Integrating empirical and AI approaches for sediment load prediction: Comparative insights and model development

Abstract This study presents a comprehensive evaluation of sediment load prediction across 64 US rivers using 12 classical empirical equations, 10 widely applied artificial intelligence and machine learning (AI/ML) models, and two newly developed empirical formulations. Among traditional methods, the Brooks (1963) equation showed the highest accuracy for suspended load prediction, while the Kalinske (1942) and Du Boys (1879) equations produced the most reliable bed‐load estimates. The newly proposed suspended‐load equation achieved R 2 = 0.946 and relative root mean square error [RRMSE] = 3.37%, yielding substantial improvements over benchmark empirical models. For bed load, the new equation reduced RRMSE by 33.2% and significantly increased R and R 2 , outperforming widely used methods such as Einstein–Brown and Cheng. Overall, the proposed formulations markedly enhanced key performance indicators—including R , R 2 , relative error, Nash–Sutcliffe efficiency, and mean bias error—demonstrating improved predictive reliability across diverse river systems. Although several AI/ML approaches, particularly the hybrid wavelet– artificial neural network, achieved high predictive accuracy ( R 2 up to 0.95) as benchmark models, the new empirical equations offer clearer interpretability, simpler implementation, and strong engineering applicability. By integrating classical hydraulic theory, established AI/ML techniques, and newly formulated empirical models into a unified comparative framework, this study emphasizes that its primary contribution lies in advancing empirical sediment‐transport prediction rather than introducing new AI/ML algorithms. The results provide practitioners and researchers with robust, interpretable, and practically applicable tools for a broad range of sediment‐transport and river‐management applications.