See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/308741238

Hydrological Modeling of Mahi Basin Using SWAT

Article · September 2016

DOI: 10.5963/JWRHE0503001

CITATIONS READS

13 6,297

4 authors:

Manish Sahu Sankalp Lahari

Indian Institute of Technology (Banaras Hindu University) Varanasi Indian Institute of Technology Delhi
1 PUBLICATION 13 CITATIONS 3 PUBLICATIONS 75 CITATIONS

SEE PROFILE SEE PROFILE

Ashvani K. Gosain Anurag Ohri

Indian Institute of Technology Delhi Indian Institute of Technology (Banaras Hindu University) Varanasi
145 PUBLICATIONS 3,777 CITATIONS 83 PUBLICATIONS 693 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Manish Sahu on 05 October 2016.

The user has requested enhancement of the downloaded file.

Journal of Water Resource and Hydraulic Engineering Sept. 2016, Vol. 5 Iss. 3, PP. 68-79

Hydrological Modeling of Mahi Basin Using SWAT

M. Sahu1, S. Lahari2, A.K. Gosain2, A. Ohri1
1
Department of Civil engineering, IIT (BHU) Varanasi, India
2
Department of Civil engineering, IIT Delhi, India
1
[email protected]

Abstract-Hydrological modeling of river basins is a valuable tool for basin management and assessment of extreme event impacts.
However, challenges with this type of modeling still remain, including issues with data acquisition, accuracy of meteorological data,
model uncertainty, and operational patterns of dams/large water structure. This paper evaluates the performance and suitability of a
Soil Water Assessment Tool (SWAT) model in predicting stream flow discharge, identifying sensitive parameters, and analyzing
uncertainty in the Mahi river basin. This kind of assessment is useful for the hydrological community, water resources engineers
involved in agricultural management and climate change concerns, as well as the government’s efforts in mitigating extreme natural
hazards such as droughts and floods. Since any model uncertainty becomes replicated in management practices, this paper addresses
the concerns of uncertainity in hydrological modelling of Water Resources in the Mahi River Basin. A GIS environment was used to
delineate the basin and its watersheds, and SWAT-CUP was used to perform the uncertainty and sensitivity analyses. The model’s
results were compared against five observation points spread across the basin. Statistical analysis showed that four of them resulted
in good Nash–Sutcliffe model efficiency and correlation.
Keywords- SWAT; Hydrological Modelling; Mahi Basin; Uncertainty Analysis; Sensitivity Analysis; SWAT-CUP

I. INTRODUCTION

A range of models, including empirical and process-based models, has been developed for water management and
watershed hydrology [1]. Great progress in watershed modeling was made as a result of advances and improvements in GIS
and remote sensing [2]. Some of the watershed models developed in recent years are used to simulate the flow of chemicals,
runoff, and erosion from Agricultural Management Systems [3], Erosion Productivity Impact Calculator [4], Agricultural Non-
Point Source Model [5], Soil and Water Assessment Tool [6], and Hydrologic Simulation Program Fortran [7]. Among these,
process-based semi-distributed models (such as SWAT) have been widely applied to evaluate the hydrologic impact of
agricultural practices, dams, and climate change [9, 10, 11]. SWAT, for instance, uses the basic principles of the hydrologic
cycle for simulating a basin’s behavior. The SWAT model divides a basin into a number of sub-basins (here called
“hydrological response units,” or HRUs) based on a combination of parameters regarding soil, land use, and slope. Similar
HRUs behave and react to precipitation in similar ways. These hydrological models are highly uncertain; a deterministic
approach is not a proper way to model any hydrological event. Uncertainty in prediction variable arises due to mathematical
models, algorithm uncertainty, and uncertainty in input data (such as rainfall). Uncertainty in data collection is a result of
measurement uncertainty. An example is uncertainty or error in measurement of stream flows. Stream flow measurements are
performed through rating curves at many stream-flow-measuring sites across India. If the rating curves are not updated
periodically, the uncertainty in the rating curves propagates itself in the hydrological model’s uncertainty. Uncertainty in the
mathematical models is well-known in modeling literature. These models use an approach to parameterize the models that
require an established correlation between the modeling parameter and the physical measurement on the field and/or a natural
phenomenon. For example, if a model uses Darcy’s approach for infiltration, then the model must be using a certain parameter
to estimate Darcy’s coefficient. However, it might not be a true representation of the natural phenomenon due to externalities
in nature. Porosity alone is unable to indicate the soil’s moisture-retention capacity or the capacity of the soil’s top layer to
evaporate water. Numerical uncertainty is another significant form of model uncertainty that is currently under consideration.
The model used is liable to depict uncertainty in the solution’s convergence. Uncertainty due to a solution’s non-uniqueness is
a particular issue in hydrological models.
In general, hydrological models such as SWAT incorporate many parameters (both statistical and of physical significance).
Most of these parameters obtain their values via extensive field surveys and experiments; the resulting values are then used to
calibrate the model. And few of these parameters have a significant impact on the modeling results, and so they are referred to
as “sensitive parameters” while others don’t have much impact or very little impact. The other parameters (“insensitive
parameters”) can be ignored in the mathematical model, leading to a simplification of the model’s structure. Sensitivity
analysis of parameters aims to explore the sensitivity of prediction variables to parameter variability. Sensitivity analysis helps
reduce the number of parameters in the calibration process, while an automatic calibration technique allows the user to avoid
tedious, time-consuming, manual calibration. The result is a computationally efficient calibration [19]. By including
uncertainty analysis in the model parameters and output variable, more information can be conveyed about the degree of risk
associated with a specific action. Uncertainty and sensitivity analyses are critical for decision-making. While calibrating the
model, two concepts must be kept in mind: (1) parameter non-uniqueness, which states that there are many other solutions
(different parameter values) that produce equally good results; and (2) parameter conditionality, which means that a calibrated
model is only locally conditional and cannot be applied globally [2]. This paper’s objective is to check the SWAT model’s

