Problems when running back geocoding/ESD on S1 SLC data


I am currently trying to generate an interfrogram from two S1 images. My steps are back geocoding > ESD > interferogram > deburst. The error I get is:

WARNING: org.esa.s1tbx.sentinel1.gpf.SpectralDiversityOp: RangeShiftOp: Cross-correlation failed for all bursts, set range shift to 0

As a result, I get a file full of NaNs.

I am running the Linux version of SNAP, running on Ubuntu 18.04. I have tried running the above xml chain using both versions 6.0.0 (which comes with the SNAP installation) and 6.0.6 (after upgrading), and get the same error regardless. I do not have this problem when running on a Windows machine on the SNAP GUI.

Any help would be much appreciated, thank you!

Nevermind, I managed to solve the problem myself. Apologies!

I have the same issue - how exactly did you solve it? I saw in some tutorials the for interferometry the Spectral Diversity Enhancement step was not even used. Not sure if it is necessarily needed?

If you are using single burst you shall not use ESD.

Thanks for the info. I am using multiple subswaths and bursts.


But according to this article the goal of the ESD is,



So, to get fine corrigisteration, what is the difference of using one burst and multi bursts?

Would you please to refer to the literature if that possible,

I also just checked with this tutorial by ASF on Interferogram generation where they use one complete subswath (including all bursts) and there is no mention of doing ESD:

Would you please to share more information, what exactly you did?

I am trying to generate VV and HV coherence from one S1A/B pair, see attached graph

What do you think, in case you create simpler graph than this, like the following one, Because, I think something went wrong during creation your graph. Putting in mind that only in this case, we could find out the source of the error,

Sure, it’s possible. I just thought that the more “complicated” one would make use of the internal parallelisation much more. snap seems to be taking a loooong time

Without ESD, very small misregistration will be translated in phase jumps between bursts. In case of stationary scenario, you can use the spectral seperation between burst on the overlapping area to find the optimal misfit between bursts and correct these phase jumps.

In the case of a single burst, you don’t meet the problem of these phase jumps (since there’s only one burst).

I agree with you, however, what about the flat area and undulated area, is that the same case, because in the case of undulated area many parameters could affect the corr. , But do you have any article refer to one burst corr. I myself didn’t find, It would be great to add up it in here.

So I am just following your recommendation for error tracing and creating that simplified graph. It is however not quite clear to me, how do I know the bursts I select for one acquisition’s subswath are also there in the other acquisition, since the images do not line up perfectly. For example I selected now IW3 burst 7-12 for one acquisition and burst 3-10 IW3 for the other acquisition. Is snap so “smart” to only show me the area where both scenes overlap? thanks for your quick response by the way!

I’m not sure why undulated areas should be more problematic in the TOPSAR acquisition mode than another.

I don’t have in mind any article about single burst coregistration. But note that ESD is there to correct a problem that arises when stitching bursts all together. This problem is not present when there’s only one burst.

Please take a look at eh threads within this topic,

Interferometric processing of products with one burst overlap

Also have a look at this tutorial,

Step by step InSAR processing

When you read these threads carefully, you’ll find the source of your error is the unmatched image selections, it means, both image should be belonged to the same slice number and orbit number, taking in account the orbit direction.

This is the source of your error.

Basically the arrival time of the signal to the antenna is different according to the objects’ characteristics, putting in mind that there are different passes, and the signal couldn’t be collected in the same pass by different antenna, adding to that the change of incidence angle, all of that and other parameters could lead to higher need to fine corr. in undulating land rather than the flat area. In case it’s one burst or multiple bursts, similar to ERS1/2 and ENVISAT, I found in my results that applying fine corr. .affects positively the results of the InSAR. (personal opinion)

The relative orbit number is the same for both acquisitions, as well as slice number.

If this is your case,

Your study area should be located in the same bursts, not as you mentioned,

But how can they be in the same bursts, if the number of bursts is not the same for both scenes? And in snap, I so far only found the way of visually doing the selection fo bursts in snap in the TOPSAR-Split step. So

I hope it’s possible to see what I mean from the screen shots. I selected burst 7-10 now for both scenes, but if you look at the region, they cover a “completely” different area.