EVALUATION OF GLOBAL ERA5 REANALYSIS PRECIPITATION DATASET FOR STREAMFLOW SIMULATION OVER THE UPPER PANGANI RIVER BASIN IN TANZANIA

Abstract

Global reanalysis precipitation products are crucial datasets for estimating precipitation over ungauged terrain, remote and inaccessible area. The study's objective intended to assess the European Centre for Medium-Range Weather Forecasts freshly published fifth-generation reanalysis products (ECMWF), i.e., ERA5, for streamflow simulation of the Pangani River Basin (PRB), Tanzania during 2010-2019 using the HEC-HMS model. This study assessed the ERA5 reanalysis precipitation reliability from two adjacent and opposite watersheds using the observed rain gauge. Both continuous and categorical verification statistics were calculated, and the results show better performance across distinct time scales. The most satisfactory precipitation agreement occurs during monthly than daily timescale in the Ruvu watershed located in lower elevation range. According to the error analysis, the ERA5 product has a more significant root mean square error (RMSE) values in the high elevation range Kikuletwa watershed (~5200 m). The mean bias error (MBE) and inter-annual basin average indicating a higher overestimated ERA5 in the Kikuletwa watershed, but with almost precisely matches in the Ruvu watershed for both rainy and dry seasons. In contrast to categorical verification indices, the results show the advanced detection capacity of ERA5 for all watersheds with variation from light, moderate and heavy precipitation. For example, the higher detection in the Kikuletwa watershed lies between 0.5 mm/day and greater than 10 mm/day, and that of the Ruvu watershed was from 0.5-10 mm/day. Furthermore, the ERA5 reanalysis' detection capacity follows a distinct seasonal trend, making it more challenging to detect moderate or higher daily precipitation occurrences (more than 10 mm/day). The hydrological model, HEC-HMS, satisfactorily projected peak discharge, volume and the observed hydrograph pattern across the upper Pangani river basin. The objective functions in terms of trends (NSE = 0.64-0.86); residual variation (RSR = 0.4-0.6); average magnitude (PBIAS < ± 10) and error variance (R2 = 0.71-0.89); also shows good model performance in line with a visual inspection from ERA5 in both watersheds. However, the hydrographs revealed an underestimate throughout the streamflow simulation period (2012-2019) in the Ruvu watershed. The global precipitation dataset such as ERA5 is helpful for water resources assessment for the data scare regions and other parts of the world.