Elsevier

Remote Sensing of Environment

Volume 139, December 2013, Pages 198-204
Remote Sensing of Environment

Time-variable 3D ground displacements from high-resolution synthetic aperture radar (SAR). application to La Valette landslide (South French Alps)

https://doi.org/10.1016/j.rse.2013.08.006Get rights and content

Highlights

  • Image correlation on High Resolution SAR images allows mapping landslide motion.

  • Application to La Valette (French Alps) site based on TerraSAR-X data is carried out.

  • Temporal analysis reveals morphological units with different displacement patterns.

  • Motion pattern identified from SAR data is consistent with local GNSS observations.

Abstract

We apply an image correlation technique to multi-orbit and multi-temporal high-resolution (HR) SAR data. Image correlation technique has the advantage of providing displacement maps in two directions; e.g. the Line of Sight direction (LoS) and the Azimuth direction. This information, derived from the two modes of data acquisition (ascending and descending), can be combined routinely to infer the three dimensional surface displacement field at different epochs. In this study, a methodology is developed to characterize the displacement pattern of the large La Valette landslide (South French Alps) using TerraSAR-X images acquired in 2010. The results allow mapping the dynamics of different units of the La Valette landslide at high spatial resolution. The study demonstrates the potential of this new application of High Resolution SAR image correlation technique for landslide ground surface deformation monitoring.

Introduction

Slope movements such as landslides are one of the most significant geo-hazards in terms of socio-economic costs. Displacement monitoring of unstable slopes is thus crucial for the prevention and the forecast. In areas where large landslides cannot be stabilized and may accelerate suddenly, remote monitoring is often the only solution for surveying and early-warning. The choice of an adequate monitoring system depends on several constraints such as the landslide type, the areal extension, the range of observed velocity, the frequency of data acquisition, the desired accuracy and the cost of data acquisition and processing. Techniques based on High Resolution (HR) Space-borne SAR could provide valuable information in terms of landslide monitoring. In this framework, the availability of HR SAR such as the German Space Agency (DLR) TerraSAR-X (TSX) data with frequent repeat cycle (11 days) represents an opportunity for constructing frequent landslide displacement maps in the perspective of an operational use.

The sub-pixel correlation technique is based on the measurement of sub-pixel offsets between SAR images acquired at different dates. The method is based on a local correlation analysis that can be performed in the Fourier domain (as is presented in the current study) or in the spatial domain (e.g. Delacourt, Allemand, Casson, & Vadon, 2004). Once the SAR data are perfectly co-registered, lines and columns (e.g. azimuth and range directions) offsets between two SAR data are converted in surface displacement estimates (e.g. de Michele et al., 2010, Michel and Avouac, 2002). Typically, the precision of the technique can reach about 1/10 of a pixel or even more (e. g. Leprince, Barbot, Ayoub, & Avouac, 2007) depending both on the characteristics of the data (acquisition geometry, changes that occurred between the acquisitions, instrumental noise) and the amplitude of ground displacement (e.g. notably its spatial wavelength with respect to the size of the correlation window). This technique has been successfully applied to optical and radar data for studying deformation patterns originated from earthquakes (e.g. de Michele et al., 2010, Michel and Avouac, 2002), glacier kinematics (e.g. Scambos et al., 1992, Wangensteen et al., 2006) and landslides (Delacourt et al., 2009, Travelletti et al., 2012). However, in regions characterized by the presence of persistent cloud cover, passive sensor data have important limitations preventing the creation of reliable archives of images for long-term monitoring. Data from active sensors such as HR SAR amplitude can be an alternative to optical imagery. The primary interests of using such data are threefold. First, the SAR amplitude is little or not even affected by the cloud cover or the atmospheric disturbances compared to SAR interferometry (InSAR). Second, the backscatter amplitude is less affected by multi-temporal vegetation changes than the phase of the signal commonly used in InSAR. Third, SAR amplitude is not affected by signal saturation in the presence of high displacement gradient.

