DOWNSCALING METHODS FOR EVALUATING THE IMPACTS OF CLIMATE CHANGE AND

VARIABILITY ON HYDROLOGICAL REGIME AT BASIN SCALE: Nguyen , V.T.V.

DOWNSCALING METHODS FOR EVALUATING THE IMPACTS OF CLIMATE

CHANGE AND VARIABILITY ON HYDROLOGICAL REGIME AT BASIN SCALE

Van-Thanh-Van Nguyen

McGill University
Department of Civil Engineering and Applied Mechanics
Montreal, Quebec, Canada H3A 2K6

E-mail: van.tv.nguyen@mcgill.ca

ABSTRACT

This paper provides an overview of various downscaling methods that could be used for assessing the po-
tential impacts of climate change and variability on hydrological regime. In general, two broad categories
of these downscaling procedures currently exist: dynamical downscaling (DD) techniques, involving the
extraction of regional scale information from large-scale circulation data based on the modeling of re-
gional climate dynamical processes, and statistical (or empirical) downscaling (SD) procedures that relied
on the empirical relationships between large-scale atmospheric variables and surface environment pa-
rameters. While neither DD methods nor SD methods were found to outperform the other, the SD tech-
niques have several practical advantages. Hence, two popular SD methods based on the Statistical Down-
scaling Model (SDSM) and the Stochastic Weather Generator (LARS-WG) were selected for testing their
feasibility in the simulation of two fundamental hydrologic processes: daily precipitation time series and
daily temperature extremes. Results of the evaluation using available climate data in the Montreal region
(Quebec, Canada) have indicated that both models were able to describe accurately the basic statistical
properties of daily maximum and minimum temperatures at local sites. However, none of these models ap-
pears to be able to simulate well the statistical properties of the daily precipitation processes.

INTRODUCTION

Climate change will have important impacts on the hydrologic cycle at different temporal and spatial
scales. The temporal scales could vary from a very short time interval of 5 minutes (for urban water cycle)
to a yearly time scale (for annual water balance computation). The spatial resolutions could be from a few
square kilometers (for urban and rural watersheds) to several thousand square kilometers (for large river
basins). General Circulation Models (GCMs) have been recognized to be able to represent reasonably well
the main features of the global distribution of basic climate parameters (Gates et al., 1999; Lambert and
Boer, 2001), but these models so far could not reproduce well details of regional climate conditions at tem-
poral and spatial scales of relevance to hydrological studies (Xu, 1999; Prudhomme et al., 2002). In other
words, outputs from GCMs are usually at resolution that is too coarse (generally greater than 2.0o for both
latitude and longitude, and greater than 200km for middle latitudes) for many climate change impact stud-
ies. Hence, there is a great need to develop tools for downscaling GCM predictions of climate change to
regional and local or station scales. In recent years, different downscaling methods have been proposed in a
number of studies around the world. Of particular importance for the management of water resources sys-
tems are those procedures dealing with the linkage of the large-scale climate variability to the historical ob-
servations of the surface parameters of interest (e.g., precipitation and temperature). If this linkage could be
established, then the projected change of climate conditions given by a GCM could be used to predict the
resulting change of the selected surface parameters for hydrological impact studies. The required linkage
can be developed using a wide range of downscaling methods.
Two broad categories of these downscaling procedures currently exist (Hughes and Guttorp 1994;
Wilks and Wilby, 1999; Xu, 1999; Yarnal et al., 2001; Wilby et al., 2002b): dynamical downscaling (DD)
techniques, involving the extraction of regional scale information from large-scale GCM data based on the
modeling of regional climate dynamical processes, and statistical (or empirical) downscaling (SD) proce-

Role of Water Sciences in Transboundary River Basin Management, Thailand, 2005 1

DOWNSCALING METHODS FOR EVALUATING THE IMPACTS OF CLIMATE CHANGE AND
VARIABILITY ON HYDROLOGICAL REGIME AT BASIN SCALE: Nguyen , V.T.V.

