Geospatial estimation of soil moisture in rain-fed paddy fields using SCS-CN-based model

Geospatial estimation of soil moisture in rain-fed paddy fields using SCS-CN-based model T.V. Reshmidevi a, R. Jana b, *, T.I. Eldho a a Department of Civil Engineering, Indian Institute of Technology Bombay, Powai, Mumbai 400076, India b Centre of Studies in Resources Engineering, Indian Institute of Technology Bombay, Mumbai 400076, India abstract Keywords: Soil moisture simulation SCS-CN model Paddy fields Drought Crop water requirement Paddy fields are characterized by standing water and saturation condition during the entire crop growth period. However, in sub-humid and semi-arid areas, scarce rainfall and intermittent dry spells often cause soil moisture depletion resulting in unsaturated condition in the fields. These distinctive characteristics of the paddy fields have significant influence on the runoff generation and soil moisture retention characteristics of the watershed. In this study, the objective is to extend the application of the Soil Conservation Services Curve Number (SCS-CN)-based models for the geospatial and temporal simulation of soil moisture to paddy field-dominated agricultural watersheds in the water scarce areas. Different SCS-CN-based models, integrated with the soil moisture balance equation, are used to estimate the surface runoff and soil moisture content wherein, the spatial variation in the soil hydraulic characteristics is used to calculate the geospatial variation in soil moisture content. Physical significance of the terms initial abstraction (I a ) and potential maximum retention (S) in these models and their influence on the estimation of runoff and soil moisture are analysed in detail. A new SCS-CN-based model for soil moisture simulation (SCS-CN-SMS), to improve the soil moisture estimation, is proposed in this paper. The proposed model is built up on the soil moisture balance equation to account for the effect of ponding condition and soil moisture variation between the dry and saturation condition. The method is tested with 3 years observed surface runoff data and crop production statistics from a part of the Gandeshwari sub-watershed in West Bengal, India. The entire study area is divided into cells of 20 m 20 m. Various components of the soil moisture balance equation are estimated for each cell as a function of the soil moisture content. Remote Sensing Technique and Geographic Information System (GIS) are used to extract and integrate the spatially distributed land use and soil characteristics. The Hortonion overland flow concept adopted in the SCS-CN method is used to estimate the soil hydraulic characteristics of each cell in which the curve number is used to infer the spatial variation of the land use and soil characteristics. Even though the original SCS-CN method and the existing modified versions are efficient for runoff estimation, these models are found to be inappropriate for the estimation of soil moisture distribution. On the other hand, the proposed SCS-CN-SMS model gives better results for both runoff and soil moisture simulation and is, therefore, more suitable for the hydrological modeling of paddy field-dominated agricultural watersheds.

448 1. Introduction In semi-arid and sub-humid regions, wherever rainfed agriculture is practiced, agricultural drought is a common problem. Assessment of agricultural drought consists of two inevitable components: simulation of soil moisture available in the crop root zone and the estimation of accumulated runoff available in the watershed. Simulation of spatial and temporal variation of soil moisture helps to identify the areas and periods of significant water stress. On the other hand, the surface runoff estimation gives an idea about the water harvesting potential of the watershed and hence the scope for supplementary irrigation. The Natural Resources Conservation Services (previously known as the Soil Conservation Services) Curve Number (SCS-CN) method is widely used for the estimation of runoff from small agricultural watersheds (SCS, 1971). The method was originally developed for the estimation of only surface runoff and hence needs modification to extend it for the simulation of soil moisture. The original SCS-CN method is based on the concept of Hortonian overland flow in which the surface runoff is assumed to occur when the rainfall rate exceeds the infiltration rate. The method has limitations as it depends on the past 5-days rainfall for the estimation of antecedent moisture conditions (AMC) and hence for the estimation of runoff. A modified form of this model was used in the Soil and Water Assessment Tool (SWAT) and ArcView SWAT (AVSWAT) by introducing the soil water holding characteristics to define the runoff generating capacity of any particular area (Arnold et al., 2000; DiLuzio et al., 2002). Mishra and Singh (2003), Mishra et al. (2004, 2005) and Michel et al. (2005) modified the original SCS-CN model by redefining the model equation considering the physical concepts of the watershed