Water Resources Management 13: 369–382, 1999.

369
© 2000 Kluwer Academic Publishers. Printed in the Netherlands.

Review Paper

Climate Change and Hydrologic Models: A Review

of Existing Gaps and Recent Research
Developments

CHONG-YU XU
Uppsala University, Department of Earth Sciences, Hydrology, Villavagen 16, S-75236 Uppsala,
Sweden, e-mail: [Link]@[Link]

(Received: 10 May 1999; in final form: 11 November 1999)

Abstract. Global atmospheric general circulation models (GCMs) have been developed to simulate
the present climate and used to predict future climatic change. While GCMs demonstrate significant
skill at the continental and hemispheric spatial scales and incorporate a large proportion of the com-
plexity of the global system, they are inherently unable to represent local subgrid-scale features and
dynamics. The existing gap and the methodologies for narrowing the gap between GCMs’ ability
and the need of hydrological modelers are reviewed in this paper. Following the discussion of the
advantages and deficiencies of various methods, the challenges for future studies of the hydrological
impacts of climate change are identified.

Key words: climate change, general circulation models, hydrological models.

1. Introduction
Increased concentration of greenhouse gases is expected to alter the radiative bal-
ance of atmosphere, causing increases in temperature and changes in precipitation
patterns and other climatic variables (Houghton et al., 1990). One of the most
important impacts on society of future climatic changes will be changes in re-
gional water availability. Such hydrologic changes will affect nearly every aspect
of human well-being, from agricultural productivity and energy use to flood con-
trol, municipal and industrial water supply, and fish and wildlife management. For
example, larger reservoir spillways and drainage waterways will be required where
runoff is expected to increase, and higher water supply storage needed where runoff
is expected to decrease. The tremendous importance of water in both society and
nature underscores the necessity of understanding how a change in global climate
could affect regional water supplies.
Global atmospheric general circulation models (GCMs) have been developed
to simulate the present climate and used to predict future climatic change. While
GCMs demonstrate significant skill at the continental and hemispheric spatial scales
and incorporate a large proportion of the complexity of the global system, they are
370 CHONG-YU XU

inherently unable to represent local subgrid-scale features and dynamics (Wigley

et al., 1990; Carter et al., 1994). When considering the impacts of global cli-
mate change the focus is primarily on societal responses to the local and regional
consequences of large-scale changes. The conflict between GCM performance at
regional spatial scales and the needs of regional-scale impact assessment is largely
related to many facts. Although individual hydrologists might consider different
topics and other views more appropriate, the following problems are considered
important and will be discussed in this paper:

• GCMs accuracy decreases at increasingly finer spatial and temporal scales,

while the needs of impacts studies conversely increase with higher resolution.
• GCMs accuracy decreases from free tropospheric variables to surface vari-
ables, while the variables at the ground surface have direct use in water balance
computations.
• GCMs accuracy decreases from climate related variables, i.e., wind, temperat-
ure, humidity and air pressure to precipitation, evapotranspiration, runoff and
soil moisture, while the later variables are of key importance in hydrologic
regimes.

Important in connection with closing these gaps are topics, such as: (1) dynamic
downscaling (nesting) approaches for generating the high-resolution meteorolo-
gical inputs required for hydrological models, (2) statistical downscaling approaches
for simulating local scale surface variables from free tropospheric variables, and
(3) macroscale hydrological modeling approaches for simulating river flows in
large river basins and for correcting perceived weaknesses in the representation
of hydrological processes in GCMs.
This paper aims at discussing the existing gaps between GCMs and hydrological
models, reviewing the recent research developments, and presenting the challenges
for the future studies of the hydrological impacts of climate change.

2. GCMs and Hydrological Models

2.1. ROLES OF GCM s IN CLIMATE CHANGE STUDY

GCMs, of which the first account dates back to Phillips (1956), were initially
developed to simulate average, synoptic-scale (i.e., 104 –106 km2 spatial scale),
atmospheric circulation patterns for specified external forcing conditions. Since
then, various atmospheric GCMs were conceptually designed to simulate average,
large-scale, atmospheric circulation (e.g., Holton, 1992). During the last 20 yr
or so, GCMs have been used to simulate climatic sensitivity to increased carbon
dioxide concentrations and other important parameters, and to predict future cli-
matic change. Some of the leading GCMs in common use today are: the Canadian
Climate Center (CCC) model, the Geophysical Fluid Dynamic Laboratory (GFDL)
model, the Goddard Institute for Space Studies (GISS) model, the National Center
CLIMATE CHANGE AND HYDROLOGIC MODELS 371

