Application of a newly developed large‐scale conceptual hydrological model in simulating streamflow for credibility testing in data scarce condition

Regionalization approach is adopted to test the credibility of a newly developed, large‐scale, conceptual hydrological model, namely, satellite‐based hydrological model (SHM), under data‐scarce conditions. SHM has a modular structure. There are five modules: surface water module (SW), which uses natural resources conservation service‐curve number and Hargreaves equation; forest module (F), which uses water balance and subsurface dynamics; snow module (S), which uses temperature index and radiation‐temperature index algorithms; groundwater module, which considers contributions from SW, F, and S modules and uses water level variation process; and routing module, which uses modified spatial distributed direct hydrograph model. Four neighboring basins of varying size are selected to form the reference (Subarnarekha) and the test (Brahmani, Baitarani, and Kangsabati) basins. The similarity index and hierarchical agglomerative cluster analysis show that Brahmani is the most homogeneous basin with respect to the Subarnarekha basin. The model simulations are run at 1 km resolution. The values of statistical indices Nash‐Sutcliffe efficiency, coefficient of determination, and percent BIAS, are found to be 0.82%, 0.84%, and 13.99%, respectively, during calibration (1993–1999; 1993 as a warm‐up period), and 0.61%, 0.63%, and 14.13%, respectively, during validation (2000–2004) at Jenapur gauging station of the Brahmani river basin. Subsequently, the calibrated parameters are transferred in the model setup of the Subarnarekha basin for simulating streamflow over a randomly chosen period 1998–2000. The results, obtained with both Brahmani and Subarnarekha parameters, confirm that the model performs well with the regionalized parameters, and is suitable for hydrologically homogeneous ungauged basins. The results show that the newly developed model may be used for the water balance simulations in ungauged or scantily gauged basins, leading to the development of water resources plans for better utilization of this vital resource. The results from the sensitivity analysis and uncertainty analysis show that the nonsensitive parameters have a negligible impact on the model results. Further study, however, may be needed to evaluate the model performance and to establish its universal credibility.

The results, obtained with both Brahmani and Subarnarekha parameters, confirm that the model performs well with the regionalized parameters, and is suitable for hydrologically homogeneous ungauged basins. The results show that the newly developed model may be used for the water balance simulations in ungauged or scantily gauged basins, leading to the development of water resources plans for better utilization of this vital resource.
The results from the sensitivity analysis and uncertainty analysis show that the nonsensitive parameters have a negligible impact on the model results. Further study, however, may be needed to evaluate the model performance and to establish its universal credibility.

Recommendations for Water Resource Managers and The Scientific Community
1. This successful testing of the newly developed hydrological model implies that it may be a useful tool for water resources assessment in data-scarce conditions. the study area in detail. Agricultural land, water bodies, and forests cover 69.7%, 7.8%, and 17.5% of the study area, respectively. Figure 1 also shows gauging stations for streamflow measurement along with the reservoirs present in the area.

