Improving precipitation forecasts in an AI weather model using observational data
Julian F. Schmitt and colleagues at X, The Moonshot Factory and Google, with Caltech and Stanford, fine-tune an AI weather model on observed precipitation rather than reanalysis and improve precipitation forecasts substantially.
Ask this paper
The problem is the training label. Global AI weather models train almost exclusively on the ERA5 reanalysis, which has known biases, and precipitation is where those biases are worst.
The fix is observational fine-tuning. A graph-transformer architecture is fine-tuned with IMERG precipitation data at 0.25 degree resolution, so the supervision comes from measurement rather than from another model's output.
Up to 19% improvement in medium-range continuous ranked probability score, with better skill on both tropical storms and drizzle, which sit at opposite ends of the intensity range.
Extreme rainfall improves 57% on Brier skill score against state-of-the-art operational models globally, and the authors state plainly that a physics-based operational model remains more reliable for the very heaviest events.
Reads as a companion to WeatherNext 3. Both replace reanalysis supervision with observations, and this one isolates precipitation specifically.
Abstract
Artificial intelligence weather prediction (AIWP) systems now surpass state-of-the-art physical models for medium-range weather forecasting. Current global AIWP models are trained almost exclusively using one reanalysis dataset, ERA5, but it has known biases, particularly for precipitation. Here we fine-tune a graph-transformer architecture with IMERG precipitation data at 0.25° resolution. The resulting model improves medium-range continuous ranked probability scores by up to 19%, while also demonstrating superior skill for tropical storms and drizzle events. Our model exceeds the Brier skill score of state-of-the-art operational models on extreme rainfall prediction by 57% globally; however, a physics-based operational model remains more reliable for the heaviest precipitation events. Our results demonstrate that incorporating observations-based precipitation data directly into training can substantially improve precipitation forecasts.