- 68 -
DOI: 10.5963/JWRHE0503001
Journal of Water Resource and Hydraulic Engineering Sept. 2016, Vol. 5 Iss. 3, PP. 68-79

performance and suitability in predicting stream flow discharge and its associated uncertainty in the Mahi river basin. Our
study is relevant and useful for the hydrological community, water resources engineers involved in agricultural management
and climate change concerns, and governmental efforts in mitigating extreme natural hazards such as droughts and floods.

II. STUDY AREA

The basin covers an area of nearly 3,74,842 sq. feet and lies between 72° 21’ E to 75° 19’ E and 21° 46’ N to 24° 30’ N.
The basin can be subdivided into two sub-basins: the Mahi upper sub-basin (65.11% of total basin area) and the Mahi lower
sub-basin (34.89% of total basin area). The basin occupies an area of 15,996 Sq. km (41.73% of the total basin area) in
Rajasthan, 15,474 Sq. km (40.36%) in Gujarat, and 6,866 Sq. km (17.91%) in Madhya Pradesh. A number of small tributaries
(from Eru, Nori, Chap, Som, Jhakham, Moran) join the Mahi River. The average annual rainfall over the Mahi Basin is about
700 mm, and most of the rainfall occurs during the four monsoon months (June-September) [18]. As per the 2001 census of
India, the total human population in the basin area is about 1,27,70,704 spreads in 15 districts [12]. There are various large
dams along the reach that affect and control river flow, Fig. 1 shows below.

Fig. 1 Mahi Basin Index Map with Gauge Discharge sites, Large Dams, River and Basin Boundary

III. DESCRIPTION OF MODELS

A. SWAT Model
Soil and Water Assessment Tool, or SWAT, is very flexible and robust river basin hydrological model that can stimulate
wide variety of watershed scenarios [8]. It is used to predict the river’s stream flow, sediment, water quality, and other
parameters. It is also used to model the effect of land management practices and irrigation and agricultural effect on the basin.
The SWAT model requires basic meteorological data such as precipitation, soil properties, land use, land cover, and DEM.
Physical processes such as infiltration, evapotranspiration associated with water movement, sediment movement, and crop
growth are modeled by SWAT with the help of input data. The rainfall runoff model used by SWAT in this study is the SCS
curve number model, which is described briefly below.
For modeling purposes, the river basin is divided into a number of sub-basins and then divided further into a number of
HRUs (Hydrological Response Units). For every model in SWAT, the first and essential step is to model the hydrological
water balance. This water balance is the driving force behind every process that happens in the river basin. A water balance
equation is given as:

- 69 -
DOI: 10.5963/JWRHE0503001
Journal of Water Resource and Hydraulic Engineering Sept. 2016, Vol. 5 Iss. 3, PP. 68-79

SWf = SWi + ∑ (Pday - Rsurf – QSeep - Ea - Dgw) (1)

Where SWf = final water content in soil (mm water); SWi = initial water content in soil on i day (mm water); Rsurf = surface
runoff on i day (mm water); Qseep = water entering the unsaturated zone of soil on i day (mm); Pday = precipitation on day i
(mm water); Dgw = return flow on day i (mm water); and Ea = amount of evapotranspiration on day i (mm water).
Fig. 2 shows a general description of the hydrological cycle. The land phase (mainly the soil layer) controls the amount of
water and sediment entering the main channel in each sub-basin. A sub-basin’s output is treated as input for the following sub-
basin.

Fig. 2 Schematic representation of hydrologic cycle

After the water balance phase, the next step is the routing of water, sediments, etc. through the streams to the basin’s outlet.
To calculate surface runoff, we used the SCS curve number method [13]. We used the Penman Monteith method to
estimate potential evapotranspiration (PET) [14], and then actual Evapotranspiration was calculated. The SCS curve equation
is described as:
( .)^

 = (2)
  .

Where Pday is rainfall depth for the day (mm); Qsurf is accumulated runoff or rainfall excess (mm); and S is the retention
parameter (mm) that can be obtained by the following:

 = 25.4  − 10! (3)


The SCS curve number (CN) depends on the soil’s permeability, infiltration, land use and, soil-water conditions. The CN
value can be defined by three conditions: dry, average moist, and wet. A full description is available in [14] and in the SWAT
technical manual [15].

B. SWAT-CUP Model
SWAT-CUP consists of a calibration and uncertainty program combined with the SWAT hydrological model. The program
links many algorithms, like SUFI-2, Particle Swarm Optimization (PSO), Generalized Likelihood Uncertainty Estimation
(GLUE), and Markov Chain Monte Carlo (MCMC) to perform uncertainty and sensitivity analyses. Parameter optimization
and calibration using an inverse problem always brings uncertainty, since it starts with the results and only then calculates the
parameter values causing those results. In hydrology, we perform this often – therefore, the model’s results are inevitably
uncertain. In this study, a sequential uncertainty fitting algorithm, referred to as SUFI-2 [23], is used for uncertainty analysis.
Starting with the initial parameter ranges, SUFI-2 is iterated many times so that 95% prediction uncertainty band (95PPU)
brackets most of the measured data (p factor), maintaining a small width band (r factor). Here is the step-by-step procedure to
apply the SUFI-2 algorithm:
1. Decide the model’s objective function. Since there is no pre-conceived, unique formulation of the objective function,
this choice depends on the purpose of the study and model. Here, we used Nash-Sutcliff Efficiency (NSE) [17] as the
objective function.
2. Select the absolute meaningful range of parameters. Due to a lack of knowledge of parameter distribution, we assumed
that all parameters are uniformly distributed (Gaussian distribution) within their respective range.

- 70 -
DOI: 10.5963/JWRHE0503001
Journal of Water Resource and Hydraulic Engineering Sept. 2016, Vol. 5 Iss. 3, PP. 68-79

3. Carry out sensitivity analysis for all parameters to identify their effects on the output variable.
4. Select initial ranges for parameters in the first round of Latin Hypercube sampling. A wide range is acceptable, since it
is changeable and frequently updated.
5. Perform Latin Hypercube sampling [22], resulting in n parameter combination where n represents the number of
simulation trials run. This number should be large (around 1000). The algorithm iterates n times and the output variable
is saved. Then, the objective function is calculated for each simulation.
6. Calculate 95PPUs of predicted/output variable (here discharge) using n Latin Hypercube simulations. After, the fit is
measured for output variable from the p factor, r factor, R2, and NSE. Ideally, p factor should be 100% and r factor 0 –
in actual circumstances, however, these numbers are near impossible. Thus, the values of p factor and r factor are
accepted as around 70% and less than the standard deviation, respectively.
7. First sampling results would not, like the high value of r factor with 100% of the data, be within the 95PPU. Hence,
further sampling with simultaneous updating of the parameters’ ranges is needed. The newer range would be less than
the original range and would be such that it reduces the initial r factor.
8. Continue procedure until the resulting parameter ranges are at the optimum range to solve the problem. Otherwise,
recheck the model’s structure and boundary conditions.

