Comparative analysis of spatial downscaling techniques for satellite-based precipitation data under varied topographic and climatic conditions

ABSTRACT The spatial resolution of satellite precipitation data influences the accuracy of hydrological modeling for smaller catchments. In this study, statistical and hydrological evaluation of downscaled precipitation data in the Kanshi River Basin, a sub-catchment of the Mangla catchment, with varied topographic and climatic conditions, has been carried out using four downscaling techniques: multiple linear regression (MLR), support random forest (RF), exponential regression (ER), and support vector machine (SVM). Statistical evaluation of the downscaled data was done using root mean square error (RMSE) and mean absolute error (MAE). The coefficient of determination (R2) and Nash–Sutcliffe efficiency (NSE) were utilized as the goodness-of-fit measures for hydrologic evaluation using the SWAT model, calibrated using the SWAT-CUP tool (SUFI-2). Model calibration revealed NSE of 0.84 and R2 of 0.96, while for validation, the NSE was 0.80 and R2 of 0.89. Gauge data simulation results revealed NSE = 0.77, R2 = 0.89, whereas for the original satellite data simulation results, NSE = 0.30, R2 = 0.61. Among the downscaling techniques, SVM gave the best results with NSE = 0.89, R2 = 0.94. The findings of the study encouraged the use of downscaled satellite-based precipitation data toward effective management of small basins.