It often happens, that Sentinel doesn’t take pictures of an exactly same place. One-two burst difference may take place.
I am wondering, if two pictures overlap covers area, that I’m interested in, what is the maximum allowed image shift, so that co-registration can be trusted?
I assume you are talking about azimuth shifts. That is common for Sentinel and it’s not a problem.
You should not have any coregistration issues.
Moreover, you can put consecutive images in the SLC folder: Yuxiao developed a simple algorithm to merge them, and you can treat them as a whole image.
I attachements you will find the screenshot of the radar footprints. Area of interest is represented by the yellow pin. I´m using 79 Sentinel 1A images acquired from the same orbit number with descending track. 12 of these images are shifted around 19km w.r.t. the median.
Based on Your answer I assume I can continue, but I am curious about the accuracy of co-registration and further analysis. Is the radar imaging geometry the same in this case?
The coregistration accuracy is independent of footprint shift.
A simple-way-to-think for how-satellite-works is that it scans along a swath. The swath has a fixed “width” (but we call it range) but as you scan along you can have a long “length” (we call it azimuth).
For the purpose of distributing products to the public, one needs to manually “cut” the long slice into shorter products. And you see the footprints. However, when ESA distributes products, how they “cut” each time is a bit random. That is why the footprints are not always the same.
Imagine you copied a piece of paper with words printed on it for ten times. Then, each time you cut one paper in half, you do a random cut. Then you try to “coregister” the cut half of the 10 pieces of paper. You just need to figure out an initial offset to start. How accurate can you “coregister” (or, as you can think of it, it is simply alignment) is independent of how you cut.