Hydrological studies for small watershed in India using the ANSWERS model

Journal of Hydrology 318 (2006) 184–199 www.elsevier.com/locate/jhydrol

Hydrological studies for small watershed in India using the ANSWERS model Ramadhar Singha,*, K.N. Tiwarib,1, B.C. Malb,1 a

Irrigation and Drainage Engineering Division, Central Institute of Agricultural Engineering, Nabibagh, Berasia road, Bhopal 462 038 (M.P.), India b Agricultural and Food Engineering Department, Indian Institute of Technology, Kharagpur 721 302 (W.B.), India Received 15 March 2004; revised 29 May 2005; accepted 3 June 2005

Abstract A study was undertaken with the objective of investigating the performance of the physically based distributed parameter Areal Non point Source Watershed Environment Response Simulation (ANSWERS) model for a 16.13 km2 small watershed in eastern India by using digital elevation model (DEM), GIS and remote sensing techniques for automatic extraction of the model input parameters. The model was calibrated by using sixteen storms of 1993 and 1994 and validated with fifteen storms of 1995 and 1996. For calibration storms, the model simulates surface runoff, peak flow and sediment yield with average per cent deviation (Dv) equal to K9.32, 1.24 and K3.04 and coefficient of efficiency (E) equal to 0.964, 0.881 and 0.884 respectively. For validation storms, the model simulates surface runoff, peak flow and sediment yield with average per cent Dv as K8.13, K2.25 and K1.63 and E as 0.991, 0.741 and 0.965 respectively. During model calibration and validation the peaks of the simulated hydrographs for majority of the storms were found to occur after the peaks of the observed hydrographs. The statistical comparisons indicate that the model simulates runoff, peak flow and sediment yield well for most of the storms with Dv less than 15% from the observed values and average value of E greater than 0.80. The model calibration and validation results indicate that the ANSWERS can be successfully used for simulating the watershed response under varying soil moisture and watershed conditions. The study reveals the suitability of the ANSWERS model application for the other Indian watersheds of similar hydro-geological characteristics. q 2005 Elsevier Ltd All rights reserved. Keywords: ANSWERS; DEM; Hydrograph; Runoff; Sediment yield; Simulation

1. Introduction

* Corresponding author. Fax: 91 755 2734016. E-mail addresses: [email protected] and rsingh_67@yahoo. co.in (R. Singh), [email protected] (K.N. Tiwari), bmal@agfe. iitkgp.ernet.in (B.C. Mal). 1 Fax: 91 3222 282244/255303.

0022-1694/$ - see front matter q 2005 Elsevier Ltd All rights reserved. doi:10.1016/j.jhydrol.2005.06.011

Degradation of water and land resources is an issue of significant societal and environmental concern. In India it has been estimated that out of the total geographical area 329 million hectare (mha), nearly 188 mha of land suffers from various form of land degradation problem (Sehgal and Abrol, 1994).

R. Singh et al. / Journal of Hydrology 318 (2006) 184–199

The Upper Damodar Valley (17,513 km2) comprising 39 sub catchments is infested with serious problems of land degradation by soil erosion affecting the agricultural, forests and waste lands of the region. Survey carried out by Damodar Valley Corporation (DVC) revealed that about 66% of total land of Upper Damodar Valley (UDV) is affected by different types of erosion. Also 35% of agricultural land is moderately to severely eroded under sheet erosion (Misra, 1999). Sedimentation studies of major reservoirs in India revealed that the annual rate of siltation from unit catchment has been 2–3 times more than the designed values (Murthy, 1980). The actual sedimentation survey of Panchet and Maithon reservoirs of Damodar valley revealed that siltation rate was as high as seven times of the designed rate (Misra and Satyanarayana, 1991; Misra, 1999). In order to preserve natural resources and the useful life of these reservoirs, soil and water conservation measures are essential and now included as major component in all river valley projects (RVP) of the country. The Soil Conservation Department of DVC, one of the prime RVP in the country, has an established network of stream gauging stations and sediment observation posts for monitoring hydrological parameters of the catchments. The purpose is to develop the region for erosion control and to enhance production on sustained basis with an eye on reducing the sediment inflow to multipurpose reservoirs. Out of 1130 small watersheds (1300–2000 ha) of UDV, the gauged watersheds are 69 only. The accurate information on watershed runoff and sediment yield are necessary for the design of conservation structures to offset the ill effects of sedimentation. These information for ungauged watersheds are to be generated through watershed simulation studies, to identify the priority watersheds for implementing watershed management programmes with the limited available funds and also for assessment of their impacts. Hydrologic models provide cost-effective means for determining the best management practices (BMPs) that minimizes water and land degradation due to soil erosion. The runoff and sediment from agricultural lands have major impact on water quality of the river. Diffuse pollution sources, such as agricultural lands are more difficult to assess and treat than point sources. Methods

185

of identifying the sources of sedimentation and quantities of runoff and sediment are therefore required. Distributed parameter watershed models are applicable for this type of assessment. The AGNPS (Young et al., 1989) and the ANSWERS (Beasley et al., 1980, 1982), are the commonly used distributed parameter models, for simulating the small watersheds (Bingner et al., 1992). These models take spatial variability of watershed characteristics into account through concept of hydrologic response units. The model used to assess the impact of diffuse pollution must simulate both runoff and sediment transport using equations that are applicable to the study region. The ANSWERS simulates both runoff and sediment transport and uses mainly physically based relationships. It is therefore an appropriate model for assessing BMPs and identifying the sources of runoff and erosion. The ANSWERS model has been used for simulating the hydrologic response of small agricultural watersheds (Montas and Madramootoo, 1991; Razavian, 1990). The integration of Geographical Information System (GIS) with distributed parameter models reduces the time needed for generating large number of input data associated with these models as compared to conventional methods. GIS provides an alternative way of manipulating the input data and preparing model input files. Raster-based GIS worked well with the ANSWERS for generating model input information (De Roo et al., 1989; Rewerts and Engle, 1991). Remote Sensing data provide real time accurate information on spatial and temporal variations in land use and land cover (LULC) of the study area, which are required for watershed simulation studies (Chakravorty, 1993; Tiwari et al., 1997). The GIS and grid digital elevation models (DEM) have been successfully used for automatic extraction of the watershed parameters (e.g. watershed delineation, drainage networks, slopes, aspects etc.) requied for hydrologic and non point source (NPS) pollution studies (Garbrecht and Martz, 1993; Tiwari et al., 1997). The planning and implementation of BMPs for effective conrol of soil erosion and development of water resources, can be improved by the integrated use of physically based distributed parameter models, GIS and remote sensing techniques. Although, physically based distributed parameter models have been successfully used in developed countries for

