Analyzing P-Surge Exceedances to Improve Storm Surge Watches and Warnings

Storm surge accounts for nearly half of all direct tropical cyclone fatalities in the United States.  In recent years, the National Hurricane Center’s Storm Surge Unit has developed products to help communicate the threat of storm surge.  A probabilistic storm surge model (P-Surge) is used by perturbing various characteristics of a tropical cyclone such as intensity, size, forward speed, and direction to create an ensemble.  Previous research was performed in an attempt to identify which exceedance(s) could be used operationally for a storm surge watch and warning.

I expanded this research by using P-Surge model runs that were run retroactively for a handful of Atlantic hurricanes.  By combining Python with Esri’s ArcGIS software, I was able to effectively process surge values for tens of thousands of grid points.  Comparisons between model outputs at various exceedances over time and observed values at landfall were used to compare accuracy.  In particular, I attempted to find if the onset of significant storm surge (> 3 feet inundation) could be accurately forecast in P-Surge.  Initial findings suggest that using a ten percent exceedance 36 to 48 hours before landfall is ideal.  However, as landfall approaches a twenty or even thirty percent exceedance may be the most useful.  More storms will need to be analyzed to yield a statistically significant sample size.