Estimating pier scour depth under combined waves and current using boosting machine-learning models

The development of pier scour in coastal environments severely affects the bridge’s stability. Therefore, estimating pier scour around the vertical cylinder is important for the safety of the bridge structure. The estimation of pier scour depth in combined wave-current conditions has become a challenging task for researchers in recent times. The existing empirical formulations that calculate scour in the combined action of current and wave are scarce and may not always provide accurate results. Machine-learning (ML) techniques have become increasingly popular for their prediction capabilities in the fields of hydraulics and coastal engineering in recent years. Therefore, the present study aims to develop Boosting ML techniques (i.e., AdaBoost, XGBoost, CatBoost, and LightGBM) of ML to estimate pier scour in combined wave-current conditions. The non-dimensional parameters, such as Keulegan-Carpenter (KC) number, Relative flow velocity (Ucw), and Absolute Froude number (Fra), are used as input parameters, whereas scour depth (S/D) is the output parameter in Boosting ML models. The sensitivity analysis has been performed to demonstrate the relative importance of the input parameter on S/D. The performance metrics show that the XGBoost model with the input combination of Fra, KC, and Ucw provides the highest accuracy of 92.47% and outperforms SVM, CatBoost, AdaBoost, and LightGBM models. The XGBoost model also outperforms the existing empirical formulations. Therefore, it can be concluded that the XGBoost techniques can be used as a reliable, accurate, and alternative tool to estimate pier scour depth in the combined action of current and wave.