186

R. Singh et al. / Journal of Hydrology 318 (2006) 184–199

simulating the hydrological processes of the watersheds, so far little attempt has been made to use them for this purpose in India However the applicability of the ANSWERS model for assessing runoff and soil loss in Indian watersheds has not been tested. The study was aimed to assess the applicability of the ANSWERS for Indian watersheds with the following objectives: (i) To construct a database for the ANSWERS using DEM, GIS and Indian Remote Sensing satellite data. (ii) To calibrate and validate the ANSWERS model for simulating the runoff, peak flow and sediment yield for the study watershed. (iii) To evaluate the simulation capabilities of the ANSWERS for rainfall/runoff events under varied soil moisture and rainfall conditions. 2. ANSWERS model overview The ANSWERS (Beasley et al., 1980, 1982) is a watershed scale, distributed parameter, eventoriented, physically based model. The ANSWERS was developed to simulate the influence of watershed management practices on runoff and sediment loss. The original model (Huggins and Monke, 1966) included surface water hydrology only. The model was expanded to include erosion and sediment transport by Beasley et al. (1980). The overall model structure consists of a hydrologic model, a sediment detachment and transport model, and several routing components necessary to describe the movement of water in overland, sub surface and channel flow phases. The model operates on cell basis. Soil moisture of a watershed is simulated by using the soil water balance equation. The inflow and outflow for a cell is solved by an explicit backward difference solution of the continuity equation. Continuity equation is solved by combining it with Manning’s equation as stage-discharge equation for both overland and channel routing. Soil detachment, transport, and deposition are modeled as a function of the precipitation and the runoff process. The various hydrological processes, flow routings, erosion and sediment transport processes used for simulations are described in details in the ANSWERS Users Manual (Beasley and Huggins, 1980).

Input information for the ANSWERS model require individual cell description containing x and y location of cell, its soil type, land use, slope, aspect, sub-surface drainage and channel features. In addition to the cell descriptions, storm rainfall, a description of infiltration parameters of each soil type and a description of the soil surface parameters of each crop and management system are required. The output from the model consists of runoff hydrograph, sediment information, net transported sediment yield or deposition for each element, and channel deposition. 3. Materials and methods 3.1. Study area and data source The study area constitutes a small watershed (1613 ha) named Banha of the UDV of Hazaribagh district in Jharkhand State of India. Fig. 1 shows the location map of the study area. The watershed lies between latitudes of 24813.85 0 N ad 24817.06 0 N and longitudes of 85812.44 0 E and 85816.15 0 E. The major part (84%) of the watershed has less than 3.0% slope with an average slope of about 1.90% (Table 1). The soils of the watershed are loam, sandy loam and clay loam covering 45.81, 27.91 and 25.23% of the watershed area respectively. The area has a humid sub tropical climate with a mean annual rainfall of 1255 mm. About 85% of the rainfall occurs during monsoon months (June–October). The elevations of the highest and lowest points are 450 m and 406 m above mean sea level respectively (Fig. 1). The land use and land cover of area comprises 29.7% land under agriculture (paddy as main crop), 28.3% under forest, 20% under wasteland and 22% under grasses and others (Table 1). The watershed is gauged at outlet for measuring hydro-meteorological data through automatic weather station. Indian Remote Sensing satellite (IRS-1B LISS-II) digital data at 23 m spatial resolution of the study area of 25th October, 1993, 11th May, 1994 and 12th October 1994 were used to generate land use and land cover (LULC) information. The EASI/PACE image processing software and IDRISI GIS were used for analysis of satellite data and generation of basic parameter layers. The Survey of India topographical maps at 1: 25,000 scale and soil survey map at 1: 10,560 scale were used as collateral data for geo-referencing of the satellite data.

R. Singh et al. / Journal of Hydrology 318 (2006) 184–199

187

Fig. 1. Location map of Banha watershed.

3.2. Acquisition of watershed data Continuous measurements of rainfall (intensity, duration and amount) and runoff were performed from 1993 to 1996 using automatic weather station and stream gauging through automatic stage recorder. Sediment yield data were obtained by manual

sampling. From the gauged watersheds data of the river valley projects (RVP) of the country, Misra and Satyanarayana (1991) reported that the bed load in stream flow was found to be 20 per cent of the suspended load in stream flow. This criteria was considered to estimate total sediment yield by adding bed load into suspended sediment load. Undisturbed

188

R. Singh et al. / Journal of Hydrology 318 (2006) 184–199

Table 1 Land use and land cover generated from satellite imageries, soil types and slope of the watershed Land use and land cover

Year-1993

Year-1994

Area (ha)

Area (%)

Area (ha)

Area (%)

Paddy Upland crops Degraded forest Open forest Dense forest Grass land Fallow land Waste land-1 (eroded land) Waste land-2 (bushes/shrubs) Water body Total Classification accuracy (%)

316.35 63.09 167.94 271.35 63.81 324.90 83.16 259.38 55.53 07.56 1613.07

19.61 3.91 10.41 16.82 3.96 20.14 5.16 16.08 3.44 0.47 100.00 88.70

342.27 59.04 168.57 224.55 63.45 340.92 78.21 271.53 56.52 8.01 1613.07

21.22 3.66 10.45 13.92 3.93 21.13 4.85 16.83 3.50 0.50 100.00 89.45

Soil textural class

Proportion (%)

Sandy loam Loam Clay loam Sand (river bed)

Total

27.91 45.81 25.23 1.05

100.00

soil samples for different depth ranges i.e. 0–150, 150–400, 400–800 and 800–1200 mm were collected at 21 locations as shown in Fig. 1 in the watershed for measurement/determination of soil physical properties. The depth of ‘A’ horizons were measured at sampling sites. Soil physical properties such as soil texture, bulk density, field capacity (FP), wilting point (WP), porosity, saturated hydraulic conductivity, infiltration characteristics were determined for all depth ranges. The infiltration parameters Fc and A for soils were estimated by using the process described by Beasley and Huggins (1980) in ANSWERS Users manual. The P values for various soil texture were obtained from the table on P values (Beasley and Huggins, 1980). Soil erodibilities (USLE-K factor) were estimated using the equation presented by Wischmeier and Smith (1978). The values of soil parameters and infiltration parameters used in simulation study are given in Table 2. The LULC information of the area were generated from IRS satellite digital data as described in this section. The forest and permanent vegetation boundaries were digitized and updated using False Colour

Slope (%)

Proportion (%)