IV. INPUT DATA

A. Digital Elevation Model (DEM)

DEM is the raster data consisting of an array of cells or pixels containing elevation values. DEM is used to delineate the
networks of river streams, sub-basins, and parameters like slopes for HRUs. Data is obtained from the Shuttle Radar
Topography Mission 90m (SRTM). In the present study, DEM is projected to coordinate the system (WGS 1984 UTM Zone
43N) and the processed [values?] in a GIS environment using a watershed delineator [16]. Fig. 3 (d) shows the digital elevation
map of the basin.

B. Land Use / Land Cover

The study area’s land use and land cover data was obtained from the National Remote Sensing Center (NRSC), ISRO. The
spatial resolution of the obtained map is 90m. We used land use and land cover data for HRU definition, and then assigned
Curve Numbers (CN) to land areas for runoff calculation and hydrological analysis. Certain standard abbreviations were used
for land use units (these are provided at the end of Table 1) indicated in Fig. 3 (c) the landuse/landcover map of the basin.

C. Soil Map
The study area’s soil map was obtained from the National Bureau of Soil Science and Land Use Planning (NBSS&LUP).
We also used the soil map for HRU definition and hydrological analysis. Fig. 3 (b) represents the soil map of the basin and
Table 2 provides the standard soil codes used in the study.
TABLE 1 LAND USE DATA TABLE 2 NBSS SOIL CODE

Code LandUse/Cover
CRDY DRYLAND CROPLAND AND PASTURE
Soil Code Hydrological GroupSoil Type
CRIR IRRIGATED CROPLAND AND PASTURE Ao81-2b-3652 D LOAM
MIXC MIXED DRYLAND/IRRIGATED CROPL
CRGR CROPLAND/GRASSLAND MOSAIC Be75-2a-3676 C LOAM
CRWO CROPLAND/WOODLAND MOSAIC
GRAS GRASSLAND Be76-2b-3677 D LOAM
SHRB SHRUBLAND
MIGS MIXED GRASSLAND/SHRUBLAND
Bv12-3b-3696 D CLAY_LOAM
SAVA SAVANNA I-Be-Lc-b-3716 C LOAM
FODB DECIDUOUS BROADLEAF FOREST
FODN DECIDUOUS NEEDLELEAF FOREST Jc51-2a-3744 D LOAM
FOEB EVERGREEN BROADLEAF FOREST
FOEN EVERGREEN NEEDLELEAF FOREST
Jc51-2a-3744 D LOAM
FOMI MIXED FOREST Lc70-1-2b-3774 C SANDY_CLAY_LOAM
WATB WATER BODIES
WEHB HERBACEOUS WETLAND Vc13-2-3b-3858 D CLAY
WEWO WOODED WETLAND
BSVG BAREN OR SPARSLY VEGETATED Vc21-3a-3859 D CLAY
TUHB HERBACEOUS TUNDRA
TUWO WOODED TUNDRA
Vc43-3ab-3861 D CLAY
TUMI MIXED TUNDRA Vc45-3a-3864 D CLAY
TUBG BARE GROUND TUNDRA
ICES SNOW OR ICE WATER-6997 D WATER

- 71 -
DOI: 10.5963/JWRHE0503001
Journal of Water Resource and Hydraulic Engineering Sept. 2016, Vol. 5 Iss. 3, PP. 68-79

D. Meteorological Data
The principal data we used in the study was hydrological and weather data of the Mahi basin. The area’s meteorological
data was obtained from the Indian Meteorological Department (IMD), and the Central Water Commission (CWC), India.

E. Large Dam Details

There are approximately 15 large water resource structures that affect and control the river flow. Various details like spatial
location, year of operation, and reservoir capacity (among others) need to be kept in mind when performing the hydrological
study. Very limited data is available regarding the dams’ operation pattern and behaviour. Fig. 3(a) offers a map showing the
spatial location of the large dams.