dures that relied on the empirical relationships between observed (or analyzed) large-scale atmospheric
variables and observed (or analyzed) surface environment parameters. Some recent comparisons of DD
and SD techniques for climate impact studies (Mearns et al., 1999; Gutowski et al., 2000; Yarnal et al.,
2001) have indicated that neither technique was consistently better than the other. In particular, based on
the assessment of the climate change impacts on the hydrologic regimes of a number of selected basins in
the United States, Gutowski et al. (2000) have found that these two methods could reproduce some general
features of the basin climatology, but both displayed systematic biases with respect to observations as well.
Furthermore, a main finding from this study was that the assessment results were dependent on the specific
climatology of the basin under consideration. Hence, it is necessary to test different, but physically plausi-
ble, downscaling methods in order to find the most suitable approach for a particular region of interest.
However, it has been widely recognized that SD methods offer several practical advantages over DD pro-
cedures, especially in terms of flexible adaptation to specific study purposes, and inexpensive computing
resource requirements (Xu, 1999; Prudhomme et al., 2002).
Several SD techniques have been developed to establish relationships between local weather variables
and the large-scale GCMs’ results. Among these techniques, the SD method based on the Statistical Down-
scaling Model (SDSM) (Wilby et al., 2002a) and the stochastic weather generator LARS-WG (Semenov et
al., 1998) are widely used for constructing climate change scenarios for daily precipitations and tempera-
ture extremes at individual sites using GCM grid point information. This paper presents therefore an over-
view of SD methods and, in particular, an assessment of the SDSM and LARS-WG models to test their fea-
sibility and adequacy in the simulation of daily time series of precipitation and temperature extremes for
four sites located in the greater region of Montreal (Quebec, Canada). These models were evaluated using
historical data available for the period from 1961 to 1990, of which the first 15 years were used for models’
calibration and the remaining 15 years for models’ validation. The performance of the selected methods
was examined by comparing the statistics of observed data time series to those of 100 simulated time se-
quences. On the basis of this comparison, the strength and weaknesses of these tools are identified and dis-
cussed in the present paper.

OVERVIEW OF STATISTICAL DOWNSCALING METHODS

Statistical (or empirical) downscaling (SD) methodologies can be classified into three categories according
to the computational techniques used (Xu, 1999; Wilby et al., 2002a): weather typing approaches; regres-
sion methods; and stochastic weather generators. In general, these SD methods require three common as-
sumptions (i) the surface local-scale parameters are a function of synoptic forcing; (ii) the GCM used for
deriving downscaled relationships is valid at the scale considered; and (iii) the derived relationships remain
valid under changing climate conditions.
The weather typing procedures consist of classifying atmospheric circulation pattern into limited num-
ber of classes; simulating weather types using stochastic models; establishing the link of rainfall occur-
rence to weather type using conditional probabilities; and simulating the rainfall process (or other hydro-
meteorological processes) using weather types (Murphy, 1999; Yarnal et al., 2001). The interesting fea-
tures of these methods are the consideration of the linkages between climate on the large scale and weather
at the local scale, and the possibility of generating long sequences of daily precipitation at a site based on
limited historical data sets. However, weather classification schemes are somewhat subjective. In particu-
lar, the main limitation of such procedures is that precipitation changes produced by changes in the fre-
quency of weather patterns could be inconsistent with the changes produced by the host GCM (Wilby et
al., 2002a).
The regression-based downscaling methods are mainly relied on the empirical statistical relationships
between large-scale predictors and local-scale parameters (Yarnal et al., 2001; Wilby et al., 2002a; Prud-
homme et al., 2002). Different approaches in this empirical downscaling category can be identified accord-
ing to the choice of the mathematical function for describing the predictor-predictand relationship, the
computational technique used, or the selection of the predictor variables considered (Wilby et al., 2002b).
In general, the main advantage of the regression downscaling procedures is that these methods are simple
and less computationally demanding as compared to other downscaling methods. However, the application
of regression-based procedures is limited to the locations where good predictor-predictand relationships
could be found. Furthermore, similar to weather typing methods, the regression-based techniques assume
validity of the estimated model parameters under future climate conditions.

