ArchesClimate: Probabilistic Decadal Ensemble Generation With Flow Matching

2025-09-19

AI Summary

Climate prediction suffers from substantial uncertainty due to system complexity and multiscale interactions, while physics-based ensemble simulations incur prohibitive computational costs. This work proposes the first flow-matching deep learning surrogate model tailored for near-term (decadal-scale) climate prediction, trained on retrospective forecasts from the IPSL-CM6A-LR Earth system model at 2.5°×1.25° resolution. The model enables autoregressive generation of arbitrarily long, physically consistent monthly state sequences, with outputs highly interchangeable with those of the original model. Experiments demonstrate excellent statistical fidelity to IPSL-CM6A-LR over ten-year integrations for key variables—including temperature and precipitation—while reducing computational cost by approximately two orders of magnitude. To our knowledge, this is the first application of flow matching to climate surrogate modeling, establishing a novel paradigm for efficient, reliable large-scale climate ensemble forecasting.

Technology Category

• Machine Learning: Deep Neural Architectures and Foundation Models
• Natural Language Processing: (Large) Language Models
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• User Modeling, Personalization and Recommendation: ML for personalized search and recommendations
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Climate projections have uncertainties related to components of the climate system and their interactions. A typical approach to quantifying these uncertainties is to use climate models to create ensembles of repeated simulations under different initial conditions. Due to the complexity of these simulations, generating such ensembles of projections is computationally expensive. In this work, we present ArchesClimate, a deep learning-based climate model emulator that aims to reduce this cost. ArchesClimate is trained on decadal hindcasts of the IPSL-CM6A-LR climate model at a spatial resolution of approximately 2.5x1.25 degrees. We train a flow matching model following ArchesWeatherGen, which we adapt to predict near-term climate. Once trained, the model generates states at a one-month lead time and can be used to auto-regressively emulate climate model simulations of any length. We show that for up to 10 years, these generations are stable and physically consistent. We also show that for several important climate variables, ArchesClimate generates simulations that are interchangeable with the IPSL model. This work suggests that climate model emulators could significantly reduce the cost of climate model simulations.

Problem

• Reducing computational cost of climate ensemble generation
• Quantifying uncertainties in decadal climate projections
• Emulating climate model simulations with deep learning

Innovation

• Flow matching model for climate emulation
• Deep learning-based decadal ensemble generation
• Auto-regressive simulation with physical consistency