Optimal scenario design for climate emulation
Abstract
As deep learning for physical systems continues to grow in popularity, efforts to improve generalizability have primarily focused on designing architectures that embed physical constraints. However, for machine-learning surrogate climate models (emulators), we show that the low structural diversity in existing scenarios commonly used to generate training data places a ceiling on predictive skill. Here, we examine whether training datasets themselves can be optimized to improve generalization. We introduce a method to create datasets that produce emulators capable of generalizing to new, structurally different scenarios absent from the training data. We use a differentiable Simple Climate Model (SCM) to calculate the sensitivity of emulator loss to perturbations in the training data, iteratively updating the training data to maximize emulator skill. For an SCM, training on one scenario optimized in this fashion outperforms an emulator trained on six standard ScenarioMIP pathways. We achieve this higher predictive skill despite training on a smaller dataset, finding that our emulator successfully isolates distinct physical behaviors of different climate forcing agents (e.g., greenhouse gases vs. aerosols) without single-forcing runs. We then demonstrate that scenarios optimized using an SCM, when used to drive an intermediate-complexity climate model, produce a training dataset that yields a more skillful emulator than training on ScenarioMIP outputs. Our results suggest that, in the compute-constrained environment of running full-scale climate models, generating a small number of dynamically rich scenarios provides greater marginal value for emulation and characterizing system responses than expanding the suite of traditional emissions pathways.
Related
- An Artificial Intelligence Integrated Geospatial Decision-Support Framework for Scenario-Based Climate-Adaptive Land-Use Optimization and Environmental Risk Reduction
- EC-EarthFlow: Probabilistic emulation of daily transient global climate model simulations with flow matching
- Variable Masking for Climate Emulation in ACE
- Rescene: stable century-long climate emulation from a frozen weather model
- METEORv1.6: Spatial climate variability and integrated impact emulation
BibTeX
@misc{womack2026optimal,
title = {{Optimal scenario design for climate emulation}},
author = {Christopher B. Womack and Shahine Bouabid and Andrei Sokolov and Popat Salunke and Glenn Flierl and Sebastian D. Eastham and Noelle E. Selin},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2606.19302},
eprint = {2606.19302},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2606.19302},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Optimal scenario design for climate emulation
AU - Christopher B. Womack
AU - Shahine Bouabid
AU - Andrei Sokolov
AU - Popat Salunke
AU - Glenn Flierl
AU - Sebastian D. Eastham
AU - Noelle E. Selin
PY - 2026
DA - 2026/09/29
PB - arXiv
DO - 10.48550/arxiv.2606.19302
UR - https://arxiv.org/abs/2606.19302
AB - As deep learning for physical systems continues to grow in popularity, efforts to improve generalizability have primarily focused on designing architectures that embed physical constraints. However, for machine-learning surrogate climate models (emulators), we show that the low structural diversity in existing scenarios commonly used to generate training data places a ceiling on predictive skill. Here, we examine whether training datasets themselves can be optimized to improve generalization. We introduce a method to create datasets that produce emulators capable of generalizing to new, structurally different scenarios absent from the training data. We use a differentiable Simple Climate Model (SCM) to calculate the sensitivity of emulator loss to perturbations in the training data, iteratively updating the training data to maximize emulator skill. For an SCM, training on one scenario optimized in this fashion outperforms an emulator trained on six standard ScenarioMIP pathways. We achieve this higher predictive skill despite training on a smaller dataset, finding that our emulator successfully isolates distinct physical behaviors of different climate forcing agents (e.g., greenhouse gases vs. aerosols) without single-forcing runs. We then demonstrate that scenarios optimized using an SCM, when used to drive an intermediate-complexity climate model, produce a training dataset that yields a more skillful emulator than training on ScenarioMIP outputs. Our results suggest that, in the compute-constrained environment of running full-scale climate models, generating a small number of dynamically rich scenarios provides greater marginal value for emulation and characterizing system responses than expanding the suite of traditional emissions pathways.
ER -