Surrogate model development for coastal dune erosion under storm conditions

Abstract

Early coastal dune erosion predictions are essential to avoid potential flood consequences but most dune erosion numerical models are computationally expensive, hence their application in Early Warning Systems is limited. Here, based on a combination of optimally sampled synthetic sea storms with a calibrated and validated XBeach model, we develop a surrogate model capable of producing fast and accurate dune erosion predictions under storm conditions when water level and wave forecasts are available. The analysis is performed on Dauphin Island, AL, where we train Multiple Linear Regression Models with oceanographic forcing from the selected sea storms (i.e., XBeach input) and predicted changes in the dune system (i.e., XBeach output). Surrogate model performance is assessed with a rigorous k-fold cross validation. Although changes in the location of dune features are not well predicted, the model attains good performance when predicting changes in dune elevation, barrier-island width and volume.