hydrology and also by relating the runoff generating capacity to the current soil moisture content. However, these models failed to produce satisfactory results when applied to paddy field-dominated watersheds, where the generated surface runoff is stored in the fields to maintain saturation condition. In paddy fields, field bunds are used to arrest the runoff during the entire crop growth period. Such areas will contribute to stream flow only after exceeding the maximum ponding depth maintained in the field using the field bunds. Therefore, in order to simulate the hydrological processes taking place in the paddy fields, Panigrahi et al. (2001) and Agrawal et al. (2004) used field scale models. These models were based on the soil moisture mass balance in the crop root zone in which various empirical equations were used to estimate different components of the mass balance. However, in order to assess the water harvesting potential of any area, it is required to estimate the accumulated runoff available at the watershed outlet and hence watershed scale models are necessary. Only a few studies are reported on the modeling of agricultural watersheds dominated by paddy fields. Portoghese et al. (2005) presented a water balance model coupled with Geographic Information System (GIS) for a paddy field-dominated watershed. Kang et al. (2006) modified the AVSWAT model by adding a paddy field component. The model was based on the water balance in the surface ponding zone. Saturation condition was considered throughout the crop growth period and a constant infiltration rate was assumed based on the soil texture. Soil hydraulic conductivity and soil water holding characteristics viz., saturation soil moisture content (SAT), field capacity (FC), wilting point (WP) were the required input for the model. However, spatial variation in topography, land use, soil characteristics (soil texture, depth, and slope), and atmospheric factors cause variation in the soil water holding characteristics, agricultural pattern of the area, and runoff generated from the watershed. In the present study, an attempt is made to extend the application of the original SCS-CN method and its modified versions for the simulation of soil moisture in the crop root zone. A new soil moisture model, which is applicable for paddy field-dominated agricultural watersheds, is proposed in this paper. The proposed soil moisture model is integrated with the SCS-CN-based methods for the estimation of runoff and spatial variation in the soil hydraulic characteristics with changes in land use and soil characteristics. Further, with the objective of improving the performance of the model, the existing SCS-CN-based models are modified by fine tuning of their participating components as well as by the modification of the physical equations. These models are applied to an agricultural watershed in India. Performance of these models, after integrating with the soil moisture model, is evaluated and validated for both runoff and soil moisture simulations. 2. Methodology 2.1. Soil moisture model with paddy component A model based on the soil moisture mass balance in the root zone is used in the present study to simulate the temporal variation in soil moisture (M) considering a daily time step (t). Rainfall (P), supplementary irrigation (IR s ), and horizontal seepage inflow (SP i ) are considered as the inflow to the system, whereas runoff (Q), evaporation and actual crop evapotranspiration (ET a ), seepage outflow (SP o ), and deep percolation (DP) are considered as the losses from the system. Various components of the model are shown in Fig. 1. Rainfall and runoff are assumed to occur at the beginning of the time step. ET a,sp i,sp o, DP, R and IR s, if any, are considered at the end of the time period. Q is estimated using SCS-CN-based models and is explained in the next section. The initial abstraction (I a ) term is also included in the soil moisture balance equation. Initial abstraction includes the part of rainfall, which accounts for interception, surface storage, evaporation and initial infiltration before the runoff begins. In order to incorporate the temporary effect of I a on soil moisture, a term I af is introduced here, which is the fraction of I a that reaches the soil. Runoff generated from the paddy fields will appear at the watershed outlet either during the non-ponding periods or when the maximum ponding depth (PD max ) is exceeded. During the ponding period, the contact time between the water and soil increases and more water infiltrates into the soil, increasing the soil moisture content to saturation. This excess infiltration from the standing water is termed as recharge (R). During the ponding period saturated overland condition is assumed and such areas will contribute to the discharge only after satisfying PD max. The soil moisture

