Cold Bias in the Central North Atlantic: Characterization and Heat Budget Analysis from CMIP6 and HighResMIP Simulations

Abstract

The Central North Atlantic (CNA) sea surface temperature cold bias is a well-documented feature in contemporary climate models, yet its underlying physical causes remain not fully elucidated. Here we systematically characterize the cold bias using historical simulations from the sixth phase of the Coupled Model Intercomparison Project (CMIP6), complemented by simulations from the High Resolution Model Intercomparison Project (HighResMIP). Our results show that the multi-model mean CNA cold bias across the 35 models employed (-2.63°C) is reduced to -1°C in high-resolution simulations from HighResMIP, which have nominal horizontal resolutions of ≤50km for the atmosphere and ≤25km for the ocean. Spatial correlation and hierarchical clustering analyses show that ocean model resolution, rather than atmospheric model resolution, decisively controls the cold bias, and coarse-resolution models sharing similar ocean model codes produce similar bias patterns. A mixed layer heat budget decomposition shows that the cold bias in coarse-resolution models arises primarily from excessive entrainment cooling and a reversal of ocean lateral heat flux convergence and diffusion from warming to cooling. High-resolution models mitigate these biases through better representation of ocean processes. Surface heat flux analysis reveals that the cold bias cannot be attributed to atmospheric heat flux biases, as models instead exhibit a positive turbulent heat flux bias that is largely a consequence rather than a cause of the cold bias. These findings open the way to targeted improvements of ocean physical schemes in coupled models, helping reduce the CNA cold bias and advance related Northern Hemisphere climate prediction skill.