Problem: Coregistration of mosaicked S1 scenes for forest change detection

I’ve given more than less detail here as maybe the problem is caused by my overall workflow. Any help would be gratefully received.

Brief project aim: To use S1 radar to detect burned areas in tropical forests.

Current workflow: following method of Verhegghen 2016 (doi:10.3390/rs8120986) where average backscatter from 3 x pre-fire scenes are differenced with the average backscatter from 3 x post-fire scenes. A threshold is then applied to the change product. The first and second aggregated scenes cover the entire Area of Interest (AOI), however the third aggregated scene does not, and requires mosaicking 2 separate S1 scenes.

SNAP workflow for individual S1 scenes which cover entire AOI for one time step.

  1. Apply orbit file
  2. Subset
  3. Speckle filter
  4. Terrain correction
  5. Calibration
  6. Coregister three scenes
  7. Stack average (of 3 scenes)

SNAP workflow for S1 scenes which partially cover AOI, therefore 2 need to be mosaicked to cover the whole AOI for each time step:

  1. Apply orbit file (x2)
  2. Remove GRD-Border Noise (x2)
  3. Slice Assembly
  4. Subset
  5. Speckle-filter
  6. terrain correction
  7. Calibration
    8. Coregistration of 2 or 3 scenes*
  8. Stack averaging

*Problem : The mosaicked scenes do not coregister well- in the resultant stack there is a mosaic for only one time-step, or if they do stack, one or more of the mosaics is mis-aligned.

Quick comment: calibration should be done very early, for example after the orbits step.

OK great, that’s a good start! Thanks.

if you coregister after terrain correction you should also select Product Geolocation as the initial offset method:
At least I had better results and less miss-alignments

I tried both of those suggestions and it seems to have worked.

Thanks kindly!