for Atmospheric Research (NCAR) model, the Oregon State University (OSU)
model, and the United Kingdom Meteorological Office (UKMO) model.
General circulation models, the only available tool for detailed modeling of
future climate evolution, are not well suited for answering the question for primary
interest to hydrologists concerning regional-scale hydrologic variability. As GCMs
operate on large spatial scale, and, furthermore, as the GCM-simulated temporal
resolution corresponds to monthly averages at best, the direct usefulness of GCM
output in impact studies and other applications is limited. The present-day free
troposphere is modeled relatively well by coarse GCMs, whereas local or even
regional characteristics of surface or near-surface climate variables, their variability
and the likelihood of extreme events cannot be obtained directly from GCMs. The
same is true in the case of climate change experiments with GCMs. Embedding
schemes linking GCMs to meteorologic and hydrologic models resolved at finer
scales have been proposed and implemented. This is the state-of-the-art approach
to bridge the gap between coarse resolution GCMs and hydrologic modeling at the
river basin scale.

2.2. ROLES OF HYDROLOGICAL MODELS IN CLIMATE CHANGE STUDY

The concept of using regional hydrologic models for assessing the impacts of cli-
matic change has several attractive characteristics (Glecik, 1986; Schulze, 1997).
First, models tested for different climatic/physiographic conditions, as well as mod-
els structured for use at various spatial scales and dominant process representations,
are readily available. This permits flexibility in identifying and choosing the most
appropriate approach to evaluate any specific region. Second, hydrologic models
can be tailored to fit the characteristics of available data. GCM-derived climate
perturbations (at different levels of downscaling) can be used as model input. A
variety of responses to climate change scenarios can hence be modelled. Third,
regional-scale hydrologic models are considerably easier to manipulate than gen-
eral circulation models. Fourth, such regional models can be used to evaluate the
sensitivity of specific watersheds to both hypothetical changes in climate and to
changes predicted by large-scale GCMs. And finally, methods that can incorpor-
ate both detailed regional hydrologic characteristics and output from large-scale
GCMs will be well situated to take advantage of continuing improvements in the
resolution, regional geography, and hydrology of global climate models.
The choice of a model for a particular case study depends on many factors
(Gleick, 1986), while the study purpose, model and data availability have been the
dominant ones (Ng and Marsalek, 1992; Xu, 1999). For example, for assessing
water resources management on a regional scale, monthly rainfall-runoff (water
balance) models were found useful for identifying hydrologic consequences of
changes in temperature, precipitation, and other climatic variables (e.g., Gleick,
1986; Schaake and Liu, 1989; Mimikou et al., 1991; Arnell, 1992; Xu and Halldin,
1997; Xu and Singh, 1998). For detailed assessments of surface flow, conceptual
372 CHONG-YU XU

lumped-parameter models are used. One of the more frequently used models in this
group is the Sacramento Soil Moisture Accounting Model (Burnash et al., 1973).
This model has been used by many researchers in the United States for studying
the impact of climate change (e.g., Nemec and Schaake, 1982; Gleick, 1987; Lett-
enmaier and Gan, 1990; Schaake, 1990; Nash and Gleick, 1991; Cooley, 1990).
Panagoulia (1992) used the same model to assess the effects of climate change on
a basin in central Greece. The HBV model (Bergström, 1976) is widely used in
Nordic countries as a tool to assess the climate change effects (e.g., Vehviläinen
and Lohvansuu, 1991; Saelthun, 1996). Several other models having a similar
structure to the above mentioned two models, but with different process conceptu-
alisations, have been used to assess the effect of climate change on many regions
of the globe (see also Leaveley, 1994). For simulation of spatial patterns of hy-
drologic response within a basin, process-based distributed-parameter models are
needed (Beven, 1989; Thomsen, 1990; Running and Nemani, 1991; Bathurst and
O’Connell, 1992). For estimating changes in the average annual runoff for differ-
ent climate change scenarios simple empirical and regression models were used.
Examples include Revelle and Waggoner (1983) in the United States, and Arnell
and Reynard (1989) in the U.K.
It is fair to say that all kinds of models find their usefulness in different applic-
ations. The models that are complex in terms of structure and input requirements
could be expected to provide adequate results for a wide range of applications; the
more simple models which have smaller range of applications can give adequate
results at greatly reduced cost, provided that the objective function is suitable. The
distinction between simple and physically-based distributed-parameter models is
not only one of lesser or greater sophistication, but also intimately bound up with
the purposes for which such models are to be used. Thus, choosing a suitable model
is equivalent to distinguishing the situation between when simple models can be
used and when complex model must be used.

