By Madeleine Lopeman
The impacts of Hurricane Sandy may have faded to the background in the memories of many New Yorkers, but for decision makers the “superstorm” still looms large. Sandy’s storm surge inundated underground tunnel infrastructure, razed entire coastal communities and caused power outages for millions. In total, the storm caused $65 billion in losses and over 100 fatalities.
Motivated by Sandy and its devastating consequences, I spent the majority of my PhD under the guidance of my adviser George Deodatis exploring new ways to estimate return periods of extreme coastal flood events. The return period, alternatively known as a recurrence interval, represents the inverse of the annual probability of observing an event. For example, there is a 1/10, or 10 percent chance of observing the 10-year event in a given year. Higher return periods correspond to rarer events and lower return periods correspond to more frequent events. It’s important to note that the 10-year event does not occur like clockwork once every 10 years. Rather, the return period is merely the average waiting time between two occurrences, but the waiting time between any two particular instances usually varies substantially.
The question I’ve been trying to answer is not a novel one—indeed, other methods to estimate extreme high water level frequencies exist in the literature. However, shortly after Sandy, the only published estimates of Sandy’s return period ranged from 500 to 3,500 years, and none was based directly on tidal gauge surge data. Since the likelihood of observing the 3,500-year event was so low, my adviser and I had a hunch that careful treatment of the tidal gauge data at the Battery in New York City may suggest a much lower return period than these other estimates, a possible indication that Sandy may have been less unlikely than previously thought.
The peak water level on record in lower Manhattan during Hurricane Sandy of 5.27 meters (just over 17 feet) was at the time—and remains today—the highest water level on record at that location since high-quality digital tidal gauge records began in 1920 (Fig. 1). Furthermore, it’s clear that Sandy’s peak surge coincided with a local high tide. This makes it particularly challenging to estimate its return period, since the estimated annual frequency for the highest value in a time series is measured empirically as the inverse length of the time series, but without more sophisticated methods and/or a longer time series, we have no idea how frequently such an event repeats.
Figure 1. Tidal gauge time history from the Battery in New York City during the days preceding and following Hurricane Sandy’s peak storm surge. Click here to view the National Oceanographic and Atmospheric Administration’s live water level data stream at this location. (Image courtesy of Springer)
Deodatis, Guillermo Franco and I recently published the details of a tidal gauge data-based method for estimation of storm surge frequencies and its application to the question of Hurricane Sandy’s return period in a paper in Natural Hazards, available here. What follows are a few highlights from our findings.
Upon application of the method to the tidal gauge data from the Battery, we found that Hurricane Sandy’s high water level has a return period of 103 years. This is an order of magnitude lower than any previously published estimates for Sandy’s return period (500-3,500 years), a discrepancy that is due in part to the fact that our method accounts for certain features of high water levels that other methods do not, but also to the fact that our method relies on a different underlying probability distribution than some of the other methods do. However, when we look at our results for a range of return periods (Fig. 2), we see a close fit of the estimated return periods to the empirical data, whereas the same cannot be said of methods relying on other distributions (for example here). For this reason, we feel that our return period estimate of 103 years for Hurricane Sandy is a reasonable one.
Figure 2. Water level return period estimates at the Battery, NY. (Image courtesy of Springer)
Here’s another way to think about it. If Sandy were the 3,500-year event, as suggested here, there would only be a 3 percent chance of observing its flood at least once since 1920; whereas if Sandy were the 103-year event, those chances rise to 60 percent over the same time period. This also means that, holding all other climatological factors constant, the chances of observing another Sandy-level flood before the end of the 21st century is anywhere from 2 percent to 56 percent.
Climatological factors, however, are not constant. For instance, we know that sea levels are rising, meaning both of these probabilities are likely lower-bounds. In fact, with 1 meter of sea level rise, our results suggest that Sandy’s peak water level would become the 28-year event, a return period low enough to factor into even low-risk decision-making. In other words, sea level rise is going to make a flood level equivalent to Sandy’s much a much more common occurrence.
A concern shared among many involved in this work was the potential that Sandy’s presence in the tidal gauge data might be forcing our technique to estimate a short return period even if it really were much longer. However, applying our technique to the data excluding Sandy’s surge data point yielded a return period of 141 years for a Sandy-level flood. This means not only that the Sandy data point does not drastically skew the results, but also that, had our method existed prior to Sandy, and had it been used to assess risk in lower Manhattan, we may have had the foresight to prepare for a Sandy-level flood ahead of time.
The results of the study have interesting implications for policy, insurance and engineering decision makers, and I look forward to future collaborations exploring how the findings can shape approaches to storm surge protection and resilience. While our method does not at present incorporate the possibility of a changing climate into its treatment of the tidal gauge data, I hope to expand the method in the future to investigate whether there are any manifestations of climate change in the data that can affect return periods as well as to project forward and assess future risk.
Madeleine Lopeman recently received her PhD in civil engineering and engineering mechanics from Columbia University; her work pertains to quantification of natural hazards risk.