Comparison of Methods for Imputing Non-Wetting Storm Surge to Improve Hazard Characterization

Abstract

Joint probability methods for characterizing storm surge hazards involve the use of a collection of hydrodynamic storm simulations to fit a response surface function describing the relationship between storm surge and storm parameters. However, in areas with a sufficiently low probability of flooding, few storms in the simulated storm suite may produce surge, resulting in a paucity of information for training the response surface fit. Previous approaches have replaced surge elevations for non-wetting storms with a constant value or truncated them from the response surface fitting procedure altogether. The former induces bias in predicted estimates of surge from wetting storms, and the latter can cause the model to be non-identifiable. This study compares these approaches and improves upon current methodology by introducing the concept of “pseudo-surge,” with the intent to describe how close a storm comes to producing surge at a given location. Optimal pseudo-surge values are those which produce the greatest improvement to storm surge predictions when they are used to train a response surface. We identify these values for a storm suite used to characterize surge hazard in coastal Louisiana and compare their performance to the two other methods for adjusting training data. Pseudo-surge shows potential for improving hazard characterization, particularly at locations where less than half of training storms produce surge. We also find that the three methods show only small differences in locations where more than half of training storms wet.