Score-Based Generative Emulation of Impact-Relevant Earth System Model Outputs

Abstract

Policy targets evolve faster than the Couple Model Intercomparison Project cycles, complicating adaptation and mitigation planning that must often contend with outdated projections. Climate model output emulators address this gap by offering inexpensive surrogates that can rapidly explore alternative futures while staying close to Earth system model (ESM) behavior. The focus is on emulators designed to provide inputs to impact models. 

Using monthly ESM fields of near-surface temperature, precipitation, relative humidity, and wind speed, it is shown that deep generative models have the potential to model the joint distribution of variables relevant for impacts. The specific model proposed uses score-based diffusion on a spherical mesh and runs on a single mid-range graphical processing unit. A thorough suite of diagnostics is introduced to compare emulator outputs with their parent ESMs, including their probability densities, cross-variable correlations, time of emergence, or tail behavior. The emulator performance is evaluated across three distinct ESMs in both pre-industrial and forced regimes. 

The results show that the emulator produces distributions that closely match the ESM outputs and captures key forced responses. They also reveal important failure cases, notably for variables with a strong regime shift in the seasonal cycle. Although not a perfect match to the ESM, the inaccuracies of the emulator are small relative to the magnitude of internal variability in ESM projections. This suggests that the generative emulators can be useful in supporting impact assessment. Priorities for future development toward daily resolution, finer spatial scales, and bias-aware training are discussed.

Plain Language Summary

Climate projections produced by Earth system models (ESMs) guide adaptation and mitigation planning, but running ESMs is slow and expensive. This makes it difficult to generate new projections fast enough to keep pace with evolving policy targets. Climate model output emulators—statistical surrogates trained to reproduce the behavior of ESMs at a fraction of the computational cost—have emerged to fill this gap. In this work, we introduce an emulator that uses score-based diffusion, a modern generative AI method, to replicate the behavior of ESMs in projecting monthly temperature, precipitation, humidity, and wind altogether. It runs efficiently on a single mid-range graphics card and produces results that closely match the original models across pre-industrial and future climate regimes. We also identify failure modes, notably for variables with strong seasonal shifts. Overall, we show that errors remain small compared with the natural variability of ESM projections, suggesting this approach may offer a fast, flexible tool to support physical risk assessment and inform evolving climate policies.