Biosystems Engineering (2003) 85 (3), 365–379doi:10.1016/S1537-5110(03)00066-7

Available online at www.sciencedirect.com

SW}Soil and Water

Identification and Prioritisation of Critical Sub-watersheds for Soil ConservationManagement using the SWAT Model

M.P. Tripathi1; R.K. Panda2; N.S. Raghuwanshi2

1Soil and Water Engineering Department, Faculty of Agricultural Engineering, Indira Gandhi Agricultural University, Raipur 492 012,Chhattisgarh, India; e-mail of corresponding author: drmpt64@hotmail.com

2Agricultural and Food Engineering Department, Indian Institute of Technology, Kharagpur 721 302, West Bengal, India;e-mail: rkp@agfe.iitkgp.ernet.in

(Received 12 September 2001; accepted in revised form 20 March 2003; published online 21 May 2003)

A few areas of the watershed are critical and responsible for high amount of soil and nutrient losses.Implementation of best management practices is required in those critical erosion prone areas of the watershedfor controlling the soil and nutrient losses. Identification of these critical areas is essential for the effective andefficient implementation of watershed management programmes. In this study, a calibrated Soil and WaterAssessment Tool (SWAT) model was verified for a small watershed (Nagwan) and used for identification andprioritisation of critical sub-watersheds to develop an effective management plan. Daily rainfall, runoff andsediment yield data of 7 years (1992–1998) were used in this study. Data related to nutrient losses for fewstorm events of 1997 were also used. Besides these data, the topographical map, soil map, land resources dataand satellite imageries of the study watershed were used in this study.

A geographical information system was used for generating the watershed and sub-watershed boundaries,drainage networks, slope, soil series and texture maps. Supervised classification method was used for land use/cover classification from satellite imageries. The weighted average values of parameters such as runoff curvenumber, surface slope, channel length, average slope length, channel width, channel depth, soil erodibilityfactor and other soil layer data were taken for each sub-watershed to verify the model.

The calibrated SWAT model was verified for the monsoon season on daily basis for the year 1997 andmonthly basis for the years 1992–1998 for both surface runoff and sediment yield. It was also tested for theavailable data on nutrient losses. Critical sub-watersheds were identified on the basis of average annualsediment yield and nutrient losses during the period of 3 years 1996–1998. The erosion rates and their classeswere used as a criterion for identifying the critical sub-watersheds. Out of the 12 sub-watersheds, one sub-watershed fell under moderate soil loss group and five sub-watersheds fell under high soil loss group of soilerosion classes whereas other sub-watersheds fell under slight erosion classes. The study revealed that theSWAT model could successfully be used for identifying and prioritising critical sub-watersheds formanagement purposes.# 2003 Silsoe Research Institute. All rights reserved

Published by Elsevier Science Ltd

1. Introduction

The Resource considerations for implementation ofwatershed management programmes or various otherreasons related to administration or even politicalconsiderations may limit the implementation of manage-ment programmes to a few sub-watersheds only. Evenotherwise, it is always better to start managementmeasures from the most critical sub-watershed, which

makes it mandatory to prioritise the sub-watershedavailable. Watershed prioritisation is thus the ranking ofdifferent critical sub-watersheds of a watershed accord-ing to the order in which they have to be taken up fortreatment and soil conservation measures

The intensive study of individual watersheds isnecessary to enable management plans to be developedand also to apply the results of one watershed, toanother with similar characteristics. Effective control of

ARTICLE IN PRESS

1537-5110/03/$30.00 365 # 2003 Silsoe Research Institute. All rights reserved

Published by Elsevier Science Ltd

soil and nutrient losses requires implementation of bestmanagement practices in critical erosion prone areas ofthe watershed. It can be enhanced by the use ofphysically based distributed parameter models, remotesensing technique and geographic information systemthat can assist management agencies in both identifyingmost vulnerable erosion prone areas and selectingappropriate management practices.

Numerous studies have indicated that, for manywatersheds, a few critical areas are responsible for adisproportionate amount of the pollution (Dickinsonet al., 1990; Dillaha, 1990; Maas et al., 1985; Storm et al.,1988). Critical areas of non-point source pollution canbe defined both from the land resources and the waterquality perspectives (Maas et al., 1985). From the landresource perspective, critical areas are those land areaswhere the soil erosion rate exceeds the soil loss tolerancevalue. Critical areas from the water quality perspectivesare areas where the greatest improvement can beachieved with the least capital investment in bestmanagement practices.

The average soil loss value of 16�4 t ha�1 yr�1 (DhruvaNarayana, 1993) and permissible soil loss value of11�2 t ha�1 yr�1 (Mannering, 1981) can be taken intoconsideration for identifying the critical sub-watershed.Priorities can be fixed on the basis of ranks assigned toeach critical sub-watershed according to ranges of soilerosion classes described by Singh et al. (1992) for theIndian condition. They categorised the soil loss rangesinto different soil erosion classes. Soil erosion classessuch as slight (0–5 t ha�1 yr�1), moderate (5–10 t ha�1

yr�1), high (10–20 t ha�1 yr�1), very high (20–40 t ha�1 yr�1), severe (40–80 t ha�1 yr�1) and very severe(>80 t ha�1 yr�1) were reported by Singh et al. (1992).

An average soil loss tolerance value of 9�0 t ha�1 yr�1

was used for Nomini Creek watershed located inWestmoreland County, Virginia by Tim et al. (1992) intheir study for identifying the critical areas from theland source prospective. They also considered a thresh-old value for the loading rate P of 1�12 kg ha�1 yr�1.This threshold value of P loading was obtained from thework of DelRegno and Atkinson (1988). Tim et al.(1992) classified watershed areas into three classes i.e.

low, medium and high potential areas from both landresource and water quality prospective. Also for nutrientlosses a threshold value of 10mg�1l for nitrate nitrogenand 0.5mg �1l for dissolve phosphorous as described byEPA (1976) can be considered as criterion for identify-ing the critical sub-watersheds.

The Soil Conservation Department of DamodarValley Corporation (DVC) Hazaribagh, Bihar (India)has demarcated 20 prioritised sub-watersheds out of 39sub-watersheds for treating them with the appropriatesoil conservation measures (Misra, 1986). The prioriti-

sation of these sub-watersheds was based on anempirical formula developed by DVC using a limitedstream flow record of only 3 years. They consideredthree priority criteria, i.e. priority I (erosion index of 30and above), priority II (erosion index lying between 15and 30) and priority III (erosion index less than 15). Theactual formula used for calculating the erosion index IE

is as follows:

IE 5Au

3þ Agw þ Adf þ

Af

5ð1Þ

where: Au is the upland area in ha; Agw is the gulliedwasteland area in ha; Adf is the denuded forestland areain ha; and Af is the wood forestland area in ha.

Several techniques, ranging from manual overlay ofspatially -index mapped data to pollutant yield model-ling, have been used to characterise and delineate criticalareas of non-point source pollution in complex land-scapes. McHarg (1969) used a manual map overlaysystem to display the common attributes of selected landareas in order to make decisions on the type and degreeof land development that is commensurate with thephysical properties and limits of an area. Recently, therehas been a shift towards the use of computerised datamanagement systems to facilitate the delineation ofcritical areas of non-point source pollution (Hession &Shanholtz, 1988; Vieux, 1991). Some of the researchworkers used the sediment yield index Isy method forprioritisation of sub-watersheds (Karale et al., 1975,1977). They used the following equation for computa-tion of Isy.