2 Role of Water Sciences in Transboundary River Basin Management, Thailand, 2005

 DOWNSCALING METHODS FOR EVALUATING THE IMPACTS OF CLIMATE CHANGE AND
VARIABILITY ON HYDROLOGICAL REGIME AT BASIN SCALE: Nguyen, V.T.V.

The stochastic weather generators have been used extensively in the planning, design, and management
of water resources systems (Wilks, 1998; Wilks and Wilby, 1999). The stochastic weather generator meth-
ods are based mainly on the stochastic weather generator models such as WGEN (Richardson, 1981) and
LARS-WG (Semenov and Barrow, 1997). These models involve typically the modelling of the daily rain-
fall occurrences, the description of the distribution of rainfall amount on a wet day, and the conditioning of
other weather variables (temperature, radiation, etc.) on the wet/dry status of the day. The climate change
scenarios are then stochastically generated based on the linkage between the stochastic model parameters
with the corresponding variable changes in the GCM. In general, both generators have a similar structure
in which observed data at a given site were used to estimate the parameters of the probability distributions
of the daily climate variables (minimum and maximum temperatures, precipitation, and solar radiation). In
addition, both models analyze dry and wet days separately and thus include a mechanism for selecting the
precipitation status of each day. The generators differ mainly in the choice of the probability distributions
used. WGEN uses standard distributions (e.g., two-parameter Gamma), whereas LARS-WG employed
semi-empirical distributions. One advantage of using a standard distribution is that it will have a smoothing
effect on the empirical frequency of the observed data and will only requires the estimation of a few pa-
rameters. However, such distribution may not provide a very good fit to the observed data. A semi-
empirical distribution, with a larger number of parameters, is more flexible and could be able to describe
accurately any shape of empirical frequency distribution. The performance of the WGEN and LARS-WG
has been tested using data from a range of diverse climates (Semenov et al., 1998). The LARS-WG genera-
tor was found to be able to describe the observed weather characteristics more accurately than the WGEN.
In general, the principal advantage of the stochastic weather generator procedures is that they are able to
reproduce many observed statistical characteristics of daily weather variables at a particular site. In addi-
tion, the stochastic weather generators could generate a large number of different climate scenarios for risk
assessment studies. However, the main disadvantage of these procedures is related to the arbitrary manner
of determining the model parameters for future climate conditions (Wilby et al., 2002b).
In summary, because of various practical advantages of SD methods over DD procedures, two popular
SD techniques based on the Statistical Downscaling Model (SDSM) (Wilby et al., 2002a) and the LARS-
WG (Semenov and Barrow, 1997) model have been selected in this study for testing their feasibility in the
simulation of daily precipitation and extreme temperature processes for the Montreal region in Quebec,
Canada. Presented in the following sections are the results of the evaluation of these two methods.

EVALUATION OF SDSM AND LARS-WG PROCEDURES

The SDSM is best described as a hybrid of the stochastic weather generator and regression-based methods.
The model permits the spatial downscaling of daily predictor-predictand relationships using multi ple lin-
ear regression techniques and generates “synthetic predictand” that represents the generated local weather.
Further details of SDSM are provided by Wilby et al. (2002a). The LARS-WG model (Semenov and Bar-
row, 1997) produces synthetic daily time series of maximum and minimum temperatures, precipitation, and
solar radiation. The model uses input observed daily weather for a given site to determine the parameters of
specifying probability distributions for weather variables as well as the correlations between these vari-
ables. As mentioned above, LARS-WG utilizes semi-empirical distributions for the lengths of wet and dry
day series, daily precipitation and daily solar radiation. The generation procedure to produce synthetic
weather data is then based on selecting values from the appropriate distributions using a pseudo-random
number generator.

