The present study evaluated five regionalization methods: global averaging; regression; spatial proximity; behavioral similarity and artificial neural network (ANN) for Soil and Water Assessment Tool (SWAT), using data from 24 river basins in monsoon dominated tropical river basins of peninsular India. Regionalization was performed for each basin using the remaining 23 basins. The performance of the calibration and thus the regionalization method is limited by the unreliable or erroneous data at the basins. Overall, we found that the regression method outperforms other regionalization methods in terms of predicting the daily as well as peak discharges. It was found that despite showing a better R2 in training, testing and validation, the ANN method performed poorly probably due to a lower number of training data. Therefore, it is suggested that the ANN should be avoided for regionalization in the absence of sufficient training data. Moreover, the regression equations developed in the present study can be utilized to predict SWAT parameters of basins located in the vicinity of the study area. However, the basins located far away from the group of catchments or having diverse characteristics should be avoided for regionalization.

  • Overall, the regression-based method showed comparatively better performance both in terms of precision and accuracy.

  • Simpler regression methods are better than complex ANNs when the number of gauged basins are limited.

  • The performance of the regionalization method is limited by the unreliable or erroneous data at the basin.

Hydrological processes are complex and dynamic, making their accurate predictions difficult. Since simple empirical equations are not proficient enough for prediction, hydrological modeling becomes necessary to simulate these complex hydrological processes. Numerous models have been developed for understanding the hydrological systems, especially over the past few decades (Devia et al. 2015). Some of the most commonly used models in water resources management are ANSWERS (Areal Non-Point Source Watershed Environment Response Simulation; Beasley et al. 1980), AGNPS (Agricultural Non-Point Source Pollution; Young et al. 1989), WEPP (Water Erosion Prediction Project; Laflen et al. 1991), SWAT (Soil and Water Assessment Tool; Arnold et al. 1993), VIC (Variable Infiltration Capacity model; Liang et al. 1994), TOPMODEL (a TOPography-based hydrological MODEL; Beven 1997), HEC-HMS (Hydrologic Modeling System; Cunderlik & Simonovic 2004) and GSFLOW (Groundwater and Surface-water FLOW; Markstrom et al. 2008). Among them, SWAT is arguably the most popular model and has been extensively used in many studies with satisfactory performance (Wilk & Hughes 2002; Tripathi et al. 2003; Dhar & Mazumdar 2009; Lakshmanan et al. 2011; Garg et al. 2012; Verma & Verma 2019; Oo et al. 2020; Swain et al. 2020).

All the models require some degree of calibration and validation to achieve adequate basin representation, which is possible only for gauged basins. However, in many basins, observed streamflow data are not available or are insufficient for developing models. Such basins are considered as ungauged (Sivapalan et al. 2003). To overcome the problem of model calibration in ungauged basins, various regionalization techniques have been developed. The process of transferring parameters from hydrologically similar basins (donors) to a basin of interest (target) is generally referred to as regionalization (Blöschl & Sivapalan 1995).

The simplest method for regionalization is to identify similar or proxy basins, based on location or behavior, known as the distance-based method. When the geographical distance is used as the basis to determine similar basins, the entire set of model parameters is transferred from the nearest basin or a combination of a number of donor basins to the target basins. The approach has been widely used in various studies and reported acceptable results (Merz & Blöschl 2004; Parajka et al. 2005; Samuel et al. 2011; Ssegane et al. 2012; Swain & Patra 2017; Yang et al. 2017; Choubin et al. 2019; Guiamel & Lee 2020). For behavioral similarity, basin attributes such as landscape, geology, climate as well as human factors, which predetermine hydrological behavior, are used to determine similar basins for transferring the entire set of model parameters. The proper definition of a similarity index and its criterion is the key basis for this type of method. A modification of the above method is the clustering-based analysis (Parajka et al. 2005; Zhang & Chiew 2009). Since the basis for determining basin similarities is not precisely known, distance-based methods have yielded mixed outcomes (He et al. 2011; Guo et al. 2020). Kittel et al. (2020) found that the behavioral similarity method performed better than the nearest-neighbor regionalization method for two African river basins. Du et al. (2020) combined behavioral similarity regionalization and satellite observations of various hydrological variables over Greater Mekong Subregion and found significant improvement in water level prediction as compared with the global regionalization parameters. Matos et al. (2020) applied behavioral similarity in the Juruena River Basin to estimate maximum, minimum and long-term mean streamflows. They found drainage area, total watercourse length, sub-basin mean altitude and perimeter to be the explanatory variables for the regionalization. According to Merz et al. (2006), compared with the method of regionalization, quantity and quality of expert judgment (meaningful catchment attributes (Skøien et al. 2006) and hydrological distance measures (Merz & Blöschl 2005)) play a very significant role in maximizing regionalization performance.

Regression, especially two-step regression, is the most popular and widely used regionalization technique. In the two-step regression, regionalization is performed in two steps: (1) calibrate model parameters for each basin and (2) relate model parameters to basin attributes using multivariate regression. A vast number of studies have successfully applied this method to the regionalization of model parameters (He et al. 2011; Guo et al. 2020). However, some of the studies reported that relationships found between model parameters and basin attributes are often weak and prediction in ungauged basins achieves limited success (Fernandez et al. 2000; Heuvelmans et al. 2006; Kim & Kaluarachchi 2008). Some of the studies also applied sequential (or multi-step) regression, in which model parameters are calibrated sequentially from the most identifiable parameter to the least. In general, sequential regression is reported to perform better than the two-step regression (He et al. 2011). However, Wagener & Wheater (2006) reported that the two-step regression performs better than the sequential regression for regionalization.

Artificial neural networks (ANNs) are nonlinear flexible model structures conceptually based on the function of the human brain (Lek & Guégan 1999). Physical catchment descriptors and model parameters are known to be interdependent. Also, it is well known that their interactions are highly nonlinear; therefore, ANN can be a suitable tool for the regionalization of model parameters (Maier & Dandy 2001; Kumar et al. 2019). Heuvelmans et al. (2006) compared linear regression analysis and ANNs (with three hidden layers) for regionalizing the most sensitive parameters of SWAT for the Scheldt river basin (25 catchments). They found ANNs to be accurate in predicting parameters if the physical catchment descriptors of the target catchment lie within the range of the descriptor values of the donor catchments used for the construction of the ANNs. However, for extrapolations outside this range, regression analysis yielded a better result. Besaw et al. (2010) found that the ANNs trained on a climate-discharge record from one basin are capable of predicting streamflow in a nearby basin as accurately as in the basin on which they were trained.

In the developing countries, such as India, there are many basins for which very little or no monitoring data are available. There have been very limited studies on this subject due to the requirement of enormous data for a large number of basins. Swain & Patra (2017) compared regionalization methods based on spatial proximity (inverse distance weighted, kriging and global mean), regression and behavioral similarity for 32 basins in Eastern and Southern India. They reported that geographical distance-based methods perform better than those based on behavioral similarity. Further, Swain & Patra (2019) categorized 32 catchments into four homogeneous groups using the self-organizing map method. They found that the prior classification of catchments into homogeneous groups helped in improving the regionalization output.

Recently, by comparing the results of 33 typical studies involving similarity- and regression-based regionalization, Guo et al. (2020) found that the accuracy of a regionalization method increases with an increasing number of gauged catchments or the runoff site density. There is still plenty of room to improve the prediction capability in data-sparse regions. The main objective of the present study is to compare different regionalization methods and it suggests a suitable method for regionalizing SWAT parameters so that the model can be applied to ungauged basins in peninsular India. Performance of the regression-based regionalization was compared with default, calibrated and other methods of regionalization (global averaging, spatial proximity, behavioral similarity and ANN). The description of the study area and datasets used is provided in the section below. This is followed by the ‘Methodology’ section which includes SWAT model description and various regression methods used. The results of the estimation of SWAT parameters by different methods and their comparison is provided in the ‘Results and discussion’ section. The summary and conclusion of the study are given in the final section.

Twenty-four basins that satisfy the following four conditions were selected for impact assessment:

  • 1.

    Basin Area: Due to the coarse resolution of input data, out of the available basins, we discarded the basins having an area less than 1,000 km2.

  • 2.

    Streamflow Records: To match with the temporal availability of input data, we chose basins having streamflow data available for the study period (2000–2013).

  • 3.

    Human Intervention: All those basins having any dam or reservoir were discarded to avoid consideration of regulated flows.

  • 4.

    Irrigated basins: Similar to the above condition, we discarded the basins having more than 50% area covered by irrigated land.

Figure 1 and Table 1 provide locations of all the 24 basins chosen for this study. Basins were numbered from B-1 to B-24 in the alphabetical order of their outlet gauging stations. Most of the gauging stations are located in the states of Chhattisgarh (five stations), Madhya Pradesh (five stations) and Odisha (five stations). Two stations are located in each of the states of Karnataka, Telangana and Kerala, whereas Gujarat, Jharkhand and Maharashtra contain one station each. The majority of the basins are part of Godavari, Krishna and Mahanadi rivers basins.