Isy 5

PEiAeiDrð Þ

Aw

100 ð2Þ

where: Isy is the sediment yield index; Ei is the weighingvalue of erosion intensity mapping unit; Aei is the area ofthe erosion intensity mapping unit in a watershed in ha;Dr is the delivery ratio; and Aw is the total area ofwatershed in ha.

Several physically based distributed parameter models(ANSWERS, AGNPS, SHE, SWRRB and SWAT) havebeen developed to predict runoff, erosion, sediment andnutrient transport from agricultural watersheds undervarious management regimes. Among these models, Soiland Water Assessment Tool (SWAT) is the most recentone used successfully for simulating runoff, sedimentyield and water quality of small watersheds. The SWATmodel is a distributed parameter, continuous modeldeveloped by the USDA-ARS (Arnold et al., 1996,1998). The SWAT model was tested mainly on monthlyand annual basis for predicting runoff and sedimentyield (Srinivasan et al., 1993; Srinivasan & Arnold, 1994;Rosenthal et al., 1995; Bingner, 1996; Bingner et al.,1997; Peterson & Hamlett, 1998). However, Tripathi

ARTICLE IN PRESSM. P. TRIPATHI ET AL.366

et al. (1999a, 1999b) tested the SWAT model on thebasis of daily runoff and sediment yield for an Indianwatershed, namely Nagwan in the Hazaribagh district ofthe Bihar State in India. Limited research work onidentification of critical sub-watersheds and assessmentof the impact of management practices on runoff,sediment yield and nutrient losses using SWAT hasbeen reported (Arnold et al. 1999; Santhi et al., 2001).Also the testing of SWAT model on daily basis for thenutrient losses has not appeared much in the literature.

Keeping the above facts in mind, the current studywas undertaken with the use of a verified model i.e.

SWAT to identify the critical sub-watersheds on thebasis of estimated sediment yield and nutrient losses of asmall watershed for the purpose of developing theeffective management plan.

2. Theoretical considerations

SWAT model is a distributed parameter model thatoperates on a daily time step. The major goal of themodel development was to predict the impact ofmanagement measures on water, sediment and agricul-tural chemical yields in large ungauged basins (Arnoldet al., 1996). It is comparatively simple, user friendly,physically based and distributed, which uses readilyavailable inputs. It is computationally efficient tooperate on large basins in a reasonable time. It is acontinuous time-scale model, capable of simulatinglong-term effects of management change.

The SWAT model uses a command structure forrouting runoff and chemicals through a watershedsimilar to the structure of problem-oriented lang-uage for hydrologic modelling (known as the HYMOmodel) (Williams & Hann, 1973). Specific commandsare there for routing flows through streams andreservoirs; adding flows and inputting measured dataor point sources. Using a routing command language,the model can simulate a basin subdivided intohydrological response units (HRU), grid cells or sub-watersheds.

The SWAT predicts surface runoff for daily rainfall byusing the Soil Conservation Service (SCS) curve number(CN) method (USDA-SCS, 1972). Sediment yield iscomputed for each subbasin with the Modified Uni-versal Soil Loss Equation (MUSLE) (Williams &Berndt, 1977). Subbasin nutrient yield and nutrientcycling were taken from the EPIC model (Williams et al.,1984) and modified as necessary for inclusion into theSWAT model (Arnold et al., 1996). The SWAT modelallows for simultaneous computations on each subbasinand routes the water, sediment, and nutrients from thesubbasin outlets to the basin outlet.

2.1. Nitrogen in subbasin

The amount of NO3-N contained in surface runoff isestimated for each subbasin by considering the first layer(10mm thickness) only. The total amount of waterleaving the layer is the sum of runoff, lateral subsurfaceflow, and percolation.

QT 5Q þ O1 þ QR1 ð3Þ

where: QT is the total water lost from the first layer inmm; Q is the runoff volume in mm; O1 is the percolationfrom the first layer in mm; and QR1 is the lateral flowfrom the first layer in mm.

Amounts of NO3-N contained in runoff, lateral flowand percolation are estimated as the products of thevolume of water lost and the average concentration ofNO3-N.

VNO35QT CNO3

ð4Þ

where: VNO3is the amount of NO3-N lost from the first

layer; and CNO3is the concentration of NO3-N in the

first layer.Leaching and lateral subsurface flow in lower layers

are treated with the same approach used in the upperlayer except that surface runoff is not considered.

A loading function developed by McElroy et al.,(1976) and modified by Williams and Hann (1978) forapplication to individual runoff events is used toestimate organic N transport by sediment. The loadingfunction estimates the daily organic N runoff loss basedon the concentration of organic N in the topsoil layer,the sediment yield and enrichment ratio. The loadingfunction

YON 5 0�001ðY ÞðCON ÞðERÞ ð5Þ

where: YON is the organic N runoff loss at the subbasinoutlet in kg ha�1; CON is the concentration of organic Nin the top soil layer in g t�1; Y is the sediment yield int ha�1; and ER is the enrichment ratio.

The value of CON is input to the model and is constantthroughout the simulation. Enrichment ratios arelogarithmically related to sediment concentration. Thelogarithmic equation estimating enrichment ratio is

ER 5 x1cx2a ð6Þ

where: ca is the sediment concentration in gm�3; and x1

and x2 are parameters set by the upper and lower limits.The enrichment ratio to approach 1�0, the sedimentconcentration would be extremely high. Conversely,a very low sediment concentration would causethe enrichment ratio to approach inverse of thesediment delivery ratio DR, which is the ratio ofsubbasin sediment yield and gross sheet erosion. Thesimultaneous solution of Eqn. (6) at the boundariesassuming sediment concentrations range from 500 to

ARTICLE IN PRESSIDENTIFICATION AND PRIORITISATION OF CRITICAL SUB-WATERSHEDS 367

250 000 gm�3 gives

x2 5log 1=DR

� �

2�699ð7Þ

x1 51

0�25ð Þx2ð8Þ

The technique used to estimate the denitrification rateis described in the user’s manual of the SWAT model(Arnold et al., 1996). The N mineralisation model is amodification of the PAPRAN, which is known as asimulation model of annual pasture production limitedby rainfall and nitrogen (Seligman & Keulen, 1981). Thedaily amount of immobilisation is computed bysubtracting the amount of N contained in the cropresidue from the amount assimilated by the microorgan-isms. Immobilisation may be limited by N availability. Ifthe amount of N available is less than the amount ofimmobilisation predicted, the decay rate constant isadjusted. To estimate the N contribution from rainfall,SWAT uses an average rainfall N concentration of8 ppm for all locations for all storms. The amount of Nin rainfall is estimated as the product of rainfall amountand concentration.

Crop use of N is estimated using a supply and demandapproach. The daily (day i) crop N demand can becomputed using

CNDi 5 ðCNBÞiBi � ðCNBÞi�1Bi�1 ð9Þ

where: CNDi is the N demand of the crop in kg ha�1; CNB

is the optimal N concentration of the crop; and B is theaccumulated N in kg ha�1.

The crop is allowed to take N from any soil layer thathas roots. Uptake starts at the upper layer and proceedsdownward until the daily demand is met or until all Nhas been depleted. If the soil cannot supply the dailyN demand for legumes, the deficit is attributed to Nfixation.