Data
Two data sets are used: station data and NCEP re-analysis data. Station data include observed daily pre-
cipitation, daily maximum temperature (tmax), and daily minimum temperature (tmin) for the period 1961-
1990 recorded at four stations in the greater region of Montreal: Dorval, Drummondville, Maniwaki, and
McGill. NCEP re-analysis data are composed of 24 daily atmospheric variables for the same period which
are selected at grid box covering each of the stations (see Table 1).

Procedures
SDSM uses NCEP reanalysis data as predictors and station data as predictands, whereas LARS-WG model
requires only the station data. Data for the 1961-1975 period were used for models’ calibration step, and
those of 1976-1990 for models’ validation. After calibrated, the calibrated models are run with the models’

Role of Water Sciences in Transboundary River Basin Management, Thailand, 2005 3

DOWNSCALING METHODS FOR EVALUATING THE IMPACTS OF CLIMATE CHANGE AND
VARIABILITY ON HYDROLOGICAL REGIME AT BASIN SCALE: Nguyen , V.T.V.

parameters and climate conditions for the period 1961-1975 to generate 100 series of local weather data,
each series has 15 years of length. At validation step, the calibrated models are run with the models’ pa-
rameters and climate conditions for the period 1976-1990 to generate 100 series of local weather data, each
of 15 years long. The outputs are statistically analyzed and compared to statistics of observed data for the
same period to evaluate the models’ performance. Evaluation statistics are the mean, the percentiles (the
90th value for daily precipitation and daily maximum temperature tmax; and the 10th value for daily mini-
mum temperature tmin), the total amount of precipitation, and the percentage of wet-days for precipitation.

Table 1: List of atmospheric variables in NCEP re-analysis data.

Measurement Level
Variable

Mean sea level pressure

Airflow strength surface 500 hPa 850 hPa

Zonal velocity surface 500 hPa 850 hPa

Meridional velocity surface 500 hPa 850 hPa

Vorticity surface 500 hPa 850 hPa

Wind direction surface 500 hPa 850 hPa

Divergence surface 500 hPa 850 hPa

Specific humidity near surface 500 hPa 850 hPa

850 hPa
Geopotential height 500 hPa

RESULTS

Model calibration
Calibration of the SDSM for daily precipitation, maximum temperature, and minimum temperature was
performed independently. Results showed that in general the daily precipitation predictand requires the fol-
lowing significant predictors: zonal velocities, meridional velocities, specific humidities, geopotential
height, and vorticity, while both daily maximum (tmax) and minimum (tmin) temperatures predictands re-
quire geopotential heights and specific humidities at all levels. The coefficients of determination (R2) after
calibration for tmax and tmin were very high, from 0.714 to 0.785, respectively, while the value for pre-
cipitation was quite very low, ranging from 0.062 to 0.098. This would indicate the difficulty in finding
significant climate variables from the NCEP data that could explain well the variability of daily precipita-
tion. For purpose of illustration, Fig. 1 shows the bias values of the mean of the daily precipitation, tmax,
and tmin produced by the SDSM and LARS-WG models for Dorval station. It can be seen that none of the
models could accurately replicate the observed mean of the daily precipitation amount. Results for other
stations (not presented here) indicate that the bias values of the mean daily precipitation for the SDSM
range from –1.8 mm to +1.8 mm, while those for the LARS-WG were smaller. Hence, LARS-WG could
provide a better description of the mean daily precipitation than the SDSM. For daily tmax and tmin, both
models could very well describe the daily mean values with bias values ranging between +0.7 oC and –1.0
o
C, with a somewhat higher accuracy for the SDSM. Similar results were found for other evaluation statis-
tics. In summary, it was found that the LARS-WG model was able to produce daily precipitation statistics
in closer agreement with those of the observed data, while SDSM provided daily extreme temperature sta-
tistics more accurate than the LARS-WG.

