Topic: Landslide detection

Dear all,

I have a Pléiades image from 16 July 2012 which I want to compare with a Pléiades image of 9 July 2013, using COSI-Corr to detect landslide displacements. Also, I have tristereo Pléiades images of the area of 9 july 2013 with which I made a georeferenced DEM. I have some questions in using COSI-Corr with this data:

1. When no pre-hazard and post-hazard (hazard=landslides) DEM's are available, it is assumed that the elevation stays constant, right? When we make this assumption, it would be no problem to analyse displacements from 2012 to 2013 with a DEM from 2013 ? In the COSI-Corr guide 2015a I read that both pre-earthquake image and post-earthquake image make use of the pre-earthquake image for selecting tie-points.

2. Why is a normal geocoded DEM not sufficient for orthorectification in COSI-Corr ? I read in the COSI-Corr guide 2015a, and all related papers, that a "shaded DEM" is used. No difficulty to create a shaded version, but I was just curious about their difference in the application.

3. COSI-Corr relies on pixel correlation, but I do not see an application for landslides here, as the pre-hazard image looks different than the post-hazard images on the places where a landslide occured, for example: from houses to debris. Or how do you see it ?

Looking forward to your answer,
Best regards



Re: Landslide detection

Dear Elise,

1. Ideally if the topography has changed between the two dates, you want to use a pre- and post- DEMs. However in practice we rarely have that luxury. The use of 1 DEM can be limiting if the area of interest sustained large elevation change AND if the images have a large difference in viewing angle. In that case, the use of 1 DEM can cause distortions that can be hard to differentiate from real ground displacement. That is why it is recommended to prefer images whose difference in incidence angle is small.

2. The use of a shaded DEM is to ease the selection of tie-points between the DEM and the image. The shaded DEM will make topographic features to stand out more than from the DEM itself. These features can then be "tie-point" to the image. It is not required though. However for the orthorectification proper, you need to use the DEM and not the shaded DEM. If you have a stereo or tri-stereo pairs that you'll use to generate a DEM, you'll also have the orthorectified images as a by-product. You should use these to select the tie-points with all other images (e.g., 16/07/12) and not the DEM (or shaded DEM) directly.

3. Pixel tracking only works for "displacement". A catastrophic landslide cannot be tracked because the surface in one image has been completely changed in the other one. You can track slow-moving landslide however. In the case of catastrophic landslide, you could potentially use pixel tracking decorrelation to locate the landslide extent.



Re: Landslide detection

Thank you François for these clarifying answers!
Best regards,