Statistical and Geospatial Analysis of InSAR data for Characterization of Processes Controlling Motion of the Slow-moving Berkeley Landslides

Understanding seasonal processes is paramount in characterising hazard of slow-moving landslides, but is often limited by the temporal and spatial resolution of ground-based and satellite data. Here we use time-series analysis of 2009-2014 TerraSAR-X InSAR data with 11-day repeat-pass, a 3m pixel size, and 463, 372 pixels covering the four slow-moving landslides in the Berkeley Hills region of northern California. These slides cause long-term structural damage to buildings and underground infrastructure, and their urban setting renders them an ideal case study in understanding slide interaction with human development.

First, we combine descending TerraSAR-X observations with ascending RSAT2 observations to isolate the vertical and downslope horizontal components of motion for each slide. We find that in average over the four slides, vertical motion accounts for only 40% of the observed ground motion. Then, we use Principal Component and Independent Component Analyses (PCA and ICA, respectively), to isolate four trends of slide motion from the high temporal resolution TerraSAR-X time series. Three of these trends exhibit seasonal components of motion. We cross-correlate these trends with temperature and precipitation data to uncover the processes controlling slide motion. We find that cumulative precipitation correlates with seasonal slide acceleration at a lag of approximately 22 days. Temperature correlates with an oscillatory seasonal trend of slide motion with zero lag, suggesting thermal expansion.

Finally, we perform a geospatial analysis of the InSAR data with soil, elevation, and optical imagery data. We analyse the independent components against distribution of soil types, slope, and landmark features to isolate the process behind each component. Together with the results of the cross-correlation, our results highlight that although precipitation is an important driver of slide motion, its impacts are not uniform across the four slides. This may be caused by variation in soil type. Our work demonstrates the effectiveness of integrating InSAR time series and geospatial data with statistical analyses to describe and quantify the processes controlling slow-moving landslide motion.