Table 1

Locations of all the selected WRIS gauging stations

Station IDGauging station
StateTributaryRiver
NameCoordinates
EastNorth
B-1 Ambabal 81.790 19.288 Chhattisgarh Indravathi Godavari 
B-2 Andhiyarkore 81.604 21.833 Chhattisgarh Hamp Mahanadi 
B-3 Arangaly 76.314 10.281 Kerala Chalakudy Periyar 
B-4 Bantwal 75.041 12.881 Karnataka – Netravathi 
B-5 Belkheri 79.339 22.929 Madhya Pradesh Sher Narmada 
B-6 Bhatpalli 79.471 19.316 Telangana Pranhita Godavari 
B-7 Burhanpur 76.235 21.299 Madhya Pradesh – Tapi 
B-8 Chhidgaon 77.311 22.402 Madhya Pradesh Ganjal Narmada 
B-9 Damercharla 79.669 16.737 Telangana Musi Krishna 
B-10 Gadarwara 78.788 22.930 Madhya Pradesh Sakkar Narmada 
B-11 Gunupur 83.809 19.088 Odisha – Banshadhara 
B-12 Jaraikela 85.110 22.321 Odisha Koel Brahmani 
B-13 Kantamal 83.723 20.653 Odisha Tel Mahanadi 
B-14 Kesinga 83.222 20.204 Odisha Tel Mahanadi 
B-15 Mahuwa 73.132 21.020 Gujarat – Purna 
B-16 Mangrul 77.986 20.189 Maharashtra Pranhita Godavari 
B-17 Muthankera 76.121 11.847 Kerala Kabini Cauvery 
B-18 Patan 79.664 23.312 Madhya Pradesh Heran Narmada 
B-19 Pathagudem 80.348 18.848 Chhattisgarh Indravathi Godavari 
B-20 Shimoga 75.577 13.919 Karnataka Tungabhadra Krishna 
B-21 Sonarpal 81.881 19.270 Chhattisgarh Indravathi Godavari 
B-22 Sundargarh 84.009 22.114 Odisha Ib Mahanadi 
B-23 Tilga 84.429 22.644 Jharkhand Sankh Brahmani 
B-24 Tumnar 81.232 19.012 Chhattisgarh Indravathi Godavari 
Station IDGauging station
StateTributaryRiver
NameCoordinates
EastNorth
B-1 Ambabal 81.790 19.288 Chhattisgarh Indravathi Godavari 
B-2 Andhiyarkore 81.604 21.833 Chhattisgarh Hamp Mahanadi 
B-3 Arangaly 76.314 10.281 Kerala Chalakudy Periyar 
B-4 Bantwal 75.041 12.881 Karnataka – Netravathi 
B-5 Belkheri 79.339 22.929 Madhya Pradesh Sher Narmada 
B-6 Bhatpalli 79.471 19.316 Telangana Pranhita Godavari 
B-7 Burhanpur 76.235 21.299 Madhya Pradesh – Tapi 
B-8 Chhidgaon 77.311 22.402 Madhya Pradesh Ganjal Narmada 
B-9 Damercharla 79.669 16.737 Telangana Musi Krishna 
B-10 Gadarwara 78.788 22.930 Madhya Pradesh Sakkar Narmada 
B-11 Gunupur 83.809 19.088 Odisha – Banshadhara 
B-12 Jaraikela 85.110 22.321 Odisha Koel Brahmani 
B-13 Kantamal 83.723 20.653 Odisha Tel Mahanadi 
B-14 Kesinga 83.222 20.204 Odisha Tel Mahanadi 
B-15 Mahuwa 73.132 21.020 Gujarat – Purna 
B-16 Mangrul 77.986 20.189 Maharashtra Pranhita Godavari 
B-17 Muthankera 76.121 11.847 Kerala Kabini Cauvery 
B-18 Patan 79.664 23.312 Madhya Pradesh Heran Narmada 
B-19 Pathagudem 80.348 18.848 Chhattisgarh Indravathi Godavari 
B-20 Shimoga 75.577 13.919 Karnataka Tungabhadra Krishna 
B-21 Sonarpal 81.881 19.270 Chhattisgarh Indravathi Godavari 
B-22 Sundargarh 84.009 22.114 Odisha Ib Mahanadi 
B-23 Tilga 84.429 22.644 Jharkhand Sankh Brahmani 
B-24 Tumnar 81.232 19.012 Chhattisgarh Indravathi Godavari 
Figure 1

Location map of 24 study basins in which the ‘star’ symbol represents basin outlet.

Figure 1

Location map of 24 study basins in which the ‘star’ symbol represents basin outlet.

Close modal

Table 2 lists input data used for the SWAT model simulations. Except for precipitation, other meteorological input data for SWAT were obtained from National Solar Radiation Database (NSRDB; Wilcox 2007) for the entire period (2000–2013), which provides hourly data at a spatial resolution of 10 km. The primary source of NSRDB data for temperature, wind speed and relative humidity is Modern-Era Retrospective Analysis for Research and Applications (MERRA; Rienecker et al. 2011), while the solar radiation data are based on the semi-empirical model of satellite data (State University of New York (SUNY) model) developed by Perez et al. (2002).

Table 2

Details of meteorological and SWAT input data

S. No.DataResolutionPeriodSourceReference
Daily streamflow Point 2000–2013 Water Resources Information System of India (WRIS) http://www.india-wris.nrsc.gov.in 
Daily precipitation 0.25° 2000–2013 India Meteorological Department (IMD) http://www.imdpune.gov.in 
Hourly temperature 10 km 2000–2013 National Solar Radiation Database (NSRDB) Wilcox (2007)  
Hourly wind speed 10 km 2000–2013 National Solar Radiation Database (NSRDB) Wilcox (2007)  
Hourly relative humidity 10 km 2000–2013 National Solar Radiation Database (NSRDB) Wilcox (2007)  
Hourly solar radiation 10 km 2000–2013 National Solar Radiation Database (NSRDB) Wilcox (2007)  
LULC 1 km 1994 USGS Global Land Cover Characterization Loveland et al. (2000)  
Soil data 10 km 1972 Food and Agriculture Organization of the United Nations FAO (1997)  
S. No.DataResolutionPeriodSourceReference
Daily streamflow Point 2000–2013 Water Resources Information System of India (WRIS) http://www.india-wris.nrsc.gov.in 
Daily precipitation 0.25° 2000–2013 India Meteorological Department (IMD) http://www.imdpune.gov.in 
Hourly temperature 10 km 2000–2013 National Solar Radiation Database (NSRDB) Wilcox (2007)  
Hourly wind speed 10 km 2000–2013 National Solar Radiation Database (NSRDB) Wilcox (2007)  
Hourly relative humidity 10 km 2000–2013 National Solar Radiation Database (NSRDB) Wilcox (2007)  
Hourly solar radiation 10 km 2000–2013 National Solar Radiation Database (NSRDB) Wilcox (2007)  
LULC 1 km 1994 USGS Global Land Cover Characterization Loveland et al. (2000)  
Soil data 10 km 1972 Food and Agriculture Organization of the United Nations FAO (1997)  

Daily gridded precipitation data at 0.25° resolution were obtained from India Meteorological Department (IMD) for the period 2000–2013 (Pai et al. 2014). Daily streamflow data were obtained from Water Resources Information Systems (WRIS) of India.

The land use map of the Global Land Cover Characterization (GLCC) version 2.0 was used to estimate vegetation parameters for the SWAT model (Loveland et al. 2000). The GLCC is part of the United States Geological Survey (USGS) database, with a spatial resolution of 1 km and 24 classes of land use land cover (LULC) representation (‘http://edc2.usgs.gov/glcc/glcc.php’) for the year 1994.

Since LULC changes after 1995 have a negligible impact on streamflow of Indian rivers (Babar & Ramesh 2015), numerous SWAT modeling studies have used GLCC data for studying recent periods (El-Sadek & Irvem 2014; Liechti et al. 2015; Sun et al. 2016; Rahman et al. 2017). The soil data were extracted from the database of the Food and Agriculture Organization (FAO 1997), which has 32 soil categories for the study area.

Table 3 provides the climatic and physical basin attributes used in the study. The basin areas range from 1,260 to 40,000 km2. As expected, basins located along the western ghats receive higher annual rainfall with B-4 having the highest mean annual rainfall of 2,940.46 mm. The 24 LULC classes were grouped into seven categories: dryland cropland and pasture (DC), irrigated cropland and pasture (IC), cropland/grassland mosaic (CG), cropland/woodland mosaic (CW), savanna, forest and other. The classified LULC maps of all the basins along with watersheds and streamlines, using shuttle radar topography mission (SRTM) digital elevation map (DEM), are provided in Supplementary Figures S1 and S2. Most of the basins are covered by DC and Savanna categories. The soil data were also reclassified into five categories: Loam (L), Clay (C), Clay Loam (CL), Sandy Loam (SL) and Sandy Clay Loam (SCL).

