Journal of Civil and Environmental Engineering

Evaluation of the Performance of High-Resolution Satellite Based Rainfall Products for Stream Flow Simulation

Abstract

Dawit Girma* and Belete Berhanu

Precipitation data is an intrinsic parameter of rainfall-runoff simulation, since it is strongly hooked into the accuracy of the spatial and temporal representation of the precipitation. In areas where rainfall gauging stations are scarce, additional data sources could also be needed. Satellite platforms have provided as a satisfactory alternative because of their global coverage. Although a good range of satellite-based estimations of precipitation is out there, not all the satellite products are suitable for all regions. In addition, in data-scarce areas where interpolation schemes are applied, it becomes difficult to get an accurate performance assessment; another comparison method is required as rainfall-runoff models. Remotely-sensed estimates are to get realistic and reliable data to be accessed in water resource assessments. Therefore, there is a requirement to evaluate the accuracy of remote sensing techniques. Inter comparison between Satellite rainfall product and observed data were done using point to grid method selecting representative metrological stations. Inter comparison between Satellite rainfall product and observed data were done using point to grid method selecting representative metrological stations. TAMSAT shows the average value of R=0.87 and NS =0.764. Considering four categorical index POD, FAR, FB and HSS, the average value 0.71, 0.22, 0.92, and 0.66 respectively. For CHRIPS average R and NS are 0.88 and 0.755 respectively and categorical index POD, FAR, FB and HSS were 0.8,0.05, 0.85 and 0.81) respectively. The study model stream flow using both CHRIPS and TAMSAT rainfall products by using the SWAT model from 1983 to 2017. The model was calibrated from 1998 to 2003 and validated from 2004 to 2007 using SUFI-2 algorithm embodied in the SWAT-CUP. The Nash-Sutcliffe Efficiency (NSE), linear correlation coefficient (R) and BIAS indices were used to benchmark the model performance and shows very good result (having R2 and NS=0.71- 0.95 during calibration and 0.72-0.97 during validation.