Accurate identification of erosion-prone areas in steep and geologically fragile mountainous watersheds is essential for effective management. This study applies the Erosion Potential Method (EPM, Gavrilović model) to assess and map erosion severity in the Portaikos mountainous watershed (Thessaly, Central Greece) using multi-source Earth Observation data. The φ coefficient, which in the EPM represents the type and extent of erosion processes and is traditionally derived from field observations, was herein estimated in an innovative way from the Bare Soil Index (BSI) calculated from Sentinel-2 imagery. This step was implemented entirely in the Google Earth Engine (GEE) cloud environment, while the remaining model inputs—CORINE Land Cover (CLC), a high-resolution bare-earth DEM (FABDEM), and the national Soil Map of Greece (MEEN)—were processed in a GIS framework for the final computation of the erosion coefficient (Z). The resulting Z values were classified into five severity categories (from excessive to very slight) to depict the spatial variability of erosion risk within the watershed. Excessive and severe severity classes together occupy approximately 7% of the watershed area, mainly on steep slopes with sparse vegetation cover. Moderate-severity zones cover about 40%, while slight and very slight classes account for the remaining 53%. Areas mapped as excessive or severe severity closely match locations of documented debris flows and landslides from historical records, supporting the model’s reliability. These results confirm that integrating open-access EO datasets with targeted cloud-based processing can enhance erosion risk assessment in mountainous watersheds, enabling more frequent updates and supporting targeted conservation and hazard mitigation measures.