Performance evaluation of multiple satellite rainfall products in mountainous catchment of Eastern Zimbabwe

Abstract

Modelling of hydrological processes in mountainous terrain is challenging and remains complex. Addressing this challenge is complicated by the non-existence or sparse distribution rain gauges. In cases where the rain gauges exist, the data is incomplete and only describe point estimates yet the mountainous topography greatly influences the spatial distribution of rainfall. Satellite rainfall estimates (SREs) present an alternative source of rainfall data since estimates are continuous, timely and spatially distributed. This work provides a first performance evaluationof four high-resolution satellite rainfall estimates from Climate Hazards Group Infrared Precipitation with Station Data (CHIRPS), Climate Prediction Center Morphing Technique (CMORPH), Precipitation Estimation from Remotely Sensed Information using Artificial Neural Network-Cloud Classification System (PERSIANNCCS) and PERSIANN-Dynamic Infrared Rain Rate near real-time (PDIR-Now). Satellite rainfall estimates were compared against ground-based measurements through statistical performance indicators and a rainfall-runoff model at a daily timescale. A daily ground-based rainfall dataset (1990–2019) obtained from Meteorological Services Department of Zimbabwe was used to evaluate the performance of SREs. Runoff data (2000–2009) obtained from Zimbabwe National Water Authority (ZINWA) was used to calibrate and validate the HEC-HMSrainfall-runoff model at three gauging stations namely. F7, F11 and F18 runoff stations along Rusitu, Nyahode and Buzi rivers respectively. Statistical performance indices applied include Pearson correlation coefficient (r), Nash Sutcliff Efficiency NSE and bias. Statistical results show that SREs are weakly correlated to rain gauge-based measurements with CMORPH exhibiting a better correlation (r = 0.34). Hydrological performance evaluation results indicate an overestimation of flow by CMORPH-driven simulations with percentage bias (PBIAS = 3.1%). An underestimation of flow was observed in CHIRPS-driven, PDIR-Now-driven and PERSIANNCCS-driven simulations which had PBIAS values of -10.4%, -12.1% and -13.6% respectively. A similar pattern was observed in RVE values obtained from SREs-driven simulations. The study provides valuable perspectivesfor hydrological modelling in mountainous catchments in the region where data ground-based rainfall measurements are limited or unevenly distributed.