!1 1–2 2–3 3–4 4–5 5–6 O6

18.5 39.39 25.93 10.80 3.30 1.45 0.64 100.00

Composits (FCC) images of summer season (11th May 1994). and ground truth information. Twenty two signature classes of LULC were generated based on ground truth of the study area. There after a two-step classification procedure was followed to achieve higher classification efficiency. Initially, the whole watershed was classified for all twenty two signature classes, using the complete watershed area as an input to a Maximum Likelihood Classifier (MLC) program. Secondly, only the forest boundary area was classified, using forest class signatures of degraded, open and dense forests, and then it was merged with the previously classified image. Classification accuracy was estimated to be 88% by using a Maximum Likelihood Report (MLR) program, which simply compares ground truth pixels with classified pixels through a confusion matrix. LULC classifications matched closely with the ground truth data. After classification, the twenty two classes were merged into ten prominent classes as given in Table 1. Important watershed characteristics such as watershed drainage area/boundary, field and channel slope, aspect and drainage network were extracted using

R. Singh et al. / Journal of Hydrology 318 (2006) 184–199

189

Table 2 Watershed soil properties used in the ANSWERS model simulations Soil texture class/soil parameter(s)

Sandy loam

Loam

Clay loam

Sand

‘A’ Horizon depth (mm) Wilting point, WP (% sat.) Total porosity, TP (%) Field capacity, FP (% sat.) Final infiltration capacity, Fc (mm/h) Maximum infiltration capacity, A (mm/h) Parameter for infiltration, P Control zone depth, DF (mm) USLE Ka factor

162 (145–170) 21 (16–25) 44 (41–46) 51 (46–53) 21.0 (19–23) 110 (95–120) 0.55 (0.5–0.6) 81.0 (57–105) 0.269 (0.26–0.32)

212 (190–225) 28 (25–32) 46 (42–49) 68 (63–71) 6.8 (5.0–8.5) 62 (48–76) 0.6 (0.5–0.65) 106.0 (72–138) 0.478 (0.47–0.49)

240 (225–252) 38 (34–41) 48 (44–52) 76 (60–70) 2.0 (1.6–2.3) 17 (12–21) 0.65 (0.6–0.7) 120.0 (84–156) 0.363 (0.36–0.43)

138 (125–140) 18 (13–25) 42 (40–45) 47 (42–51) 30.0 (24–34) 155 (137–169) 0.50 (0.4–0.55) 69.0 (48–90) 0.170 (0.16–0.19)

a USLE K factor varies with initial soil moisture condition prior to storm % sat. Percentage saturation () Values in parentheses indicate range of parameter values for Banha watershed.

DEM of varying resolutios from 30 to 150 m through EASI/PACE and IDRISI GIS. The spatial resolution variation from 30 to 150 m influences the accuracy of watershed characteristics extracted from DEM. The flow path length and watershed slope decrease as cell size (resolution) increases. DEM of 30 m resolution resulted into automatic extraction of watershed characteristics most accurately with variations less than 10%. Hence watershed parameters extracted from DEM of 30 m resolution were used for the model simulation studies. Stream ordering of the extracted drainage network was carried out using the criteria of stream ordering suggested by Strahler (1964). There is no sub surface drainage system used in the watershed. Therefore tile drainage coefficient was considered to be zero for simulations. The typical values of the crop parameters such as USLE C and P factors for watershed land uses/crops and management practices adopted in the study watershed were considered from the available literature (Wischmeier and Smith, 1978; Beasley and Huggins, 1980; Park and Mitchell, 1983; Chakravorty, 1993). The relative erosiveness (C 0 ) parameter used in the ANSWERS is a direct combination of USLE cover and management (C) factor and practice (P) factor with seasonal adjustment. The values of C 0 parameter for the ANSWERS were obtained by multiplying the USLE C factor value with USLE P factor value at the time of storm occurrence. The percentage cover was assumed to vary during the monsoon season for different land uses and crops.

The antecedent soil moisture (ASM) values prior to occurrence of simulation storms were estimated using the moisture balance equation and procedure suggested by Beasley and Huggins (1980) based on rainfall pattern of 30 days prior to storm under simulation. Different values of ASM for various soil types of the watershed were used in the model calibration process assuming that uniform rainfall occurs over the entire catchment/watershed. The paddy is the main crop in the watershed which is cultivated under terraced and contoured management practice. The other upland crops namely maize and millets are cultivated under straight row and contour bunded management practices. The most important channel properties are slope, width and roughness. The measured average width for each order stream was considered for simulation. The Manning’s roughness coefficient for each channel was estimated by using the equation given by Cowan (1956). The model input files were prepared at 30 m cell size using text [ASCII] files of input parameters GIS layers and developed computer programs. 3.3. Model performance evaluation and storm classification The model performance was evaluated using the criteria suggested by ASCE Task Committee (1993), Haan et al. (1995) and Legates and McCabe (1999). Statistical parameters such as per cent deviation, Dv

190

R. Singh et al. / Journal of Hydrology 318 (2006) 184–199

(Martinec and Rango, 1989), Nash and Sutcliffe’s coefficient of efficiency, E (Nash and Sutcliffe, 1970), coefficient of performance (CPA 0 Z1KE), coefficient of determination (R2), root mean square error (RMSE), index of agreement of difference, IOA-d (Willmott, 1981) and Student’s t test for significance difference at 95% confidence level were used to test the model simulation results. For the purpose of describing the model simulation behaviour under varying conditions, considering rainfall pattern of the study area, the used storms were divided tentatively into four classes based on rainfall amount namely, very small storm (rainfall%25 mm), small storm (25 mm!rainfall%50 mm), medium storm (50 mm!rainfall%75 mm) and large storm (rainfallO75 mm). The storms were also categorized as very low intensity storm (I30%15 mm/h), low intensity storm (15 mm/h!I30%30 mm/h), medium intensity storm (30 mm/h!I30%45 mm/h), high intensity storm (45 mm/h!I30%60 mm/h) and very high intensity storm (I30O60 mm/h).The underprediction /overprediction by the model within or equal to 20% of observed values as criteria of success suggested by Bingner et al. (1989) is considered acceptable level of accuracy for the simulations. The underpredictions/overpredictions with per cent deviation less than 10% are considered as low (slight), 10–20% as moderate, 20–30% as severe and greater than 30% as very severe.

