Ice-jam flood hazard risk assessment under simulated levee breaches using the random forest algorithm

At higher latitudes, lower winter temperatures can cause ice jams to form in rivers, leading to levee breaches and significant economic losses, injuries, and deaths. This study examines the portion of the Inner Mongolia section of the Yellow River where ice-jam floods typically occur. Drawing on hazard system theory and a comprehensive analysis of hazard risk indices, including hazard-inducing factors and hazard-pregnant environments, as well as the vulnerability of the hazard-bearing bodies, we selected eight risk assessment indices to construct an ice-jam flood hazard risk assessment model that uses the random forest (RF) algorithm. The three hazard-inducing factors consisted of max water depth, max overbank flow velocity, and max inundation time, which were derived from the ice-jam flood backwater-levee break-inundation coupling model. The three hazard-vulnerable environment indices were elevation, terrain slope, and the distance to the river channel. The two hazard-bearing body indices were population density and Gross Domestic Product density. The modeling results show that compared with the K-nearest neighbor algorithm, the RF model performed better on both the Precision (P) and the area under the curve in assessing the four study areas. The RF model has significant advantages in classifying multi-dimensional ice-jam flood hazard data. It can provide support for ice-jam flood hazard prevention and mitigation.