Analyse de la cinématique de la zone d’accumulation : pied de la coulée

Multi-date correlation of Terrestrial Laser Scanning data for the characterization of landslide kinematics

Julien Travelletti

, Jean-Philippe Malet

, Christophe Delacourt

a Institut de Physique du Globe de Strasbourg, CNRS UMR 7516, Université de Strasbourg / EOST, 5 rue René Descartes, 67084 Strasbourg Cedex, France

b GEOPHEN-LETG, CNRS UMR 6554, Université de Caen Basse-Normandie, 14032 Caen Cedex, France

c Université Européenne de Bretagne, Institut Universitaire de la Mer, CNRS UMR 6538, Université de Brest, Brest, France

A b s t r a c t

A detailed characterization of landslide displacements is an important prerequisite to understand the failure mechanism and quantify the hazard. In the last decade, the potential of Terrestrial Laser Scanning (TLS) to monitor slow-moving landslides has been largely demonstrated but accurate processing methods are still needed to extract useful information available in multiple TLS point clouds.

This work presents a new and simple method to obtain 3D deformation and displacement maps from repeated TLS acquisitions by taking full advantage of the geometric information available in consecutive point clouds. The performance of the method is tested on datasets acquired at the toe of the Super-Sauze landslide (South French Alps) over a period of 3 years.

The method is based on the simplification of a 3D matching problem in a 2D matching problem by using a 2D statistical normalized cross correlation function. First, the point clouds are filtered from vegetation and co-registered in a common local coordinate system by aligning the TLS acquisitions on stable parts in the surroundings of the landslide. Second, a perspective projection is applied to project the 3D point clouds on a 2D regular grid perpendicular to the viewing direction. In order to emphasize the relief morphology projected in the 2D grid, the 2D gradient of the distance separating the point clouds from the TLS location is computed and then correlated. Third, a re-projection of the correlated displacements in the 3D local coordinate system is used to compute the 3D displacement field and to evaluate the strain field.

Comparisons with the 3D amplitudes of displacement computed (1) with the Iterative Closest Point algorithm and (2) with dGPS observations of benchmarks indicate an average accuracy of the method of 0.008 m and a standard deviation of 0.04 m.

The observed landslide displacements are heterogeneous, and range from 0.04 m to 10.76 m between consecutive TLS acquisitions; the maximal cumulated displacement observed over the period October 2007 – May 2010 reaches 21.80 m. Within the landslide, sub-areas presenting different kinematic and deformation patterns (extension, compression) are detected by a strain analysis. It is demonstrated that pore water pressure changes within the landslide is the main controlling factor of the kinematics.

Key words:Terrestrial Laser Scanning, Point clouds, Image correlation, Landslide, Kinematics, Strain analysis

1. Introduction

Techniques of slope monitoring have made a lot of progress in the last decade, especially in the field of ground-based remote sensing platforms (e.g. Ground- Based Synthetic Aperture Radar Interferometry, Terrestrial Laser Scanning, Terrestrial Optical Photogrammetry). These techniques allow to discriminate stable and unstable areas from safe and remote places and to map sectors with different kinematics within a landslide (Corsini et al., 2006; Delacourt et al., 2007). The different instruments provide the necessary information to carry out a quantitative analysis of the deformation field and a geomechanical understanding of the failure mechanism (Casson et al., 2005; Teza et al., 2008;

Oppikofer et al., 2008).

The acronym LIDAR stands for LIght Detection and Ranging. When operated from a ground-based platform, this instrument is also known as a Terrestrial Laser

Scanning (TLS). This type of instrument is currently used in a large variety of applications in earth and environmental sciences, and among them for landslide analysis as underlined by the considerable increase in the number of publications in the last years (Slob and Hack, 2004; Sturzenegger and Stead, 2009; Jaboyedoff et al., 2010). TLS instruments allow a fast (typically thousands of points per seconds), distributed, high resolution (millimetric to centimetric) and dense (several millions) acquisition of 3D information of the terrain (triplets of XYZ points). The instruments typically use ‘time-of-flight’ (also known as ‘pulse based’), ‘phase based’ or ‘waveform processing’

technology to determine distance and collect a massive amount of raw data called a ‘point cloud’. There are significant differences in laser light wavelengths, amount and velocity of point data collection, field acquisition procedures, data processing and possible error sources which are detailed in Hiremagalur et al.