(a) (b)

(c) (d)
Fig. 3 Basic spatial data input to the model (a) Discharge Gauge Sediment Quality stations (GDSQ) (b) Soil map
(c) Land use map (d) Digital Elevation Model (DEM)

- 72 -
DOI: 10.5963/JWRHE0503001
Journal of Water Resource and Hydraulic Engineering Sept. 2016, Vol. 5 Iss. 3, PP. 68-79

V. MODEL SETUP

A. SWAT Model
The input data needed for the SWAT model requires the use of a Digital Elevation Model to generate slope, basin boundary,
topography, land use, soil layer data and climate data. The latter includes daily precipitation values, daily minimum and
maximum air temperature, and solar radiation, which can be derived using theoretical formulae, relative humidity, and wind
speed. Daily precipitation and minimum and maximum temperature are entered in the model while the other parameters are
simulated based on the nearest climate station using a SWAT weather generator.
The model is built completely in a GIS environment using a SWAT extension [16]. A watershed delineation tool is used to
delineate the sub-basins based on DEM data. Subsequently, a stream network is also generated using DEM data. Accordingly,
the basin is divided into 148 sub-basins. After this, Hydrological Response Units (HRUs) are created based on slope, land use,
and soil data. HRU creation in SWAT requires land use and soil threshold inputs in order to derive HRUs. A threshold value of
10% of the basin’s total area is selected. The set threshold values divide the HRUs into a number of sub-units, each with a
unique ID. Areas that respond to incoming precipitation similarly are therefore areas with similar hydrological response.
Runoff for individual HRUs is calculated separately and then used to obtain the total runoff. Fig. 4 below shows the workflow
of the SWAT Model.

Fig. 4 Flow diagram: setting up SWAT Model

B. SWAT-CUP Model
Automated model calibration requires changes in iterations and model parameters. When the model is run, the required
outputs are evaluated from model output files. The SWAT-CUP’s main function is to provide a link between the model and the
input and output of calibration. A schematic diagram below shows the linkage between SWAT and SWAT-CUP. The SWAT
model’s output is fed as input for SWAT-CUP. Fig. 5 below represents the combined workflow of SWAT and SWAT-CUP
models.

- 73 -
DOI: 10.5963/JWRHE0503001
Journal of Water Resource and Hydraulic Engineering Sept. 2016, Vol. 5 Iss. 3, PP. 68-79

Fig. 5 Flow diagram: setting up SWAT-CUP SUFI-2 model

(Source: SWAT-CUP user manual, K.C. Abbaspour)

VI. PERFORMANCE EVALUATION

The SWAT model’s performance is calculated using a graphical representation of observed and simulated discharge values,
as well as on the basis of various statistical parameters such as Nash-Sutcliff Efficiency (NSE) and coefficient of determination
(R2) [17]. We used the R2 coefficient to evaluate the trend between observed and simulated values. The range of NSE values
can vary from -∞ to 1, where NSE = 1 is the most optimum value. NSE values from 0.6 to 1 are considered to be a good fit,
whereas values less than 0 imply that the observed values’ mean is a better predictor than the simulated values. The R2 varies
between -1 and 1, where 1 represents perfect positive trend and -1 represents perfect negative / opposite trend. The coefficient
of determination and NSE can be calculated by the equations given below:
[∑($%,' $% )($(,' $( )] *

=∑ * (4)
'($%,' $% ) ∑'($(,' $( ) *

∑($ $( ) * '
+, = 1 − ∑($ % (5)
%,' $% ) *

Where Qm is the measured discharge value; Qs is the simulated discharge; Qm(bar) is the average measured discharge; and
Qs(bar) is the average discharge simulated.
We calculated uncertainty and sensitivity performance using p factor and r factor values. As establishd earlier, the model will
be most effective if p factor is closer to 100% and r factor is less than the standard deviation. A large p factor can be achieved
at an expense of a large r factor. Therefore, balancing these two factors is required, and when acceptable values of r factor and
p factor are reached, the parameter uncertainties become the desired parameters. R factor can be calculated as:
 23'(45.6%) 23'(*.6%)
 -./0 = ∑18<= (6)
1 8 (9:;)

Where Yti(97.5%) and Yti(2.5%) are the upper and lower boundaries of the 95PPU; and Std(obs) is the observed data’s standard
deviation.

