Climate Change and Water Resources in South Asia - Chapter 2 doc

2
Hydrologic Modeling Approaches for Climate
Impact Assessment in South Asia
2.1 INTRODUCTION
The hydrologic and water resources problems in South Asia are discussed in Chapter 1.
It is anticipated that the problems will be exacerbated if basin-wide temperature and
precipitation would change due to climate change. Quantification of possible changes in
river discharge (mean or peak) is achieved with the application of hydrologic models. Four
types of hydrologic model - empirical, water-balance, conceptual lumped-parameter and
process-based distributed models - are used for hydrologic modeling.
A model is usually selected depending on the purpose of the application which
includes: runoff-simulation; sediment transport and morphological changes; estimating
ground water and changes in ground water volume; forecasting flood volume, depth and
duration; assessing changes in land-use; and assessing impacts of changes in climate.
Availability of data and resources are also governing factors in a model selection process.
This chapter discusses the comparative advantages and limitations of various hydrologic
models and their suitability for estimating changes in mean annual and mean peak
discharge under selected climate change scenarios for the river basins in South Asia. It
examines reduction of input variables for empirical modeling through the sensitivity
analysis of runoff to changes in temperature and precipitation. This chapter also discusses
application of hydrologic models in Bangladesh as a case study to assess climate change
impacts.
2.2 HYDROLOGIC MODELS
In planning for water resources and extreme events like floods and droughts, it is essential
to know the precipitation-runoff processes in the vegetation, land surface and soil
components of the hydrologic cycle. These processes differ in arid, semi-arid and humid
climates. Even within a single climate zone, physical processes can vary widely because of
the diversity of vegetation, soils and microclimates.
Hydrologic models describe these processes by partitioning the water among the
various pathways of the hydrologic cycle (Dooge, 1992). Mathematically, hydrologic models
incorporate a set of assumptions, equations and procedures intended to describe the

performance of a prototype (real-world) system (Linsley et al., 1988). Because of the
M. MONIRUL QADER MIRZA
Copyright © 2005 Taylor & Francis Group plc, London, UK
increase in computing capacity, complex mathematical descriptions of the physical
processes of the hydrologic cycle can now be incorporated into hydrologic models.
However, because of variations in physical parameters and the limitations of our
knowledge and understanding about the complexity of the hydrological processes, no
‘hydrologic model’ is able to reproduce fully the prototype processes. Accuracy of the
model is highly dependent on factors such as: adequacy of empirical, statistical and
mathematical descriptions of the physical processes; the quantity and quality of input data;
the extent of basin coverage; and the magnitude of variability in physical parameters.
There are two main aims for using simulation modeling in hydrology. The first is to
explore the implications of making certain assumptions about the nature of the real-world
system. The second is to predict the behavior of the real-world system under a set of
naturally occurring circumstances (Beven, 1989). In order to meet these aims, different
types of hydrologic models are required.
There are four types of hydrologic models - empirical, water-balance, conceptual
lumped-parameter and process-based distributed models. The choice of model type
depends partly on the purpose of the application including: simulating runoff, sediment
transport and morphological changes; estimating ground water and changes in ground
water volume; forecasting flood volume, depth and duration; assessing changes in
land-use; and assessing impacts of changes in climate. The choice of model also depends
on the availability of data and resources. The various types of hydrologic models and their
advantages and limitations are discussed below.
2.2.1 EMPIRICAL MODELS
In hydrological modeling, empirical models are generally developed and used for
prediction and estimation purposes. These models do not explicitly consider the physical
laws governing the processes involving precipitation, temperature, vegetation and soils
(Singh, 1988). However, they do implicitly incorporate the fundamental physical fact that,
generally, variations in runoff tend to respond proportionally to the variations in climate.

Empirical models are developed based on a ‘black box’ modeling approach where
empirical equations are used to relate runoff and rainfall, and only the input (rainfall) and
output (runoff) have physical meanings. Through statistical techniques, empirical models
reflect only the relations between input and output for the climate and basin conditions
during the time period for which they were developed. These models provide a much
more simplified view of reality, particularly when regression techniques are employed
(Kirkby et al., 1987). The accuracy of models largely depends on the magnitude of error
inherent in the input and output data. As the empirical models are developed with input
and output data within a certain range and time period, caution should be exercised
regarding the extension of the relationship for climate conditions different from those used
for the development of the function (Leavesley, 1994). Models developed for a particular
river basin cannot be applied to a different basin. Although empirical models are often
criticized for these limitations, they are widely used compared to other models.
Despite their limitations, empirical models have some distinct advantages over other
types of hydrologic models. For example, they are relatively easy to develop, require less
data, can be calibrated simply, require fewer resources, and do not need a huge computing
capacity. When other models cannot be developed or used because of the paucity of data,
empirical models can be developed for various purposes. In many situations, empirical
models can yield accurate results and can, therefore, serve a useful purpose in
24 HYDROLOGIC MODELING APPROACHES
Copyright © 2005 Taylor & Francis Group plc, London, UK
decision-making (Singh, 1988). In hydrology, empirical models are generally useful in
estimating the mean annual flood, monthly and annual mean discharge and bankful
discharge (Garde and Kothyari, 1990; Kothyari and Garde, 1991; Mosley, 1979 and 1981;
Schumm, 1969; Thomas, 1970; Rodda, 1969; Leopold and Millier, 1956; Natural
Environment Research Council (NERC), 1975; Beable and Mckerchar, 1982).
There are two important issues which need to be taken into account before
developing an empirical model for estimating discharge and floods. First, empirical models
require very good spatial distribution of precipitation. Ideally, this can be achieved by
acquiring long-term records of precipitation for a large number of stations uniformly

