Abstract: The upper reaches of the Jinsha River, located on the eastern margin of Tibet Plateau, is characterized by intense tectonic activity and frequent landslide disasters in alpine gorge region. However, Current landslide hazard assessment models demonstrate limited predictive accuracy, failing to meet the precision requirements for practical disaster prevention and mitigation efforts. this study focuses on the Diwu Township section in the upper reaches of Jinsha River. Through the integration of remote sensing interpretation and field geological surveys, and based on the regional characteristics of landslide development, we refined the hydrological analysis method to delineate slope units guided by a watershed-geomorphology-slope structure framework. Thirteen evaluation factors were selected, including topographic and geomorphological indicators (elevation, slope, aspect, and terrain relief), lithology, active faults, rainfall, hydrographic network, anthropogenic engineering activities, vegetation coverage, and SBAS-InSAR surface deformation velocity. A novel Random Forest-Continuous Frequency Ratio (RF-OFR) model was employed to conduct regional landslide hazards assessment. Results demonstrate that slope unit-based evaluation achieves significantly higher accuracy (AUC=0.902) compared to grid unit analysis (AUC=0.858), with delineated high-risk zones showing strong spatial correspondence with documented landslide occurrences. Moreover, within the slope unit framework, the combined ascending and descending SBAS-InSAR deformation results yielded the highest predictive accuracy (AUC=0.902), representing improvements of 6%, 5%, and 0.6% over models using no deformation input, ascending-only, and descending-only data, while significantly improving detection capability for hidden creeping landslides. These findings provide enhanced scientific support for landslide hazard assessment in alpine gorge region and offer valuable references for regional geohazard prevention and risk management strategies.