4. Results and discussion The ANSWERS version 4.840815 was used for hydrological simulations of the watershed. In this version rainfall detachment coefficient was increased by two times, flow detachment coefficient by five times and transport coefficients by 10% of the original version values. 4.1. Model calibration Sixteen storms (8 storms of 1993 and 8 storms of 1994) were used for model calibration. The model was calibrated using trial and error procedure of parameter adjustment and optimisation. After each parameter adjustment, the simulated and observed

runoff, peak flow and sediment yield values were compared to judge the improvement in simulations. The model runs successfully for the simulation control parameters: the maximum project dischargeZ50.8 mm, time increment step for solving continuity equationZ30 s, Number of hydrograph print linesZ101, time increment step for infiltration simulationZ90 s and number of increments on segment curveZ350. During initial runs of the ANSWERS, the spatial performance of the model was evaluated by varying the grid size from 30 to 150 m with the increment of 30 m (i.e 30, 60, 90, 120 and 150 m) for simulating runoff, peak flow and sediment yield. The runoff, peak flow and sediment yield simulations by the model decrease as cell size increases from 30 to 150 m. The runoff and sediment yield simulations are not observed to be significantly different from the observed values up to 120 m cell size. However, model simulates peak flow at acceptable accuracy for 30 m cell size only. Based on these findings, the grid resolution of 30 m was used for the ANSWERS’s simulations during model calibration and validation process. In order to determine the relative sensitivity of the ANSWERS input parameters on model outputs, sensitivity analysis was performed. The parameters such as ASM, control zone depth (DF), field slope, Manning’s roughness coefficient for overland flow (n) and channel flow (nC), soil erodibility (K) factor and relative erosiveness (C 0 ) parameter were considered for parameters sensitivity analysis. The selected parameters were varied within the prescribed range keeping the other constant. The ANSWERS input parameters sensitivity analysis revealed that runoff simulations were most sensitive to ASM followed by DF, Manning’s roughness coefficients and field slope. The peak flow simulations were most sensitive to ASM followed by DF, Manning’s roughness coefficients for channel and overland flow and field slope. The variables most significantly affecting the sediment yield in descending order of significance are C 0 parameter, K factor, ASM, DF, field slope and Manning’s roughness coefficient (n). During the model calibration process, the sensitive parameters such as ASM and field slope were not adjusted as these were known and measured parameters. The LULC based parameters were varied according to vegetative growth stages (VGS). Among

R. Singh et al. / Journal of Hydrology 318 (2006) 184–199

these parameters, the C 0 parameter and Manning’s n for overland flow were found to be the sensitive parameters influencing model outputs significantly. Manning’s roughness coefficients for channel were varied according to the condition of channel with respect to vegetation growth and types of channel. The calibrated values of Manning’s roughness coefficient (n) for channel flow were found to be 0.035, 0.045, 0.055 and 0.065 for channels of order 1, 2, 3, and 4 respectively. The values of all LULC based parameters increased with the increase in vegetative growth in the watershed, except C 0 parameter, which decreased with the increase in vegetative growth. The calibrated values of land use and land cover based input parameters of the ANSWERS such as potential interception volume (PIT), percentage of vegetative cover (PER), roughness coefficient (RC), maximum roughness height (HU) and relative erosiveness parameter (C 0 ) are given in Table 3. The USLE is not applicable for single rainfall event. The erodibility factor (K) varies significantly with storms and initial soil moisture conditions. During model calibration, USLE-K factor values for different soil types were varied linearly from 35 to 100% with respect to corresponding soil moisture from wilting point to 100% saturation level. The control zone depth (DF) was varied from 35 to 65% of ‘A’ horizon during simulation. Among soil parameters, ASM and DF were found to be potential variable influencing model simulations significantly. For the calibration trials (the rainfall and flow detachment coefficients equal to the ANSWERS original version values and transportation coefficients 10% higher than the ANSWERS original version values, DFZ50% of the ‘A’ horizon depth and K-factors adjustment from 35 to 100%), the simulated sediment yields were in close agreement with the observed values. The model simulation results and performance evaluation statistics for the storms used in model calibration are presented in Table 4. In the present study, the simulation results are considered within the acceptable level of accuracy/deviation if the average Dv%20% for a particular type of storms. 4.1.1. Runoff and peak flow simulation The model simulates the runoff and peak flow less than the observed values with deviations less than 10% for very small storms of medium intensity

191

(June/18/1993 and October/09/1993) which occurred under dry antecedent soil moisture condition (AMCI). The runoff and peak flow values are underpredicted for the small storms of medium intensity (June/25/1993 and October/12/1993) which occurred under average soil moisture condition (AMC-II). This indicates that under AMC-I and AMC-II, the runoff and peak flow simulations by the ANSWERS for very small to small storms of medium intensity are reasonably well with the acceptable level of deviation. The model also underpredicts the runoff and severely overpredicts the peak flow for the small and large storms of very high rainfall intensity (August/21/1994 and July/23/1994) occurring under soil moisture conditions close to upper range of AMC-II. Peak flow overprediction by the model for large storm of July/23/1994 is also shown in Fig. 3. The model underpredicts peak flow with Dv!5% and runoff moderately for small and medium storms of medium intensity (September/10/1994 and October/07/1994) occurring under the lowest range of wet soil moisture condition (AMC-III). The runoff values are moderately underpredicted with Dv!15% for most of the small to medium storms of medium intensity under AMC-III. For majority of the medium storms of medium and high rainfall intensities occurring under AMC-III the model overpredicts the peak flows with deviations ranging from 12.24 to 20. 72%. Under AMC-III, the model simulates the peak flows well with Dv!7% for small storms of medium and high intensities. Fig. 3 shows the hydrograph for one such storm of August/07/1994 which indicates good match of the simulated hydrograph with the observed one (CPA 0 !0.15). The model slightly overpredicts the runoff and peak flow with Dv!10% for large storm of high intensity (September/14/1993) occurring under very wet soil condition (Table 4). However, the runoff is slightly overpredicted and peak flow is severely overpredicted for the storm of August/24/1993 of similar soil condition. The underprediction and overprediction of runoff and peak flow by the ANSWERS is also reported by Bingner et al. (1989) and Montas and Madramootoo (1991) for similar storm conditions. The reason of overprediction as suggested by them could be inability of Holtan equation to simulate infiltration rate accurately under very wet soil condition. It is evident from

192

R. Singh et al. / Journal of Hydrology 318 (2006) 184–199

Table 3 Calibrated values of land use and surface parameters for the ANSWERS model Land use/land cover

Paddy

Upland crops

Degraded forest

Open forest

Dense forest

Grass land

Fallow land

Waste land-1

Waste land-2

Water

Vegetation growth stage

Land use and surface parameters PIT

PER

RC

HU

Overland Manning’s ‘n’

Relative erosiveness C 0

VGS-1 VGS-2 VGS-3 VGS-4 VGS-1 VGS-2 VGS-3 VGS-4 VGS-1 VGS-2 VGS-3 VGS-4 VGS-1 VGS-2 VGS-3 VGS-4 VGS-1 VGS-2 VGS-3 VGS-4 VGS-1 VGS-2 VGS-3 VGS-4 VGS-1 VGS-2 VGS-3 VGS-4 VGS-1 VGS-2 VGS-3 VGS-4 VGS-1 VGS-2 VGS-3 VGS-4 –