distributed over a river basin, covering high and low elevations. Similarly, long-term records
of temperature are also necessary if temperature effects are to be considered.
Second, a fairly good record of discharge (or runoff) from downstream stations is
needed. However, if there is any diversion of flows through the distributary (ies) or by any
other means in the upstream areas, this has to be taken into account depending on the
magnitude of the diversion.
2.2.2 WATER-BALANCE MODELS
Water-balance models were first developed by Thornthwaite (1948) in the 1940s and were
subsequently revised by Thornthwaite and Mather (1955) and by others. Palmer (1965)
used a water-balance model similar to that of the Thornthwaite model while developing an
index of meteorological drought. Thomas (1981) presented an alternative water-balance
model with several new features. These water-balance models have very simple structures
and are characterized by a limited number of parameters. This kind of model is essentially
a ‘book-keeping procedure,’ which uses the following fundamental equation to estimate
the balance between the precipitation (as rain and snowmelt), loss of water by
evapo-transpiration, stream flow and recharge into the ground water:
where P is the precipitation, R is the runoff, G is ground water runoff,
∆S is the changes in
storage (snow and soil water) and E is evapo-transpiration. The typical structure of a
water-balance model is shown in Figure 2.1.
The models can be simple to complex depending on the details of each of the
components of the equation (2.1). Most water-balance models calculate direct runoff from
precipitation and lagged runoff from the basin storage in the computation of the total
runoff (R). The sensitivity and accuracy of water-balance models often depend on the
method of calculating potential evapo-transpiration (PET). Various PET-models are
available among which Penman (1948), Thornthwaite (1948), Blaney and Criddle (1950),
Monteith (1964), Priestley and Taylor (1972), and Hargreaves (1974) are important (see
cited references for descriptions of these models). The selection of the PET model is largely
dependent on the availability of sufficient climate data, which varies from place to place.
Most models compute E as a function of potential ET and water available in soil storage

(S). Various methods are in use for calculating E from the PET and soil moisture deficit
relationship, including linear, layered and exponential methods.
One advantage of water-balance models is that they can potentially be used to
determine changes in seasonal snow storage and melt. Within a water-balance model, the
storage and melting processes of snow are described by two types of model:
energy-balance and temperature-index models.
M. M. Q. MIRZA 25
Copyright © 2005 Taylor & Francis Group plc, London, UK


Evapo-Transpiration

Precipitation

Snow Storage

Soil Storage

Soil Storage
Base Flow/
Delayed Flow

Surface Runoff

Total Runoff
Fig. 2.1 Typical structure of a water-balance model.
The energy-balance models simulate the flow of mass and energy in the snow cover.
The energy-balance approach for calculating snowmelt applies the law of conservation of
energy to a control volume. The control volume has its lower boundary as the
snow-ground interface and its upper boundary as the snow-air interface. The use of a

volume allows the energy fluxes into the snow to be expressed as internal energy changes
(Gray and Prowse, 1993). The energy balance model is physically or meteorologically
more explicit than the temperature-index model. It contains parameters that can be
extrapolated to a certain degree of confidence from weather maps or from regional climate
models (Kuhn, 1993). Various studies have used energy-balance models to estimate runoff
from snowmelt (Fitzharris and Grimmond, 1982; Granger and Gray, 1990; Gray and O’Neill,
1974; and Gray and Landine, 1987). Details of an energy-balance model can be found in
Gray and Prowse (1993).
The second type of model is the temperature-index snowmelt model (Equation (2.2)).
Despite its simplicity, the model is widely used in forecasting discharge in snow-covered
basins. Using monthly data, for example, Kwadijk (1993) applied a temperature-index
snowmelt model in order to assess the impact of climate change on the Rhine River basin
and found close fit between the simulated values and observed data. While modeling the
effects of climate change on water resources in the Sacramento River basin in the USA,
Gleick (1987) found poor performance of a temperature-index snowmelt model using
monthly data. The temperature-index models for rain-free and rain conditions are as
follows:
(i) Rain-Free Condition
26 HYDROLOGIC MODELING APPROACHES
Copyright © 2005 Taylor & Francis Group plc, London, UK
where M = snowmelt in mm
M
f
= snowmelt factor
T
i
= index temperature
T
b
= base temperature (set as 0

o
C)
(ii) During Rain
For a rain event, the melt factor is modified as follows:
M
f
= (0.74 + 0.007P) (T
i
– T
b
)
where P = precipitation (in mm)
Snowmelt is calculated by:
Overall, water-balance models incorporate soil-moisture characteristics of regions,
allow monthly, seasonal, and annual estimates of hydrologic parameters, and use readily
available data on meteorological phenomena, soil, and vegetation characteristics. They
can often provide efficient estimates of surface runoff when compared to measured runoff,
reliable evapo-transpiration estimates under many climate regimes, and estimates of ground
water discharge and recharge rates. Typical data requirements are precipitation,
temperature, sunshine hour, wind speed, information on characteristics of vegetation (which
may include type of vegetation for estimating rooting depths), and soil (such as field
capacities and wilting points). While generally the water-balance models require huge
amounts of data, they can nevertheless be applied in reasonably large areas with sparse
data (Hare and Hay, 1971; Brash and Murray, 1980). For example, Hare and Hay (1971)
applied the Lettau’s (1969) empirical model to approximate precipitation in order to
analyze the anomalies in the large-scale annual water-balance over Northern North America.
Brash and Murray (1980) estimated adjusted equilibrium precipitation from an
energy-balance equation. The estimated precipitation was then used to estimate water
yield in the Taieri catchment in New Zealand and found to be very closely matched with the
measured data. Note, however, that these energy balance techniques require reliable net

