APPLYING ARTIFICIAL NEURAL NETWORKS FOR PREDICTING BEACH VULNERABILITY TO STORM-INDUCED EROSION

Abstract

 This study investigates the potential of Artificial Neural Networks (ANN) in predicting beach vulnerability to storm-induced erosion. Long-term morphology and hydrodynamic data (24 years) from Hasaki beach in Japan were used to identify storm events and quantify beach erosion. First, to compare the performance of an ANN model with a Multiple Linear Regression (MLR) model in predicting the shoreline change (dSL) during storms, we used initial shoreline position, storm power, maximum surge, and beach slope as input variables. Next, the model predictions of the dSL were utilized to quantify the beach vulnerability on a scale of 1 to 5, resulting in the creation of the Beach Vulnerability Index: BVI_ANN for the ANN model and BVI_MLR for the MLR model. While MLR performed well for short-term beach erosion predictions (8 years) as Thilakarathne et al. (2022) showed, our results indicate that it was less effective when using long-term storm data. In contrast, ANN demonstrated superior performance, resulting in more accurate predictions of beach vulnerability. Specifically, the Mean Absolute Errors for BVI_MLR were 1.33, 0.83, 0.78, 0.90, and 1.07, while for BVI_ANN were 1.00, 0.20, 0.69, 1.05, and 0.57 for indexes 1-5, respectively. The ANN model also achieved higher R² Scores for both training (0.65) and testing (0.62) data in predicting dSL, compared to the MLR model (0.26 on training and 0.35 on testing). The study findings suggest that using ANN or other Machine Learning (ML)-based algorithms for coastal/beach vulnerability studies has significant potential for capturing and representing the dynamic nature of beach morphology changes with increased accuracy. The study's contribution adds to the growing body of research on using ANN and ML algorithms for predicting coastal morphology changes and beach vulnerability, highlighting the potential of these methods for future coastal engineering applications.