0.1 0.3 0.8 0.9 0.1 0.3 0.9 1.0 0.7 0.9 1.7 2.0 0.8 0.8 1.8 2.0 1.0 1.0 1.8 2.0 0.2 0.5 0.7 0.8 0.1 0.2 0.2 0.3 0.1 0.2 0.2 0.3 0.2 0.3 0.4 0.4 0.0

0.10 0.30 0.75 0.80 0.10 0.30 0.75 0.80 0.20 0.40 0.55 0.70 0.30 0.45 0.65 0.75 0.40 0.60 0.75 0.80 0.30 0.50 0.65 0.75 0.10 0.20 0.30 0.45 0.15 0.20 0.30 0.40 0.20 0.30 0.40 0.50 0.00

0.33 0.34 0.35 0.40 0.33 0.34 0.35 0.40 0.50 0.51 0.52 0.54 0.50 0.51 0.52 0.54 0.55 0.55 0.56 0.58 0.40 0.41 0.41 0.43 0.32 0.33 0.33 0.35 0.36 0.37 0.38 0.40 0.41 0.42 0.42 0.43 0.10

30 40 60 80 30 40 70 80 80 80 85 100 90 90 95 105 100 100 105 110 10 20 20 25 30 30 40 50 40 40 45 50 50 50 55 60 0.10

0.038 0.080 0.150 0.201 0.038 0.051 0.090 0.100 0.112 0.135 0.151 0.200 0.152 0.176 0.201 0.260 0.203 0.260 0.280 0.300 0.050 0.070 0.080 0.100 0.038 0.038 0.041 0.042 0.032 0.035 0.038 0.040 0.061 0.072 0.080 0.080 0.001

0.153 0.150 0.140 0.140 0.141 0.115 0.085 0.060 0.211 0.198 0.191 0.186 0.009 0.007 0.006 0.005 0.004 0.003 0.003 0.003 0.145 0.110 0.091 0.090 0.146 0.146 0.139 0.139 0.801 0.701 0.601 0.601 0.151 0.141 0.131 0.115 0.000

VGS-1: rough/fallow and crop pre sowing stage, VGS-2: crop seeding/sowing and branching stage; VGS-3: crop establishment stage VGS-4: crop growth and maturity stage. PIT: Potential Interception Volume (mm) PER: Percentage of vegetative cover; RC: Roughness coefficient HU: Maximum roughness height (mm); C 0 : Relative erosiveness parameter.

the discussion that under AMC-III the ANSWERS simulates the runoff reasonably well with the acceptable accuracy for small to large storms of medium to high intensity. The simulated runoff values for the storms used in calibration are found to be distributed on lower side of the 458 line (1:1 line) for most of the storms which indicate the trend of underprediction (Fig. 2a).

The model simulated runoff values with average DvZK9.32%, R2Z0.976, RMSEZ3.98 mm, and EZ0.964. The statistical tests indicate that the model simulates runoff well within the acceptable level of accuracy. Student’s t test for difference (t-diff) indicates that the difference between the mean values of observed and simulated runoff is significant at 95% level of confidence. Here the interesting point of

R. Singh et al. / Journal of Hydrology 318 (2006) 184–199

193

Table 4 Runoff, peak flow and sediment yield simulations for the storms used in calibration of the ANSWERS model Storm date (Mo/ Da/Yr) 06/18/93 10/09/93 06/25/93 08/22/93 08/24/93 10/12/93 06/10/94 08/07/94 08/21/94 09/10/94 09/05/93 07/02/94 08/25/94 10/07/94 09/14/93 07/23/94

Storm size (mm) 0–25 O25–50

O50–75

O75

Rainfall (mm)

I30 (mm/h)

20.9 21.6 30.3 37.7 29.3 31.3 37.1 43.1 49.6 49.1 58.4 54.9 52.5 59.5 123.8 99.1

41.5 37.8 43.0 43.2 34.8 41.9 29.8 56.8 69.9 40.2 37.8 41.4 53.8 44.0 55.6 75.6

AMC

I I II III III II I III III III III III III III III III

Observed

Simulated

Hydrograph CPA 0

API (mm)

RO (mm)

Qp (m3/s)

SY! 10K3 (t/ha)

Deviation, Dv (%)

0.9 3.8 24.1 48.9 62.1 21.6 0.0 43.6 35.2 33.7 57.1 41.5 50.6 33.5 66.4 36.3

3.09 3.05 7.98 19.67 12.24 8.13 5.01 23.21 32.73 25.87 32.67 33.06 29.92 34.23 80.71 62.83

4.41 4.17 11.21 26.62 15.23 8.81 5.90 32.26 41.96 29.38 36.53 36.51 31.56 40.10 78.25 68.95

76.99 70.42 154.11 238.05 127.33 122.31 90.38 334.51 490.98 260.99 302.59 426.01 386.27 352.98 1019.78 1229.31

K7.77 K5.22 K21.18 0.251 K8.52 K4.32 K24.21 0.313 K10.65 K1.16 K4.18 0.092 K12.86 K6.87 14.45 0.080 6.13 K22.19 11.10 0.161 K2.46 K22.70 K5.02 0.246 K11.58 K30.68 K23.25 0.684 K7.93 K3.81 15.96 0.110 K15.83 28.46 1.65 0.103 K14.22 K1.77 K9.00 0.153 K6.95 12.24 7.72 0.067 K17.60 20.68 4.18 0.239 K13.60 20.72 2.89 0.112 K17.18 K2.59 K9.69 0.115 6.49 5.21 20.18 0.362 K14.58 33.87 K30.23 0.125

RO

Qp

SY

Performance evaluation statistical parameters Mean mm, m3/s, !10K3 t/ha Average Dv (%) Coefficient of efficiency (E) Coefficient of determination (R2) Root mean square error (RMSE) mm, m3/s, !10K3 t/ha Index of agreement for difference (IOA-d) Student’s t test for difference (t-diff) t-table value (t0.975,15) for two tailed distribution

25.90 K9.32 0.964 0.976 3.980 0.991 2.717 2.131

discussion is that most of the statistical techniques for performance evaluation of hydrologic models suggest the suitability of the model, whereas most commonly used Student’s t does not support the statement. Therefore, it emphasizes the need of using more than one statistical tests for performance evaluation of hydrologic and NPS pollution models before drawing any conclusion about their suitability in actual applications. Similar view about the model performance evaluation was also expressed by Legates and McCabe (1999). The scattergram between the observed and simulated peak flow values for the storms used in calibration is shown in Fig. 2b. The high value of the coefficient of determination (R2Z0.963) indicates a close agreement between the measured and the simulated peak flow values. The model simulated peak flow with average DvZ1.24%, RMSEZ