2.2. Phosphorus in subbasin

The SWAT approach is based on the concept ofpartitioning pesticides into the solution and sedimentphases (Knisel, 1980). As P is mostly associated with thesediment phase, the soluble P runoff equation can beexpressed in the simple form

YSP 50�01ðCLPPÞðQÞ

kd

ð10Þ

where: YSP is the soluble P in kg ha�1 lost in runoffvolume Q in mm; CLPP is the concentration of soluble Pin soil layer in g t�1; and kd is the P concentration inthe sediment divided by that of the water in m3 t�1.

The value of CLPP is input to the model and remainsconstant. The value of kd used in SWAT is 175.

Sediment transport of P is simulated with a loadingfunction. The P loading function is

YP 5 0�01ðY ÞðCPÞðERÞ ð11Þ

where: YP is the sediment phase of P loss in runoff inkg ha�1 and Cp is the concentration of P in the topsoillayer in g t�1.

The P immobilisation model, also developed by Joneset al. (1984) are similar in structure to the Nimmobilisation model. The daily amount of immobilisa-tion is computed by subtracting the amount of Pcontained in the crop residue from the amountassimilated by the microorganisms.

2.3. Channel sediment routing

The sediment routing model consists of two compo-nents operating simultaneously (deposition and degra-dation). Deposition in the stream channel is based onthe fall velocity of the sediment particles (Arnold et al.,1990). With a temperature of 228C and a sedimentdensity of 1�2 tm�3, Stokes’ Law for fall velocitybecomes

Vf 5 411 d2 ð12Þ

where: Vf is the fall velocity in mh�1 and d is thesediment particle diameter. The depth yf that sedimentof particle size d will fall during time, TT is

yf 5Vf TT ð13Þ

The sediment delivery ratio DR through the reach isestimated with the following equations.

DR 51 � 0�5yf

dq

; yf � dq ð14Þ

DR 50�5 dq

� �

yf

; yf > dq ð15Þ

where: dq is the depth of flow.Finally, deposition SD is calculated with the equation:

SD 5SIN ð1 � DRÞ ð16Þ

where SIN is the sediment entering the reach.Stream power is used to predict degradation in the

routing reaches. Williams (1980) used Bagnold’s (1977)definition of stream power to develop a method fordetermining degradation in channels. Bagnold definedstream power PS with the equation

PS 5 gqSw ð17Þ

where: g is the density of the water; q is the flow rate;and SW is the water surface slope. By applying streampower to bed load predictions (Bagnold, 1977) and

ARTICLE IN PRESSM. P. TRIPATHI ET AL.368

estimating model parameters (Williams, 1980), theequation for sediment reentrained, Dr, is

Dr 5 aspg15FdurwðdqSW VcÞ15 ð18Þ

where: asp is a parameter dependent on maximumstream power for the reach; Fdur flow duration in h;and Vc is the velocity in the channel.

The parameter asp can be estimated with the equation

asp 5 ðgwqScÞ�05mx ð19Þ

where: Sc is the slope of the channel and the subscriptmx refers to the maximum flow expected in the reach forextreme events. The value of q is assumed to equal somemaximum rainfall intensity (250mmh�1) and asp be-comes

asp 5 ð69�44gADScÞ�05 ð20Þ

where: AD is the drainage area into the reach in km2. Allof the stream power is used for reentrainment of looseand deposited material until all of the material has beenremoved. When this occurs, degradation of the bedmaterial DB begins and is calculated by

DB 5KCDr ð21Þ

where: K and C are MUSLE (Williams & Berndt, 1977)factors for the stream channel.

Total degradation DT is the sum of the reentrainmentand bed degradation components. This amount is alsoallowed to be redeposited before reaching the basinoutlet.

DT 5 ðDr þ DBÞð1 � DRÞ ð22Þ

Finally, the amount of sediment reaching the basinoutlet, SOUT, is

SOUT 5SIN � SD þ DT ð23Þ

where: SIN is the sediment entering the reach.

2.4. Nitrate and phosphorus routing

Once NO3-N enters a stream it is considered aconservative material for the duration of an individualrunoff event (Williams, 1980). Thus, NO3-N routing issimply a matter of adding the yields from all subbasinsto determine the basin yield.

The loading function approach is also used in routingorganic N from the subbasin outlets to the basin outlet:

ðYON ÞBj 5 001 YBð Þj YCONð Þj ERRð Þj ð24Þ

where: (YON )B is the organic N runoff loss at the basinoutlet in kg ha�1; YB is the sediment yield reaching thebasin outlet from subbasin j in t ha�1; YCON isthe concentration of organic N in the sediment reachingthe subbasin j outlet in g t�1; and ERR is the enrichment

ratio for the channel routing from subbasin j to thechannel outlet.

The delivery ratio for the channel routing is calculatedfrom

DR 5YSBð ÞjYBð Þj

ð25Þ

where: YSB is the sediment yield at the subbasin outlet int ha�1; and YB is that sediment yield from subbasin j

after it has been routed to the basin outlet in t ha�1.As with NO3-N routing, once soluble P enters a

stream it is considered a conservative material androuting is accomplished by adding the yields from allsubbasins to determine the basin yield.

Again, the loading function approach is used inrouting P from the subbasin outlets to the basin outlet.

ðYPÞBj 5 001 YBð Þj CPSBð Þj ERRð Þj ð26Þ

where: (YP)B is the P yield at the basin outlet in kg ha�1;and CPSB is the P concentration in the sediment reachingthe subbasin j outlet in g t�1.

3. Materials and methods

3.1. Study area and data collection

The selected Nagwan watershed (92�46 km2) is locatedin Upper Damoder Valley Corporation (DVC) in theHazaribagh district of Bihar, India. Location map of thestudy area is shown in Fig. 1. The watershed receives anaverage annual rainfall of 1256mm, out of which themonsoon season (June–October) contributes more than80% rainfall. Rainfall and runoff data for 7 years (1992–1998) from the gauging station at Nagwan sedimentobservation post were collected from DVC, Hazaribagh.IRS-1B (LISS II) satellite data with date of pass 19thOctober 1996 were collected and used for land use/landcover classification. Topographic maps (1:50 000) werecollected from Survey of India, Calcutta, and the soilresources data were collected from DVC, Hazaribagh,for use in the study.

The software available at Regional Remote SensingService Center (RRSSC), Kharagpur, such as theEnvironmental Analysis and Scientific Interface (EASI)and Picture Analysis, Correlation and Enhancement(PACE) commonly known as EASI-PACE (PCI Inc.,1994) for terrain analysis and image processing wereused. Extracted data were processed with the help ofExcel package of Microsoft Office 97 and all the inputdata file were generated in DOS using UTIL programmewhich was built-up with the SWAT model used in thestudy (Arnold et al., 1996, 1998).

ARTICLE IN PRESSIDENTIFICATION AND PRIORITISATION OF CRITICAL SUB-WATERSHEDS 369

3.2. Extraction of watershed parameters for the model

The digitised contour vectors were gridded using theEASI-PACE and converted to grided data. The gridswere interpolated using the k-nearest neighbours meth-od for generating the digital elevation model (DEM).The DEM of the watershed was prepared in 30m by30m resolution. Many researchers have also used DEMof 30m by 30m resolution and obtained satisfactoryresults (Bingner, 1996; Sharma et al., 1996; Tiwari et al.,1997; Wang & Hjelmfelt, 1998).

