Evidence of Instability in Previously-Mapped Landslides as Measured Using GPS, Optical, and SAR Data between 2007 and 2017: A Case Study in the Portuguese Bend Landslide Complex, California

Velocity dictates the destructive potential of a landslide. A combination of synthetic aperture radar (SAR), optical, and GPS data were used to maximize spatial and temporal coverage to monitor continuously-moving portions of the Portuguese Bend landslide complex on the Palos Verdes Peninsula in Southern California. Forty SAR images from the COSMO-SkyMed satellite, acquired between 19 July 2012 and 27 September 2014, were processed using Persistent Scatterer Interferometry (PSI). Eight optical images from the WorldView-2 satellite, acquired between 20 February 2011 and 16 February 2016, were processed using the Co-registration of Optically Sensed Images and Correlation (COSI-Corr) technique. Displacement measurements were taken at GPS monuments between September 2007 and May 2017. Incremental and average deformations across the landslide complex were measured using all three techniques. Velocity measured within the landslide complex ranges from slow (> 1.6 m/year) to extremely slow (< 16 mm/year). COSI-Corr and GPS provide detailed coverage of m/year-scale deformation while PSI can measure extremely slow deformation rates (mm/year-scale), which COSI-Corr and GPS cannot do reliably. This case study demonstrates the applicability of SAR, optical, and GPS data synthesis as a complimentary approach to repeat field monitoring and mapping to changes in landslide activity through time.


Introduction
The destructive capabilities of a landslide depend on its velocity and proximity to assets deemed valuable to human livelihood [1]. Unfortunately, many landslides occur in areas that put human life and societal assets (e.g., homes, infrastructure, transportation networks) at risk [2,3]. It is important for communities to identify areas susceptible to landslides and perform necessary preventative measures, which may in some form include spatial identification (e.g., landslide inventory) and temporal monitoring (e.g., displacement measurements), to establish a community landslide mitigation plan [4,5]. Observational landslide identification and monitoring can take many forms [6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23]: repeat mapping expeditions with qualitative descriptions, quantitative field assessments and installation of in situ monitoring equipment (e.g., inclinometers or GPS monuments), and remote sensing surveys (terrestrial, aerial, or satellite-based). The authors utilize similar data-the California landslide inventory [24][25][26][27], annual displacement measurements at GPS monuments [28], and two satellite-based remote sensing techniques: Persistent Scatterer Interferometry (PSI) and Co-registration of Optically Sensed Images and Correlation (COSI-Corr)-to map and update the extent of recent landslide activity Table 1. Variables to consider prior to mapping and monitoring of landslides using Persistent Scatterer Interferometry (PSI), Co-registration of Optically Sensed Images and Correlation (COSI-Corr), and GPS. The authors present this approach with a specific focus to update the California landslide inventory in an area with continuously-moving landslides and high population density. The Portuguese Bend landslide complex is well-known because of its seemingly endless deformation and destruction of local assets, mainly transportation infrastructure and residential buildings. The region's high population density makes mapping the full extent of these landslides quite difficult since natural geologic features are paved over and built upon. Thus, the novel approach presented in this paper assists on two fronts: (1) to quantify maximum landslide velocity with GPS and COSI-Corr in the high hazard regions Remote Sens. 2019, 11, 937 3 of 20 of the landslide complex where deformation is observable, and (2) to quantify and map areas with extremely slow deformation with PSI (mm/year-scale) near the perimeter of the landslide complex where residential neighborhoods have expanded to and may be in danger of long-term deformation or creep.

