Thank you for your interest in COSI-Corr and for participating in this forum.
1- If you're expecting dune displacement to be around 5 pixels, you can use 32x32 pixels correlation windows. I would however suggest using multiscale 64-32 windows. Results should the same, but more robust against noise and decorrelation.
2- The ortho-rectification quality is always the heart of all solutions and problems. The best things is to be in a situation where the DEM quality doesn't matter much, and it's usually possible with ASTER images. You can have a look at figure 2 of this paper:
http://www.tectonics.caltech.edu/slip_h … SE2008.pdf
In short, if the images are taken with similar viewing angles and if the two images are orthorectified using the same DEM, potential errors will cancel out. With ASTER images, if the image footprint overlap very closely, it's usually a good indication that parallax effects will be minimum (you can try to quantify it roughly with the formula in the paper).
3- The automatic destriping method we're proposing assumes that you generated the ASTER ortho-images using COSI-Corr, starting from the L1A images. I am guessing it could still work even with a correlation deduced from the ortho-ready images, but we've never tried, I can't guaranty the results will be good. For destriping, the idea is to chose areas were ground motion didn't occur, average them out in the direction of satellite jitter, and subtract it from the correlation. There are two ways you can go:
a- The easy way: the same file is used to determine the filtering model, and to apply the waveform subtraction. This would work only if you don't have dunes going all the way across (from East to West) your image. When the image appears in memory, you select a subset that encompasses all the height of the image, but that do not include any dunes (it's usually good practice to remove bad correlation points before doing this using the discard/replace values tool. Threshold SNR values less than 0.9 and EW and NS displacements that are too large to be physical. You can use the histogram in ENVI to help you determine where to cut the values) . I think most of this should be explained in the manual.
b- The more difficult way: if dunes are so widespread in your image that (a) cannot be applied, you need to mask out all moving dunes with NaN values in the correlation. You will then determine the correction model from a masked correlation, and you will apply it to the original correlation. The trick is to create a proper mask. If the dunes have largely moved, you can simply threshold all displacements and only keep the waveforms (assuming dunes have moved by a lot more than the amplitude of the waveforms to be removed). if you can't make this assumption, you'll have to mask things out manually...
4- Usually, images are acquired along similar orbits, hence wave artifacts are all aligned and you cannot distinguish whether artifacts come from one image, the other one, or a combination of the two. In that case, it doesn't matter which ancillary file you use, they will both produce similar destriping and will get rid of the waveform all in once. It's good to use the right ancillary if you have artifacts in several directions (CCD artifacts from SPOT and jitter from ASTER images, etc...), but in your case, you don't need to do it twice. Anyway, if you can't use the automatic destripe tool you can use the manual tool, you'll have to provide an average rotation angle to align the wave artifacts with the vertical.
5- We usually assume that the correlation file to be destriped was produced by a pair of two L1A data. The destriping may not be as accurate if that's not the case. I'm not sure I really understand your question here..
All the best,