Observing sharp transition in phase and coherence data

[sharath-velishala]

Hello

I recently started working on dolphin to do time series analysis for land subsidence observations and I have used sentinel 1 slc data.
I had 38 slcs lined up and used isce2 (stackSentinel.py) to stack and coregister the slc files (Only 1 swath and 3 bursts were used as they covered my region of interest). Later I pointed the coregistered merged vrt files for the dolphin run. Initially I had RAM issues when using strides x, y as 1,1 respectively with default ministack (15), so I changed the strides from (1,1) to (5,5) just to make sure I could run the dolphin end to end.
Later I geocoded unwrapped phase, coherence, velocity (I had to create the vrts as the merged vrt from isce2 outputs shape wasn't matching with dolphin outputs given I have chosen 5,5 stride).
I started visualizing velocity and found that there is a sharp transition happening as shown below

Then I started visualizing the unwrapped phases and found the same sharp changes in the exact same position in all the unwrapped phases (as shown below)

I started to wonder if I had issue with unwrapping, so I wanted to check how the coherence files were looking and there were also having this issue, at the same exact position I see black lines appearing (its like we have stitched 2 different rasters together but then I did combine 3 bursts but these lines are no way aligned with how the bursts will be aligned). Following are few coherence images

also I want to show how the non geocoded coherence looked

I am currently stuck at this stage and unsure if this is a processing artifact or an issue with the input stack. If anyone has recognized this pattern or has suggestions on what to check next, I would be very grateful. Thank you!
Note : Kindly Ignore the regions outside the bursts and inside bbox, I tried giving nodata but wasnt successful.

Regards
Sharath

[scottstanie]

So just checking- you did all dolphin processing in radar coordinates (with inputs looking like this shape) and then after dolphin was finished, you geocoded with something and see these artifacts? What made this "coherence" image, is it something in isce2?

I'm thinking that it doesn't look like an artifact that would come from dolphin: in that last "coherence" image, it almost looks like the artifact line is already there (but warped into radar coordinates, so it's skewed).

I might try to just do make a simple interferogram with each of the SLCs that you coregistered and see how the look- i'm guessing it's either an isce2 issue (or maybe if you used it to grab a few bursts and stitch them, a burst2safe issue?)

[scottstanie]

yep that will be instructive to see the interferogram, you could run the isce2 version or do something simple like

from dolphin import io, utils
# whatever the path to the merged SLCS are:
file1 = "merged/20230106.slc.vrt"
file2 = "merged/20230118.slc.vrt"
slc1 = io.load_gdal(file1)
slc2 = io.load_gdal(file2)
ifg = utils.take_looks(slc1 * slc2.conj(), 6, 20) # not sure the best number for visualization

# then plot the data
fig, ax = plt.subplots()
ax.imshow(np.angle(ifg), cmap='twilight', interpolation='none')
fig.savefig("test_ifg.png")

[sharath-velishala]

Ran the above code snippet and we can see those lines clearly in the below image

Thanks for your valuable suggestions Scott :) . I shall run the isce2 coregistration (or may be full interferogram) again, check how the interferograms are looking and then go for dolphin run.

[scottstanie]

oh what are the bounds of your DEM? kinda looks like the DEM only was used in the upper left to flatten, and i bet there is no data in the rest of the image.
(you can check rio bounds --bbox dem.tif and compare to the bounds of the geocoded images)

[sharath-velishala]

Ohh yeah damn you are absolutely correct. So this was the first time where I passed the exact bbox (or rather 10% buffered) of my region of interest as I wanted to decrease the number of bursts and swaths. Previously I used to pass the bbox of the whole slc tile (all 3 swaths and 8-9 bursts).
Following are the bounds ( in the order - SNWE)

1. DEM - 26.028132, 26.085079, 74.052571, 74.182181
2. geocoded - 25.687294, 26.360280, 73.391812, 74.409303

so the yellow line in the above image was my roi and I gave a 10 percent buffer to get my bbox, which was then sent into stackSentinel.py -b.

Found the culprit :D

[sharath-velishala]

Hi Scott,

I have corrected the DEM and these errors are no more present.

Once again thanks 😀