| Data
Daily rainfall and maximum and minimum temperatures are obtained from India Meteorological Department (IMD), Pune at 1°× 1°resolution. Data are interpolated to 1 km × 1 km resolution by using bilinear interpolation technique to use as input into SHM. Soil and land use land cover (LULC) maps are collected from the Food and Agriculture Organisation (FAO) website (http://www.fao.org/soils-portal/soil-survey/soil-maps-and-databases/harmonizedworld-soil-database-v12/en/) at 1 km × 1 km scale. Digital Elevation Model (DEM) of 30 m × 30 m resolution is taken from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) website (https://asterweb.jpl.nasa.gov/gdem.asp). DEM is resampled into 1 km × 1 km resolution for use in SHM. Observed streamflow data, for various gauging stations of the study area, are collected from the Central Water Commission (CWC), Bhubaneswar, India. The observed streamflow is available for 1995-2004, 1992-2004, 1987-2005, and 1987-2005 for Jenapur, Muri, Jamshedpur, and Ghatshila, respectively. The observed streamflow at Muri and Jamshedpur gauging stations are naturalized to reduce the effect of reservoirs. The detailed description of the naturalization process is provided below.  There are two reservoirs, Getalsud and Chandil, in Subarnarekha basin. Since we have gauged flow data (at three gauging stations of Subarnarekha basin) for the post-reservoir period, these may represent a different hydrologic regime of the basin from the prereservoir period (Biemans et al., 2011;Magilligan & Nislow, 2005;Risley, Constantz, Essaid, & Rounds, 2010). As SHM presently does not take into account the effect of the reservoirs during model simulation, we need to reconstruct the flow data to minimize the impact of reservoir storage on streamflow simulations. There are several techniques available for streamflow naturalization, for example, drainage area ratio method (Hirsch, 1979;Wurbs, 2006), modified soil conservation service (SCS)-curve number (CN) technique (Wurbs, Muttiah, & Felden, 2005), generic equations (Wurbs, 2006), and regional statistical technique (Hirsch, 1979). Among these, regional statistical technique and drainage area ratio methods are well-suited when the observed flow data for the prereservoir period are not available. The regional statistical technique, however, performs better if one or more base stations have long continuous streamflow records available (Hirsch, 1979). Thus, we have used this technique to naturalize the flow at Muri and Jamshedpur gauging stations. The most downstream, and hence the least hydrologically affected gauging station, Ghatshila has been as the base station.

| Regional statistical technique
The technique assumes that the standardized flow at the base station and the gauging station, chosen for data reconstruction, are equal for every month. The reconstruction equation is as follows: where A q ( ) j is the estimated mean flow for month j at the gauging station; B q ( ) j is the estimated standard deviation of flow for month j at the gauging station; A p ( ) j is the mean flow for month j at the base station; B p ( ) j is the standard deviation of flow for month j at the base station; P ij is the observed streamflow (day i, month j); and Q ij is the reconstructed streamflow (day i, month j).
The regional regression equations between streamflow and basin characteristics are used to estimate means and standard deviations of the monthly flows at Muri and Jamshedpur gauging stations (Thomas & Benson, 1970). The regression equations used are of the following form: where X is the dependent variables (A q ( ) j and B q ( ) j , stated in Equation (1)); A is the drainage area; S is the slope of the main channel; L is the length of the main channel; S t is the percentage area of water bodies; E is the mean basin elevation; F is the percentage of forest area in the basin; P is the mean annual precipitation; I 24,2 is the 2-year, 24-hr maximum rainfall intensity; and a, b 1 , b 2 , b 3 , b 4 , b 5 , b 6 , b 7 , and b 8 are the regression coefficients. However, to estimate the values of regression coefficients, A p ( ) j and B p ( ) j are used on the left-hand side of Equation (2) as dependent variables. The values of independent variables, on the right-hand side of Equation (2), are considered for the Ghatshila subbasin. The estimated values of the regression coefficients are subsequently used to determine means and standard deviations of monthly flows at Muri and Ghatshila gauging stations. From the regression equation, used in Hirsch (1979), terms related to temperature and snowfall, are neglected as snowfall does not occur in the study area.

| Description of SHM
SHM works on 1 km × 1 km spatial grid resolution and properties at the center of a cell are assumed to be the properties of the cell. SHM has five modules: surface water (SW), forest (F), snowmelt (S), groundwater (GW), and routing (ROU). SHM grid cells corresponding to forest and snow land cover are modeled using the F and S modules, respectively; whereas other grid cells are modeled using the SW module. SHM uses MySQL (open source software) as a relational database management system (RDBMS). The details of the model are given below.