The watershed and sub-watershed boundaries, drai-nage networks and slope map were generated using theprocedure described by Jenson and Domingue (1988).The area delineated by the algorithm was 90�23 km2

against the manually judged area of 92�46 km2. Theautomatically delineated watershed was used forfurther study. Since SWAT works on sub-watershedsbasis, the delineated watershed was subdivided into 12sub-watersheds on the basis of topography usingprocedure similar to that used for delineation of mainwatershed (Fig. 2). In this study Nagwan watershed wascoded as WS and sub-watershed were coded as WS1 toWS12.

Supervised land use classification method was usedfor land use/land cover classification. The identified landuse classes were upland paddy (rice), low land paddy,

orchards, deep water, shallow water, closed forest, openforest, fallow land, grasses/shrubs, upland crops andsettlements. Soil textures and soil series maps were alsogenerated. There are mainly three soil series (Harina,

Bhuswa and Atia) in the watershed. The predominantsoil of the watershed is silt loam. Sandy loam, clay loam,

ARTICLE IN PRESS

WS1

WS2

WS3WS8

WS11

WS6

WS9

WS7

WS5

WS4

WS10

WS12

Outlet

N

Watershed boundary

Sub-watershed boundary

Drainage channel

Fig. 2. Geographic information system delineated sub-water-sheds

Nagwan Watershedin India

DELHIPATNA

DVCHAZARIBAGH

CALCUTTA

LEGEND:

O-Outlet/gauging stationNot to the scale

N

85.43°E24.12°N

85.43°E23.99°N

85.25°E23.99°N

Fig. 1. Location of Nagwan watershed in India

M. P. TRIPATHI ET AL.370

loam, loamy sand and silty clay loam are the other types(as per the USDA nomenclature) of soil found in thewatershed. Layerwise soil properties under different soilseries of the watershed are given in Table 1.

The weighted curve number for each sub-watershedwas calculated using land use/land cover map, soiltexture maps and standard curve numbers for the Indianconditions (Dhruva Narayana, 1993). Other inputparameters of the delineated sub-watersheds, such asoverland and channel slope, channel length and averageslope length were extracted using the various mapsincluding contour map, sub-watershed map, slope mapand drainage map (Table 2).

3.3. Model calibration and validation

The SWAT model has already been calibrated andvalidated by Tripathi et al. (1999a, 1999b) for Nagwanwatershed on monthly and daily basis for runoff andsediment yield for the year 1991 and 1992, respectively.The calibrated values for hydraulic conductivity ofalluvium for surface runoff and channel runoff were foundto be 6�4 and 1�0, respectively, and values of Manning’sroughness coefficient n for overland flow and channel flowwere found to be 0�065 and 0�040, respectively.

Since data for nutrient losses were available for fewevents during the monsoon season of the year 1997 only,

ARTICLE IN PRESS

Table 1Soil properties under different soil series of Nagwan watershed

SoilSeries

Depth,mm

Bulk,density,g/cm3

Availablewater

capacity

Organiccarbon,

%

Coursesand,

%Fine

sand, %Silt,%

Clay,%

Harina 150 1�55 0�14 0�47 40�1 38�8 11�7 7�9370 1�45 0�14 0�41 38�3 39�7 11�8 12�4690 1�31 0�12 0�21 12�6 28�8 19�1 39�6

1050 1�25 0�11 0�16 17�9 26�7 10�3 43�8Bhushwa 150 1�36 0�20 0�55 4�7 20�4 54�4 18�8

370 1�31 0�18 0�51 11�6 31�5 36�2 19�4690 1�33 0�14 0�25 4�8 28�1 25�3 40�5

1050 1�47 0�11 0�16 8�4 23�3 23�5 43�4Atia 150 1�31 0�20 0�54 5�4 33�3 35�7 23�6

370 1�31 0�18 0�53 5�8 29�0 32�8 30�9690 1�33 0�14 0�24 2�2 22�8 28�2 45�7

1050 1�40 0�14 0�21 1�9 23�5 27�4 46�1

Table 2

Sub-watershed wise input data for the SWAT model

Sub-watershed Area,km2

Slope,%

Curve number Av. slopelength, m

Channellength, km

Channelslope

Kfactor

Pfactor

Cfactor

1991 1996

WS1 17�19 2�2 73�1 83�6 464�3 9�60 0�005 0�28 0�60 1�0WS2 9�33 3�0 68�7 71�0 493�8 5�28 0�008 0�19 0�50 1�0WS3 6�27 2�1 76�9 79�7 481�6 1�80 0�001 0�22 0�60 1�0WS4 9�89 2�2 58�9 55�0 456�4 5�40 0�004 0�26 0�60 1�0WS5 14�67 2�1 66�8 68�9 395�8 6�00 0�005 0�21 0�60 1�0WS6 3�54 2�8 77�1 80�1 492�3 2�25 0�001 0�19 0�50 1�0WS7 9�46 3�1 71�5 69�0 517�0 5�76 0�005 0�24 0�50 1�0WS8 4�24 2�3 67�7 68�9 574�3 2�94 0�006 0�19 0�60 1�0WS9 3�10 2�9 73�1 74�7 437�8 2�25 0�008 0�23 0�50 1�0WS10 7�23 3�3 67�1 66�5 454�7 5�40 0�009 0�23 0�50 1�0WS11 4�80 2�9 77�7 79�2 479�4 3�36 0�009 0�17 0�50 1�0WS12 0�51 9�1 73�1 66�8 290�8 0�90 0�006 0�25 0�60 1�0WS* 90�23 2�3 71�0 72�0 461�7 13�86 0�005 0�21 0�60 1�0

Note: K, soil erodibility factor;P, soil conservation practice factor;C, crop management factor. *Whole Nagwan watershed.

IDENTIFICATION AND PRIORITISATION OF CRITICAL SUB-WATERSHEDS 371

the model was again tested for the year 1997 for dailyrunoff and sediment yield beside nutrient losses for thefew available events at the outlet of the Nagwanwatershed.

Since model was used to identify critical sub-water-sheds according to the annual sediment yield. Thereforeit was also validated on monthly basis for simulating thesurface runoff and sediment yield for monsoon seasonsof multiple years (1992–1998). The model performancewas evaluated on the basis of test criterion recom-mended by ASCE Task Committee (1993). Graphicaland statistical methods were also used for evaluating themodel performance. The numerical performance criteriaare briefly described below.

Martinec and Rango (1989) recommended that thecriteria should be as simple as possible. The percentdeviation of runoff volumes DV is one goodness-of-fitcriterion.

DV 5V � V 0

V100 ð27Þ

where: V is the measured yearly or seasonal runoffvolume and V 0 is the model computed yearly or seasonalrunoff volume. DV can take any value. However, smallerthe number better the model results. DV would equalzero for a perfect model.

The second basic goodness-of-fit criterion recom-mended by ASCE Task Committee (1993) is the Nash–Sutcliffe coefficient or coefficient of simulation efficiencyCOE (Nash & Sutcliffe, 1970):

COE 5 1 �

Pni5 1 Qi � Q0

i

� �2Pn

i 5 1 Qi � Qð Þ2ð28Þ

where: Qi is the measured daily discharge; Q0i is the

computed daily discharge; and Q is the averagemeasured discharge values. The values for COE can bevaried from 0 to 1, with 1 indicating a perfect fit. A valuefor COE equal to zero indicates that the model wassimulating no better than using the average of theobserved data. Martinec and Rango (1989) recom-mended using Q for the year or season to avoidunrealistically high values of COE in low runoff years.

