Estimation of Bedload Yield in Mountain Streams Using Field Observation Data and Machine Learning Techniques

The estimation of the sediment yield in Korea relies primarily on universal soil loss equation-based empirical models. However, validating these models remains difficult, and their application is limited to mountain streams where bedload transport is dominant. This study applied machine-learning techniques to predict bedload yields in mountain streams as an alternative to empirical models. Analysis of global datasets revealed that tree-based models achieved high accuracy (R ≥ 0.94). When the best-performing models were applied to three domestic mountain streams, significant predictive capability was confirmed, particularly in the Gimcheon site (R > 0.87, NSE > 0.73). These findings suggest that machine-learning-based models hold sufficient potential as quantitative decision-making tools for the design and management of check dams.