Mapping of soil erosion susceptibility using advanced machine learning models at Nghe An, Vietnam

Abstract

Soil Erosion Susceptibility Mapping (SESM) is one of the practical approaches for managing and mitigating soil erosion. This study applied four Machine Learning (ML) models, namely the Multilayer Perceptron (MLP) classifier, AdaBoost, Ridge classifier, and Gradient Boosting classifier to perform SESM in a region of Nghe An province, Vietnam. The development of these models incorporated seven factors influencing soil erosion: slope degree, slope aspect, curvature, elevation, Normalized Difference Vegetation Index (NDVI), rainfall, and soil type. These factors were determined based on 685 identified soil erosion locations. According to SHapley Additive exPlanations (SHAP) analysis, soil type emerged as the most significant factor influencing soil erosion. Among all the developed models, the Gradient Boosting classifier demonstrated the highest prediction power, followed by the MLP classifier, Ridge classifier, and AdaBoost, respectively. Therefore, the Gradient Boosting classifier is recommended for accurate SESM in other regions too, taking into account the local geo-environmental factors. HIGHLIGHTS Soil erosion has been modeled and a soil erosion susceptibility map was generated.; Several ML models, including the MLP classifier, Ada Boost, Ridge classifier, and Gradient Boosting classifier were implemented.; Developed models were tuned using the Grid Search CV technique.; The Gradient Boosting classifier performed the best.; About 33% of the study area has a high and very high susceptibility to soil erosion occurrence.;

Keywords

• gradient boosting classifier
• machine learning
• grid search
• soil erosion
• vietnam