449 Fig. 1 Schematic diagram showing the soil moisture balance model. balance can be represented by the Eq. (1), where the suffixes t and t 1 are used to represent the time period. M ðtþ ¼ M ðt 1Þ þ½p ðtþ þ IR sðt 1Þ þ SP iðtþ Š ½Q ðtþ þði a I af Þ ðtþ þ ET aðtþ þ SP oðtþ þ DP ðtþ ŠþR ðtþ (1) Eq. (1) is developed to capture the soil moisture deficit condition prevailing in the field. Therefore, it gives the soil moisture condition prior to the irrigation on the t th day. Evaporation and transpiration losses are determined using the Food and Agricultural Organization (FAO)-56 Penman-Monteith method, selecting the crop coefficients for different vegetation types during different growth stages (Allen et al., 1998). In addition to this, direct evaporation from the water surface is also considered when standing water is available in the field. Evaporation from the standing water is assumed to vary with respect to the vegetative cover over the area. It is assumed to range from the free water evaporation rate (1.05 times the reference evapotranspiration as recommended in FAO-56) to zero, as the vegetation cover increases. An empirical relation in terms of reference evapotranspiration (ET o ) and crop coefficient (K c ) is used to represent this variation, which is shown in Eq. (2). The equation is derived assuming maximum vegetation cover corresponding to the K c value of 1.25, which is the highest value given in FAO-56 (Allen et al., 1998). 1:05 ET o ; if K c 1 ½1:05 4:2ðK c 1ÞŠET o ; Free water evaporation ¼ if 1 < K c 1:25 0; if K c > 1:25 Seepage and percolation losses are determined based on the soil moisture content and the hydraulic conductivity using Darcy s equation (Stephens, 2000). Supplementary irrigation may be provided when the soil moisture reaches the critical moisture content, which is the minimum allowable soil moisture content below which the plants undergo critical (2) moisture stress, reducing the yield. All the components of the model are varied with the moisture content available in the soil. 2.2. Estimation of runoff using SCS-CN-based models The soil moisture model [Eq. (1)] is integrated with the original SCS-CN model and its variants and these models are presented here as Models 1 6, explaining the physical concept and their applicability for geospatial soil moisture modeling. The model equations, and the similarities and differences between these models are presented in Table 1. When integrated with the soil moisture balance equation, these models consider the effect of ponding and its duration in the paddy fields while estimating runoff. In Model 1, the original SCS-CN model is used for the estimation of runoff, whereas the modified SCS-CN model proposed by Mishra and Singh (2003) is used in Model 2. Accordingly, the static infiltration, which is a part of the initial abstraction in Model 1, is considered separately in Model 2. Model 3 uses the original SCS-CN model equation for calculating runoff, whereas the term potential maximum retention (S) is determined as a function of M and the water holding capacity of the soil, as proposed in SWAT (Arnold et al., 2000). When the SCS-CN-based methods are used for the estimation of surface runoff, I a is generally considered as the loss, which occurs before the water enters the soil, and hence I af is taken as zero in Models 1 3. On the other hand, when these models are used along with the soil moisture model, it is logical to assume that the components of I a may be added to the soil moisture temporarily (Michel et al., 2005). Accordingly, I af is assumed to be equal to I a and this modified form of Model 2 is presented as Model 4 in the current study. Model 3 is also modified in a similar way and is presented as Model 5. In Models 2 and 4, S is used to represent the volumetric water holding capacity of the soil as shown in Table 1. Therefore, areas with higher S value tend to show higher M at saturation compared to areas with lower S value. This

450 Table 1 Comparison of SCS-CN-based models integrated with the soil moisture model Soil moisture balance M ðtþ ¼ M ðt 1Þ þ½p ðtþ þ IR sðt 1Þ þ SP iðtþ Š ½Q ðtþ þði a I af Þ ðtþ þ ET aðtþ þ SP oðtþ þ DP ðtþ ŠþR ðtþ Characteristics Parameters Remarks Model 1 Q ¼ ðp I aþ 2 P I a þ S I af ¼ 0 S t ¼ 1000 CN t 10 2:54 S ðt¼0þ ¼ S1 ¼ 1000 CN1 10 2:54 AMC I, II and III are considered based on the last 5 days rainfall (SCS, 1971) Model 2 Q ¼ ðp I a F c ÞðP I a F c þ MÞ P I a F c þ S þ M I af ¼ 0 S t ¼ S t 1 ðm t M t 1 Þ S ðt¼0þ ¼ S1 ¼ 1000 CN1 10 2:54 F c l Static and dynamic infiltrations are considered separately (Mishra and Singh, 2003) Model 3 Q ¼ ðp I aþ 2 P I a þ S Model 4 S t ¼ S1 1 FFC FFCþexpðw1 FFC:w2Þ Q ¼ ðp I a F c ÞðP I a F c þ MÞ P I a F c þ S þ M S t ¼ S t 1 ðm t M t 1 Þ Model 5 Q ¼ ðp I aþ 2 P I a þ S FFC S t ¼ S1 1 FFC þ expðw1 FFC:w2Þ Model 6 Q ¼ ðp I a F c Þ 2 ðp I a F c þ SÞ SCS-CN-SMS FFC S t ¼ S1 1 FFC þ expðw1 FFC:w2Þ I af ¼ 0 S ðt¼0þ ¼ S1 ¼ 1000 CN1 10 2:54 I af ¼ I a S ðt¼0þ ¼ S1 ¼ 1000 10 2:54 F c CN1 l I af ¼ I a S ðt¼0þ ¼ S1 ¼ 1000 10 2:54 CN1 0:9I a ; if K c < 1 I af ¼ ð4:1 3:2K c ÞI a ; if 1 K c 1:25 0:1I a ; if K c > 1:25 S ðt¼0þ ¼ S1 ¼ 1000 10 2:54 F c CN1 l S is a function of M and water holding capacity of the soil (Arnold et al., 2000) Ia is assumed to