Document Type : Original Research Article
Authors
1
Department of Natural Resources Engineering, Faculty of Agriculture and Natural Resources, University of Hormozgan, Bandar Abbas, Iran
2
Department of Watershed Management, Faculty of Natural Resources and Environment, University of Birjand, Birjand, Iran
Abstract
Soil erosion poses a significant threat to environmental stability, land fertility, and food security, necessitating precise assessment and mapping. In this study, three widely used machine learning algorithms, including Random Forest (RF), Generalized Linear Model (GLM), and Multivariate Adaptive Regression Splines (MARS), to predict the sensitivity of various water erosion forms in Gonabad County, an arid and semi-arid region in Iran. Thirty-one environmental variables (physiographic, climatic, hydrological, pedological, and land cover) were initially prepared using remote sensing and GIS data. Following collinearity analysis via the Variance Inflation Factor (VIF), 21 variables were selected as model inputs. Erosion types, including rill, channel, streambank, and gully erosion, were recorded using the BLM method and extensive field surveys.The results showed that the RF model performed better in predicting different forms of water erosion than the GLM and MARS models. Specifically, the RF model achieved superior performance with Overall Accuracy (OA) ranging from 0.866 to 0.994, and Kappa coefficients between 0.723 and 0.988 across various erosion facies. The ROC and TSS indices also confirmed the higher predictive power of the RF algorithm compared to its counterparts. Sensitivity analysis identified geology, geomorphology, land use, and soil properties (litter, gravel, gypsum, clay, acidity, hydrological groups, evaporation) as the primary drivers of erosion. Furthermore, channel erosion was identified as the dominant type, covering 24.85% of the study area. This research underscores that integrating machine learning with GIS and remote sensing provides a robust framework for soil erosion assessment and sustainable natural resource management.
Keywords
Main Subjects
)