When Physics Gets It Almost Right: Building an ML Correction Layer for ECMWF Temperature Forecasts (opens in new tab)

How I used 5 million operational records, a Random Forest, and a FastAPI service to shave systematic bias off numerical weather predictions