Chapter 09 · Sustainability & AISustainability & AI

AI weather forecasting

Meaning statusEstablishedSource recordDirect document linkedWhy these are different

Definition

Machine-learning models that predict weather from atmospheric data, now matching or exceeding conventional numerical prediction on key metrics at a fraction of the computing cost.

References

Science 2023 / arXivLam et al. — GraphCast: skillful medium-range global weather forecasting

This source provides part of the technical or institutional basis for the definition.

Nature 2023Bi et al. — Pangu-Weather: 3D neural networks for global weather forecasting

This source supports the explanation of how the term is applied, measured or governed in practice.

ECMWFAIFS operational machine-learning forecasting (Feb 2025)

This source supports the wider sustainability significance and context described in the entry.

ECMWF / arXivLang et al. — AIFS: ECMWF's data-driven forecasting system

This source supports the wider sustainability significance and context described in the entry.

Overview

What it means

Landmark systems include GraphCast (Science 2023) and Pangu-Weather (Nature 2023), trained on decades of reanalysis data. The field crossed into operations in February 2025, when ECMWF began running its machine-learning model AIFS alongside its physics-based system, adding an ensemble version later that year.

How it is used

Forecasts support renewable-energy scheduling, agriculture, disaster preparedness and logistics. AI systems generate medium-range global forecasts in seconds on modest hardware, enabling wider access and experimentation by national meteorological services.

Why it matters

Weather prediction is the first planetary-scale physical system where AI has matched supercomputing incumbents — a proof point for AI-for-climate claims, and a template for hybrid physical–statistical modelling elsewhere in Earth science.