Probabilistic multi-fidelity climate model parameterization for better generalization
Abstract
Machine-learning-based parameterizations (i.e. representation of sub-grid processes) of global climate models have recently been proposed as a powerful alternative to physical, but empirical, representations, offering a lower computational cost and higher accuracy. Yet, those approaches still suffer from a lack of generalization and extrapolation beyond the high-fidelity training data, which is however critical to projecting climate change. Here we show that a multi-fidelity approach can provide the best of both worlds: it generalizes to a warmer climate for which high fidelity data are not available, guided by abundant low-fidelity data from an even warmer climate, while retaining the accuracy of high-fidelity data, by learning a nonlinear mapping from low-to high-fidelity. In an application to climate modeling, the multi-fidelity framework yields more accurate climate projections without requiring major increase in computational resources, while providing uncertainty estimates that increase coherently with prediction error. Even though the multi-fidelity model was only tested offline, our approach paves the way for the use of machine-learning based methods that can optimally leverage historical observations or high-fidelity simulations and extrapolate to unseen regimes such as warmer climates.
Related
- Regional climate risk assessment from climate models using probabilistic machine learning
- EC-EarthFlow: Probabilistic emulation of daily transient global climate model simulations with flow matching
- PepC-Global: A Basin-Tuned Probabilistic Tropical Cyclone Model with Enhanced Out-of-Sample Skill and Climate-Sensitive Over-Land Decay
- Hybrid models generalize better to warmer climate conditions than process-based and purely data-driven models
- ClimateBench v2.0: Probabilistic Climate Model Benchmarking
BibTeX
@article{balogh2026probabilistic,
title = {{Probabilistic multi-fidelity climate model parameterization for better generalization}},
author = {Blanka Balogh and Mohamed Aziz Bhouri and Liran Peng and Michael S Pritchard and Pierre Gentine},
journal = {Machine Learning: Earth},
year = {2026},
month = oct,
doi = {10.1088/3049-4753/aea999},
url = {https://meteofrance.hal.science/meteo-05796490v1},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Probabilistic multi-fidelity climate model parameterization for better generalization
AU - Blanka Balogh
AU - Mohamed Aziz Bhouri
AU - Liran Peng
AU - Michael S Pritchard
AU - Pierre Gentine
PY - 2026
DA - 2026/10/12
JO - Machine Learning: Earth
DO - 10.1088/3049-4753/aea999
UR - https://meteofrance.hal.science/meteo-05796490v1
AB - Machine-learning-based parameterizations (i.e. representation of sub-grid processes) of global climate models have recently been proposed as a powerful alternative to physical, but empirical, representations, offering a lower computational cost and higher accuracy. Yet, those approaches still suffer from a lack of generalization and extrapolation beyond the high-fidelity training data, which is however critical to projecting climate change. Here we show that a multi-fidelity approach can provide the best of both worlds: it generalizes to a warmer climate for which high fidelity data are not available, guided by abundant low-fidelity data from an even warmer climate, while retaining the accuracy of high-fidelity data, by learning a nonlinear mapping from low-to high-fidelity. In an application to climate modeling, the multi-fidelity framework yields more accurate climate projections without requiring major increase in computational resources, while providing uncertainty estimates that increase coherently with prediction error. Even though the multi-fidelity model was only tested offline, our approach paves the way for the use of machine-learning based methods that can optimally leverage historical observations or high-fidelity simulations and extrapolate to unseen regimes such as warmer climates.
SN - 3049-4753
SN - 3049-4753
ER -