Fixing a Labrador Sea Bias in a fully coupled model

LeAnn Conlon1, Luke Van Roekel2, Andrew Roberts3, Qing Li4, Kristin Hoch4, Kevin L Rosa5, Jon Wolfe4 and Steven R Brus4, (1)United States, (2)Los Alamos National Laboratory, Fluid Dynamics and Solid Mechanics Group, Los Alamos, United States, (3)Naval Postgraduate School, Monterey, United States, (4)Los Alamos National Laboratory, Los Alamos, NM, United States, (5)University of Rhode Island, Narragansett, RI, United States
Abstract:
The Labrador Sea impacts global circulation and climate due to the deep convection and deep water formation that occurs there. However, the Labrador Sea and North Atlantic subpolar gyre can be difficult to model accurately. Model output of the North Atlantic can vary widely, even given similar initial conditions. Biases could originate from a number of sources. In particular, resolution in the North Atlantic has been linked to changes in circulation and model biases, though it's likely that temperature and salinity variability also play a role. To understand the Labrador Sea and determine causes for biases in the region, we examine fully coupled model output using different ocean unstructured mesh resolutions: a low resolution mesh (30 to 60 km), a high resolution eddy resolving mesh (6 to 18 km), and a mesh locally refined in the Labrador sea (10 to 60 km) in the U.S. Department of Energy (DOE) Energy Exascale Earth System Model (E3SM). We examine differences in circulation patterns between meshes and the effect of eddies and their parameterization on model biases. The increased resolution in select areas improved model output without the need to run a more costly, time intensive mesh. Biases appear to be created through a positive feedback mechanism, with a lack of eddies to transport heat into the Labrador Sea at low resolution. However, while increasing mesh resolution prevents biases from occurring, we recognize that this may not always be feasible, and as a result we have also attempted to better understand how correctly choosing a GM parameterization may be able to prevent model biases.