Improvement in wintertime mixed layer depth simulation by CMIP6 compared to CMIP5 climate models

Abstract

In this study, we assessed the simulated winter mixed layer depth (MLD) in the CMIP5 and CMIP6 models by comparing to Argo observations. Both the CMIP5 and CMIP6 models have well simulated the pattern of the winter MLD; however, there are large biases in the subtropical and subpolar regions, where the Subtropical Under Waters, Subtropical and Subantarctic Mode Waters, and North Atlantic Deep Water are formed. The CMIP6 models have not eliminated the simulation biases of winter MLD in subtropical and subpolar regions. However, Taylor Diagram (TD) analysis indicates that the CMIP6 models improved winter MLD simulation, showing less scatter and better TD scores than the CMIP5 simulations. Biases in the CMIP6 models were reduced by 11.60–27.55% in the subtropical to middle latitudes, and by 12.64–67.64% in the south/north subpolar Atlantic Ocean. With both the CMIP5 and CMIP6 models, the biases in winter MLD simulation are mainly attributed to inaccurate convection and stratification associated with heat/freshwater fluxes and thermohaline structures. Improving mixing parameterization is one of the most important updates in the CMIP6 models; however, it remains uncertain whether these improvements caused the winter MLD biases to decrease.