Evidence-Based Land Degradation Assessment with Earth Observation Data Products

Land degradation (LD) is one of the most pressing environmental problems on a global scale, directly threatening ecosystem resilience, food security, and sustainable land use. Traditional methods used to assess land degradation are often limited by high labor intensity and insufficient integration of heterogeneous geospatial datasets. In this study, we propose an evidence-based approach to LD mapping that integrates multi-source Earth observation (EO) data products with the Dempster–Shafer theory of evidence. A geospatial data cube was constructed based on precipitation, soil moisture, terrain slope, land surface temperature, land cover transitions, vegetation productivity, and soil organic carbon indices. Our classification workflow combined expert knowledge with probabilistic evidence weighting to define LD classes at a regional scale, and our methodology was tested in the Kryvyi Rih Iron Ore Basin (Ukraine), a region under intense anthropogenic and natural pressure. Field-based validation demonstrated the high reliability of the proposed approach, achieving a Kendall rank correlation coefficient of 0.832, which outperforms alternative methods based on Support Vector Machines (SVMs) and Trends.Earth.