2.3. GAPS BETWEEN CLIMATE MODELING AND HYDROLOGIC MODELING

While the atmospheric components of the GCMs are often very sophisticated (di-
viding the atmosphere into many layers), Kite et al. (1994) has shown that the
land-phase parameterizations in current GCMs do not agree on predictions of most
hydrological variables, even when all atmospheric forcings are identical. Many
gaps in the relationship between hydrologic modeling and climate modeling exist.
The gaps shown in Table I will be discussed in the following sections, while other
hydrologists might consider other views more appropriate to discuss. It is hoped
that all views will be discussed in the literature so that a significant improvement
in the hydrologic component of climate models will be achieved.
Gap 1: The spatial and temporal scale mismatches between GCMs ability and
hydrology need: General circulation models (GCMs) are the primary tools today
to study and estimate the nature of climate change. Based on the physical laws for
CLIMATE CHANGE AND HYDROLOGIC MODELS 373
Table I. Some existing gaps between GCMs’ ability and hydrology need

Better simulated Less-well simulated Not well simulated

Spatial scales Global Regional Local

Mismatch 500×500 km 50×50 km 0–50 km

Temporal scales Mean annual Mean monthly Mean daily

Mismatch and seasonal

Vertical scale 500 hPa 800 hPa Earth surface

Mismatch

Working variables Wind Cloudiness Evapotranspiration

Mismatch Temperature Precipitation Runoff
Air pressure Humidity Soil moisture

GCMs’ ability declines I

Hydrological importance increases I

the atmopsheric composition and behavior, they attempt to provide a calculable

model of the earth’s climate system, including internal and external forcing as
well as feedback in the climate system. The size of the climate system (atmo-
sphere, oceans, land) and the time range of climate experiments (several decades
to thousands of years) places a heavy constraint on the design of the GCMs. This
leads to spatial and temporal coarseness. For example, hydrological models are
frequently concerned with small, sub-catchment (even hillslope) scale processes,
occurring on spatial scales much smaller than those resolved in GCMs. While
GCMs deal most proficiently with fluid dynamics at the continental scale and oper-
ate on horizontal grid resolutions ranging from 200 to 600 km. Operation on such
large spatial scales prevents explicit modeling of such climate-modifying local
geographic factors as topography and land/water-distribution or vegetation type.
Moreover, although GCMs use short time steps, commonly 10–30 min, cascading
through 10 or more atmospheric layers and then providing information for a range
of climatic variables (e.g., T, P), most verifications of the models have been based
on long term mean simulations for base cases similar to present conditions, and
the most reliable temporal scale to date remains seasonal (e.g., Schulze, 1997).
Hydrological impact models, on the other hand, typically use a time step of one day,
commonly cascading rainfall through two to three soil layers to produce output on
hydrological variables, such as Q, E and 1S. Table I reveals that GCMs ability to
predict spatial and temporal distributions of climatic variables decline from global
to regional to local catchment scales, and from annual to monthly to daily amounts.
374 CHONG-YU XU

However, the hydrological importance of climate predictions increases from global

