Topic: COSI-Corr for sand dune movement in Qatar

Dear Sebastien, Dear Colleagues,

I was registered here two years ago and even posted once, while I was working on my Bachelor thesis. As I had time pressure I couldn't use COSI-Corr in my thesis. Now I am involved in a project, where we want to measure the sand dune movement in South-East-Qatar over different time lines, using COSI-Corr. My assignment is to get familiar with COSI-Corr and ENVI, but I don't have much experience with both programs.
I have satellite data from 1995 and 2004 (IKONOS) with a resolution of 1 m, so I cut out two images of the same area with a resolution of 800 x 800. They are georeferenced in ArcGIS (same coordinate system), but COSI-Corr tells me they are not georeferenced. I think this shouldn't be a problem because a pixel per pixel comparison should work here as well. While one of the satelite images is just black and white, the newer one has 3 bands. I also don't think this should be a problem. (In my next step I anyway want to compare the IKONOS data of 2004 with around 50 years old CORONA images.)
I tried to run the images through COSI-Corr with a lot of different options to compute the correlation, but I always just got images with a lot of noise as a result. I tried the Frequency Correlator with a lot of different settings as the Statistical Correlator, but I didn't get any results and I couldn't find the mistake I am doing. Maybe somebody has an idea what I could try to get some results. For example using the Freq correlator... what for a setting would you suggest for the data? (The dune displacement is around 100 pixel and the expansion of the dune is a little bit below 150 x 150 pixel.) I tried many different window and step sizes for example. Please help me, as I belive I am just missing out something.

Best regards, Fabian


Re: COSI-Corr for sand dune movement in Qatar

Hi Fabian,

You are using images that are very difficult to process and I would advise agaisnt it for someone who's starting to lear about COSI-Corr. For now, I would suggest you use Landsat 8 images, the panchromatic band only, and use the frequency correlation only. Then try to get a first result, and go from there.

Do not expect to be able to track dunes over more than about 10 years as the shape usually changes over time and dunes merge and split.