Abstract Gully erosion poses a significant threat to land stability in many regions of the world, including the Babai Basin in Nepal. One of the steps in the mitigation planning process is determining areas prone to this type of soil erosion. This study used high-resolution PlanetScope satellite images and data from several Nepali national projects to identify areas susceptible to gully erosion. Fourteen geomorphological, topographical, and hydrological variables were used as predictors in various shallow machine learning algorithms, such as the Random Forest (RF), Extreme Gradient Boost, and Support Vector Machine models. These models were optimized using GridSearchCV. Model validity was assessed using the metrics root mean square error (RMSE), mean square error (MSE), median absolute error (MeDAE), mean absolute error, (MAE), the coefficient of determination (R 2 ), explained variance, the receiver operating characteristics (ROC) curve, and area under the curve (AUC). The RF model performed optimally (RMSE = 0.42, MSE = 0.18, MeDAE = 0.36, MAE = 0.38, R 2 = 0.30, explained variance = 0.30, AUC = 0.82), demonstrating its ability to detect intricate space patterns. The resulting susceptibility map indicates high-risk areas and offers practical assistance for land use planning and erosion control in specific areas. This research demonstrates that blending multi-source geospatial data with optimized machine learning models effectively identifies locations vulnerable to gullies, making it possible to increase the effectiveness of mitigation measures.
Parajuli et al. (Tue,) studied this question.