To apply split-bandding techniques, it is required that I worked on coresgistered SAR pairs where the azimuthal phase ramp is removed. This is something we already discussed and, in the end, the option of deramping data and not reramping after coregistration has been added to SNAP. For that, I would like to express my thanks for having listened to the community needs.
Unfortunately, with the most recent SNAP version (8.0.3), the signals properties of deramped SAR coregistered pairs seem disturbed. Forming an interferogram on deramped SAR pairs (Grandin, 2015) leads to data gaps (see image below). This was not the case in previous SNAP versions.
I’d like to share more details about the problem I encountered. While coherence was the major issue, filtering the interferometric signal using goldstein phase filtering brings additional clues to the situation.
The below image show the intensity and phase of the filtered interferogram. We notice an abundance of gaps.
FInally, when comparing the interferogram of deramped data compaired to ramped data, the phase information is completely different (all other parameters being equals).
The below image shows the classical inteferogram (left) and interferogram following Grandin’s procedure 2015 (right, based on deramped master and slave images). You can observe the fringe pattern being completely different, with important discontinuities between bursts. All this indicate a probable mis-computation of the deramping operation. Please correct me if I’m wrong.
Sorry about the late reply. The gaps in the coherence and interferogram images are from the gaps in the original Sentinel-1 TOPS data in which the burst data are separated by the gaps. I think we should deburst the interferogram before filtering it. A proper way to compute the interferogram and coherence should be like the procedure described by the following graph:
No “Disable Rerramp” is needed in Back-geocoding
Which gaps are you talking about? I am not talking about the gaps between bursts, but the gaps inside the burst. We can see hundred of them in my figures.
Unfortunatelly, this is an unrecommended procedure. Each burst is an independent scene. Deburst, and geoprojection operators are supposed to be performed at the very end, when possible.
It depends on the application. You might need coregistered images without the azimuthal phase bias.
In this below figure, I compare the interferometric coherence between a coregistered SAR pair following a traditional approach (left), with a coregistered pair where the reramping has been disabled, following Grandin’s approach (right)
You can observe vertical stripes everywhere and a lot of regions where the processing stopped.
The error is systematic for a given SAR pair, but highly different from different SAR pairs. In some case, it just “works” (aka no vertical stripes). Still, local data gaps persist, which are amplified with filtering operations.
The problem was caused by that the valid zeros in the source bands are treated as invalid pixels because the source band’s noDataValue is always set to 0. We have fixed Back Geocoding to avoid this problem. But we are still looking for better solution so that it can be fixed fundamentally. With the current fix, you should see no gaps in the interferogram and coherence. But you need to disable the “Mask out areas with no elevation” option in Back Geocoding when you coregister the images. The fix should be available with the next patch release. Thank you for pointing out the problem