Conditional Diffusion Models for Global Precipitation Map Inpainting

Abstract

Incomplete satellite‐based precipitation presents a significant challenge in global monitoring. For example, the Global Satellite Mapping of Precipitation (GSMaP) from the Japan Aerospace Exploration Agency (JAXA) contains regions without valid microwave observations because Low Earth orbiting (LEO) satellites cannot observe the entire globe simultaneously, and its current interpolation methods often result in spatial discontinuities. In this study, we formulate the completion of the precipitation map as a video inpainting task and propose a machine learning approach based on conditional diffusion models. Our method employs a three‐dimensional (3D) U‐Net with a 3D condition encoder to reconstruct complete precipitation maps by leveraging spatio‐temporal information from infrared images, latitude‐longitude grids, and physical time inputs. Training was carried out on hourly precipitation data from the fifth‐generation European Center for Medium‐Range Weather Forecasts (ECMWF) reanalysis (ERA5) from 2020 to 2023. We generated a pseudo‐GSMaP data set by randomly applying GSMaP masks to the ERA5 maps. Performance was evaluated for the calendar year 2024, and our approach produces more spatio‐temporally consistent inpainted precipitation maps compared to conventional methods. These results indicate the potential to improve global precipitation monitoring using the conditional diffusion models.