Collocation algorithm irregularities, former Artefacts in Sentinel-1A data?

Microwave Toolbox
#1

Hi everybody,
I’am working with SNAP V5.0 on Sentinel-1 data for some weeks now on a small Area of Interest (AoI).
I’ve download several S1A_IW_SLC products encompassing the AoI and have done the following processing with SNAP:

  1. find the swath of the AoI and split to this Swath
  2. deburst the splitted product
  3. apply the orbit-file
  4. apply the single speckle filter Gamma-Map 3x3
  5. collocate the results of 3. and 4.
  6. build the quotient of Intensities_VH before and after despeckling (looking for mean and sigma in the statistics) with band-maths

The results of the calculated band for an ascending (AoI_S1A_IW_SLC__1SDV_20161225_IW2) and a descending passage (AoI_S1A_IW_SLC__1SDV_20161229_IW3) together with the coordinates of the AoI are shown:

ascending.jpg1126×895 462 KB

(Since I couldn’t upload two pictures as a new user I describe the “descending picture”: in the upper half of the image three smaller ellipsoids (in comparison to the picture above) show up)

Due to the regularity I think the ellipsoids are artefacts. But of what kind?

  1. I’ve found no hint about possible deficiencies of sentinel-1-data
  2. Speckle-Filtering works locally so (imho) can’t produce such big regular feature
  3. Importing the artefacts by the other processing steps with SNAP looks very improbable

Edit:
But in the meantime I found the following

  1. The regular pattern were seen in the whole sub-swath of a product so the problem is independent of the AoI.
  2. (Mis-)Using the stack-Operator of SNAP to collacate the intensities before and after despeckling and building the quotient shows no regular structure.
  3. Using the Scatter-Operator of SNAP to compare the despeckled intensity with the collocated one shows heavy aberration from the expected line (in contrary to the same procedure with the stacked products).
  4. Using bilinear interpolation resampling (instead of nearest neighbor resampling) produces comparable but less contrasted images for the quotient.

Did I miss some preconditions for using the collocation-operator in SNAP?
or
Is there some unexpected behavior of the collocation algorithm?

Since I heavily use the collocation for comparisons I am very interested in an answer.

Thanks
HD.

#2

I guess you have used the normal Collocation operator, but for radar data you should use one of the Coregistration operators you find under Radar / Coregistration in the menu. But I don’t know which one, I’m not a radar expert.
Maybe @lveci can elaborate on this.

#3

Thanks for the reply.
Indeed I’ve got undisturbed results when using corregistration/stacking. But why? And why doesn’t collocate work as expected (e.g. delivering the same results as stacking)?
Furthermore I want to inspect sentinel-2 data too. Will the collocate-operatin work as expected for this data?
Perhaps someone can give me some explanation.

#4

For S2 it should work. Radar data needs always special treatment. Unfortunately I can’t explain it to you, as I’m not a radar guy.