reach the soil surface (Michel et al., 2005) Ia is assumed to reach the soil surface (Michel et al., 2005) Modified proportionality relation is used (the new model proposed) CN, SCS curve number; CN1, curve number corresponding to dry condition; F c, static infiltration or final constant rate of infiltration; FFC, current soil moisture content as a fraction of field capacity; K c, crop coefficient; S, potential maximum retention; S1, potential maximum retention corresponding to dry condition; t and t 1 are used to represent the time period. w1 and w2 are shape factors used to calculate S in SWAT model; l, coefficient used to estimate initial abstraction in SCS-CN model. produces inappropriate results when geospatial variation of soil moisture is considered over a heterogeneous surface. Sandy areas, which produce very less runoff owing to the higher S values, show higher soil moisture at saturation compared to clayey areas with lower S value. This is contradictory with the soil hydraulic properties depicted based on the soil texture. In order to rectify this, the proportional equality hypothesis proposed by Mishra and Singh (2003) is further modified and is presented as a new model, SCS-CN model with soil moisture simulation (SCS-CN-SMS). This model is presented as Model 6 in Table 1. Further, the term I af is modified to account for the interception losses. In this model, the entire volume of rainfall, leaving the interception part, is assumed to reach the soil surface. In Model 6, I af is assumed to vary from 0.1 to 0.9 times I a, based on the leaf area coverage of plants. An empirical equation in terms of K c is used to represent the variation of I a as shown in Table 1. This empirical equation for I af is derived to get the maximum value (0.9 times I a ) when K c is less than or equal to 1 and the minimum value (0.1 times I a ) when K c is greater than or equal to 1.25. 2.3. Inferring spatial distribution of soil parameters Soil characteristics like hydraulic conductivity and porosity are mostly related to the soil texture. However, Golson et al. (2001) studied the influence of other parameters like slope and land use on the soil characteristics related to the water holding capacity. Soils with trees are found to have 5.2% more porosity than those with grass type vegetation, indicating the influence of land use on the hydraulic soil characteristics. Therefore, in addition to soil texture, the spatial variation of the soil hydraulic characteristics with change in land use is also considered in the present study. For this, the watershed area is divided into cells of uniform dimensions and standard values of the soil hydraulic parameters viz., SAT, FC and WP are initially adopted for each cell based on the soil texture. These standard values are assumed as the bare soil characteristics. In the absence of observed values, assuming that the curve number reasonably reflects the soil hydraulic characteristics with respect to the changes in soil and land use, spatial variation of the soil parameters are extrapolated from the corresponding standard values by using the CN. For the same

451 soil texture, water-holding capacity of the soil is assumed to increase with the increase in CN and vice versa. CN for each cell is assigned from the NEH-630 1 tables corresponding to different soil and land use classes. The actual values of SAT, FC and WP are calculated for each cell from the bare soil values by using a coefficient x, based on the difference in CN for the selected area and the bare land area in the same soil. The value of x is calculated as shown in Eq. (3). x ¼ 1 þ ðcn for the same soil group and for bare soil present CNÞ 100 This empirical relation gives the value of x greater than 1, if the CN of the area is less than that of the bare soil, giving higher values of SAT, FC and WP. Higher values of CN results in x less than 1 and hence SAT, FC and WP are less than that of the bare soil area. Static infiltration rate (F c ) and saturated hydraulic conductivity (K s ) of the soil in each cell are calculated from the physical concepts of the SCS-CN method. In the original SCS- CN method F c is considered as a part of I a. During the wet condition all the components of I a, except F c, are normally assumed as zero. Therefore, in the present study, F c is assumed to be equal to I a during the wet condition and is determined from the corresponding value of S. Further, at saturation SP i and SP o are assumed to be equal. Therefore, F c is assumed to be equal to DP and hence the resulting K s is calculated from F c using Darcy s equation with (SAT-FC) as the soil moisture available for drainage. Hydraulic conductivity of the soil layer is maximum when all the pores are filled with water and decreases with decrease in soil moisture. Unsaturated hydraulic conductivity (K) is therefore calculated as a function of M, using the following relation (Todd, 1995). M WP 3 K ¼ K s (4) SAT WP 3. Application of the models 3.1. Study area The described models are applied to a part of the Gandeshwari sub-watershed in