Integrating RUSLE, Remote Sensing, and Machine Learning for Precise Soil Erosion Assessment in Semi-Arid Regions

Land degradation is a critical environmental issue that poses significant threats to ecosystem stability, especially in semi-arid regions, which are highly susceptible to erosion. Addressing this challenge necessitates innovative technologies for accurate and efficient prediction. This study leverages the Revised Universal Soil Loss Equation (RUSLE) framework, remote sensing, and soil analyses to identify factors driving soil erosion. By integrating these methodologies with machine learning, the research offers a novel approach for precise, real-time monitoring and detection of soil degradation.

The study characterises each RUSLE factor encompassing soil physicochemical properties, rainfall intensity, soil type, land cover, and topography to estimate soil loss and identify erosion-prone areas. Two complementary data sources were utilised: digital data, comprising time-series satellite imagery (LANDSAT, DEM, CHIRPS) processed through Google Earth Engine, Earth Explorer, and ArcGIS to generate RUSLE factor maps spanning 1987 to 2023; and field data, consisting of soil samples collected from various locations to validate digital results and calculate average soil loss across the study area.

Results indicate that the average annual soil loss during this period is approximately 20.65 tons/ha/year, significantly higher than findings from comparable studies. By combining field and digital datasets using the Random Forest model, a predictive map was developed to highlight erosion-prone areas, providing detailed visualisations of spatial erosion patterns across the region. The analysis further identifies the region's geographic characteristics and irregular, extreme climatic conditions as primary drivers of soil erosion. These findings underscore the critical role of advanced data integration and machine learning techniques in developing effective strategies for soil degradation management.