EC-EarthFlow: Probabilistic emulation of daily transient global climate model simulations with flow matching

2026-10-07

AI Summary

This study addresses the high computational cost of physical climate models and their inefficiency in simulating global daily transient temperatures by proposing a generative autoregressive forecasting framework based on flow matching. Trained on EC-Earth3 data, the model predicts next-day states using only the temperature field as input, enabling probabilistic climate simulation. Results demonstrate that the approach accurately reproduces the spatial distribution, annual cycles, and long-term trends of surface air temperature while maintaining stability during extended rollouts. The core contribution lies in showing that data-driven methods can effectively learn complex physical relationships, preserving high fidelity while substantially reducing computational demands, thereby offering a new paradigm for low-cost climate modeling.

Abstract

We introduce EC-EarthFlow, a generative flow matching model that emulates simulations from the physical climate model EC-Earth3. The model is trained on transient simulations from EC-Earth3 (1950-2166, SSP2-4.5) to predict the day ahead temperature field from the previous days temperature as well as annual mean temperature. Predictions are made auto-regressively with rollout periods of between a month and an extended season. Using only this variable of interest, we are able to reproduce the daily variability, spatial patterns, annual cycle and long-term trend from EC-Earth3 at a substantially lower computational cost than the physical model. We demonstrate that EC-EarthFlow is stable for long inference periods, and that it can learn the physical relationships as simulated in EC-Earth3.

Problem

• climate model emulation
• flow matching
• transient simulations
• computational cost
• daily variability

Innovation

• flow matching
• generative model
• climate emulation
• autoregressive prediction
• probabilistic modeling