Uncertainty Quantification of Multi-Satellite Precipitation Products with Deep Learning: A Case Study over Taiwan

Abstract

Uncertainties prevalently exist in satellite precipitation retrieval processes at different stages since passive microwave (PMW) sensors and infrared (IR) channels have difficulties in resolving shallow and/or heavy rain. These uncertainties often lead to underestimation or overestimation of rainfall in various composite satellite products. In this study, we developed a deep convolutional neural network (CNN) model to characterize the uncertainties in satellite precipitation retrievals, and a demonstration study over Taiwan has been performed. In particular, precipitation estimates derived from the ground-based operational Quantitative Precipitation Estimation and Segregation Using Multiple Sensors (QPESUMS) system are used as reference labels in training the deep CNN model to correct various satellite precipitation products, including NOAA Climate Prediction Center (CPC) morphing technique (CMORPH), NASA Global Precipitation Measurement (GPM) Integrated Multi-satellitE Retrievals for GPM (IMERG), and Taiwan Central Weather Bureau Infrared Quantitative Precipitation Estimation (IRQPE). The results show that the designed machine learning model is capable of capturing and correcting the precipitation pattern and intensity derived from satellite retrievals.