3 Image quality summary as of 12072012

The guiding/tracking problem causes troubles for us to get good quality images. Among the many nights, we have 11/23/2012, 11/24/2012 and 12/02/2012 get relatively less trouble from the guiding and tracking and therefore we will show the summary of the image quality based on these nights. I am here summarize the results corresponding to

  • R-18: The PSF FWHM in each of the r, i, and z bands should, over all exposures and over the survey area, have a median of less than 0.9”. Moreover, for 95% of the survey area, there should be at least one exposure in each of these bands for which the mean PSF FWHM is 0.9” or smaller (DES-doc-20-v32).
  • R-19: The mean PSF whisker length for stars per exposures must be below 0.2" in the r, i and z bands

Though the requirements look straightforward, the actual measurement of the fwhm and whisker has some caveats. Let me first put some summary results and then go into the tricky parts. Note that the red vertical lines are the specs, but for the R50 case, red lines are half of the fwhm spec, 0.45.

For the fwhm, I plot the distribution of the median from each exposure. The spec requires the median of all exposures, which should lie between the minimum and maximum of the medians from each exposure.

For the R50, the red lines are 0.45, which is half of the fwhm spec, by assuming Gaussian profile. But clearly, this is not an good assumption.


  • FWHM: There could be different ways to measure the fwhm, namely, weighted moments, adaptive moments, Sech2 profile fitting, Moffat profile fitting, Gaussian Profile fitting, and the results from Sextractor. A clear result is that these different methods do not give consistent results. Gary then suggest that we should compare the R50 which is the radius that enclose half of the total flux instead of fwhm. This turns out work very well in bring consistent results from different methods. So, we should translate our spec from fwhm to R50 in the next phase. I show two plots below to show the situation:

  • Whisker length: Similarly, there are different ways to measure the moments and therefore the whisker length. The difference between my weighted moments and that from the sextractor has two aspects:
    • we using different weight kernel size. I am using constant 2 pix kernel size while sextractor using something like the R50. ** After the measurement of the moments, I subtract the contribution from the weight kernel while sextractor does not. That make the sextractor results consistently smaller than my weighted moments.

Whisker diagnostics of the images:

The whisker pattern reflects how the optics system aligned/focused. By comparing the measured whisker pattern, we can reverse to get the deviation of the optics system from the perfect. At the current stage, this method still gives large uncertainties on the reversed hexapod parameters, which is mainly a result of the actual measurement uncertainty. For details of this work, you can see this page:

Nevertheless, as we are improving the above mentioned optics diagnosis based on the image based PSF distribution, the current whisker plots from each exposures do serve as clues for us to understand how close the system is to the perfect optics alignment/focus. I attached two whisker plots as well as the expansion in terms of the zernike polynomials. The whisker pattern from the optics model with know deviations can be found from here:

  • exposure 158309

  • exposure 158319

More such kind of diagnostics for the exposures taken on 11/23, 11/24 and 12/06 can be found from the attached PDF files:

Remaining Issues:

In addition to the previous Caveats, the major issue we have now is to get the measurements from Image health, Quick reduce and my measurement on the same page. Quick reduce essentially use the sextractor moments, which is consistent with the results from my run of sextractor. But the Image health uses different star finding mechanism and measured the harmonic mean instead of median of the fwhm and the ellipticity rather than whisker length. The following page shows the differences of the results when using different methods:

We now have a framework by choosing the same set of stars and get the quantities measured from all methods, and then find out the way to get consistent and stable measurement for the Image health. Details can be found from