Abstract: Landslides pose a serious threat to the safety of ultra-high-voltage (UHV)transmission corridors.In mountainous areas with large relief and complex geology,displacement signals are strongly nonlinear,multi-stage,and prone to abrupt mutations; reliable short-term prediction is therefore essential for risk mitigation.We propose a multi-stage landslide displacement forecasting framework that fuses variational mode decomposition (VMD)and a bidirectional long short-term memory (BiLSTM)with time-series InSAR observations.In a high-hazard segment of the Lingzhou-Shaoxing UHV line (Shaanxi,China),SBAS-InSAR is used to retrieve surface deformation; VMD decomposes and denoises the displacement series,and the BiLSTM captures bidirectional temporal dependencies for prediction.The InSAR results reveal up to 110 mm of cumulative subsidence and >50 mm displacement during single accelerated episodes.The proposed model accurately forecasts large-amplitude,multi-stage motions within -120 to 60 mm,achieving a maximum absolute error ≤2.6 mm and a root-mean-square error of 1.0~1.5 mm across typical points,with pronounced advantages during rapid acceleration and impending instability.The method faithfully characterizes multi-phase evolution and extreme deformation of landslides and provides robust technical support for proactive monitoring and intelligent early warning along UHV transmission corridors.

Key words: landslide displacement prediction, InSAR, VMD, BiLSTM, multi-stage evolution, ultra-high voltage transmission lines