Wintertime Atmosphere‐Forced and Ocean‐Driven Sea Surface Temperature Variability in the Kuroshio–Oyashio Extension in <scp>HighResMIP</scp> Simulations

ABSTRACT Using observational and reanalysis data and coupled model simulations with varying resolutions from High‐Resolution Model Intercomparison Project (HighResMIP), sea surface temperature anomalies (SSTAs) in the Kuroshio–Oyashio Extension (KOE) are decomposed into atmosphere‐forced and ocean‐driven components through a simple stochastic climate model and a transition of dominance in wintertime air–sea interaction over the KOE region is revealed. Focusing on the lead–lag relationship relative to January KOE SSTAs, results show that at a 1‐month lead, high‐pressure anomalies over the North Pacific weaken the Aleutian Low and generate easterly anomalies over the KOE, which suppress evaporation and induce downward turbulent heat flux (THF) anomalies. This process, dominated by large‐scale atmospheric forcing, sustains the development of positive SSTAs. At a 1‐month lag, the positive SSTAs in the KOE region tend to force the atmosphere, inducing equivalent barotropic low‐pressure anomalies and upward THF, indicating a transition towards ocean‐driven dominance. Both high‐ and low‐resolution models of ECMWF‐IFS and EC‐Earth3P reproduce the observed reversal of midlatitude atmospheric circulation anomalies associated with the KOE SSTAs. However, clear differences emerge in the simulated lagged, ocean‐driven atmospheric response. The ECMWF‐IFS‐HR models (atmosphere ~25 km; ocean ~25 km) better capture the spatial pattern and relative contribution of ocean‐driven SST variability to atmospheric anomalies, whereas the ECMWF‐IFS‐LR models (atmosphere ~50 km; ocean ~100 km) do not clearly exhibit ocean‐driven dominance in the lagged atmospheric response. Further analyses of EC‐Earth3P (atmosphere ~100 km; ocean ~100 km) and EC‐Earth3P‐HR (atmosphere ~50 km; ocean ~25 km) models indicate notable inter‐model uncertainty in the lagged atmospheric response, suggesting the influence of model dependence beyond horizontal resolution.