Bradley Stanley-Clamp

arXiv2026

How Do AI Climate Models Respond to Warming Across Climate Zones?

Regional climate zones are expected to shift under global warming. Whether AI climate models have learned to generalize climate-zone distributions under warming in a physically meaningful way affects their suitability for climate projection. We address this question by applying a Köppen-Geiger climate-zone decomposition to AIMIP Phase 1 models under prescribed +4K SST forcing and comparing their responses to physics-based AMIP models. Using this diagnostic, we compare baseline classification skill, per-zone responses in temperature, precipitation, and near-surface specific humidity, and the spatial structure of departures from physics-based models. All AI models considered reproduce the 1979-2014 ERA5 climatology within the physics-based models' range, but only the hybrid physics-AI model NeuralGCM-HRD reorganizes zones in agreement with established thermodynamic and hydrological scaling relations. The remaining emulators have distinct failure modes traceable to their architectural treatment of land cells. A physically consistent climate-zone response is therefore necessary for AI models intended for climate projection.

physics.ao-ph↗

arXiv2026

No Epoch Like the Present: Robust Climate Emulation Requires Out-of-Distribution Generalisation

Climate emulation is an out-of-distribution (OOD) projection task. This is precisely the challenge where modern Machine Learning (ML) methods are most prone to failure. Consequently, while current ML emulators trained on present climate achieve high in-distribution performance, their future reliability under the inevitable distribution shifts of a changing climate remains a critical, poorly understood blind spot. Addressing this challenge requires a fundamental shift in how we understand, evaluate, and design climate emulators. In this work, we first confirm that climate change drives a statistically significant and progressively growing shift in atmospheric state distributions, rendering standard evaluation protocols insufficient. We empirically establish that seasonal variation serves as an effective proxy for these long-term climate shifts, providing access to $\textit{real-world}$ distribution shifts without recourse to heuristics like synthetic perturbations. Motivated by this link, we introduce a novel evaluation framework that leverages seasonal shifts as a rigorous, zero-overhead testbed for emulator robustness. Our systematic characterisation confirms that current state-of-the-art hybrid-ML emulators degrade significantly under these realistic shifts. Finally, we chart a path forward by identifying compositional generalisation, the ability to form novel combinations from observed elementary components, as a principled route towards robust climate emulation. We demonstrate that physically motivated decompositions substantially improve OOD performance with only modest trade-offs against in-distribution performance, providing an avenue towards ML-driven climate emulators robust to an unknown future.

cs.LG↗