radiation data, which are not readily available for the major river basins in South Asia.
By integrating hydrologic advances with existing water-balance techniques, new
insights into hydrologic processes and environmental impacts can be gained for climate
impact assessments. Furthermore, water-balance models are well suited to the current
generation of microcomputer software and hardware. A number of water-balance models
have been developed to assess the impact of climate change on river runoff and soil
moisture stress from wet to dry regions (Mather and Feddema, 1986; McCabe and Wolock,
1992; Thompson, 1992; Flaschka et al., 1987; McCabe and Ayers, 1989; Conway, 1993
and Kwadijk, 1993). These studies show various magnitudes of runoff and soil moisture
sensitivities on monthly time-scales to possible changes in climate. Overall, such studies
demonstrate that the water-balance approach holds good potential for application in the
river basins of South Asia (subject to availability of the required data) in order to assess
effects of climate change on hydrology and water resources.
M. M. Q. MIRZA 27
Copyright © 2005 Taylor & Francis Group plc, London, UK
2.2.3 CONCEPTUAL LUMPED-PARAMETER MODELS
Conceptual lumped-parameter models are developed based on approximations or
simplifications of physical laws. These models embody a series of functions which are
considered to describe the relevant catchment processes. The algorithms are usually
simplified by the use of empirical relations in order to speed the solution and to adapt the
model to cope with the point-to-point variations in the hydrologic processes within the
catchment (Crawford and Linsley, 1968; Boughton, 1968; Linsley et al., 1988; Leavesley,
1994). They contain parameters, some of which may have direct physical significance and
can, therefore, be estimated by using concurrent observations on input and output. Some
widely-used models of this category are: the Sacramento Soil Moisture Accounting model
(Burnash et al., 1973), the Institute of Royal Meteorology Belgium (IRMB) model (Bultot
and Dupriez, 1976), the HBV model (Bergstorm, 1976), the Hydrologic Simulation
Program-FORTRAN (HSPF) model (USEPA, 1984), the Erosion Productivity Impact
Calculator (EPIC) model (Williams et al., 1984) and the MODHYDROLOG (Chiew and
McMahon, 1993). A schematic diagram of a conceptual lumped-parameter model

(MODHYDROLOG) is shown in Figure 2.2.
In the conceptual lumped-parameter models, the vertical and lateral movement of
water with respect to time is incorporated. Variations in respect of space are ignored. The
vertical processes of water movement include interception storage and evaporation,
infiltration, soil-moisture storage, evapo-transpiration, percolation to ground water
storage, snow-pack accumulation and melt, and capillary rise. The horizontal processes
include surface runoff, interflow, ground water flow, and stream flow. Components of the
vertical and lateral processes are integrated. The model development starts with the
vertical processes. Interception storage is assumed and calibrated usually by trial and
error. Empirical algorithms are used for calculating the evaporation from the surface
storage.
For calculating infiltration calculation, two methods are in practice. First, the
maximum infiltration rate is assumed from the field observations and then the infiltration
rate is expressed as a function of soil storage (Boughton, 1968). Second, some prominent
infiltration models, such as Green-Ampt (1911), Philip (1957 and 1969) and Holtan (1961),
can be used directly. For example, the Hydrologic Engineering Center’s HEC-1 model
uses the Green-Ampt and Holtan’s infiltration models. One of the important limitations of
using these models is the need to estimate a number of parameters, some of which have to
be estimated either from laboratory experiments or from field observations.
The other vertical and horizontal components that need to be developed are
evapo-transpiration, percolation and base flow. Evapo-transpiration is usually calculated
as a function of soil moisture storage, soil moisture storage capacity and potential
evapo-transpiration (Chiew and McMahon, 1993). A constant is used to calculate the
percolation to ground water storage. Another constant is used to estimate the base flow
from the ground water storage. The base flow constant is usually determined by calibrating
the estimated flows with the observed values.
Lumped-parameter models have some distinct advantages. They do not necessarily
require direct use of mathematical equations of physical processes and they take into
account more physical processes than water-balance models. They also have been shown
to be capable of making acceptable estimates of stream flow, evapo-transpiration, soil

moisture deficits, and storage changes, including changes in ground water storage, for
smaller river basins.
28 HYDROLOGIC MODELING APPROACHES
Copyright © 2005 Taylor & Francis Group plc, London, UK
Fig. 2.2 Schematic representation of the MODHYDROLOG daily rainfall-runoff model. Source:
Courtesy of Chiew and McMahon, 1994.
Although lumped-parameter models are widely used, they have a number of
limitations. These include: (1) the equations of a lumped-parameter model can only be
approximate representations of the real world and must introduce some error arising from
the model structure; (2) spatial heterogeneities in system responses may not be well
reproduced by catchment-averaged parameters (Sharma and Luxmore, 1979; Freeze, 1980);
(3) the accuracy with which a model can be calibrated or validated is very dependent on
the observations of both inputs and outputs (Ibbit, 1972; Hornberger et al., 1985). Since
input variables, particularly evapo-transpiration estimates, may be subject to considerable
uncertainty; (4) there is a great danger of over-parameterization if attempts are made to
simulate all hydrological processes thought to be relevant and to fit those parameters by
optimization against an observed discharge record (Hornberger et al., 1985), so three to
five parameters should be sufficient to reproduce most of the information in a hydrological
record; and (5) the calibrated parameters of such models may be expected to show a
degree of interdependence, so that equally good results may be obtained with different sets
M. M. Q. MIRZA 29
Copyright © 2005 Taylor & Francis Group plc, London, UK
of parameter values, even though a model has only a small number of parameters
(Ibbitt and O’Donnell, 1971; Pickup, 1977; and Sorooshian and Gupta, 1983).
Another potential disadvantage is that the use of lumped-parameter rainfall-runoff
models depends essentially on the availability of sufficiently long meteorological and
hydrological records for their calibration. Such records are not always available. Their
calibration also involves a significant element of curve fitting, thus making any physical
interpretation of the fitted parameter values extremely difficult. There are other
limitations, too. Because of their inherent structure, these models also make very little use