| Core theory
In the SW module, NRCS-CN method (Chow, Maidment, & Mays, 2005) and Hargreaves method (Hargreaves & Samani, 1985) are used to estimate the surface runoff and the potential evapotranspiration (PET), respectively. A water balance method is used to determine soil moisture. The soil profile is considered as a single-layered zone of 300 mm depth, and moistureholding and moisture-transmitting characteristics of the soil and underlying layer are considered while accounting for the soil moisture. The wetting of the soil layer depends on the amount of infiltration. The amount of water, more than the maximum capacity of the zone, percolates, and contributes to the groundwater. The soil moisture is depleted by evapotranspiration, at a potential rate or actual rate, depending on the moisture condition. The potential rate is considered as long as sufficient moisture is available.
The F module functions based on water balancing and the dynamics of the subsurface to provide output in the form of runoff, soil moisture, evapotranspiration, and contribution to groundwater using the technique and parameters stated in Das, Islam, Dutta, Dubey, and Sarkar (2014). Subsurface is considered of having soil matrix and macropores. There are two types of macropores: main bypass and internal catchments, with the main bypass being responsible for direct contribution to groundwater. Soil matrix is assumed to have three layers, which are crucial for water balance and changes in the soil moisture. After infiltration, the saturation of three layers gets started from the top in batch, and after complete saturation of the three layers, the excess water goes to the groundwater. Runoff generation, after a precipitation event, occurs according to the antecedent conditions in the subsurface.
The S module determines the snow density from snow albedo (Smith & Halverson, 1979) for estimating snowmelt depth by using two different algorithms, namely, temperature index algorithm and radiation-temperature index algorithm. Since the study area does not have any snow-covered land, S module is not considered in this study.
GW module uses the contribution from SW, F, and S modules and generates baseflow, following the water level variation process described in Sekhar and Rasmi (2004). The resultant baseflow along with the surface runoff generated from other modules is routed up to the outlet as streamflow.
For routing of generated runoff and baseflow, modified spatial distributed direct hydrograph (SDDH; Paul et al., 2018) is used. This is a cell-to-cell routing technique. The details of the modification of ROU module are described in Paul et al. (2018).

| Model parameters
There are 23 parameters in the model, which are listed in Table 2.

| Model applications
SHM is applied in three applications so far. SHM is applied in the Kabini basin for testing the compatibility of the ROU module (Paul et al., 2018). Subsequently, SHM is tested using hierarchical operational testing of Klemes (1986) (Paul, Gaur, et al., 2019). The performance of SHM is further compared with the widely used Soil and Water Assessment Tool (SWAT) model (Paul, Zhang, Mishra, Panigrahi, & Singh, 2019).

| Methodology
To test the capability of SHM for data-scarce conditions, Subarnarekha river basin, a gauged basin, is selected as the reference (data-scarce) basin. The Brahmani, Baitarani, and Kangsabati are used as the test basins (for details please see Section 2) to check the regionalization capability of the model. Homogeneity test is conducted for identifying the most homogeneous test basin with respect to the reference basin, Subarnarekha. For homogeneity test, we have used "similarity index" (Burn & Boorman, 1993) technique because of its simplicity, and "hierarchical agglomerative cluster analysis" (Ouyang et al., 2010) [1998][1999][2000]. SHM simulation results obtained using the transferred parameters from the most homogeneous test basin are compared with the SHM simulation results with Subarnarekha parameters at Muri, Jamshedpur, and Ghatshila stations. Finally, to understand the role of sensitive SHM parameters in regionalization process and model competency, sensitivity analysis, and parametric uncertainty analysis are performed for the reference basin and the most homogeneous basin.

| Regionalization approach
For regionalization, the homogeneity test is performed first to identify the most homogeneous basin among the test basins (Brahmani, Baitarani, and Kangsabati) with respect to the reference basin (Subarnarekha). After determining the most homogeneous basin, the model is calibrated and validated for the chosen basin. Subsequently, the calibrated parameters are transferred to the model setup of the reference basin for streamflow simulation. Burn and Boorman (1993) defined similarity index as the "sum of absolute differences of k selected physiographic attributes of the gauged X G basin and the ungauged X U basin, normalized by the range of attributes (ΔX)". Mathematically, it is expressed as follows:

| Similarity index technique
where Φ is the similarity index.
Here, for performing the similarity index test, we have used the combination of attributes based on LULC (FAO data), soil type (FAO data) and ten equal intervals of elevation, ranging from 0 m to 1,041 m, for all four basins. The test basin with the lowest value of the similarity index is chosen as homogeneous to the reference basin. Besides LULC, soil and elevation, land management features related to agricultural management, urban management, and hydraulic structures may also play a role in driving the basin similarity to a certain extent. We, however, did not consider these features here due to unavailability of the detailed information.