Ionospheric disturbances could, in specific cases (e.g. Quegan & Lamont, 1986), produce anomalous signatures on the azimuth offset estimations due to azimuth misregistration errors of the processed images. As the ionosphere is a dispersive medium for microwaves, such an effect mainly concerns the radar data obtained by sensors based on longer wavelength (e.g. L-band sensors). It is thus assumed that results derived from X-band data are little disturbed by the ionosphere except for the geographic locations that are affected by very high total electron content (such as the polar regions; Mattar & Gray, 2002). It has been shown that a posteriori comparison of several azimuth offset maps allows detecting and rejecting offset results affected by ionospheric disturbances (e.g. Raucoules & de Michele, 2010).

A further advantage of using SAR imagery is the high attitude control of the platform, which provides very similar geometrical conditions for image acquisitions (e.g. perpendicular baselines generally shorter than few hundreds of meters). In such conditions, where the base to height ratio (B/H) ~ 10 3, the topographic contribution to the range offsets is rather moderate (for 1 m pixel size, a 100 m height variation would correspond to about 0.1 pixel). By selecting smaller baselines, the topographic component can even be decreased (in the presence of steep relief, for instance) and in several cases, as the presented one, there is no need to correct the topographic component.

With this basis, the ability of TSX (as well as other HR SAR sensors) of providing amplitude images with resolution equivalent to optical remote sensing data (~ 1 m) is of major interest. Given the high spatial resolution of the TSX data, displacement fields in the azimuth and range directions can be measured with an expected precision of about 0.1 m at different acquisition dates.

The objectives of this work are threefold. First, we intend to evaluate the capability of the SAR amplitude offset technique to obtain landslide displacement maps with TSX data. Second, we explore the possibility of combining offset maps into time series using least-square approaches (Casu et al., 2011, Le Mouélic et al., 2005, Usai, 2003). Third, we intend to combine ascending and descending modes to retrieve the multi-temporal 3D surface displacement fields. The study could be of particular interest for detecting changes in surface displacement rates (e.g. acceleration and deceleration) at high spatial resolution. This could help for the forecast of large movements. To reach the aforementioned goals, we have decided to plan the acquisitions of TSX Spotlight data (1 m resolution) at La Valette landslide (South French Alps) for one year, from April 2010 to March 2011. This landslide is characterized by displacement rates of about a dam.yr 1 (e.g. Colas and Locat, 1993, Travelletti et al., 2013). Eight TSX Spotlight data in ascending mode and thirteen in descending mode have been acquired.

The La Valette landslide is one of the largest and more complex slope movement in the South French Alps, and has been triggered in March 1982. The landslide features two styles of behaviors; a translational slide type with the development of a flow tongue in the medium and lower part, and a slump-type with the development of multiple rotational slides in the upper part at the main scarp. The landslide extends over a length of 2 km. It features a variable width ranging from 0.2 km in the lower and medium parts to 0.4 m in the upper part. The maximum depth, estimated by seismic and electrical resistivity tomography and geotechnical boreholes, varies from 25 m in the lower and middle parts (e.g. Hibert et al., 2011, Samyn et al., 2012) to 35 m in the upper part (Le Mignon, 2004, Travelletti et al., 2013). The mean slope gradient is ca. 30° in the scarp area and ca. 20° in the translational slide area. The volume of the landslide is estimated at 3.5 106 m3 (Fig. 1).

The landslide affects a hill slope located uphill of the municipality of St-Pons (Alpes-de-Haute-Provence Department). It represents a significant threat for the 170 community housings located downhill (Le Mignon & Cojean, 2002). The occurrence of rapid mudflows triggered in the scarp area in the 1980s and 1990s has motivated the development of an early-warning system since 1991 composed of benchmark topographical monitoring, optical and infra-red camera monitoring and installation of debris height detection sensors in the torrent, and drainage of the lower part of the landslide.

