Large Sample Basin Experiments for Hydrological Model Parameterization: Results of the Model Parameter

Experiment—MOPEX. IAHS Publ. 307, 2006. 149

Flow simulation in an ungauged basin: an

alternative approach to parameterization of a
conceptual model using regional data

MONOMOY GOSWAMI & KIERAN MICHAEL O’CONNOR

Department of Engineering Hydrology, National University of Ireland, Galway, Ireland
kieran.oconnor@nuigalway.ie

Abstract An alternative approach to parameterization of the conceptual Soil

Moisture Accounting and Routing (SMAR) model for continuous simulation
of flow in an ungauged basin, using daily discharge data from neighbouring
gauged catchments, is presented. The methods are suited for situations where
adequate data on soil, geology, land use, etc., which are required for relating
model parameters to catchment-specific hydrometeorological and physio-
graphical characteristics, are not available. Four methods, namely: regional
averaging of data, regional pooling of data, transposition of data, and averag-
ing the parameters of the model calibrated for different gauged catchments,
are applied to a group of 12 catchments in France. The concept of a pseudo-
ungauged basin is used whereby each catchment is first considered as
ungauged for the purpose of flow estimation, and only subsequently consid-
ered as gauged to evaluate the performance of the flow estimation procedures.
Performance evaluation using the Nash-Sutcliffe efficiency index shows that
pooling of data is the best in this approach.
Key words conceptual model; flow simulation; rainfall–runoff; regional analysis;
ungauged basin

INTRODUCTION AND OBJECTIVE

Parameters of a conceptual model, when calibrated for a catchment, are generally

regarded as a potential source of information about the physiographic and hydromet-
eorological characteristics of that catchment. Explicit relations between the descriptors
of relevant features of the catchment and perceptibly predominant processes of the
hydrometorological systems on one hand, and the calibrated values of parameters on
the other, are often difficult to prescribe. Yet efforts are made to link the parameters of
the model with some descriptors of catchment characteristics and hydrometeorological
behaviour using physically plausible and usually empirical relations. This assumed
capability of producing meaningful relationships for transferring information from the
gauged to the ungauged basin is often considered a prospective tool for application in
ungauged basins for rainfall–runoff simulation. However, in many situations, partic-
ularly where an ungauged catchment is located in the remote headwater region of a
river, adequate and reliable data about soil, geology, land use, land cover, etc., may not
be available. Although important topographical and hydrometeorological data can
generally be obtained with some effort, lack of complete data sets representative of all
major hydrological processes, and consequent difficulty in relating, empirically or
otherwise, the parameters of a conceptual model to the catchment-characteristics
150 Monomoy Goswami & Kieran Michael O’Connor

greatly limit the application of the conceptual model for such purposes. In such
situations, however, it may not be difficult to collect hydrometeorological data from a
number of gauged catchments from a homogeneous region to which the ungauged
catchment under study may be considered to belong. In this context, the term “region”
may not be restricted to geographical proximity. Judging on the nature of the response
behaviour to rainfall inputs, and broad hydrometeorological conditions, contiguous or
“local” catchments may also be included in the “region”. An approach of regionaliza-
tion for evaluation of parameter values of the conceptual model for subsequent use in
ungauged catchments in such cases may be useful. Three methods involving
regionalization of discharge data of the gauged catchments, and one method in which
parameters calibrated for each of the gauged catchments are combined are presented in
this study. The physically-inspired lumped conceptual Soil Moisture Accounting and
Routing (SMAR) model has been used for transferring the values of the parameters
from gauged basins to the ungauged basin for this purpose.
Seven years of continuous daily hydrometeorological data from 12 French catchments
were generously provided by Météo France and the Direction de l’Eau, through Dr
Vazken Andréassian, of Cemagref, Paris, for application in the MOPEX (Model Parameter
Estimation Experiment) research project, and made available to the present authors for
their contribution to the 2004 MOPEX Workshop held in Paris. The Nash-Sutcliffe
efficiency index (R2) is used for assessing the performance of the SMAR model. Results
are presented, and conclusions are drawn on the efficacy of the procedure.