| Cluster analysis
In this study, we applied the hierarchical agglomerative clustering algorithm, considering complete linkage dissimilarity and the Euclidean distance measure (Ouyang et al., 2010). We used an open-source Cluster 3.0 software package for the purpose (de Hoon, Imoto, Nolan, & Miyano, 2002Eisen, Spellman, Brown, & Botstein, 1998). The Euclidean distance measurement is performed using the following equation: The distance (d) between x and y is computed directly, where x and y are values of the same attribute for reference and test basins, respectively. While using the Euclidean distance measure, the data, however, is suitably normalized.
The software contains Cluster and TreeView programs to provide a computational and a graphical environment, respectively, for analysing data. The LULC, soil type, and 10 equal intervals of elevation ranging from 0 m to 1,041 m, covering the study area, are combined to form hydrological response units (HRUs) for four basins and used as input to the Cluster 3.0 program. TreeView program is subsequently run with output files from cluster 3.0 program as input, and the output is visualized and analysed. Figure 3 presents the essential components of TreeView (Status Panel, Array tree, Array names, Global, and zoom pixels) in Cluster 3.0. The visual interpretation of the Global pixels of Array tree (a zoomed-in view is presented in the right panel of the figure) helps us in identifying the most homogeneous basin with respect to the reference basin. For example, based on the oval-marked similar regions of the global pixels in the figure, Basin D would be chosen as the most homogeneous basin with respect to the reference Basin A.

| Transfer of parameters from the homogeneous basin to reference basin
There are two types of parameters in SHM-spatially distributed and static (Table 2). Both types of parameters are calibrated in the homogeneous basin and transferred to the reference basin. For sensitive spatially distributed parameters, the calibration is performed for various combinations of watershed characteristics. For example, we explain the calibration of the parameter "n o ." "n o " values are dependent on different LULC classes. Therefore, "n o " values are calibrated over the specified range for different LULC classes in the homogeneous basin. Later, the calibrated values are transferred from the database of the homogeneous basin to the database of reference basin for respective LULCs. The details are described in Figure S1. On the other hand, the static parameters are lumped over the basin. For nonsensitive parameters, we use developers' default values/values from standard references.

| Evaluation criteria
Nash-Sutcliffe efficiency (NSE; Nash & Sutcliffe, 1970) and coefficient of determination (R 2 ; Moriasi et al., 2007) are used as the goodness-of-fit parameters to assess the model simulation accuracy, along with percent bias (PBIAS; Moriasi et al., 2007) for measuring the performance for the water balance.

| Sensitivity analysis
Sensitivity analysis is conducted to identify the model output variability with respect to changes in the values of 20 SHM parameters (Table 2) for the period 1998-2000 in Jenapur subbasin of Brahmani basin and three subbasins of the Subarnarekha basin (Baird, 1989;Paul, Gaur, et al., 2019;Paul et al., 2018). Three parameters of the S module (Table 2) are excluded from the sensitivity analysis exercise as the study basins are free from snow. The model output variability is analysed against variations in the value of a single parameter around its calibrated value. NSE F I G U R E 3 TreeView containing various clusters formed in Cluster 3.0 is chosen as the objective function for this analysis. Parameters are perturbed by ±5% and ±10% from their calibrated values, one at a time, while keeping the other parameters constant.

