Reducing GPM IMERG uncertainty with ANN models using GK-2A IR data and atmospheric indices over South Korea

The frequent occurrence of extreme weather events has drawn increasing attention to climate change. Torrential rainfall in East Asia, including Korea, causes natural disasters with significant effects on human life and property. The Integrated Multi-Satellite Retrievals for Global Precipitation Measurement (IMERG) product provides global precipitation data by combining multiple satellite-based observations, including microwave, infrared (IR), and other complementary datasets from various satellites. However, it has limitations in accurately describing precipitation in Korea, which is influenced by complex seasonal and topographical characteristics. In this study, we used IMERG precipitation data in conjunction with optical satellite data to estimate various indicators of atmospheric conditions. Data used to improve precipitation estimation included the 500-hPa geopotential height, atmospheric instability index, horizontal wind field, and water vapour (WV 6.9), IR 11.2 and IR 12.3 infrared channels of the Geostationary Korea Multi-Purpose Satellite 2A (GK-2A). These data were input into a deep learning model that was used to estimate precipitation. The results showed that 500-hPa geopotential height had a significant impact on the performance of the artificial neural network (ANN). Atmospheric data from optical satellites are anticipated to compensate for the shortcomings of IMERG, to improve the accuracy of remote sensing precipitation estimation.