Bias Estimates from Ocean Analyses and Forecasts with Different Atmospheric Forcing Products

Ocean analyses and forecasts from operational prediction systems are known to contain errors from a variety of sources. Some of those sources are introduced by errors in the surface boundary conditions, i.e. the wind and heat flux forcing. These errors in turn introduce errors, or biases, in the corresponding ocean forecasts. We have forced a high resolution global ocean model with atmospheric forcing from 4 different sources: the Navy Global Atmospheric Prediction System (NOGAPS) (now decommissioned), NAVy Generalized Environmental Model (NAVGEM), the Climate Forecast Reanalysis System (CFRS), and the Global Forecast System (GFS), the latter two made available by NOAA's National Centers for Environmental Prediction. The ocean model used here is based on the HYbrid Coordinate Ocean Model (HYCOM), run with 1/12 degree horizontal resolution and 41 layers in the vertical. The ocean model is coupled to the Community Ice Model (CICE), and both the ocean and ice ingest observations to perform a 3D variational analysis using the Navy Ocean Coupled Data Assimilation system (NCODA). The overall system configuration is known as the Navy Global Ocean Forecast System v. 3.1.

We will describe how we use the difference between a forecast of sea surface temperature (SST) and a corresponding best analysis (or hindcast) to estimate the total heat flux offset required to minimize those SST differences. The heat flux offsets are calculated in space and time, and once estimated, are used in future forecasts. The NAVGEM forced oceans apply a monthly heat flux offset of -245 W/m^2 per 1°C SST error, which results in a significant reduction in the 5-day SST forecast error. Overall, both GFS and NAVGEM show broadly similar patterns and amplitudes of SST forecast error. For example, during summer of 2014 there were warm biases in the North Pacific, North Atlantic, and Indian oceans, and cold bias in the Eastern and South Pacific Oceans. Results presented include patterns and amplitudes of SST and heat flux bias, and forecast improvement after bias corrections are applied.