| Uncertainty analysis
For each parameter, 95 percent predictive uncertainty (PPU) band is developed to analyse the uncertainty. For the purpose, simulated streamflows obtained using the calibrated values and the four perturbed values, that is at ±5% and ±10% of the calibrated values, of the parameters are considered. Consequently, 97.5% upper band and 2.5% lower band are estimated to develop to 95 PPU band. Then, values of P (measured data bracketed by the 95 PPU band) and R (the relative length of the 95 PPU band with respect to the model simulated values) are determined to quantify the uncertainty of each parameter, following the procedure described in Paul, Zhang, et al. (2019). Consequently, the p value versus R value scatter plots are used for individual subbasins to quantify the uncertainty of individual parameters.
The contribution in uncertainties from different sources (sensitive parameters, nonsensitive parameters, their interactions and other sources) in the simulated streamflow is obtained through analysis of variance (ANOVA). Though, it is a widely used technique in climate projection analysis of streamflow (Bosshard et al., 2013;Gaur, Bandyopadhyay, & Singh, 2020;Yip, Ferro, Stephenson, & Hawkins, 2011); it is rarely used for identifying uncertainty sources in hydrological simulations. In the process, the difference between simulated values and references values, Y ∆ (Equation 5) is considered as the input in the ANOVA method for the partitioning of uncertainty due to individual sources as: where Y sim and Y obs represent the simulated and observed values for a given variable Y. An ANOVA model is constructed with different contributing sources, that is, sensitive parameters (represented by α), nonsensitive parameters (represented by β), and the error term (ε) representing the internal variability (Equation 6). The nonlinear term (αβ) shows the interaction between these two sources and is also considered a factor for uncertainty.

Y µ α β αβ ε
ANOVA partitions the total variance into components due to different sources of variation. Therefore, the total variance in simulated streamflow is represented by the sum of squares from contributions and interactions as follows: where SS Total (Equation 7) presents the total sum of squares and SS Contribution and SS Interaction present the sum of squares from contributors and interactions, respectively. Equation (7) may also be expressed as follows: Uncertainty is estimated by the ratio of variance contribution due to individual factors to total variance (SS /SS

| RESULTS AND DISCUSSION
Results of the regionalization approach and sensitivity analysis for both reference and the most homogeneous basins are presented and discussed in this section.

| Calibration and validation of SHM in Subarnarekha basin
Figures S2 and S3 in the Supporting Information section present the time series plots of the simulated and observed streamflows and scatter plots of simulated versus observed streamflows at the gauging stations of Subarnarekha basin, respectively. The values of statistical indices, NSE, R 2 , and PBIAS are found to be 0.65%, 0.61%, and 13.12%, respectively, during Muri calibration, and 0.5%, 0.51%, and 16.37%, respectively, during Muri validation. Similarly, NSE, R 2 , and PBIAS values are 0.8%, 0.82%, and 14.31%, respectively, during Jamshedpur calibration, and 0.67%, 0.77%, and 14.15%, respectively, during Jamshedpur validation. For Ghatshila, NSE, R 2 , and PBIAS are 0.66%, 0.61%, and 13.13%, respectively, during calibration, and 0.5%, 0.5%, and 16.38%, respectively, during validation. The values of statistical indices show that the performance of SHM is satisfactory (i.e., NSE > 0.36, R 2 > 0.5, and −25% ≤ PBIAS ≤ 25%) during the calibration and validation periods for Muri, Jamshedpur, and Ghatshila gauging stations (Moriasi et al., 2007;Van Liew, Veith, Bosch, & Arnold, 2007). The results observed here are better than those reported in Paul, Gaur, et al. (2019). The improved model performance may be attributed to a finer grid resolution of one km grid cell resolution used here as compared to 5 km grid cell resolution in Paul, Gaur, et al. (2019). However, as evident from PBIAS values, underestimation persists throughout the model simulation (though within acceptable limit).

