Enhancing predictive modeling of interrill and rill erosion susceptibility in the Eastern Mediterranean using stacking ensemble machine learning algorithms

Soil erosion assessment is essential for effective conservation planning, particularly through the development of accurate susceptibility maps using advanced modeling techniques. Despite this importance, the Eastern Mediterranean remains underexplored in terms of hybrid modeling approaches for predicting interrill and rill erosion in environmentally sensitive areas. This study aims to develop a robust spatial prediction model for soil erosion susceptibility in the Eastern Mediterranean using a stacking ensemble machine learning framework. The performance of Extreme Gradient Boosting (XGBoost), Random Forest (RF), and Multilayer Perceptron (MLP) models was evaluated individually and in stacked combinations based on 751 erosion and non-erosion events and 15 erosion-related conditioning factors. The results identified slope and land use/land cover (LULC) as the most influential drivers of soil erosion. Among the standalone models, XGBoost showed the highest predictive performance, while the hybrid XGB–RF ensemble achieved the best overall accuracy and reliability. These findings demonstrate the effectiveness of hybrid modeling in enhancing soil erosion prediction and provide a reliable decision-support tool for sustainable land management. The proposed approach offers valuable insights for preventive planning and natural resource protection in the Eastern Mediterranean, particularly in Syria, and represents an important step toward improved erosion modeling in complex topographic and climatic environments.