of m ≥ 5:5 is plotted because the catalog
and Petersen (2013) adopted a similar value for the main does not include many smaller earthquakes. Up to m 7.2,
branch (weighted 0.5) of the logic tree in the 2014 U.S. na- the first model (dashed curve) is fairly consistent with both
tional seismic-hazard maps. Lower weightings were assigned the Global CMT and PDE catalogs. The PAGER-CAT catalog
to deeper (top of the ETS zone) and shallower alternatives has a slightly lower rate than the other two catalogs but is
(primarily based on thermal models) based on workshop more consistent with the second model (solid curve). The larg-
consensus results. We use an average vertical seismogenic est event in the PAGER-CAT catalog, the 1946 Vancouver Is-
thickness of 20 km and dip of 9° in the tectonic moment rate land earthquake, has a magnitude of 7.6 (the 2012 m 7.7 Haida
calculation. We assume the up-dip rupture extends to the sea- Gwai and the 1949 m 8.1 Queen Charlotte fault earthquakes
floor. Thus, the average seismogenic width is about 127 km, are outside the rectangular box in Fig. 2). The difference be-
consistent with the value calculated using the scaling rela- tween catalogs is caused by magnitude uncertainties and dis-
tionship by Kagan (2002c) for a magnitude 9 earthquake. crepancies, and the different time periods covered by different
Some GPS-based models show coupling of the outer accretion- catalogs.
ary wedge (e.g., McCaffrey et al., 2013), consistent with our The most recent great earthquake in the CSZ is the Janu-
assumption. But it is also possible that significant parts of the ary 1700 earthquake, estimated to be m 8.7–9.2 (Satake et al.,
up-dip accretionary wedge are uncoupled, as discussed in 2003). Turbidite data show an average recurrence of ∼500
Clarke and Carver (1992), Goldfinger et al. (1992, 1997), and years for great earthquakes along CSZ (Goldfinger et al.,
Priest et al. (2009). For the rigidity, we use 30 GPa. The M_ T is 2012). None of the catalogs includes such large earthquakes
20
estimated to be 1:9 × 10 N·m=year. Using the zone-specific given the short times covered by the catalogs. Besides, the
β-value of 0.59, mc is 9.02. For the generic β-value of 0.65, mc region has been very quiet at least since 1900. It is not clear if
is 9.58. The mc values derived from the moment conservation the post-1900 rate well represents the long-term rate. The
principle are slightly larger than the value derived from purely turbidite data provide a rare opportunity to verify the tail of
catalog statistics using the combined instrumental and turbidite the TGR distributions. Hence we plot the distribution of tur-
earthquakes. The difference can be ascribed to the incomplete- bidite events together with the TGR distributions in Figure 5.
ness of turbidite events, the magnitude uncertainty of the tur- Although the turbidite events with estimated magnitudes < 8:0
bidite events, the uncertainty of the parameters in the tectonic may not be complete (Goldfinger et al., 2012), we plot them to
moment rate calculation, and the slight negative bias in our demonstrate that events of magnitude 7.5–8.0 did occur along
maximum-likelihood method discussed above. this subduction zone. The 95% confidence limits for the TGR
We construct two theoretical TGR magnitude–frequency distributions, which are calculated using the Monte Carlo
distributions for the Cascadia region. In the first, we use technique discussed below, are also shown. Although there
mc 9.02 and β-value 0.59 (the dashed curve in Fig. 4), and are differences between the four sets of estimated magnitudes,
in the second, we use mc 9.58 and β-value 0.65 (the solid for the magnitude range of m ≥ 8:0, all the sets fall into the

by gmontalva
Figure 5. Earthquake magnitude–frequency distributions for the Cascadia region. The solid dots represent the empirical relationship
based on the Global CMT catalog. The black curves illustrate TGR distribution with β 0:59 0:05 and mc 9:02 0:27 and its 95%
confidence limits, and the gray curves illustrate TGR distribution with β 0:65 0:05 and mc 9:58 0:27 and its 95% confidence limits.
The thick long-dashed curve represents the modeled occurrence of great earthquakes in the 2008 U.S. Geological Survey national seismic-
hazard maps (Petersen et al., 2008). The other distributions are based on magnitudes inferred from turbidite data: circles by Goldfinger et al.
(2012); squares, triangles, and diamonds are based on columns malt1 to malt3 in Table 3. The two vertical dashed lines separate the empirical
magnitude–frequency distributions based on the instrumental earthquake data, incomplete turbidite data, and turbidite data, respectively. The
color version of this figure is available only in the electronic edition.
95% confidence limit of the TGR with β-value of 0.65. For any evidence of events that ruptured only the northern
m ≥ 8:3, all the sets also fall into the 95% confidence limit CSZ. On the other hand, Atwater and Griggs (2012) dis-
of the TGR with β-value of 0.59. cuss additional tsunami evidence from Discovery Bay,
In the USGS 2008 national seismic-hazard maps (Pe- along the eastern Juan de Fuca strait in northern Wash-
tersen et al., 2008), the primary constraint in the hazard model ington, for an event that does not correspond in time with
is that every location in the CSZ is ruptured on average every Pacific coastal evidence of whole-CSZ rupture events. A
500 years, based on the return time of large earthquakes. Two greater frequency of far northern plate boundary earth-
sets of rupture scenarios were considered for the events: quakes would help explain the roughly 300-year average
(1) m 9:0 0:2 events that rupture the entire CSZ every interval between inferred tsunamis at Discovery Bay (At-
500 years on average and (2) m 8.0–8.7 events with rupture water and Griggs, 2012). However, Goldfinger, Beeson,
zones that fill the entire zone over a period of about 500 years. et al. (2013) searched archive cores for such evidence and
A weighing factor of 0.67 was assigned to the m 9:0 0:2 found only a small number of potential northern events;
scenario, and a weighing factor of 0.33 was assigned to the these had been previously reported by Adams (1990). At-
m 8.0–8.7 one. For comparison, we plot the USGS modeled water and Griggs (2012) argued that gentle slopes and a
rate in Figure 5. The USGS model lies entirely between the broad continental shelf may put northern Cascadia cores
lower bound of 95% confidence limit and median of the TGR at a disadvantage in recording small turbidite deposits.
with β 0:65. For the magnitude range of 8.4–9.2, the USGS However, a new analysis of continental slope cores in
model falls between the lower bound of 95% confidence limit Washington shows that the upper slope cores, in areas
with steeper slopes and higher sediment supply, contain
and median TGR models of β 0:59.
Nevertheless, the discrepancy between the paleoseismic only a few potential northern events not previously reported
(Goldfinger, Beeson, et al., 2013; Goldfinger, Morey, et al.,
data and the theoretical TGR models cannot be ignored: the
2013). Furthermore, Enkin et al. (2013) show that in a
median TGR models predict more m < 8:5 earthquakes than
highly sensitive fiord site with very steep sidewalls, no ex-
observed by turbidite data. There are several possible reasons
tra turbidites are observed beyond those correlated to the
for the discrepancy:
offshore record, other than the 1946 Vancouver Island
1. The turbidite record may not be complete at magnitudes earthquake. Nevertheless, if more events of m 7–8 are even-
less than 8.5. Goldfinger et al. (2012) believed that the tually confirmed, the m < 8:5 gap between turbidite data
turbidite record is complete at m ∼ 8. They reported and TGR curves in Figure 5 could be reduced.
about 20 earthquakes that ruptured only the southern 2. Some turbidite flows believed caused by magnitude 9
and/or southern and central CSZs, but they did not find events might have been caused by multiple smaller events