METHODOLOGY AND REGIONALIZATION STRATEGIES

Data gaps in discharge series of six catchments in the sub-group are first filled
synthetically by using rainfall, evaporation and available discharge records with the
SMAR model in an iterative scheme. With the starting values of –9.99 for the missing
data, the model is calibrated, and the iterations continued until the model performance
in two successive iterations converged, the discharge data used for the gaps in each
iteration being the corresponding estimates obtained in the previous iteration.
For assessing relative performance of the methods in the regionalization approach,
each of the 12 gauged catchments considered in the region is used initially in turn as if
it was ungauged, but subsequently, after simulation of its flows from the regional
analysis, as gauged (using its measured flow data) for the purpose of evaluating the
simulation efficiency of the procedure of flow simulation in an ungauged catchment in
the region. The catchment, thus considered, is called “pseudo-ungauged” in this study.
With the underlying assumption that regionalization reflects the general
characteristics of the region, which are also representative of the ungauged basin in the
region, four regionalization methods are applied for flow estimation in pseudo-
ungauged catchments. The methods are described below.

Regional averaging of data

In this method, the naïve “no-model” discharge estimate of a pseudo-ungauged

(equally applicable to ungauged) catchment is obtained as the average of the
 Flow simulation in an ungauged basin 151

synchronous discharges of N–1 number of gauged catchments excluding the pseudo-

ungauged catchment in the selected region comprising N catchments. The average
discharge series, thus generated, is used for rainfall–runoff simulation by calibration of
the SMAR model. The rainfall and the evaporation data series, as required by a model,
are those observed for the pseudo-ungauged catchment, i.e. it is assumed that rainfall
and evaporation data are available for the pseudo-ungauged catchment and ultimately
for any ungauged catchment in the region.

Regional pooling of data

In this method, the observed hydrological data series are combined by putting the m
years of data from all gauged catchments (N–1 in a group of N) in a region, in series,
end to end, and appending the data of the pseudo-ungauged (Nth) catchment as the last
one, thereby making N × m years of data in all. Hydrological models are then
calibrated by maximizing the combined R2 value over the calibration period using the
corresponding combined (end to end) input to the N–1 gauged catchments in a region
as inputs to the “regional model” to simulate flows in this calibration period. Finally,
considering the entire data set of the Nth pseudo-ungauged catchment as that belonging
to the “validation” period of the “regional model”, and using the rainfall and
evaporation inputs to that catchment as inputs to the model, the discharge series for the
pseudo-ungauged catchment is estimated.

Transposition of data

This “nearest neighbour” approach is used when very few catchments in a homogeneous
region in the neighbourhood of an ungauged catchment are gauged. The flow data series
of a gauged catchment measured in volume of flow per unit time, i.e. m3 s-1, are scaled
up or down in the proportion of catchment areas depending on whether the ungauged
(pseudo-ungauged in this study) catchment is larger or smaller in area than the gauged
catchment considered. Noting that for the catchments used in the present study the
rainfall values within each subgroup are similar, the mean annual rainfall data of the
gauged and the ungauged catchments are not used for additional scaling. By scaling of
flow rates from the gauged catchments by area, it is thus assumed that for all catchments
in a homogeneous region, the flow depth over the catchment during any given time
interval is uniform. Taking the rainfall and the evaporation data series as those observed
for the ungauged catchment, and the flow data series as that obtained by scaling of the
data series of the nearest-neighbour gauged catchment, i.e. of the “index basin”, the
SMAR model is calibrated. For validation, the output of the model is compared with the
corresponding measured discharge of the pseudo-ungauged catchment.