of contour, soil, and vegetation maps, or of the increasing body of information related to
soil physics and plant physiology. These models are not suitable for predicting the effects
of land-use changes on the hydrological regime of a catchment, particularly when only a
part of the catchment is affected.
In the case of the lumped models, parameter values are highly dependent on both the
model structure and the period of calibration (Beven and O’Connell, 1982). Therefore, as
with other hydrologic models, it is not advisable to extrapolate events that are outside the
conditions over which the model parameters are estimated.
2.2.4 PHYSICALLY-BASED DISTRIBUTED MODELS
Neither the empirical nor the lumped models are capable of addressing the physical
processes of the basin which control the basin response, as they do not account for the
spatial distribution of basin parameters. This limitation prompted the development of
physically-based models aimed at improving the understanding of catchment processes. A
schematic diagram of the Système Hydrologique Européen (SHE) distributed model is
shown in Figure 2.3.
Fig. 2.3 Schematic representation of the SHE model. Source: Adapted from Abbot et al., 1986.
30 H
YDROLOGIC MODELING APPROACHES
Copyright © 2005 Taylor & Francis Group plc, London, UK
Physically-based distributed models require descriptive equations for the hydrological
processes involved (Freeze and Harlan, 1969). The equations on which distributed models
are developed generally involve one or more space coordinates. They thus have the
capability of forecasting the spatial pattern of hydrologic conditions within a catchment as
well as the simple outflows and bulk storage volumes. In general, the descriptive equations
are non-linear differential equations that cannot be solved analytically for cases of practical
interest. Therefore, for simplification, some empirical discretization is made. Indeed, the
complexities of hydrological systems are such that all the model components ultimately
rely on an empirical relationship.
As discussed by Freeze and Harlan (1969), the development of a computational model
to simulate physical processes is carried out by: (1) defining a physical system isolating a

region of consideration with simplified boundaries and neglecting all physical processes
non-essential to the phenomenon being studied; (2) representing the idealized and
simplified physical system by a mathematical model, including governing differential
equations and boundary/initial conditions; (3) converting the mathematical model into a
numerical model using one of the numerical methodologies (finite difference, finite
element, boundary element, and characteristics methods) which is most appropriate to the
problem; and (4) writing a computer code based on the selected computational algorithm
to obtain numerical results in still graphic or animated form. In other words, before
the computational model is developed, numerous idealizations, simplifications,
approximations and discretizations have to be made.
Regarding calibration of the physically-based model, the theoretical idea is that the
model has the potential to estimate parameter values by field measurements without
having to carry out parameter optimization as required by the simpler models of the lumped,
conceptual type (Abbott et al., 1986). But in reality, the situation is different. Such an ideal
situation requires comprehensive field data covering all parameters and a model discretization
to an appropriate scale (Refsgaard et al., 1992). For example, the SHE model was applied
to the Wye catchment in England and in six small catchments in the Narmada basin in India
(Bathurst, 1986 and Refsgaard et al., 1992). In these catchments, during the application,
optimizations were carried out because of inadequate representation of the hydrological
processes, insufficient data, and the possible difference in scale between the measurement
and the model grid scale (Bathurst, 1986 and Refsgaard et al., 1992).
The distributed nature of physically-based models offers some advantages over other
types of models. For example, they are capable of forecasting the effects of land-use changes,
the effects of spatially variable inputs and outputs, the movement of pollutants and
sediments, and the hydrological response of ungauged catchments. Regarding land-use
changes in a catchment, deforestation rarely takes place abruptly over a complete basin;
it is more common for piecemeal changes to take place over a considerable period of time.
In a distributed model the effects of such changes can be examined in their correct spatial
context.
It is clear from the above discussion that physically-based models require much more

information than their empirical, water-balance or lumped-conceptual counterparts. Thus,
calibration and validation emerge as major tasks. Extensive field measurements require
huge amounts of resources and time, and computing capacities are high. Finally, despite
the greater effort required to parameterize, validate and run physically-based models, the
simulated results often provide only slightly better (or sometimes worse) correspondence
with measured values than lumped-conceptual models (Beven, 1987; Logue, 1990; and
Wilcox et al., 1990). Perhaps this results from the equations used to describe the physical
variability and the high degree of temporal and spatial variability of critical input
M. M. Q. MIRZA 31
Copyright © 2005 Taylor & Francis Group plc, London, UK
parameters. Ironically, the description of physical variability is presumed to be a strength
for physically-based models (Beven, 1985; Bathurst and O’Connell, 1992).
Regarding extrapolation of physically-based distributed models, Beven and O’Connell
(1982) mentioned that, because of the physical basis of the model parameters, the
measured parameters’ values might be extrapolated to other locations or time periods.
However, response of the physical parameters at other locations or other time periods may
not be same. Therefore, physically-based distributed models also have limitations
regarding extrapolation. Comparisons of various hydrologic models are tabulated in
Table 2.1.
In this section, the advantages and limitations of various hydrologic models have been
discussed. In the next section, the applicability of some hydrologic models for assessing
the impact of climate change on water resources is discussed.
2.3 ADVANTAGES AND LIMITATIONS OF HYDROLOGIC MODELS IN
CLIMATE CHANGE APPLICATION
A number of studies have been carried out to assess the impacts of climate changes using
empirical, water-balance and lumped-parameter models (Revelle and Waggoner, 1983;
Mather and Feddema, 1986; McCabe et al., 1990; McCabe and Wolock, 1992;
Thompson, 1992; Flaschka et al., 1987; MaCabe and Ayers, 1989; Conway, 1993; and
Kwadijk, 1993). All these studies used monthly precipitation and temperature time-series
data for the assessment. Models were calibrated to the observed data and then validated

against the other observed dataset in order to assess the capacity of the model to generate
current hydrological output (for example, runoff). Finally, the models were used to predict
the possible effect of future climate change on water resources. Most of the models used
GCM-based and hypothetical climate scenarios for sensitivity analysis.
In the applications noted above, the model parameters were estimated from the
current climate as a basis for predicting future conditions. This is one of the major
limitations of modeling the effects of climate change. The behavior of physical
parameters of a catchment is not necessarily stationary overtime. For example, most
pedological processes operate over a very long time-scale, but changes in organic matter
content and soil structure may become apparent over a time-scale of less than 10 years
(Climate Change Impact Review Group (CCIRG), 1991). Higher temperatures and
increased rainfall would lead to a loss of soil organic matter and hence a decrease in ability
of the soil to hold moisture; higher temperatures would also encourage clayey soil to
shrink and crack, thus assisting the passage of water into and through the soil profile
(CCIRG, 1991). Another issue is the response of vegetation to climate changes. For
example, Idso and Brazel (1984) estimated that plant evapo-transpiration may be decreased
by one-third for a doubling of carbon-dioxide due to partial stomatal closure in plants,
increasing their water use efficiency and conserving soil moisture for increased runoff
to rivers and streams. Thus, as CO
2
concentrations change over time, so might the
relationships between climate and hydrology. Indeed, Dooge (1992) suggested that
research should not be used to develop more complex models until the issue of the
“antitranspirant effect” of higher atmospheric CO
2
enrichment is effectively resolved.
Which type of model should be chosen for assessing changes in runoff from scenarios
of climate change? Empirical models can be applied successfully if the processes are
ignored and the objective is limited to predicting runoff or discharge on monthly or annual
time-scales. Empirical models require less data than the other models. The model