Study Area
An active landslide complex is located on the south-central coast of the Palos Verdes Peninsula in California. Residential neighborhoods of Rancho Palos Verdes, Rolling Hills, and Rolling Hills Estates delineate the perimeter of previously mapped landslides. At least eight major landslides form this complex (Figure 1), including Ancient Portuguese Bend (AnPB), Active Portuguese Bend (AcPB), Valley View Graben (VVG), Parcel 4 (P4), Abalone Cove (AC), Klondike Canyon (KC), Beach Club (BC), and Flying Triangle (FT), as mapped by the California Geological Survey [24]. Landslide mapping was originally performed through analysis of aerial photographs and then validated through field reconnaissance and topographic map interpretations [25][26][27]. Landslides were classified based on specific characteristics using terminology from Varnes [60], Wieczorek [61], Keaton and DeGraff [62], and Cruden and Varnes [29]. Characteristics include landslide type, thickness, activity (as shown in Figure 1), movement direction, and confidence of interpretation [24]; see Table 2 for characteristics of notable landslides. The landslide complex is sliding south-southwest down the Palos Verdes Hills, a northwestsoutheast trending ridge located north of Rolling Hills and Rolling Hills Estates. All major landslides are classified as rock slides [24] where the moving mass includes bedrock and younger alluvium, the main body generally stays intact, and movement can be described as either translational or rotational, although larger landslides exhibit complex movements (both translational and rotational). The basal surface of rupture on these deep-seated landslides (>15 m in thickness) typically occurs along bedding planes of the tuffaceous unit of the Altamira Shale, the oldest member of the middle to upper Miocene Monterey Formation, parts of which have been altered to bentonite and montmorillonite [63,64]. Relatively impermeable tuff beds rest between clay-altered, highly absorbent bedding planes that act as a conduit for groundwater (the basal surface of rupture) and studies have observed a direct correlation between precipitation and landslide activity [24,[65][66][67]. Landslide activity (Figure 1 and Table 2) was defined by the California Geological Survey using aerial photographs from 1952-1959 and 1994, and field work in the 1990s [24]; it is based on the terminology proposed by Keaton and  DeGraff [62]. Dormant slides are those that have not moved for at least 100 years-old/relict slides have not moved in the last 10,000 years. They show evidence of erosion and are covered with vegetation. Active/Historic slides are those that have occurred recently (since the 1950s) or "within historic time," defined as occurring within the last 100 years.
The landslide complex was dormant prior to reactivation in 1956 and possibly caused by two anthropogenic factors that may have increased groundwater pore pressure, although Kayen et al. [67] acknowledge a lack of hydrologic data to support this hypothesis. First, Rancho Palos Verdes approved the construction of Palos Verdes Drive South, a road running parallel to the coastline, which cuts through the slope toe. Material and fill used for construction were piled nearby, potentially causing rapid loading. Second, irrigation practices from nearby neighborhoods may have contributed to elevated groundwater levels [68]. The 1956 reactivation occurred within portions of AcPB. In February 1974, southern AC also began moving [69]. Heavy rainfall in early 1978 accelerated deformation within the entirety of both AcPB and AC landslides [66]. A 1979 field investigation by Proffer [69] concluded that short-term instability of AC was caused by increased groundwater levels and long-term instability by wave erosion of the toe. Eight dewatering wells were installed within the AC landslide boundary in 1980, significantly mitigating landslide hazard [69]. In 1984, dewatering wells were also installed in the AcPB landslide [67,68]. Other active landslides in the area were moving by the early-to-mid-1980s [68,69] and have been continuously moving since, exhibiting accelerated deformation rates (> 2.6 m/year) during rainy months and decelerated deformation rates (< 1 m/year) during dry months [65,67]. In recent years, coastline roads (including Palos Verdes Drive South) had to be repaired, replaced, or rerouted [70] and mitigation of the landslide complex is a continued topic of debate [71]. and long-term instability by wave erosion of the toe. Eight dewatering wells were installed within the AC landslide boundary in 1980, significantly mitigating landslide hazard [69]. In 1984, dewatering wells were also installed in the AcPB landslide [67,68]. Other active landslides in the area were moving by the early-to-mid-1980s [68,69] and have been continuously moving since, exhibiting accelerated deformation rates (> 2.6 m/year) during rainy months and decelerated deformation rates (< 1 m/year) during dry months [65,67]. In recent years, coastline roads (including Palos Verdes Drive South) had to be repaired, replaced, or rerouted [70] and mitigation of the landslide complex is a continued topic of debate [71].

