eva‎luation of Drag Coefficient of Mangrove Coastal Protection Systems Against Tsunami Waves Using Machine Learning

12/22/2025 12:00:00 AM

Coastal vegetation, particularly mangrove forests, is recognized as an effective natural solution for mitigating the energy of destructive waves such as tsunamis. However, quantifying the hydrodynamic resistance of these systems, represented by the drag coefficient, remains a scientific challenge due to the complexity of interactions between flow an‎d vegetation structures. Previous studies have often limited knowledge extraction to individual experimental datasets. This research aims to address potential biases by integrating data from two independent sources, encompassing diverse protection systems, including solitary trees an‎d hybrid systems with reef balls an‎d perforated pipes, to develop an accurate an‎d interpretable engineering model for predicting the drag coefficient using advanced machine learning techniques. A comprehensive dataset comprising 126 laboratory samples from two independent studies was compiled. Based on fluid mechanics principles an‎d similarity theory, eleven dimensionless physical parameters were defined to describe various aspects of incoming flow, vegetation structural characteristics, an‎d their interactions. To identify the most effective an‎d least redundant variables, a four-stage feature selec‎tion framework was implemented, including low-variance feature elimination, correlation analysis, mutual information analysis, an‎d recursive feature elimination, resulting in the selec‎tion of an optimal subset of five key parameters. Subsequently, the performance of six machine learning models from different families—Linear Regression, Support Vector Machine, Neural Network, Ran‎dom Forest, XGBoost, an‎d CatBoost—was eva‎luated on the combined dataset. Test set results demonstrated that the XGBoost model outperformed others, achieving a coefficient of determination (R²) of 0.9526, a mean absolute error of 0.3667, an‎d a root mean square error of 0.2903. The robustness an‎d reliability of the results were assessed using cross-validation an‎d bootstrap uncertainty analysis. Finally, using SHapley Additive exPlanations (SHAP) analysis, the XGBoost model was examined as a "glass-box" model, revealing nonlinear relationships between key physical parameters an‎d the drag coefficient.

ضريب پسا , مانگرو , سونامي , يادگيري ماشين , راهكارهاي مبتني بر طبعيت

Drag Coefficient , Mangrove , Tsunami , Machine Learning , Nature Based Solutions

Amirreza Shahmiri

Dr. Abbas Yeganeh Bakhtiary - Dr. Naser Shabakhty