Successful application of a calibrated hydrologicwatershed model depends on how well the model isverified. The parameters required for the model wereextracted from the analysis of DEM, soil map andsatellite imagery. Those parameters were then used inthe SWAT model. The observed daily rainfall andtemperature data for 1 year (1997) and multiple years(1992–1998) were used for the validation of the modelon daily and monthly basis, respectively. The model wasalso validated for the nutrient losses for the 12 events of1997. Since land use data for the year 1997 was notavailable, estimated curve number (CN) values of 1996

were given as input to the model. The soil series and soiltexture were different for different sub-watersheds.Therefore, weighted average of all the soil resource datawere given as input to the model.

3.4. Identification and prioritisation of critical

sub-watersheds

The verified model was applied for identifying andprioritising the critical sub-watersheds of the Nagwanwatershed. Since average rainfall data of 3 years (1996–1998) were significantly similar to the average rainfalldata of 13 years (1986–1998), simulations were per-formed using rainfall data of 3 years for identificationand prioritisation of the critical watersheds on thebasis of average annual sediment yield and nutrientlosses. The ranges of erosion rates and their classessuggested by Singh et al. (1992) were inferred foridentification of critical sub-watersheds. The criticalsub-watersheds were then prioritised and proposed forevaluating the management scenarios to reduce therunoff rate, sediment yield and nutrient losses from thewatershed.

A particular sub-watershed may get top priority dueto various reasons but often, the intensity of landdegradation is taken as the basis. This approach ofprioritising watersheds based on actual sediment yieldrates may be possible only when the number of sub-watersheds to be prioritised is less and necessary dataare available. Further, this method will be helpful whenthe sediment yield potentials of different sub-watershedsdo not have considerable variation.

The critical sub-watersheds were identified on thebasis of average annual sediment and nutrient lossesfrom the sub-watersheds during the period of 1996–1998. In this context, mean annual sediment yields weresimulated for each sub-watershed of Nagwan watershedusing SWAT model. Priorities were fixed on the basis ofranks assigned to each critical sub-watershed accordingto ranges of soil erosion classes described by Singh et al.(1992) (Table 3). Soil loss values of identified criticalsub-watersheds were compared with the average soil loss(16�35 t ha�1 yr�1) of India (Dhruva Narayana, 1993)and prescribed permissible upper limit of soil loss(11�2 t ha�1 yr�1) (Mannering, 1981) for justification ofresults. Also for nutrient losses a threshold value of10mg l�1 for nitrate nitrogen and 0�5mg l�1 for dis-solved phosphorous as described by EnvironmentalProtection Agency (EPA) were considered as criterionfor identifying the critical sub-watersheds (EPA, 1976).Identified critical sub-watersheds were arranged indescending order and then priorities were fixed for theirmanagement.

ARTICLE IN PRESSM. P. TRIPATHI ET AL.372

4. Results and discussion

4.1. Surface runoff

The graphical representation of validation results forthe daily runoff is shown in Fig. 3. The graphs show thatthe magnitude and temporal variation of simulatedrunoff matched closely with the observed runoff valuesfor the entire monsoon season of 1997. Timings ofoccurrence of the peaks for both observed and simulatedrunoff matched well. However, the model overpredictedrunoff for few rainfall events of high magnitude.

The descriptive statistics for both measured andsimulated daily runoff are given in Table 4. The meanvalues of observed and simulated runoff were notsignificantly different at 95% level of confidence, the t-calculated (0�53) being less than the t-critical (1�98).Similarity in means and standard deviation revealedsimilarity in the frequency distributions of observed andsimulated runoff. However, the highest peak runoffvalue predicted by the model was slightly higher thanthat of the observed value. The per cent deviation Dv

value indicated that the model was underpredictingrunoff by 4�6%, whereas a high value (0�87) of theNash–Sutcliffe simulation efficiency COE showed a closeagreement between the measured and simulated runoff.

Daily predicted runoff values for the monsoon seasonof 1997 were plotted against the measured values andtheir distribution along with the 1:1 line is shown inFig. 4. The results showed that the distribution ofobserved and simulated runoff was uniform throughoutthe season. Regression analysis between the observedand simulated runoff values resulted in a high value(0�91) of the coefficient of determination r2 indicating aclose relationship between measured and simulatedrunoff (Fig. 4, Table 4).

4.2. Sediment yield

The time series of observed and simulated dailysediment yield of the Nagwan watershed for the

validation period (June 1–October 31, 1997) werecompared graphically as shown in Fig. 5. The time topeaks of simulated sediment yield matched consistentlywell with the time to the measured peak sediment yieldthroughout the season. However, the model slightlyunder predicted few peak events of sediment yield.

Daily predicted sediment yields were plotted againstthe measured values and their distribution along withthe 1:1 line as shown in Fig. 6. The simulated sedimentyields were distributed uniformly about the 1:1 line forboth lower and higher values of observed sediment yield.Regression analysis was also performed between theobserved and simulated sediment yield values and thebest-fit line is shown in Fig. 6. A value 0�89 of r2

indicates a close relationship between the measured andsimulated sediment yields. The Nash–Sutcliffe simula-tion efficiency of 0�89 indicated that there was goodagreement between the observed and simulated sedimentyields during the validation period of 1997. The overalldeviation between the simulated and observed sedimentyields was found to be 14�3%.

The descriptive statistics for both measured andpredicted daily sediment yields are given in Table 4. Aclose agreement between means and standard deviationof measured and simulated sediment yields indicates thatthe frequency distributions for the occurrence of sedi-ment yields were similar. Comparison of means usingStudent’s t-test (t-calculated value of 0�06 and t-criticalvalue of 1�98) revealed that the mean values of observedand simulated sediment yields were not significantlydifferent at 95% confidence level. However, the max-imum sediment yield predicted by the model was slightlylower than the observed maximum sediment yield.

4.3. Monthly distribution of runoff and sediment yield

The results of measured and simulated monthly runofffor the monsoon seasons of 1992–1998 were compared

ARTICLE IN PRESS

Table 3Area under different classes of soil erosion by water in India

Serial no.Soil erosion

classesSoil erosion

range, t ha�1 yr�1 Area, km2

1 Slight 0–5 801 3502 Moderate 5–10 1 405 6403 High 10–20 805 0304 Very high 20–40 160 0505 Severe 40–80 83 3006 Very severe >80 31 895

0

25

50

75

100

1-Jun 1-Jul 1-Aug 1-Sep 1-OctTime, days

Run

off,

mm

Rai

nfal

l, m

m

0

50

100

150

200

250

Fig. 3. Observed and simulated runoff hydrograph for modelvalidation (June to October, 1997); ( ), rainfall; ( ),

observed runoff; ( ), simulated runoff

IDENTIFICATION AND PRIORITISATION OF CRITICAL SUB-WATERSHEDS 373

and the results are given in Table 5. Results showed thatthe model was underpredicting the monthly runoffduring monsoon season by 6�0%. Though there wassignificant difference between the observed (60�63mm)and simulated (56�96mm) means of monthly runoffbecause t-calculated (2�39) was higher than t-critical(2�03) at 95% level of confidence, a high value (0�97) ofcoefficient of determination r2 showed close relationshipbetween observed and simulated monthly surface runofffor the validation period.

