Basic Subsidence Formation

[gkour]

Greetings all!

I am initiating this post to ask about basic steps in creating a subsidence map from numerus consecutive interferometric steps.
I have followed the information in here (and every other thread I could find):

https://forum.step.esa.int/t/subsidence-map-in-3d-view/3135
https://forum.step.esa.int/t/subsidence-map/10621/61
https://eo-college.org/resource/insar_deformation/

I work with ALOS-2 PALSAR images:
What would be the best pol for interferograms? HH or VV
Should the original product be calibrated or not?
Should I apply any speckle filtering/ and at which stage?

In my understanding I have to:

1. Coregister 2 products:
   -Would that be meaningful if I coregister the scenes in their full extent? (that might contain sea area)
2. Create Interferogram in SNAP:
   -I have used the defaults values for removing flat-earth/topographic phase as well as the coherence evaluation
   Do I need to play with the values? (for which I have no understanding)
3. Apply Goldstein Filter:
   -Again with default values: Do I have to change them? Do I have to apply the filter 2 or 3 times?
4. SNAPHU:
   -Again default values (DEFO & MCF)
5. Unwrapped Phase to displacement:
   -The end results seem non physical in the sense that I get a displacement map that linearly increments displacement along the image (from lets say 0.5m to 3m)

Here follows a small snip-shot of the first flattened interferogram

The bottom right area corresponds to sea, so I would expect such phase behavior.
The top left area is the actual land and the fridges appear as such, and I have no idea if they seem “good” or not. This behavior of the fringes on the land area is general to the whole scene. As you can imagine the following steps follow this “origin” behavior.

Bellow I present 4 images of the full process on a small land subset:
with InSAR Stack Overview:

Top-left: Initial flattened interferogram
Top-tight: After the application of the G/S filter
Bottom-left: Unwrapped interferogram
Bottom-right: displacement map (coh>0.5)

I have displayed the displacement map in gradient color red to show that the displacement itself follows a gradient like behavior (displacement [-0.18m, 0.26m]) from one edge of the image to the other. It is not assured that in that particular patch I expect to find some subsidence, but still I do not knot how to assess these results.

Any inputs would be much appreciated!
Thank you all in advance

[ABraun]

the interferogram seems to be affected by ionospheric effects. We discussed this in here, please have a look at the solution proposed by mdelgado: Interferogram from Alos palsar 1.1 data

[gkour]

Hello again dear @ABraun!

Thank you for your reply, it was a big read the last post.
I did what is mentioned there with the parameters and I got rid of those slow varying waves of the interferograms. (by increasing the flat earth orbit degree)
I also did that with DEM assisted coregistration, which resulted in non varying waves in phase.

Below is the parameters used for the interferogram:

Here are some results (same pair as previous post, different arbitrary subset):

Top: Coregistration: Create Stack
Left: Filter G/S Wrapped phase
Right: Filter G/S Un-wrapped phase
Bottom: Coregistration: DEM-Assisted
Left: Filter G/S Un-wrapped phase
Right: Filter G/S Wrapped phase

I have also translated them into displacement, as you can see bolow:

Left: Create Stack + Coherency>0.5
Right: DEM-Assisted + Coherency>0.5

The low coherence on the right makes me unease with DEM-Assisted cor.
(How am I supposed to evaluate this based on: (Interferogram from Alos palsar 1.1 data) ?)

The left displacement image makes some sense. However, if you look at the first figure, the wrapped and unwrapped interferograms look the same. The only difference I observe is that the wrapped ranges in [-π, +π] while the unwrapped ranges in [-2π, +2π]

At the end though, how am I to be sure to extract the ionospheric effect so that the remainder is the actual subsidence?

[ABraun]

@gkour

The left images makes some sense. However, if you look at the first figure, the wrapped and unwrapped interferograms look the same. The only difference I observe is that the wrapped ranges in [-π, +π] while the unwrapped ranges in [-2π, +2π]

I’d say this is a good sign, because short temporal baselines will not result in interferograms with massive fringes, especially when the topographic phase was removed. So these look quite nice actually - good job!