performance during the calibration and validation period is highly dependent on good
spatial and temporal coverage of the input data.
32 HYDROLOGIC MODELING APPROACHES
Copyright © 2005 Taylor & Francis Group plc, London, UK
M. M. Q. MIRZA 33
Copyright © 2005 Taylor & Francis Group plc, London, UK
34 HYDROLOGIC MODELING APPROACHES
Copyright © 2005 Taylor & Francis Group plc, London, UK
Water-balance models are equally suitable for monthly and annual time-scales, but
require more data. Calibration and validation are relatively more time and resource
consuming as compared to empirical models. Conceptual lumped-parameter models need
more data than either empirical or water-balance models. These models can be applied at
shorter time-scales (say on a daily or hourly basis), a distinct advantage over empirical and
water-balance models. However, calibration and validation procedures for these models
are much more complicated. Physically-based distributed models require a vast amount
of data, which are often impracticable to collect. These models need laboratory
experimentation to estimate parameter values. Calibration and validation procedures are
much more complex than for other models and computing (and other resource) demands
are higher. Finally, physically-based models may not necessarily improve the accuracy of
outputs compared to other models.
This last point is particularly important to consider when choosing between complex
and simpler models. For example, while applying the SHE model in the Narmada basin in
India, Refsgaard et al. (1992) concluded that the simulated results of the rainfall-runoff
were of the same degree of accuracy as would have been expected with similar
hydrological models of the lumped-parameter type. They concluded that the results
obtained in the Narmada basin do not justify the application of an advanced model, such as
the SHE, where the objective is limited to rainfall-runoff modeling.
2.4 APPLICATION OF HYDROLOGIC MODELS FOR CLIMATE CHANGE
IMPACT ASSESSMENT IN BANGLADESH
Based on the comparative advantages, limitations, and suitability of various hydrologic

models with respect to research purposes, data availability, scale and resources, Mirza
(1997) applied a suite of empirical model and MIKE 11-GIS hydrodynamic model for:
(1) determining the sensitivity of mean annual and mean peak river discharges in the Ganges,
Brahmaputra and Meghna (GBM) basins (Fig. 2.4) in Bangladesh to future climate changes;
and (2) estimating the consequent changes in flood magnitude, depth and extent.
Fig. 2.4 The Ganges, Brahmaputra and Megna basins.
M. M. Q. M
IRZA 35
Copyright © 2005 Taylor & Francis Group plc, London, UK
Broadly, the objectives of the application were centered on relationships between
precipitation, temperature and discharge, and the time-scale of concern was annual. Based
on the above discussion the use of lumped-parameter or process-based distributed models
for these purposes was not found practicable. Note that complex hydrologic models
generally operate over small time intervals, such as an hour or day. Consequently, the field
measurement, determination, estimation and optimization procedures can be onerous. For
this reason, complex models are usually more suitable for smaller, more manageable
catchments. For example, in the application of the SHE model to the small Wye catchment,
it was still necessary to specify about 2,400 parameter values (Beven, 1989). By
comparison, the combined GBM basins are approximately 1.75 million sq. km in area. For
these river basins, the number of parameter values that would need to be specified using
the same modeling approach is unmanageably large. The water-balance approach had a
good potential for application in the GBM basins in order to determine the effects of
climate change on annual discharge and flooding in Bangladesh. But the use of the
water-balance approach was hindered by the lack of adequate hydro-meteorological
(radiation, wind speed and humidity) and land-use data. Although the water-balance
approach has been employed successfully in smaller basins with sparse data, as
discussed in Section 2.2, the shear size and geographical diversity of the GBM basins
(1.75 million sq. km) dictates against its use and it would probably create more
uncertainties than it would resolve. Therefore, given the time and resources available,
Mirza (1997) decided to apply a simpler empirical approach in combination with the

MIKE 11-GIS hydrodynamic model. It is worth noting that simple empirical models have
already been developed and applied successfully for similar purposes in the Himalayan
region (Khosla, 1994; Garde and Kothyari, 1990 and Kothyari and Garde, 1991).
Following sub-sections describe the processes involved in empirical models and simulating
their results in the MIKE 11-GIS hydrodynamic model.
2.4.1 THE RELATIVE SENSITIVITY OF RUNOFF TO PRECIPITATION AND
TEMPERATURE
What is the relative importance of precipitation and temperature in affecting runoff in the
Ganges, Brahmaputra and Meghna basins? Precipitation and temperature (as it affects
evaporation and transpiration) are the principal climate driving forces in generating runoff.
Therefore, runoff is mainly sensitive to these two meteorological inputs.
The sensitivity of runoff to temperature and precipitation changes can be
approximately determined by using the water-balance equation (see equation (2.1)). In
many cases it is assumed that the catchment is watertight and that no inflow or outflow of
ground water occurs. On an annual basis it is often assumed that no change in storage
takes place from year to year. Therefore, equation (2.1) reduces to:
In equation (2.3), E cannot be determined directly. However, if the annual open-water
evaporation in any place is known or determined using the PET model of Penman (1948),
E can be estimated by the following empirical model (Pike, 1964):
36 HYDROLOGIC MODELING APPROACHES
Copyright © 2005 Taylor & Francis Group plc, London, UK
The equation for the future runoff under the climate change can be written as:
The percentage change from the present day runoff can be determined using the
following expression:
In order to estimate the sensitivity of runoff to temperature and precipitation, one
station has been selected from each of the river basins - the Ganges, Brahmaputra and
Meghna. These stations are New Delhi, Gauhati and Sylhet. For the sensitivity analysis an
approximately 5% change in precipitation is associated with each degree change in global
mean temperature, in accordance with global estimates from GCM simulations
(IPCC, 1990), in order to select the range of precipitation change. The sensitivity of runoff

