Stacked-based hybrid gradient boosting models for estimating seepage from lined canals

Abstract

Accurate seepage loss estimation from lined canals is crucial for effective water management, especially in water-scarce regions. This study explores seepage loss prediction using advanced gradient boosting models optimized through Bayesian optimization (BO) and introduces a novel Stacked Multiple Linear Regression (SM-MLR) approach. Four hybrid base models were evaluated: Extreme Gradient Boosting (XGB), Light Gradient Boosting Machine (LGB), CatBoost Gradient Boosting (CGB), and Natural Gradient Boosting (NGB), alongside SM-MLR. A dataset of 600 records, using key canal and liner characteristics as inputs, was analyzed. Model performance was assessed using R² and RMSE, with uncertainty evaluations and feature analysis through Partial Dependence Plots (PDPs) and SHapley Additive exPlanations (SHAP). Results showed that the SM-MLR model achieved the highest predictive accuracy, with an R² of 0.998 and RMSE of 0.161 during testing, significantly outperforming all base models and setting a new benchmark for seepage loss prediction in lined canals. Among the base models, BO-CGB demonstrated strong and consistent performance, but it slightly lagged behind the SM-MLR model in terms of overall accuracy. SHAP and PDP analysis identified liner hydraulic conductivity (k_L) as the most influential factor in seepage loss, followed by canal geometry and liner thickness. Further insights were gained from sensitivity analysis, which validated that optimizing k_L is the most effective strategy for minimizing seepage. Thus, the potential of hybrid ensemble modeling was highlighted in significantly improving predictive accuracy, providing a robust and reliable tool for water resource management. The SM-MLR model, by outperforming previous best predictive models, offers actionable insights for real-world canal design, emphasizing the need to consider both hydraulic and geometric parameters to optimize canal performance and support water conservation efforts in water-scarce regions.