Key points are not available for this paper at this time.
Today, coastal storms represent one of the most formidable environmental challenges, causing significant impacts on coastal communities. This situation underscores both the importance and urgency of studying storms and their characterization. This study proposes an innovative approach combining Principal Component Analysis (PCA) and machine learning (Classification and Regression Trees, CART) to characterize and distinguish damaging storms from non-damaging ones along the coast of Dakar, Senegal. The analysis revealed that among several hydrometeorological variables studied (wave height, period, direction, runup, wave energy, sea level anomaly, tide, etc.), the variables SLA and tide play a central role in the occurrence of damage, although they are weakly correlated with the others. By cross-analyzing these variables, critical thresholds were established, such as Tide > 0.53 m combined with SLA ≥ 0.061 m, Tide > 0.53 m and ECWL ≥ 1.3 m, as well as Runup ≤ 0.64 m associated with a high wave period (Tp), allowing accurate differentiation of potentially damaging storms. The CART method validated these results and identified three key combinations: (1) Tide–SLA, where no damage is observed if Tide 0.53 m and ECWL ≥ 1.3 m; (3) Runup–Tp, where storms are damaging if Runup ≤ 0.64 m or if Runup > 0.82 m with Tp ≥ 16 s. These results constitute the first application of machine learning for storm classification on the Senegalese coast, providing a novel quantitative foundation for better understanding the hydrodynamic conditions associated with damaging storms. The findings of this study could be valuable for risk management and the development of early warning systems
Cheikh Omar Tidjani Cissé (Wed,) studied this question.