How to assess the quality of the unwrapped phase

Hello everyone,

I am working on land deformation over a mining site using SBAS (MintPy). I have created 120 unwrapped interferograms where I am visually assessing them trying to discard the low quality ones (e.g phase jump). Below we can see the coherence, wrapped and unwrapped phase of the AOI where numbers correspond the the following dates. 1). 2021/01/29 - 2021/02/10, 2). 2021/01/29 - 2021/02/22, 3). 2021/01/29 - 2021/03/06

Coherence

Wrapped phase

Unwrapped phase

Looking at the unwrapped phase of (1) it looks good to me but for the other two, the unwrapped interferogram is completely different despite the fact hey share the same reference date. the unwrapped phase of number (2) is above -10 radians for the whole scene (it does not look normal to me) and for the number (3) we see lots of variation between -10 to 10 radians. Can this be due to tropospheric noise ?

My first though is to discard number (2) and (3) from SBAS processing in MintPy. I do not understand why there is such a huge difference amongst all three unwrapped interferograms. Does this has to do with 1) the low coherence of the area , 2) atmospheric noise or anything else?

There are quite a few interferograms in my analysis that look like the ones above. Am I right to assume that there is something wrong with those unwrapped phases and need to be discarded ?

One last thing, Does the positive values on the unwrapped phase indicate subsidence (signal travelling longer distance) and negative ones uplift? or is it the other way round

I would appreciate any comments or thoughts on this, especially by people that are more experienced in that filed such as @EJFielding @ABraun @qglaude

1 Like

You are correct to be suspicious of the unwrapping of interferograms 2 and 3. The problem here is coherence is very low and degrades quickly with the time interval length, which is a typical situation in areas of significant vegetation cover. Your interferogram 1 has a 12-day interval and acceptable coherence so the unwrapping is reasonable except in a few small areas. Interferograms 2 and 3, with intervals of 24 and 36 days have very low coherence, and the unwrapped phase is not usable. You can see that the wrapped phase looks like almost pure noise, so the unwrapping has no information. It seems that only the 12-day intervals in this area are usable. You can try to make a MintPy time series with only the 12-day pairs, but it will not be very accurate because there is only one pair connecting each date.

Because the interferograms have low coherence, it is difficult to see whether atmospheric effects are having any effect.

4 Likes

Thank you very much for your response.
I suspected that coherence might be the major issue on getting good quality interferograms. I suppose, there is no much I can do given that I cannot collect sentinel-1 data more frequently than every 12 days
for my AOI.

You may be able to extract some pixels with high coherence with one of the persistent scatterer InSAR (PS-InSAR) methods. I have not done that processing myself, but there are many who do that. The SNAP stack processing can be loaded into StaMPS that is one of the PS-InSAR systems.

1 Like

This is the wrapped (top image )and unwrapped phase, the transitions between colors in the unwrapped phase (middle) looks smooth with minimal noise nor wave-like atmospheric patterns, also I see minimal resemblance between the wrapped, unwrapped phase pattern and the topographic phase pattern (bottom image).

Please correct me if I am wrong, I am not expert in assessing the quality of unwrapping and also I am not sure why those square areas were added to both sides of the image of the unwrapped phase, I have aligned windows vertically while synchronizing all views and those squares represent blank regions on the equivalent wrapped phase

Also those individual spikes in the unwrapped phase’s histogram looks weird, what error could they reflect?

Thanks in advance for taking a look

The spikes in your histogram of phase are caused by the areas of zeros on the two sides of your interferogram. The SNAPHU program unwraps those areas into a series of steps.

Your wrapped interferogram looks very strange with the huge number of fringes. It cannot be correct. Maybe the coregistration in not good.

1 Like

This is how the RGB composite of the slave and master intensity images look like.
Master>> Red
Slave>> Green
Master>> Blue
I see a lot of pixels highlighted with pure green color which means the coregistered stack is dominated by the intensity from the slave image. Should I retry doing the coregisteration separately instead of including it in a graph builder ?


this is how an interferogram from different pairs of exactly the same fram but with a smaller temporal baseline looks like. A little too noisy

That other interferogram does not look correct either. The very large number of fringes in the along-track direction is a symptom of bad coregistration. Are you doing the Enhanced Spectral Diversity (ESD) estimation? It is possible that the low coherence is not allowing an accurate ESD estimate. It has been a while since I ran SNAP on Sentinel-1 data, so I don’t remember the details.

It would be a good idea to try processing one pair with the regular set of steps instead of the Graph Builder.

yes
I did regual set of steps and got the same results actually. The earthquake happened in this area along the Anatolian fault was 7.8 magnitude of Richter scale and it was really destructive, Maybe that’s why there are so many fringes and also reason for low coherence.

Here’re the results from a different area affected by the same earthquake movements, I suspect unwrapping error at the area inside the ellipse shape

Even though I have created a subset using a mask to isolate the zero date on the image borders from the non-zero data but still got grid like pattern on the sides of the unwrapped phase.
I set a value of 200 for both row and column overlap.
I didn’t know how to mask out this area which is a part of a water body (inside the pink polygon) before creating the interferogram so if you have got some idea, would be appreciated.

Here’s how the displacement profile looks like across the other area of low coherece (which is a vegetated land)



I see now you are trying to unwrap the Sentinel-1 interferograms for the 8 February 2023 Mw 7.8 and Mw 7.5 earthquakes. That is why you have so many fringes. I don’t think it is possible to correctly unwrap these interferograms because there is too much displacement. You might have seen that people have run pixel offset tracking on the coseismic Sentinel-1 pairs, and there is over 6 meters of displacement on the northern fault that ruptured in the Mw 7.5 earthquake. Since a C-band fringe is a little less than 3 cm, there would have to be 200 fringes to measure that displacement, but the fringes near the fault were aliased or lost, so the unwrapping will never be correct.

2 Likes

Very strong notice! I appreciate it. I was wondering why most if not all the unwrapped phase have observable error. Maybe that’s why ongoing projects for Turkey’s February 6th are working on L-Band ALOS PalSAR data pairs