West Bengal, India. The area is located at the lower part of the Chhotanagapur Plateau in Eastern India. The sub-humid climate of the area is associated with a mean annual rainfall of around 1200 mm, which occurs mainly during the south-west monsoon period, i.e., June to September. Topographically the landscape of the area can be divided into three: upper terraces, middle terraces and lower terraces (Banik et al., 2004). The undulating terrain and sloping subsurface strata contribute to the low water holding capacity of the area. Paddy is the major crop of the area and is mainly cultivated as rainfed crop. Even though large numbers of small water bodies are present, supplementary irrigation is a rare 1 NEH-630 (2004). NRCS National Engineering Handbook, Hydraulics and Hydrology Technical References, USDA. http:// www.wcc.nrcs.usda.gov/hydro/hydro-techref-neh-630.html (Last accessed July 24, 2006). (3) phenomenon in the area. Water availability is crucial for cultivation in the upland and medium land. Lowlands, on the other hand, are favourable with comparatively higher soil moisture for a longer period. 3.2. Generation of geospatial and temporal input parameters In the present study, the spatial variation in soil, land use and cropping pattern are considered. The total catchment area is divided into cells of 20 m 20 m size. Digital soil and land use map of the area, crop type and agricultural practices over different terrains are the input required for the model. The digital soil map of the area is prepared using the soil map from the National Bureau of Soil Survey & Land Use Planning (NBSS & LUP), Nagpur, India (Fig. 2) and the land use map is generated from the satellite imagery (LISS III image, dated April, 1998) obtained from the National Remote Sensing Agency (NRSA), India. The total catchment area of 105.6 km 2 is classified into various classes viz., water bodies, cultivable paddy fields, scrub land, dense forest, medium forest and other areas as shown in Fig. 2. These maps are brought to a same frame using GIS. During the kharif season (June to October) entire fallow land and crop land are assumed to be under paddy cultivation. For various hydrological, soil and land use conditions, suitable CN values are assigned from the NEH-630 tables. Soil moisture availability controls the selection of crop varieties over different terrains. As observed in the field, in the upper ridge land fields, drought tolerant, dry seeded, short duration (90 days) varieties are usually cultivated, whereas transplanted long duration crops (140 days) are cultivated in the low laying areas with good soil moisture condition. Middle land areas are generally cultivated with both dry seeded and transplanted varieties of medium duration (106 days). To facilitate maximum storage of water in the field, the common practice is to maintain high field bunds with a height of 200 mm. Considering the occurrence of frequent dry spells, drought tolerant varieties of the paddy crops are considered. These varieties are assumed capable to withstand soil moisture depletion up to 20% of FC, which is termed as the critical soil moisture (M critical ). As irrigation is found negligible in the study area, IR s is assumed as zero in the present study. In order to arrive at the spatial distribution of the soil water holding parameters with changes in the land use and soil characteristics, values available in literature are initially selected for each soil class, assuming bare soil condition. SAT, FC and WP selected for the bare soil condition in the area vary from 9.0%, 21.6%, and 43.0%, respectively, for sandy loam to 21.2%, 44.0%, and 53.0%, respectively, for clay loam. 2 For each cell, the actual values are extrapolated from the bare soil values using the coefficient x. The CN and the extrapolated values of SAT, FC and WP are further used to estimate F c and K s for each cell. 2 Beasley, D.B., Huggins, L.F., 1991. Soil parameters. From ANSWERS user s manual. In: http://pasture.ecn.purdue.edu/ aggrass/models/answers/ (Last accessed, August 3, 2006). Agen Grass Account, Department of Agricultural & Biological Engineering, Purdue University, USA.

452 Fig. 2 Digital map layers used in the model. Temporal data used in the present study include runoff and meteorological data. Daily variation in rainfall, temperature, relative humidity, wind velocity, and sunshine duration are collected from the meteorological observation station within the watershed. Runoff observations arecollectedfrom the gauging station located at the watershed outlet (Fig. 2). While applying the model to the study area, total dry condition is assumed prior to the rainfall and hence M is assumed to be equal to WP. Runoff generated at each cell is calculated separately and is added at the watershed outlet withtherespectivetimelagtogeneratethesurfacerunoff hydrograph. The flow path from each cell to the watershed outlet and the respective time lag are calculated through an algorithm, using the digital elevation model of the area (Jana et al., 2007). 