I have Sentinel-1 IW GRD images; for each capture date, one geotif with the VV and one with the VH values. I assume these are intensity values for the co-polarization and cross-polarization bands. To perform image segmentation (estimating parcel boundaries), I was advised to use the SAR intensity values. Should I use VV, VH, or a combination of these (maybe squared sum: VV^2 + VH^2)?

I have tried using both and their squared sum, but the resulting image segments do not correlate well with what I see on the S2 rgb-images.



The two intensities are indeed related to the polarization (V for vertical, H for horizontal; first letter for emission, second letter fo reception)I recommand you to use both and their ratio (not square sum), because it is the standard way to make rgb with Sentinel-1 data, which accounts for the polarization. This will however remain a challenging task because of the speckle, so if you have the opportunity to use Sentinel-2 images in you area this will be more efficient (of course, clouds will be a problem, then)

  • It is also worth to mention that Sentinel-1 potentially contains thermal noise. I recommend to import the data into SNAP and performa radiometric calibration followed by range doppler terrain correction in order to get radiometrically and geometrically correct data. You might also consider converting the data to the logarithmic scale [dB] and check if the parcel boundaries are clearer visible then. – AndyB Jun 2 '20 at 17:02
  • then it is better to run all the preprocessing from the SLC images, but I assumed that the GRD images are well processed. – radouxju Jun 2 '20 at 17:06
  • Indeed, GRD images are multilook and supposed to be well-processed already (projected to the ground and terrain-corrected). I knew about the VV/VH ratio, and tried that too, but it puzzles me. I think it would be better to have something like VV/(VV+VH); ratio variable tend to show a nicer distribution when defined this way. – Petró Bálint Jun 2 '20 at 21:12
  • GRD noise can easily be removed, it makes little sense to use SLC products because of this. I just wanted to point out that tiffs of Sentinel1 products are not ready to use for many applications (potential reasons m why you are unsatisfied with the first results) – AndyB Jun 3 '20 at 4:23

Normally, to reproduce the RGB image, you'd use different values for the different RGB bands: Red: VV
Green: VH
Blue: VH - VV

There's a video worth watching on the subject by the ESA:

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.