Verification of a Multimodel Storm Surge Ensemble around New York City and Long Island for the Cool Season

Abstract

Abstract Three real-time storm surge forecasting systems [the eight-member Stony Brook ensemble (SBSS), the Stevens Institute of Technology’s New York Harbor Observing and Prediction System (SIT-NYHOPS), and the NOAA Extratropical Storm Surge (NOAA-ET) model] are verified for 74 available days during the 2007–08 and 2008–09 cool seasons for five stations around the New York City–Long Island region. For the raw storm surge forecasts, the SIT-NYHOPS model has the lowest root-mean-square errors (RMSEs) on average, while the NOAA-ET has the largest RMSEs after hour 24 as a result of a relatively large negative surge bias. The SIT-NYHOPS and SBSS also have a slight negative surge bias after hour 24. Many of the underpredicted surges in the SBSS ensemble are associated with large waves at an offshore buoy, thus illustrating the potential importance of nearshore wave breaking (radiation stresses) on the surge predictions. A bias correction using the last 5 days of predictions (BC) removes most of the surge bias in the NOAA-ET model, with the NOAA-ET-BC having a similar level of accuracy as the SIT-NYHOPS-BC for positive surges. A multimodel surge ensemble (ENS-3) comprising the SBSS control member, SIT-NYHOPS, and NOAA-ET models has a better degree of deterministic accuracy than any individual member. Probabilistically, the ALL ensemble (eight SBSS members, SIT-NYHOPS, and NOAA-ET) is underdispersed and does not improve after applying a bias correction. The ENS-3 improves the Brier skill score (BSS) relative to the best deterministic member (SIT-NYHOPS), and the ENS-3 has a larger BSS and better reliability than the SBSS and ALL ensembles, thus illustrating the benefits of a multimodel storm surge ensemble.