4. Results and discussion 4.1. Geospatial variation in soil hydraulic characteristics In order to show the spatial variation in the hydraulic characteristics with changes in the terrain and soil texture, three sample locations (Fig. 2) are selected in the present study (all three locations are paddy fields with location 1 on high slope area with sandy clay, location 2 on moderate slope area with clay loam, and location 3 on low slope area with clay loam). In order to show the efficiency of the model to calculate spatially distributed parameters for each cell, variations of the soil hydraulic characteristics in these locations are presented in Table 2. The higher K s and F c together with lower SAT values for sandy clay compared to that of clay loam show good Table 2 Soil hydraulic parameters of the sample plots Location 1 Location 2 Location 3 Type of soil Sandy clay Clay loam Clay loam Land use class Paddy field Paddy field Paddy field Distributed slope (%) (from the soil map) 0 3% 1 5% 3 15% SAT (% of volume) 51.2 55.1 56.2 FC (% of volume) 25.7 43.5 46.6 WP (% of volume) 10.7 21.2 22.5 F c (mm/d) 14.9 5.9 5.1 K s (mm/day) 12.0 5.3 4.6

453 Table 3 Performance comparison of various SCS-CN-based models integrated with the soil moisture model Parameter Model performance 1114.7 mm a 988.9 mm a 1043.1 mm a 1048.9 mm a 69.8 mm b 69.8 mm b 39.7 mm b 59.8 mm b 1999 c 2000 c 2001 c Average Model 1 Simulated runoff (mm) 52.5 42.4 43.2 46.0 RMSE (mm) 1.19 0.67 0.59 0.82 Correlation coefficient 0.53 0.83 0.74 0.70 Model 2 Simulated runoff (mm) 36.4 13.4 14.1 21.3 RMSE (mm) 1.22 0.96 0.60 0.93 Correlation coefficient 0.48 0.79 0.73 0.67 Model 3 Simulated runoff (mm) 27.7 8.2 9.3 15.0 RMSE (mm) 1.15 1.04 0.69 0.96 Correlation coefficient 0.51 0.81 0.72 0.68 Model 4 [Simulated runoff (mm) 82.5 59.4 58.2 66.7 RMSE (mm) 1.42 0.48 0.64 0.85 Correlation coefficient 0.53 0.90 0.69 0.71 Model 5 Simulated runoff (mm) 66.2 41.9 39.9 49.3 RMSE (mm) 1.29 0.60 0.58 0.83 Correlation coefficient 0.54 0.89 0.69 0.71 Model 6 Simulated runoff (mm) 70.5 40.4 42.6 51.1 RMSE (mm) 1.33 0.59 0.61 0.84 Correlation coefficient 0.55 0.89 0.68 0.71 a Rainfall. b Observed runoff. c Year. agreement with the theoretical concepts of the texture based soil characteristics. In addition to the variation with soil texture, the model gives variation of these parameters with change in land use as well. The porosity and hence the SAT of clay loam soil in forest area and agricultural land are found to be 56.2% and 59.4%. The forest area is showing 5.7% more porosity than that of agricultural area, similar to the trend observed in the literature (Golson et al., 2001). 4.2. Estimation of runoff The original SCS-CN method when applied for the paddy-field dominated watershed is found to overestimate the runoff significantly. It has been observed that runoff from the paddy fields is the main source of error in the output. Runoff estimated from the small water bodies also does not reach the outlet. Field observations revealed the common practice of arresting the runoff at the paddy fields to meet the water requirement of the crops and hence to maintain the saturation condition for a longer period. Accordingly, the concept of ponding depth is incorporated in Models 1 6 through the soil moisture balance equation. Runoff simulated using these models are compared with the observed surface runoff for the monsoon season of the 3 years 1999, 2000 and 2001. Model performance is compared in terms of total runoff, root mean square error (RMSE) and correlation coefficient, and is shown in Table 3. From Table 3 it can be observed that overall performance of Models 1, 4, 5 and 6 are comparable in terms of RMSE and correlation coefficient. However, Model 1 shows fairly poor performance during the year 1999 in terms of total runoff. This can be explained due to the dependency on the past-5 days rainfall for the estimation of AMC and hence the runoff. In this particular case, AMC dependent models fail to account for the prolonged effect of a high intense rainfall, which is coupled with poor soil water drainage resulting in high soil moisture and runoff. Models 2 and 3 show the poorest performance among the six models with the estimated runoff much less than the observed values. This is due to the improper accounting of I af and the direct relation between S and M, resulting in low soil moisture and in turn less runoff. Even though the runoff estimated using Model 4 shows good correlation with the observed value, the model is found to overestimate runoff during the years 1999 and 2001. Model 1 gives the best performance in terms of RMSE, whereas Models 5 and 6 give best correlation between the observed and simulated runoff. The average runoff estimated using Model 4 is closest to the observed value followed by Model 6. Fig. 3 shows the runoff hydrograph produced by Model 6 along with the observed values of rainfall and surface runoff from the study area.

454 Fig. 3 Runoff estimated using Model 6. Fig. 4 Spatio-temporal variation in soil moisture simulated using Models 1 5 in 2000.

