The explanation of small- and medium-watershed-scale bias variability in IMERG in Chinese humid regions

Abstract

The absence of in situ precipitation data in remote small and medium watersheds (SMWs) highlights the need for reliable satellite precipitation estimations (SPEs). This study evaluates and compares the updated Integrated Multi-satellitE Retrievals for Global Precipitation Measurement (IMERG) V07 Final Run uncalibrated (V07F-Uncal) and calibrated (V07F-Cal) products against their predecessors (V06F-Uncal and V06F-Cal). The comparison is conducted across 339 SMWs in Chinese humid regions during the summers from 2015 to 2020, using rain gauge observations as benchmarks. Conditional multivariate regression is employed to examine the relationships between satellite precipitation bias values and factors such as precipitation intensity, surface temperature, and fraction of vegetation cover (FVC). Results show that V07F-Uncal outperforms V06F-Uncal in terms of CC and RMSE in most mountainous and coastal SMWs, but it consistently underestimates precipitation, particularly in the mountainous regions. Bias for both V07F-Uncal and V06F-Uncal transitions from positive to increasingly negative values with rising precipitation intensity. V07F-Uncal exhibits a tighter distribution of bias values across all intensity categories compared to V06F-Uncal, but it shows a pronounced negative bias in high-intensity categories. In terms of performance metrics and the distribution of bias values, V06F-Cal demonstrates marked improvements over V06F-Uncal. However, the enhancements observed in V07F-Cal relative to V07F-Uncal are not substantial. As for the variability of bias values associated with changes in precipitation intensity, surface temperature and FVC, the explained bias variability in V07F-Uncal is significantly higher than in V06F-Uncal, averaging approximately 43 % versus 22 %. In western mountainous SMWs, this variability is also greater than in the eastern region (52 % versus 34 %). Precipitation intensity is the primary factor explaining bias variability for both V07F-Uncal and V06F-Uncal, although in specific regions, the bias variability of V06F-Uncal may relate to the surface temperature or its interaction with precipitation intensity. FVC exerts minimal influence (<3 %) on bias variability for both products. This research is crucial for improving the accuracy of SPEs in SMWs, which are vital for flood simulation and disaster adaptation in ungauged SMWs.