Evaluation and Sensitivity of Climate Model Representation of Upper Arctic Hydrography

The satellite-derived rate of Arctic sea ice extent decline for the past decades is faster than those simulated by the models participating in the Coupled Model Intercomparison Project (CMIP5). In addition, time-varying Arctic sea ice concentration and thickness distribution in those models are often poorly represented, suggesting that predicted sea ice decline might be modeled in the wrong place or time and for the wrong reasons. We hypothesize that these limitations are in part the result of an inadequate representation of critical high-latitude processes controlling the accumulation and distribution of sub-surface oceanic heat content and its interaction with the sea ice cover, especially in the western Arctic. For the purpose of this study, we define the sub-surface ocean as that below the surface mixed layer and above the Atlantic layer. Those limitations are evidenced in the CMIP5 multi-model mean exhibiting a cold temperature bias near the surface and a warm bias at intermediate depths. In particular, CMIP5 models are found to be inadequately representing the key features of the upper ocean hydrography in the Canada Basin, including the near-surface temperature maximum (NSTM) and the secondary temperature maximum associated with Pacific Summer Water (PSW). To identify the sensitivity of upper Arctic Ocean hydrography to physical processes and model configurations, a series of experiments are performed using the Regional Arctic System Model (RASM), a high-resolution, fully-coupled regional climate model. Analysis of RASM output suggests that surface momentum coupling (air-ice, ice-ocean, and air-ocean) and brine-rejection parameterization strongly influence thermohaline structure down to 700 m. The implementation of elastic anisotropic plastic sea ice rheology improves mixed layer properties, which is also sensitive to changes in numerical convective viscosity and diffusivity. Sea ice formation during model spin-up essentially destroys the initial climatological upper ocean stratification suggesting the need for full-column numerical restoration or for realistic sea ice initial conditions during spin-up to prevent this. Finally, we quantify model skill by using a metric that combines both variance and correlation between modeled and observed quantities.