Equilibrium scour depth evaluation using well-established regression models in non-cohesive sediment desilting through hydro-suction

Abstract Jaiswal A. 2026. Equilibrium scour depth evaluation using well-established regression models in non-cohesive sediment desilting through hydro-suction. Lake Reserv Manage. 42:73–87. Sediment inflow into reservoirs gradually reduces storage capacity over time. Hydro-suction is an innovative and cost-effective technique for sediment removal that employs siphonic action to desilt reservoirs. Equilibrium scour depth during hydro-suction is directly related to the desilted volume, and its accurate estimation is essential to optimize hydro-suction performance. This study evaluates the predictive capability of 5 machine-learning regression models, artificial neural network (ANN), k-nearest neighbors (KNN), decision tree (DT), random forest (RF), and Gaussian process (GP), for computing equilibrium scour depth during hydro-suction desilting of non-cohesive sediment. A scour depth dataset of 252 laboratory experiments, covering suction pipe diameter (D), suction inlet height (C), sediment median size (d50), and suction velocity (V) as key hydraulic parameters, was used to train and validate the models. Performance was evaluated based on various evaluation parameter metrics, supported by percentage error plots and a Taylor diagram. Results show that GP and RF outperform the other models, with GP providing the most stable predictions, achieving correlation coefficient (CC) values of 0.965 and 0.984 across training and testing datasets (MAE = 0.004 m, NSE = 0.965). Sensitivity analysis identified D as dominant parameter influencing the scour depth prediction, followed by V, C, and d50, exerting secondary but relevant effects. The findings highlight the potential of data-driven approaches to accurately predict scour depth under suction flow, reducing dependence on expensive physical modeling and facilitating efficient hydro-suction design and operation.