Table 3

Basin attributes used for regionalization

BasinElevationSlopeAreaTemperature (°C)
RainfallRHLULC (%)
Soil (%)
(m)(m/m)(km2)(mm)(%)DCICCGCWSavanaForestOtherLCCLSLSCL
B-1 543 0.030 1,968 50.90 4.50 1,458.04 56.80 30.44 3.60 14.71 3.46 44.11 3.56 0.13 0.00 0.00 0.00 100.00 0.00 
B-2 265 0.055 2,210 43.50 10.20 1,123.02 75.62 27.59 3.74 33.77 0.21 30.38 4.31 0.00 7.12 3.60 27.90 61.39 0.00 
B-3 0.206 1,432 50.40 7.70 1,967.41 71.46 0.00 3.81 0.00 7.72 0.06 86.17 2.24 37.70 0.00 9.32 0.00 52.97 
B-4 0.153 3,184 50.30 11.10 2,940.46 70.71 0.00 2.91 0.00 18.67 0.43 77.54 0.45 0.00 0.00 33.19 0.00 66.82 
B-5 368 0.048 1,508 49.00 6.70 1,146.16 68.97 48.84 23.83 17.33 1.15 8.85 0.00 0.00 0.00 100.00 0.00 0.00 0.00 
B-6 156 0.077 3,100 49.70 4.80 1,220.43 64.39 18.67 8.71 36.93 0.57 35.08 0.04 0.00 0.00 7.68 92.32 0.00 0.00 
B-7 213 0.080 8,487 39.90 12.40 1,067.59 84.51 37.60 5.05 29.24 0.12 27.70 0.28 0.01 0.00 22.83 77.17 0.00 0.00 
B-8 296 0.079 1,729 52.10 6.40 1,283.63 67.62 44.08 0.00 11.94 1.60 42.38 0.00 0.00 0.00 40.54 59.46 0.00 0.00 
B-9 63 0.021 11,501 51.60 6.40 821.72 67.59 39.63 38.36 19.31 0.11 0.19 0.00 2.41 0.00 0.58 99.41 0.00 0.00 
B-10 342 0.086 2,270 40.90 14.60 1,049.48 84.75 55.61 1.62 11.02 0.24 30.89 0.52 0.10 0.00 98.34 1.66 0.00 0.00 
B-11 80 0.193 6,740 51.80 8.50 1,323.89 60.60 5.06 28.10 1.83 11.95 27.21 23.14 2.72 0.00 0.00 0.00 0.00 100.00 
B-12 185 0.060 9,160 50.50 8.00 1,313.10 61.56 14.51 39.41 13.51 1.84 24.54 4.63 1.56 38.50 2.96 0.00 4.02 54.52 
B-13 118 0.092 19,600 51.40 6.60 1,475.67 68.53 10.94 31.88 13.11 8.98 28.21 5.91 0.97 0.00 0.00 48.69 24.52 26.79 
B-14 166 0.080 11,960 41.40 11.80 1,500.99 84.20 14.64 34.65 13.76 1.21 31.74 3.66 0.36 0.00 0.00 57.14 27.51 15.36 
B-15 0.096 1,995 49.10 7.10 1,305.71 70.16 15.58 2.43 41.40 0.18 39.21 0.05 1.15 4.72 31.66 63.60 0.00 0.00 
B-16 279 0.026 2,500 51.10 3.90 873.73 64.71 40.74 23.39 35.74 0.00 0.00 0.00 0.13 0.00 86.97 13.03 0.00 0.00 
B-17 705 0.096 1,260 38.80 23.50 1,305.71 77.39 0.00 3.90 0.00 12.98 0.00 83.12 0.00 0.00 0.00 5.63 0.00 94.37 
B-18 412 0.036 3,950 49.30 3.90 1,194.97 64.83 65.81 0.42 2.91 0.13 28.48 1.24 1.01 13.07 86.93 0.00 0.00 0.00 
B-19 104 0.069 40,000 50.50 8.50 1,497.89 72.33 11.11 10.89 8.71 2.13 34.62 32.10 0.46 0.00 7.31 23.46 53.34 15.89 
B-20 556 0.104 2,831 51.40 4.40 2,684.73 57.74 0.08 8.79 0.02 24.02 6.41 59.31 1.38 0.00 0.00 0.50 0.00 99.50 
B-21 541 0.030 1,523 51.70 8.30 1,613.39 63.61 21.59 11.22 10.73 1.73 48.58 6.15 0.00 0.00 0.00 0.00 100.00 0.00 
B-22 214 0.058 5,870 52.30 7.00 1,319.08 57.61 31.58 16.78 19.08 0.83 27.33 3.78 0.63 22.15 0.00 4.73 73.12 0.00 
B-23 372 0.087 3,160 50.80 10.10 1,255.23 70.25 27.78 20.37 13.48 1.53 33.55 1.19 2.10 51.20 0.00 0.00 48.80 0.00 
B-24 330 0.072 1,700 44.30 13.00 1,525.77 79.16 21.83 4.33 6.51 3.89 58.16 5.25 0.04 0.00 0.00 0.00 50.76 49.24 
BasinElevationSlopeAreaTemperature (°C)
RainfallRHLULC (%)
Soil (%)
(m)(m/m)(km2)(mm)(%)DCICCGCWSavanaForestOtherLCCLSLSCL
B-1 543 0.030 1,968 50.90 4.50 1,458.04 56.80 30.44 3.60 14.71 3.46 44.11 3.56 0.13 0.00 0.00 0.00 100.00 0.00 
B-2 265 0.055 2,210 43.50 10.20 1,123.02 75.62 27.59 3.74 33.77 0.21 30.38 4.31 0.00 7.12 3.60 27.90 61.39 0.00 
B-3 0.206 1,432 50.40 7.70 1,967.41 71.46 0.00 3.81 0.00 7.72 0.06 86.17 2.24 37.70 0.00 9.32 0.00 52.97 
B-4 0.153 3,184 50.30 11.10 2,940.46 70.71 0.00 2.91 0.00 18.67 0.43 77.54 0.45 0.00 0.00 33.19 0.00 66.82 
B-5 368 0.048 1,508 49.00 6.70 1,146.16 68.97 48.84 23.83 17.33 1.15 8.85 0.00 0.00 0.00 100.00 0.00 0.00 0.00 
B-6 156 0.077 3,100 49.70 4.80 1,220.43 64.39 18.67 8.71 36.93 0.57 35.08 0.04 0.00 0.00 7.68 92.32 0.00 0.00 
B-7 213 0.080 8,487 39.90 12.40 1,067.59 84.51 37.60 5.05 29.24 0.12 27.70 0.28 0.01 0.00 22.83 77.17 0.00 0.00 
B-8 296 0.079 1,729 52.10 6.40 1,283.63 67.62 44.08 0.00 11.94 1.60 42.38 0.00 0.00 0.00 40.54 59.46 0.00 0.00 
B-9 63 0.021 11,501 51.60 6.40 821.72 67.59 39.63 38.36 19.31 0.11 0.19 0.00 2.41 0.00 0.58 99.41 0.00 0.00 
B-10 342 0.086 2,270 40.90 14.60 1,049.48 84.75 55.61 1.62 11.02 0.24 30.89 0.52 0.10 0.00 98.34 1.66 0.00 0.00 
B-11 80 0.193 6,740 51.80 8.50 1,323.89 60.60 5.06 28.10 1.83 11.95 27.21 23.14 2.72 0.00 0.00 0.00 0.00 100.00 
B-12 185 0.060 9,160 50.50 8.00 1,313.10 61.56 14.51 39.41 13.51 1.84 24.54 4.63 1.56 38.50 2.96 0.00 4.02 54.52 
B-13 118 0.092 19,600 51.40 6.60 1,475.67 68.53 10.94 31.88 13.11 8.98 28.21 5.91 0.97 0.00 0.00 48.69 24.52 26.79 
B-14 166 0.080 11,960 41.40 11.80 1,500.99 84.20 14.64 34.65 13.76 1.21 31.74 3.66 0.36 0.00 0.00 57.14 27.51 15.36 
B-15 0.096 1,995 49.10 7.10 1,305.71 70.16 15.58 2.43 41.40 0.18 39.21 0.05 1.15 4.72 31.66 63.60 0.00 0.00 
B-16 279 0.026 2,500 51.10 3.90 873.73 64.71 40.74 23.39 35.74 0.00 0.00 0.00 0.13 0.00 86.97 13.03 0.00 0.00 
B-17 705 0.096 1,260 38.80 23.50 1,305.71 77.39 0.00 3.90 0.00 12.98 0.00 83.12 0.00 0.00 0.00 5.63 0.00 94.37 
B-18 412 0.036 3,950 49.30 3.90 1,194.97 64.83 65.81 0.42 2.91 0.13 28.48 1.24 1.01 13.07 86.93 0.00 0.00 0.00 
B-19 104 0.069 40,000 50.50 8.50 1,497.89 72.33 11.11 10.89 8.71 2.13 34.62 32.10 0.46 0.00 7.31 23.46 53.34 15.89 
B-20 556 0.104 2,831 51.40 4.40 2,684.73 57.74 0.08 8.79 0.02 24.02 6.41 59.31 1.38 0.00 0.00 0.50 0.00 99.50 
B-21 541 0.030 1,523 51.70 8.30 1,613.39 63.61 21.59 11.22 10.73 1.73 48.58 6.15 0.00 0.00 0.00 0.00 100.00 0.00 
B-22 214 0.058 5,870 52.30 7.00 1,319.08 57.61 31.58 16.78 19.08 0.83 27.33 3.78 0.63 22.15 0.00 4.73 73.12 0.00 
B-23 372 0.087 3,160 50.80 10.10 1,255.23 70.25 27.78 20.37 13.48 1.53 33.55 1.19 2.10 51.20 0.00 0.00 48.80 0.00 
B-24 330 0.072 1,700 44.30 13.00 1,525.77 79.16 21.83 4.33 6.51 3.89 58.16 5.25 0.04 0.00 0.00 0.00 50.76 49.24 

All the meteorological variables (rainfall, temperature and relative humidity) are annual averages over the study period (2000–2013). The 24 LULC classes were grouped into seven categories: dryland cropland and pasture (DC), irrigated cropland and pasture (IC), cropland/grassland mosaic (CG), cropland/woodland mosaic (CW), savanna, forest and other. Similar to LULC, soil data were also grouped into five categories: Loam (L), Clay (C), Clay Loam (CL), Sandy Loam (SL) and Sandy Clay Loam (SCL).

Five regionalization methods (global averaging, regression, spatial proximity, behavioral similarity and ANN) were evaluated in terms of their abilities to improve SWAT predictions in ungauged basins of peninsular India (Figure 2). Analyses were conducted using results of a previous modeling study (Soni 2018) that reported 13 calibrated SWAT parameters (Table 4), namely: SCS runoff curve number for moisture condition II (CN2), available water capacity of the soil layer (SOL_AWC), compensation factor for soil evaporation (ESCO), compensation factor for plant uptake (EPCO), base flow recession constant (ALPHA_BF), groundwater delay (GW_DELAY), threshold water level in shallow aquifer for base flow to occur (GWQMN), groundwater revap coefficient (GW_REVAP), threshold depth of water in shallow aquifer for revap to occur (REVAPMN), Manning's n value for overland flow (OV_N), average slope steepness (HRU_SLP), average slope length (SLSUBBSN) and surface runoff lag coefficient (SURLAG). Calibration (2003–2009) and validation (2010–2013) were performed using daily observed discharge data after removing 3 years of warm up period (2000–2002). Out of the 13 model parameters, eight were found to be sensitive in 70% of the basins (Soni 2018).

