Estimating the storm surge magnitude and annual exceedance probability is a key element in the siting and design of coastal nuclear power plants in both the U.S. and France. However, differences in storm climatology, specifically the relative importance of tropical cyclones (TCs) versus extratropical storms (XTCs), have driven differences in estimation method development. This work compares purely statistical modeling with combined statistical and numerical simulation modeling approaches for extreme storm surge applied to the U.S. North Atlantic coast which is subject to both tropical and extratropical storms. Two frequency analysis methods are applied to observed water levels and compared to a copula-based joint probability analysis of TCs and automated frequency analysis of XTCs that is enriched with numerically simulated storms. One frequency analysis method is applied using (1) hourly at-site data and (2) hourly at-site data enriched with additional data from a homogeneous region. The other frequency analysis method is applied using (1) hourly at-site data and (2) hourly at-site data enriched with monthly water level maxima. Variables of interest used in the comparison are skew storm surge, maximum instantaneous storm surge, non-tidal residual and maximum seal level. The performance of the methods (mean surge and water level estimates and confidence intervals) depend on the variable of interest and, to some extent, on return period. Inclusion of additional information (e.g., regional water levels, and monthly maxima) in the frequency analysis methods does not have a large impact on estimated mean surge and water levels, but significantly reduces resulting confidence intervals (over 40% reduction in some cases). However, the confidence intervals still grow with increasing return period. Inclusion of simulated storms in the joint probability analysis results in significantly different mean surge and water level estimates (up to 25% higher than the frequency analysis in some cases). The joint probability analysis confidence intervals are wider than those for the frequency analysis methods lower return periods (e.g., 60%–80% wider at 100 years), but they grow much more slowly and are significantly narrower for higher return periods (e.g., 40%–60% narrower at 1 000 years). Although there are appreciable differences between the results documented in this paper, these are reasonable due to differences in the data and methods used in this comparison.
Read full abstract