The results of measured and simulated monthlysediment yield for the monsoon season of 1992–1998were also compared (Table 5). Results showed that mostof the time the model predicted sediment yield was closeto the observed sediment yield during each month of

ARTICLE IN PRESS

Table 4Statistical analysis of the observed and simulated daily surface runoff and sediment yield for monsoon season (June–October) of the

year 1997

Statistics Runoff, mm Sediment yield, t ha�1

Observed Simulated Observed Simulated

Mean 2�80 2�68 0�026 0�022Standard deviation 7�06 8�16 0�072 0�067Maximum peak 50�33 57�00 0�538 0�460Total 429�07 409�40 3�885 3�330Count 153 153 153 153t-calculated 0�53 0�06t-critical (two tailed) 1�98 1�98Coefficient of determination r2 0�91 0�89Deviation, % 4�6 14�3Coefficient of simulation efficiency COE 0�87 0�89

0.0

0.2

0.4

0.6

0.8

1.0

1-Jun 1-Jul 1-Aug 1-Sep 1-Oct

Time, days

Sedi

men

t yie

ld, t

ha−1

0

50

100

150

200

250

Rai

nfal

l, m

m

Fig. 5. Observed and simulated sediment yield for modelvalidation (June to October, 1997); ( ), rainfall; ( ),

observed sediment yield; ( ), simulated sediment yield

0

10

20

30

40

50

60

70

0 10 20 30 40 50 60 70Observed runoff, mm

Sim

ulat

ed r

unof

f, m

m

Fig. 4. Comparison between observed and simulated daily runofffor model validation (June to October, 1997); ( ),

regression line; ( ), 1:1 line

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.0 0.1 0.2 0.3 0.4 0.5 0.6

Observed sediment yield, t ha−1

Sim

ulat

ed s

edim

ent y

ield

, t h

a−1

Fig. 6. Comparison between observed and simulated dailysediment yield for model validation (June to October, 1997);

( ), regression line; ( ), 1:1 line

M. P. TRIPATHI ET AL.374

monsoon season (r2 5 0�89). In general, the simulatedmonthly sediment yield compared well with measuredvalues during monsoon season of the validation yearsbecause means were similar at 95% confidence level(Table 5). The results also showed that the model wasoverpredicting monthly sediment yield by 4�0%.

Based on the above results, it could be inferred thatthe model was accurately validated for predicting dailyas well as monthly surface runoff and sediment yieldfrom the Nagwan watershed.

4.4. Nutrient losses

Data for the nutrient losses were not available for theyear 1991. Hence the prediction capability of nutrientlosses of the model could not be calibrated (Tripathi,1999). Therefore, the model was evaluated by perform-ing validation only for the year 1997 with the availablenutrient loss data of 12 events in this study. The

nutrients considered for validation were organic nitro-gen, phosphorus, nitrate nitrogen (NO3-N) and solublephosphorus.

The graphical comparison between observed andsimulated nutrient losses is shown in Figs 7–10 anddescriptive statistics is given in Table 6. It was seen thatthe observed and simulated values of nutrients includingorganic nitrogen, phosphorus, NO3-N and soluble Pwere uniformly distributed about 1:1 line (Figs 7–10).Observed and simulated means of organic nitrogen,phosphorus, NO3-N and soluble P were not significantlydifferent at 95% level of confidence, since t-calculatedwere found to be less than t-critical in all the cases(Table 6). The Dv values were found to be 15�8, 11�7, 3�7and 12�5%, respectively, for organic N, P, NO3-N andsoluble P indicating that the model was predictingnutrient losses satisfactorily. For organic N, P, NO3-Nand soluble P the r2 values were 0�82, 0�86, 0�89 and 0�82,respectively,which is a good agreement between ob-served and simulated values of nutrient losses.

ARTICLE IN PRESS

Table 5Results of statistical analysis for observed and simulated monthly surface runoff and sediment yield (1992–1998)

Statistics Surface runoff, mm Sediment yield, t ha�1

Observed Simulated Observed Simulated

Mean 60�63 56�96 0�77 0�80

t-calculated 2�39 �0�52t-critical (two tailed) 2�03 2�03Coefficient of determination r2 0�97 0�89Deviation, % 6�0 �4�0Coefficient of simulation efficiency COE 0�98 0�79

0

0.2

0.4

0.6

0.8

0.0 0.2 0.4 0.6 0.8

Observed organic N, kg ha−1

Sim

ulat

ed o

rgan

ic N

, kg

ha−1

Fig. 7. Comparison of observed and simulated organic nitrogenfor model validation; ( ), regression line; ( ), 1:1 line

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.00 0.05 0.10 0.15 0.20 0.25 0.30

Observed nitrate, kg ha−1

Sim

ulat

ed n

itrat

e, k

g ha

−1

Fig. 8. Comparison of observed and simulated nitrate nitrogenfor model validation; ( ), regression line; ( ), 1:1 line

IDENTIFICATION AND PRIORITISATION OF CRITICAL SUB-WATERSHEDS 375

On the basis of above validation results for thenutrient losses, it was confirmed that model waspredicting reasonably well and can be used for manage-ment and planning of the Nagwan watershed for furtheragricultural development.

4.5. Identification and prioritisation of critical

sub-watersheds

Identification and prioritisation of critical sub-water-sheds based on actual sediment yield rate may be

possible only when sediment data are available.Mean annual sediment and nutrient losses weresimulated for each sub-watershed of Nagwan watershedusing SWAT model (Table 7) and priorities werefixed as described earlier. Results showed that out ofthe 12 sub-watersheds, the WS5 fell under moderate soilloss group of soil erosion classes (5–10 t ha�1 yr�1). TheWS6, WS7, WS9, WS10 and WS12 fell under high soilloss group of soil erosion classes (10–20 t ha�1 yr�1),whereas other sub-watersheds fell under slight erosionclasses.

None of the sub-watersheds fell under very high,severe or very severe erosion classes. This may bebecause of the fact that the study watershed is having anaverage slope of 2�3% only. The study watershed mighthave got stabilised as contour and graded bunds andterraces already exist in the watershed. However, thesub-watershed WS12 resulted in maximum sedimentyield, which is also more than average soil loss16�35 t ha�1 yr�1 of India (Dhruva Narayana, 1993).This may be due to the high average surface slope of9�1% with undulating topography. Sub-watershedsWS6, WS7, WS9 and WS10 exceeded the prescribedpermissible upper limit of soil loss of 11�2 t ha�1 yr�1

(Mannering, 1981).The results indicated that the dissolved nutrient

losses including NO3-N and soluble P were within thepermissible limit of 10 and 0�5mg l�1, respectively (EPA,1976; Tim et al., 1992). The trend showed that losses ofnutrient attached with the sediment were proportionalto losses of sediment from the watershed (Table 7).Annual losses of nutrient attached with sediment werefound to be more in case of WS12 followed by WS9,WS7, WS10 and WS6. The nitrate nitrogen and solublephosphorous in runoff were not high as compared toother sub-watersheds.