The landslide exhibits a complex style of activity in space and time. It has developed first as a rotational slide affecting the Autapie thrust sheet in relation to a major fault system (Colas & Locat, 1993). The failed mass has progressively loaded the underlying black marls formation, and the landslide has progressed downhill by a series of rapid mudflows triggered in the marls such as in March 1982, April 1988, March 1989 and March 1992. The most important acceleration occurred in 1988 when a mudflow of 50,000 m3 triggered at the elevation of 1400 m over a distance of ca. 500 m. For the moment, these mudflows did not generate a cascade mobilization of the entire landslide mass.

The ground displacements are monitored permanently with topometric benchmarks since 1991 (Squarzoni et al. 2005), differential dual-frequency GPS (Malet, Déprez, Ulrich, & Masson, submitted for publication) and an extensometer since 2008. At regular periods the monitoring is performed also by digital correlation of terrestrial photos (Travelletti et al., 2012, Travelletti et al., 2013) and satellite radar interferometry (Squarzoni, Delacourt, & Allemand, 2003). Two main aspects can be pointed out from these past monitoring studies and from the observations by the local risk managers. First, they observed a decrease of velocity (from 0.4 m.day 1 to about 0.01 m.day 1) in the middle and lower part of the landslide caused by the local groundwater drawdown since the installation of a drainage system in the 1990s. Second, they pointed out an important activity of the upper part at the Soleil Boeuf crest since year 2000. This activity is characterized by a rapid retrogression of the main scarp towards the North-East and an enlargement of the landslide towards the North-West. In response to this worrying situation, the RTM Service has installed several additional benchmarks along profiles both in the unstable and stable parts of the Soleil-Boeuf crest to monitor the displacements in the crown area. Actually, an accumulation of material and a steepening of the slope are observed in the upper part because of the retrogression of the scarp (Travelletti et al., 2013). Consequently, the possible hazard scenario consists in the untrained loading of underlying black marls formation and the triggering of new rapid and mobile mudflows.

Table 1 lists the TSX Spotlight images used for the analysis. The objective of the satellite programmation was to obtain a sufficient amount of data to estimate the changes in displacement rates during the study period. Thus the image acquisition rate has been deliberately increased for the periods between March and June as changes in the displacement regime (due to possible changes in the sub-surface water circulation in spring) were expected.

The ascending data set is incomplete due to failure in the data acquisition. However, the period between April and November 2010 is globally well covered. We preferred to plan data acquisitions with incidence angles of 41°.2 (ascending mode) and 49°.3 (descending mode) in order to minimize the surface affected by lay-over and shadowing phenomena.

Section snippets

Sub-pixel image correlation

The objective is to estimate local changes in the position of elements at the ground surface by comparing two images acquired at different dates. The observed position change on the image is interpreted as displacement. The estimation of such offsets (both in azimuth and range directions) is obtained by local correlation processing on the image pair. The method applied to SAR amplitude images has been firstly described by Michel et al. (1999) and is today widely used in characterization of

Results: kinematic analysis of the La Valette landslide

Fig. 2 presents the estimated 3D displacement map for the complete investigated period (April 2010–November 2010). The map indicates that the most active part of the landslide is the upper part, and that the displacement rates decrease downslope. The maximum measured horizontal displacement rate is 14 m.yr 1, while the maximum measured vertical velocity is 11 m.yr 1.

For identifying temporal variations in the behavior of the landslide, displacement rate maps were produced for three periods

Conclusions and perspectives

This work demonstrates the interest of sub-pixel image correlation techniques applied to series of HR X-band SAR images for mapping and quantifying landslide displacement patterns. The characteristics of these data in terms of spatial resolution, geometry, repetitiveness and low dependence to the weather conditions are very suitable for landslide monitoring. It appears as a performing alternative to optical HR image correlation whose data can be hampered by atmospheric conditions.