Regional averaging of parameters

In this method, unlike the previous three, the SMAR model is first applied to N–1
gauged catchments individually in a group of N catchments in the region obtaining the
152 Monomoy Goswami & Kieran Michael O’Connor

best possible fit for each catchment. The weighted average of N–1 values of each
parameter from the N–1 parameter sets is obtained. The weights used in this study are
based on the R2 efficiency values, which reflect the degree of fit of the model. These
weighted-average parameters are then applied, without further calibration of the SMAR
model, to the Nth catchment which is considered pseudo-ungauged, and the simulated
discharge is compared with the discharge observed at the pseudo-ungauged catchment.
For simplicity, despite its shortcomings (Kachroo & Natale, 1992), only the
dimensionless efficiency index R2 (Nash & Sutcliffe, 1970) is used in this study for
judging the relative performance of the SMAR model while using different methods of
regionalization. Whereas R2 = 100% would denote an ideal or “perfect” fit, it is
generally agreed that R2 > 90% is indicative of a very good model fit, while that in the
range of 80–90% is a fairly good fit, and a range of 60–80% is unsatisfactory. For
consistency, a “warm-up” period corresponding to first year’s data is not considered
for performance evaluation in calibration as well as in validation in all four methods.

ASSESSMENT OF REGIONAL HOMOGENIETY

Assessment of regional homogeneity is very important for reducing the uncertainty of

estimation of flow in an ungauged catchment by regional analysis. In the group of 12
catchments, the three in the northwest are in the humid seaboard climatic zone, three in
the southeast are in the Mediterranean zone, one in the northeast is in the semi-continen-
tal zone, and the remaining five are characterized by an intermediate climatic zone.
A1522020 in the east is the wettest and H3613020 is the driest. The southeastern
catchments, namely, V6035010, Y3514020 and Y5615030 have significant evapora-
tion, with evaporation exceeding rainfall for almost 80% of the time. The northwestern
catchments, namely, J2034010, J3024010 and J4124420 constitute a hydrometeoro-
logically homogeneous region with very little variability in hydrometeorological data
values. The contributing catchment at station K0744010 is contained within that of
station K0753210, both stations being located on the same river. These two catchments
are therefore considered as being hydrometeorologically homogenous for the purpose
of this study.
Topographically, the mean altitude, the altitude at the highest point, and the
altitude at the outlet of the three catchments in the west are of the same order of
magnitude, whereas Y3514020 in the sub-group of three catchments in the southeast is
located at a much lower altitude in comparison with V6035010 and Y5615030 in that
sub-group. A1522020 is located at a significantly high altitude in the northeast France.
Among the three catchments in the central region, H5723011 and H3613020 are
located at nearly the same altitude, whereas H2001020 is at a higher altitude.
Characteristics of the hydrological data of all 12 catchments are presented in Table 1.
It is seen from the table that the rainfall in three southeastern catchments, namely,
V6035010, Y3514020 and Y5615030, exceeds the evaporation for only about 15–20%
of the time. Although conscious that just seven years of data is hardly sufficient to
adequately characterise the rainfall–runoff process in these drier catchments, this
number was adopted for uniformity in application, the same number being used in the
analysis for consistency with the lengths of data available for the other catchments in
 Flow simulation in an ungauged basin 153

Table 1 Characteristics of hydrological data (1 August 1995–31 July 2002) 2557 data values.
Station code % days R>E Rmean – Qmean Rmean − Qmean E mean Rmean – Emean
no. (mm day-1) Emean Rmean (mm day-1)
J2034010 37.3 1.74 0.90 0.74 0.69
J3024010 38.7 1.33 0.71 0.68 0.90
J4124420 38.1 1.95 0.99 0.58 1.41
A1522020 40.0 2.52 1.25 0.44 2.55
H5723011 30.7 1.54 0.75 0.94 0.14
H3613020 31.6 1.87 0.93 0.91 0.20
H2001020 37.0 1.18 0.58 0.57 1.53
K0744010 34.1 1.66 0.83 0.74 0.69
K0753210 34.6 1.56 0.78 0.72 0.78
V6035010 20.1 2.07 0.70 1.02 –0.06
Y3514020 15.4 1.61 0.51 1.37 –0.86
Y5615030 17.1 1.59 0.52 0.97 0.10

