Identifying landslide precursors using multiscale InSAR coherence change detection

Early and accurate identification and monitoring of landslide precursors are crucial for disaster prevention and mitigation, playing a significant role in safeguarding human lives and property. While radar coherence maps can effectively detect subtle surface changes, accurately distinguishing the coherence anomalies caused by landslides and identifying landslide precursors of different sizes remains a key challenge. This study proposes a multi-scale spatial coherence change detection method for extracting information on potential landslides. The method utilizes stable temporal coherence to remove inherent decorrelation caused by factors such as soil moisture. It employs a hollow window combined with topographic and vegetation indices to fit the coherence of the central pixel, thereby eliminating transient decorrelation induced by factors like vegetation growth. Multi-scale results are computed based on information entropy-weighted averaging to derive coherence anomaly features. Validation based on the Mud Creek and Heifangtai landslides demonstrates that these features can successfully identify the onset of landslide acceleration and provide continuous warnings. The identification results are consistent with deformation detection, offering new insights and methodologies for the early screening and monitoring of regional landslide hazards.