- 74 -
DOI: 10.5963/JWRHE0503001
Journal of Water Resource and Hydraulic Engineering Sept. 2016, Vol. 5 Iss. 3, PP. 68-79

VII. RESULTS AND DISCUSSIONS

A. Sensitivity Analysis
Parameter sensitivities are determined by performing a multiple regression analysis, which regresses the parameters
generated by Latin Hypercube against the objective function [2]. t-test and p-value are used to identify each parameter’s
relative significance. The t-stat is the ratio of parameter coefficient to its standard error. Parameters with p value less than or
equal to 0.05 are taken as sensitive. We found that, out of 14 total parameters, seven were sensitive to the output variable
(discharge). This reduces the number of parameters for calibration. Table 3 shows the list of parameters used in sensitivity
analysis and each parameter’s individual sensitivity. The sensitive parameters are: SOL_AWC, SOL_K, CH_K2,
GW_DELAY, CN2, ALPHA_BF, ESCO.
TABLE 3 SENSITIVITY ANALYSIS TABLE WITH T-STAT AND P-VALUE

Abbreviated Descriptions of parameters t-stat p-value Rank

SOL_AWC Available water capacity of the soil layer. 17.64949 0 1

SOL_K Saturated hydraulic conductivity. -16.3289 0 2
Effective hydraulic conductivity in the main
CH_K2 channel alluvium. 3.432463 0.000649972 3

GW_DELAY Groundwater delay (days). 2.915269 0.003720346 4

CN2 SCS runoff curve number -2.1156 0.034893979 5
ALPHA_BF Baseflow alpha factor (days). -2.11401 0.035030255 6
ESCO Soil evaporation compensation factor. -1.92773 0.054480029 7

CH_N2 Manning's "n" value for the main channel. 1.236497 0.216879971 8
GW_REVAP Groundwater "revap" coefficient. 0.914357 0.360989077 9
SOL_BD Moist bulk density. -0.69713 0.486061802 10

Threshold depth of water in the shallow aquifer

GWQMN required for return flow to occur (mm). -0.46446 0.642532698 11

ALPHA_BNK Baseflow alpha factor for bank storage. -0.39213 0.695139901 12

SLSUBBSN Average slope length. -0.35082 0.725878202 13

Threshold depth of water in the shallow aquifer for

REVAPMN "revap" to occur (mm). -0.22826 0.819540499 14

B. Calibration Process
Using the sensitive parameters, we carried out model calibration in a SWAT-CUP environment. Calibration is performed
using observed discharge data from five stations spread across the basin with monthly time steps. This process gave us the
optimum parameter ranges. Moreover, since we were doing inverse optimization, the ranges were applicable to our basin. We
computed the model’s performance using NSE and R2 values. The statistical results we computed showed that most of the
stations are giving good results as compared with observed values. Table 4 shows the statistical coefficients and parameter’s
optimum ranges. Only the Rangeli station is not giving good results in neither R2 nor NSE coefficients. One of the possibilities
for this deviation from optimum values may be that the Rangeli station lies just below several tributaries with four large dams.
Due to data limitation regarding the dams’ operational patterns, a large deviation from optimum at Rangeli station is observed.
This might be corrected by incorporating the dams’ operational patterns in the SWAT model. This limitation is due to lack of
data availibility for the Rangeli dam. Additionally, this is a scale issue in Hydrology. The scale at which modeling is performed
also gives rise to uncertainty in Hydrological models.
TABLE 4 STATISTICAL RESULTS OF THE ANALYSIS AT DIFFERENT OBSERVATION SITES

Station Name p-factor r-factor R2 NSE p-factor r-factor R2 NSE

Rangeli 0.34 1.22 0.69 0.43 0.24 0.44 0.66 0.53
Mataji 0.65 0.57 0.86 0.81 0.42 0.26 0.8 0.73
Chakaliya 0.33 0.22 0.84 0.58 0.17 0.12 0.81 0.52
Khanpur 0.13 0.19 0.87 0.82 0.13 0.17 0.89 0.66
Pingalwada 0.63 0.91 0.84 0.76 0.21 0.35 0.8 0.75
Calibration using complete parameter set Calibration using sensitive parameter set

C. Uncertainty Analysis in Streamflow Prediction

We derived the performance indices of uncertainty analysis for the model parameters using monthly time steps. As
discussed earlier, uncertainty analysis was done using p factor and r factor. Results show the 95% prediction band (95PPU), p
factor, and r factor values. The p factor and r factor values for three sites are too small. In addition, the uncertainty band did not

