Incident wave run-up prediction using the response surface methodology and neural networks

Abstract

Submerged breakwaters (SBs) protect coastal areas from intense wave actions, such as inundation and erosion, by controlling the wave run up. The effective regulation of wave run-up heights depends on the accuracy of predictions made by the forecasting model and on understanding the relation between incident wave characteristics, SBs' geometry, and their configuration. This paper proposes models based on the Artificial neural network (ANN) and the Response surface methodology (RSM) to predict the maximum wave run-up heights over a series of rubble mound and caisson-type SBs under varying incident wave conditions. The data for the ANN and RSM models are obtained through physical modeling in the laboratory flume. The objectives of the study are to (1) provide robust tools for the prediction of the maximum wave run up under complex wave-structure interactions; (2) explore the optimum conditions for reducing the maximum run-up height and examine the interdependence of wave-structure characteristics; and (3) investigate the run-up prediction efficacy of the ANN and RSM models. Assessment of the prediction quality of the ANN and RSM models reveals that both techniques establish powerful tools for wave run-up prediction; however, the former offers a slightly better statistical performance. Well-trained ANN model such as Multi-layer perceptron and well-tested statistical methods have considerable potential for application in the development of climate adaptive coastal resilience plans because of their rapid and robust predictive capability.