| Outcome of homogeneity test
Results of the homogeneity test to identify the most homogeneous test basin with respect to the reference basin are presented below. Table 3 presents the results of the homogeneity test using the similarity index technique. As evident, Brahmani is the most homogeneous basin with respect to Subarnarekha basin as the similarity index is the lowest (1.04) for this basin as compared to Baitarani and Kangsabati basins. Figure 4 presents the zoom pixels extracted from TreeView of Cluster 3.0 software, obtained using the data of the reference (Subarnarekha) and test basins (Brahmani, Baitarani, and Kangsabati). The similar regions of the pixels in Figure 4, marked with blue-colored ovals and circles, show that Brahmani basin is the most homogeneous basin with respect to Subarnarekha basin. Thus, both the Similarity Index technique and Hierarchical Agglomerative Cluster Analysis technique produce similar results.

| Model simulation results with regionalization approach
SHM setup for Brahmani river basin is calibrated over 1993-1999 (1993 as a warm-up) and validated over 2000-2004 at Jenapur gauging station. The values of statistical indices, NSE, R 2 , and PBIAS, are found to be 0.82%, 0.84%, and 13.99%, respectively, during calibration, and 0.61%, 0.63%, and 14.13%, respectively, during validation. The values of the statistical indices show that SHM is calibrated satisfactorily at Jenapur gauging station of Brahmani basin (Moriasi et al., 2007;Van Liew et al., 2007). We have, however, calibrated the model against discharge at the outlet only. Thus, our calibration technique includes only the temporal variability of the system and neglects the spatial component as the aggregated discharge at the outlet neglects the sensitivity to spatially explicit hydrological states and fluxes. Figure 5 presents the time series plot of the simulated and observed streamflows, and the scatter plot of simulated versus observed streamflow at Jenapur gauging station. Figure 5a depicts that the simulated flows capture the overall temporal pattern of the observed flow well. It is also evident from Figure 5b that the linear relationship between the observed and simulated streamflows is acceptable. Subsequently, the calibrated parameters of Brahmani basin are transferred in the   (Figure 6b) that the linear relationship between the observed and simulated streamflows is acceptable. Besides, Figure 7 summarizes the statistical indices of simulation results at three gauging stations of Subarnarekha basin with Brahmani parameters. The statistical indices obtained using previously calibrated parameters for Subarnarekha basin are also included for better comparison. As evident, NSE and R 2 values are satisfactory at all three gauging stations with both Brahmani and Subarnarekha parameters (Moriasi et al., 2007;Van Liew et al., 2007). PBIAS values show flow underestimation at all three stations, though, within the acceptable limit (Moriasi et al., 2007). However, at Muri gauging station, medium flows are overestimated at a few instances Brahmani parameters (Figure 6a(i)), whereas, high flows are overestimated with Subarnarekha parameters (Figure 6a(ii)). At Ghatshila gauging station, however, medium and high flows are underestimated with both Brahmani and Subarnarekha parameters (Figure 6a(v), (vi)). The flow underestimation at Ghatshila gauging station may be due to nonnaturalized observed flow (Paul, Gaur, et al., 2019). Since regionalized model parameters cannot account for all the variations in the basin, under-or over-estimation of the observed flow to a certain extent are expected. Overall results, however, show that SHM performs credibly with regionalized parameters at 1 km grid cell resolution. Thus, it is proved to be a useful modeling tool for flow simulations in data-scarce conditions. F I G U R E 6 (a) Simulation results (as daily time series plot) with (i) Brahmani parameters at Muri gauging station, (ii) Subarnarekha parameters at Muri gauging station, (iii) Brahmani parameters at Jamshedpur gauging station, (iv) Subarnarekha parameters at Jamshedpur gauging station, (v) Brahmani parameters at Ghatshila gauging station, and (vi) Subarnarekha parameters at Ghatshila gauging station. (b) Simulation results (as scatter plot) with (i) Brahmani parameters at Muri gauging station, (ii) Subarnarekha parameters at Muri gauging station, (iii) Brahmani parameters at Jamshedpur gauging station, (iv) Subarnarekha parameters at Jamshedpur gauging station, (v) Brahmani parameters at Ghatshila gauging station, and (vi) Subarnarekha parameters at Ghatshila gauging station