by gmontalva
Table 4
Results of mp T from Different TGR Models for the CSZ
In 50 years In 100 years In 250 years In 500 years In 1000 years In 10,000 years
TGR Models mp σ mp σ mp σ mp σ mp σ mp σ
β 0:59 mc 9:02 7.94 0.20 8.25 0.21 8.59 0.20 8.80 0.20 8.96 0.21 9.29 0.23
β 0:65 mc 9:58 7.69 0.18 8.00 0.20 8.40 0.22 8.69 0.22 8.95 0.22 9.56 0.22
closely spaced in time. In fact, event T5 is such a candidate Results

(Goldfinger et al., 2012).
3. The TGR curves represent the occurrence of all the shal- Uncertainty Estimation
low earthquakes in the rectangular box in Figure 2, in- We construct TGR distributions for the circum-Pacific
cluding events related to structures other than the CSZ. F-E subduction zones. For each region, we calculate the rate of
There is potential of m 8 earthquakes along the Mendo- events larger than the threshold magnitude (mt 5:0) by sum-
cino and Blanco fracture zones and the fault structures to ming up the smoothed rate density in the Whole Earth model
the north. However, the turbidite-based events were likely on each 0:1° × 0:1° grid multiplied by the area of the grid. For
only related to the subduction zone. all the regions but the CSZ, the β-value and the corner magni-
4. The large earthquakes in the CSZ might not follow the tude listed in Table 2 were used to construct the TGR distribu-
TGR distribution. However, they do not follow obviously tions. Because the largest earthquakes are usually subduction
slip or time-predictable models, either (Goldfinger, interplate earthquakes, only shallow earthquakes (depth less
than 70 km) in the catalogs are considered in this study.
Ikeda, et al., 2013). Earthquake clusters and gaps are ob-
Because mp is determined by a TGR curve, it is a function
served during the 10,000-year event history (Goldfinger
of three input variables: rt (the rate of earthquakes larger than a
et al., 2012). Goldfinger, Ikeda, et al. (2013) suggested threshold magnitude), the TGR β-value, and mc . We propagate
that those features represent long-term supercycles in the the uncertainties in the three input variables into the uncertainty
subduction zone. of mp . Among the three input variables, rt and β are positively
5. There could be substantial temporal variation in the earth- correlated. We performed a sensitivity test on the correlation of
quake rate at all magnitudes. Goldfinger, Ikeda, et al. (2013) the two parameters. Our results show that, from noncorrelated
observe temporal variations in large magnitude events, (correlation coefficient is 0) to fully correlated (correlation co-
but similar variations may exist for other magnitudes that efficient is 1), the change of mp is less than 0.03 magnitude
have not been observed in the seismic recordings of the units. The change of σmp is less than 0.1 for all the zones
last century. except the CSZ, in which σmp reduces from 0.28 to 0.14
6. It is very difficult to infer magnitudes of paleoearthquakes when correlation is changed from weak to strong. Thus, we
from the turbidite data. Because many assumptions were adopt a correlation coefficient of 0.5 for our calculations.
involved, the inferred magnitudes are subject to large We use Monte Carlo simulation to calculate the uncer-
tainty of mp for each of the F-E zones. We assume normal
uncertainties.
distributions for β and mc . The mean and uncertainty of β and
Whatever the cause of the apparent discrepancies, there mc are from Table 2. The mean value of rt is calculated from
is no question that the turbidite data and onshore paleoseis- the Whole Earth model (Kagan and Jackson, 2012). The
p

mic evidence imply that huge earthquakes, magnitude 9 or standard deviation of rt is estimated by rt T cat =T cat, in
greater, have occurred along the CSZ. Any magnitude– which T cat is the historical earthquake catalog duration used
frequency model must accept large uncertainties because of in the Whole Earth model. The underlying assumption of this
the limited data and reliance on proxy information, but such standard deviation is that the occurrence of m ≥ 5:0 earth-
models are required to estimate probable magnitudes over quakes during the catalog time follows the Poisson distribu-
different time periods. tion. We simulate 2000 rt , β, and mc sets for each of the
We summarize the results of CSZ mp T derived from subduction zones. For each of the 2000 simulations, a
different TGR models in Table 4. The two TGR models in- TGR distribution is created, and mp for each return period
tersect at mp 1000 but diverge at both shorter and longer of interest is calculated so that we obtain a distribution of
return periods. The result of mp 500 8:7 or 8.8 does mp at each of the return periods (e.g., Fig. 6).
not conflict with the general statement that “the return time
of m ∼ 9 earthquakes is ∼500 years” because the former
Maximum Probable Earthquake Magnitude Using
comes from a TGR distribution and the latter implies a char-
Zone-Specific β-values
acteristic model. In fact, the median TGR models predict the
same rate as the USGS model at m 9.0 and a higher rate at Table 5 summarizes the estimated median of mp and un-
magnitudes lower and higher than 9.0 (Fig. 5). certainties for return periods of 50, 100, 250, and 500 years

