A time-dependent surrogate model for storm surge prediction based on an artificial neural network using high-fidelity synthetic hurricane modeling

Abstract

Expedient prediction of storm surge is required for emergency managers to make critical decisions for evacuation, structure closure, and other emergency responses. However, time-dependent storm surge models do not exist for fast and accurate prediction in very short periods on the order of seconds to minutes. In this paper, a time-dependent surrogate model of storm surge is developed based on an artificial neural network with synthetic simulations of hurricanes. The neural network between six input hurricane parameters and one target parameter, storm surge, is trained by a feedforward backpropagation algorithm at each of 92 uniform time steps spanning 45.5 h for each storm. The basis data consist of 446 tropical storms developed from a joint probability model that was based on historical tropical storm activity in the Gulf of Mexico. Each of the 446 storms was modeled at high fidelity using a coupled storm surge and nearshore wave model. Storm surge is predicted by the 92 trained networks for approaching hurricane climatological and track parameters in a few seconds. Furthermore, the developed surrogate model is validated with measured data and high-fidelity simulations of two historical hurricanes at four points in southern Louisiana. In general, the neural networks at or near the boundary between land and ocean are well trained and model predictions are of similar accuracy to the basis modeling suites. Networks based on modeling results from complex inland locations are relatively poorly trained.