The only errors I see are some strong phase jumps in the unwrapped interferogram. Maybe you can test if the unwrapping works better in SMOOTH mode instead of DEFO.

[gkour]

Yes I could try that.
I will share the results as soon as possible.
I however have a few follow up questions:
while playing with the ifg parameters, if I input high polynomial degrees i can get the elevation band generated from the ifg look similar to the elevation band generated by SNAP (with right click on product, generate elevation band).
I have not found any combinations that would result in an exact match of the elevation bands, which makes me think that the coregistration failed?
I will post different combination results, with DEFO and SMOOTH also.
But how can I readily assess the coregistration outcome ?

[EDIT]: Also, how to interpret the fact that with the Stack Cor. the wrapped and unwrapped phases were very similar? (top row of figure one)

[ABraun]

why should the interferogram look like the elevation band?
According you your screenshot, you selected to remove topographic phase, so any pattern which remains is no longer related to the underlying topography.

You can visually check the quality of the coregistration by creating an RGB band with the both intensity bands. The RMS of the coregistration can be found in the InSAR tool. Please check this tutorial InSAR Displacement mapping with ERS data (page 11) for more explanations.

[gkour]

Thank you for your reply!
now I got a better idea how to assess the coregistration.

Regarding the elevation band I did not mean that the interferogram should look like the elevation band.
Perhaps I miss communicated this part. What I mean is the following:
In the parameters of the interferogram generation, after one selects to remove topographic phase he can elect to create 2 output bands

1. Output elevation Lets call that elevation_ifg
2. Output orthorectified lat/lon.
   like in this screenshot (do not mind the parameters, they are just the default ones)
   
   
   
   I output both of them.

We can also get the elevation band directly from the product itself with right click on the product name + add elevation band as show here. Lets call that elevation_add

Hence I have 2 elevation bands for the same product!
here follows the 2 elevations bands:

On the left we have the elevation_add while in the middle we have elevation_ifg. Comparing these two it seems that the elevation_ifg is shifted to the left by almost 500m.
The elevation band on the right has been created in the same manner as elevation_ifg but with using the default coregistration/interferogram parameters and it seems totally off. Presumably due to bad coregistration or bad topographic phase removal. I thought to present it here just for display and comparison purposes. i will disregard it from further processing.

Next, I thought to orthorectify the product, along with its two elevation bands, to see if any of both would match the topography of the scene. Of course, I also saved the output band DEM during range doppler t.c., lets call this one elevation_DEM. Below you can see these images:

On the top: Elevation_DEM. Google projection suggested that this band matches the ground very good, so we will keep this as reference.
Middle: Elevation_add
Bottom: Elevation_ifg
It is observed that both elevations_ add/ifg do not match very good the DEM one. They are both adjacent to it, with a shift of about 200 or 300 meters each. This is apparent by looking at the cross symbol that apparently is placed at exactly the same spot in all displayed images.

-What to make of this?
-Since I get such a big shift between elevation_add and elevation_ifg, and since both of the them are not well placed on earth, can I really trust the end results of the coregistration and interferogram generation, and after all the end-displacement map?

[ABraun]

now I undestand. Interesting…

Was the same color scheme (min/max) applied to all three DEM outputs?

[gkour]

@ABraun

The only errors I see are some strong phase jumps in the unwrapped interferogram. Maybe you can test if the unwrapping works better in SMOOTH mode instead of DEFO.

Also, can you please tell me where are the jumps located in the 2 top interferograms of the

So I can see if SMOOTH will resolve this issue? and to understand what jumps are

[ABraun]

@gkour

Also, can you please tell me where are the jumps located in the 2 top interferograms of the

such abrupt changes should not happen in unwrapped interferograms

[gkour]

@ABraun

such abrupt changes should not happen in unwrapped interferograms

aah I see! not this is the bad interferogram, that I will not use, derived from DEM-Assisted coregistration and low degree polynomial order in interferogram creation.

