Water erosion in semi-arid mountain watersheds reflects both sediment-source potential and structural connection to drainage networks. We developed a screening workflow for the 105.20 km² Deli watershed, southwestern Iran, 2005–2025. Landsat 5/7 indices were transformed to an Operational Land Imager-equivalent scale using near-coincident Landsat 7–8 pairs. Cover-management factors were derived, rainfall erosivity represented with a CHIRPS–modified Fournier proxy, soil erodibility estimated from SoilGrids, and topographic controls derived from a 30 m DEM. RUSLE products were integrated with connectivity, hotspot persistence, and conditional support-practice scenarios. Uncertainty represents the low-to-high structural sensitivity range across alternative R, K, C, and P assumptions, not a probabilistic confidence interval. Results are unitless comparative index values, not soil-loss or sediment-yield rates. The long-term mean potential soil-loss index was 582.11 units; annual means ranged from 138.33 in 2010 to 1040.20 in 2018, with no significant monotonic watershed-scale trend. Response closely followed rainfall erosivity (Spearman ρ = 0.98, p < 0.001), whereas association with mean C was weak and non-significant (ρ = −0.07, p = 0.758). Persistent hotspots occupied 10.42 km², with 37.40% relative structural uncertainty. Moderate and strong conditional P scenarios reduced the index within 4.85 km² of cropland by 36.87% and 68.38%, respectively, but lowered the watershed-wide mean by only 0.34% and 0.63%, because treated areas covered about 4.6% of the watershed. Maps provide first-tier evidence for field verification, monitoring, and sediment pre-treatment. CHIRPS intensity limitations, SoilGrids’ 250 m support, and absent sediment observations preclude calibrated rates or stand-alone siting decisions.