On the basis of annual sediment yield and nutrientlosses, sub-watersheds WS6, WS7, WS9, WS10 andWS12 were found to be critical. After arranging thecritical sub-watersheds in ascending order, consideringthe annual sediment yield and nutrient losses from eachsub-watershed, priorities were fixed. The sub-watershedthat comes first is given the top priority for developingthe management plan to reduce the soil and nutrientlosses.

As a result the critical sub-watersheds WS12, WS9,WS7, WS10 and WS6 were selected and recommendedto adopt the management measures in that order toreduce the sediment and nutrient losses and to conservethe rainwater within the watershed for sustainable cropproduction. Other sub-watersheds were not consideredfor management practices because those sub-watershedswere not yielding sediment more than moderate or slighterosion classes.

ARTICLE IN PRESS

0.0

0.1

0.2

0.3

0.4

0.5

0.0 0.1 0.2 0.3 0.4 0.5

Observed P, kg ha−1

Sim

ulat

ed P

, kg

ha−1

Fig. 9. Comparison of observed and simulated organic P formodel validation: ( ), regression line; ( ), 1:1 line

0

0.01

0.02

0.03

0.04

0.05

0.00 0.01 0.02 0.03 0.04 0.05

Observed soluble P, kg ha−1

Sim

ulat

ed s

olub

le P

, kg

ha−1

Fig. 10. Comparision of observed and simulated soluble P formodel validation; ( ), regression line; ( ), 1:1 line

M. P. TRIPATHI ET AL.376

5. Conclusions

The study confirmed that the Soil and WaterAssessment Tool (SWAT) model could accuratelysimulate runoff, sediment yield and nutrient lossesparticularly from small agricultural watersheds. Thesimulated data closely matched with their observedcounterparts in the study. The study also revealed thatall the sub-watersheds of a small agricultural watersheddo not contribute to the discharge, sediment yield andnutrient losses measured at the outlet. The SWAT modelcould identify the critical sub-watersheds, which aremajor contributors of these parameters. The SWATmodel can successfully be used for prioritisation of thecritical sub-watersheds in order to develop multi-yearmanagement plan to reduce the runoff, sediment andnutrient losses from a small agricultural watershed.

Acknowledgements

Authors wish to acknowledge the CSIR, New Delhi,for providing financial assistance to conduct this study.

Er. Kamal Misra, Director (Soil Conservation) DVC,Hazaribagh, Bihar, and Project Co-ordinator, IGBP‘‘Watershed Management’’, New Delhi, are also ac-knowledged by the authors for providing the data toconduct the above study. The facilities and supportprovided by the Department of Agricultural and FoodEngineering, IIT, Kharagpur and RRSSC, IIT Campusare sincerely acknowledged.

References

Arnold J G; Srinivasan R; Muttiah R S; Williams J R (1998).Large area hydrologic modeling and assessment part I:model development. Journal of the American WaterResources Association, 34(1), 73–89

Arnold J G; Srinivasan R; Ramanarayanan T S; DiLuzio M

(1999). Water Resources of the Texas Gulf Basin. WaterScience and Technology, 39(3),121–133

Arnold J G; Williams J R; Nicks A D; Sammons N B (1990).SWRRB: A Basin Scale Simulation Model for Soil andWater Resources Management, pp 125. Texas A & MUniversity Press, College Station, TX

Arnold J G; Williams J R; Srinivasan R; King K W (1996). Soiland Water Assessment Tool, User’s Manual USDA,

ARTICLE IN PRESS

Table 7Model output for identification of the critical sub-watersheds of the Nagwan watershed (1996–1998)

Sub-watershedArea,km2

Runoff,mm

Sedimentyield, t ha�1

OrganicN, t ha�1

OrganicP, t ha�1 NO3-N, t ha�1 Soluble P, t ha�1

WS1 17�23 467�68 4�41 5�00 2�38 2�71 0�27WS2 9�29 274�78 3�66 4�15 1�99 1�42 0�17WS3 6�32 422�14 4�39 5�03 2�39 2�29 0�25WS4 9�93 143�59 1�68 1�96 0�94 0�75 0�09WS5 14�71 284�49 7�00 7�50 3�58 1�15 0�18WS6 3�52 456�13 12�87 13�26 6�24 2�04 0�28WS7 9�47 292�96 13�47 13�85 6�56 1�19 0�18WS8 4�24 331�56 3�33 3�84 1�84 1�74 0�20WS9 2�98 373�98 14�63 14�86 6�99 1�59 0�23WS10 7�22 262�27 12�80 13�23 6�27 1�05 0�17WS11 4�78 416�65 4�67 5�30 2�53 2�25 0�25WS12 0�54 256�34 18�82 18�40 8�79 1�26 0�16

Table 6Statistical analysis of the observed and simulated nutrient losses for the 12 events of the year 1997

Statistics Organic N Phosphorous NO3 –N Soluble P

Obs. Sim. Obs. Sim. Obs. Sim. Obs. Sim.

Mean, kg ha�1 0�326 0�275 0�168 0�148 0�100 0�096 0�017 0�015Standard deviation, kg ha�1 0�233 0�255 0�145 0�135 0�071 0�085 0�015 0�014Maximum, kg ha�1 0�686 0�750 0�440 0�400 0�220 0�240 0�047 0�040Total, kg ha�1 3�917 3�300 2�015 1�780 1�194 1�150 0�206 0�180Count 12 12 12 12 12 12 12 12t-calculated 1�66 1�47 0�39 1�15t-critical (two tailed) 2�20 2�20 2�20 2�20Coefficient of determination r2 0�82 0�86 0�89 0�82Deviation, % 15�8 11�7 3�7 12�5

Note: Obs., observed; Sim., simulated.

IDENTIFICATION AND PRIORITISATION OF CRITICAL SUB-WATERSHEDS 377

Agriculture Research Service, Grassland, Soil and WaterResearch Laboratory, 808 East Blackland Road Temple,TX 76502

ASCE Task Committee (1993). Criteria for Evaluation ofWatershed Models. Journal of Irrigation and DrainageEngineering, 119(3), 429–442

Bagnold R A (1977). Bedload transport in natural rivers. WaterResources Research, 13(2), 303–312

Bingner R L (1996). Runoff simulation from Goodwin Creekwatershed using SWAT. Transactions of the ASAE, 39(1),85–90

Bingner R L; Garbrecht J; Arnold J G; Srinivasan R

(1997). Effect of watershed division on simulation of runoffand fine sediment yield. Transactions of the ASAE, 40(5),1329–1335

DelRegno K J; Atkinson S F (1988). Nonpoint pollution andwatershed management: a remote sensing and geographicinformation system approach. Lake and Reservoir Manage-ment, 4(2), 17–25

Dhruva Narayana V V (1993). Soil and Water ConservationResearch in India, pp 146–151. Indian Council of Agricul-tural Research, Krishi Anusandhan Bhavan, Pusa, NewDelhi

Dickinson W T; Rudra R P; Wall G J (1990). Targetingremedial measures to control nonpoint source pollution.Water Resource Bulletin, AWRA, 26(3), 499–507

Dillaha T A (1990). Role of best management practices inrestoring the health of the Chesapeake Bay: 1990-advancesin estuarine sciences. Report No. CBP/TRS4/90, USEnvironmental Protection Agency, Chesapeake Bay Pro-gram/Chesapeake Res. Consortium, Gloucester Point, VApp 57–81