29.49 1.24 0.881 0.963 7.290 0.976 K1.645 2.131

355.19 K3.04 0.884 0.885 108.54 0.969 0.287 2.131

0.201

7.29 m3/s and EZ0.881. The statistical comparison indicates that the model simulates peak flow well within the acceptable level of accuracy(average Dv%20%). For most of the storms, the simulated runoff hydrographs recede rapidly as compared to the observed hydrographs. Similar result is also reported by Montas and Madramootoo (1991). The peaks of the simulated hydrographs for model calibration are found to occur after the observed values (10 out of 16 storms). The simulated and observed runoff hydrographs are found to match well within the acceptable level of accuracy as indicated by the CPA 0 values less than 0.20 for 10 storms occurring under AMC-II and AMC-III. However, the CPA 0 values are found to be greater than 0.20 for the hydrographs of storms occurring under dry and very wet soil conditions indicating not good matching of the simulated hydrographs with the observed ones.

194

R. Singh et al. / Journal of Hydrology 318 (2006) 184–199

Fig. 2. Comparison between the observed and simulated runoff, peak flow and sediment yield values for calibration and validation storms. (a) Runoff simulations, (b) Paeak flow simulations, (c) Sediment yield simulations.

4.1.2. Sediment yield simulation The model severely underpredicts the sediment yield values for very small storms of medium intensity and small storm of low intensity (June/10/1994) occurring under AMC-I. The simulated sediment yield values match well with the observed values (Dv%5%) for

small storms of medium intensity occurring under AMC-II (Table 4). Under antecedent soil moisture slightly higher than AMC-II, the model underpredicts sediment yield values with Dv!10% for small and medium storms of medium intensity (September/10/ 1994 and October/07/1994). The sediment yield is very

R. Singh et al. / Journal of Hydrology 318 (2006) 184–199

severely underpredicted with DvZK30.23% for large storm of very high intensity. Similar to runoff and peak flow simulations, the model overpredicts the sediment yield for large storms of high intensity (September/14/ 1993) and for small storm of medium intensity (August/ 24/1993) under very wet soil condition. For most of the medium storms of medium and high rainfall intensities occurring under AMC-III the sediment yield simulations are found to be close to the observed values. Based on above discussion, it is inferred that the ANSWERS underpredicts the sediment yield values for the storms occurring under dry, average and slightly more than the average soil moisture conditions. The sediment yield underpredictions by the ANSWERS were observed in other studies as well (Bingner et al., 1989; Montas and Madramootoo, 1991). The probable reasons of underpredictions may be non consideration for re-suspension of deposited soil particles and channel erosion by the ANSWERS. The sediment

195

yield is overpredicted for the storms of medium and high intensities occurring under wet and very wet soil conditions. The simulated sediment yield values are found to be distributed about the 1: 1 line showing almost unbiased simulation (Fig. 2c). In general the simulated sediment yield values are found to be well within the acceptable level of deviation leaving few exceptions. The average Dv of the simulated sediment yield values is found to be K3.04% which indicates the trend of slight underprediction. Statistical tests (R2Z0.885, RMSEZ108.54!10K3 t/ha and EZ 0.884) indicate that the model simulates the sediment yield well within the acceptable level of accuracy. 4.2. Model validation For model validation fifteen storms (8 storms of 1995 and 7 storms of 1996) were used. The input parameters values optimized during model calibration

Table 5 Runoff, peak flow and sediment yield simulations for the storms used in validation of the ANSWERS Storm date (Mo/ Da/Yr) 06/19/95 09/05/95 07/10/96 07/06/95 07/10/95 08/18/95 08/20/95 08/25/95 09/04/95 07/17/96 08/17/96 08/18/96 09/03/96 07/18/96 08/02/96

Storm size (mm) 0–25

O25–50

O75

Rainfall (mm)

I30 (mm/h)

AMC

25.0 23.7 25.0 32.7 33.0 48.6 30.7 27.2 35.8 26.5 27.5 34.1 26.1 75.7 87.8

36.6 13.2 36.0 55.4 64.0 62.0 53.8 52.4 41.0 51.4 51.4 28.0 50.2 68.2 44.8

I III I I III III III III III III III III III III III

Observed

Simulated

Hydrograph CPA 0

API (mm)

RO (mm)

Qp (m3/s)

SY!10K3 (t/ha)

Deviation, Dv (%) RO

Qp

SY

2.1 57.5 7.5 10.5 52.5 43.3 75.6 54.3 33.4 38.0 64.0 76.3 39.3 54.8 43.5

4.12 8.18 4.21 7.33 19.16 27.80 14.11 11.51 16.09 11.03 11.90 16.86 10.56 48.86 55.40

5.68 9.35 5.89 10.01 26.17 38.10 18.13 14.54 19.46 14.72 15.10 18.11 12.13 50.39 52.20

82.33 59.20 81.95 171.04 314.92 424.77 192.49 171.16 166.76 174.76 169.24 135.76 149.40 698.03 617.44

K16.75 K11.00 K15.68 K3.14 K11.69 K3.31 K4.46 K16.42 K15.16 K9.07 K6.64 1.19 K5.49 1.27 K5.54

K9.68 K17.54 K8.32 16.38 20.83 20.87 10.76 K7.70 K16.19 K11.62 K11.99 K15.07 K12.45 39.59 K31.65

K23.64 K22.24 K18.61 1.99 8.19 6.04 13.88 4.71 K8.70 13.55 14.84 K17.99 5.10 11.71 K13.24

0.255 0.180 0.365 0.166 0.125 0.085 0.105 0.209 0.199 0.087 0.074 0.061 0.128 0.271 0.257

Performance evaluation statistical parameters Mean mm, m3/s, !10K3 t/ha Average Dv (%) Coefficient of efficiency (E) Coefficient of determination (R2) Root mean square error (RMSE) mm, m3/s, !10K3 t/ha Index of agreement for difference (IOA-d) Student’s t test for difference (t-diff) t-table value (t0.975,14) for two tailed distribution