In the case of

Acknowledgments

The TerraSAR-X data used for this study was provided by DLR in the framework of the LAN0666 project. Processing was carried out in the framework of the project SafeLand “Living with landslide risk in Europe: assessment, effects of global change, and risk management strategies” (Grant Agreement No. 226479) funded by the 7th Framework Programme of the European Commission. The GNSS data are provided by the French National Landslide Observatory (OMIV: Observatoire Multidisciplinaire des

References (28)

  • M. de Michele et al.

    Three-dimensional surface displacement of the 2008 May 12 Sichuan earthquake (China) derived from synthetic aperture radar: Evidence for rupture on a blind thrust

    Geophysical Journal International

    (2010)
  • C. Delacourt et al.

    Velocity field of the “La Clapière” landslide measured by the correlation of aerial and QuickBird satellite images

    Geophysical Research Letters

    (2004)
  • C. Delacourt et al.

    Observation of a large scale landslide in La Reunion Island using differential SAR interferometry (JERS and Radarsat) and correlation of optical images (Spot 5)

    Sensors

    (2009)
  • C. Hibert et al.

    Characterizing landslides through geophysical data fusion: Example of the La Valette landslide (France)

    Engineering Geology

    (2011)
  • Cited by (91)

    • Patch-Like Reduction (PLR): A SAR Offset Tracking amplitude filter for deformation monitoring

      2022, International Journal of Applied Earth Observation and Geoinformation
    • Polarimetric SAR pixel offset tracking for large-gradient landslide displacement mapping

      2022, International Journal of Applied Earth Observation and Geoinformation
      Citation Excerpt :

      The interval time of average velocity observed by GPS is from 1 August 2011 to 10 April 2012, and the standard errors of the GPS results are less than 1 cm on horizontal and 1.5 cm on vertical (Coe et al., 2000, 2003). According to the principle of SAR imaging geometry, the GPS monitoring results (east to west, north to south, and vertical displacements) were transformed into the radar coordinate system (range and azimuth displacements) (Raucoules et al., 2013; Hu et al., 2014; Li et al., 2019). The average velocities in the radar coordinate system estimated by PCM-POT and GPS were compared because of the noncontemporaneous observations.

    • Hydrological control shift from river level to rainfall in the reactivated Guobu slope besides the Laxiwa hydropower station in China

      2021, Remote Sensing of Environment
      Citation Excerpt :

      The fastest rates of movement occur in the middle section (~0.4 m/yr in the vertical and ~0.7 m/yr in the horizontal). Similar phenomenon has been observed in other steep slopes, e.g., the Baige landslide moving at similar rates of ~0.1 m/day in both vertical and horizontal directions between August 2017 and February 2018, prior to damming the Jinshajiang River on October 10th 2018 (Li L. et al., 2020, Li Y. et al., 2020); the Shuping landslide moving at 0.6–0.8 m/yr in both vertical and horizontal directions from February 2009 to April 2010 (Shi et al., 2015); the La Valette landslide moving at up to 11 m/yr vertically and 20 m/yr horizontally during April–November 2010 (Raucoules et al., 2013). The active slope surface mainly moves horizontally to the west (Fig. 4A).

    • When image correlation is needed: Unravelling the complex dynamics of a slow-moving landslide in the tropics with dense radar and optical time series

      2021, Remote Sensing of Environment
      Citation Excerpt :

      While optical images are most commonly used – amongst other due to availability of medium resolution data going back to decades, these data are typically limited by the presence of cloud cover that can limit the temporal resolution and prevent the identification of transient deformation signals. In those cases, the use of images from active sensors such as SAR amplitude (i.e., a measure of the backscattered SAR signal strength) can be a good alternative (Raucoules et al., 2013; Singleton et al., 2014). Further advantages of using SAR (or radar) amplitude images include a lower sensitivity to vegetation cover and a sensitivity to the vertical component of displacement.

    View all citing articles on Scopus
    View full text