The Yellow River source region (SRYR), a climate transition zone highly sensitive to human and environmental changes, faces growing compound soil erosion risks. However, the spatiotemporal distribution of compound soil erosion zones remains unclear. This study integrate machine learning with decision rules to propose a zone framework for compound soil erosion zones, revealing the spatiotemporal variation of the compound erosion zone and synergistic mechanisms. The results showed:(1) The freeze-thaw(FT) erosion was the most widely distributed in the SRYR, water erosion intensity decreased significantly(691.1 t km-2 yr-1) from 1980 to 2020, wind erosion showed decadal fluctuations; (2) The XGBoost-optimized compound soil erosion zone (XGCSEZ) and rule-based compound erosion zone (CSEZ) showed partial consistency (OA = 71.4 %, Kappa = 68.5 %), better identifying in compound soil erosion(i.e.water-wind erosion: 2.0 % increased to 7.3 % in 1980s, water-FT erosion:6.0 % increased to 11.8 % in 2020s);(3) Soil texture plays the most important role (28.1 %) in the compound erosion zone, followed by terrain (27.0 %), climate (25.4 %), landform (15.5 %) and human activities (4.0 %), with clay(SHAP value = 0.54), elevation(0.51), FTCD(0.39), NDVI(0.34), and slope(0.34) as important feature; (4) The XGCSEZ showed vertical distribution of elevation and undergoes significant spatiotemporal changes due to grain for green project(GGP) divided into three stages: binary compound transformation (1980-1995) (I), single-force erosion (water and wind) expansion (1995-2000) (II), and compound soil erosion diversification (2000-2020) (III) due to warm and humid climate and sustainable ecological management. Vegetation is an important stabilizing regulator. This work advances science-based management of compound erosion in high-altitude regions.