sarproz-group

APS Processing parameters

This topic contains 7 replies, has 2 voices, and was last updated by  periz 5 months, 4 weeks ago.

  • Author
    Posts
  • #2233

    moathmk
    Participant

    Hello!

    I have been processing some RADARSAT-2 data recently in SARPROZ and I would like to know how to choose the best parameters in APS processing. I attached the parameters I used and the results I got. I would also appreciate it if someone could explain to me how I can assess if the parameters I chose were good or not.

    This is oilfield data processing. The problem with these results is that I expected around 0 mm/yr velocity everywhere apart from the two subsidence and 1 uplift areas. However, the results show uplift velocities on the left and some small subsidence velocities on the right. I suspect that these are caused by the parameters I chose but I have no idea how to correct it.

    Thank you very much!

    Attachments:
    You must be logged in to view attached files.
  • #2238

    moathmk
    Participant

    more screenshots..

    Attachments:
    You must be logged in to view attached files.
  • #2246

    periz
    Keymaster

    So, you have ramps both on the estimated height and on the estimated velocity
    this is why u see non-zeros velocity.
    What I would do is the following:
    firstly, check interferograms and see if they are well flattened or if they have ramps themselves. If interferograms have residual orbital ramps, then u can remove them with this function: http://sarproz.com/manual/refine_orbits.html
    If interferograms are ok, then it means that you are integrating errors spatially. You can see if increasing the NL weights helps for this. You have used values 0.7 0.85 1. However, the histogram of the connections coherence shows higher numbers. Try with 0.85 0.9 0.95. You can play a bit with these numbers and then re-process the APS (and re-run the final test) to see the effects (u can keep the estimated connections, no need to re-process them to test the non-linear weights).
    A final note: the histogram of the estimated height shows values in the range -20 20m: no need to search from -100 100.
    Just now I noticed a last thing: you are using a delaunay images graph. But you did not check the coherence as a weight for processing connections. This is probably the reason for the errors. I suggest you to firstly switch back to star graph (single master). If you are afraid of non-linear movements, try smart=5 instead of linear velocity.
    If then you want to try the delaunay images graph, check the coherence as weight (and keep the linear velocity).

  • #2282

    moathmk
    Participant

    Hi Periz

    Thank you very much for your response.I checked the interferograms and they had ramps. I obtained better results (please find attachments) when I used the function you mentioned the the parameters you specified.

    Is there a way to know how good these results are?

    Many thanks
    Regards,
    Moath

    Attachments:
    You must be logged in to view attached files.
  • #2287

    moathmk
    Participant

    I also have two more questions:

    1- Although the results obtained after residual orbit removal were better, there are still interferograms that are affected by ramps (see attachment). Could this still have influence on my results?

    2- The other issue is with the GCP selection. I identified a lamp post as a GCP and when opening mean KML, everything looks correctly placed where it should be (roughly) as shown in the second image (follow the roads on the top). The problem is that when I zoom in, I cannot identify structures and small roads as easy as in the (plot mean) image. Therefore, the reflectivity map could be off by a few metres. So, is there a way to improve my GCP selection/checks? and is it normal that the reflectivity map will not 100% match all features because only one GCP point was chosen?

    Thanks a lot
    Moath

    Attachments:
    You must be logged in to view attached files.
  • #2296

    periz
    Keymaster

    I don’t see easily a reason for that residual ramp. This would deserve some deeper investigation. The results anyway are now much better.
    About the geocoded reflectivity map: are you keeping a 1×1 downsampling? Otherwise, u’ll degrade the resolution. The mean kml generated from the GCP module is downsampled 8×8. That map is generated just to give you an idea of the relative orientation between SAR and geo coordinates.

  • #2298

    moathmk
    Participant

    Hi Periz,

    I viewed the reflectivity map with 1×1 downsampling from (Extended geocoding). There appears to be a slight shift (several metres) between the reflectivity map and the actual structures from Google Earth. Do you think this can/should be improved by choosing better GCP?

    Thank you very much!
    Moath

  • #2301

    periz
    Keymaster

    that’s exactly the purpose of using a GCP. Anyway in Sarproz you can correct such shifts also later with the geographic corrections module

You must be logged in to reply to this topic.