Attempts to more accurately forecast weather are beginning to use machine learning technology – or artificial intelligence – more extensively.
Historical forecasts and observations are used to train machine learning models to find past forecasts that are most similar to a current target forecast. These past observations are used as inputs to generate ensemble forecasts without the same high costs and large computational resources used for current ensemble forecasts.
Three experimental artificial intelligence (AI)-driven models, though without available ensembles, are now available for public viewing by the European Center for Medium-range Weather Forecasting (ECMWF). These models include the FourCastNet, GraphCast, and Pangu-Weather machine learning models. They are trained by the ERA5 reanalysis, which is produced by the Copernicus Climate Change Service and implemented by ECMWF.
According to a manuscript submitted by lead author Zied Ben-Bouallegue at ECMWF, these models are already showing encouraging results with some displaying skill comparable to the ECMWF Integrated Forecasting System (IFS).
Forecast results for all three of these AI-driven machine learning models are available on the charts page for the ECMWF for various parameters and regions worldwide.