- 75 -
DOI: 10.5963/JWRHE0503001
Journal of Water Resource and Hydraulic Engineering Sept. 2016, Vol. 5 Iss. 3, PP. 68-79

cover many of the observed values, although the NSE and R2 coefficient were satisfactory. A reason behind this might be that
these stations are located downstream of a few large dams. These dams capture enormous amounts of water. Due to the
unavailability of the dams’ operational data, SWAT is unable to capture the true discharge values. SWAT models at monthly
scale with such streamflow predictions may be considered for application in Planning scale models. This may also be used for
Water Resources development on a larger scale of both time and space.
Also, SWAT has the limitation that it does not thoroughly simulate groundwater flow. A careful examination shows that
much of the unbracketed data in 95PPU lies in the low flow or baseflow region. Fig. 6 shows an example of such a case. If this
flow had been simulated properly, then the p factor would have been much higher. NSE also has the limitation that it strongly
overestimates larger values while neglecting lower values, since it calculates by squaring the differences between observed and
simulated [20] therefore showing good NSE in spite of poor low flow simulation.

VIII. LIMITATIONS AND FUTURE WORK

In this study, we performed uncertainty and sensitivity analyses of model parameters. Our main objective was to improve
the calibration process and achieve more accurate results. We found that calibration using sensitive parameters resulted in
slightly poorer results compared to calibration using a complete set of parameters. A potential explanation might lie with the
limitations of SWAT, which does not model groundwater flow properly. Our sensitivity analysis also indicated that
groundwater parameters (GW_delay Alpha_bf that directly impacts groundwater flow, and Sol_AWC that indirectly affects
groundwater flow) are sensitive. Future research based on this study must overcome these SWAT limitations. This can be
achieved by integrating SWAT and Modflow [21] models. Modflow is a widely used model in groundwater modeling.

IX. CONCLUSIONS
A thorough model calibration is required to obtain accurate results when performing prediction using a hydrological model
(such as discharge). Along with the results, it is always advisable to report model uncertainty in the model predictions. In this
paper, we applied the SWAT model in the Mahi River Basin to simulate discharge in the period of 1992-2005. We did this at a
monthly scale with data from large dams incorporated at a few sites. The model was then calibrated and uncertainty and
sensitivity analyses were done to obtain sensitive parameters. The common SUFI-2 algorithm was employed here to carry out
uncertainty and sensitivity analyses. Our study’s results indicate that the SWAT and SAWT-CUP models are useful in
forecasting flow and performing uncertainty and sensitivity analyses. SWAT, however, has a limitation in groundwater
modeling (this is illustrated in Fig. 6, where the base flow is poorly predicted by the model). The model is at a monthly scale
and it can be used to plan Mahi River Basin’s future project development, as long as sufficient data and details are
incorporated and the model is validated and identified. Additionally, this study’s model can be used in further assessment of
climate change and land use/land cover impact assessment on river basin. For improving Ground Water Modeling, SWAT-
MODFLOW is a new platform that can be used for future studies of this nature.

X. GRAPHICAL REPRESENTATION OF RESULTS

A. Calibration and Uncertainty Analyses Using Complete Set of Parameters

Fig. 6 showing 95PPU range not covering baseflow

- 76 -
DOI: 10.5963/JWRHE0503001
Journal of Water Resource and Hydraulic Engineering Sept. 2016, Vol. 5 Iss. 3, PP. 68-79

B. Calibration and Uncertainty Analyses Using Only the Sensitive Set of Parameters

- 77 -
DOI: 10.5963/JWRHE0503001
Journal of Water Resource and Hydraulic Engineering Sept. 2016, Vol. 5 Iss. 3, PP. 68-79

