Examination of Analysis and Forecast Uncertainty Generated from an Ocean Reanalysis

A series of ocean ensembles has been performed with the goal of estimating the analysis and forecast uncertainty in a high resolution global ocean circulation model. Initial states from different years from a 20-year ocean reanalysis using the 1/12 degree HYCOM-NCODA system (Global Ocean Forecast System 3.0) were used as proxies for perturbed initial conditions. The 20 ensemble members were integrated January-March 2014 assimilating the same observations, than run in forecast mode April-June 2014 using identical surface atmospheric forcing. A second set of ensembles was run using the same method, but for 10 members (2003-2012) and covering the time frame July-September 2014 (assimilating observations) and October-December (forecast mode with identical surface forcing). A goal of these experiments was to characterize and track the evolution of the ensemble spread throughout the analysis and forecast periods, under the assumption that the errors in the initial condition would decrease during the first 3 months due to the assimilation of the observations and subsequently grow during the ensuing 3-month forecast. The ensembles generated from the perturbed initial conditions are compared to the “climatological” ensemble statistics and a corresponding 20-year non-assimilative twin. A variety of diagnostic quantities will be shown to demonstrate and characterize the ensemble system behavior, such as the timescales for the collapse of the ensemble spread with (unperturbed) observation data assimilation as well as the subsequent spread levels maintained during the extended forecast.