the sample. It is also observed that the ratio of mean values of evaporation to rainfall is
significantly greater than unity (at 1.37) for the Y3514020 catchment in the southeast,
very close to unity for the two neighbouring catchments, i.e. V6035010 and
Y5615030, and less than unity for the remaining nine catchments. These statistics,
while indicating higher evaporation levels in the three catchments in the south-east of
the country, also suggest that the catchment Y3514020 is an outlier within the group of
12 catchments. As a further indication of the heterogeneity effect of Y3514020, the
mean evaporation for this catchment is seen to be higher than the mean rainfall by
0.86 mm day-1, whereas it is either less than or very near to the mean rainfall in the
case of the other eleven catchments. A comparison of the values in Table 1 of [(Rmean –
Qmean)/Emean], i.e. the ratio of the difference between the mean values of rainfall and
discharge to the corresponding value of evaporation, shows that, for catchment
A1522020, this ratio is significantly greater than unity (at 1.25) whereas it is less than
unity for the other 11 catchments. This suggests that, in the transformation of rainfall
to precipitation, quite apart from evapotranspiration losses, there is substantial
unaccounted-for loss in this catchment, so that the hydrological system, represented by
the observed rainfall, evaporation, and discharge data, is apparently non-conservative.
On estimation of regionally averaged “no-model” discharge for each pseudo-ungau-
ged catchment for regional analysis, it is observed that the ratio of the mean regionally
averaged discharge to the mean observed discharge for H3613020 is very high (3.79)
followed by H5723011 (1.99), Y3514020 (1.77) and V6035010 (1.51), whereas for
other catchments it is near to 1. This shows that in each of these four catchments, actual
discharge production is much less than the regionally averaged discharge, and there may
have been some component of rainfall which is not accounted for in the water-balance
expression dependent on the observed data series. In the light of the above comments,
the six catchments namely, A1522020, H3613020, H5723011, Y3514020, Y5615030
and V6035010 are considered heterogeneous in the group of 12 catchments.
Although initial analyses were conducted using all 12 catchments without
considering regional heterogeneity, in the light of the observations regarding
homogeneity presented above, the catchments A1522020, H3613020, H5723011,
154 Monomoy Goswami & Kieran Michael O’Connor

V6035010, Y3514020 and Y5615030 were excluded from the subsequent analyses
having the objective of the parameterization of the SMAR model variant applicable to
conservative systems. Thus the remaining six catchments: J2034010, J3024010,
J4124420, H2001020, K0753210 and K0744010, having discarded the six apparently
heterogeneous catchments, as indicated above, were used only for exploratory tests to
assess the effect of homogeneity, as reflected in the sub-group within the whole sample
of 12 catchments, on the overall performance of the chosen methods.
Clearly, it would be desirable from the perspective of drawing a generalized
conclusion on the performance of the regional methods tested in this study to include
more catchments in the whole sample and hence in the sub-group. However, for the
purposes of this heuristic study, the number used was deemed sufficient and was also
used to demonstrate the applicability of the methods, their relative efficiencies, and the
importance of homogeneity in selection of catchments for the simulation of flow in the
case of ungauged basins.

