Description
A revolution is underway in weather prediction, with AI providing new opportunities to produce high-quality forecasts. AI-driven approaches leverage decades of investment into physics-based models that have generated multi-decadal datasets ideal for machine learning. Data-driven weather forecasts have been shown to be more skilful for lead times of up to 10–15 days, including applications such as extreme-event prediction and renewable-energy planning. AI-based solutions transform a task that typically requires a high-performance computing system into one that can be run in minutes on a single GPU. ECMWF, the European Centre for Medium-Range Weather Forecasts, is pioneering the development and use of these data-driven models for operational forecasts. Within Destination Earth, ECMWF is further developing this to expand data-driven models to other components of the Earth system—such as land, ocean and sea ice—as well as to longer, climate-relevant timescales. We will showcase the revolution, discuss the methodologies, highlight results, and demo the ability to easily run a complete forecasting system on varied computing infrastructures (“forecast in a box”), opening the door to wider global access and broader democratisation of advanced forecasting capabilities.
