Strange "amp_dispersion" output (cities are more unstable than water?)

[Shima-shoki]

Hi, I've been testing dolphin with the "burst2safe" module, which allows us to download a subset of the original huge Sentinel-1 SLC file. https://github.com/ASFHyP3/burst2safe?tab=readme-ov-file

I chose to download a single subswath, and two-bursts in Japan. The target area is composed of forest (~ 40%), agricultural fields (~ 20%), water surface (~ 10%) and settlements (~20%). The study period is 3 years (2019 ~ 2021). This area is snowy region, so I didn't include the S1 data during winter seasons (December ~ February) after the data download. There are 69 scenes in total.

!burst2stack --rel-orbit 46 --start-date '2019-01-01' --end-date '2022-01-01' --pols 'VV' --swaths 'IW3' --extent 138.78593 37.55812 139.1150 37.7795 --output-dir 'BurstDataDownload-ISCE2-InSAR/BurstData' --all-anns

I followed standard preparation steps to execute InSAR processing with dolphin by using the ISCE2 stackSentinel.py module without significant errors. Then, I executed the command below to prepare a dolphin config file (I'm using Google Colab to run the code).

!dolphin config --slc-files /content/drive/MyDrive/SatelliteImageProcessings/InSAR_Python/BurstDataDownload-ISCE2-InSAR/merged/SLC/*/*.vrt --strides 8 2 --threads-per-worker 8  --n-parallel-bursts 2 --enable-gpu 

The problem is that even though most of the PS-related files were successfully created without errors, the "amp_dispersion.tif (and amp_dispersion_looked.tif)" showed strange values. My target area contains forested areas, cities, agricultural lands, and ocean, so I expected low-amplitude dispersion values in urbans, and high values in other zones. However, the result was opposite amp_dispersion_looked (cities) > amp_dispersion_looked (ocean) 👀 .

amp_dispersion_looked
Center: cities, rivers, and paddies. Right: Ocean

The actual area of interest (It is rotated 90-degrees counter clockwise in the amp_dispersion_looked image)

I was confused at first when I saw that no PS points were chosen, and I noticed that the amp_dispersion threshold is 0.25, which is much lower than most of the land features in this area. Maybe something is not correctly working when I used the burst-wise data download?

I attached the relevant files in my processing. Any help is very much appreciated!

co-registered SLC files by ISCE2 software
PS files from dolphin output
dolphin_config.yaml file

*mean amplitude looks generally fine

*but the temporal coherence very low 😱 Maybe the data have issues...?

[scottstanie]

This is interesting... I don't have an obvious answer here.

What city is it? I wonder if there's one/a couple messed up images that drastically affect the dispersion and temporal coherence.

[Shima-shoki]

@scottstanie
Thank you for the prompt feedback!
This city is called "Sanjo city" in Niigata prefecture, Japan. I used the bounding box of [138.78593 37.55812 139.1150 37.7795] (W S E N) to get the S1 data.

This is not the first time I've got the bad PS estimate in Japan, when I tried dolphin in other region (Boso peninsula in Chiba prefecture), a similar output (no-PS points) was returned. Therefore, my current guess is something wired is happening in the burst-wise data download, but not certain 😢

Let me try this burst-wise data processing approach in Jakarta, where I've already got a nice result by using the full-SLC dataset. If dolphin works there, the data may not have problems, but if dolphin returns the same strange amplitude dispersion result, maybe the burst-wise data download affects the InSAR processing procedure (If that is the case, we need to retrieve the entire huge SLC dataset to run the InSAR things 😭 ). Even though the burst2safe developer team is said to have tested the module upon ISCE, I believe it is worth to further investigate...

[scottstanie]

Hmmmm this might be some bad interaction with burst2safe / burst2stack and ISCE2 actually...

I gave it a shot (burst2stack worked for me, thanks for providing the command 👍 ) but I used sweets / isce3's geocoded SLC module (via COMPASS).

It looks much more like what you'd expect- cities are high temporal coherence/low amplitude dispersion

mkdir data
 burst2stack --rel-orbit 46 --start-date '2019-01-01' --end-date '2022-01-01' --pols 'VV' --swaths 'IW3' --extent 138.78593 37.55812 139.1150 37.7795 --output-dir 'data/' --all-anns
sardem --data-source COP --output-type float32 --output-format GTiff -o dem.tif --bbox 138.5 37.4 139.3 37.9
eof --search-path data/ --save-dir orbits/ --force-asf

then the sweets config (which... kinda failed a little, but not badly. i just need to see what it didn't like. Hadn't used sweets in a year)

sweets config --bbox 138.78593 37.55812 139.1150 37.7795 --start 2019-01-01

but it successfully skipped the data download/orbit/dem downloading. so on my machine with a gpu:

taskset -c 0-16 sweets run sweets_config.yaml
# it runs, i forgot to turn off unwrapping since i'm doing to do dolphin anyway... oh well

dolphin config --subdataset /data/VV --slc gslcs/t046_09750*/*/t*.h5 --output-bounds 138.78593 37.55812 139.1150 37.7795 --enable-gpu -s 4 2 --threads-per-worker 2 --unwrap-method spurt
XLA_PYTHON_CLIENT_MEM_FRACTION=.20 taskset -c 0-8 dolphin run dolphin_config.yaml

[Shima-shoki]

@scottstanie
Thank you!
I haven't used isce3, so it's impressive to see that it gave the results what I'd expect to get (very clean coherence & amplitude dispersion!). So maybe the ISCE2 was built to handle the entire SLC dataset at once, rather than the burst-wise approach. I also know that there is a similar approach to only download & process necessary bursts (pyGMTSAR), and it worked fine when I tested before, so this problem looks specific to the ISCE2 software. However, ISCE2 returns a lot of processing log messages, so it is difficult for me where the problem occurs unless the processing is halted😢

Btw, it is fortunate to narrow down the problem that the burst-wise data & dolphin looked okay, so I only need to change the SLC co-registration process 👍 Let me have a try with another software to get the CSLC. I think that it'd be nice to give it a try first for me with ISCE3 (it's new to me, but thanks for your commands I think I'm already at 90% of the goal).