MODEL DESCRIPTION

Whereas the standard form of the model, generally indicated by SMARG and referred
in this study by SMAR, was designed for conservative systems, modifications were
incorporated in its structure to make it applicable to both conservative and “apparently
non-conservative” systems. These variants are named SMAR-NC1 and SMAR-NC2,
where “NC” denotes “non-conservative”. Detailed description of the structure of the
model, which is not included in this paper for brevity, may be found in Kachroo (1992)
for the original version of SMAR, Goswami et al. (2002) for SMARG, and Goswami
& O’Connor (2005) for SMAR-NC1 and SMAR-NC2.
Briefly, the SMAR model is a parsimonious nine-parameter lumped quasi-physical
conceptual rainfall–evaporation–runoff model, with distinct water-balance and routing
components. Using a number of empirical and intuitively assumed relations which are
considered to be at least physically plausible, the nonlinear water balance (i.e. soil
moisture accounting) component ensures satisfaction of the continuity equation over
each time-step, i.e. it preserves the balance between the rainfall, the evaporation, the
generated runoff, and the changes in the various elements (layers) of soil moisture
storage. Five parameters, namely, Z (moisture holding capacity of soil layers), T
(evapotranspiration conversion factor), H (fast response separation factor), Y
(infiltration excess separation term), and C (factor for soil moisture depletion by
evapotranspiration), control the overall operation of the water-budget component of the
SMAR model. The four parameters G (saturation excess separation factor), N (shape
factor of the Nash cascade model for surface water routing), NK (lag of the Nash
cascade model) and Kg (linear reservoir constant for groundwater routing) control the
operation of the routing component.

RESULTS AND DISCUSSION

A summary of the result of application of the SMAR model using different

regionalization methods, first considering all 12 catchments and subsequently
 Flow simulation in an ungauged basin 155

considering the sub-group of six catchments, are given in Tables 2 and 3. These tables
also show the “no model” efficiencies, which correspond to the discharge series
generated for a pseudo-ungauged catchment by averaging the discharges of all gauged
catchments (except the pseudo-ungauged) in the region. The efficiency of the model,
when applied to the actual observed discharge data of the pseudo-ungauged basin,
considering this basin as gauged, is given for comparison. It is observed from Table 2
that, in comparison with the method of regional averaging of data, the regional pooling
method performs significantly better in calibration and in 10 out of 12 catchments in
validation. For the sub-group of six catchments, the pattern is similar, the R2 efficiency
in this case being significantly better in validation in five out of six catchments. The R2
efficiency in validation, which really reflects the ability of the model to perform in an
ungauged catchment, has a median value of 70.21% and 70.65% for all 12 catchments
and for the sub-group of six catchments, respectively, which is reasonably good for an
ungauged catchment. The improvement in performance in this method is attributed to
the use of catchment-specific data series from each catchment in a group without

Table 2 R2 efficiency values (%) by different regionalization methods considering all 12 catchments.
Catchment H2001020 J2034010 (125) J3024010 J4124420 (32.1) K0744010 K0753210
(station code) (98) (43) (181) (371)
and area (km2) Calib. Verif. Calib. Verif. Calib. Verif. Calib. Verif. Calib. Verif. Calib. Verif.
Regional 68.92 17.89 60.01 63.81 51.33 65.68 55.80 66.15 62.01 64.97 61.86 67.82
averaging of data
Regional pooling 74.45 72.41 74.60 68.21 73.35 73.21 75.51 55.08 74.65 70.57 74.90 66.76
of data
Regional --- 67.86 --- 49.75 --- 73.57 --- 45.97 --- 72.96 --- 72.30
averaging of
parameters
“No model” --- 10.03 --- 23.56 --- 34.33 --- 39.63 --- 37.16 --- 44.01
Best fit with --- 73.42 --- 83.07 --- 83.43 --- 87.69 --- 75.76 --- 75.09
actual observed
discharge data of
the pseudo-
ungauged basin
Catchment and A1522020 H3613020 H5723011 V6035010 Y3514020 Y5615030
area (km2) (68.1) (252) (104) (150) (291) (279)
Calib. Verif. Calib. Verif. Calib. Verif. Calib. Verif. Calib. Verif. Calib. Verif.
Regional 50.78 21.72 58.74 –1722 57.57 26.89 42.48 42.68 6.18 26.42 10.79 15.11
averaging of data
Regional pooling 74.58 74.42 75.01 54.57 73.77 64.82 74.91 69.85 74.31 82.37 72.17 72.68
of data
Regional --- 68.38 --- –685.31 --- 54.43 --- 72.15 --- 68.88 --- 78.34
averaging of
parameters
“No model” --- 15.65 --- –2766 --- 7.84 --- 15.62 --- –110 --- 11.89
Best fit with --- 78.07 --- –78.23 --- 68.06 --- 77.50 --- 79.14 --- 89.30
actual observed
discharge data of
the pseudo-
ungauged basin
Calib: efficiency in calibration with regionally derived data, Verif: efficiency (in verification, wherever applicable)
when used for pseudo-ungauged basin
 156 Monomoy Goswami & Kieran Michael O’Connor