4 Role of Water Sciences in Transboundary River Basin Management, Thailand, 2005

 DOWNSCALING METHODS FOR EVALUATING THE IMPACTS OF CLIMATE CHANGE AND
VARIABILITY ON HYDROLOGICAL REGIME AT BASIN SCALE: Nguyen, V.T.V.

Mean of precipitation Dorval 1961-1975

(mm) SDSM LARS
2.0
1.5
1.0
0.5
0.0
-0.5 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
-1.0
-1.5
-2.0

Mean of Tmax
(oC) SDSM LARS
Dorval 1961-1975
1.0

0.5

0.0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
-0.5

-1.0

-1.5

Mean of Tmin
(oC) SDSM LARS
Dorval 1961-1975
0.6
0.4
0.2
0.0
-0.2 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
-0.4
-0.6
-0.8
-1.0

Fig. 1: Bias values for the mean of daily precipitation, and maximum and minimum temperatures for
Dorval station for the calibration 1961-1975 period

Mean of precipitation Dorval 1976-1990

(mm) SDSM LARS
3.0
2.0
1.0
0.0
-1.0 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
-2.0
-3.0
-4.0

Fig. 2: Bias values for the mean of daily precipitation for Dorval station for the validation
1976-1990 period

Model validation
Statistics of weather data for the validation period from 1976 to 1990 given by the two models are com-
pared to those of observed data for the same period. Results of the validation of the SDSM and LARS-WG
for daily precipitations at the four selected sites indicated that none of these models was able to generate
perfectly the statistics of daily precipitation. Bias for the mean of daily precipitation ranges from –3.3 mm
to +2.0 mm for SDSM, and –0.9 mm to +1.3 mm for LARS-WG; for the 90th percentile of daily precipita-
tion from -5.6 mm to 7.2 mm for SDSM, and from -5.0 mm to 5.2 mm for LARS-WG; and for the monthly
total amount of precipitation from -37.1 mm to 35.6 mm for SDSM, and from -10.6 mm to 10.2 mm for
LARS-WG. For purpose of illustration, Fig. 2 shows the bias values of the mean of daily precipitation for

Role of Water Sciences in Transboundary River Basin Management, Thailand, 2005 5

DOWNSCALING METHODS FOR EVALUATING THE IMPACTS OF CLIMATE CHANGE AND
VARIABILITY ON HYDROLOGICAL REGIME AT BASIN SCALE: Nguyen , V.T.V.

the SDSM and LARS-WG models for Dorval station. Hence, LARS-WG can reproduce more accurate ob-
served statistics of daily precipitation than the SDSM.
For assessment of the validation results for daily maximum temperature tmax, the box plots of the
evaluation statistics were drawn using 100 series generated by the SDSM and LARS-WG models. The box
is formed by the first and third quartiles; the upper and lower whiskers are the maximum and minimum val-
ues, respectively; and the center of the box is the mean value of the statistic. The box-plots are plotted to-
gether with the mean of the statistic computed from the observed data (presented by a bar) in the same
graph for comparison. Labels on horizontal axis of the graphs are composed of two letters, e.g., O-J, of
which the first letter represents the type of data (O: observed, S: SDSM-generated, L: LARS-WG-
generated) and the second letter stands for the first letter of a month (J: January, F: February, and so on).
Fig. 3 presents the box-plots of the means of tmax for Drummondville station for different seasons in a
year. The graph suggests that SDSM can reproduce better the observed means of tmax as compared to the
LARS-WG model. Results for other stations indicated that the maximum bias of the mean tmax ranges
from –2oC to +2 oC for SDSM, and from –3 oC to +2.5 oC for LARS-WG. Moreover, the box-plot widths
for SDSM-generated mean tmax are all narrower than those for LARS-WG-generated one. Similar results
were found for other evaluation statistics (e.g., the 90th percentile value). These results clearly show that
the SDSM could provide more accurate and more robust estimates of the observed statistics of daily maxi-
mum temperature than the LARS-WG model.
Similarly, for daily minimum temperature, the performance of the SDSM and LARS-WG models was
assessed by comparing the mean and other statistics (e.g., the 10th percentile) of generated daily tmin series
with those of the observed data. Fig. 4 presents the mean of observed daily tmin together with the box-plots
of the mean of daily time generated by SDSM and LARS-WG model for Maniwaki station. It can be seen
that SDSM was able to reproduce the mean of the observed daily tmin for each month more accurately and
more reliable than the LARS-WG. Similar results were found for other stations and for other evaluation
statistics.

