A Hybrid GEE–Random Forest Framework for Soil-Erosion Mapping in Andalusia: A Two-Reference-Year Assessment of 2018 and 2025 for Sustainable Land Management

Abstract

Soil water erosion is one of the most serious environmental problems worldwide, with major consequences for agricultural output, food security, and terrestrial ecosystems, particularly in the Mediterranean basin. This study compares modelled soil-loss conditions across Andalusia, Spain (87,268 km2), between the 2018 and 2025 reference years using a cloud-based implementation of the Revised Universal Soil Loss Equation (RUSLE) in Google Earth Engine (GEE). The framework couples daily precipitation data from the Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) at its native information scale of approximately 5.5 km, 10 m satellite imagery, and a machine learning-derived, year-specific soil-erodibility update on a common 30 m output grid. The regional mean annual soil loss in 2025 (45.86 t ha−1 yr−1) was 24.98% higher than in 2018 (36.70 t ha−1 yr−1). The 2025 reference year also showed a 12.28% higher R-factor and a 7.83% higher C-factor. The area under Severe erosion (>50 t ha−1 yr−1) increased from 16,330 to 20,636 km2 (+26.37%). Exact signed Shapley attribution on the common erodible support assigned +7.43, +3.52, +0.23, and −2.70 t ha−1 yr−1 to R, C, P, and K, respectively. These results describe a marked contrast between two modelled reference years without establishing a continuous trend or causal change, and demonstrate a transparent framework for regional erosion screening to support sustainable land-use planning and soil-conservation strategies.