Table 3 R2 efficiency values (%) by different regionalization methods considering six catchments.
Catchment H2001020 J2034010 J3024010 J4124420 K0744010 K0753210
(station code) and (98) (125) (43) (32.1) (181) (371)
area (km2) Calib. Verif. Calib. Verif. Calib. Verif. Calib. Verif. Calib. Verif. Calib. Verif.
Regional 55.42 37.71 71.84 31.84 61.47 74.85 43.54 84.13 59.25 59.52 60.02 65.28
averaging of data
Regional pooling 74.45 65.69 72.05 75.04 71.58 77.75 73.55 58.26 72.09 71.42 72.61 69.87
of data
Transposition
Index basin
J2034010 --- --- --- --- 82.97 54.08 75.28 56.68 --- --- --- ---
J3024010 --- --- 76.22 43.12 --- --- 76.78 82.53 --- --- --- ---
J4124420 --- --- 77.17 32.90 84.09 76.76 --- --- --- --- --- ---
K0744010 --- --- --- --- --- --- --- --- --- --- 55.90 35.75
K0753210 --- --- --- --- --- --- --- --- 42.04 58.52 --- ---
Regional --- 63.37 --- 60.00 --- 75.75 --- 59.39 --- 69.34 --- 69.29
averaging of
parameters
“No model” --- 19.42 --- -9.75 --- 39.90 --- 44.20 --- 36.00 --- 50.47
Best fit with --- 73.42 --- 83.07 --- 83.43 --- 87.69 --- 75.76 --- 75.09
actual observed
discharge data of
the pseudo-
ungauged basin
Calib: efficiency in calibration with regionally derived data, Verif: efficiency (in verification, wherever
applicable) when used for pseudo-ungauged basin.

any averaging, and hence without dilution of the response characteristics of each of the
catchments. Models calibrated to the data series generated by pooling simulate the
response from an ungauged catchment in the region better because of the response of
the ungauged catchment being similar to many, if not all, catchments of the region.
The method of transposition, being suitable for a small sub-group of catchments,
was applied only to two relatively more homogeneous sub-groups, one comprised of
the catchments J2034010, J3024010 and J4124420 in the northwest, and the other
consisting of two catchments, K0744010 and K0753210. The results of the method of
transposition are given in Table 3. From the results of transposition for catchment
J4124420 considering J3024010 as the index basin, and vice versa, it may be observed
that for catchments having areas of identical order of magnitude good results may be
obtained by transposition. For a large difference in areas of the two catchments
considered for transposition (one gauged and the other pseudo-gauged), the scaling of
discharge series of the index catchment lowers the performance. This is seen from the
performances of the pairs K0744010 and K0753210, J2034010 and J3024010, and
J2034010 and J4124420, when considered for transposition.
Comparison of results of the method of regional averaging of parameters with
those of the regional pooling method in Table 2 shows that, for the case with 12
catchments, the former performs better in five out of 12 catchments, and that the
efficiency for the catchment H3613020 by the method of regional averaging of para-
meters is unacceptably negative. This reflects that the method of regional pooling is
clearly best for the group of 12 catchments. It also shows that catchment H3613020
 Flow simulation in an ungauged basin 157