Table 4

Calibrated model parameters for all the basins

Basinr_CN2r_SOL_AWCv_LPHA_BFv_EPCOv_ESCOv_GWQMNv_GW_DELAYv_GW_REVAPr_HRU_SLPr_OV_Nv_REVAPMNr_SLSUBBSNv_SURLAG
mmdaysmmdays
B-1 − 0.013 0.373 0.079 0.514 0.434 1,867 16.00 0.11 − 0.100 4.004 368.13 0.148 10.29 
B-2 − 0.210 0.380 0.220 0.921 0.560 4,203 17.00 0.08 0.405 3.660 292.59 − 0.233 6.10 
B-3 − 0.195 − 0.460 0.350 0.256 0.500 3,195 15.00 0.05 0.059 2.200 378.29 − 0.070 3.47 
B-4 − 0.350 − 0.167 0.413 0.611 0.706 2,109 16.00 0.06 − 0.309 3.325 322.63 0.248 4.05 
B-5 − 0.310 0.300 0.260 0.822 0.450 396 45.90 0.15 0.223 2.070 383.94 − 0.026 4.36 
B-6 − 0.240 0.491 0.300 0.766 0.410 2,192 31.00 0.08 0.196 4.000 435.06 − 0.186 9.90 
B-7 − 0.160 0.370 0.357 0.415 0.648 143 18.78 0.09 − 0.107 3.862 251.86 0.054 2.44 
B-8 − 0.157 0.331 0.190 0.706 0.480 196 19.15 0.16 0.071 2.614 193.01 0.013 5.91 
B-9 − 0.353 0.272 0.570 0.266 0.330 183 171.00 0.12 − 0.281 5.772 204.00 0.010 11.94 
B-10 − 0.182 0.235 0.185 0.232 0.720 278 12.69 0.15 0.128 1.863 489.32 − 0.091 6.05 
B-11 − 0.320 0.100 0.260 0.422 0.480 1,263 79.62 0.04 0.880 4.100 417.19 0.163 6.35 
B-12 0.094 0.484 0.240 0.474 0.510 500 170.88 0.03 − 0.131 4.870 410.13 0.187 2.26 
B-13 − 0.061 0.300 0.390 0.310 0.500 391 88.00 0.08 0.113 5.900 408.75 0.160 5.18 
B-14 − 0.061 0.330 0.330 0.357 0.630 464 94.02 0.11 0.075 5.150 490.31 0.215 1.81 
B-15 − 0.110 0.530 0.260 0.611 0.540 1,817 13.84 0.05 − 0.068 3.180 322.63 0.136 4.05 
B-16 − 0.380 0.410 0.300 0.345 0.450 1,699 65.00 0.07 0.074 2.660 351.87 0.078 3.49 
B-17 − 0.370 − 0.580 0.280 0.829 0.910 230 23.00 0.04 0.221 4.000 421.08 0.024 2.78 
B-18 − 0.190 0.204 0.220 0.555 0.448 1,683 7.51 0.09 − 0.216 3.730 284.43 0.165 6.73 
B-19 − 0.167 − 0.040 0.438 0.164 0.560 871 20.00 0.12 0.146 6.731 357.07 − 0.124 4.17 
B-20 − 0.318 − 0.070 0.270 0.639 0.460 368 20.00 0.04 − 0.158 4.090 488.72 0.099 6.02 
B-21 − 0.070 0.355 0.070 0.670 0.620 1,494 22.19 0.13 − 0.170 4.775 471.84 0.190 2.75 
B-22 − 0.135 0.334 0.240 0.894 0.485 251 45.73 0.14 − 0.013 4.820 236.85 0.023 5.13 
B-23 − 0.043 0.310 0.180 0.533 0.580 85 30.60 0.09 − 0.019 4.734 372.82 0.090 7.71 
B − 24 − 0.015 0.279 0.040 0.566 0.665 1,747 17.63 0.08 0.835 4.230 260.09 − 0.084 8.81 
Basinr_CN2r_SOL_AWCv_LPHA_BFv_EPCOv_ESCOv_GWQMNv_GW_DELAYv_GW_REVAPr_HRU_SLPr_OV_Nv_REVAPMNr_SLSUBBSNv_SURLAG
mmdaysmmdays
B-1 − 0.013 0.373 0.079 0.514 0.434 1,867 16.00 0.11 − 0.100 4.004 368.13 0.148 10.29 
B-2 − 0.210 0.380 0.220 0.921 0.560 4,203 17.00 0.08 0.405 3.660 292.59 − 0.233 6.10 
B-3 − 0.195 − 0.460 0.350 0.256 0.500 3,195 15.00 0.05 0.059 2.200 378.29 − 0.070 3.47 
B-4 − 0.350 − 0.167 0.413 0.611 0.706 2,109 16.00 0.06 − 0.309 3.325 322.63 0.248 4.05 
B-5 − 0.310 0.300 0.260 0.822 0.450 396 45.90 0.15 0.223 2.070 383.94 − 0.026 4.36 
B-6 − 0.240 0.491 0.300 0.766 0.410 2,192 31.00 0.08 0.196 4.000 435.06 − 0.186 9.90 
B-7 − 0.160 0.370 0.357 0.415 0.648 143 18.78 0.09 − 0.107 3.862 251.86 0.054 2.44 
B-8 − 0.157 0.331 0.190 0.706 0.480 196 19.15 0.16 0.071 2.614 193.01 0.013 5.91 
B-9 − 0.353 0.272 0.570 0.266 0.330 183 171.00 0.12 − 0.281 5.772 204.00 0.010 11.94 
B-10 − 0.182 0.235 0.185 0.232 0.720 278 12.69 0.15 0.128 1.863 489.32 − 0.091 6.05 
B-11 − 0.320 0.100 0.260 0.422 0.480 1,263 79.62 0.04 0.880 4.100 417.19 0.163 6.35 
B-12 0.094 0.484 0.240 0.474 0.510 500 170.88 0.03 − 0.131 4.870 410.13 0.187 2.26 
B-13 − 0.061 0.300 0.390 0.310 0.500 391 88.00 0.08 0.113 5.900 408.75 0.160 5.18 
B-14 − 0.061 0.330 0.330 0.357 0.630 464 94.02 0.11 0.075 5.150 490.31 0.215 1.81 
B-15 − 0.110 0.530 0.260 0.611 0.540 1,817 13.84 0.05 − 0.068 3.180 322.63 0.136 4.05 
B-16 − 0.380 0.410 0.300 0.345 0.450 1,699 65.00 0.07 0.074 2.660 351.87 0.078 3.49 
B-17 − 0.370 − 0.580 0.280 0.829 0.910 230 23.00 0.04 0.221 4.000 421.08 0.024 2.78 
B-18 − 0.190 0.204 0.220 0.555 0.448 1,683 7.51 0.09 − 0.216 3.730 284.43 0.165 6.73 
B-19 − 0.167 − 0.040 0.438 0.164 0.560 871 20.00 0.12 0.146 6.731 357.07 − 0.124 4.17 
B-20 − 0.318 − 0.070 0.270 0.639 0.460 368 20.00 0.04 − 0.158 4.090 488.72 0.099 6.02 
B-21 − 0.070 0.355 0.070 0.670 0.620 1,494 22.19 0.13 − 0.170 4.775 471.84 0.190 2.75 
B-22 − 0.135 0.334 0.240 0.894 0.485 251 45.73 0.14 − 0.013 4.820 236.85 0.023 5.13 
B-23 − 0.043 0.310 0.180 0.533 0.580 85 30.60 0.09 − 0.019 4.734 372.82 0.090 7.71 
B − 24 − 0.015 0.279 0.040 0.566 0.665 1,747 17.63 0.08 0.835 4.230 260.09 − 0.084 8.81 

Significantly sensitive parameters are shown in bold. The symbols ‘r’ and ‘v’ with each parameter denote the relative and replace method of parameter calibration.

Figure 2

The flowchart of methodology.

Figure 2

The flowchart of methodology.

Close modal
Regionalization was performed using these 13 model parameters for the 24 selected river basins. For each basin, the parameters were estimated using all the five regionalization methods. The estimated parameters were used to simulate daily streamflow for the entire period (2000–2013) and their performance was compared with the default and calibrated parameter sets using three error statistics.

Here, Qm and Qo are modeled and observed discharge, respectively.

A brief description of the SWAT model is provided in the ‘SWAT model description’ section. Methodologies used to regionalize model parameters are given in the ‘Global averaging’, ‘Regression’ and ‘Spatial proximity’ sections.

SWAT model description

SWAT is developed jointly by the United States Department of Agriculture-Agricultural Research Services (USDA-ARS) and Agricultural Experiment Station in Temple, Texas (Arnold et al. 1998; Arnold et al. 2001; Arnold & Fohrer 2005). It is a physically based, continuous-time, long-term simulation, lumped parameter, deterministic model, which originated from agricultural models. There are eight major computational components in SWAT: hydrology, weather, sedimentation, soil temperature, crop growth, nutrients, pesticides and agricultural management. The water balance equation used in SWAT is given by:
(1)
where SWt is the final soil water content (mm); SW0 is the initial soil water content (mm); t is time (days); Ri is the amount of precipitation on day i (mm); Qisurf is the amount of surface runoff on day i (mm); Eia is the amount of evapotranspiration on day i (mm); Wiseep is water entering the vadose zone from soil profile on day i (mm); Qigw is the amount of return flow on day i (mm).
For computing surface runoff, SWAT uses Soil Conservation Service Curve Number (CN) method as shown below:
(2)
where S is the retention parameter.
In Equation (2), the retention parameter is related to CN as:
(3)

Global averaging

It involved computing the mean of each parameter listed in Table 4, over 23 donor basins (excluding the test basin). The estimated mean value is treated as the parameter for the test basin. The model simulations were carried out for all the 24 basins, considering each basin as a test basin individually (He et al. 2011).

Regression

The backward elimination method of stepwise regression was employed to determine functional relationships between the basin attributes (Table 3) and model parameters (Table 4). The procedure starts by including all the basin attributes in the regression model. In each step, a basin attribute is considered for subtraction from the set of variables based on the significance of its coefficient. If the coefficient of the attribute was found to be insignificant (t-stat < 1), the variable was removed from the regression. If the procedure results in the removal of all the variables, the corresponding globally averaged value obtained in the ‘Global averaging’ section was used (He et al. 2011).

Spatial proximity

Geographical distance () between the centroids of the basins is used as a similarity measure (Equation (4)). For each test basin, parameters of the donor basin having the least geographical distance were selected and model simulations were carried out using parameters of that similar basin (He et al. 2011):
(4)
where LAT is the latitude of the centroid of the basin; LON is the longitude of the centroid of test basin; t is the index for text basin and d is the index for donor basin.

Behavioral similarity

This method is similar to spatial proximity (Section ‘Spatial proximity’), except that the distance between basins was calculated using basin attributes (see Table 3) instead of geographical coordinates (He et al. 2011).
(5)
where
(6)

Artificial neural network

The SWAT model parameters for each test basin were predicted using an ANN model trained using calibrated model parameters of the remaining 23 basins as outputs (Table 4) and their corresponding basin attributes as inputs (Table 3).

ANN was employed to predict each SWAT model parameter for each test basin using the model parameters of the remaining 23 basins as outputs (Table 4) and basin attributes (of corresponding basins; Table 3) as inputs. Thus, there were 18 neurons in the input layer (corresponding to 18 basin attributes) and 1 in the output layer (for each SWAT model parameter). The input and output data were normalized between 0 and 1. Apart from that, the ANN architecture consisted of two hidden layers with a varying number of neurons (up to 25 and 30 for the first and the second hidden layers, respectively). The ANN models were trained by keeping 60% data for training, 20% for testing and the remaining 20% for the validation purpose. The number of hidden layers were limited to two, to strike a balance between the lesser number of training data sets (23 basins) and complexity of the process (Rogers & Dowla 1994; Heuvelmans et al. 2006).

The best architecture having the highest average coefficient of determination (R2) (over training, testing and validation period) was selected for predicting each model parameter for each test basin. A total of 312 (24 × 13) ANN models were trained. If any ANN model had a root mean square error (RMSE) value more than 0.2 or the coefficient of determination (R2) was less than 0.50 (whether in training, testing or validation), then the model was not used to predict the parameter, instead the corresponding globally averaged value was used (Section ‘Global averaging’). The various options selected for training the ANN model are given in Table 5.

Table 5

Training parameters of ANN

S. No.OptionValue
Type of network Feedforward backpropogation network 
Transfer functions Tan-sigmoid (hidden) and Log-sigmoid (input) 
Backpropagation network training function Levenberg–Marquardt backpropagation 
Backpropagation weight/bias learning function Gradient descent weight and bias learning function 
Performance function Mean square error (MSE) 
Learning Rate 10−4 
Maximum fails in Validation 50 
Minimum learning Gradient 10−10 
S. No.OptionValue
Type of network Feedforward backpropogation network 
Transfer functions Tan-sigmoid (hidden) and Log-sigmoid (input) 
Backpropagation network training function Levenberg–Marquardt backpropagation 
Backpropagation weight/bias learning function Gradient descent weight and bias learning function 
Performance function Mean square error (MSE) 
Learning Rate 10−4 
Maximum fails in Validation 50 
Minimum learning Gradient 10−10 

For each of the five regionalization methods, first the SWAT model parameters were estimated for each basin, and then their performance was compared with the corresponding calibrated model for simulating daily streamflow during the study period.