REFERENCES

[1] V. P. Singh and D. A. Woolhiser, “Mathematical Modeling of Watershed Hydrology,” Journal of Hydrologic Engineering, Vol. 7, No. 4,
pp. 270-292, 2002. doi:10.1061/(ASCE)10840699(2002)7:4(270).
[2] K. C. Abbaspour, “SWAT-CUP2: SWAT Calibration and Uncertainty Programs—A User Manual,” Department of Systems Analysis,
Swiss Federal Institute of Aquatic Science and Technology, Dübendorf, 2008.
[3] Knisel WG. CREAMS, a field scale model for chemicals, runoff, and erosion from agricultural management systems. USDA
Conservation Research Report Washington, D.C.: USDA, 1980, No. 26.
[4] Williams JR. The EPIC model. In Singh VP. (Ed). Computer models of watershed hydrology. Water resources Publications, Highlands
Ranch, CO, 1995, Chapter 25, pp. 909-1000.
[5] Young RA, Onstad CA, Bosch DD, Anderson WP. Agnps - a Nonpoint- Source Pollution Model for Evaluating Agricultural
Watersheds. J Soil Water Conserv 1989; 44(2): 168-173.
[6] Arnold JG, Srinivasan R, Muttiah RR, Williams JR. Large Area Hydrologic Modeling and Assessment Part I: Model Development. J
Am Water Resour Assoc 1998; 34(1): 73-89.
[7] Bicknell BR, Imhoff JC, Kittle JL, Jobes TH, Donigian AS. Hydrologic Simulation Program – FORTRAN (HSPF), user’s manual for
version 12.0, 2001, USEPA.
[8] Gassman PW, Reyes MR, Green CH, Arnold JG. The soil and water assessment tool: Historical development, applications, and future
research directions. Trans ASABE 2007; 50(4): 1211-1250.
[9] J.G. Arnold and N. Fohrer. “SWAT2000: current capabilities and research opportunities in applied watershed modeling,” Hydrological
Processes, vol. 19, no. 3, pp. 563-572, 2005.
[10] D.K. Borah, and M. Bera, “Watershed-scale hydrologic and nonpoint-source pollution models: Review of applications,” Trans. ASAE,
vol. 47, no. 3, pp. 789-803, 2004.
[11] USDA-ARS (U.S. Department of Agriculture, Agricultural Research Service). The automated geospatial watershed assessment tool
(AGWA). Available at: www.tucson.ars.ag.gov/agwa/. Accessed 23 August, 2006.
[12] Central Water Commission of India report on Mahi Basin, Version 2.0, Mar. 2014.
[13] United States Department of Agriculture (1986). Urban hydrology for small watersheds. Technical Release 55 (TR-55) (Second ed.),
Natural Resources Conservation Service, Conservation Engineering Division.

- 78 -
DOI: 10.5963/JWRHE0503001
Journal of Water Resource and Hydraulic Engineering Sept. 2016, Vol. 5 Iss. 3, PP. 68-79

[14] Richard G. Allen, Luis S. Pereira, Dirk Raes & Martin Smith (1998). Crop Evapotranspiration – Guidelines for Computing Crop Water
Requirements. FAO Irrigation and drainage paper 56. Rome, Italy: Food and Agriculture Organization of the United Nations. ISBN
92-5-104219-5.
[15] S.L. Neitsch, J. G. Arnold, J. R. Kiniry, and J. R. Williams, Soil and Water Assessment Tool – Theoretical Documentation, Version
2009. Texas, USA, 2009.
[16] SWAT, Soil and Water Assessment Tool: ArcSWAT, College Station, Texas: Texas A&M University. Available at:
www.brc.tamus.edu/swat/arcswat.html.
[17] J.E. Nash, and J.V. Sutcliffe, “River flow forecasting through conceptual models, Part I: A discussion of principles,” J. Hydrol., vol. 10,
no. 3, pp. 282-290, 1970.
[18] Rajasthan irrigation department, www.waterresources.rajasthan.gov.in.
[19] Kurt K. Benke, et. al., “Parameter uncertainty, sensitivity analysis and prediction error in a water-balance hydrological model”,
Mathematical and computer modelling, www.elsevier.com/locate/mcm, 2007.
[20] P. Krause, et. al., “Comparison of different efficiency criteria for hydrological model assessment”, Advances in Geosciences, 2005.
[21] Banta, Edward R., Mary C. Hill, and Michael G. McDonald. MODFLOW-2000, the US Geological Survey modular ground-water
model: User guide to modularization concepts and the ground-water flow process. Reston, VA, USA: US Geological Survey, 2000.
[22] McKay, M. D., Beckman R. J., and Conover W. J. “A Comparison of Three Methods for Selecting Values of Input Variables in the
Analysis of Output from a Computer Code.” Technometrics, vol. 21, no. 2 (1979): 239-45.
[23] K. C. Abbaspour, C. A. Johnson, and M. Th. van Genuchten, “Estimating Uncertain Flow and Transport Parameters Using a
Sequential Uncertainty Fitting Procedure”, 2004.

- 79 -
DOI: 10.5963/JWRHE0503001

View publication stats