A weather model that keeps up with the supercomputers

orig. “GraphCast: Learning skillful medium-range global weather forecasting” · Remi Lam, Alvaro Sanchez-Gonzalez, Matthew Willson, Peter Battaglia

AI for Climate Intermediate 4 min read Written, reviewed by Marginalia Editorial

In the margin

Heads up

This model predicts ten days of global weather in under a minute, and it is often more accurate than the traditional forecast.

What's going on

Standard forecasts run huge physics simulations on supercomputers. GraphCast instead learned from decades of past weather and predicts the next steps directly. It produces a ten-day global forecast in well under a minute, and in tests it matched or beat the leading physics-based system on most measures, including the tracks of major storms.

Why it matters

Faster, cheaper forecasts help with everything from daily planning to early warnings for extreme weather. It also shows learned models can rival decades of carefully built physics code in a serious scientific field. That makes it a strong example of AI pointed at a real-world problem.

Source

Remi Lam, Alvaro Sanchez-Gonzalez, Matthew Willson, Peter Battaglia, Google DeepMind

View on arXiv ↗ Read the PDF ↗

Still fuzzy on a sentence?

Paste it and we'll explain it even more simply.

Test your understanding

Member notes Sign in ↗

ME Marginalia Editorial TEAM

The editorial team picked this paper because it is worth your time, and an AI drafted this note. We check each one and rewrite it when the draft misses the mark.