Estimation of SWAT parameters

For the global averaging, the mean of each calibrated parameter was computed over all donor basins after removing the test basin. For spatial proximity, the closest basins were identified based on location and for behavioral similarity, the closest basins were identified based on basin attributes (see Table 6). It can be seen that out of the 24 basins, nine were having common donor basins for both spatial proximity and behavioral similarity methods.

Table 6

Similar basins based on the spatial proximity (SP) and behavioral similarity (BS)

BASINSPBSBASINSPBSBASINSPBS
B-1 B-21 B-21 B-9 B-6 B-16 B-17 B-4 B-20 
B-2 B-18 B-7 B-10 B-5 B-7 B-18 B-5 B-5 
B-3 B-17 B-4 B-11 B-13 B-13 B-19 B-1 B-13 
B-4 B-20 B-3 B-12 B-23 B-22 B-20 B-4 B-4 
B-5 B-10 B-16 B-13 B-14 B-14 B-21 B-1 B-1 
B-6 B-16 B-15 B-14 B-13 B-13 B-22 B-23 B-23 
B-7 B-8 B-14 B-15 B-7 B-6 B-23 B-22 B-22 
B-8 B-7 B-15 B-16 B-7 B-5 B-24 B-21 B-2 
BASINSPBSBASINSPBSBASINSPBS
B-1 B-21 B-21 B-9 B-6 B-16 B-17 B-4 B-20 
B-2 B-18 B-7 B-10 B-5 B-7 B-18 B-5 B-5 
B-3 B-17 B-4 B-11 B-13 B-13 B-19 B-1 B-13 
B-4 B-20 B-3 B-12 B-23 B-22 B-20 B-4 B-4 
B-5 B-10 B-16 B-13 B-14 B-14 B-21 B-1 B-1 
B-6 B-16 B-15 B-14 B-13 B-13 B-22 B-23 B-23 
B-7 B-8 B-14 B-15 B-7 B-6 B-23 B-22 B-22 
B-8 B-7 B-15 B-16 B-7 B-5 B-24 B-21 B-2 

Figure 3 shows the percentage difference in LULC, soil, climate and all 18 basin properties at each basin from their respective global averages. Overall, the basins located in the central and east-central India (located in the states of Chhattisgarh and Odisha) dominated globally averaged basin characteristics, as seen from the large blue region there. Moreover, as shown by dark red colors, the coastal basins (B-3, B-4, B17 and B-20) have different LULC and overall basin attributes as compared with other basins. The basins located in the Far East (B-12, B-22 and B-23) have larger differences in terms of soil properties, but they are similar in overall terms. This may have an impact while selecting the basin based on behavioral similarity, as the basin may be similar overall but some model parameters may be dominated by particular catchment descriptors, resulting in the different performance of the method, depending on the parameter sensitivity. For example, the most sensitive parameter CN2 depends only on soil and land use characteristics, but while using behavioral similarity, the donor basin is selected based on overall similarity. Thus, it is possible that the donor basin is similar in terms of overall similarity (due to meteorological characteristics) but has different soil and LULC properties, and thus a different CN2 value.

Figure 3

Spatial plot showing the percentage difference in LULC, soil, climate and all 18 basin attributes from the global average of 24 basins. Please refer to the online version of this paper to see this figure in color: https://doi.org/10.2166/wcc.2021.298|0|0|2021.

Figure 3

Spatial plot showing the percentage difference in LULC, soil, climate and all 18 basin attributes from the global average of 24 basins. Please refer to the online version of this paper to see this figure in color: https://doi.org/10.2166/wcc.2021.298|0|0|2021.

Close modal
For regression-based regionalization, regression equations were developed for all the parameters using basin attributes of 23 donor basins after excluding the test basin. For non-sensitive parameters, the globally averaged parameter values were used. The general regression equations developed by considering all 24 basins are given in Equations (7)–(14) (description and ranges of all the basin attributes used for generating these equations are given in Table 3). The soil category, Loam (L) is an important explanatory variable that is present in all the equations, except r_CN2. The irrigated cropland and pasture (IC) is the next important explanatory variable present in five of the equations. It can also be noted that the maximum temperature (Tmax) and the land-use category, dryland cropland and pasture (DC), are not present in any of these regression equations.
(7)
(8)
(9)
(10)
(11)
(12)
(13)
(14)

Figure 4 shows the scatter plots of calibrated and regression-based parameters obtained from the regional regression equations. For all the parameters, regression equations were obtained with R2 more than 0.7, except r_CN2 (R2 = 0.497) and v_GWQMN (R2 = 0.332). Equations have the best performance for r_SOL_AWC, v_GW_DELAY and r_OV_N with R2 more than 0.9. The performance of regression is better than a previous study by Swain & Patra (2017), wherein they found minimum and maximum R2 values of 0.29 and 0.73, respectively.

Figure 4

Scatter plots between the calibrated and regression-based values of model parameters.

Figure 4

Scatter plots between the calibrated and regression-based values of model parameters.

Close modal

ANN models were trained for each basin and model parameter by varying the number of neurons in hidden layers. The architecture that gave the highest average coefficient of determination (in training, testing and validation) was chosen to predict the SWAT model parameter for each test basin. The number of neurons in the best ANN model corresponding to each basin and parameters are shown in Table 7. The first value in each cell represents the number of neurons in the first hidden layer and the second value (after ‘’) is the number of neurons in the second hidden layer.

Table 7

Best ANN architecture for each basin and parameter (number of neurons in first and second hidden layers)

BASINCN2SOL_AWCALPHA_BFEPCOESCOGWQMNGW_DELAYGW_REVAPHRU_SLPONREVAPMNSLSUBBSNSURLAG
B-1 9 × 20 22 × 15 2 × 19 8 × 18 14 × 6 4 × 1 2 × 2 25 × 4 3 × 21 7 × 23 – 20 × 26 – 
B-2 16 × 1 12 × 18 5 × 30 1 × 20 18 × 18 15 × 28 12 × 5 11 × 11 8 × 16 19 × 10 – 23 × 19 21 × 4 
B-3 25 × 29 9 × 15 14 × 2 11 × 14 17 × 21 3 × 11 12 × 2 5 × 1 16 × 4 22 × 1 19 × 27 10 × 7 – 
B-4 6 × 29 18 × 1 18 × 8 22 × 28 21 × 15 18 × 1 3 × 12 4 × 25 17 × 18 13 × 3 1 × 2 6 × 10 16 × 28 
B-5 2 × 22 21 × 27 9 × 15 20 × 16 9 × 23 12 × 7 5 × 4 14 × 1 13 × 29 19 × 25 4 × 18 16 × 13 – 
B-6 4 × 27 5 × 6 7 × 20 5 × 30 13 × 3 6 × 1 1 × 3 12 × 26 14 × 7 16 × 5 8 × 27 21 × 17 – 
B-7 1 × 23 24 × 24 14 × 22 18 × 17 11 × 27 7 × 2 20 × 1 8 × 8 10 × 5 25 × 21 18 × 13 4 × 24 – 
B-8 3 × 3 15 × 14 15 × 28 17 × 12 20 × 4 1 × 11 5 × 2 7 × 7 1 × 7 6 × 2 25 × 25 14 × 9 6 × 14 
B-9 18 × 12 8 × 2 7 × 22 23 × 13 2 × 9 11 × 6 20 × 1 4 × 18 – 10 × 7 – 13 × 26 – 
B-10 13 × 28 25 × 11 19 × 25 13 × 14 15 × 30 – 14 × 11 16 × 5 12 × 14 13 × 28 21 × 19 16 × 17 – 
B-11 13 × 26 20 × 5 6 × 6 14 × 7 8 × 20 10 × 9 25 × 1 6 × 17 16 × 10 17 × 30 17 × 21 4 × 27 10 × 3 
B-12 7 × 13 16 × 24 5 × 2 5 × 4 14 × 12 13 × 3 2 × 27 5 × 3 4 × 5 23 × 5 – 10 × 9 – 
B-13 11 × 17 19 × 24 22 × 25 16 × 17 17 × 1 15 × 6 11 × 6 1 × 1 – 13 × 16 – 3 × 9 1 × 15 
B-14 17 × 14 17 × 1 23 × 2 25 × 21 12 × 16 14 × 27 2 × 19 13 × 4 16 × 11 24 × 16 3 × 18 9 × 29 – 
B-15 17 × 18 13 × 27 16 × 2 19 × 27 23 × 15 12 × 9 24 × 2 15 × 20 7 × 3 24 × 8 – 13 × 23 – 
B-16 10 × 14 16 × 3 15 × 21 4 × 5 23 × 19 3 × 30 23 × 6 8 × 26 2 × 18 16 × 14 11 × 21 2 × 17 18 × 8 
B-17 21 × 14 13 × 15 11 × 10 2 × 25 23 × 20 24 × 19 23 × 1 3 × 22 25 × 1 24 × 6 1 × 26 9 × 23 – 
B-18 19 × 23 14 × 1 25 × 12 16 × 3 7 × 4 9 × 18 3 × 6 17 × 2 14 × 15 15 × 14 17 × 15 10 × 7 – 
B-19 2 × 6 13 × 16 6 × 28 13 × 3 24 × 14 14 × 13 25 × 8 5 × 19 17 × 3 23 × 21 25 × 23 12 × 17 9 × 2 
B-20 18 × 25 7 × 8 16 × 10 – 22 × 28 3 × 11 4 × 3 4 × 24 5 × 5 20 × 11 – 11 × 24 – 
B-21 13 × 24 23 × 8 21 × 2 – 6 × 19 18 × 21 14 × 7 4 × 27 23 × 9 7 × 26 25 × 1 25 × 23 1 × 14 
B-22 23 × 16 7 × 4 24 × 11 2 × 10 24 × 20 21 × 20 19 × 2 7 × 20 17 × 21 10 × 22 8 × 25 17 × 18 – 
B-23 10 × 1 17 × 7 7 × 8 17 × 2 2 × 22 – 2 × 29 24 × 3 – 21 × 22 – – – 
B-24 2 × 9 24 × 5 8 × 25 6 × 9 9 × 4 24 × 26 3 × 1 4 × 7 – 3 × 9 2 × 13 4 × 6 5 × 25 
BASINCN2SOL_AWCALPHA_BFEPCOESCOGWQMNGW_DELAYGW_REVAPHRU_SLPONREVAPMNSLSUBBSNSURLAG
B-1 9 × 20 22 × 15 2 × 19 8 × 18 14 × 6 4 × 1 2 × 2 25 × 4 3 × 21 7 × 23 – 20 × 26 – 
B-2 16 × 1 12 × 18 5 × 30 1 × 20 18 × 18 15 × 28 12 × 5 11 × 11 8 × 16 19 × 10 – 23 × 19 21 × 4 
B-3 25 × 29 9 × 15 14 × 2 11 × 14 17 × 21 3 × 11 12 × 2 5 × 1 16 × 4 22 × 1 19 × 27 10 × 7 – 
B-4 6 × 29 18 × 1 18 × 8 22 × 28 21 × 15 18 × 1 3 × 12 4 × 25 17 × 18 13 × 3 1 × 2 6 × 10 16 × 28 
B-5 2 × 22 21 × 27 9 × 15 20 × 16 9 × 23 12 × 7 5 × 4 14 × 1 13 × 29 19 × 25 4 × 18 16 × 13 – 
B-6 4 × 27 5 × 6 7 × 20 5 × 30 13 × 3 6 × 1 1 × 3 12 × 26 14 × 7 16 × 5 8 × 27 21 × 17 – 
B-7 1 × 23 24 × 24 14 × 22 18 × 17 11 × 27 7 × 2 20 × 1 8 × 8 10 × 5 25 × 21 18 × 13 4 × 24 – 
B-8 3 × 3 15 × 14 15 × 28 17 × 12 20 × 4 1 × 11 5 × 2 7 × 7 1 × 7 6 × 2 25 × 25 14 × 9 6 × 14 
B-9 18 × 12 8 × 2 7 × 22 23 × 13 2 × 9 11 × 6 20 × 1 4 × 18 – 10 × 7 – 13 × 26 – 
B-10 13 × 28 25 × 11 19 × 25 13 × 14 15 × 30 – 14 × 11 16 × 5 12 × 14 13 × 28 21 × 19 16 × 17 – 
B-11 13 × 26 20 × 5 6 × 6 14 × 7 8 × 20 10 × 9 25 × 1 6 × 17 16 × 10 17 × 30 17 × 21 4 × 27 10 × 3 
B-12 7 × 13 16 × 24 5 × 2 5 × 4 14 × 12 13 × 3 2 × 27 5 × 3 4 × 5 23 × 5 – 10 × 9 – 
B-13 11 × 17 19 × 24 22 × 25 16 × 17 17 × 1 15 × 6 11 × 6 1 × 1 – 13 × 16 – 3 × 9 1 × 15 
B-14 17 × 14 17 × 1 23 × 2 25 × 21 12 × 16 14 × 27 2 × 19 13 × 4 16 × 11 24 × 16 3 × 18 9 × 29 – 
B-15 17 × 18 13 × 27 16 × 2 19 × 27 23 × 15 12 × 9 24 × 2 15 × 20 7 × 3 24 × 8 – 13 × 23 – 
B-16 10 × 14 16 × 3 15 × 21 4 × 5 23 × 19 3 × 30 23 × 6 8 × 26 2 × 18 16 × 14 11 × 21 2 × 17 18 × 8 
B-17 21 × 14 13 × 15 11 × 10 2 × 25 23 × 20 24 × 19 23 × 1 3 × 22 25 × 1 24 × 6 1 × 26 9 × 23 – 
B-18 19 × 23 14 × 1 25 × 12 16 × 3 7 × 4 9 × 18 3 × 6 17 × 2 14 × 15 15 × 14 17 × 15 10 × 7 – 
B-19 2 × 6 13 × 16 6 × 28 13 × 3 24 × 14 14 × 13 25 × 8 5 × 19 17 × 3 23 × 21 25 × 23 12 × 17 9 × 2 
B-20 18 × 25 7 × 8 16 × 10 – 22 × 28 3 × 11 4 × 3 4 × 24 5 × 5 20 × 11 – 11 × 24 – 
B-21 13 × 24 23 × 8 21 × 2 – 6 × 19 18 × 21 14 × 7 4 × 27 23 × 9 7 × 26 25 × 1 25 × 23 1 × 14 
B-22 23 × 16 7 × 4 24 × 11 2 × 10 24 × 20 21 × 20 19 × 2 7 × 20 17 × 21 10 × 22 8 × 25 17 × 18 – 
B-23 10 × 1 17 × 7 7 × 8 17 × 2 2 × 22 – 2 × 29 24 × 3 – 21 × 22 – – – 
B-24 2 × 9 24 × 5 8 × 25 6 × 9 9 × 4 24 × 26 3 × 1 4 × 7 – 3 × 9 2 × 13 4 × 6 5 × 25 

