Evaluations of threshold and curvature mixed layer depths by various mixing schemes in the Mediterranean Sea

Abstract: Predictive ability of five different embedded turbulent mixing models that range from second-order turbulent closure to bulk mixing parameterization is examined in the Mediterranean Sea. Each is embedded in the HYbrid Coordinate Ocean Model (HYCOM). Mixed layer depth (MLD), which is one of the most important upper ocean variables, is used to evaluate the treatment of turbulent processes in each model run. In addition to overall spatial and temporal variability, analyses of MLD are presented using an extensive set (3976) of temperature and salinity profiles from various data sources during 2003–2006. Results obtained from simulations (with no data assimilation and relaxation only to salinity) for the five mixing models are compared with observed MLDs obtained from in situ temperature and salinity profile observations. To ensure the robustness of the validation statistics MLD is computed using both curvature and threshold based methodologies. Results indicate that while all mixing schemes represent the MLD well, the bulk mixing models have substantial accuracy deficiencies relative to the higher order mixing models. The modeled MLDs are slightly deeper than observed MLDs with the mean bias error ∼10m for the higher order mixing models while the bulk mixing model bias error is 15m or more. The RMS error for the higher order mixing models is ∼40m while it is ∼50m for the bulk mixing models. The bulk mixing models had substantially larger errors particularly for the curvature MLD definition. [Copyright &y& Elsevier]