I am sorry if this post could seem too “basic” for most of you, but I am approaching to the “world” of Geospatial data analysis very recently.
For the moment, I am concentrating on the analysis of Sentinel-1 data and my particular interest is the Co-Registration of SAR images to use them for Interferometry.
Thus, I am trying to trace the state of the art for the Co-Registration methodologies and I started this topic trying to clear my mind about the topic and, possibly, to share recent studies/researches about it.
Till now, I understood that there are 2 main possibilities to Co-Register SAR images:
- With orbit information (+ DEM)
The first approach performs pixel-to-pixel co-registration and basically maps the SAR coordinates to Cartesian coordinates. It is what is typically done with a standard (?) pipeline in SNAP that more or less can be summarized in the following:
TOPS Split → Apply Orbit → TOPSAR Back Geocoding + ESD
However, the results of this approach strictly depend on the precision of orbit files. As far as I understood, the main problem here is that the precise orbits are available after 20 days for SAR products. So if you wish to analyse data acquired in the last 20 days, you need to use restituted orbits and it is very likely that this approach will not produce good results (maybe here I am speaking too broadly since the goodness of results may depend on the application).
On the other hand, the second approach does not rely on Orbit information but it should use Cross-Correlation to estimate the offset between the images, then you can perform a rigid image transformation (parameters should be estimated, too) to Co-Register the products.
I did not explore further this approach till now, and I did not even understand whether or not there are Operators in SNAP that perform this type of co-registration too.
Both of these approaches need an ESD correction (Improving co-registration accuracy of interferometric SAR images using spectral diversity by R. Scheiber and A. Moreira) that should estimate both the Azimuth and Range shifts comparing the two phases of the images to further correct the process of Co-Registering them.
Exception made for these 2 general approaches (and all their possible variations based on data knowledge and application), it is not clear (to me at least) if there are other deep-rooted approaches to handle the task.
I would really appreciate if you could comment this post to underly all the “errors” in my considerations above, but also if you could suggest me some interesting reading to further understand the problem and its newest possible solutions.