2 Data and Methods
2.1 Interferometric Synthetic Aperture Radar Processing and Analysis
Open-access SAR data from the C-band (~5.6 cm radar wavelength) Copernicus Sentinel-1 A/B satellites were automatically processed to standardized geocoded, unwrapped interferograms (GUNW) by the ARIA project (Bekaert et al., 2019). ARIA uses the open-source JPL InSAR Scientific Computing Environment (ISCE) software to process the interferograms (Rosen et al., 2012). These standardized interferograms are corrected for topographic contributions to phase and geocoded to a ~90 m (3 arc second) pixel spacing using the Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM) (Farr et al., 2007). ARIA provides key data needed for deformation analyses and time series inversions including geocoded unwrapped interferograms, coherence, incidence and azimuth angles, and the SRTM DEM and water mask.
We used the ARIA-tools open-source package in Python (Buzzanga et al., 2020) to download and prepare 1689 interferograms covering California (Table S1). We inverted the interferograms to deformation time series using the Miami InSAR Time-series software in PYthon (MintPy) (Yunjun et al., 2019). To remove longer time span interferograms that often have low coherence and are more likely to contain unwrapping errors for persistently moving features such as slow-moving landslides (e.g., Handwerger, Huang, et al., 2019), we used only 2 consecutive interferogram pairs in the time series inversion. We quantified InSAR uncertainty using a bootstrapping technique (Efron & Tibshirani, 1986; Bekaert et al., 2020) with 400 iterations for each time series. More information on the InSAR data processing can be found in the Supporting Information.
2.2 Landslide Reconnaissance and Metrics
We identified active landslides by examining the 5-year time-averaged InSAR velocity maps. Active slow-moving landslides displayed localized deformation zones with relatively high velocity (Figures S2 and S3). We then confirmed that the InSAR signals corresponded to true landslides by overlaying the InSAR velocity maps onto DEMs, Google Earth imagery, and previously published landslide inventories. We began searching for active landslides by examining the InSAR velocity in well-known landslide areas in northern California (Bennett, Miller, et al., 2016; Handwerger, Fielding, et al., 2019; Kelsey, 1978), central California (Booth et al., 2020; Cohen-Waeber et al., 2018; Finnegan et al., 2019; Scheingross et al., 2013; Wills et al., 2001), and southern California (Calabro et al., 2010; Jibson, 2006; Merriam, 1960; Swirad & Young, 2021; Young, 2015). We also examined the InSAR data alongside the California Geologic Survey statewide landslide inventory (Wills et al., 2017). After examining these known landslide areas, we then systematically expanded outward from these regions to identify active landslides in all mountainous regions of California. We quantified landslide metrics such as area, length, width, and slope angle using the SRTM ~30 m (1 arc second) DEM.
To estimate landslide thickness (and volume), which is often thought to be one of the key length scales that controls landslide response to rainfall (e.g., Handwerger et al., 2013), we applied recently developed geometric scaling relations for slow-moving landslides in California (Handwerger et al., 2021). These are particularly useful for estimating thickness of slow-moving landslides, which is difficult to do without numerous ground-based instruments. Landslide scaling relations take the form of a power function as
\(h\ =\ c_{h}A^{\zeta}\) and \(V\ =\ c_{V}A^{\gamma}\) , (1)
where A is the landslide area, h is the estimated mean thickness, V is the estimated volume, γ and ζ are the scaling exponents and \(c_{V}\) and \(c_{h}\) are fit intercepts (see parameters in Table S2). Landslide geometric scaling relations also vary by landslide type and material (Bunn et al., 2020; Larsen et al., 2010; Handwerger et al., 2021). To apply the most appropriate scaling relations (Handwerger et al., 2021), we classified slow-moving landslides as slumps (one primary kinematic zone and low length/width aspect ratios), earthflows (one primary kinematic zone and medium aspect ratios), and landslide complexes (amalgamations of landslides with multiple kinematic zones and high aspect ratios).
To further assess the kinematic behavior of the slow-moving landslides in wet and dry environments during wet and dry years, we selected a subset of landslides to perform detailed time series investigation (Figure 1d). These landslides were selected based on their relatively high velocity signal (i.e., strong InSAR signal) and their location within California’s different hydroclimatic regimes. We characterized the landslide motion by calculating the spatial mean of the fastest moving kinematic zone and used a moving median temporal smoothing filter to further reduce noise and highlight the seasonal and annual deformation signals (Figures S2 and S3). We explored environmental controls on landslides by examining the rock type and precipitation data in active landslide areas. Rock type data are provided by the California Geologic Survey (Jennings et al., 2010) and precipitation data are provided by the Parameter-elevation Regressions on Independent Slopes Model (PRISM) (see Open Research). We then quantified landslide sensitivity to rainfall by exploring relative changes in precipitation and landslide velocity. To explore relative changes in precipitation and velocity, we defined the Precipitation Ratio as the total water year precipitation divided by the 30-year mean water year precipitation (calculated from WY1990-WY2019) at each landslide (Figure 1d-i), and the Velocity Ratio as the water year velocity divided by the average velocity from WY2016-WY2019.