Data and Methodology
Three deformation monitoring techniques-GPS, COSI-Corr, and PSI-have been used to study the continuously-moving Portuguese Bend Landslide complex. These three techniques were chosen to complement each other, as described in Table 1. Complete descriptions of the geodetic (GPS) and remote sensing (COSI-Corr and PSI) data and the methodology used in this case study are provided in the subsections below. GPS data are used to ground-truth COSI-Corr and PSI displacement and velocity measurements.

GPS
GPS surveys within the Portuguese Bend Landslide complex were conducted by Michael McGee, of McGee Surveying Consulting, on behalf of the City of Rancho Palos Verdes. Sixty-six GPS monuments were placed in a~4 km 2 area, with a focus on the more active AcPB, KC, BC, and AC landslides, and partial coverage of AnPB and FT ( Figure 1). This GPS survey was a continuation of an original survey which began in 1994 and included 149 monuments, but 89 monuments were lost or destroyed, some due to rapid landslide deformation mainly in AcPB [28]. McGee resumed annual monitoring of all GPS monuments (60 found from 1994 survey and six new) in September 2007 and continued through May 2017 (most recent dataset available). A subset of monuments (about 30) were chosen for semiannual monitoring beginning February 2012 and triannual monitoring beginning April 2014. All the information for this project, entitled 'Portuguese Bend Landslide Monitoring Surveys,' including project history, datums and reference system, data collection, equipment and processing, Global Navigation Satellite System (GNSS) network diagram and description, accuracy, and quality control/quality assurance may be found in a series of reports from the City of Rancho Palos Verdes [28]. These GPS measurements are accurate to 1 cm (in relatively stable areas) and 2 cm (in active areas) between annual readings [28].

COSI-Corr
Co-registration of Optically Sensed Images and Correlation (COSI-Corr) is an optical remote sensing technique and ENVI software module [73,74]. It was originally created to measure ground deformation from seismic activity with satellite or aerial optical image pairs-a pre-event image and post-event image [75]. COSI-Corr measures two-dimensional (horizontal) ground deformation between image pairs, with displacement vectors in the north/south and east/west directions. Although designed to measure earthquake-induced deformation, COSI-Corr has been successfully used to measure geomorphological and surficial processes [47,[76][77][78][79].
Eight WorldView-2 high resolution (50 cm) panchromatic optical images were acquired between 20 February 2011 and 16 February 2016. Images were obtained at an incidence angle of 20 • in the 450-800 nm spectral range. Operated by DigitalGlobe, images were processed to the map scale 1:12,000 orthorectified level. Full-size WorldView-2 images were cut to only include the spatial extent covered by the landslides identified in Figure 1 and Table 2 and distributed by ESA through a written proposal (ID 36617). The final output to COSI-Corr processing is a deformation map which includes two-dimensional horizontal displacement measurements (north/south and east/west components) and a signal-to-noise ratio (SNR) at each pixel. SNR values range from 0 (all noise) to 1 (no noise). This procedure was then repeated for all image pairs and yielded seven deformation maps, which were then summed to produce a total deformation map.
Although the accuracy of COSI-Corr displacement measurements depends on many factors, Leprince [80] shares an accuracy estimation based on image pixel size. Leprince [80] states, "The typical uncertainty on the displacement measurement is on the order of 1/10 of the nominal image pixel size." Thus, the accuracy of displacement measurements in a deformation map created with a pair of WorldView-2 images should be approximately 5 cm. The accuracy of the total deformation map, Remote Sens. 2019, 11, 937 6 of 20 the summation of seven deformation maps from seven pairs of WorldView-2 images, should be approximately 35 cm.

