Predicting tsunami-like solitary wave run-up over fringing reefs using the multi-layer perceptron neural network

Abstract

Modeling of tsunami wave interaction with coral reefs to date focuses mainly on the process-based numerical models. In this study, an alternative machine learning technique based on the multi-layer perceptron neural network (MLP-NN) is introduced to predict the tsunami-like solitary wave run-up over fringing reefs. Two hydrodynamic forcings (incident wave height, reef-flat water level) and four reef morphologic features (reef width, fore-reef slope, beach slope, reef roughness) are selected as the input variables and wave run-up on the back-reef beach is assigned as the output variable. A validated numerical model based on the Boussinesq equations is applied to provide a dataset consisting of 4096 runs for MLP-NN training and testing. Results analyses show that the MLP-NN consisting of one hidden layer with ten hidden neurons provides the best predictions for the wave run-up. Subsequently, model performances in view of individual input variables are accessed via an analysis of the percentage errors of the predictions. Finally, a mean impact value analysis is also conducted to evaluate the relative importance of the input variables to the output variable. In general, the adopted MLP-NN has high predictive capability for wave run-up over the reef-lined coasts, and it is an alternative but more efficient tool for potential use in tsunami early warning system or risk assessment projects.