has been calculated applying equations (2.6), (2.7) and (2.8).
The results of the sensitivity analysis are presented in Figures 2.5a, 2.5b, and 2.5c.
In general, the gentle slopes of equal percentage change lines show that runoff change is
more sensitive to precipitation change than to temperature change. The results also show
that, in percentage terms, runoff is more sensitive to precipitation and temperature changes
in relatively dry stations than wet stations. As an example, in the case of the New Delhi
station (a drier station) no change in temperature and a 4% increase in precipitation changes
runoff by +11%, while for the Gauhati and Sylhet (the wetter stations) the changes in
runoff are +6% and +8%, respectively. In the extreme case, a 5
o
C increase in temperature
and a 20% increase in precipitation could increase runoff by 29% at the New Delhi station,
whereas for Gauhati and Sylhet stations the expected changes are 22% and 21%,
respectively.
2.4.2 THE EMPIRICAL MODEL DEVELOPMENT PROCESS
The sensitivity analysis in preceding sub-section shows that runoff in the Ganges,
Brahmaputra and Meghna River basins appears to be much more sensitive to changes in
precipitation than to changes in temperature. Based on this analysis, it was decided to use
only precipitation as the independent variable in developing empirical models for the three
river basins.
2.4.2.1 STEP I: DATA IDENTIFICATION AND ACQUISITION
The sensitivity analysis described above has facilitated the reduction of variables by
excluding temperature. The main data requirements are identified as annual precipitation,
annual mean discharge and annual peak discharge. Annual precipitation can be derived
from daily or monthly records, or by directly using annual totals depending on availability.
Similarly, annual mean discharge can be calculated from the daily observations or monthly
mean values. Peak discharge will be the highest daily observed value in a year.
M. M. Q. MIRZA 37
Copyright © 2005 Taylor & Francis Group plc, London, UK
Fig. 2.5 Sensitivity of runoff to temperature and precipitation changes in the: (a) Ganges basin (New

Delhi), (b) Brahmaputra basin (Gauhati) and (c) Meghna basin (Sylhet).
(a)
(b)
38 H
YDROLOGIC MODELING APPROACHES
(c)
Copyright © 2005 Taylor & Francis Group plc, London, UK
2.4.2.2 STEP II: DATA QUALITY ASSESSMENT
Data quality requires an assessment of the length of records, homogeneity and missing
observations. In India, Bangladesh and Nepal, station history is seldom available, which
makes an assessment of homogeneity difficult. Standard procedures (Salinger, 1980) can
be applied to fill in missing observations. If more than one dataset is available, the best
dataset can be selected based on various criteria including the length of record, spatial
coverage and volume of missing observations.
2.4.2.3 STEP III: EMPIRICAL MODEL BUILDING
This step involves building empirical models for the purpose of assessing possible changes
in flood discharge in Bangladesh due to changes in precipitation. Three empirical
relationships are required: a) the relationship between annual precipitation and annual mean
discharge; b) the relationship between annual mean discharge and annual peak discharge;
and c) the stage-discharge rating equations. Future changes in flood discharge and stages
can be determined using the empirical relationships for (a), (b), and (c).
The empirical relationships in (a) and (b) can be linear or non-linear while in (c) the
stage-discharge relationship is non-linear. These can be checked by plotting the y variable
against the x variable(s). Linearity or non-linearity can be bi-variate or multi-variate
depending on the number of independent variables. For the regression model building,
non-linearity can be transformed to linearity by applying standard transformation methods
(McCuen and Snyder, 1987; Box and Cox, 1964). Standard procedures were followed in
developing the empirical models and their adequacy for prediction purpose were also
examined (Mirza, 1997; Mirza et al., 2003). Results of climate change impacts are
presented and discussed in Chapter 6 on Bangladesh Country Study. Stage-discharge

relationships were not developed because these are already built into the MIKE 11 model,
which is used to estimate the effects of changes in peak discharge and local precipitation
on flood extent and depth as a consequence of climate change.
2.4.3 SIMULATION WITH THE MIKE 11-GIS MODEL
The French Engineering Consortium (FEC) (1989) first proposed to use geographic
information systems (GIS) for flooded area and inundation depth mapping for Bangladesh
in the late 1980s. Subsequently, GIS was widely used in the Flood Action Plan (FAP)
studies for various purposes, including flood mapping. Reasonably accurate flood
mapping requires inputs (such as discharge and water levels) from flood models.
In Bangladesh, by and large, flood modeling has been carried out using the MIKE 11
flood model, maintained by the Institute for Water Modeling (IWM in Dhaka). The
MIKE 11 software package models the flows and water levels in rivers and estuaries. It is
used as a tool to simulate flooding behavior of rivers and floodplains. The models
numerically represent the river and floodplain topography and are calibrated to recorded
flood levels and discharges (FAP 25, 1994). The MIKE 11 model is based on an efficient
numerical solution of the complete non-linear equations for 1-D flows. A network
configuration represents the rivers and floodplains as a system of connected branches. The
inputs are daily discharge and water levels at the boundary of Bangladesh and rainfall over
the area covered by the model. At discrete points along the branches, flood levels
(at h-points) and discharges (at Q-points) are calculated at hourly time-steps as a function
of time (Fig. 2.6). However, flood models do not themselves generate the flood maps.
M. M. Q. MIRZA 39
Copyright © 2005 Taylor & Francis Group plc, London, UK

Branch
h-point
Q-point
Figure 2.6 A MIKE 11 network is an interconnected system of branches representing rivers and
floodplains. Along the branches h-points and Q-points are located. Flood levels are calculated at
h-points and discharge at Q-points. Source: FAP 25, 1994.

