چكيده لاتين
Soil, as a valuable natural resource, plays a fundamental role in sustaining and improving human life, and its proper conservation is essential. Soil erosion, as a geomorphological process, leads to the displacement of soil particles and organic materials to other locations. Among the different types of erosion, water erosion—particularly gully erosion—holds special importance due to its destructive impacts. Therefore, the present study aims to spatially model the gully erosion hazard in the Lamerd watershed, located in Fars Province, using three machine learning algorithms: Random Forest (RF), XGBoost, and TreeNet. In this research, 8,440 gully points were used as the dependent variable, while 30 environmental variables related to erosion were employed as independent variables, including topographic indices (elevation, slope, slope aspect, terrain ruggedness, cross curvature, longitudinal curvature, slope length, and topographic wetness index), environmental factors (soil texture, lithology, land use, distance from roads, distance from rivers, soil erodibility, fault density, and road density), hydrological indices (precipitation, stream power index, drainage density, and flow accumulation), and vegetation indices (NDVI, SAVI, and RVI). For model implementation, 70% of the identified gully data were used for model training, and the remaining 30% were used for model testing. The modeling results indicated that the XGBoost model achieved the highest predictive performance with an AUC value of 96.15. The most influential variables in gully erosion were identified as vegetation indices (NDVI, SAVI, RVI), precipitation, elevation, land use, and drainage density. Finally, the results revealed that the central part of the watershed, extending from west to east, exhibited the highest susceptibility to gully erosion.