455 Fig. 5 Spatio-temporal variation in soil moisture simulated using Model 6. 4.3. Estimation of soil moisture Fig. 4 shows the daily variation in M estimated using Models 1 5 for the three sample locations under no-irrigation condition in the year 2000. It can be observed that Model 1 could not estimate the soil moisture satisfactorily during the initial period. The output shows dry condition persisting as long as the mid of July, wherein the actual field information indicates field condition favouring the initiation of paddy cultivation. Models 2 and 3 are underestimating runoff significantly due to the improper accounting of I af.inmodels 2 and 3, where I af is taken as zero, infiltration is considered Table 4 Drought severity of the area as estimated from the soil moisture models Year Paddy production statistics available Predicted water stress condition for the revenue block (100 t) 1 a 69 b 2 a 80 b 3 a 110 b 1 a 69 b 2 a 80 b 3 a 110 b 1 a 69 b 2 a 80 b 3 a 110 b Remarks Number of days M < M critical Model 4 Model 5 Model 6 1999 980 June Area under paddy cultivation (% of total watershed area): Upland = 15; Middle land = 8; Lowland = 52 July 0 5 6 5 5 5 11 5 5 Non-paddy area = 25 August 0 0 0 0 0 0 0 0 0 September 0 0 0 0 0 0 0 0 0 October 0 0 0 Total 0 5 6 5 5 5 11 5 5 2000 568 June July 0 8 13 0 5 5 18 13 13 August 0 4 11 7 0 0 17 4 3 September 0 0 1 3 0 0 6 0 0 October 0 0 0 Total 0 12 25 10 5 5 41 17 16 2001 770 June July 0 0 0 0 0 0 0 0 0 August 0 0 0 0 0 0 21 0 0 September 0 0 3 0 0 0 8 0 0 October 6 0 0 a Location. b Total number of days considered. Total 0 0 9 0 0 0 29 0 0

456 only when the rainfall intensity is greater than I a. However, I a is a function of S and in turn of M. It continues to remain high as the infiltration occurs and increases M. ThisresultsinM much less than the SAT throughout the simulation period. Thus,itisobservedthatModels2and3failedtosimulate both surface runoff and soil moisture in dry areas. On the other hand, in Model 1 I a and S change with AMC, irrespective of M, and thus results in higher infiltration rates and M compared to Models 2 and 3. Model 4 in which I af is taken as I a, is found to show gradual increase in M with rainfall. Nevertheless, due to the direct relation between S and the volumetric water storage in the soil, sandy clay area (location1)withhighers value shows higher soil moisture and longer ponding period compared to the clay loam areas (locations 2 and 3). This is contradictory with the texturebased soil hydraulic characteristics. Model 5 shows a reasonable variation of M with change in soil texture. Simulated values of M in the clay loam area are higher than those in the sandy clay area. The soil moisture variation simulated using Model 6 for the 3 years (1999, 2000 and 2001) are shown in Fig. 5. Assuming WP as the initial condition, larger WP results in higher initial soil moisture for clay loam. Sandy clay, for which hydraulic conductivity is high, but total porosity is less, attains FC faster compared to clay loam. When both kind of soil reach saturation, M remains less in the sandy clay than that in the clay loam due to the smaller porosity. After saturation, due to the lower FC and higher drainage rate, M remains less in the sandy clay than that in the clay loam. Models 5 and 6 are thus found to give the best output among the six models. Comparison between Models 5 and 6 shows that during the dry period, when soil generally fails to reach the SAT, Model 5 is found to give higher M against Model 6. Models 4 6 are further validated by comparing the simulated water stress condition with that actually observed in the field. In order to assess the water stress condition in the paddy fields, suitable agricultural practices are assumed over different terrains. Dry seeding in the 1st week of July is assumed in the upper lands, whereas in the middle land and lowland areas crop transplantation date is selected as the 6th of July. Even though the normal practice is to maintain the ponding depth throughout the crop growth period, the last stage of the crop period is not considered as critical, as the crop is allowed to dry out during this stage. Therefore, while analysing the critical water stress in the fields, soil moisture stress during the last stage of the crop is not considered. Thus, the critical periods considered are 69, 80 and 110 days from the seeding or transplantation date for the upper lands, middle lands and lowlands, respectively. Table 4 shows the number of days, within the critical period, during which the soil moisture reaches M critical as estimated by Models 4 6, assuming noirrigation condition in the area. From Table 4 it can be observed that Model 4 predicts more water stress in location 3 than in locations 1 and 2, which is against the field observations. Models 5 and 6 predict a slight moisture stress during the initial stage of the crop in the year 1999, which may turn out in a small reduction of the production. During the year 2000, Model 5 shows mild water stress in the fields, whereas Model 6 shows severe stress, which may affect the crop production. Model 5 identified the year 2001 as the best period with no soil moisture stress in the field. On the other hand, Model 6 shows long moisture stress period in the upland fields in 2000, pointing out the possible reduction in yield over 20% of the total paddy area. Thus, from the soil moisture simulation, Model 5 predicts maximum possibility of crop production in 2001 followed by the years 1999 and 2000, whereas Model 6 predicts maximum crop production in 1999 even though a small reduction in the production is expected due to the slight soil moisture stress during the initial period. Model 6 further predicts moderate crop production in 2001 and very poor in 2000, owing to the severe soil moisture stress during the entire crop growth period. The available statistics on crop production in the revenue block containing the study area as a part of it is also shown in Table 4. The field data shows maximum crop production in 1999 followed by 2001 and 2000, as predicted by Model 6. Even though, Models 1, 4 and 5 give reasonably accurate runoff estimation, these models fail to produce satisfactory results when applied for the geospatial estimation of soil moisture. By incorporating the physical concepts in a more appropriate way, the proposed SCS-CN-SMS model integrated with the soil moisture model (Model 6) is found to give the best results when both runoff and soil moisture simulations are considered. 5. Conclusions An attempt is made to extend the SCS-CN-based models for the estimation of both surface runoff and soil moisture from saturated/ponded paddy areas and a new SCS-CN-SMS model is presented. All the models are generalized for the different land use classes in the watershed considered. Spatial variation in the soil and land use characteristics is incorporated in the models using remote sensing and GIS. The physical interpretation of the SCS-CN method is attempted to link with the soil water holding characteristics and accordingly the spatial variation in the soil hydraulic characteristics are inferred based on the variation in CN. An integrated model, based on the soil moisture balance, is the most efficient model when both runoff estimation and soil moisture simulation are considered. The original SCS-CN-based method and its variants showed poor performance in dry areas either due to the improper accounting of initial abstraction or the direct relation between S and the volumetric water storage in the soil. The models, in which the temporary effect of initial abstraction on soil moisture is considered, are found to give better results for soil moisture and runoff simulation. The proposed SCS-CN-SMS model is superior to the other models as it gives a runoff output comparable with the observed runoff. The geospatial and temporal variation in the soil moisture stress condition simulated using the proposed model is most satisfactory with respect to the physical soil characteristics. The soil moisture stress is also validated using the actual crop production statistics from the area. The model can be checked further with the direct spatio-temporal measurement of soil moisture during the crop-growing season.