SUMMARY AND CONCLUSIONS

Several features distinguish DD and SD methods for regional climate simulation. DD procedures are
mainly based on regional-scale limited-area models that describe the climate physical processes using fun-
damental conservation laws for mass, energy and momentum. DD methods contain thus more complete
physics than SD techniques. However, the more complete physics significantly increases computational
cost, which limits the simulation of a climate by these models to typically a single realization. On the other
hand, SD approaches are relatively fast and less expensive than computationally intensive DD methods.
These convenient features of the SD allow the users to develop a large number of different climate realiza-
tions and thus to be able to quantify the confidence interval of simulated climate variables. In addition, SD
methods can directly account for the observed climate and weather data available at the study site. The re-
sults are thus consistent with the local climate conditions as described by the observations.
In this study, because of various practical advantages of SD methods over DD procedures, two popular
SD techniques based on the LARS-WG and the SDSM models have been selected in this study for testing
their feasibility in the simulation of daily precipitation and extreme temperature series for four raingage
stations in the Montreal region in Quebec, Canada. Calibration of the SDSM suggested that local precipita-
tion was mainly related to the large-scale climate variables such as the zonal velocities, meridional veloci-
ties, specific humidities, geopotential height, and vorticity, while the local maximum and minimum tem-
peratures were strongly related to the geopotential heights and specific humidities at all levels.
The comparison between some selected statistics of observed weather data and those of weather data
generated by the two models indicates that the LARS-WG model can provide the daily precipitation statis-
tics more comparable to those of the observed data than the SDSM. However, both models were unable to
reproduce accurately theses observed statistics.
The SDSM and LARS-WG were found to be able to describe adequately the observed statistics of daily
temperature extremes, and the SDSM was found to be somewhat more accurate than the LARS-WG. In
terms of practical applications, calibration of the LARS-WG model is much simpler than that of the SDSM.
The calibration of the SDSM is based on a complex procedure in order to be able to establish successfully
the good relationships between large-scale predictor variables and the surface weather variables at a local
site.

6 Role of Water Sciences in Transboundary River Basin Management, Thailand, 2005

 DOWNSCALING METHODS FOR EVALUATING THE IMPACTS OF CLIMATE CHANGE AND
VARIABILITY ON HYDROLOGICAL REGIME AT BASIN SCALE: Nguyen, V.T.V.

Further studies are planned to compare the performance of these two models using data from other sites
with different climatic conditions. In addition, the performance of these two models should be evaluated
using data from GCMs in order to be able to assess the reliability of generated future climate scenarios for
a local site.

Fig. 3: Box plots of the mean of daily tmax in Winter (DJF), Spring (MAM), Summer (JJA), and
Autumn (SON) months for Drummondville for the validation 1976-1990 period

Fig. 4: Box plots of the mean of daily tmin in Winter (DJF), Spring (MAM), Summer (JJA), and Au-
tumn (SON) months for Maniwaki station for the validation 1976-1990 period

REFERENCES

Gates, W.L., Boyle, J.S., Covey, C., Dease, C.G., Doutriaux, C.M., Drach, R.S., Fiorino, M., Glecker, P.J.,
Hnilo, J.J., Marlais, S.M., Phillips, T.J., Potter, G.L., Santer, B.D., Sperber, K.R., Taylor, K.E., and
Williams, D.N. (1999) An overview of the results of the atmospheric model intercomparison project
(AMIP I). Bulletin of the American Meteorological Society, 80: 29-55.

