Gully erosion susceptibility prediction in Mollisols using machine learning models

In recent years, gully erosion has caused soil loss, land degradation, and a large sediment yield in the Mollisols northeastern China, threatening agricultural development national food security. Moreover, prediction of remains great challenge owing to difficulty determining suitable environmental indicators identifying best models for predicting prone areas. Therefore, objective this study was quantify contributions main factors controlling identify model areas susceptible Hailun City, China. Initially, spatial distribution investigated through visual interpretation GaoFen-1 satellite images. The analyzed gullies were evenly distributed region, we selected 70% as training data set remaining 30% validation set. Subsequently, 12 variables, including elevation, slope, aspect, plan curvature, profile topographic wetness index (TWI), type, use, normalized difference vegetation (NDVI), precipitation, distance from rivers, existing gullies, erosion. Then, multicollinearity analysis conducted determine without linearity. Finally, determined using machine learning models, support vector (SVM), multilayer perceptron neural network (MLPNN), random forest (RF), extreme gradient boosting (XGBoost) models. results revealed that there no among indicators, so they all employed susceptibility prediction. XGBoost had highest R2 lowest root mean square error (RMSE) values stage (0.81 0.60, respectively), followed by RF (0.78 0.61, MLPNN (0.65 0.70, SVM (0.62 respectively). largest relative importance score (>35%) erosion, which scores 10% 15%. map central part area more than other regions. These can help managers regions are design conservation practices slow down process.