by gmontalva
2
10 of the upper bound of mh . The GLHECAT includes global
large (m ≥ 7:0) earthquakes from A.D. 1000 to A.D. 1903. If
Annual Cumulative Number of Earthquakes
1
10
the maximum magnitude in the pre-1900 catalog is larger
10
0 than the historical mh from the PAGER-CAT and Centennial
catalogs, the pre-1900 magnitude is listed in the table and
−1
10 shown in the figure.
Using the preferred magnitude in the PAGER-CAT cata-
log, 6 of the mh after 1900 are larger than the mp 100,
−2
10
−3
whereas 12 are smaller. Using the largest magnitude in
10
the Centennial catalog, the mh since 1900 in 11 zones falls
10
−4 between the 50-year and 250-year expected mp . Is the ob-
served mh consistent with the estimated mp ? To answer this
10
−5 question, we apply the two-tail paired Student’s t-test to the
mp 100 and mh since 1900. Using the PAGER-CAT mh, a
−6
10 p-value of 0.23 is obtained. Using the Centennial mh , a
5 6 7 8 9 10 11
Magnitude p-value of 0.36 is obtained. This means that we cannot reject
the hypothesis at the 95% confidence level that the observed
Figure 6. TGR distributions for the Alaska–Aleutian subduction mh is consistent with the estimated mp for either catalog.
zone (F-E zone 1). Each of the thin gray curves represents one of the
2000 simulations. The solid black curve is created using the mean
values of β, mc , and rt . The two dashed black curves represent the Maximum Probable Earthquake Magnitude Based on
95% confidence limits of the TGR distribution based on the 2000 the Generic β-Value
simulations.
We adopted the generic β-value of 0.65 for all the zones
and the corresponding mc to estimate mp . The results are
estimated using zone-specific β and the corresponding mc shown in Table 6. We again applied the two-tail paired Stu-
values. Except for Guam–Japan (F-E zone 18), the estimated dent’s t-test to the mp 100 based on the generic β-value and
mp exceeds 7.9 at 50-year, 8.2 at 100-year, 8.5 at 250-year, mh since 1900. Using the PAGER-CAT’s mh , a p-value of
and 8.6 at 500-year return periods. The 50-year mp is equal to 0.06 was obtained. Using the Centennial’s mh , a p-value of
or larger than 8.4 for the Alaska–Aleutian arc, Andean South 0.80 was obtained. Therefore, we cannot reject the hypoth-
America, New Hebrides Islands, Bismarck–Solomon Islands, esis at 95% confidence level that the observed mh is consis-
New Guinea, Japan–Kamchatka, and Sunda arc regions. For tent with the mp estimated from the generic β-value.
the Andean South America and Japan–Kamchatka regions,
the 100-year mp is about 8.9. The 250-year mp is greater
Discussion
than 9.0 for Andean South America, New Guinea, Japan–
Kamchatka, and Sunda arc. The 500-year mp is about 9.0 The b-value (or β-value) is an important parameter in
or larger for the Alaska–Aleutian arc, Andean South America, seismic-hazard analysis because small changes in b-value
New Guinea, Japan–Kamchatka, southeast Japan–Ryukyu can result in large changes in the projected numbers of major
Islands, Borneo–Celebes, and Sunda arc. Most of the uncer- earthquakes. The debate of universal or nonuniversal b-values
tainties of mp are 0.2–0.3 magnitude units, which is about the across different regions has been going on for years. Schor-
uncertainty of mc . However, the uncertainty is larger for the lemmer et al. (2005) showed the b-value varies systematically
Caribbean Loop, Guam–Japan, and southeast Japan–Ryukyu for different styles of faulting. Kagan (1999) and Kagan and
Islands due to relatively high uncertainties on the β-values and Jackson (2013) state that statistically β-values are approxi-
mc and relatively low seismic activity. mately the same for all regions. Figure 8 compares the mp esti-
We compare the estimated mp with the available histori- mated from zone-specific and generic β-values at 50-, 100-,
cal maximum magnitudes mh (Table 5 and Fig. 7). The mh is 250-, 500-, 1000-, and 10,000-year return periods. The dif-
collected from three sources: the PAGER-CAT catalog (Al- ferences are small for most of the regions but significant for
len et al., 2009), the Centennial catalog (Engdahl and Villa- a few regions. For example, in Andean South America, where
señor, 2002), and the Global Large Historical Earthquake the 1960 m 9.6 Chile earthquake occurred, the mp 10;000 is
catalog (GLHECAT; Albini et al., 2014) by the Global Earth- about 9.7 using the generic β-value and only about 9.3 if the
quake Model (GEM) project. For each event listed in the zone-specific β-value is used. The Guam–Japan region is an-
PAGER-CAT and Centennial catalogs, multiple magnitudes other example of a large difference: the mp based on the generic
are reported, and a preferred magnitude is selected by the β-value is 0.7–0.9 magnitude units larger than that based on the
catalog authors. In the table and figure, we use the preferred zone-specific β-value for 50- to 1000-year return periods.
magnitude from the PAGER-CAT catalog and the largest re- The faulted Earth working group of the GEM project
ported magnitude from the Centennial catalog. We take the characterized geometry, coupling coefficients, upper and
former as a proxy of the median mh and the latter as a proxy lower depth, plate motions, maximum earthquake magnitudes,

by gmontalva
Table 5
Estimated mp Using Specific β-Values for the Circum-Pacific Subduction Zones
Centennial
In 50 years In 100 years In 250 years In 500 years PAGER-CAT Cat. GLHECAT
F-E ID mp σ mp σ mp σ mp σ Year mh Year mh Year mh Name
1 8.43 0.17 8.66 0.17 8.90 0.18 9.04 0.19 1964 9.2 1964 9.2 Alaska–Aleutian arc
3* 7.94 0.20 8.25 0.21 8.59 0.20 8.80 0.20 1946 7.6 1946 7.6 1700 9.0 Cascadia
5 8.19 0.22 8.38 0.21 8.56 0.22 8.67 0.22 1985 8.0 1902 8.4 1787 8.4 Mexico–Guatemala
6 7.93 0.21 8.21 0.22 8.54 0.22 8.75 0.21 1950 7.8 1904 8.3 Central America
7 7.95 0.35 8.22 0.34 8.51 0.31 8.67 0.29 1946 7.9 1900 8.4 1842 8.2 Caribbean Loop
8 8.74 0.19 8.89 0.20 9.03 0.22 9.12 0.23 1960 9.6 1960 9.6 Andean South America
12 8.08 0.14 8.33 0.15 8.66 0.17 8.90 0.18 1917 8.5 1917 8.7 Kermadec–Tonga–Samoa
14 8.39 0.20 8.49 0.22 8.59 0.23 8.65 0.24 1901 7.9 1901 8.4 New Hebrides Islands
15 8.39 0.19 8.49 0.21 8.59 0.23 8.65 0.23 1971 8.1 1906 8.4 Bismarck–Solomon Islands
16 8.47 0.23 8.75 0.24 9.08 0.23 9.28 0.22 1996 8.2 1906 8.4 New Guinea
18 7.34 0.27 7.57 0.30 7.88 0.33 8.11 0.36 1914 7.9 1902 8.1 1498 8.3 Guam–Japan
19 8.72 0.18 8.87 0.19 9.01 0.21 9.10 0.22 2011 9.0 1952 9.0 Japan–Kamchatka
20 8.11 0.44 8.41 0.43 8.75 0.40 8.95 0.37 1911 8.1 1946 8.4 1707 8.6 S.E. Japan–Ryukyu Islandsa
21 8.03 0.25 8.27 0.24 8.53 0.23 8.68 0.23 1920 7.9 1920 8.0 Taiwan
22 8.26 0.19 8.5 0.19 8.76 0.19 8.92 0.20 1918 8.2 1918 8.3 Philippines
23 8.31 0.20 8.55 0.20 8.82 0.20 8.99 0.20 1965 8.2 1939 8.6 Borneo–Celebes
24 8.53 0.17 8.77 0.17 9.03 0.18 9.18 0.19 2007 8.5 1938 8.6 1833 8.7 Sunda arc
46 8.03 0.28 8.30 0.29 8.66 0.30 8.90 0.29 2004 9.0 2004 9.0 Andaman Island Sumatra
*Results are for the redefined Cascadia region (Fig. 2).
and other parameters for all 55,000 km of the subduction to the F-E zones. We compare the GEM preferred and maxi-
zones in the world (Berryman et al., 2013). They divided mum mx , McCaffrey’s mx , and mp 1000 and mp 10;000
the subduction zones into 65 individual segments and esti- from this study (Fig. 9). The comparison reveals that:
mated minimum and maximum mx for each segment, based
• Most of the GEM preferred mx are lower than the mp 1000
on the largest observed magnitudes and tectonic moment rate,
from this study, but most of the GEM maximum mx are
respectively. They then assigned a preferred mx , which is the
between mp 1000 and mp 10;000.
average of the minimum and maximum mx . We map their seg-
• The mx values by McCaffrey (2008) are much higher than
ments into the F-E zones and read the preferred and maximum
our mp 1000 values but fairly consistent with mp 10;000.
mx values among the segments within one F-E zone. McCaf-
However, most of the recurrence times of the great earth-
frey (2008) also estimated the maximum earthquake magni-
quakes by McCaffrey (2008) are in the range of hundreds
tudes and their recurrence time for all of the subduction
to thousands of years. This implies that McCaffrey (2008)
zones in the world. We relate the values in McCaffrey (2008)
Figure 7. The estimated mp for 50-, 100-, 250- and 500-year return periods for the subduction zones. Zone-specific β-values are used for
estimating mp T values. Uncertainties (1σ; black bars) are shown only for 50- and 500-year return periods. The historical maximum
magnitudes are from the Centennial, PAGER-CAT, and GLHECAT catalogs.