The ‘–’ represents the case when no satisfactory ANN model was found.

Box plots showing the total number of neurons (sum of neurons in the two hidden layers) needed in the best ANN architecture are presented in Figure 5. We can see that on average, around 30 neurons are needed for each variable. For the variables corresponding to groundwater processes (GWQMN, GW_DELAY, GW_REVAP), fewer neurons gave better results. The best ANN models were used to predict the SWAT model parameters for each basin and for those cases when a suitable ANN model was not found, a corresponding globally averaged value of the parameter was used.

Figure 5

Total number of hidden layer neurons for the best ANN architecture for eight sensitive parameters.

Figure 5

Total number of hidden layer neurons for the best ANN architecture for eight sensitive parameters.

Close modal

Figure 6 shows the performance of ANN in terms of predicting the eight sensitive parameters, during calibration, validation and testing. We can see that during calibration, except GWQMN, all other parameters have a high value of the coefficient of determination. Overall, ANN shows satisfactory performance both in validation and testing. The model has significantly better performance for SOL_AWC and GW_DELAY.

Figure 6

Coefficient of determination of the ANN model in training, testing and validation.

Figure 6

Coefficient of determination of the ANN model in training, testing and validation.

Close modal

The model parameters estimated from each method were compared with the calibrated parameters. The percentage difference from the calibrated parameter for the sensitive parameters of each basin was calculated and the results are shown in Figure 7. We can see that for all the parameters, there is a significant difference in the Inter Quartile Range (IQR) between different methods. The model parameters estimated from the ANN have comparatively larger width, which can be attributed to the small sample size used for training the ANN models. These results question the applicability of ANN for regionalization studies that typically have data from the limited number of gauging stations for training. Overall, the model parameters estimated from the regression-based regionalization have a smaller IQR, but the median shows a larger deviation from the calibrated values as compared with other methods.

Figure 7

Box plots showing the percentage difference from the calibrated values of model parameters in different regionalization methods.

Figure 7

Box plots showing the percentage difference from the calibrated values of model parameters in different regionalization methods.

Close modal

Comparison of regionalization methods

Once the model parameters were estimated by all the regionalization methods, their performance was compared in terms of predicting the observed streamflow at daily timescale. Figure 8(a) and 8(b) presents the box plots of MNS values obtained from different regionalization methods for the calibration period and the entire simulation period, respectively. The percentage improvement (with respect to the default model parameters) in the streamflow prediction by these regionalization methods is shown in Figure 8(c). Furthermore, to compare the performance of these methods in simulating peak flows, the analyses were done for only those days for which streamflows have 5% exceedance probability. The % improvement (in terms of RMSE) for predicting such flows with respect to default set is shown in Figure 8(d).

Figure 8

Performance comparison of the SWAT model having parameters obtained from calibration and different regionalization methods. (a) Box plots showing MNS values obtained from different regionalization methods for the calibration period; (b) Same as (a) but for the entire simulation period. (c) The box plots showing the percentage improvement in the simulation of daily streamflow during the entire period and (d) Same as (c) but only for the peak streamflows.

Figure 8

Performance comparison of the SWAT model having parameters obtained from calibration and different regionalization methods. (a) Box plots showing MNS values obtained from different regionalization methods for the calibration period; (b) Same as (a) but for the entire simulation period. (c) The box plots showing the percentage improvement in the simulation of daily streamflow during the entire period and (d) Same as (c) but only for the peak streamflows.

Close modal

As expected, the calibrated model shows the highest median MNS values (nearly 0.55) for both calibration and the entire simulation period. This is relatively smaller than the globally averaged median value of 0.75, found from 33 studies reviewed by Guo et al. (2020). The poor performance of the calibrated model can be attributed to unreliable or erroneous data for Indian basins, which is also highlighted by a smaller NSE value of 0.52 for Indian basins by Guo et al. (2020).

The mean and the median MNS values do not show major differences between different regionalization methods for both the periods. However, there are some differences in terms of box width and whisker limits. Compared with other methods, the spatial proximity has a larger box width and for the calibration period, the whiskers even go beyond the zero MNS value for the ANN method. In terms of % improvement, the regression and behavioral similarity show slightly better performance, with regression having a leading edge with a higher median value. Previous studies have also reported better performance of regression over other methods (Waseem et al. 2015; Pugliese et al. 2016), but globally the regression shows slightly poor performance (NSE = 0.66) as compared to behavioral similarity (NSE = 0.69) and spatial proximity (NSE = 0.686) (Guo et al. 2020).

On comparing the performance based on peak flow prediction, it can seen that both the ANN, and the regression-based regionalization are comparatively better in predicting the peak streamflow (regression median = 7.99% and ANN median = 10.03%). Moreover, behavioral similarity shows marginally better performance (median = 4.82%) than spatial proximity (2.16%). Thus, these methods (which involve physical parameters) are able to capture peak flows more accurately.

Figure 9 shows spatial plots of percentage improvement (in terms of daily MNS) with respect to default parameters by all the regionalization methods along with calibration. The larger areas covered by red indicate significant improvement in streamflow prediction by regionalization methods. However, some difference over the regional level can be observed. In the basins located near the Western Ghats (B-3, B-4, B-17 and B-20), the behavioral similarity and regression method outperformed other methods. For these basins, ANN and global averaging both showed relatively poorer performance.

Figure 9

Spatial plot showing percentage improvement (in terms of daily MNS) from the default simulation by using SWAT parameters obtained from (a) calibration (b) global averaging, (c) regression, (d) spatial proximity, (e) behavioral similarity and (f) ANN. Please refer to the online version of this paper to see this figure in color: https://doi.org/10.2166/wcc.2021.298|0|0|2021.

Figure 9

Spatial plot showing percentage improvement (in terms of daily MNS) from the default simulation by using SWAT parameters obtained from (a) calibration (b) global averaging, (c) regression, (d) spatial proximity, (e) behavioral similarity and (f) ANN. Please refer to the online version of this paper to see this figure in color: https://doi.org/10.2166/wcc.2021.298|0|0|2021.

