Coastal Hurricane Inundation Prediction for Emergency Response Using Artificial Neural Networks

Abstract

Emergency managers require both fast and accurate estimates of hurricane inundation to make critical decisions about evacuations, structure closures, and other emergency response activities before, during, and after events. Probability analyses require multiple simulations which, generally, cannot be performed with the physics-based models under the time constraints during emergency conditions. To obtain highly accurate results with a fast turnaround computation time a “surrogate” modeling approach is employed. This surrogate modeling approach uses an extensive database of storms and storm responses and applies “smart” pattern recognition tools such as Artificial Neural Networks (ANN) as well as interpolation techniques. The goal is to provide forecasts of hurricane inundation and waves with the accuracy of high-resolution, high-fidelity models but with very short execution time (minutes). The city of New Orleans as well as surrounding municipalities along the Gulf of Mexico coastal area encompasses the region used to demonstrate this approach. The results indicate that the developed surge prediction tool could be used to forecast both magnitude and duration to peak surge for multiple selected points in a few minutes of computational time once the storm parameters are provided. In this paper, only results of surge magnitude are presented.