PSI
Persistent Scatterer Interferometry (PSI) is a remote sensing technique that measures geophysical and geometric changes of ground targets using synthetic aperture radar (SAR) image stacks [57,58]. Coherent points, known as persistent scatterers (PS), are identified in every image to produce a point cloud, and each PS within the point cloud includes enough data to create a displacement-time series. PSI has been widely used for long-term monitoring of dynamic processes, with many recent studies focusing on landslides [36,52,[81][82][83][84][85][86][87][88][89][90][91].
Forty descending COSMO-SkyMed SAR images (level 1A: single-look complex slant products) were initially acquired between 19 July 2012 and 27 September 2014 by the Italian Space Agency (ASI). Images were acquired at a frequency of 9.6 GHz-corresponding wavelength of 3.1 cm (X-band)-in STRIPMAP HIMAGE mode (~26 • incidence angle) with a spatial resolution of 3 m. Images were provided by the European Space Agency (ESA) through a written proposal (ID 31684).
Level 1A products arrive as focused data in slant range, complex form with no Doppler projection, and include the following pre-processing steps (performed by ASI from Level 0 RAW products): gain receiver compensation, internal calibration, data focusing, statistics estimation of the output data, and data formatting into output [92]. COSMO-SkyMed SAR images were processed with the ENVI + SARscape PSI software package [93]. The result is a PS point cloud. Every PS in the point cloud contains the following information: displacement (mm) at each acquisition; average velocity (mm/year); coherence; location within three-dimensional, geocoded coordinate system (x, y, z); line-of-sight incidence angle and azimuth direction of SAR signal; original location within slant range coordinate system (azimuth, range); precision estimates of height (m) and velocity (mm/year).

GPS and COSI-Corr: Measuring cm-to m-scale Deformation
Maps illustrating the spatial extent of incremental displacement as measured by 66 GPS stations [28] between 24 September 2007 and 3 May 2017 are provided in Figure 2. A maximum displacement > 20 m (average velocity > 2 m/year) was measured at the toe of AcPB. The most active region of the landslide complex is within the AcPB landslide block, which experienced incremental displacements > 1.5 m between annual surveys (orange and red arrows in Figure 2). Other regions of the landslide complex that experienced displacements between 1 m and 1.5 m (light green and yellow arrows in     GPS, COSI-Corr, and precipitation data can be combined to create a unique, descriptive timeline of annual deformation in and around the Portuguese Bend landslide complex (as shown in Table 3).
GPS and COSI-Corr measurements can also be directly compared when analyzing temporal changes at a single location in the landslide complex. A velocity time series from the AcPB toe is provided in Figure 5. AcPB toe maintains a deformation rate between 0.     Temporal results across the landslide complex (when plotted like Figure 5) show deformation patterns identical to those expected from shallow-dipping translational landslides [94]; recall that landslides in this complex slip along bedding planes dipping at 5 • south-southwest toward the ocean. Therefore, most of the deformation will be measured in the horizontal components rather than the vertical component. At GPS station UB-2, GPS and COSI-Corr measurements show a similar average velocity time series trend ( Figure 5). Remember, COSI-Corr measures deformation in the two horizontal planes and GPS measures deformation in three dimensions. Thus, if COSI-Corr measurements equal GPS measurements, as shown in Figure 5, then deformation primarily occurs nearer the two horizontal planes (north-south, east-west) and there is little or no vertical deformation in the region. These measurements match previous observations [24][25][26][27].  A comparison of total displacement measured by GPS and COSI-Corr, as shown in Figure 6, can be used as an accuracy assessment tool. In terms of accuracy, it should be expected that COSI-Corr measurements are less than or equal to GPS measurements at the same location (as discussed in the previous paragraph). As shown in Figure 6, there are five instances (out of 55 total) where COSI-Corr measurements exceed GPS measurements by at least 0.25 m (i.e., the five plot points and respective error bars lie below the red 1:1 line in Figure 6). Thus, COSI-Corr measurements at these five GPS stations may be inaccurate due to additional sources of noise that are not present at the other GPS stations. Table 4. GPS displacement and COSI-Corr displacement comparison between 2012 and 2016 at 55 GPS stations (information from Figure 6).   Table 4 for additional details). Line of 1:1 slope shown as red line (GPS displacement = COSI-Corr displacement). Table 4. GPS displacement and COSI-Corr displacement comparison between 2012 and 2016 at 55 GPS stations (information from Figure 6).  Table 4 for additional details). Line of 1:1 slope shown as red line (GPS displacement = COSI-Corr displacement).

