Striping after terrain flattening


Further to the discussion on terrain flattening, I found serious problems with striping in some terrain flattened products.
I am running tests with ascending and descending images. Striping occurs in ascending images only (in my set).
In particular this scene:

(but also the neighbouring one)

suffers from striping when running the sequence:

  • Thermal Noise Removal (TNR) -> Calibration -> Terrain Flattening (TF)

using either SRTM 3Sec or SRTM 1Sec HGT (and either BICUBIC_INTERPOLATION or BILINEAR_INTERPOLATION DEM resampling).

Striping disappears when the sequence is adapted to:

  • TNR -> Calibration -> Multi-Look (3x2) -> TF

but only when using SRTM 3Sec. Striping remains when using SRTM 1Sec.

Striping is regularly spaced in azimuth direction and runs across the whole range. It looks like it is somehow related to DEM sampling.

Btw. when comparing imagery that is terrain corrected with SRTM 3Sec or SRTM 1Sec there is a very clear shift between products towards the north-east (3Sec vs. 1 Sec). Apparently this is due to differences in the definition of the origin of the UL pixel coordinates (center or top left) for the respective SRTM tiles. This does not seem to be a trivial problem.


Guido, thermal noise has already been removed from all detected (GRD) products, therefore you should not remove it again.


Ok. Thanks for that tip. I did not know that. So, TNR is really for SLC processing then?

Anyway, without TNR, striping is still there, though that’s not a surprise.


(Snap 2.0 Beta 5)

For the record: Snap 2.0 Beta 6: problem with striping persists.

However, the values are suddenly several 100s of times higher in the output of the same script (2.6 KB) when comparing Snap 2.0 Beta 6 with Snap 2.0 Beta 5.

This is not expected.

1 Like

I am afraid I cannot help with the striping problem, but are you sure thermal noise is removed from all S1 GRD products?

In glemoine’s image and in all the other images I have looked at including images acquired today the parameter field “thermalNoiseCorrectionPerformed” in the annotation files is set to false. According to the documentation (Sentinel-1 Product Specification) that means that “thermal noise correction has not been performed”. I am assuming that removal and correction are synonyms here

As far as I know, thermal noise correction is neither applied to SLC nor GRD products during their generation.
I checked a few hundred recent GRDH products, and all had “False” under thermalNoiseCorrectionPerformed in their respective annotation XML files.
The noise vectors are generated and annotated by the IPF, but application of those vectors is generally done in post-processing (e.g. optionally by SNAP).