Characterising Sensor-Level Errors of Global Satellite Precipitation Estimates for Different Rainfall Events

Abstract

Revealing sensor-level errors of global satellite precipitation estimates for different rainfall events is vital for understanding the error sources and components of ultimate satellite precipitation products at various rainfall events and is beneficial for improving the availability of satellite precipitation products in intensity-based hydro-meteorological applications such as drought analysis, extreme precipitation, typhoon monitoring and so on. However, investigations into sensor-level errors for various rainfall events are still lacking. To address this research gap, this study investigated the sensor-level errors in the Global Satellite Mapping of Precipitation for Global Precipitation Measurement (GPM-GSMaP) for mainland China by separating the total rainfall into six rainfall events (i.e., light rainfall, moderate rainfall, heavy rainfall, rainstorm, heavy rainstorm and extraordinary storm). The results indicated that the multi-sensor precipitation merging approach effectively reduces individual sensor errors but also inevitably propagates the shortcomings of various sensors into merging precipitation estimates, making it not the best way in most rainfall events. The performance rankings of the sensors varied depending on the error metrics, rainfall events, topography categories and climate types. In measuring light rainfall, the Advanced Microwave Sounding Unit-A/Microwave Humidity Sounder (AMSU-A/MHS) and infrared sensors were the major error sources of satellite precipitation products in most areas. Additionally, the AMSU-A/MHS sensors showed large normalised root mean square errors (> 3.5) and biases (> 80%) in estimating light, moderate, and heavy rainfall events in coastal areas and were dominant contributors that resulted in high measurement uncertainty of ultimate satellite precipitation products in coastal areas. As a core GPM sensor, the GPM Microwave Imager (GMI) sensor showed the worst performance in capturing light, moderate and heavy rainfall events, demonstrating that its current retrieval algorithms failed to leverage the hardware advantage fully. Finally, our research results highlighted that inversion algorithms of the satellite sensors need to consider the impact of different rainfall events on the inversion results to improve the accuracy of the sensor inversion.