by gmontalva
Table 6
Estimated mp for 50-, 100-, 250-, and 500-year Return Periods for the Circum-Pacific Subduction Zones Using the Generic β-value
Centennial
In 50 years In 100 years In 250 years In 500 years PAGER-CAT Cat. GLHECAT
F-E ID mp σ mp σ mp σ mp σ Year mh Year mh Year mh Name
1 8.43 0.17 8.66 0.17 8.90 0.18 9.04 0.19 1964 9.2 1964 9.2 Alaska–Aleutian arc
3* 7.69 0.18 8.00 0.20 8.40 0.22 8.69 0.22 1946 7.6 1946 7.6 1700 9.0 Cascadia
5 8.07 0.23 8.31 0.23 8.58 0.22 8.74 0.22 1985 8.0 1902 8.4 1787 8.4 Mexico–Guatemala
6 8.05 0.22 8.31 0.21 8.59 0.21 8.76 0.21 1950 7.8 1904 8.3 Central America
7 7.84 0.35 8.12 0.36 8.46 0.34 8.67 0.32 1946 7.9 1900 8.4 1842 8.2 Caribbean Loop
8 8.57 0.19 8.83 0.19 9.11 0.19 9.28 0.19 1960 9.6 1960 9.6 Andean South America
12 8.54 0.17 8.69 0.18 8.85 0.20 8.94 0.22 1917 8.5 1917 8.7 Kermadec–Tonga–Samoa
14 8.41 0.16 8.57 0.18 8.72 0.20 8.81 0.22 1901 7.9 1901 8.4 New Hebrides Islands
15 8.42 0.16 8.57 0.18 8.71 0.20 8.79 0.22 1971 8.1 1906 8.4 Bismarck–Solomon Islands
16 8.52 0.24 8.79 0.24 9.09 0.23 9.28 0.22 1996 8.2 1906 8.4 New Guinea
18 8.09 0.44 8.39 0.44 8.76 0.42 9.00 0.40 1914 7.9 1902 8.1 1498 8.3 Guam–Japan
19 8.70 0.17 8.88 0.18 9.07 0.19 9.18 0.21 2011 9.0 1952 9.0 Japan–Kamchatka
20 7.98 0.44 8.28 0.45 8.66 0.44 8.91 0.42 1911 8.1 1946 8.4 1707 8.6 S.E. Japan–Ryukyu Islands
21 7.99 0.25 8.25 0.24 8.52 0.23 8.68 0.23 1920 7.9 1920 8.0 Taiwan
22 8.34 0.19 8.54 0.19 8.75 0.20 8.87 0.21 1918 8.2 1918 8.3 Philippines
23 8.39 0.20 8.60 0.20 8.82 0.20 8.94 0.21 1965 8.2 1939 8.6 Borneo–Celebes
24 8.53 0.17 8.77 0.17 9.03 0.18 9.18 0.19 2007 8.5 1938 8.6 1833 8.7 Sunda arc
46 8.27 0.28 8.52 0.27 8.78 0.25 8.93 0.23 2004 9.0 2004 9.0 Andaman Island –Sumatra
*Results are for the redefined Cascadia region (Fig. 2).
estimated much higher rates for such earthquakes than place mx We develop a method to estimate mp T for each of
this study. the circum-Pacific subduction zones. In our method, we fit a
• The GEM preferred mx is extremely low for the Central TGR distribution to historical catalogs for each of the zones.
America and Guam–Japan regions. Both subduction zones The two parameters of the TGR distribution are determined
are characterized by rapid plate convergence and high rates by maximum-likelihood method and moment conservation
of seismicity but have been associated with few great principle. The rate of earthquakes for each zone, which is also
(m > 8:0) interplate earthquakes. Pacheco et al. (1993) ob- required to estimate mp T, is calculated using the Whole
tained low seismic coupling for both subduction zones Earth model (Kagan and Jackson, 2012). We use Monte-Carlo
based on the observed seismicity. The weak coupling along simulation to estimate uncertainties in the TGR parameters.
the Central America subduction zone is also supported by Our results show that most of the circum-Pacific subduc-
GPS studies (Correa-Mora et al., 2009). Nevertheless, the tion zones can generate m ≥ 8:5 earthquakes over a 250-year
estimation of coupling is far from reliable, especially based return period. Over a 500-year return period, most of the
on the short history of earthquakes, and even very low rates zones can generate m ≥ 8:8 earthquakes. Over a 10,000-year
of strain accumulation may result in a large earthquake, return period, almost all of the zones can generate m ≥ 9:0
given enough time (Goldfinger, Ikeda, et al., 2013). earthquakes. The β-value (or b-value) is an important param-
Although the statistical tests cannot reject the hypothesis eter in statistical seismology and seismic-hazard analysis.
that the estimated mp is consistent with the historical obser- The debate of universal or locally varying β-values (or b-val-
vations, the historical earthquake catalog is not long enough ues) is still ongoing. In general, the mp T determined using
to verify the mp . However, paleoseismic data can extend the universal and local β-values are similar.
earthquake record back thousands of years. The turbidite We compare our results with the largest historical earth-
sediments extracted from the deep sea channels of the quake magnitudes in each of the subduction zones. We con-
CSZ preserve about a 10,000-year earthquake history for that clude that the hypothesis that the estimated mp 100 is
subduction zone, which can be used to verify and constrain consistent with the observations since 1900 cannot be re-
the recurrence rate of great earthquakes. It would be useful to jected by statistical tests. Unfortunately, the historical earth-
collect paleoseismic data for all of the major subduction quake catalogs are too short to verify our results for longer T.
zones to obtain return times. Submarine turbidite records unearth a ∼10;000-year pa-
leoseismic history of great earthquakes along the CSZ. We
Conclusions collate the paleoseismic data and estimate the magnitudes.
We compare the magnitude–frequency distribution of the
Because the traditional methods for estimating mx are great earthquakes with our TGR models. At 95% confidence
inadequate and mx itself is ill-defined, we use mp T, prob- level, our models and the data are consistent, which gives us
able maximum magnitude within a time period of T, to re- confidence about our methodology.