to local scales and from annual to daily amounts.
Gap 2: The vertical level mismatches between GCMs ability and hydrology
need: Due to the fact that the free troposphere is more spatially and temporally
homogeneous than the earth’s surface, GCMs are more skillful in simulating the
free troposphere climate than the surface climate. However, hydrological models
have to work with surface variables. It is known that the higher the altitude, the
better prediction can be expected from GCMs, but the less correlation with ground
surface variables exists. GCMs output from 500 or 700 mb level is commonly used
(see Table I).
Gap 3: The mismatches between GCMs accuracy and the hydrological import-
ance of the variables: GCMs were conceptually designed to simulate average,
large-scale, atmospheric circulation. Variables, such as wind, temperature and air
pressure field, can be predicted quite well. Precipitation and cloudiness are less
well predicted variables. There are other variables of key importance in hydrologic
regimes, such as runoff, soil moisture and evapotranspiration which are not well
represented by GCMs (e.g., Loaiciga et al., 1996). Table I shows that GCMs sim-
ulation skills decrease from climate variables to hydrological variables, while the
hydrological importance increases along the same direction.
Precipitation is less well simulated because its common events, such as hur-
ricanes and thunderstorms, occur at smaller spatial scale than GCM’s grid size.
Evapotranspiration is not well represented by GCMs since it occurs at model bound-
aries, i.e., it represents an exchange of latent heat and water mass between the
earth’s surface and the atmosphere. Consequently, estimation of runoff from GCM
output as the difference between precipitation and evapotranspiration is bound to
be inaccurate (e.g., Loaiciga et al., 1996). There is an even more important problem
with many GCMs from a hydrological point of view (e.g., Kite et al., 1994). The
problem is that most GCMs contain no lateral transfer of water within the land
phase. Such models carry out a vertical water distribution at each grid point at each
time interval using precipitation, evapotranspiration, and groundwater storages.
However, any ‘water excess’ or overflow is simply discarded and plays no further
role in model computations, so that even if GCMs were able to simulate water
excess correctly, they would still be operating with an incomplete hydrological
cycle.

3. Recent Research Developments and Achievements

The problem of mismatches between GCMs and hydrological models is a chal-
lenging one for both the meteorological and hydrological modeling communities.
The meteorologists who work with large complex models are forced, through com-
putational limitations, to operate on a coarse grid scale but seek to move to finer
scales, i.e., they wish to ‘scale down’. The hydrological community, on the other
hand, is used to modelling at much smaller scales and is having to consider ‘scal-
CLIMATE CHANGE AND HYDROLOGIC MODELS 375

ing up’. To circumvent the problems and narrow the gaps between GCMs ability
and hydrology needs, various methodologies have been developed during the last
20 yr.
(a) Dynamic downscaling (nesting) approaches for generating high-resolution me-
teorological inputs and narrowing gap 1.
(b) Statistical downscaling approaches for simulating local-scale surface variables
based on large-scale free tropospheric variables and/or surface patterns and for
narrowing gaps 1 and 2.
(c) Macroscale hydrological modeling approaches for narrowing gap 3, i.e., for
correcting perceived weaknesses in the representation of hydrological pro-
cesses in GCMs.
(d) Hypothetical scenarios have been used as input to hydrological models to show
the sensitivity to climate change within a reasonable interval.

3.0.1. Area (a): Dynamic Downscaling

Dynamical downscaling has been attempted with three approaches (Rummukainen,
1997): (1) running a regional scale limited area model with the coarse GCM data as
geographical or spectral boundary conditions; (2) performing global-scale experi-
ments with high-resolution AGCMs (atmosphere GCMs), with coarse GCM data as
initial (and partially also boundary) conditions; and (3) use of a variable-resolution
global model (with the highest resolution over the area of interest).
The goal of dynamic downscaling, i.e., to extract local-scale information from
large-scale GCM data, is achieved by developing and using the limited area models
(LAMs) or regional climate models (RCMs). Regional climate models (RCMs)
have recently been developed that can attain horizontal resolution on the order of
tens of kilometres, over selected areas of interest. They have been applied with
relative success to numerous regions (e.g., Giorgi, 1990; Giorgi and Mearns, 1991;
Giorgi et al., 1990, 1994; Jones et al., 1995; Jenkins and Barron, 1997). Compared
with GCMs the resolution of these RCMs is much closer to that of landscape-scale
hydrologic models (LSHMs) and makes coupling of RCMs and LSHMs potentially
suitable for evaluating the effects of hydrologic systems. Coupling between the
scales can be one way (e.g., Leavesley et al., 1992; Hostetler and Giorgi, 1993) or
bi-directional (e.g., Giorgi and Mearns, 1991; Resso and Zack, 1994).
The main shortcomings of the dynamic modeling include that RCMs still re-
quire considerable computing resources and are as expensive to run as a global
GCM; these models still cannot meet the needs of spatially explicit models of
ecosystems or hydrological systems, there will remain the need to downscale the
results from such models to individual sites or localities for impact studies (Wilby
and Wigley, 1997).