may have been an outlier in the group, and that the inclusion of this catchment in the
regional analysis is likely to reduce the efficiency of flow modelling by any regionali-
zation method. A similar comparison of the method of regional averaging of
parameters with the regional pooling method given in Table 3 shows that, in the case
of the sub-group of six catchments, the efficiencies in the pseudo-ungauged basins,
although generally lower, are comparable. In the context of the regional averaging
method, however, it may be noted that due to equifinality (Beven & Freer, 2001), the
parameter values in the optimum parameter set, as obtained by the modelling exercise,
may differ significantly in some cases from catchment to catchment in the group, and
meaningful averaging of parameters may not be possible. A number of tests may be
required in such a case to choose the appropriate set of parameter values giving the
best fit of the model to be considered for averaging.
From Tables 2 and 3, it is found that in five out of the six catchments indicated in
Table 3, the method of regional pooling of data, considering these six catchments as a
homogeneous sub-group, generally performs better than when the whole group of 12
catchments is selected. Thus, although a larger volume of data is used in the case of 12
catchments, the homogeneity in the sub-group of six catchments yields better
efficiency of the modelling method in the case of the test with the homogeneous sub-
group of only six catchments in comparison with that with all 12 catchments.
Similarly, in three out of these six catchments, the method of regional averaging of
parameters, considering the six catchments in the homogeneous sub-group, performs
better in comparison with the case when the whole group of 12 catchments is chosen.
From Tables 2 and 3 it is observed that, as expected, efficiency values obtained by
all regionalization methods range between those achieved by “no model” and by using
the model best fitted to the actual observed discharge series of the pseudo-ungauged
basins except in the case of catchments H3613020 and Y3514020. In the case of these
two catchments, the efficiencies in verification obtained by the method of regional
pooling are higher than the corresponding values obtained by considering the observed
discharge data series. This inconsistency is attributed to the peculiarity in hydrological
characteristics of these two catchments as explained earlier in the section regarding
assessment of regional homogeneity. As demonstrated in this study on regionalization
approaches for continuous flow simulation in an ungauged basin, the conceptual
SMAR model proved to be a suitable choice.

CONCLUSIONS

Assessing hydrometeorological and physiographical homogeneity is very important for

reducing uncertainty of estimation of flow in an ungauged catchment by regional
analysis. The method of regional pooling of data is considered to be the best when a
number of gauged catchments are available in the region. When only a few gauged
catchments are available, transposition of data from the gauged to the ungauged basin
may be used provided the catchments are similar in characteristics and their areas are
of similar order of magnitude. As expected, the lack of homogeneity in a larger sample
of catchments generally reduces the efficiency of the regionalization methods for flow
modelling in ungauged basins. The conceptual SMAR model is seen to be a suitable
choice for regionalization studies in continuous flow simulation in an ungauged basin.
158 Monomoy Goswami & Kieran Michael O’Connor

REFERENCES
Kachroo, R. K. (1992) River flow forecasting, Part 5: Applications of a conceptual model. J. Hydrol. 133, 141–178.
Goswami, M., O’Connor, K. M. & Shamseldin, A. Y. (2002) Structures and performances of five rainfall–runoff models
for continuous river-flow simulation. In: Proc. 1st Biennial Meeting of Int. Env. Modeling and Software Soc.
(Lugano, Switzerland), 1, 476–481.
Goswami, M. & O’Connor, K. M. (2005) Application of a conceptual rainfall–runoff simulation model to three European
catchments characterised by non-conservative system behaviour. In: Proc. Int. Conf. Hydrological Perspectives for
Sustainable Development (Roorkee, India), 1, 117–130.
Beven, K. & Freer, J. (2001) Equifinality, data assimilation, and uncertainty estimation in mechanistic modelling of
complex environmental systems using the GLUE methodology. J. Hydrol. 249, 11–29.