Item

Abstract

We present our recent paper “GraphCast: Learning skillful medium-range global weather forecasting” (https://arxiv.org/abs/2212.12794). GraphCast is a machine-learning (ML) based weather simulator, trained from the ERA5 reanalysis archive, which can make forecasts, at 6-hour time intervals, of five surface variables and six atmospheric variables (37 vertical pressure levels), on a 0.25-degree grid (~25 km at the equator). GraphCast can generate a 10-day forecasts (35 gigabytes of data) in under 60 seconds, while outperforming ECMWF's deterministic operational forecasting system, HRES, on 90.0% of the 2760 variable and lead time combinations we evaluated, as well as all other ML baselines. These results represent a key step forward in complementing and improving weather modeling with ML, opening new opportunities for fast, accurate forecasting. In this talk we will go into the details of the model architecture, as well as providing a detailed evaluation against HRES.