Flow Matching for Convective-Scale Precipitation Downscaling
Researchers have developed a new generative machine learning model using flow matching for downscaling precipitation data. This model was trained to increase the resolution of daily precipitation from 8 km to 2 km over a specific region. When benchmarked against a diffusion model, the flow matching approach demonstrated superior spatial skill in capturing precipitation patterns, though it slightly underestimated extreme rainfall events. AI
IMPACT This research introduces a novel generative model that could improve climate and weather forecasting accuracy.