Role of Water Sciences in Transboundary River Basin Management, Thailand, 2005 7

DOWNSCALING METHODS FOR EVALUATING THE IMPACTS OF CLIMATE CHANGE AND
VARIABILITY ON HYDROLOGICAL REGIME AT BASIN SCALE: Nguyen , V.T.V.

Giorgi, F. and Mearns, L.O. (1999) Introduction to special section: Regional climate modeling revisited.
Journal of Geophysical Research, 104: 6335-6352.
Gutowski, W.J., Jr., Wilby, R.L., Hay, L.E., Anderson, C.J., Arritt, R.W., Clark, M.P., Leavesley, G.H.,
Pan, Z., Da Silva, R., and Takle, E.S. (2000) Statistical and Dynamical Downscaling of Global Model
Output for US National Assessment Hydrological Analyses. 11th Symposium on Global Change Stud-
ies, Long Beach, CA, January 9-14, 2000.
Hughes, J.P. and Guttorp, P. (1994) A class of stochastic models for relating synoptic atmospheric patterns
to regional hydrologic phenomena. Water Resources Research, 30: 1535-1546.
Lambert, S. J., and Boer, G.J. (2001) CMIP1 Evaluation and Intercomparison of Coupled Climate Models.
Climate Dynamics, 17: 83-106. Prudhomme C, Reynard N, Crooks S. 2002. Downscaling of global cli-
mate models for flood frequency analysis: where are we now? Hydrological Processes 16: 1137-1150.
Mearns LO, Bogárdi I, Matyasovszky I, Palecki M. (1999) Comparison of climate change scenarios gener-
ated from regional climate model experiments and statistical downscaling. Journal of Geophysical Re-
search 104: 6603-6621.
Murphy J. (1999) An evaluation of statistical and dynamical techniques for downscaling local climate.
Journal of Climate 12: 2256-2284.
Richardson, C. W. (1981) Stochastic Simulation of Daily Precipitation, Temperature, and Solar Radiation.
Water Resources Research, 17(1): 182-190.
Richardson, C.W., and Wright, D.A.. (1981) WGEN: A model for generating daily weather variables. US
Department of Agriculture Research Service, ARS-8, USDA, Washington, DC.
Semenov, Mikhail A., and Barrow, E.M.. (1997) Use of Stochastic Weather Generator in the Development
of Climate Change Scenarios, Climatic Change, 35: 397-414.
Semenov, M.A., Brooks, R.J., Barrow, E.M., and Richardson, C.W. (1998) Comparison of the WGEN and
LARS-WG stochastic weather generators in diverse climates. Climate Research, 10: 95-107.
Wilby, R.L., Hay, L.E., and Leavesley, G.H. (1999) A comparison of downscaled and raw GCM output:
implications for climate change scenarios in the San Juan River Basin, Colorado. Journal of hydrology,
225: 67-91.
Wilby, R.L., Dawson, C.W., and Barrow, E.M. (2002a) SDSM – a decision support tool for the assessment
of regional climate change impacts. Environmental Modelling & Software, 17: 147-159.
Wilby R.L., Conway, D., Jones, P.D. (2002b) Prospects for downscaling seasonal precipitation variability
using conditioned weather generator parameters. Hydrological Processes 16: 1215-1234.
Wilks, D.S. (1998) Multisite Generalization of a Daily Stochastic Precipitation Generation Model, Journal
of Hydrology, 210: 178-191.
Wilks, D.S., and Wilby, R.L. (1999) The Weather Generation Game: A Review of Stochastic Weather
Models, Progress in Physical Geography, 23(3): 329-357.
Xu C-Y. (1999) From GCMs to river flow: a review of downscaling methods and hydrologic modeling ap-
proaches. Progress in Physical Geography 23(2): 229-249.
Yarnal, B., Comrie, A.C., Frakes, B., and Brown, D.P. (2001) Developments and Prospects in Synoptic
Climatology, International Journal of Climatology, 21: 1923-1950.

8 Role of Water Sciences in Transboundary River Basin Management, Thailand, 2005