A multi-temporal image correlation method to characterize landslide displacements with a terrestrial camera

The objective of this work is to present a methodology to infer the displacement pattern of slow moving landslides from terrestrial multi-temporal images analyzed with an image correlation technique. The method is being tested on the Super-Sauze mudslide (South French Alps), which exhibits velocities from 0.01 to 0.4 m.day-1. The performance of the method is discussed by analyzing the displacement pattern observed over the period June 2007-July 2009. The monitoring system consists in a low cost high resolution optical camera installed on a concrete pillar located on a stable crest in front of the mudslide, and controlled by a datalogger. Every 4 days, photographs are registered at 11:00 AM, 12:00 PM, 13:00 PM and 14:00 PM. The processing method consists in applying an image correlation technique using a multi-resolution approach. The correlation results are first qualitatively interpreted in terms of pixel displacement, pixel velocity and direction in the image plane of the camera. The correlation quality is primarily controlled by: (i) variations of illumination angles and intensities between two dates, (ii) surface texture changes between two dates and (iii) slight movements of the camera. The average uncertainty is estimated at less than 1 pixel. For a quantitative analysis, control points in the field visible in the photographs are used to orthorectify the image and to derive a transformation matrix for converting the pixel coordinates into metric coordinates. The transformation matrix is applied on the pixel coordinates of each correlation results assuming that the global landslide morphology is nearly invariant in time. The displacement pattern is clearly identified in the image plane of the camera. The translation of the pixel displacement in terms of ground displacement is highly sensitive on the accuracy of the conversion matrix. Thus orthorectification must be done regularly after important topography changes to get a more accurate conversion matrix. The obtained displacement fields are realistic with good correlation coefficients and homogenous directions and amplitudes of displacements. The displacement maps show a complex displacement field of the landslide. The landslide body is very well distinguished from the stable parts around the landslide. Displacement maps indicate that the landslide kinematics is mainly controlled by the buried topography. During an acceleration period observed in May-July 2008, a difference of horizontal velocities between the lower (< 1 m.day-1) and the upper part of the landslide (~ 4 m.day-1) is clearly noticeable .