In order to generate flood maps, GIS techniques (available with ARC/INFO GIS) are
applied in combination with the MIKE 11 model. In a MIKE 11-GIS model, the inputs from
the MIKE 11 model include: information on flood model network, cross-section databases,
and results from flood simulations (which include water levels and discharge over time
throughout the river system). A wide variety of data can be held in the GIS, such as: ground
elevations in the form of Digital Elevation Model (DEM); rivers; roads; beels/lakes;
settlements; and satellite images. Critical information on river and floodplain topography,
rainfall, discharges and water levels are fed into the MIKE 11 model from the GIS (Fig. 2.7).
Thus, for the purpose of floodplain mapping, information from both systems (MIKE11 and
GIS) is related in the combined MIKE 11-GIS model (Fig. 2.8). More particularly, the
MIKE 11 cross-sectional databases are exported to the MIKE 11-GIS model to display
cross-sectional profiles, and for merging river cross-sections with the DEM of floodplain
profiles. Imported flood model simulation results are used for flood mapping, graphing, and
statistical output (FAP 25, 1994).
The elevations in a DEM used by the MIKE 11-GIS model are derived from three
basic types of topography: floodplains; high ground; and features which depress or raise
the floodplain (for example, rivers, khals, beels, roads, embankments and settlements).
The floodplain is characterized by very flat topography, while the high ground is typically
steep.
2.4.4 SIMULATION OF CHANGES IN FLOOD DEPTH AREAL EXTENT
Empirical models were used to calculate changed mean and 20-year peak discharge values.
It is obviously not realistic, however, to carry out flood simulation with constant high
discharges, as this will lead to an overestimate of inundation areas and depths. Rather, the
peak discharge must be associated with a realistic temporal distribution of discharges
throughout the season, as input to the model. In order to overcome this problem, boundary
inflows for the three rivers were selected for a “typical” year. (The term “typical” is
attributed to temporal distribution of discharge rather than the magnitude). Available records
show that the inflows of the 1991 monsoon represent a temporal distribution which may be
considered fairly “typical” with regard to the usual peaking time of the three rivers
40 HYDROLOGIC MODELING APPROACHES

Copyright © 2005 Taylor & Francis Group plc, London, UK
(Chapter 1). In 1991, discharges of the Ganges and Brahmaputra peaked in September and
July, respectively.

River Network
(Line Coverage)
Rain Gauges
(Point Coverage)

Catchments
(Polygon Coverage)
DEM
(Grid)
Fig. 2.7 Organization of the GIS data. Source: FAP 25, 1994.
For model simulations, the 1991 discharge values were multiplied by the scaling
factors given in Table 2.2. The scaling factors were calculated by dividing the mean
discharge values (current and determined from the GCM scenario applications) of each
river by the 1991 peak discharges of the Ganges, Brahmaputra and Meghna Rivers at
Hardinge Bridge, Bahadurabad and Bhairab Bazaar, respectively. The recorded peak
discharge in 1991 was: Ganges, 56,000 cumecs; Brahmaputra, 84,100 cumecs; and Meghna,
14,500 cumecs (BWDB, 1995).
For simulating floods, the MIKE 11 model needs local rainfall as an input in addition
to the discharge values of the major rivers entering Bangladesh. In the rainfall-runoff model
contained within the MIKE 11, rainfall data for a total of 86 rainfall stations maintained
by the Bangladesh Water Development Board (BWDB) were used. These stations are
distributed over the 100,000 sq. km. area covered by the General Model for Bangladesh.
The rainfall dataset covers the 25-year period 1967-1992. The rainfall-runoff model
consists of 48 catchments, of which only 7 actually represent the catchment areas of the
Ganges, Brahmaputra and Meghna Rivers in Bangladesh. Mean areal rainfall for each
catchment was calculated by the IWM from the 86 rainfall stations using the Thiessen

polygon method. The mean areal rainfall was then used as direct input to the model. The
rainfall-runoff model was run using a 24 hourly time-step.
The procedure (Annex 2.1) for calculating daily rainfall values for the “control run”
and “climate change scenarios” (mean and 20-year rainfall), was developed by a BDCLIM
team of researchers at the International Global Change Institute (IGCI), University of
Waikato, New Zealand. For the calculation, the rainfall records of the 86 stations for the
period 1967-1992 were considered. For the control run, mean rainfall for each of the
365 days was calculated. For calculating 20-year rainfall values, the ratio of the 20-year
and mean rainfall was determined and the mean rainfall for each of the 365 days was then
multiplied by the computed ratio. For the climate change simulations, each of the current
values was multiplied by the scaling factors.
M. M. Q. MIRZA 41
Copyright © 2005 Taylor & Francis Group plc, London, UK
Fig. 2.8 Steps being followed in the MIKE 11-GIS model to produce flood maps. Source: F
AP 25, 1994.

Time

0.00
6.00
12.00
18.00
24.00
h
t


M
I
K

E

11
M
I
K
E

11
-
G
I
S
Flood Maps Flood Simulation
River System
h

5.35
5.75
5.90
6.20
5.80
42 HYDROLOGIC MODELING APPROACHES
Copyright © 2005 Taylor & Francis Group plc, London, UK
M. M. Q. MIRZA 43
Copyright © 2005 Taylor & Francis Group plc, London, UK
44 HYDROLOGIC MODELING APPROACHES
Copyright © 2005 Taylor & Francis Group plc, London, UK
2.5 APPLICATION OF HYDROLOGIC MODEL IN INDIA
Gosain and Rao (2003) applied the SWAT (Soil and Water Assessment Tool) distributed

hydrologic model on major river basins in India. For simulation of discharge, daily weather
data from HadRM2 was used. A total of 40 years of simulation over 12 river basins were
conducted (Fig. 2.9). The simulation period was splitted into two equal 20 years period
belonging to “control” and “future”. Each river basin was also subdivided into reasonable
sized sub-basins in order to account for spatial variability of possible change in climate.
Fig. 2.9 The 12 river basins in India where the Soil and Water Assessment Tool (SWAT) model was
applied. Source: Shukla et al., 2003. Reproduced with permission.
2.5.1 DATA USED FOR STUDY
For making assessment of water resource availability at particular locations of the river
basin, the data inputs for the SWAT model (Box 2.1) are: terrain, land-use, soil and weather.
Data (spatial scale 1:250,000) for all the river basins of India (except for the Brahmaputra
and Indus Rivers) was used in the model. The snowbound areas of the Ganges River basin
could not include in the model due to the lack of required data. The following data elements
were used.
DEM (Digital Elevation Model): A DEM represents a digital file consisting of terrain
elevations for ground positions at regularly spaced horizontal intervals. Contours taken
from a 1:250,000 scale ADC world topographic map was used to generate DEM for the
study.
M. M. Q. MIRZA 45
Copyright © 2005 Taylor & Francis Group plc, London, UK

