How Do AI Climate Models Respond to Warming Across Climate Zones?
In one sentence. Köppen-Geiger diagnostic of AIMIP AI climate emulators under +4K warming shows only hybrid NeuralGCM-HRD reorganizes climate zones consistently with thermodynamic scaling.
Abstract
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.
Labels
- Topic
- Climate emulation, Benchmarks, datasets & evaluation
- Type
- Evaluation or intercomparison of existing models
- Time scale
- Climate projections / centennial
- Domain
- Global
- Data
- ERA5 / reanalysis, CMIP / climate model output, Other
- Status
- Research
- Identifiers
- arXiv:2608.17986 OpenAlex:W7203755093
- Added
- 2026-10-22
Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.