Prediction of Total Sediment Load Using a Novel Data-Driven Fusion of Semi-Empirical and Physically-Based Models

Accurate prediction of total sediment load can play a crucial role in water resources engineering for managing river systems and mitigating sediment-related risks. In this study, we introduce a new and efficient data-driven fusion framework that employs the outputs from the semi-empirical and physically-based Yang (۱۹۷۳, ۱۹۷۹) sediment transport equations as inputs of two ensemble machine learning (ML) techniques, Random Forest (RF) and extreme gradient boosting (XGBoost), to predict total sediment concentrations in the Nestos River, Greece, using ۱۱۱ field measurements. ML models fuse the results of Yang models to achieve more accurate estimations. Both models performed well. During testing, both models demonstrated strong predictive power and generalization capability, significantly outperforming Yang's models. In the testing phase, XGBoost recorded an R² of ۰.۷۶۶۹ and an RMSE of ۰.۲۷۴۲. Random Forest also yielded an R² of ۰.۷۵۱۵ and an RMSE of ۰.۲۸۳۱. Our results indicate that the proposed cutting-edge approach significantly improved the predictive accuracy of traditional Yang equations used in previous studies. This hybrid approach provided higher accuracy despite its simplicity and low computational costs, which makes it beneficial for engineering and management purposes, especially in data-scarce rivers.

کلیدواژه ها:

Total sediment load ، Random Forest ، XGBoost ، Yang sediment transport models ، Nestos River