457 Acknowledgement The work presented in this paper is a part of the research project carried out at the Centre of Studies in Resources Engineering, IIT Bombay, sponsored by the Department of Science and Technology, Government of India and is gratefully acknowledged. references Agrawal, M.K., Panda, S.N., Panigrahi, B., 2004. Modeling water balance parameters for rain-fed rice. J. Irrig. Drain. Eng. 130 (2), 129 139. Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop evapotranspiration guidelines for computing crop water requirements. FAO Irrigation and Drainage Paper 56. FAO, Rome. Arnold, J.G., Williams, J.R., Srinivasan, R., King, K.W., 2000. Soil and Water Assessment Tool Manual. USDA Agricultural Research Service, Texas. Banik, P., Edmonds, C., Fuwa, N., Kam, S.P., Villano, L., Bagchi, D.K., 2004. Natural resources endowments, subsistence agriculture, and poverty in the Chhotanagpur Plateau. Discussion Paper 47, International Rice Research Institute, Philippines. DiLuzio, M., Srinivasan, R., Arnold, J.G., Neitsch, S.L., 2002. ArcView interface for SWAT 2000, user s guide. TWRI Report TR-193, Texas Water Resources Institute, Collage Station, Texas. Golson, K.F., Tsegaye, T.D., Green, T.H., Mays, D., Coleman, T.L., 2001. Spatial distribution of soil physical properties under various forested and agricultural land use management practices. In: IEEE 2001 International, Geoscience and Remote Sensing Symposium, 2001-IGRASS 01. Proceedings of the Symposium, 9 13 July 2001, at University of New South Wales, Sydney, Australia. Jana, R., Reshmidevi, T.V., Arun, P.S., Eldho, T.I., 2007. An enhanced technique in the construction of the discrete drainage network from low resolution spatial database. Comput. Geosci. 33 (6), 717 727. Kang, M.S., Park, S.W., Lee, J.J., Yoo, K.H., 2006. Applying SWAT for TMDL programs to a small watershed containing rice paddy fields. Agric. Water Manage. 79, 72 92. Michel, C., Andre assian, V., Perrin, C., 2005. Soil Conservation Service Curve Number method: how to mend a wrong soil moisture accounting procedure? Water Res. Res. 41 (2). Mishra, S.K., Singh, V.P., 2003. Soil Conservation Service Curve Number (SCS-CN) Methodology. Kluwer Academic Publication, The Netherlands. Mishra, S.K., Jain, M.K., Singh, V.P., 2004. Evaluation of the SCS- CN-based model incorporating antecedent moisture. Water Res. Manage. 18 (6), 567 589. Mishra, S.K., Jain, M.K., Bhunya, P.K., Singh, V.P., 2005. Field applicability of the SCS-CN-based Mishra-Singh general model and its variants. Water Res. Manage. 19 (1), 37 62. Panigrahi, B., Panda, S.N., Mull, R., 2001. Simulation of water harvesting potential in rainfed ricelands using water balance model. Agric. Syst. 69, 165 182. Portoghese, I., Uricchio, V., Vurro, M., 2005. A GIS tool for hydrogeological water balance evaluation on a regional scale in semi-arid environments. Comput. Geosci. 31, 11 27. SCS, 1971. Hydrology, National Engineering Handbook. USDA- Soil Conservation Services (1971), Washington, DC. Stephens, D.B., 2000. Vadose Zone Hydrology. Lewis Publishers, Boca Raton, Florida. Todd, D.K., 1995. Groundwater Hydrology. John Wiley & Sons, Singapore.