Prediction of wave runup on beaches using interpretable machine learning

Abstract

Wave runup estimation is of major interest to coastal engineers for identifying vulnerable and safe areas in coastal regions. Recently, prediction using machine learning (ML) techniques for finding statistical structures from nonlinear analyses and input/output data has attracted significant research attention. However, ML is a black-box model wherein model interpretation becomes difficult with an increase in complexity. Therefore, for selecting a suitable runup database model, we consider ML models such as support vector machine, random forest, and XGBoost (XGB) in this study. In addition, we perform model analysis including variable importance and correlation related to runup prediction through interpretable ML. In comparison to the existing empirical formula, the prediction accuracy of the XGB model is superior, with significantly reduced errors. However, the model exhibits increased error for some unseen data outside the data training range. Moreover, model analysis shows the dominant effect of the beach slope and the wave steepness factor on wave runup prediction. © 2024 Elsevier Ltd