Evaluation and modelling of accuracy of satellite‑based CHIRPS rainfall data in Ruvu subbasin, Tanzania

In data scarce regions, satellite-derived products can be used as alternatives to ground observed rainfall. However, satellite-derived data quality assurance and its quick use are of concern. In this study, Climate Hazards Infra-Red Precipitation with Station data (CHIRPS) satellite-based rainfall data were evaluated against ground observations during 1981–2015 in Ruvu subbasin. Commonly, a BIAS threshold is used for accuracy evaluation. However, for the rainfall data, there are no dedicated BIAS threshold for accuracy evaluation of rainfall products against insignificant differences from observations. Therefore, the study evaluated CHIRPS data and developed regression modelling to capture both measures of accuracy for establishment of the BIAS threshold. The model was calibrated and validated using rainfall gauging stations in and outside the subbasin. Results showed that CHIRPS monthly data correlated well with ground rainfall observations with Pearson correlation coefficient, r > 0.7 and BIAS < 25%. Based on t test, the CHIRPS data were proved to be the same as ground rainfall (p > 0.05), with exception of some stations with higher absolute percent of bias (ABIAS > 17%). The developed innovative linear model capturing both measures of accuracy proved to have high predictive skills. Subsequently, the developed linear models on a month and monthly basis also indicated ABIAS > 17% for significant differences between CHIRPS and observation data. Therefore, it is recommended to use ABIAS < 17% threshold as acceptable BIAS for CHIRPS data. Values above the threshold (> 17%) provide a warning point or level, indicating large deviation from observed rainfall data.