Machine Learning Techniques for Gully Erosion Susceptibility Mapping (Case Study: Mukhtaran Watershed, South Khorasan Province, Iran)

Gully erosion, a significant environmental issue, can lead to severe consequences like soil loss, habitat destruction, and water pollution. To mitigate its impact, accurate mapping of land sensitivity to gully erosion is crucial. Machine learning models offer a powerful approach to predict and map gully erosion susceptibility. This study focuses on the Mukhtaran basin in South Khorasan province, Iran. By employing various machine learning techniques, including GLM, GBM, CTA, ANN, SRE, FDA, MARS, RF, and MaxEnt, the researchers aimed to identify the most suitable model for predicting gully erosion. Twenty-two environmental factors were selected and analyzed, with a focus on physiographic, climatic, hydrological, soil, land surface/cover, and geological variables. The results showed that the random forest (RF) and ensemble (ESMs) models demonstrated the highest accuracy in predicting gully erosion susceptibility, with a TSS index of 0.97. Sensitivity analysis revealed that the digital elevation model, soil electrical conductivity, bare soil percentage, land unit components, geology, runoff coefficient, and maximum storage capacity were the most influential factors. The study emphasizes the potential of machine learning models in generating accurate gully erosion susceptibility maps. However, further research is needed to explore additional factors and improve data quality. By combining topographic/hydrologic indices with machine learning models, more precise estimates of gully paths can be obtained for use in process-based models.

Machine Learning Techniques for Gully Erosion Susceptibility Mapping (Case Study: Mukhtaran Watershed, South Khorasan Province, Iran) Keywords:

Arid watershed , ML performance , Random forest , Spatial prediction , Vulnerability assessment