(2007) and in Vosselman and Maas (2010). Time-of-flight scanners are the most common type of laser scanner used in geological research because of their longer effective maximum range (typically 100-800m) and data collection rates. They combine a pulsed laser emitting the beam, a mirror deflecting the beam towards the scanned area, and an optical receiver subsystem, which detects the laser pulse reflected from the object. Since the speed of light is known, the travel time of the laser pulse can be converted to a precise distance measurement (Hiremagalur et al., 2007; Vosselman and Maas, 2010). The precision of the technique is mainly affected by instrumental errors, point resolution and laser beam divergence (Vosselman and Maas, 2010). It can be represented by the standard deviation of each single point data measurement which amounts typically to a centimetric accuracy at a distance range of 100 m (Lichti and Jamtso, 2006). However, the accuracy of the global point cloud data is higher than the accuracy of a single point data because of the very high spatial resolution and density of collected points (Linendenberg and Pfeifer, 2005; Abellan et al., 2009).

This amount of data is the main advantage of TLS compared to classical geodetic techniques (e.g, tacheometry, GPS).

The potential of TLS for the monitoring of geomorphologic processes has been demonstrated in the last years, mainly for defining the structure of rocky slopes susceptible to rockfalls and rockslides (Abelan et al., 2009; Oppikofer et al., 2009; Sturzenegger and Stead, 2009) or for characterizing the dynamics of slow-moving (typically a few centimeters to a few meters per year) processes such as ice glaciers (Bauer et al., 2003;

Schwalbe et al., 2008; Avian et al., 2009) and landslides (Teza et al., 2008; Prokop and Panholzer, 2009;

Travelletti et al., 2008). The kinematics and, more generally, the geomorphologic changes (depletion, transport, accumulation of sediment) can be monitored and characterized with different analysis methods such as (1) multi-date DEM comparisons (Bitelli et al., 2004) yielding a 1D information on elevation changes, (2) point clouds comparisons using the shortest distance approach (Oppikofer et al., 2009) or (3) point/object matching approaches (Travelletti et al., 2008; Prokop and Panholzer, 2009) yielding 2D or 3D information through the computation of displacement vectors.

In the last few years, automatic matching algorithms applicable to TLS data have started to be developed because of their capability to fully exploit all the geometric information available in the point clouds (Teza et al., 2008; Monserrat and Crosetto, 2008). The objective of these techniques is to find correspondences among typical features or objects located in multi-temporal point clouds assuming that the tracked object has a constant geometry in time and/or a perfectly rigid behaviour.

The objective of this work is to propose a new method to measure the 3D displacement field of a slow-moving

clayey landslide, and derive displacement and deformation maps from repeated TLS acquisitions. The method is based on the application of a normalized cross-correlation function in order to exploit the complete geometrical information available in the point clouds. The hypothesis is that for objects scanned from a unique view point, simple 2D correlation functions (largely used in digital photogrammetry analyses) can be applied on multi-temporal point clouds and yield a range of accuracy comparable to complex and time- consuming 3D Surface Matching algorithms. Numerous examples demonstrated the efficiency of such type of statistical function to detect the displacement field of landslides from satellite and aerial optical images (Casson et al., 2003; Delacourt et al., 2004; LePrince et al., 2008), but only a few work has been carried out to apply this approach to TLS point clouds (Travelletti et al., 2008; Schwalbe et al., 2008).

The performance of the method is tested on datasets acquired at the toe of the Super-Sauze mudslide (South French Alps) over a period of three years (October 2007 - May 2010). First, the main geomorphological and kinematical characteristics of the Super-Sauze mudslide are presented. Second, the principles of the method are explained. Third, the application to the landslide dataset is detailed and the performance is evaluated among other measures of displacement. Finally, a strain analysis is applied in order to define the deformation and displacement regime of the landslide, and identify some possible controlling factors.

2. Experimental site: the Super-Sauze mudslide

The proposed methodology has been developed to monitor the displacements and the deformation pattern of the Super-Sauze mudslide, which exhibits cumulated displacements of a few meters per year and is representative of landslides developed in clay-shales.

Continuous movement of the mudslide may be maintained over long distances (hectometric to kilometric) because of the availability of a continuous sloping path in a stream valley, and over long periods of time because of the continuous enrichment of material from the main scarp.

The Super-Sauze mudslide has developed in the Callovo-Oxfordian black marls of the Barcelonnette basin (Alpes-de-Haute-Provence, France) (Fig. 1A, B), in the upper part of the Sauze torrent characterized by a gully-type morphology. The area is featured by a fault system affecting the black marls acting as predisposing factors for the formation of the mudslide scarp in the 1970s. Quaternary deposits (moraine, rock glacier debris) are also overlying partly the marls in some parts of the catchment.