PSI to Measure mm-Scale Deformation
Neither GPS nor COSI-Corr can accurately measure extremely slow deformation of < 16 mm/year [29], so PSI is required for these measurements. Figure 7 shows PSI average velocity measurements between 19 July 2012 and 27 September 2014 across the Portuguese Bend landslide complex. PSI average velocity is measured in the line-of-sight direction (26 • from nadir, N85 • W), with negative values indicating ground movement away from the satellite (which corresponds to downward and/or westward directions). There are five areas of interest that stand out when comparing PSI results with GPS and COSI-Corr: (1) AcPB Body, (2) AC and AcPB Toe, (3) KC, (4) AnPB, and VVG Scarp Reactivation, and (5) P4.
AcPB Body. As mentioned before, this is one of the more active areas within the landslide complex. A lack of PS suggest decorrelation due to rapid deformation (which occurs at a velocity > 2.5 cm/year). PS that border AcPB exhibit velocity around −8 mm/year. PS presence within AcPB appear to act as a boundary around high landslide activity areas, which are approximated well by GPS displacement measurements (yellow, orange, and red arrows in Figure 2). AC and AcPB Toe. This region also shows high landslide activity (Figures 2 and 3). However, presence of PS suggests there is no decorrelation due to rapid deformation and, instead, high landslide activity is localized (e.g., GPS stations are placed in areas with local instability and nearby areas may not be as unstable, hence PS presence). In fact, there are no PS along the coast (Figure 7), which is where most GPS monuments are located (see Figures 1 and 2). PS presence in this region once again act as a boundary to the high landslide activity occurring along the coast.

Final Landslide Deformation Map
A final landslide deformation map is provided in Figure 8. This map, which segments landslide activity based on Cruden and Varnes [29] landslide velocity scale, is divided as follows: slow (velocity > 1.6 m/year), very slow (velocity between 16 mm/year and 1.6 m/year), and extremely slow (velocity < 16 mm/year). Stable areas, those with no sustained deformation throughout the study period, are

Final Landslide Deformation Map
A final landslide deformation map is provided in Figure 8. This map, which segments landslide activity based on Cruden and Varnes [29] landslide velocity scale, is divided as follows: slow (velocity > 1.6 m/year), very slow (velocity between 16 mm/year and 1.6 m/year), and extremely slow (velocity < 16 mm/year). Stable areas, those with no sustained deformation throughout the study period, are not labeled. Additional notes on how PSI, COSI-Corr, and GPS deformation data were converted into the final landslide deformation map are provided in Table 5.

