Machine learning-assisted check dam planning on the Chinese Loess Plateau

Abstract

Check dams, as an effective soil and water conservation measure, have intercepted billions of tons of eroded sediment on the Chinese Loess Plateau (CLP), significantly reducing the Yellow River’s sediment load. However, uncertainty regarding the optimal sites and appropriate number of check dams for future planning limits their potential ecological and economic benefits. Here, we employ a machine learning model trained on hydrological, topographic, and economic factors to identify suitable watersheds for check dam construction across 437,630 watersheds on the CLP. Additionally, we use the check dam system planning method to determine the appropriate number of check dams for future construction. Our analysis indicates that 14,280 watersheds are suitable for check dam construction, primarily located in the High-plain Gully Region and Loess Hilly and Gully Region of the CLP. In these watersheds, constructing 4,551 key dams and 24,816 small and medium-sized check dams is feasible. Validation using the receiver operating characteristic curve shows an area under the curve value of 0.972, demonstrating excellent model accuracy. Additionally, the Mean Decrease Gini index indicates that, among the numerous factors we considered, the soil erosion rate is the most influential factor in determining optimal watersheds. These findings will assist decision-makers in developing plans for the largest soil and water conservation projects on the CLP, and provide methodological insights for dam siting studies in other regions.