Regarding the usage of correct unwrapping error (bridging)

[Lincoln Olayta]

Hi again everyone,

I'm currently using Mintpy to process a two-year stack of Sentinel-1 data from ISCE (created using stackSentinel.py)

My area of interest is located on a volcanic island, and I am aware that unwrapping errors would affect it. I explored the usage of phase_closure and bridging by scanning the pages of Yunjhun's 2019 paper. 

I'm aware that the unwrapping error correction algorithm relies on the connected component output by SNAPHU. I've attached the connectComponent picture files. I removed some interferograms that get masked when creating a "non-zero" mask for maskConnComp.h5.

Here are my problems:

1. I can see that even removing a large amount of pairs (33 out of 160 pairs) with "zeroed" connected components on the island of interest, it still masks (almost entirely) the island. Is there any other way to do this aside from removing large amounts of pairs (ifg index)? since my area is highly noisy. For example, increase the threshold of the connected components? or is it possible to use other masks (like Spatial or temporal coherence) as connComp to use for unwrapping error?

2. After removing some pairs and proceeding with unwrapping error correction, I would like to know how the bridging (which I think fits my problem relatively well) can be performed. I know that I can create a waterMask (with --roi-poly) to the select area, but I'm unsure how big the area should I include with it (and how the algorithm will bridge it). In addition, is it advisable to apply a ramp in bridging method on my specific usecase?

I removed some interferograms with zeroed out connComp on the island, and created a maskPoly.h5 which includes the island and the reference area. How big will this affect the correction on bridging? I'm assuming that bridging relies on MST of the connComponents to perform the correction, but I didn't understand the "breadth-first" algorithm to choose which bridge to connect on the reliable regions.



I attached the connectComponent picture file; the maskPoly.h5 I used (on Problem #2); and the timeseries wrap files, the commonregionsample, numTriNonZeroAmbiguity, network.pdf . If there are any other files needed, please don’t hesitate to ask me. 


Any advice (from using the bridging, mitigation of the masked island from ConnectComponent masking); and the whole concept of unwrapping error correction [like considering to use bridging+phase_closure] on MintPy) is welcome :) 

Thank you!!

Lincoln

[Lia Joo]

Hi Lincoln .

Did you use a notebook available on opensciencelab ? Or write your own code to get these results?

Thanks

--
You received this message because you are subscribed to the Google Groups "MintPy" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mintpy+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/mintpy/3b515b3c-ba24-4e9f-8ab5-a91ad62d719an%40googlegroups.com.

[Lia Joo]

Hi Lincoln.

Thank you for your clarification. Yes it helps a lot.

Lia

On Thu, Mar 28, 2024 at 9:34 PM Lincoln Olayta <lincoln...@gmail.com> wrote:

Hi Lia,

I have not used the opensciencelab to produce these results. I used the smallbaselineapp.py (and inputted the files from the results of stackSentinel.py). The removal of pairs with "zeroed" island (area-of-interest) is done by viewing the connectComponent from ifgramStack.h5. The maskPoly.h5 was created using the prepared waterMask.h5 and the generate_mask.py --roi-poly interactive command. My area is located in a tropical country, which is usually plagued by the noise created from dense vegetation etc. when using C-band SAR, probably causing the noisy interferograms pair. In addition, since the island is disconnected, I think the unwrapping software SNAPHU (used by ISCE) is prone to unwrapping errors as observed in connectedComponents results. 

If this will help add more information, I also used the latest build of MintPy & ISCE built (using cmake) on macOS ARM architecture with its dependencies provided by conda.

I hope this clears things up. 

Kind regards,

Lincoln

To view this discussion on the web visit https://groups.google.com/d/msgid/mintpy/814e0949-097b-4606-b895-ea9f7c344239n%40googlegroups.com.