A Statistical Approach to Understanding the Long-Term Variability of Storm Surge in the New York Bight

In late October of 2012, a powerful post-tropical cyclone (termed Superstorm Sandy) made landfall near Atlantic City, New Jersey incurring billions of dollars in damage to the NY/NJ Bight region. Numerous tidal stations along the NY Bight coastline broke historical flood stage water levels causing extensive low-lying flooding and infrastructure damage to the region. Given the magnitude of damage and flooding, this cyclone raised scientific and geophysical engineering interest into how often major surge events would occur in the future in this region given various climate change and sea-level rise scenarios. A generalized parametric downscaling technique is presented here to create a point-based storm surge time series for the cool season using global climate model data as part of the latest Coupled Model Intercomparison Project (CMIP5). The statistical model is trained and evaluated using Oct-March ('cool-seasons') 10-meter and mean sea level pressure NARR renanalysis data between 1979-2012. The statistical model is created for three stations along the NY/NJ Bight: Montauk, New York, Battery Park, New York, and Atlantic City, New Jersey. In each station, the multi-linear regression technique features a coefficient of determination (R squared) close to 0.60. The regression technique has near zero bias and a mean absolute error of 0.23 m (Atlantic City), 0.27 m (Battery Park), and 0.19 m (Montauk) in predicting a set of independent 'cool-season' surge data greater than 0.6 m from the years 1979-2012. This statistical model is then applied to predictions from 6 CMIP5 models between 1979-2100. The model data is partitioned into a historical (1979-2005) and a future period (2020-2100). In order to assess the skill of the statistical model in representing historical surge levels, the distributions of historical period surges exceeding particular thresholds are compared with historical surge level observations. Non-parametric hypothesis testing using Kolmogorov-Smirnov two-sample tests are used to compare the exceedance distributions. In all models, the historical period surges greater than 0.6 m at Battery Park feature distributions similar to that of observations at a significance level of 10 %. The presentations will also highlight surge time series using future CMIP5 model data. The CMIP5 GCM models with the greatest historical skill will be used to identify trends in the decadal variability of surge at Battery Park, New York into the future. A Generalized Extreme Value model will be fit to each annual maxima distribution using maximum likelihood estimation . A set of sea-level rise scenarios will be added to the surge level data to quantify how much sea-level rise could impact the return-period of coastal-flooding events.