17.81 K8.13 0.991 0.996 1.399 0.998 3.940 2.141

20.67 K2.25 0.741 0.829 7.306 0.944 K0.195 2.141

240.62 K1.63 0.965 0.970 34.95 0.992 K0.425 2.141

0.171

196

R. Singh et al. / Journal of Hydrology 318 (2006) 184–199

process were used for hydrological simulations. The simulated runoff, peak flow, sediment yield and model performance evaluation parameters values for the storms used in model validation are presented in Table 5. 4.2.1. Runoff and peak flow simulation The ANSWERS model simulates the runoff and peak flow reasonably well although underpredicted for very small storms of medium intensity (June/19/1995 and July/10/1996) occurring under AMC-I. The model is found to behave in the similar ways as in the case of calibration storms for very small storms occurring under AMC-I. The model underpredicts the runoff and peak flow with an acceptable level of deviation for small storm of medium intensity (September/04/1995) occurring under soil moisture conditions close to AMC-II (Table 5). Under AMC-III, the model simulates the runoff well and overpredicts peak flow for majority of

the small storms of high and very high intensities. The observed and simulated runoff hydrographs of one such storm of August/18/1995 (Fig. 3) matched satisfactorily (CPA 0 Z0.085). The runoff and peak flow were simulated well though underpredicted for small storms of high intensities (July/17/1996 and September/03/ 1996) occurring under wet conditions close to AMC-II. The model overpredicts the runoff with DvZ1. 27% and peak flow with Dv of 39.59% for large storm of very high intensity (July/18/1996). However, model underpredicts slightly runoff and peak flow severely for large storm of medium intensity (August/02/1996) under AMC-I. The model simulated runoff well (slightly overpredicted with DvZ1.19%) and peak flow with moderate underpredicton for a large duration small storm of low intensity (August/18/1996). Fig. 3 illustrates the peak flow underprediction for the storm of August/18/1996. It is evident from the discussion

Fig. 3. Comparison between the simulated and observed hydrographs for the storms used in ANSWERS model calibration and validation.

R. Singh et al. / Journal of Hydrology 318 (2006) 184–199

that under initial wet soil condition, the ANSWERS simulates the runoff well and peak flow reasonably well within the acceptable level of accuracy for the small storms of low to high rainfall intensity. Similar to calibration storms, the high value of the coefficient of determination (R2Z0.996) indicates a close agreement between the measured and the simulated runoff values for the validation storms also (Fig. 2a). The average Dv of the simulated runoff values is found to be K8.13% indicating trend of underprediction. Other performance evaluation statistical parameters for runoff simulation are RMSEZ1.399 mm and EZ0.991. In general, the peak flows are underpredicted by the model for the storms of all sizes with low, medium and high intensities occurring under dry and wet soil moisture conditions. However, the peak flows are overpredicted by the model for the small and large storms of high and very high intensities occurring under similar moisture conditions. The overall simulated peak flow values are found to be less than the observed values with an average DvZK2.25%. The simulated peak flow values are found to be distributed on both sides of the 458 line (1:1 line) indicating a small bias in the simulated peak flow values (Fig. 2b). The R2 value equal to 0.829 indicates a satisfactory relation between the measured and the simulated peak flow values. Statistical parameters values (RMSEZ7.306 and EZ0.741) indicate the peak flow simulation within the acceptable level of accuracy. Similar to calibration storms, the simulated hydrographs recede rapidly as compared to that of observed hydrographs for the most of the storms used in model validation. This effect is probably due to the fact that the ANSWERS does not adequately simulate the sub surface flow. The simulated and observed runoff hydrographs match well as indicated by the CPA 0 values less than 0.20 for 9 storms occurring under wet soil moisture conditions. The peak of simulated hydrographs is found to occur after the observed peak for 9 storms out of the 15 storms used for model validation. 4.2.2. Sediment yield simulation The sediment yield values are underpredicted with deviations ranging from K18.61% to K23.64% for very small storms of medium intensity occurring under AMC-I. The sediment yield is found to be in

197

close agreement (DvZ1.99%) to the observed value for small storm of high intensity (July/6/1995) which occurred under relatively dry soil condition (Table 5). Under AMC-III, the sediment yield was severely underpredicted for very small storm of very low intensity (September/05/1995).The model overpredicts the sediment yield values with Dv!15% for small storms of high and very high intensities. The model overpredicts the sediment yield for large storm of very high intensity (July/18/1996). However, the sediment yield for large storm of medium intensity (storm of August/2/1996) is underpredicted with DvZK13.24%. The storms of September/4/1995, July/17/1996 and September/03/1996 had antecedent soil moisture contents only marginally higher than that of AMC-II. For these storms also the simulation results are found to be satisfactory with acceptable level of deviations indicating the validity of the model for storms occurring under average soil moisture condition (AMC-II). The simulated sediment yield values for validation storms are found to be distributed about the 1:1 line showing almost unbiased simulation (Fig. 2c). The average deviation of K1.63% of the simulated sediment yield values indicates the trend of slight underprediction. The performance evaluation parameters values (R2Z 0.970, RMSEZ34.95!10K3 t/ha, and EZ0.965) indicate that the model simulates the sediment yield well within the acceptable level of deviation. As a whole it is inferred that the model simulates the runoff, peak flow and sediment yield satisfactorily for majority of the storms used in model calibration and validation with average deviations (Dv) less than 10% and average value of coefficient of efficiency (E) greater than 0.800. By and large the model simulation results remain consistent during calibration and validation processes with only few exceptions. On the basis of calibration and validation results, it is inferred that the ANSWERS model can be successfully used for simulating the watershed response under varied soil moisture and watershed conditions.

5. Conclusions The following conclusions are drawn from the present study of the ANSWERS model:

198

R. Singh et al. / Journal of Hydrology 318 (2006) 184–199

(i) The ANSWERS model simulates surface runoff, peak flow and sediment yield with the average per cent Dv equal to K9.32, 1.24 and K3.04 and the coefficient of efficiency (E) equal to 0.964, 0.881, and 0.884 respectively for the storms used in model calibration. For the ANSWERS model validation, average per cent Dv equal to K8.13, K2.25 and K1.63 and E equal to 0.991, 0.741, and 0.965 were obtained for surface runoff, peak flow and sediment yield simulations respectively. The model showed consistency in simulations during calibration and validation processes. (ii) The peaks of the simulated hydrographs were found to occur after the observed values for majority of the storms considered for model calibration and validation. The model underpredicts runoff, peak flow and sediment yield for majority of storms within acceptable level of deviation. (iii) The ANSWERS model is capable of simulating runoff, peak flow and sediment yield from a watershed with the acceptable level of deviation (average Dv!20%) under varied soil moisture and rainfall conditions. This indicates the suitability of the model application for the ungauged watersheds of similar hydro-geological characteristics. Acknowledgements The authors are thankful to Dr. S. Sudhakar, former Head, Regional Remote Sensing Service Centre (Indian Space Research Organisation, Dept. of Space, Government of India), Kharagpur for providing the necessary facilities to carry out this work. We also extend our thanks to Director (Soil Conservation), DVC, Hazaribagh (Jharkand) and Project Coordinator, Indo German Bilateral Project on watershed management New Delhi for providing hydrological and soil resource data.

