Polarimetry analysis vs. Terrain Correction

I have several quad-pol SLC SAR images which I would like to analyse by different decompositions/classification processes dedicated to quad-pol. For this, I am using (or trying to use…) PolSARPro but also SNAP.

So far, my workflow was strictly following polarimetric tutorials of PolSARPro. So, for example, for H-alpha Wishart classification in PolSARPro it was:

Import SAR (generation of LUT - Scalling Lookup Table) -> Extract SAR data (includes multilooking and T3 matrix generation) -> Georeference T3 matrix -> H-A-alpha Decomposition -> H-alpha Segmentation

Question one: does Calibration take place at any stage of this workflow (somewhere hidden under e.g. Importing SAR data)?

I also tested SNAP for polarimetric analysis. In the meantime, I have found excellent tutorial Polarimetry with Radarsat-2. Based on this, my workflow in SNAP was as follows:

Calibration -> T3 matrix generation -> Polarimetric Speckle Filtering -> H A Wishart classification -> Terrain Correction

In the Polarimetry with Radarsat-2 tutorial is written that the Terrain Correction should take a place at the end of processing chain. However, if that’s the case, the results of H-alpha Wishart classification are no longer 0/1/2/3/4/5/6/7/8/9 classes but values from range 0-9. Some pixels migrate from one class to another (comparing results of the classification before and after the Terrain Correction).
Question two: is there a way to geocode results of the H-alpha classification while preserving the classes?
Question three: the Range-Doppler Terrain Correction also corrects the pixels values based on the local incidence angles. Therefore, could you please explain why this operation should not be done before the classification (=PolSARPro tutorial) but at the end of the processing chain (=SNAP tutorial)?

Select nearest neighbor resampling of the output raster im the Range Doppler Terrain Correction to preserve the pixels as integers. Bilinear calculated new values by averaging several pixels.

I don’t know if PolSARpro applies any complex calibration, but it does not seem so.

It’s always good to work with original values as long as possible, among other reasons because of the mentioned resampling during the geocoding. If you select nearest neighbor resampling, you could technically apply it earlier but that’s not very common for SAR processing.


Of course, I forgot about different resampling methods…

@ABraun, again, thanks a lot for your help!