Evaluation of weather forecast systems for storm surge modeling in the Chesapeake Bay

Accurate forecast of sea-level heights in coastal areas depends, among other factors, upon a reliable coupling of a meteorological forecast system to a hydrodynamic and wave system. This study evaluates the predictive skills of the coupled circulation and wind-wave model system (ADCIRC+SWAN) for simulating storm tides in the Chesapeake Bay, forced by six different products: (1) Global Forecast System (GFS), (2) Climate Forecast System (CFS) version 2, (3) North American Mesoscale Forecast System (NAM), (4) Rapid Refresh (RAP), (5) European Center for Medium-Range Weather Forecasts (ECMWF), and (6) the Atlantic hurricane database (HURDAT2). This evaluation is based on the hindcasting of four events: Irene (2011), Sandy (2012), Joaquin (2015), and Jonas (2016). By comparing the simulated water levels to observations at 13 monitoring stations, we have found that the ADCIR+SWAN System forced by the following: (1) the HURDAT2-based system exhibited the weakest statistical skills owing to a noteworthy overprediction of the simulated wind speed; (2) the ECMWF, RAP, and NAM products captured the moment of the peak and moderately its magnitude during all storms, with a correlation coefficient ranging between 0.98 and 0.77; (3) the CFS system exhibited the worst averaged root-mean-square difference (excepting HURDAT2); (4) the GFS system (the lowest horizontal resolution product tested) resulted in a clear underprediction of the maximum water elevation. Overall, the simulations forced by NAM and ECMWF systems induced the most accurate results best accuracy to support water level forecasting in the Chesapeake Bay during both tropical and extra-tropical storms.