[gkour]

@ABraun

now I undestand. Interesting…

Was the same color scheme (min/max) applied to all three DEM outputs?

You are right, here follows the same image as before, with the same color coding of the elevation DEM

Again the order is the same,

1. elevation_DEM
2. elevation_add
3. elevation_ifg

[gkour]

@ABraun

now I undestand. Interesting…

Was the same color scheme (min/max) applied to all three DEM outputs?

Now they are! in the previous post,
So, any ideas what to make of it?
Would that make the end displacement map not-trustworthy?

[ABraun]

that looks better. Indeed, smaller differences of the used DEM can occur depending on the resampling and pixel size of the inputs. So the variations you see here are totally okay, because the main pattern of displacement will still be derived from the phase difference of the two input products. Topographic phase is just removed to prevent unwrapping errors, so it does not harm if it is differently estimated for the inputs.

[gkour]

OK I see! So after all I can trust the end result? -Would I have to apply any shift to the end result so as to make it match its true ground position considering the elevation_DEM?

I will play a little with the interferogram parameters and post the results again to discuss them, so I can understand how to evaluate them

[ABraun]

you can check if there are preprocessed interferograms of your area in this portal: https://comet.nerc.ac.uk/COMET-LiCS-portal/

They are a good reference regarding how good ones look like. Click on a tile, then “Link” then “interferograms” and select a date. There are png files for comparison.
The differences regarding the DEM do not cause shifts, but you have to level all displacements at one stable point. We had a discussion on this here: Different results in the study of land subsidence using the Sentinel-1 satellite image

[gkour]

Sadly there are no available ifgs for my study area, in that site.

Here follows an image of ifg generation:

the ordering is top row left to right and then bottom row left to right (1 2 3 4 5)

1. Polynomial Degree 6
   Orbit Degree 1
2. Polynomial Degree 6
   Orbit Degree 2
3. Polynomial Degree 6
   Orbit Degree 3
4. Polynomial Degree 6
   Orbit Degree 4
5. Polynomial Degree 6
   Orbit Degree 5

I have also tried these combinations with varying Polynomial degree from 2 to 8 taking even number, resulting in highly similar ifgs (if not identical).

By the looks of 4 and 5 it seems that the ionospheric effect is eliminated.
-Is it true?

I then chose to proceed with 4, so as not to take the extreme values of combinations.
Then i performed the goldstein filtering and phase unwrapping as shown:

left: Wrapped right:Unwrapped

I am impressed on how similar these two are.
i do not know if that is expected.

Next follows the displacement map, based on the unw/ ifg.:

Left: coherence estimation, right:Displacement (for coh>0.5)

What do you think of the process so far?
Am I lead astray by seemingly “good” results? or are they actually useful results?

[gkour]

@ABraun

The differences regarding the DEM do not cause shifts, but you have to level all displacements at one stable point. We had a discussion on this here: Different results in the study of land subsidence using the Sentinel-1 satellite image

I think I got what you said here, i will take this to the big scale. I will also take two consecutive pair this time, based on the approach mentioned above. Subtract the relative difference and see what happens.
I am thinking of using the Polynomial Degree 5 and Orbit Degree 4 combination as being the closest match to the default values of snap, while eliminating those slow varying waves.

After all, i would see no relative displacement in rivers and in urban like areas, right?
just trying to imagine how to assess the end result

PS. Do you think it would be best to coregister non-calibrated products?

[ABraun]

Can’t tell which areas are suitable as stable reference point. Once you identified one, make sure it’s value is zero in all pairs for best comparability.

What do you mean with ‘uncalibrated products’?

[gkour]

@ABraun

What do you mean with ‘uncalibrated products’?

I mean Radiometric Calibration (or even Speckle filtering).
So far I have been using non-processed products