Multi-scale sensing and ML fusion reveal the accessibility paradox driving soil degradation in alpine pastures

Alpine soil cover and water resources face compounding pressures from climate extremes and land-use intensification. Yet, monitoring degradation remains challenging as coarse satellite data often mask localized hydrological failures. We address this by integrating high-resolution UAV imagery, historical aerial surveys, and Sentinel-2 time-series with meteorological networks in an alpine silvopastoral system. Using Machine Learning to fuse topographic, climatic, and spectral datasets, we reveal a critical divergence in detection: while satellites suggest landscape greening, sub-meter UAV sensing unmasks degradation hotspots hidden by sub-pixel heterogeneity. Crucially, these hotspots are not driven by steep-slope erosion, but by a "climate-accessibility mismatch" concentrated on gentle, convergent slopes. Here, livestock congregation overwhelms physical soil resilience, particularly during pulse drought events. Coupling this framework with CMIP6 projections further demonstrates that future degradation risk is structurally constrained by topographic accessibility rather than linearly coupled with warming. This implies a "saturation" trajectory defined by landscape morphology. Our findings highlight how fusing high-resolution topography with AI modeling precisely identifies "tipping point" zones, translating complex sensing data into targeted decision support for sustainable alpine management.