Characterizing the dynamics of precipitation-induced landslide via multi-scale decomposition of time-series InSAR: a case study of Xinmo landslide, China

Landslide activity in mountainous regions is increasingly driven by cumulative precipitation, especially in slow-moving landslides where precursory signals are subtle and delayed. Therefore, this requires an approach that not only links precipitation to deformation, but also uses multi-scale decomposition to resolve the layered temporal signals driving the evolution of deep-seated landslides. To address this challenge, this paper develops an integrated approach that combines Time Series Interferometric Synthetic Aperture Radar (TS-InSAR) with multi-scale signal decomposition techniques to analyse. Firstly, nearly a decade of Sentinel−1 ascending and descending data is used to derive east-west and vertical deformation. Subsequently, Principal Component Analysis (PCA), Independent component analysis (ICA), K-means clustering, and wavelet analysis are then applied to extract dominant motion features and quantify the precipitation-deformation coupling. Results reveal strong interannual‑scale wavelet coherence between deformation and precipitation, with precipitation typically leading deformation by 50−80 days. Furthermore, after applying wavelet decomposition, precipitation and deformation maintained consistently high Wavelet Transform Coherence (WTC) over an extended period. Vertical deformation shows sustained precipitation-driven behaviour, particularly during the three months preceding the June 2017 failure, while east-west deformation exhibits localized. This modular approach provides transferable insights into the spatial and temporal variability and the delayed hydro-mechanical responses of precipitation-driven landslides.