Close modal

We can see that in the basins for which calibration shows higher improvement, the regionalization methods also show comparatively higher improvement (i.e. basins located in the states of Chhattisgarh and Odisha). For the basin located at the most extreme distance (B-15), none of the methods (including calibration) showed any significant improvement. This highlights the importance of homogeneous catchment groups in regionalization (Swain & Patra 2019).

There are some exceptions such as for B-12, for which all the methods show significant improvement except for the spatial proximity method. Similarly for B-22, all the methods show improvement except the similarity methods (spatial proximity and behavioral similarity). These (B-12 and B-22) are the basins having diverse characteristics of LULC and soil categories (see Table 3 and Figure 3). Thus, we must be careful while selecting the range of basin attributes for regionalization. For the basins B-7 and B-23, none of the regionalization methods are able to show any improvement probably due to dissimilar climatic conditions (lower maximum temperature and higher relative humidity) and soil categories (Largest % of Loam for B-23), respectively.

In this study, parameters of the SWAT model for 24 river basins in peninsular India obtained from five different methods of regionalization (global averaging, regression, spatial proximity, behavioral similarity and ANN) are compared with the default and calibrated parameter sets.

Out of 24 basins, 9 basins were such that the closest basins are the same for both spatial proximity and behavioral similarity methods. All the regression equations developed during regionalization showed an R2 value greater than 0.7 except for r_CN2 (R2 = 0.497) and v_GWQMN (R2 = 0.332), with equations of r_SOL_AWC, v_GW_DELAY and r_OV_N having R2 greater than 0.9. For the best ANN architectures, nearly 30 neurons were needed in hidden layers for each variable. For the variables corresponding to groundwater processes (GWQMN, GW_DELAY, GW_REVAP), comparatively fewer neurons gave better results.

The median MNS value for all the calibrated models (about 0.55) is relatively smaller than the globally averaged median value of 0.75, found from 33 studies reviewed by Guo et al. (2020). This can be attributed to unreliable or erroneous data for Indian basins.

Compared with other methods, the spatial proximity has a larger box width and for the calibration period and for the ANN, the whiskers go below the zero MNS mark. In terms of % improvement, the regression and behavioral similarity show slightly better performance, with regression having a leading edge with a higher median value. On comparing the performance based on peak flow prediction, we found that methods (regression, behavioral similarity and ANN) which involve physical parameters are able to capture peak flows more accurately compared with other methods of regionalization.

On analyzing the regional patterns, we found that the basins for which calibration shows higher improvement, the regionalization methods also exhibit improved performance. For the basin located at the most extreme distance, none of the methods (including calibration) showed any significant improvement. For some of the basins (e.g. basins B-7 and B-23) having dissimilar climatic or soil characteristics, none of the regionalization methods are able to show any improvement. Based on the above analysis, the conclusions of the present study are:

  • 1.

    The performance of the regionalization method (and calibration) is limited by the unreliable or erroneous data at the basin.

  • 2.

    The regression method outperforms other regionalization methods in terms of predicting the daily as well as peak discharges. Thus, the equations developed in the present study can be utilized to predict SWAT parameters of basins located in the vicinity of the study area.

  • 3.

    The basins located at far away distance or having diverse characteristics should be avoided.

  • 4.

    The ANN should also be avoided for regionalization, as in the absence of sufficient training data, the performance is not adequate.

Future scope

Based on the results obtained from our study, we found that there is still a lot of future scope that must be explored in the field of regionalization, specially for Indian basins. The most important point to consider in future studies should be to include a sufficiently larger number of basins. With a larger number of basins, homogeneous groups should be formed and regionalization should be done according to that. Moreover, future studies can also explore nonlinear and nonparametric regression methods.

The authors thank IMD and TRMM for the precipitation data. We also thank NSRDB for providing various meteorological data at high resolution. We gratefully acknowledge the financial support by AICTE CRS Project under TEQIP-III to conduct this research work.

Data cannot be made publicly available readers should contact the corresponding author for details.