by gmontalva
(a)
9.80
50-year
100-year
9.30
250-year
Magnitude
8.80 500-year
50-year
8.30
100-year
250-year
7.80
500-year
7.30
(b)
9.9
9.7
1000-year
9.5
Magnitude
10000-year
9.3
9.1 1000-year
8.9 10000-year
8.7
8.5
8.3
Figure 8. The estimated mp for different return periods for the subduction zones. The squares (connected by solid lines) are based on
zone-specific mc and β-values, and the circles (connected by dashed lines) are based on generic β-values and the corresponding mc : (a) For
50-, 100-, 250-, and 500-year return periods; and (b) for 1000- and 10,000-year return periods.
10
GEM pref. mx mp(1000) GEM max. mx mp(10,000) McCaffrey
9.8
9.6
9.4
9.2
9
8.8
8.6
8.4
8.2
Figure 9. Comparisons of mp T from this study to the mx by McCaffrey (2008) and GEM (Berryman et al., 2013). The mp T values are
based on the zone-specific β-values.

by gmontalva
McCaffrey (2008) examined worldwide subduction Atwater, B. F., A. R. Nelson, J. J. Clague, G. A. Carver, D. K. Yamaguchi, P.
zones and stated that current evidence cannot rule out that T. Bobrowsky, J. Bourgeois, M. E. Darienzo, W. C. Grant, and E.
Hemphill-Haley et al. (1995). Summary of coastal geologic evidence
any subduction zone may produce a magnitude 9 or larger for past great earthquakes at the Cascadia subduction zone, Earthq.
earthquake. Our study demonstrates that m ∼ 9 earthquakes Spectra 11, 1–18.
are expected along any of the circum-Pacific subduction Bell, A. F., M. Naylor, and I. G. Main (2013). Convergence of the frequency-
zones given a sufficient period of time (∼10;000 years). size distribution of global earthquakes, Geophys. Res. Lett. 40, 2585–
2589, doi: 10.1002/grl.50416.
Bender, B. (1983). Maximum likelihood estimation of b values for magni-
Data and Resources tude grouped data, Bull. Seismol. Soc. Am. 73, 831–851.
Berryman, K., L. Wallace, G. Hayes, P. Bird, K. Wang, R. Basili, T. Lay,
The Global Centroid Moment Tensor catalog (Dziewon- R. Stein, T. Sagiya, C. Rubin, S. Barreintos, C. Kreemer, N. Litchfield,
ski et al., 1981; Ekström et al., 2005, 2012) is available at M. Pagani, K. Gledhill, K. Haller, and C. Costa (2013). The GEM
http://www.globalcmt.org/CMTfiles.html. The Centennial Faulted Earth Subduction Characterisation Project, Version 1.0, June
catalog (Engdahl and Villaseñor, 2002) is available at 2013, GEM Faulted Earth Project, available at http://www.nexus
.globalquakemodel.org/gem‑faulted‑earth/posts (last accessed October
http://earthquake.usgs.gov/data/centennial/. The PAGER-CAT 2013).
earthquake catalog (Allen et al., 2009) is available at http:// Bird, P., and Y. Y. Kagan (2004). Plate-tectonic analysis of shallow seismic-
earthquake.usgs.gov/research/pager/data/. The PDE catalog ity: Apparent boundary width, beta, corner magnitude, coupled litho-
(National Earthquake Information Center [NEIC], 1970; Sip- sphere thickness, and coupling in seven tectonic settings, Bull.
kin et al., 2000) is available at ftp://hazards.cr.usgs.gov/pde/ or Seismol. Soc. Am. 94, no. 6, 2380–2399.
Bird, P., and Z. Liu (2007). Seismic hazard inferred from tectonics: Califor-
as an online search at http://neic.usgs.gov/neis/epic/epic_ nia, Seismol. Res. Lett. 78, no. 1, 37–48.
global.html. Flinn–Engdahl regions are explained and their Bird, P., C. Kreemer, and W. E. Holt (2010). A long-term forecast of shallow
coordinates as well as FORTRAN files to process them are seismicity based on the Global Strain Rate Map, Seismol. Res. Lett. 81,
available at http://earthquake.usgs.gov/learn/topics/flinn_ no. 2, 184–194, doi: 10.1785/gssrl.81.2.184.
engdahl_list.php. All the electronic addresses referenced here Chapman, J. S., and T. I. Melbourne (2009). Future Cascadia megathrust
rupture delineated by episodic tremor and slip, Geophys. Res. Lett.
were last accessed August 2014. 36, doi: 10.1029/2009GL040465.
Clarke, S. H., and G. A. Carver (1992). Breadth of interplate coupling in the
Acknowledgments southern Cascadia subduction zone: Implication for earthquake mag-
nitudes, Geol. Soc. Am. Abs. Prog. 24, 15.
The authors are grateful for the constructive comments from FM Correa-Mora, F., C. DeMets, D. Alvarado, H. L. Turner, G. Mattioli,
Global colleagues: Hosam Ali, Franco Tamanini, and Lou Gritzo. We thank D. Hernandez, C. Pullinger, M. Rodriguez, and C. Tenorio (2009).
Yan Kagan and Peter Bird from the University of California, Los Angeles GPS-derived coupling estimates for the Central America subduction
(UCLA), and Ross Stein from U.S. Geological Survey (USGS) for useful zone and volcanic arc faults: El Salvador, Honduras and Nicaragua,
discussions. We also thank Kelvin Berryman from GNS Science of New Geophys. J. Int. 179, 1279–1291. doi: 10.1111/j.1365-246X.2009.
Zealand, Marco Pagani from Global Earthquake Model (GEM), and Paola 04371.x.
Albini from Istituto Nazionale di Geofisica e Vulcanologia (INGV) of Italy Dziewonski, A. M., T. A. Chou, and J. H. Woodhouse (1981). Determination
for generously sharing data and results. We appreciate the USGS John Wes- of earthquake source parameters from waveform data for studies of
ley Powell Center for Analysis and Synthesis for providing us opportunities global and regional seismicity, J. Geophys. Res. 86, 2825–2852.
to discuss the project at meetings supported by the Center. We are grateful to Earthquake Research Committee (2005). National Seismic Hazard Maps for
Associate Editor Matt Gerstenberger and two anonymous reviewers for their Japan, available at http://www.jishin.go.jp/main/index‑e.html (last ac-
reviews and comments, which significantly improved the manuscript. This cessed October 2013).
work was funded by FM Global. Ekström, G., A. M. Dziewonski, N. N. Maternovskaya, and M. Nettles
(2005). Global seismicity of 2003: Centroid-moment-tensor solutions
for 1087 earthquakes, Phys. Earth Planet. Int. 148, 327–351.
References
Ekström, G., M. Nettles, and A. M. Dziewonski (2012). The global CMT
Adams, J. (1990). Paleoseismicity of the Cascadia subduction zone: Evi- project 2004–2010: Centroid-moment tensors for 13017 earthquakes,
dence from turbidites off the Oregon-Washington margin, Tectonics Phys. Earth Planet. Int. 200/201, 1–9.
9, 569–583, doi: 10.1029/TC009i004p00569. Engdahl, E. R., and A. Villaseñor (2002). Global seismicity: 1900–1999, in
Aki, K. (1965). Maximum likelihood estimate of b in the formula logN International Handbook of Earthquake Engineering and Seismology,
a–bM and its confidence limits, Bull. Earthq. Res. Inst. Tokyo Univ. W. K. Lee, H. Kanamori, P. C. Jennings, and C. Kisslinger (Editors),
43, 237–239. Academic Press, Amsterdam, Netherlands, 665–690.
Albini, P., R. M. W. Musson, A. Rovida, M. Locati, A. A. Gomez Capera, Enkin, R. J., A. Dallimore, J. Baker, J. R. Southon, and T. Ivanochko (2013).
and D. Viganò (2014). The global earthquake history, Earthq. Spectra A new high-resolution radiocarbon Bayesian age model of the
30, 607–624. Holocene and Late Pleistocene from core MD02-2494 and others,
Allen, T. I., K. Marano, P. S. Earle, and D. J. Wald (2009). PAGER-CAT: A Effingham Inlet, British Columbia, Canada; with an application to the
composite earthquake catalog for calibrating global fatality models, paleoseismic event chronology of the Cascadia Subduction Zone, Can.
Seismol. Res. Lett. 80, no. 1, 50–56. J. Earth Sci. 50, 746–760.
Atwater, B. F. (1987). Evidence for great Holocene earthquakes along the Flinn, E. A., and E. R. Engdahl (1965). A proposed basis for geographical
outer coast of Washington State, Science 236, 942–944. and seismic regionalization, Rev. Geophys. 3, 123–149.
Atwater, B. F., and G. B. Griggs (2012). Deep-sea turbidites as guides to Flinn, E. A., E. R. Engdahl, and A. R. Hill (1974). Seismic and geographical
Holocene earthquake history at the Cascadia Subduction Zone— regionalization, Bull. Seismol. Soc. Am. 64, 771–993.
Alternative views for a seismic-hazard workshop, U.S. Geol. Surv. Flück, P., R. D. Hyndman, and K. Wang (1997). 3-D dislocation model for
Open-File Rept. 2012-1043, 58 pp., available at http://pubs.usgs great earthquakes of the Cascadia subduction zone, J. Geophys. Res.
.gov/of/2012/1043/ (last accessed October 2013). 102, 20,539–20,550.

