An assessment of the skill of real-time models of Mid-Atlantic Bight continental shelf circulation

Prescribing open boundary conditions for regional coastal ocean models encounters the challenge of imposing information on sea level, velocity and tracers that characterize the unrepresented far field ocean. Deriving such information from a larger domain model without communicating information from the "nested" model back to the exterior model is " downscaling". We evaluate whether real-time models presently in operation for the Mid-Atlantic Bight (MAB) can deliver useful predictions of subtidal frequency currents and subsurface temperature and salinity for this downscaling purpose. The MAB is a broad continental shelf region where several models run in real time and there is a dense observational data set available for skill assessment. We examine seven real-time models that cover the MAB: three global models, and four regional models. A regional climatology is included as an eighth model. Skill metrics with respect to model bias, centered root mean square error and cross correlation are computed for temperature and salinity profile data from 16 autonomous underwater glider vehicle missions and four hydrographic voyages in 2010-2011. Two years of hourly HF-radar surface current observations that span the shelf are used to evaluate modeled mean surface currents and daily time scale variability in speed and direction. Skill metrics, with uncertainty estimates, are reported for inner and outer shelf subregions, and for stratified and unstratified seasons. A group of models is identified that offers useful skill for the purposes of providing open boundary data to inner shelf and estuary models for real-time applications. Key Points Real-time models of Mid Atlantic Bight assessed with gliders and HF-radar data Bias, RMS error and correlation quantified for inner and outer shelf, and season Models with skill for coastal downscaling applications are identified