Box 2.1 The SWAT Model

SWAT (Soil and Water Assessment Tool) is a conceptual continuous model. The
model is useful to the water resource managers in assessing the impact of
management on water supplies and non-point source pollution in large river
basins. The model operates on a daily time step and allows a basin to be
subdivided into grid cells or natural sub-watersheds. Major components of the
hydrologic balance and their interactions are simulated including surface runoff,
lateral flow in the soil profile, ground water flow, evapo-transpiration, channel

routing, and pond and reservoir storage. The primary considerations in model
development were to stress land management, water quality loadings, flexibility
in basin discretization and continuous time simulation.
Stream Network Layer: Large-scale contour/DEM data was not available. Therefore, the
actual stream network was used. This option helped in conforming to the shapes of the
sub-basins, which were close to the prototype situation. Appropriate threshold values
were used for generating the stream networks for various river basins.
Watershed (sub-basin) Delineation: Automatic delineation of watersheds was done by
using the DEM as input and the final outflow point on each river basin as the pour point.
Weather Data: The weather data generated in transient experiments of the climate carried
out by the Hadley Center for Climate Prediction, U.K. The data at a resolution of 0.44
o
x 44
o
latitude by longitude grid points was obtained from the Indian Institute of Tropical
Meteorology (IITM). The daily weather data on maximum and minimum temperature, rainfall,
solar radiation, wind speed and relative humidity at all the grid locations were processed to
use in the hydrologic model. The RCM grid was superimposed on the sub basins in order
to derive the weighted means of the inputs for each of the sub basins. The centroid of each
sub basin was then taken as the location for the weather station to be used in the SWAT
model. This procedure was used for both the “present/control” (1981-2000) and
“future/GHG” (2041-2060) climate data.
Land Cover/Land-Use Layer: Classified land cover using remote sensing by the
University of Maryland Global Land Cover Facility with resolution of 1 km grid cell size
was used (Hansen et al., 1999).
Soil Layer: Soil map was adopted from FAO Digital Soil Map of the World and Derived
Properties (Version 3.5, November, 1995) with a resolution of 1:5,000,000 was used (FAO,
1995).
2.6 APPLICATION OF MODELS IN PAKISTAN
Masood and Ullah (1991) applied the UBC-Mangla watershed model to forecast inflows to

the Mangla reservoir, Indus basin (Box 2.2), Pakistan. The UBC model was developed by
the University of British Columbia, Canada and was extensively used there in assessing
impacts of climate change on water resources (Morrison et al., 2002; Micovic and Quick,
1999; Loukas and Quick, 1999). The general flow chart of the UBC model is shown in
Figure 2.10. In assess climate impacts on average inflows, three climate scenarios were
46 HYDROLOGIC MODELING APPROACHES
Copyright © 2005 Taylor & Francis Group plc, London, UK
used (for details see the Chapter 8 on Pakistan). The model has seven major components
which form a logical subdivision of the hydro-meteorological modeling and evaluation
process. Details on the model can be seen at />main.htm.
• The meteorological sub-model distributes the input data to all elevation zones of
the watershed. This distribution process controls the total volume of moisture
which is input to the model, and specifies the variation of temperature with
elevation, which controls whether precipitation falls as rain or snow and also
controls the melting of the snow packs and glaciers.
• The soil moisture sub-model controls the evaporation losses and the subdivision
of the rainfall and snowmelt into the four components of runoff: fast, medium,
slow and very slow components. The model computes the soil moisture deficit
and which controls the non-linear subdivision of rain and melt into the runoff
components.
• The watershed routing sub-model determines the time distribution of runoff.
Each of the four components of runoff determined by the soil moisture sub-model
is subjected to storage routing using either cascades or single linear reservoirs.
Because these reservoirs are linear, conservation of mass is guaranteed and an
accurate water budget is maintained.
• The output and evaluation sub-model is designed to give flexible access to
many aspects of the calculated watershed behavior.
• The semi-automatic calibration sub-model requires some user guidance to ensure
that parameters are restricted to reasonable ranges. The calibration process is a
constrained iterative search optimization which evaluates a maximum of four

parameters at a time.
• The updating sub-model is based on a combination of feedback information from
flow measurement and snow cover data from snow course or satellite.
• The routing sub-model, based on the UBC Flow Model, combines watershed
flows and routes these flows through a river, lake and reservoir system.
A generalized interactive model called MODSIM
1
was also used to model climate
impacts on water resource systems in the Indus basin. MODSIM is a capacitated network
flow model in which components of the system are represented by an interconnection of
nodes (diversion points, reservoirs, points of inflow/outflow, demand locations, stream
gauges, etc.) and links that have a specified direction of flow and maximum capacities
(canals, pipelines, and natural reaches). In order to consider the demands, inflows, and
desired reservoir operating rules, MODSIM internally creates a number of “accounting”
nodes and linkages that are intended to ensure mass balance throughout the network. The
network can be visualized as a resource allocation system through which the available
water resource can be moved from point to point to meet various demands.
The Indus basin was divided into 57 nodes and 70 links, including 30 demand nodes
which represented the canal system in the region. All major contributors to the Indus
1
MODSIM is a river basin network simulation model developed by Dr. John Labadie at Colorado
State University, USA based on an earlier model, SIMYLD II, developed for the Texas Water
Development Board. MODSIM was developed to enable the simulation of large-scale, complex
water resource systems, including considerations for water rights priorities, reservoir operations,
and important institutional and legal factors that affect river basin planning functions.
M. M. Q. MIRZA 47
Copyright © 2005 Taylor & Francis Group plc, London, UK