Arnold
J. G.
Fohrer
N.
2005
Swat2000: current capabilities and research opportunities in applied watershed modelling
.
Hydrological Processes: An International Journal
19
(
3
),
563
572
.
doi:10.1002/hyp.5611
.
Arnold
J. G.
Allen
P. M.
Bernhardt
G.
1993
A comprehensive surface-groundwater flow model
.
Journal of Hydrology
142
(
1–4
),
47
69
.
https://doi.org/10.1016/0022-1694(93)90004-s
.
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 Water Resources Association
34
(
1
),
73
89
.
doi:10.1111/j.1752-1688.1998.tb05961.x
.
Arnold
J. G.
Allen
P. M.
Morgan
D. S.
2001
Hydrologic model for design and constructed wetlands
.
Wetlands
21
(
2
),
167
178
.
doi:10.1672/0277-5212(2001)021[0167:HMFDAC]2.0.CO;2
.
Babar
S.
Ramesh
H.
2015
Streamflow response to land use–land cover change over the Nethravathi river basin, India
.
Journal of Hydrologic Engineering
20
(
10
),
05015002
.
doi:10.1061/(asce)he.1943-5584.0001177
.
Beasley
D.
Huggins
L.
Monke
A.
1980
ANSWERS: a model for watershed planning
.
Transactions of the ASAE
23
(
4
),
938
944
.
https://doi.org/10.13031/2013.34692
.
Besaw
L. E.
Rizzo
D. M.
Bierman
P. R.
Hackett
W. R.
2010
Advances in ungauged streamflow prediction using artificial neural networks
.
Journal of Hydrology
386
(
1–4
),
27
37
.
doi:10.1016/j.jhydrol.2010.02.037
.
Beven
K.
1997
TOPMODEL: a critique
.
Hydrological Processes
11
(
9
),
1069
1085
.
https://doi.org/10.1002/(sici)1099-1085(199707)11:9%3C1069::aid-hyp545%3E3.0.co;2-o
.
Blöschl
G.
Sivapalan
M.
1995
Scale issues in hydrological modelling: a review
.
Hydrological Processes
9
(
3–4
),
251
290
.
doi:10.1002/hyp.3360090305
.
Choubin
B.
Solaimani
K.
Rezanezhad
F.
Roshan
M. H.
Malekian
A.
Shamshirband
S.
2019
Streamflow regionalization using a similarity approach in ungauged basins: application of the geo-environmental signatures in the Karkheh River Basin, Iran
.
Catena
182
,
104128
.
doi:10.1016/j.catena.2019.104128
.
Cunderlik
J.
Simonovic
S. P.
2004
Calibration, Verification and Sensitivity Analysis of the HEC-HMS Hydrologic Model
.
Department of Civil and Environmental Engineering, The University of Western
.
Devia
G. K.
Ganasri
B.
Dwarakish
G.
2015
A review on hydrological models
.
Aquatic Procedia
4
,
1001
1007
.
https://doi.org/10.1016/j.aqpro.2015.02.126
.
Dhar
S.
Mazumdar
A.
2009
Hydrological modelling of the Kangsabati river under changed climate scenario: case study in India
.
Hydrological Processes
23
(
16
),
2394
2406
.
doi:10.1002/hyp.7351
.
Du
T. L.
Lee
H.
Bui
D. D.
Arheimer
B.
Li
H.-Y.
Darby
S. E.
Kim
D.
2020
Streamflow prediction in ‘geopolitically ungauged’ basins using satellite observations and regionalization at subcontinental scale
.
Journal of Hydrology
,
125016
.
doi:10.1016/j.jhydrol.2020.125016
.
El-Sadek
A.
Irvem
A.
2014
Evaluating the impact of land use uncertainty on the simulated streamflow and sediment yield of the Seyhan river basin using the SWAT model
.
Turkish Journal of Agriculture and Forestry
38
(
4
),
515
530
.
doi:10.3906/tar-1309-89
.
FAO
1997
Digital Soil Map of the World and Derived Soil Properties
.
Fernandez
W.
Vogel
R.
Sankarasubramanian
A.
2000
Regional calibration of a watershed model
.
Hydrological Sciences Journal
45
(
5
),
689
707
.
doi:10.1080/02626660009492371
.
Garg
K. K.
Bharati
L.
Gaur
A.
George
B.
Acharya
S.
Jella
K.
Narasimhan
B.
2012
Spatial mapping of agricultural water productivity using the SWAT model in Upper Bhima Catchment, India
.
Irrigation and Drainage
61
(
1
),
60
79
.
doi:10.1007/s40030-015-0113-3
.
Guiamel
I. A.
Lee
H. S.
2020
Watershed modelling of the Mindanao River Basin in the Philippines using the SWAT for water resource management
.
Civil Engineering Journal
6
(
4
),
626
648
.
https://doi.org/10.28991/cej-2020-03091496
.
Guo
Y.
Zhang
Y.
Zhang
L.
Wang
Z.
2020
Regionalization of hydrological modeling for predicting streamflow in ungauged catchments: a comprehensive review
.
Wiley Interdisciplinary Reviews: Water
e1487
.
doi:10.1002/wat2.1487
.
He
Y.
Bárdossy
A.
Zehe
E.
2011
A review of regionalisation for continuous streamflow simulation
.
Hydrology and Earth System Sciences
15
(
11
),
3539
.
doi:10.5194/hess-15-3539-2011
.
Heuvelmans
G.
Muys
B.
Feyen
J.
2006
Regionalisation of the parameters of a hydrological model: comparison of linear regression models with artificial neural nets
.
Journal of Hydrology
319
(
1
),
245
265
.
doi:10.1016/j.jhydrol.2005.07.030
.
Kim
U.
Kaluarachchi
J. J.
2008
Application of parameter estimation and regionalization methodologies to ungauged basins of the Upper Blue Nile River Basin, Ethiopia
.
Journal of Hydrology
362
(
1
),
39
56
.
doi:10.1016/j.jhydrol.2008.08.016
.
Kittel
C. M.
Arildsen
A. L.
Dybkjær
S.
Hansen
E. R.
Linde
I.
Slott
E.
Tøttrup
C.
Bauer-Gottwein
P.
2020
Informing hydrological models of poorly gauged river catchments – a parameter regionalization and calibration approach
.
Journal of Hydrology
,
124999
.
doi:10.1016/j.jhydrol.2020.124999
.
Kumar
S.
Roshni
T.
Himayoun
D.
2019
A comparison of emotional neural network (ENN) and artificial neural network (ANN) approach for rainfall-runoff modelling
.
Civil Engineering Journal
5
(
10
),
2120
2130
.
doi:10.28991/cej-2019-03091398
.
Laflen
J.
Elliot
W.
Simanton
J.
Holzhey
C.
Kohl
K.
1991
WEPP: soil erodibility experiments for rangeland and cropland soils
.
Journal of Soil and Water Conservation
46
(
1
),
39
44
.
Lakshmanan
A.
Geethalakshmi
V.
Rajalakshmi
D.
Bhuvaneswari
K.
Srinivasan
R.
Sridhar
G.
Sekhar
N.
Annamalai
H.
2011
Climate change adaptation strategies in the Bhavani Basin using the SWAT model
.
Applied Engineering in Agriculture
27
(
6
),
887
893
.
doi:10.13031/2013.40623
.
Lek
S.
Guégan
J.-F.
1999
Artificial neural networks as a tool in ecological modelling, an introduction
.
Ecological Modelling
120
(
2–3
),
65
73
.
doi:10.1016/S0304-3800(99)00092-7
.
Liang
X.
Lettenmaier
D. P.
Wood
E. F.
Burges
S. J.
1994
A simple hydrologically based model of land surface water and energy fluxes for general circulation models
.
Journal of Geophysical Research: Atmospheres
99
(
D7
),
14415
14428
.
https://doi.org/10.1029/94jd00483
.
Liechti
T. C.
Matos
J.
Boillat
J.-L.
Schleiss
A.
2015
Influence of hydropower development on flow regime in the Zambezi river basin for different scenarios of environmental flows
.
Water Resources Management
29
(
3
),
731
747
.
doi:10.1007/s11269-014-0838-1
.
Loveland
T. R.
Reed
B. C.
Brown
J. F.
Ohlen
D. O.
Zhu
Z.
Yang
L.
Merchant
J. W.
2000
Development of a global land cover characteristics database and IGBP discover from 1 km AVHRR data
.
International Journal of Remote Sensing
21
(
6–7
),
1303
1330
.
doi:10.1080/014311600210191
.
Maier
H. R.
Dandy
G. C.
2001
Neural network based modelling of environmental variables: a systematic approach
.
Mathematical and Computer Modelling
33
(
6–7
),
669
682
.
doi:10.1016/S0895-7177(00)00271-5
.
Markstrom
S. L.
Niswonger
R. G.
Regan
R. S.
Prudic
D. E.
Barlow
P. M.
2008
GSFLOW-coupled ground-water and surface-water flow model based on the integration of the precipitation-runoff modeling system (PRMS) and the modular ground-water flow model (MODFLOW-2005)
.
US Geological Survey Techniques and Methods
6
,
240
.
https://doi.org/10.3133/tm6d1
.
Matos
T. S.
Uliana
E. M.
Martins
C. A. d. S.
Rapalo
L. M. C.
2020
Regionalization of maximum, minimum and mean streamflows for the Juruena River basin, Brazil
.
Revista Ambiente & Água
15
(
3
).
doi:10.4136/ambi-agua.2418
.
Merz
R.
Blöschl
G.
2004
Regionalisation of catchment model parameters
.
Journal of Hydrology
287
(
1–4
),
95
123
.
doi:10.1016/j.jhydr
.
Merz
R.
Blöschl
G.
2005
Flood frequency regionalisation – spatial proximity vs. catchment attributes
.
Journal of Hydrology
302
(
1–4
),
283
306
.
doi:10.1016/j.jhydrol.2004.07.018
.
Merz
R.
Blöschl
G.
Parajka
J. D.
2006
Regionalization methods in rainfall-runoff modelling using large catchment samples
.
IAHS Publication
307
,
117
.
Oo
H. T.
Zin
W. W.
Kyi
C. C. T.
2020
Analysis of streamflow response to changing climate conditions using SWAT model
.
Civil Engineering Journal
6
(
2
),
194
209
.
doi:10.28991/cej-2020-03091464
.
Pai
D.
Sridhar
L.
Rajeevan
M.
Sreejith
O.
Satbhai
N.
Mukhopadhyay
B.
2014
Development of a new high spatial resolution (0.25 × 0.25) long period (1901–2010) daily gridded rainfall data set over India and its comparison with existing data sets over the region
.
Mausam
65
(
1
),
1
18
.
Parajka
J.
Merz
R.
Blöschl
G.
2005
A comparison of regionalisation methods for catchment model parameters
.
Hydrology and Earth System Sciences Discussions
9
(
3
),
157
171
.
Perez
R.
Ineichen
P.
Moore
K.
Kmiecik
M.
Chain
C.
George
R.
Vignola
F.
2002
A new operational model for satellite-derived irradiances: description and validation
.
Solar Energy
73
(
5
),
307
317
.
Pugliese
A.
Farmer
W. H.
Castellarin
A.
Archfield
S. A.
Vogel
R. M.
2016
Regional flow duration curves: geostatistical techniques versus multivariate regression
.
Advances in Water Resources
96
,
11
22
.
doi:10.1016/j.advwatres.2016.06.008
.
Rahman
K.
Ray
N.
Giuliani
G.
Maringanti
C.
George
C.
Lehmann
A.
2017
Breaking walls towards fully open source hydrological modeling
.
Water Resources
44
(
1
),
23
30
.
Rienecker
M. M.
Suarez
M. J.
Gelaro
R.
Todling
R.
Bacmeister
J.
Liu
E.
Bosilovich
M. G.
Schubert
S. D.
Takacs
L.
Kim
G.-K.
Bloom
S.
2011
Merra: Nasa's modern-era retrospective analysis for research and applications
.
Journal of Climate
24
(
14
),
3624
3648
.
Samuel
J.
Coulibaly
P.
Metcalfe
R. A.
2011
Estimation of continuous streamflow in Ontario ungauged basins: comparison of regionalization methods
.
Journal of Hydrologic Engineering
16
(
5
),
447
459
.
doi:10.1061/(asce)he.1943-5584.0000338
.
Sivapalan
M.
Takeuchi
K.
Franks
S.
Gupta
V.
Karambiri
H.
Lakshmi
V.
Liang
X.
McDonnell
J.
Mendiondo
E.
O'connell
P.
Oki
T.
2003
IAHS decade on predictions in ungauged basins (PUB), 2003–2012: shaping an exciting future for the hydrological sciences
.
Hydrological Sciences Journal
48
(
6
),
857
880
.
doi:10.1623/hysj.48.6.857.51421
.
Skøien
J. O.
Merz
R.
Blöschl
G.
2006
Top-kriging – geostatistics on stream networks
.
Hydrology and Earth System Sciences
10
(
2
),
277
287
.
doi:10.5194/hess-10-277-2006
.
Soni
P.
2018
Implications of Data Uncertainty on Trend Analysis and Effects of Renewed Solar Dimming on the Hydrology of Indian River Basins
.
PhD Thesis
,
Indian Institute of Technology
,
Kanpur
.
Ssegane
H.
Tollner
E.
Mohamoud
Y.
Rasmussen
T.
Dowd
J.
2012
Advances in variable selection methods I: causal selection methods versus stepwise regression and principal component analysis on data of known and unknown functional relationships
.
Journal of Hydrology
438
,
16
25
.
doi:10.1016/j.jhydrol.2012.01.008
.
Sun
L.
Seidou
O.
Nistor
I.
Goïta
K.
Magagi
R.
2016
Simultaneous assimilation of in situ soil moisture and streamflow in the SWAT model using the Extended Kalman Filter
.
Journal of Hydrology
543
,
671
685
.
doi:10.1016/j.jhydrol.2016.10.040
.
Swain
J. B.
Patra
K. C.
2017
Streamflow estimation in ungauged catchments using regionalization techniques
.
Journal of Hydrology
554
,
420
433
.
doi:10.1016/j.jhydrol.2017.08.054
.
Swain
J. B.
Patra
K. C.
2019
Impact of catchment classification on streamflow regionalization in ungauged catchments
.
SN Applied Sciences
1
(
5
),
456
.
doi:10.2166/wcc.2018.161
.
Swain
S. S.
Mishra
A.
Sahoo
B.
Chatterjee
C.
2020
Water scarcity-risk assessment in data-scarce river basins under decadal climate change using a hydrological modelling approach
.
Journal of Hydrology
125260
.
doi:10.1016/j.jhydrol.2020.125260
.
Tripathi
M.
Panda
R.
Raghuwanshi
N.
2003
Identification and prioritisation of critical sub-watersheds for soil conservation management using the SWAT model
.
Biosystems Engineering
85
(
3
),
365
379
.
doi:10.1016/S1537-5110(03)00066-7
.
Verma
M. K.
Verma
M. K.
2019
Calibration of a hydrological model and sensitivity analysis of its parameters: a case study of Seonath river basin
.
International Journal of Hydrology Science and Technology
9
(
6
),
640
656
.
doi:10.1504/IJHST.2019.103444
.
Wagener
T.
Wheater
H. S.
2006
Parameter estimation and regionalization for continuous rainfall-runoff models including uncertainty
.
Journal of Hydrology
320
(
1
),
132
154
.
doi:10.1016/j.jhydrol.2005.07.015
.
Waseem
M.
Ajmal
M.
Kim
T.-W.
2015
Ensemble hydrological prediction of streamflow percentile at ungauged basins in Pakistan
.
Journal of Hydrology
525
,
130
137
.
doi:10.1016/j.jhydrol.2015.03.042
.
Wilcox
S.
2007
National Solar Radiation Database 1991–2005 Update: User's Manual. Technical Report
.
National Renewable Energy Laboratory (NREL)
,
Golden, CO
.
Wilk
J.
Hughes
D. A.
2002
Simulating the impacts of land-use and climate change on water resource availability for a large south Indian catchment
.
Hydrological Sciences Journal
47
(
1
),
19
30
.
doi:10.1080/02626660209492904
.
Yang
X.
Magnusson
J.
Rizzi
J.
Xu
C.-Y.
2017
Runoff prediction in ungauged catchments in Norway: comparison of regionalization approaches
.
Hydrology Research
49
(
2
),
487
505
.
doi:10.2166/nh.2017.071
.
Young
R.
Onstad
C.
Bosch
D.
Anderson
W.
1989
AGNPS: a nonpoint-source pollution model for evaluating agricultural watersheds
.
Journal of Soil and Water Conservation
44
(
2
),
168
173
.
https://doi.org/10.21273/horttech.6.4.344
.
Zhang
Y.
Chiew
F.
2009
Evaluation of regionalisation methods for predicting runoff in ungauged catchments in southeast Australia
. In:
18th World IMACS Congress and MODSIM09 International Congress on Modelling and Simulation
.
Modelling and Simulation Society of Australia and New Zealand
, pp.
3442
3448
.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits copying, adaptation and redistribution, provided the original work is properly cited (http://creativecommons.org/licenses/by/4.0/).

Supplementary data