Abstract
In developing countries, hydrological modeling in support of decision-making and planning of hydrological resources is often hindered by the scarcity of in situ hydro-meteorological data and uneven distribution of observation stations. Satellite-based precipitation products (SPPs) can potentially play a role in overcoming the challenge posed by insufficient and inconsistent in situ precipitation measurements. However, their performance in estimating observations needs to be evaluated before their application in hydrological modeling. This paper evaluates and compares four high-resolution SPPs [ARC2 (African Rainfall Climatology version 2), CHIRPS (Climate Hazard Group Infrared Precipitation with Station data), TMPA 3B42v7 (Tropical Rainfall Measuring Mission Multi-satellite Precipitation Analysis 3B42 product), and PERSIANN-CDR (Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks)] with measurements from rain-gauge stations (2007-2016) over the Upper Tana River Basin (UTRB). Performance of the SPPs is evaluated using continuous [correlation coefficient (R), mean absolute error (MAE), root mean square error (RMSE), Nash-Sutcliffe efficiency coefficient (NSE), and percent bias (PBIAS)] and categorical [POD (probability of detection), FAR (false alarm ratio), CSI (critical success index), and ETS (equitable threat score)] metrics at daily, dekadal, and monthly time steps. The daily, dekadal, and monthly R of ARC2 SPP were 0.8, 0.92, and 0.94, respectively. The R of CHIRPS (TMPA 3B42v7) is 0.53 (0.63), 0.82 (0.83), and 0.93 (0.95). The daily RMAEs for ARC2, CHIRPS, TMPA 3B42v7, and PERSIANN-CDR are 0.73, 1.27, 1.17, and 1.04, while the RRMSEs are 2.94, 4.52, 4.03, and 3.85, respectively. PBIAS indicates that all the SPPs except CHIRPS underestimated precipitation. It is shown that ARC2 and CHIRPS are better in reproducing the occurrence frequency of daily events for the different intensity ranges. Also, ARC2 was the best performing SPP with regard to the ability to detect precipitation events. It is observed that all SPPs were able to detect low-intensity precipitation events but poor in detecting intensity events higher than 25 mm/day. Overall, ARC2 provided the most accurate precipitation estimates, although CHIRPS and TMPA 3B42v7 also showed good performance in reproducing and detecting precipitation.