by gmontalva
Frankel, A., C. Mueller, T. Barnhard, D. Perkins, E. Leyendecker, N. Dickman, Kagan, Y. Y. (2002c). Aftershock zone scaling, Bull. Seismol. Soc. Am. 92,
S. Hanson, and M. Hopper (1996). National Seismic Hazard Maps— 641–655.
Documentation, U.S. Geol. Surv. Open-File Rept. 96-532, 110 pp. Kagan, Y. Y., and D. D. Jackson (2012). Whole Earth high-resolution earth-
Frankel, A. D., and M. D. Petersen (2013). Appendix P: Models of Earthquake quake forecasts, Geophys. J. Int. 190, no. 1, 677–686, doi: 10.1111/
Recurrence and Down-Dip Edge of Rupture for the Cascadia Subduction j.1365-246X.2012.05521.x.
Zone, http://pubs.usgs.gov/of/2013/1165/ (last accessed August 2014). Kagan, Y. Y., and D. D. Jackson (2013). Tohoku earthquake: A surprise?
Frankel, A. D., M. D. Petersen, C. S. Mueller, K. M. Haller, R. L. Wheeler, Bull. Seismol. Soc. Am. 103, 1181–1194, doi: 10.1785/0120120110.
E. V. Leyendecker, R. L. Wesson, S. C. Harmsen, C. H. Cramer, D. M. Kagan, Y. Y., and F. Schoenberg (2001). Estimation of the upper cutoff
Perkins, and K. S. Rukstales (2002). Documentation for the 2002 up- parameter for the tapered Pareto distribution, J. Appl. Probab. 38A,
date of the National Seismic Hazard Maps, U.S. Geol. Surv. Open-File 158–175.
Rept. 2002-420, 39 pp. Kanamori, H. (2006). Lessons from the 2004 Sumatra–Andaman earth-
Furlong, K. P., T. Lay, and C. J. Ammon (2009). A great earthquake rupture quake, Phil. Trans. Roy. Soc. A. 364, 1927–1945, doi: 10.1098/
across a rapidly evolving three-plate boundary, Science 324, 226–229, rsta.2006.1806.
doi: 10.1126/science.1167476. Knopoff, L. (2000). The magnitude distribution of declustered earthquakes
Gaull, B. A., M. O. Michael-Leiba, and J. M. W. Rynn (1990). Probabilistic in southern California, Proc. Natl. Acad. Sci. Unit. States. Am. 97,
earthquake risk maps of Australia, Aust. J. Earth Sci. 37, 169–187. 11,880–11,884.
Goldfinger, C., J. Beeson, H. Nelson, J. R. Patton, A. Morey, and S. Galer Kreemer, C., W. E. Holt, and A. J. Haines (2003). An integrated global
(2013). Cascadia seismoturbidities: A landlubber critiqued (Abstract), model of present-day plate motions and plate boundary deformation,
2013 Fall Meeting, AGU, San Francisco, California, 9–13 December. Geophys. J. Int. 154, 8–34.
Goldfinger, C., Y. Ikeda, R. S. Yeats, and J. Ren (2013). Superquakes and Mazzotti, S., H. Dragert, J. Henton, M. Schmidt, R. Hyndman, T. James,
supercycles, Seismol. Res. Lett. 84, 24–32, doi: 10.1785/0220110135. Y. Lu, and M. Craymer (2003). Current tectonics of northern Cascadia
Goldfinger, C., L. D. Kulm, R. S. Yeats, B. Appelgate, M. MacKay, and G. F. from a decade of GPS measurements, J. Geophys. Res. 108, no. B12,
Moore (1992). Transverse structural trends along the Oregon conver- doi: 10.1029/2003JB002653.
gent margin: Implications for Cascadia earthquake potential, Geology McCaffrey, R. (2008). Global frequency of magnitude 9 earthquakes,
20, 141–144. Geology 36, 263–266.
Goldfinger, C., L D. Kulm, R. S. Yeats, L. C. McNeill, and C. Hummon McCaffrey, R., R. W. King, S. J. Payne, and M. Lancaster (2013). Active
(1997). Oblique strike-slip faulting of the central Cascadia submarine tectonics of northwestern US inferred from GPS-derived surface veloc-
forearc, J. Geophys. Res. 102, 8217–8243. ities, J. Geophys. Res. 118, 709–723, doi: 10.1029/2012JB009473.
Goldfinger, C., A. E. Morey, B. Black, J. Beeson, C. H. Nelson, and J. Patton McCaffrey, R., M. D. Long, C. Goldfinger, P. Zwick, J. Nabelek, and
(2013). Spatially limited mud turbidites on the Cascadia margin: C. Smith (2000). Rotation and plate locking at the southern Cascadia
Segmented earthquake ruptures? Nat. Hazards Earth Syst. Sci. 13, subduction zone, Geophys. Res. Lett. 27, 3117–3120.
2109–2146, doi: 10.5194/nhess-13-2109-2013. McCaffrey, R., A. I. Qamar, R. W. King, R. Wells, G. Khazaradze,
Goldfinger, C., A. E. Morey, C. H. Nelson, J. Gutierrez-Pastor, J. E. Johnson, C. A. Williams, C. W. Stevens, J. J. Vollick, and P. C. Zwick (2007).
E. Karabanov, J. Chaytor, and A. Ericsson (2007). Rupture lengths and Fault locking, block rotation and crustal deformation in the Pacific
temporal history of significant earthquakes on the offshore and north northwest, Geophys. J. Int. 169, 1,315–1,340, doi: 10.1111/j.1365-
coast segments of the northern San Andreas fault based on turbidite 246X.2007.03371.x.
stratigraphy, Earth Planet. Sci. Lett. 254, 9–27. Minoura, K., F. Imamura, D. Sugawara, Y. Kono, and T. Iwashita (2001).
Goldfinger, C., C. H. Nelson, A. E. Morey, J. E. Johnson, J. R. Patton, E. The 869 Jogan tsunami deposit and recurrence interval of large-scale
Karabanov, J. Gutiérrez-Pastor, A. T. Eriksson, E. Gràcia, G. Dunhill, tsunami on the Pacific coast of northeast Japan, J. Nat. Disast. Sci. 23,
R. J. Enkin, A. Dallimore, and T. Vallier (2012). Turbidite event history no. 2, 83–88.
—Methods and implications for Holocene paleoseismicity of the Cas- Mitchell, C. E., P. Vincent, R. J. I. Weldon, and M. A. Richards (1994).
cadia subduction zone, U.S. Geol. Surv. Profess. Pap. 1661-F, 170 pp. Present-day vertical deformation of the Cascadia margin, Pacific north-
Gomberg, J., and Cascadia 2007, and Beyond Working Group (2010). Slow- west, U.S.A., J. Geophys. Res. 99, 12,257–12,277.
slip phenomena in Cascadia from 2007 and beyond: A review, GSA Musson, R. M. W. (2000). Generalised seismic hazard maps for the Panno-
Bulletin 122, 963–978, doi: 10.1130/B30287.1. nian basin using probabilistic methods, Pure Appl. Geophys. 157,
Hanks, T. C., and H. Kanamori (1979). A moment magnitude scale, J. Geo- nos. 1/2, 147–169.
phys. Res. 84, 2348–2350, doi: 10.1029/JB084iB05p02348. National Earthquake Information Center (NEIC) (1970). Preliminary deter-
Hayes, G. P., D. J. Wald, and R. L. Johnson (2012). Slab1.0: A three-dimen- mination of epicenters monthly listing, U.S. Department of the
sional model of global subduction zone geometries, J. Geophys. Res. Interior, 8 pp.
117, no. B01302, doi: 10.1029/2011JB008524. Nelson, A. R., H. M. Kelsey, and R. C. Witter (2006). Great earthquakes of
Holschneider, M., G. Zöller, and S. Hainzl (2011). Estimation of the maxi- variable magnitude at the Cascadia subduction zone, Quaternary Res.
mum possible magnitude in the framework of a doubly truncated 65, 354–365.
Gutenberg–Richter model, Bull. Seismol. Soc. Am. 101, 1649–1659, Oppenheimer, D., J. Eaton, A. Jayko, M. Lisowski, G. Marshall, M. Murray,
doi: 10.1785/0120100289. R. Simpson, R. Stein, G. Beroza, and M. Magee et al. (1993). The
Hyndman, R. D., and K. Wang (1995). The rupture zone of Cascadia great Cape Mendocino, California, earthquakes of April, 1992: Subduction
earthquakes from current deformation and the thermal regime, J. Geo- at the triple junction, Science 261, 433–438.
phys. Res. 100, 22,133–22,154. Pacheco, J. F., L. R. Sykes, and C. H. Scholz (1993). Nature of seismic
Jackson, D. D., and M. Matsu’ura (1985). A Bayesian approach to nonlinear coupling along simple plate boundaries of the subduction type, J. Geo-
inversion, J. Geophys. Res. 90, 581–591. phys. Res. 98, no. B8, 14,133–14159, doi: 10.1029/93JB00349.
Kagan, Y. Y. (1997). Seismic moment-frequency relation for shallow earth- Petersen, M. D., A. D. Frankel, S. C. Harmsen, C. S. Mueller, K. M.
quakes: Regional comparison, J. Geophys. Res. 102, 2835–2852. Haller, R. L. Wheeler, R. L. Wesson, Y. Zeng, O. S. Boyd, D. M.
Kagan, Y. Y. (1999). Universality of the seismic moment-frequency relation, Perkins, N. Luco, E. H. Field, C. J. Wills, and K. S. Rukstales
Pure Appl. Geophys. 155, 537–573. (2008). Documentation for the 2008 Update of the United States
Kagan, Y. Y. (2002a). Seismic moment distribution revisited: I. Statistical National Seismic Hazard Maps, U.S. Geol. Surv. Open-File Rept.
results, Geophys. J. Int. 148, no. 3, 520–541. 2008-1128, 61 pp.
Kagan, Y. Y. (2002b). Seismic moment distribution revisited: II. Moment Pisarenko, V. F., and D. Sornette (2004). Statistical methods of parameter
conservation principle, Geophys. J. Int. 149, no. 3, 731–754. estimation for deterministically chaotic time series, Phys. Rev. E 69,

