Global spatio-temporal ERA5 precipitation downscaling to km and sub-hourly scale using generative AI

Abstract

The spatial and temporal distribution of precipitation significantly impacts human lives. While reanalysis datasets provide consistent long-term global precipitation information that allows investigations of rainfall-driven hazards like larger-scale flooding, they lack the resolution to capture the high spatio-temporal variability of precipitation and miss intense local rainfall events. Here, we introduce spateGAN-ERA5, the first deep learning-based spatio-temporal downscaling of precipitation data on a global scale. SpateGAN-ERA5 enhances ERA5 precipitation data from 24 km and 1 h to 2 km and 10 min, delivering high-resolution rainfall fields with realistic spatio-temporal patterns and accurate rain rate distribution, including extremes. Its computational efficiency enables the generation of a large ensemble of solutions, addressing uncertainties inherent to downscaling challenges and supports practical applicability for generating high-resolution precipitation data for arbitrary ERA5 time periods and regions on demand. Trained solely on data from Germany and validated in the US and Australia, considering diverse climates, including tropical rainfall regimes, spateGAN-ERA5 demonstrates strong generalization, indicating robust global applicability. It fulfills critical needs for high-resolution precipitation data in hydrological and meteorological research.