Direct prediction of wave transmission by coastal vegetation using artificial neural networks

Accurate prediction of wave attenuation by coastal vegetation remains a critical challenge in coastal engineering, with no generally accepted empirical formulas currently available. Existing theoretical or numerical models rely on the drag coefficient Cd as an intermediate variable when calculating wave attenuation, but the available empirical formulas for this coefficient lack universality, and have limited prediction accuracy. The present study addresses this problem by applying artificial neural network (ANN) techniques to build models that predict the wave attenuation coefficient directly. Based on the experimental data of rigid and flexible vegetation compiled from the open literature, including both submerged and emergent conditions, the model directly learns the mapping from wave and vegetation parameters to the wave attenuation coefficient Kv without requiring the drag coefficient as an intermediate variable. The highest prediction accuracy is achieved by the model that contains both the Ursell number and Keulegan-Carpenter number in the input features. Sensitivity analysis based on the SHAP method revealed that the length of the vegetation zone x/h and the geometric parameter φ of the vegetation density have much greater impact on the transmission coefficient than hydrodynamic parameters and vegetation flexibility properties.