| Sensitivity analysis
Sensitivity analysis is carried out following the procedure described in Section 2.4, and only three parameters, CN, n o and n c are found to be sensitive (Figure 8). Figure 8 presents the results of the sensitivity analysis. The results show that CN is the most sensitive parameter, followed by n o and n c , for both Subarnarekha and Brahmani basins. In Jenapur subbasin, streamflow simulations are more sensitive to negative perturbation than positive perturbation, specifically for perturbation of −5% for CN and −10% for n o and n c . In Muri subbasin, however, streamflow simulations are sensitive to both positive and negative perturbations in the calibration parameters, with higher sensitivity to negative perturbations. Besides, CN appears to be sensitive at Jamshedpur gauging station for negative perturbation and at Ghatshila gauging stations for both positive and negative perturbations. However, n c and n o show limited sensitivity at Jamshedpur gauging station. The figure also depicts that at Ghatshila gauging station, n c and n o show limited sensitivity for positive perturbations, and negligible sensitivity for negative perturbations. Besides, it is also visible that parameter sensitivities are unidirectional at Jenapur and Muri as compared to the other two subbasins. Overall, the variability of NSE, however, is not uniform due to complex interactions between parameter sets (Table 2) and the hydrological responses for the three subbasins of Subarnarekha basin and Jenapur subbasin of Brahmani basin (Beven, 2006). The nonsensitivity of F module parameters may be due to the low forest cover in the study area. The lower sensitivity of GW module parameters shows that the groundwater contribution to the streamflow generation is low. However, the nonsensitive behavior of the ROU module parameter "B" is contradictory to the earlier finding of Paul et al. (2018). The larger size of the study area may be responsible for this behavior.

| Uncertainty analysis
The 95 PPU bands, along with the observed data, for all parameters and all the subbasins, are produced in Figure S4. Figure 9 presents the p value versus R value plots for various parameters in different subbasins. It is clear from Figure 9 that the sensitive parameters show more uncertainty as compared to the nonsensitive parameters. Figure 10 presents the uncertainty contribution of the sensitive parameters, nonsensitive parameters, their interactions and other sources to the total

With Brahmani parameters
With Subarnarekha parameters F I G U R E 7 Statistical indices during satellite-based hydrological model testing using regionalization approach. PBIAS, percent bias model simulation uncertainty for different subbasins. It is evident from Figure 10 that nonsensitive parameters have the lowest contribution to the total model uncertainty. The sensitive parameters are the primary source of uncertainty in all subbasins, followed by the other sources. The other sources may include model structure and input data uncertainty. The next contributor is, however, the interaction between sensitive and nonsensitive parameters. Though the order of contribution from different sources of uncertainty to the total uncertainty remains the same for all subbasins, the magnitudes vary over the subbasin. The lower contribution of nonsensitive parameters to the total model uncertainty, obtained from the ANOVA analysis corroborates the findings from the p value versus R value plots (Figure 9).

| CONCLUSIONS
Regionalization approach is adopted to test the credibility of a newly developed large-scale, conceptual, hydrological model, SHM, for data-scarce conditions. Model is applied at 1 km grid cell resolution. The similarity index and hierarchical agglomerative cluster analysis show that Brahmani is the most homogeneous basin with respect to the Subarnarekha basin. Overall results of the regionalization approach demonstrate that SHM performs credibly in simulating streamflow with regionalized parameters in a hydrologically homogeneous region at 1 km grid  Table 2 cell resolution. The sensitivity analysis shows that three parameters, CN, n o , and n c , are sensitive. The uncertainty analysis shows that the sensitive parameters are the major contributors to the total model uncertainty, with contributions of nonsensitive parameters being negligible. Therefore, SHM may be adopted in ungauged basins of India and other parts of the world for studying the streamflow characteristics, leading to better management of water resources. Further applications, however, may establish the finding with more details.