This website requires JavaScript.

GraphCast: Learning skillful medium-range global weather forecasting

Remi LamAlvaro Sanchez-GonzalezMatthew Willson ...+14 Peter Battaglia
Dec 2022
摘要
We introduce a machine-learning (ML)-based weather simulator--called"GraphCast"--which outperforms the most accurate deterministic operationalmedium-range weather forecasting system in the world, as well as all previousML baselines. GraphCast is an autoregressive model, based on graph neuralnetworks and a novel high-resolution multi-scale mesh representation, which wetrained on historical weather data from the European Centre for Medium-RangeWeather Forecasts (ECMWF)'s ERA5 reanalysis archive. It can make 10-dayforecasts, at 6-hour time intervals, of five surface variables and sixatmospheric variables, each at 37 vertical pressure levels, on a 0.25-degreelatitude-longitude grid, which corresponds to roughly 25 x 25 kilometerresolution at the equator. Our results show GraphCast is more accurate thanECMWF's deterministic operational forecasting system, HRES, on 90.0% of the2760 variable and lead time combinations we evaluated. GraphCast alsooutperforms the most accurate previous ML-based weather forecasting model on99.2% of the 252 targets it reported. GraphCast can generate a 10-day forecast(35 gigabytes of data) in under 60 seconds on Cloud TPU v4 hardware. Unliketraditional forecasting methods, ML-based forecasting scales well with data: bytraining on bigger, higher quality, and more recent data, the skill of theforecasts can improve. Together these results represent a key step forward incomplementing and improving weather modeling with ML, open new opportunitiesfor fast, accurate forecasting, and help realize the promise of ML-basedsimulation in the physical sciences.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答