Discussion of Multi-Sensor Approaches for Landslide Monitoring
Previous studies have utilized multi-sensor approaches to monitor landslides. In fact, many of the publications previously cited in this paper use more than one remote sensor for landslide monitoring [30][31][32][33]36,37,42,43,[50][51][52]54,78,81,83,88,89]. Typically, multiple sensors are used to increase spatial coverage or measure deformation in three dimensions (e.g., acquisition of ascending and descending SAR images), to increase temporal coverage (e.g., combining ERS-1/-2 and ENVISAT images for almost 18 years of observation), or to optimize certain sensor parameters (e.g., X-Band SAR for high spatial resolution or L-Band SAR for vegetation penetration).
This study takes a novel approach to demonstrate the capabilities of using a multi-sensor approach (COSI-Corr and SAR along with GPS) to monitor a landslide complex undergoing deformation at three orders of magnitude: mm-, cm-, and m-scale. This type of application allows for in-depth landslide monitoring at a site with widespread displacement rates and land uses (residential, industrial, environmental, etc.) by optimizing the advantages of each technique and minimizing their limitations.
• GPS provides accurate, three-dimensional displacement measurements at a scale ranging from cm to m. In remote sensing studies, GPS data are used as a source of ground-truthing and validation. GPS data are spatially limited as measurements are only available as point sources (e.g., GPS stations). • COSI-Corr processing of optical images allows for two-dimensional displacement measurements at a scale like GPS (cm-to m-scale). COSI-Corr provides a major advantage with respect to spatial coverage, increasing the extent of displacement measurements from point sources (from GPS) to an area equal to the optical image swath. This technique is affected by noise ( Figure 6) and cannot accurately measure displacements less than 1/10-pixel size of the optical images. • PSI processing of SAR images allows for one-dimensional (line-of-sight) displacement measurements at a scale ranging from mm to 1-2 cm. Although PSI cannot measure rapid deformation, it excels at measuring slow-moving deformation (e.g., landslide creep) especially over urban areas, which may be difficult to map in the field.
Monitoring landslides at various orders of magnitude can provide an asset for at-risk communities because almost all types of landslides can be detected. Continuous movement on rotational and translational slides can be annually monitored by GPS surveys or, preferably, by continuous GPS stations. These types of landslides can also be monitored through repeat acquisition of optical images from myriad of sensors on current satellites (e.g., Deimos-2, GeoEye-1, IKONOS, KOMPSAT-2, Landsat, Pleiades, RapidEye, SPOT, and WorldView-1/-2/-3). Slower moving landslides-such as the initial stages of rockfalls or topples (possibly, prior to failure), the perimeter of rotational and translational landslides, or slopes experiencing creep-may be monitored by SAR techniques, such as PSI and others. There are also many current satellites that provide coverage around the world (e.g., ALOS-2, COSMO-SkyMed, RADARSAT-2, SAOCOM-1/-2, Sentinel-1, and TerraSAR-X).

Conclusions
Continuously-moving landslides, such as the Portuguese Bend landslide complex on the Palos Verdes Peninsula in Southern California, are ideal locations for multi-sensor monitoring. The premise is that each technique (PSI, COSI-Corr, and GPS), when analyzed together, provides an advantage where the others might be limited (see Table 1). Forty COSMO-SkyMed SAR images (2012-2014) were processed using PSI to measure average velocity, eight WorldView-2 (2011-2016) optical images were processed using COSI-Corr to measure average horizontal downslope velocity, and 66 GPS monuments (2007-2017) were used to measure incremental displacement. This approach allowed for delineation of active zones within the landslide complex (during the study period between 2007 and 2017). A final landslide deformation map was produced (Figure 8), which divides the landslide complex into three activity categories based on the Cruden and Varnes [29] velocity scale: slow (> 1.6 m/year), very slow (between 16 mm/year and 1.6 m/year), extremely slow (< 16 mm/year), and stable. Average velocity measurements obtained in this study match those of previous studies [65,70] and older observations [24,[66][67][68][69]96]. Landslides are a complex natural hazard and it may require the use of all available resources to fully detect, monitor, understand their geographic and temporal components. Hopefully one day we can use this information to confidently and accurately predict their occurrence as well. Acknowledgments: COSMO-SkyMed images were acquired by the Italian Space Agency and provided by the European Space Agency (ESA, proposal ID 31684). ENVISAT images were acquired and provided by ESA (proposal ID 82169). WorldView-2 images were acquired by DigitalGlobe and provided by ESA (proposal ID 36617). California landslide inventory shapefile was provided by the California Geological Survey and the California Department of Conservation. GPS survey reports were provided by the City of Rancho Palos Verdes, California. Digital elevation models were provided by the Jet Propulsion Laboratory and the United States Geological Survey. Precipitation data for Los Angeles, California were provided by the National Oceanic and Atmospheric Administration.

Conflicts of Interest:
The authors declare no conflict of interest.