EPA (1976). Quality Criteria for Water. EnvironmentalProtection Agency, Washington, DC

Hession C W; Shanholtz V O (1988). A geographic informationsystem for targeting non-point source agricultural pollution.Journal of Soil and Water Conservation, 43(3), 264–266

Jenson S K; Domingue J O (1988). Extracting topographicstructure from digital elevation data for geographic infor-mation system analysis. Photogrammetric Engineering andRemote Sensing, 54(11), 1593–1600

Jones C A; Cole C V; Sharpley A N; Williams J R (1984). Asimplified soil and plant phosphorus model, I. Documenta-tion. Soil Science Society of the American Journal, 48(4),800–805

Karale R L; Bali Y P; Narula K K (1977). Priority watershedfor soil conservation works in Matatila catchment. Journalof Indian Society of Soil Science, 25, 207

Karale R L; Bali Y P; Singh C P (1975). Photo interpretationfor erosion assessment in the Beas catchment. PhotoNirvachak, 4, 30

Knisel W G (1980). CREAMS: a field scale model forchemicals, runoff, and erosion from agricultural manage-ment systems. Conservation Research Report No. 26,USDA-SEA, Washington, DC, pp 643

Maas R P; Smolen M D; Dressing S A (1985). Selecting criticalareas for nonpoint source pollution control. Journal of Soiland Water Conservation, 40(1), 68–71

Mannering J V (1981). The Use of Soil Tolerances as Strategyfor Soil Conservation. Soil Conservation Problem andProspect, pp 337–349. R. P. C. Morgan John Wiley &Sons, Chichester, England

Martinec J; Rango A (1989). Merits of statistical criteria forthe performance of hydrological models. Water Resources

Bulletin, American Water Resource Association, 25(20),421–432

McElroy A D; Chiu S Y; Nebgen J W, et al. (1976). Loadingfunctions for assessment of water pollution from nonpointsources. Environment Protection Technology Service, EPA600/2–76–151

McHarg I L (1969). Design with Nature. Natural HistoryPress, New York, NY

Misra N (1986). Modelling runoff and sediment yield fromsmall watersheds. PhD Thesis, Agricultural EngineeringDepartment, Indian Institute of Technology, Kharagpur,India, pp 55–56

Nash J E; Sutcliffe J V (1970). River flow forecasting throughconceptual models, part 1}a discussion of principles.Journal of Hydrology, 10(3), 282–290

PCI Inc. (1994). EASI-PACE Image Analysis System Manual,Version 5�3. PCI Incorporated, Richmond Hill, Ontario,Canada; Vols. I & II, variously paged

Peterson J R; Hamlett J M (1998). Hydrological calibration ofthe SWAT model in a watershed containing fragipan soils.Journal of the American Water Resources Association,34(3), 531–544

Rosenthal W D; Srinivasan R; Arnold J G (1995). Alternativeriver management using a linked GIS-hydrology model.Transactions of the ASAE, 38(3), 783–790

Santhi C; Arnold J G; Wliiiams J R; Dugas W A; Srinivasan R;

Hauck L M (2001). Validation of the SWAT model on alarge river basin with point and nonpoint sources. Journal ofAmerican Water Resources Association, 37(5), 1169–1188

Seligman N G; Keulen H V (1981). PAPRAN-A simulationmodel of annual pasture production limited by rainfall andnitrogen. In: Simulation of Nitrogen Behaviour of Soil-PlantSystems, (M. J. Frissel; J. A. VanVeen, eds) pp 192–221.Proceeding of the Workshop, Wageningen, The Nether-lands, January–February 1980

Sharma K D; Menenti M; Huygen J; Vich A (1996). Modelingspatial sediment delivery in an arid region using thematicmapper data and GIS. Transactions of the ASAE, 39(2),551–557

Singh G; Ram Babu; Pratap Narain; Bhushan L S; Abrol I P

(1992). Soil erosion rate in India. Journal of Soil and WaterConservation, 47(1), 97–99

Srinivasan R; Arnold J G (1994). Integration of a basin scalewater quality model with GIS. Water Resources Bulletin,AWRA, 30(3), 453–462

Srinivasan R; Arnold J G; Rosenthal W; Muttiah R S (1993).Hydrologic modelling of Texas Gulf Basin using GIS. In:Proceedings of Second International GIS and Environmen-tal Modelling, Breckinridge, CO pp 213–217

Storm D E; Dillaha T A; Mostaghimi S; Shanholtz V O (1988).Modeling phosphorous transport in surface runoff. Trans-actions of the ASAE, 31, 117–127

Tim U S; Moostaghimi S; Shanholtz V O (1992). Identificationof critical nonpoint pollution source areas using geographicinformation systems and water quality modeling. WaterResources Bulletin, AWRA, 28(5), 877–887

Tiwari K N; Kanan N; Singh R; Ghosh S K (1997). Watershedparameters extraction using GIS and remote sensing forhydrologic modelling. ASIAN-PACIFIC Remote Sensingand GIS Journal, 10(1), 43–52

Tripathi M P (1999). Hydrological modelling for effectivemanagement of a small watershed. PhD Thesis, Agriculturaland Food Engineering Department, Indian Institute ofTechnology, Kharagpur, India, variously paged

ARTICLE IN PRESSM. P. TRIPATHI ET AL.378

Tripathi M P; Panda R K; Raghuwanshi N S (1999a). Runoffestimation from a small watershed using SWAT model.Hydrological Modelling, Proceeding of International Con-ference on Water, Environment, Ecology, Socio-Economicsand Health Engineering, held at Seoul, Korea from October18 to 21, 1999, pp 143–152

Tripathi M P; Panda R K; Raghuwanshi N S (1999b).Estimation of Sediment Yield from a small Water-shed Using SWAT Model. In: Proceeding of Civiland Environmental Engineering Conference NewFrontiers and Challenges, held at Asian Institute ofTechnology, Bangkok, Thailand from November 8 to 12,pp I-87–I-96

USDA-SCS (1972). National Engineering Handbook, Hydrol-ogy Section 4, Chapters 4–10

Vieux B E (1991). Geographic information system andnonpoint source water quality and quantity modelling.Hydrological Processes, 5(1), 101–113

Wang M; Hjelmfelt A T (1998). DEM based overland flowrouting. Journal of Hydrologic Engineering, 3(1), 1–8

Williams J R (1980). SPNM, A model for predicting sediment,phosphorous and nitrogen yields from agricultural basins.Water Resources Bulletin, AWRA, 16(5), 843–848

Williams, J R; Berndt H D (1977). Sediment yield predictionbased on watershed hydrology, Transactions of the ASAE,20(6), 1100–1104

Williams J R; Hann R W (1973). HYMO: Problem-orientedLanguage for Hydrologic Modelling}User’s Manual.USDA, ARS-S-9

Williams J R; Hann R W (1978). Optimal operation of largeagricultural watersheds with water quality constraints.Texas Water Resource Institute, Texas A & M University,Technical Report No. 96

Williams J R; Jones C A; Dyke P T (1984). A modelingapproach to determining the relationship between erosion andsoil productivity. Transactions of the ASAE, 27(1), 129–144

ARTICLE IN PRESSIDENTIFICATION AND PRIORITISATION OF CRITICAL SUB-WATERSHEDS 379