Simulating the Southern Ocean: Cross-generational Improvements and Persistent Model Biases to Target with New Diagnostics.

Given its disproportionate influence on the planetary heat and carbon budget, changes in Southern Ocean (SO) properties and circulation patterns will directly influence the amount of atmospheric warming felt throughout the 21st century and beyond. Additionally, future changes in the mass balance of the Antarctic ice sheet are the largest source of uncertainty in sea level rise projections and are particularly sensitive to changes in the thermal properties of waters in the subpolar SO. The SO is among the least explored of the worlds ocean basins, but recent observations demonstrate that coupled climate models struggle to accurately reproduce both the mean state and historical trends of several key SO properties. Much work has been done to document individual model biases, often focusing on individual metrics with an emphasis on the CMIP ensemble mean performance. However, to improve climate models, a process-level understanding of the contributors to such biases is required, i.e., are the models producing an accurate simulation for the right reasons? Are biases compensating for one another? Is one process improving at the expense of another? This can only be achieved through assessing biases in combination with one another. In such a dynamically complex region as the SO, this requires consideration of many metrics at once, utilization of new observational datasets and tools, and robust documentation of sub-grid scale parameterizations used in the simulations made available to the wider community. Knowledge gained from such large intercomparisons can be used to drive process-focused idealized experiments to understand differing model responses, producing improved simulations, and thus reducing the uncertainty in future projections. A brief overview is provided of areas of improvement in simulating the SO across CMIP generations as well as critical model biases that persist which the community needs to focus on in future model development. Simulations performed using two GFDL coupled models, CM4 and ESM4, are discussed to provide an example of how differing representations of important SO processes can influence a models response to surface buoyancy and momentum perturbations.