sarproz-group

APS processing parameters

This topic contains 1 reply, has 2 voices, and was last updated by  periz 1 year, 8 months ago.

  • Author
    Posts
  • #1350

    jsal
    Participant

    Dear Daniele,
    I’m looking at almost one year of TSX spotlight data on cooling lava which sits on the summit of a volcano. I have a relatively good idea of the displacements already: the cumulative displacements over the year are between 40 and -140 cm, the velocities however decrease over the year, only a small proportion happens in the last 4 months. I have a Lidar DEM for the processing, but because of summit morphology changes it is not that accurate there. I calculated the Amplitude On/Off model because I think the PS may get lost inbetween due to hurricane or volcanic activity. Even though the area is small, there is very strong atmospheric, mostly stratified signal.

    I did the APS processing with the parameters as in the attached .png
    This worked fine, and when I plotted the estimated parameters it looked ok and nicely within the boundaries I set. However, when I then estimate the APS it gives an error:
    >>Encountered nonpositive pivot.
    >>trying with cholinc
    >>inverti_dif: error with cholinc
    And 4 of the APS estimates are empty when plotting (not just the master).
    When I reprocess the APS using “None” for the weights it works fine, the estimated parameters and APS Estimates are similar to the ones estimated using the Amplitudes and I can finish using the IR+Stratified for the APS estimate.

    2 questions: Considering the decreasing velocity of my displacements, are these the correct parameters to use? I wasn’t sure if I should change the order of the polynomial or do something else. Secondly I wonder what the problem is when I use the Amplitudes (although it isn’t an issue here, since the volcano was quiet and most of the PS stayed anyway).

    Thank you!

    Attachments:
    You must be logged in to view attached files.
  • #1352

    periz
    Keymaster

    Hi Jackie,

    it’s a bit a complex matter, there are many things to discuss.

    first of all, in the APS it makes no sense to give asymmetric ranges in the parameters estimation. You are processing connections and you have no idea whether you are going to process A-B or B-A.
    secondly, again because you are processing connections, you may want to reduce the ranges (you should be processing close points).
    but this depends on the kind of PSC graph you are using (this is an important variable)
    then, the On-Off model uses a rectangular function to estimate temporary targets. so, the behavior is off-on-off. I do not think this matches well the case of hurricanes in between.
    in any case, if you use such an option (an amplitude model to decide the life period of each scatterer) before running the analysis I would suggest you to check well the results to make sure you are doing the right thing. in the scatter plot module you can display all results and also plot time series together with the model. I may need to show you how to do it (I am not sure if this is reported in some tutorials).
    then, again I doubt this is a good choice to be used in the APS estimation. Usually you want to have your best permanent targets in the APS estimation to carry out a robust inversion and get a good atmospheric delay. Only in the next step you may want to include many more points (also unstable and un-permanent ones) and thus use some tricks as amplitude models to process them.

    the error you are getting is about the failure of the spatial graph inversion. that operations is the main step of the APS estimation. probably, you are trying to invert the graph in some images where you have no “alive target”. but this is something you have to investigate yourself (e.g. through the scatter plots module)

    then, if you say that you have a non-linear movement, I suggest you to try the “smart” function. this is a kind of temporal filter which is not using any displacement model. You can activate it by typing 5 in the corresponding box. that number is the length of the filter (number of image samples).

    I assume you are using the star images graph (single master). if this is not the case, then the estimation is even a bit more complex.

    let me know if my words help you in understanding the case a bit better

    cheers

You must be logged in to reply to this topic.