Appendix A. List of notations AGNPS Agricultural Non Point Source ANSWERS Areal Non point Source Watershed Environment Response Simulation

Agril Agricultural AMC Antecedent Soil Moisture Condition API Antecedent Precipitation Index ASAE American Society of Agricultural Engineers ASCE American Society of Civil Engineers AWRA American Water Resources Association Bull Bulletin CPA 0 Coefficient of Performance Dept. Department Dv Per cent deviation E Nash and Sutcliffe’s coefficient of efficiency EASI Environmental Analysis and Scientific Interface FESHM Finite Element Storm Hydrograph Model Fig Fig. h Hour ha Hectare IRS Indian Remote Sensing IDRISI Window based GIS developed by Clark University, USA I30 Rainfall Intensity based on 30 min duration J Journal km Kilometre LISS Linear Imaging and Self Scanning m Metre mm Milli metre m3/s Cubic metre per second n Manning’s roughness coefficient nC Manning’s roughness coefficient for channel flow PACE Picture Analysis Correction and Enhancement R2 Coefficient of determination Res Research Resour Resources s Second t Ton t-diff Student’s t-test for significant difference Trans Transactions USLE Universal Soil Loss Equation

R. Singh et al. / Journal of Hydrology 318 (2006) 184–199

References ASCE Task Committee on Definition of Criteria for Evaluation of Watershed Models, 1993. Criteria for evaluation of watershed models. J. Irr. Drain. Eng. 119 (3), 429–442. Beasley, D.B., Huggins, L.F., 1980. ANSWERS User’s Manual. Department of Agricultural Engineering, Purdue University, West Lafayette, In. Beasley, D.B., Huggins, L.F., Monke, E.J., 1980. ANSWERS: a model for watershed planning. Trans. ASAE 23 (4), 938–944. Beasley, D.B., Huggins, L.F., Monke, E.J., 1982. Modeling sediment yield from agricultural watersheds. J. Soil Water Conserv. 37 (2), 113–117. Bingner, R.L., Murphee, C.E., Mutchler, C.K., 1989. Comparison of sediment yield models on watershed in Mississippi. Trans. ASAE 32 (2), 529–534. Bingner, R.L., Murphee, C.E., Mutchler, C.K., 1992. Predictive capabilities of erosion models for different storm sizes. Trans. ASAE 35 (2), 505–513. Chakravorty, A.K., 1993. Strategies for watershed management planning using remote sensing techniques. Photonirvachak: J. Indian Soc. Remote Sens. 21 (2), 87–98. Cowan, W.L., 1956. Estimating hydraulic roughness coefficients. Agric. Eng. 37 (7), 473–475. De Roo, A.P.J., Hazelhoff, I., Burroh, P.A., 1989. Soil erosion modeling using ANSWERS and GIS. Earth Surf. Proc. Land. 14 (6–7), 517–532. Garbrecht, J., Martz, L.W., 1993. Network and subwatershed parameters extracted from digital elevation models: the Bill Creek experience. Water Resour. Bull. AWRA 29 (6), 909–916. Haan, C.T., Allred, B., Storm, D.E., Sabbagh, G.J., Parbhu, S., 1995. Statistical procedure for evaluating hydrological/water quality models. Trans. ASAE 38 (3), 725–733. Huggins, L.F., Monke, E.J., 1966. The mathematical simulation of the hydrology of small watersheds. Technical Report No. 1. Water Resour. Res. Center Purdue University. (130 p). Legates, D.R., McCabe Jr.., G.J., 1999. Evaluating the use of ‘goodness-of-fit’ measures in hydrologic and hydroclimatic model validation. Water Resour. Res. 35 (1), 233–241. Martinec, J., Rango, A., 1989. Merits of statistical criteria for the performance of hydrological models. Water Resour. Bull., AWRA 25 (2), 421–432.

199

Misra, K., 1999. Watershed management activities in Damodar Valley Corporation at a glance, Soil Conservation Department. Damodar Valley Corporation, Hazaribagh, India. Misra, N., Satyanarayana, T., 1991. A new approach to predict sediment yield from small ungauged watersheds. Agric. Eng. Div. J. Inst. Eng. (India) 37, 30–36. Montas, H.J., Madramootoo, C.A., 1991. Using the ANSWERS model to predict runoff and soil loss in South Western Quebec. Trans. ASAE 34 (4), 1751–1762. Murthy, B.N., 1980. Sedimentation Studies in Reservoirs, Technical Report No. 20, Research Scheme applied to River Valley Projects. Central Board of Irrigation and Power, New Delhi Vol. 1. 10 p. Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models, Part-1: A discussion of principles. J. Hydrol. 10 (3), 282–290. Park, S.W., Mitchell, J.K., 1983. MODANSW (A Modified ANSWERS) Model Users Guide. College of Agriculture, Agricultural Experiment Station, Agricultural Engineering Department, University of Illinois at Urbana Champaign. Razavian, D., 1990. Hydrologic responses of an agricultural watershed to various hydrologic and management conditions. Water Resour. Bull. AWRA 26 (5), 777–785. Rewerts, C.C., Engel, B.A., 1991. ANSWERS on GRASS: Integrating watershed simulation with a GIS. ASAE Paper No. 91-2621,ASAE, St. Joseph, Michigan. Seghal, J.L., Abrol, I.P., 1983. Soil Degradation in India: Status and Impact. Oxford and IBH Publishing Company Pvt. Ltd., New Delhi, India p. 80. Strahler, A.N., 1964. Quantitative geomorphology, drainage basin and channel networks. In: Chow, V.T. (Ed.), Handbook of Applied Hydrology, Section 4-II. McGraw Hill, New York, pp. 4–39 (see also pages 4–76). Tiwari, K.N., Kannan, N., Singh, R., Ghosh, S.K., 1997. Watershed parameter extraction using GIS and remote sensing for hydrologic modeling. ASIAN-PACIFIC Rem. Sens. GIS J. 10 (1), 43–52. Willmott, C.J., 1981. On the validation of models. Phys. Geogr. 2, 184–194. Wischmeier, W.H., Smith, D.D., 1978. Predicting rainfall Erosion losses, Agriculture Handbook No. 537. U.S. Department of Agriculture, Washington DC. Young, R.A., Onstad, C.A., Bosch, D.D., Anderson, W.P., 1989. AGNPS: a non-point source pollution model for evaluating agricultural watersheds. J. Soil Water Conserv. 44 (2), 68–73.