Baroclinic Coupling Improves Depth‐Integrated Modeling of Coastal Sea Level Variations Around Puerto Rico and the U.S. Virgin Islands

This study applies a baroclinic‐coupled depth‐integrated modeling system to the North Atlantic Ocean, where an unstructured mesh is used to focus resolution down to ∼30 m along the coasts of Puerto Rico and the U.S. Virgin Islands. Ocean baroclinicity is incorporated through one‐way coupling from operational data‐assimilated Global Ocean Forecasting System 3.1 temperature and salinity fields at just 12% additional computational time. The main objectives are to provide a comprehensive analysis of observed and modeled coastal sea levels (spanning from seasonal to supertidal variations) in Puerto Rico and the U.S. Virgin Islands during 2017 and to evaluate the associated model performance with and without baroclinic coupling at 14 National Oceanic and Atmospheric Administration/National Ocean Service tide gauges deployed in the region. It is found that baroclinic coupling increases modeled energy across the entire frequency spectrum, which is more commensurate with observations. In particular, density‐driven effects such as the seasonal cycle and sea level setdown due to trailing cold wakes from passing hurricanes are largely reproduced. Supertidal shelf‐resonant seiching at one to two cycles per hour is observed and modeled at a number of locations, where excitation of these modes is often promoted by the baroclinic coupling. Baroclinicity improves the yearlong model skill at every tide gauge, where the mean total skill is increased from 87% to 93% accuracy (54% to 85% for the nontidal residual). In September 2017 during Hurricanes Irma and Maria, baroclinicity increases model skill at 10 out of 14 tide gauges even when the barotropic mode is adjusted to have no mean offset from the observations. Plain Language Summary Computer simulations (models) with fine scales at the coast are required to accurately forecast coastal water levels potentially causing flooding. Often these computer models only account for the effects of winds and air pressure (creating surge), and gravity from the Moon and the Sun (creating tides), ignoring the effects of ocean density because of the large computational expense required. In this study we present an approach to incorporate ocean density effects into the fine‐scale model by using information from an existing coarse‐scaled ocean model already in operation. This results in a small and manageable 12% increase to computational run‐times. The approach is assessed by comparing model results to measured coastal sea levels during 20