An evaluation of real-time forecast models of Middle Atlantic Bight continental shelf waters

John L. Wilkin and Elias J. Hunter,
Institute of Marine and Coastal Sciences, Rutgers University.

Setting open boundary conditions for regional coastal ocean models has two related challenges: formulating computational conditions that allow motions generated within the domain to escape, while imposing information on sea level, velocity and tracers that characterize the unrepresented far field ocean. We might expect that comprehensive descriptions of the exterior ocean could be obtained from larger domain models that assimilate observations and are driven by skillful meteorological analyses or forecasts. Providing output from one model as open boundary condition data to a ‘nested’ model, without communicating information back to the exterior model, is essentially the ‘downscaling’ problem. We evaluate whether existing real-time models can deliver useful predictions of sub-tidal frequency currents and subsurface temperature and salinity for this downscaling purpose. We do so by example, focusing on shelf waters of the Middle Atlantic Bight (MAB) – a broad (~100 km) continental shelf region with several models operate in real-time and a dense observational data set acquired by the Mid-Atlantic Regional Association Coastal Ocean Observing System (MARACOOS; maracoos.org) is available for skill assessment. We examine 7 real-time models of the MAB: 3 global models (HyCOM, NCOM, Mercator), and 4 regional models (COAWST, UMassHOPS, ESPreSSO, NYHOPS). A regional climatology (MOCHA) is included as an 8th model. Skill metrics with respect to model bias and centered root mean square error are computed for 16 autonomous glider missions and 4 hydrographic voyages in 2010-2011, and 4 years of CODAR currents. Few of the models regularly outperform a prediction based on climatology T/S. Aggregated skill metrics, with uncertainty estimates, are reported for inner and outer shelf sub-regions, and for stratified and unstratified seasons.