Machine learning-driven approach to quantify the beach susceptibility to storm-induced erosion

Abstract

ABSTRACT This study focuses on quantifying the susceptibility of sandy beaches to storm-induced erosion by analyzing 14 key morphometric indicators. We used a 24-year morphological and metocean dataset, enabling the identification of 347 storms and their beach responses. Four beach profile patterns were identified according to the presence and characteristics of sandbars. Subsequently, four XGBoost models were individually trained for each profile pattern to discern the complex morphodynamics during storm conditions. SHAP analysis was then employed to identify the most influential morphometrics, revealing variations in the morphometrics contributing to beach susceptibility based on the beach profile type. Storm cases were divided into three groups based on storm power to validate the beach erosion susceptibility number (BESN). Evaluation of BESN for each profile pattern and storm group showed that unbarred profiles under average storm conditions exhibited the highest correlation, with a Pearson correlation coefficient (r) of 0.75. There was at least a 0.40 correlation coefficient between observed beach erosion and BESN in eight of the 12 scenarios studied. These findings highlight the effectiveness of BESN in quantifying the susceptibility of different beach profile patterns to erosion. BESN provides valuable insights regarding the vulnerability of sandy beaches by considering key morphometrics.