by gmontalva
036122, available at http://arXiv.org/abs/physics/0308059 (last ac- Utsu, T. (1999). Representation and analysis of the earthquake size distri-
cessed October 2013). bution: A historical review and some new approaches, Pure Appl. Geo-
Pisarenko, V. F., A. Sornette, D. Sornette, and M. V. Rodkin (2008). phys. 155, 509–535.
New approach to the characterization of Mmax and of the tail of the Vilanova, S. P., and J. F. B. D. Fonseca (2007). Probabilistic seismic-hazard
distribution of earthquake magnitudes, Pure Appl. Geophys. 165, assessment for Portugal, Bull. Seismol. Soc. Am. 97, 1702–1717.
847–888. Wang, K., R. Wells, S. Mazzotti, R. D. Hyndman, and T. Sagiya (2003). A
Priest, G. R., C. Goldfinger, K. Wang, R. C. Witter, Y. Zhang, and A. M. revised dislocation model of interseismic deformation of the Cascadia
Baptista (2009). Confidence levels for tsunami-inundation limits in subduction zone, J. Geophys. Res. 108, no. B1, 2026, doi: 10.1029/
northern Oregon inferred from a 10,000-year history of great earth- 2001JB001227.
quakes at the Cascadia subduction zone, Nat. Hazards 54, no. 1, Wang, L., C. K. Shum, F. J. Simons, A. Tassara, K. Erkan, C. Jekeli, A.
27–73, doi: 10.1007/s11069-009-9453-5. Braun, C. Kuo, H. Lee, and D. Yuan (2012). Coseismic slip of the
Reiter, L. (1990). Earthquake Hazard Analysis: Issues and Insights, Colum- 2010 M w 8.8 great Maule, Chile, earthquake quantified by the inver-
bia University Press, New York, New York, 254 pp. sion of grace observations, Earth Planet. Sci. Lett. 335, 167–179.
Rong, Y., D. D. Jackson, and Y. Y. Kagan (2003). Seismic gaps and earth- Wech, A. G., K. C. Creager, and T. I. Melbourne (2009). Seismic and geodetic
quakes, J. Geophys. Res. 108, doi: 10.1029/2002JB002334. constraints on Cascadia slow-slip, J. Geophys. Res. 114, no. B10316,
Ruff, L., and H. Kanamori (1980). Seismicity and the subduction process, doi: 10.1029/2008JB006090.
Phys. Earth Planet. Int. 23, no. 3, 240–252, doi: 10.1016/0031-9201 Witter, R. C., Y. Zhang, K. Wang, C. Goldfinger, G. R. Priest, and J. C. Allan
(80)90117-X. (2012). Coseismic slip on the southern Cascadia megathrust implied by
Satake, K., K. Shimazaki, Y. Tsuji, and K. Ueda (1996). Time and size of a tsunami deposits in an Oregon lake and earthquake-triggered marine tur-
giant earthquake in Cascadia inferred from Japanese tsunami records bidites, J. Geophys. Res. 117, no. B10303, doi: 10.1029/2012JB009404.
of January, 1700, Nature 379, 246–249. Yoshida, Y., H. Ueno, D. Muto, and A. Aoki (2011). Source process of the
Satake, K., K. Wang, and B. F. Atwater (2003). Fault slip and seismic mo- 2011 off the Pacific coast of Tohoku Earthquake with the combination
ment of the 1700 Cascadia earthquake inferred from Japanese tsunami of teleseismic and strong motion data, Earth Planets Space 63, no. 7,
descriptions, J. Geophys. Res. 108, no. B11, 2535, doi: 10.1029/ 565–569.
2003JB002521. Zöller, G. (2013). Convergence of the frequency–magnitude distribution of
Savage, J. C., and M. Lisowski (1991). Strain measurements and potential global earthquakes: Maybe in 200 years, Geophys. Res. Lett. 40, 3873–
for a great subduction earthquake off the coast of Washington, Science 3877, doi: 10.1002/grl.50779.
252, 101–103. Zöller, G., M. Holschneider, and S. Hainzl (2013). The maximum earth-
Scholz, C. H., and J. Campos (2012). The seismic coupling of subduction quake magnitude in a time horizon: Theory and case studies, Bull.
zones revisited, J. Geophys. Res. 117, no. B05310, doi: 10.1029/ Seismol. Soc. Am. 101, 860–875, doi: 10.1785/0120120013.
2011JB009003.
Schorlemmer, D., S. Wiemer, and M. Wyss (2005). Variations in earthquake-
size distribution across different stress regimes, Nature 437, doi:
Center for Property Risk Solutions
10.1038/nature04094.
FM Global Research Division
Schwartz, S. Y., and J. M. Rokosky (2007). Slow slip events and seismic 1151 Boston-Providence Highway
tremor at circum-Pacific subduction zones, Rev. Geophys. 45, doi: Norwood, Massachusetts 02062
10.1029/2006RG000208. (Y.R., H.M.)
Secanell, R., D. Bertil, C. Martin, X. Goula, T. Susagna, M. Tapia, P. Dominique,
D. Carbon, and J. Fleta (2008). Probabilistic seismic hazard assessment of
the Pyrenean region, J. Seismol. 12, 323–341.
Department of Earth and Space Sciences
Sipkin, S. A., W. J. Person, and B. W. Presgrave (2000). Earthquake bulle-
University of California
tins and catalogs at the USGS National Earthquake Information
Los Angeles, California 90095
Center, IRIS Newsletter 2000, 2–4. (D.D.J.)
Stein, S., and E. Q. Okal (2011). The size of the 2011 Tohoku earthquake
need not have been a surprise, Eos Trans. AGU 92, no. 27, doi:
10.1029/2011EO270005.
Strasser, F. O., M. C. Arango, and J. J. Bommer (2010). Scaling of the source College of Earth, Ocean and Atmospheric Sciences
dimensions of interface and intraslab subduction-zone earthquakes Oregon State University
with moment magnitude, Seismol. Res. Lett. 81, no. 6, 941–950. Corvallis, Oregon 97331
(C.G.)
Sugawara, D., K. Goto, F. Imamura, H. Matsumoto, and K. Minoura (2012).
Assessing the magnitude of the 869 Jogan tsunami using sedimentary
deposits: Prediction and consequence of the 2011 Tohoku-oki, Manuscript received 7 November 2013;
Sediment. Geol. 282, 14–26. Published Online 16 September 2014

