<p>Bridge pier scour, the erosion of sediment around bridge foundations caused by hydraulic forces, remains a critical cause of bridge failures. Traditional methods, such as empirical formulas and computational fluid dynamics models, struggle with limited accuracy and high computational costs, especially under complex conditions. In contrast, machine learning (ML) models, including genetic programming, decision trees , artificial neural networks (ANNs), and support vector machines, offer improved predictive capabilities by leveraging large datasets and modeling intricate relationships among various factors. However, the “black-box” nature of ML models often limits their interpretability and practical use in engineering. This study integrates explainable artificial intelligence techniques, such as Shapley additive explanations (SHAP) and partial dependence plots (PDPs), into ML models for predicting bridge scour depth. By incorporating these methods, the study enhances the accuracy and interpretability of the models, providing engineers with transparent and reliable insights. The results indicate that the GEP model outperforms other approaches, achieving superior predictive power with a coefficient of determination (R²) of 0.699 and root mean square error (RMSE) of 0.257 in training, and excelling in testing with an R² of 0.894, RMSE of 0.154, and mean absolute percentage error (MAPE) of 9.705%, demonstrating strong generalization. SHAP values revealed that the most influential features for scour depth predictions were b/D₅₀ (pier width to sediment size ratio), V/V_c (flow velocity to critical sediment velocity ratio), and Fr (Froude number), in this order. The PDPs demonstrate the nonlinear relationships between these features and scour depth, revealing the complex behavior of the system.</p>