Evaluating performance of 20 global and quasi-global precipitation products in representing drought events in Ethiopia I: Visual and correlation analysis

Abstract

Increased availability and access to satellite observation and the advanced blending techniques for generating reanalysis data have improved accuracy of precipitation products for data poor regions. However, all data products should be evaluated for their performance before being used for specific applications in different regions or locations. In this study, we evaluated 20 precipitation products for their drought monitoring performance over Ethiopia. These datasets are produced by different state-of-the-art techniques, including: 1) gauge-interpolated, 2) satellite-estimate, 3) reanalysis and 4) merged from multi-sources. The Ethiopian gauge-satellite gridded precipitation dataset and gauge records from 126 stations were used for ground truthing. Drought indices were generated for all data products using the Standardized Precipitation Index (SPI) method at 3- and 12-month time scales. We evaluated accuracy of the data products in representing occurrence and spatial and temporal patterns of the 3- and 12-month droughts using visual and correlation analysis techniques. The data were visually compared against reference data in representing four selected major drought episodes in the country (1984, 2002, 2009 and 2015). The Spearman's correlation was used to quantify relationships between the 20 precipitation products and two reference datasets using SPI values generated at 3- and 12-month time scales. Results showed considerable discrepancies and poor performance for most datasets. The ability to represent the spatial pattern and severity of major drought events varied between drought years. Although the correlation result for areal average SPI time series of the 3- and 12-month droughts for all data products showed statistically significant correlation with the reference data, there is discrepancy between the data products across space. Overall, only three out of the 20 products (CHIRPS, FLDAS and GPCC) performed relatively better. Our results provide important information to guide the choice of precipitation products for drought research, and operational drought risk management as well as useful feedback to data developers to further improve their products.