by gmontalva

Magnitude Limits of Subduction Zone Earthquakes

Uploaded by

Copyright:

Available Formats

Magnitude Limits of Subduction Zone Earthquakes

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

Magnitude Limits of Subduction Zone Earthquakes

Uploaded by

Copyright:

Available Formats

Bulletin of the Seismological Society of America, Vol. 104, No. 5, pp. 2359–2377, October 2014, doi: 10.

Magnitude Limits of Subduction Zone Earthquakes

Abstract Maximum earthquake magnitude (mx ) is a critical parameter in seismic

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

address the deviations from the GR relationship at the tail of Table 1

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

Sources: Esri, USGS, NOAA

Slab 1.0, Depth (km)

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

1 Alaska-Aleutian arc 280 0.65 0.04 9.01 0.28 9.01

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

130° W 125° W 120° W 115° W

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

1 250 A 1000 83 9.00 9.15 9.00 9.00

*From Goldfinger et al. (2012).

Magnitude–Frequency Distributions for the earthquake occurrence above a lower-threshold magnitude in

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

(a) 9 10,000-year medium-to-large event set (with magnitudes

7.5 likelihood mc is 8.84 with a 95% confidence range of 8.6 and

7 0 −3 −6 to solve for magnitude limits for a region.

9.6 hmi − mt − Δm2bin

+3 in which Δmbin is the magnitude binning width of the catalog,

8.6 0 mean magnitude of the earthquakes equal to or larger than mt

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

Cumulative Annual Rate of Occurence

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

closely spaced in time. In fact, event T5 is such a candidate Results

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

*Results are for the redefined Cascadia region (Fig. 2).

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

*Results are for the redefined Cascadia region (Fig. 2).

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

Downloaded from http://pubs.geoscienceworld.org/ssa/bssa/article-pdf/104/5/2359/2678433/2359.pdf

You might also like

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