Early Identification of Reservoir-Bank Landslides in Deeply Incised Mountain Canyon Areas with Interferometric Baseline Optimization

Abstract. The complex geological conditions in deeply incised mountainous canyon areas make reservoir-bank landslides a frequent hazard. Accurate interferogram selection and baseline network configuration are crucial for SBAS-InSAR-based landslide monitoring, yet are severely challenged by seasonal vegetation decorrelation. To overcome this limitation, this study proposes a novel Vegetation-Adaptive WCTM that integrates time-series vegetation dynamics into interferometric baseline optimization. This approach establishes a vegetation–coherence coupling model to dynamically adjust coherence thresholds based on quantified vegetation coverage levels and synergizes ERA5 meteorological data with tropospheric delay modeling for atmospheric correction. The results demonstrate significant advancements:（1）The deformation rate standard deviation is reduced by 0.520 and 0.192 compared to traditional short-temporal baseline and average coherence threshold methods, respectively, corresponding to a 29.1 % improvement (1.2668 vs. 1.7865).（2）140,146 additional valid phase-unwrapping points were obtained, indicating substantially improved interferometric processing quality.（3）39 landslides were successfully identified, representing a 22 % increase compared to conventional methods (32 landslides), with 7 new high-risk sites discovered even during low-coherence vegetation seasons. Based on field verification with drone surveys, typical landslides were selected to analyze their spatial distribution and temporal evolution patterns, demonstrating the applicability of the method in deeply incised mountainous canyon areas. These findings provide theoretical and technical support for regional disaster prevention and mitigation efforts.