A combined semantic segmentation and time-series InSAR method for debris flow mobilized material volume evaluation

Accurate assessment of debris flow mobilized material volume (DFMMV) is essential for effective hazard prevention. This study proposes a novel framework that integrates semantic segmentation with time-series InSAR to enable catchment-scale DFMMV estimation using satellite remote sensing data from multiple orbits and time windows. Using a Mixed Vision Transformer U-Net (MVT-UNet), debris flow source material types were classified across 21 debris-flow catchments in the Wenchuan region. Subsequently, a quantitative estimation equation for DFMMV was established by combining multi-track time-series InSAR (MTTS-InSAR) and SAR geometric decomposition techniques. Compared with other models, MVT-UNet outperformed with a 3.2% increase in mIoU and the lowest loss value (0.029), thereby enhancing the source material types classification capabilities. The DFMMV in the study area was quantitatively evaluated using two-dimensional (2D) deformation components in the U-D and normal directions, extracted by the MTTS-InSAR technique. Spatial statistics indicated that 71.09% of the DFMMV were distributed on 20-45° slopes and 61.59% at elevations of 1,200-3,000 m, predominantly in middle to upper catchment reaches and along channel margins. Field observations and historical event data validation demonstrate that the model exhibits strong consistency. The proposed framework substantially reduces the need for extensive field surveys while providing reproducible, quantitative DFMMV maps to support debris-flow forecasting, hazard zoning, and emergency planning.