›

Questionable Data Fit with PCC

craig smith
in PixInsight
Just wondering if anyone can help... if I run PCC and the curve fitting plots seems to indicate (by eye at least) that the fit isn't terrific, what kind of conclusions can that be telling me about my raw data?  The input data is calibrated normally, but otherwise unmolested.  Sample plot is attached.

Thanks!
Craig

Comments

  • BlockHeadBlockHead
    This looks like you have very noisy data. The stars are either very faint in your data (you do not have many bright ones)  or your exposures are very short. OSC data in particular is prone to this kind of thing (you didn't specify). Under these conditions, you can try to force the detection of more stars..but I am not convinced that would help if you do not have them to begin with (change the auto magnitude to a set value...perhaps 12th if this is widefield data (again you did not specify)). I am going to guess there isn't much color in your data to begin with? But if there is, you can always fallback on regular Color Calibration and see if the result you get there is significantly different than the PCC result- Although I agree by eye the fit isn't great...but the error in the fit still might not be enough to make a difference given your data.

    -the Blockhead
  • craig smith
    Sorry to not be more specific...  And thanks very much for the reply.

    My rig is a TEC140 reduced with an AP quad TCC so that the scope is operating at about 705 mm at f/5.  Sensor is a monochrome Sony sensor in a QSI690 with 4.54um pixels if I remember correctly, but the resultant resolution is right at 1 arc-sec/pixel.  Exposures for each color filter were 600 seconds.  I collected something on the order of 16-18 frames for each color, but tossed 5-6 of each based on a visual inspection, so there is probably around two hours of actual data for each color channel.  The target is M109 and I was shooting it in Florida as the last target of the night, and transparency may not have been ideal.

    So to clarify the points you mentioned, the data is neither OSC or particularly widefield.  One other factor which probably doesn't matter but I'll mention anyhow is that there is a bright star (Phecda, I think) just outside my field of view dumping a lot of blue glow over about 1/3 of the field.

    When you suggest that there "isn't much color in the data to begin with," I am not sure that I understand how to assess whether that is the problem or exactly what it means.  I am happy to share one of the sub-frames or the resultant integration if that would be helpful in diagnosing what is going on.

    The data isn't amazing, but also didn't strike me as *excessively noisy.  Admittedly, I don't have a lot of experience with PCC, and am trying to learn what those questionable fits are telling me.

    Again, I appreciate the reply!
    Craig
  • craig smith
    Poking around a bit more and listening to your hints, I am guessing that my color sub frames are just underexposed at 10 minutes - at least for this target on those nights.  

    I have to learn to do a better job of looking at image statistics while I can still do something about them.

    Hopefully, I can collect some more data at May new moon and prove my theory.

    If you have other thoughts, Adam, I would love to hear them, but in any event I'll update this thread if I learn something more.

    Thanks,
    Craig

  • BlockHeadBlockHead
    I see... yes, M109 is a particularly star poor field. 
    And 600 second exposures is definitely healthy! Most times when people post problems like this...it is due to very poor data. So, to begin with you can make the RGB channel combine image available..that would be good to look at. 

    I still think comparing with Color Calibration is a good thing to do. PCC is giving you the weights (the color ratios)... if Color Calibration (using M109 as the white source) gives you effectively the same answer- you are safe to assume that that *is* the answer and the fits are not terrible.

    Lets see that unprocessed RGB!

    -the Blockhead
  • craig smith
    Attached is a freshly made RBG combination.  The individual channels were touched up with MURE denoise before combination, but no other processing or balancing or anything.

    Actually, I couldn't seem to upload the xisf file; maybe it is too big? Here is a link to the file on my google drive:


    Thanks again,
    Craig
  • BlockHeadBlockHead
    Craig,

    I am going to guess you have not installed the local GAIA DR3 and both APASS databases?
    We are getting different results. Although the scatter is still there...it appears the fit is more determined and believable. Note the weight coefficients on the right... so that you can compare. 

    When I used color calibration with the galaxy as the white reference...the result was a little more pleasing. I think this is because your image is lacking blue. It appears to me the seeing most have been worse for your blue data- green and red are dominant in your image. Color Calibration "corrects" this... since it is based on a fixed white reference as it is in PCC. (It is determined by the colors in your image.)

    So...bottom line- there is nothing wrong with your data or the process. It might be that you need to get the full catalog or force a better catalog for better results.  (Please see my catalog section in FastTrack training).

    -the Blockhead
    Capture.JPG
    1920 x 1160 - 334K
  • craig smith
    Adam,

    You are correct that I have not installed the local catalogs.  I will go ahead and install them... maybe the SSD space they consume will help remind me to get some of my unprocessed data moved off my computer!

    In comparing the calibration results, I see that although our PCC settings appear to be the same, your results were based on over 350 stars, whereas mine was on only 50 stars.   Is the difference in the number of stars that were used is based just on having the local catalogs available?  The White Balance factors were not wildly different but they are a little different. 


    I guess I am still not clear on the cause of the data not being a better fit.  As you say, the fit of the data looks better and more believable, but it is still relatively scattered.  Is that still potentially due to noisy data or at least a relatively low SNR?  Are there other possible causes?

    Thanks for taking the time to look at my data and offering advice.

    Best regards,
    Craig
  • BlockHeadBlockHead

    I wouldn't worry about the scatter too much. The fit that I show isn't terrible (or unusual). There is some solid clumping there!

    And again, as I first mentioned, even with so few stars..the and was not that bad right? That should really give you some confidence that PCC gives you a reproducible/consistent answer.

    As I mentioned, I do think your S/N in your color data is variable... and likely lacking in blue (perhaps due to poor seeing or focus). This might introduce some scatter... 

    -the Blockhead

  • craig smith
    Adam,

    As a follow up, I installed the catalogs, and not too surprisingly, the results matched those that you shared.

    I want to bring out the differences in color; more blue in the arms yellow in the center, and at least some hints of brownish dust detail.  I don't think I can see those differences in the individual channels with the color data, at least at this point.   I'll keep plugging away and hope to get some additional color data at the May new moon and see if I can get some of that detail.

    Thanks again for your help.  I appreciate all the reference information you have made available.  It is an amazing resource.

    Craig


  • BlockHeadBlockHead
    M109 in particular is not a very contrasty galaxy in color... so look for any subtle color bits to bring out.
    -the Blockhead
Sign In or Register to comment.