Refined InSAR landslides deformation monitoring with tropospheric delay correction using multi-temporal moving-window linear model
<p>SAR Interferometry (InSAR) has been proven to be effective for measuring landslides deformation. However, the accuracy of InSAR application of landslide mapping and monitoring is limited by the complex atmospheric distortion in alpine valley areas. The sparse external atmospheric data cannot accurately reflect the complex heterogeneous atmosphere in alpine valley areas. The conventional atmospheric delay corrections based on InSAR phase are weakened by the presence of confounding signals (e.g., tropospheric delays, deformation signals, and topographic errors) and spatial heterogeneity of the troposphere.<br>In this study, we propose the multi-temporal moving-window linear model to estimate the stratified tropospheric delay. The linear relationship between the multi-temporal interferometric phases and the local terrain is estimated using moving windows and then we retrieve the vertical stratified atmospheric phase over the whole scene. Taking into account the deformation information and phase unwrapping errors, the model is solved by an iterative robust estimation algorithm weighted by both deformation rates and residual unwrapping phases.<br>&#160;We first compared our model with four other InSAR atmospheric delay correction methods: traditional empirical linear model, REA5 numerical atmospheric model, GACOS, and temporal/spatial filtering method. The results demonstrated that our proposed model has the best performance on atmospheric delay correction over the reservoir of the Lianghekou hydropower station using Sentinel-1 datasets. Meanwhile, our model was less affected by randomly turbulent phase and phase unwrapping errors, which significantly improves the accuracy of landslide deformation detection and monitoring.<br>Then, we integrated the multi-temporal moving window atmospheric delay correction model into the STAMPS-SBAS program. The high-precision wide-area time series deformation over the reservoir area can be obtained through iterating phase unwrapping and atmospheric delay correction. In particular, the phase unwrapping errors were gradually corrected during the iteration process. The improvement of the proposed model on the landslide investigations was validated through using UAV images and field surveys. Some landslides that have not been identified in the traditional time-series InSAR results can be identified after atmospheric delay correction by the proposed model.</p>