Prediction and early-warning of bank erosion in the Middle Yangtze River, China

Bank erosion can cause serious damage to flood control infrastructures in alluvial rivers, and thus threaten the safety of riparian residents and industry in heavily populated river systems, such as the Middle Yangtze River (MYR), China. The current study proposes a novel framework for the prediction and early-warning of bank erosion. The prediction of bank erosion is implemented by coupling a dynamic model with a data-driven random forest model. To determine the early-warning level, four indices are proposed corresponding to bank erosion intensity and dangerous degree, respectively. These indices are combined into a final early-warning level of bank erosion. The proposed framework is applied to the MYR, with its performance being evaluated by the corresponding flow, sediment, and topographic measurements. The results show that: (1) the dynamic model reproduces the flow and sediment transport process well in the MYR, with relative errors being less than 6%, 33%, and 4% for the water discharge, sediment discharge, and river stage, respectively. The model is also able to capture the major bank erosion sites well; (2) The performance of the data-driven model is increased when the input data groups are balanced, with a recall rate of 0.67 and an precision of 0.80 being obtained; (3) the calculated distributions of early-warning sites are generally in accordance with the observations in 2020, and the dangerous areas locate close to the outlet of the Jingjiang Reach of MYR. Besides, a model ensemble is probably a better way to improve the prediction of bank erosion, as compared with solely relying on the refinement of a dynamic model. However, some major gaps are also identified in the current framework.