Extratropical Atmospheric Circulation Response to ENSO in Deep Learning Pacific Pacemaker Experiments

Abstract Coupled atmosphere–ocean deep learning (DL) climate emulators are a new frontier but are known to exhibit weak ENSO variability, raising questions about their ability to simulate its teleconnections. Here, we present the first Pacific pacemaker (PACE) experiments using a coupled DL Earth system model (DLESyM) to bypass this weak variability and isolate the atmospheric response to observed ENSO forcing. We find that while the emulator realistically captures internal atmospheric variability, it produces a significantly amplified forced teleconnection response to ENSO. This amplified response leads to biases in simulating extremes, notably an overestimation of atmospheric blocking frequency and duration and an underestimation of peak intensity. Our findings underscore that coupled DL climate models require in-depth, process-based evaluation, analogous to traditional numerical models, to build confidence in their use for climate simulation and analysis. Significance Statement Deep learning (DL) ocean–atmosphere coupled climate models offer computational efficiency for climate science but exhibit weak El Niño and La Niña variability, raising questions about whether they realistically simulate remote weather impacts. We bypassed this limitation by constraining tropical Pacific Ocean temperatures to match historical observations, directly testing the atmospheric response in a DL coupled climate model. The model captures realistic spatial patterns but produces responses approximately 50% stronger than conventional-physics-based models. These amplified responses cause biases in simulating persistent high pressure weather systems by overestimating their frequency while underestimating intensity. As DL models are increasingly used for weather prediction and climate projections, our evaluation approach provides essential benchmarks for assessing their trustworthiness.