[jaroslav]

Hello everyone!
During co-registration of my dataset I often get a warning, my area of interest doesn’t have enough coherence.

 WARNING!!! WARNING!!! WARNING!!! WARNING!!! WARNING!!! WARNING!!!
 Your AOI does not have enough coherence for accurate subpixel coregistration!!!
 A linear azimuth ramp may appear on your interferograms as the result!!!
 Possible solution is to increase lines or samples by steps of 1500 pixels!
 Please manually check the interferograms regarding date 20170214!!!
 WARNING!!! WARNING!!! WARNING!!! WARNING!!! WARNING!!! WARNING!!!

I understand, that possible actions are already mentioned in the warning, but maybe solution isn’t that linear. I’d like to ask, how should I adequately react to this warning? Do I change master and try one more time, or try to increase an AOI to include more coherent areas, even though there are already one small town inside, or take a failed image out of the dataset right away, or look at the interferograms, at first, and then remove uncoherent images and try co-registration one more time, or should I try everything for each different case?
Could you give an advice?

Jaroslav

[periz]

Hi Jaroslav,
first of all, take this as a temporary message. We are working at general solutions for avoiding these problems.
Secondly, when you get such messages, the first thing you should do is checking if this has an impact on the phase.
The quickest solution is temporarily discarding images that generate this problem.
Otherwise, the problem could be fixed by processing a wide area. Ideally, extracting the whole subswath should help.
But once more, it is a temporary problem, we’ll soon provide some improvements

[venera]

Hi Jaroslav,
How did you overcome your issue?
I am facing these warnings, however I am processing a whole subswath..

Thanks.

[jaroslav]

Hello,
I didn’t actually completely overcome this issue yet.
However, the master choice seems to help sometimes. In cases of problematic co-registration I usually try to manually evaluate image list in order to exclude the most noisy ones and choose the most contrast image as a master.
Somebody, correct me if I’m wrong, please.

[venera]

Thanks for the quick response.

[Yuxiao]

Hi Venera,

First of all, please make sure you are using the newest version.

Second, it would be helpful to check the coherence of the pair that shows this warning. The goal is to check that if the coherence of interferogram is indeed so bad that accurate subpixel coregistration could not be performed.

At last, it would be helpful if you pass me (qyuxiao@sarproz.com) the logs to your processing site, in which case I would understand better what is going on there.

Q

[venera]

Hi Yuxiao,

I am using the newest version.
I solved my problem with co-registration. I’ve just realized that it was due to corrupted downloading of Sentinel images.

However, do you have some general recommendations or tips to get reliable results from Sentinel data processing (parameters and so on)?
I am doing processing Sentinel 1 data over oilfield.

Thank you,
V

[Yuxiao]

Hi V,

For the data corruption part, it is a known issue. In sarproz we implemented several ways to detect corruption of SLC images. You should get warnings/errors if they are corrupted.

In the case where no warnings/errors were given, it is always helpful to check:

(1) The JPGs of SLCs in EXT/yyyymmdd folders and in FITTED folders. If the JPGS are showing strange/abnormal things it should be a warning that something went wrong;
(2) The interferograms. The interferograms should always be able to tell you something. If you spot something abnormal from interferograms then we should stop there and investigate the problem.

In general it is always important to check the outcome each time you click a button and make sure nothing has gone seriously wrong before moving to the next step.

Q

[jaroslav]

Hi Yuxiao,
Doesn’t master choice influence co-registration process a lot? I have often noticed, that outcome may be significantly improved just by choosing right master.
Additionally, I’d like to ask about Sentinel master choice methodology. Usually, master is chosen such, that the dispersion of the perpendicular baseline is as low as possible, but Sentinel already has small baseline difference. So can the baseline dispersion be ignored? What master choice criteria may be more influential?

Jaroslav

[Yuxiao]

Hi Jaroslav,

Yes and No. A few explanations:

Yes Part:
1. My experience is that spatial baseline is not a big problem for tops in general. Since (a) tops has such good control in its orbits precision & accuracy; (b) the baseline is usually quite small (within 200 meters in general). So geometric decorrelation is not really a big issue here. (BUT please note this is a general idea. Different cases should still be examined carefully.)
HOWEVER, it is always nice to choose an image close to the middle in the sense of spatial baseline.

2. In the sense of the traditional coregistration that is amplitude based (which is also required for tops before spectral diversity), it is also suggested to choose an image in the middle in the sense of temporal baseline.

NO Part:
3. However, in the sense of tops spectral diversity, the choice of master actually does not matter much in the sense of temporal baseline. The code will find an MST and do spectral diversity in that order. Regardless the master you choose, the MST tree will be the same.

In the older version of sarproz, as it is flawed, different masters will change how other slaves are resampled, which could indirectly influence the spectral diversity part hence the coregistration successful rate for tops. However, in recently release versions, since much has been improved, as long as (1) Master image is roughly in the middle in temoporal-spatial-baseline coordinate system; (2) No significant precipitation/snow on the date of master image; the result should be quite consistent regardless the master image you choose.

If in a newer version of sarproz, you do find a contradiction to my previous statement, please let me know and I will be happy to investigate.

Q

[Author]

Posts