QC/Kentools and Image Health Seeing

Steve Kent has expanded his kentools toolkit to calculate seeing, sky level, cloud coverage and effective transparencies. Currently (early November 2013) these quantities are calculated offline by a separate script and the results can be viewed "here:". It is planned to integrate this script with the online system and to generate the information automatically similar to the pointing offsets that are also determined on an per exposures basis using kentools. Until this is done we will occasionally backfill the information manually into the exposure table. This makes the qc information available to our standard data mining/Db query tools and it becomes straightforward to compare it to image health results.

QC Information in the Exposure Table

Six new columns have been added to the exposure table to present the QC information.

Field Type Default Description
qc_done Boolean False True if QC data is available
qc_accept Boolean False Result of QC assessment
qc_fwhm Float null QC seeing (FWHM) in arcsec
qc_sky Float null QC sky brightness
qc_cloud Float null QC cloud coverage
qc_teff Float null QC effective transparency


IH measures the half flux radius r50 of objects it identifies as stars and then uses an empirical scaling relation to convert this to FWHM to match sextractor seeing for the exposure (as is implemented in QR and was done by DM). IH uses only a very simple star/galaxy separation algorithm as processing speed was an important design requirement for IH to be able to operate on every exposure. This leads to known issues in crowded fields or very long exposures where IH mistakes galaxies for stars. For standard DES exposures IH FWHM results match very well with sextractor. IH analyses the entire focal plane and determines the average seeing. QC uses only the innermost CCDs. Studies have shown that the seeing gradient across the focal plane is less than 2% when the instrument is focused (and the AOS takes care of that)

Telemetry DB Access

This simple SQL query can be used to extract both the QC and IH seeing information from the database.

select, exposure.exptime, exposure.filter, exposure.qc_accept, exposure.qc_fwhm, image_health_fp.fhwm[1], image_health_fp.r50[1] from exposure.exposure full join telemetry.image_health_fp on where exposure.qc_accept=true 

Seeing Comparison

Here are some initial plots to show the difference between the two software packages. A more detailed analysis will be required to determine the scale factor(s).

The first plot shows FWHM from QC and IH for all DES exposures up to Nov. 3 that have the QC information available. Short and long exposures are shown separately. The large scatter for long exposures reflects the problems of the IH algorithm with exposures that have a large galaxy count. The blue line marks FWHM = FWHM. The need for a scale factor is obvious.
FWHM from QC and IH for all DES Exposures until Nov. 3

A simple scaling relation can be extracted from this plot. Obviously a more detailed analysis is required including higher order terms as the relationship is not linear. This scaling relation does not address the problem with some of the images with long exposure times.
The plot below shows that IH_FWHM * 0.8 + 0.26 matches QC FWHM reasonably well.
Scaled IH FWHM

The next plot shows the difference in QC and IH seeing before and after scaling the IH FWHM values. The same simple scaling relation is used.
Difference in QC and IH FWHM before and after Scaling

Scaling Directly from R50 (Gary B 5 Nov 2013)

The FWHM values reported by IH are already a scaling of its actual measured quantity, R50. The scaling was derived by Jiangang Hao from some very early SV data. So it makes sense for us to adjust this scaling relation to produce a new FWHM instead of scaling a scaling. Here is the plot of Image Health R50 vs QC FWHM. I split off the long exposures as Klaus did above. The yellow points are binned median values of the short exposures.

The dashed black line is the best linear fit:

FWHM = 1.586 * R50 + 0.077

This yields a standard deviation of 0.013 arcsec for the binned points, with maximum deviations about about twice as large.
These residuals are cut in half from allowing the slope of the line to change at R50=0.6:
FWHM = 1.373*(R50-0.6) + 1.011   (for R50<0.6)
FWHM = 1.661*(R50-0.6) + 1.011   (for R50>=0.6)

I would suggest implementing the 2nd form. The individual measurements of QC FWHM have a standard deviation of 0.023 arcsec from this fit.

Update 10 Nov: Ken Patton has updated the IH code. With data on hand, the new R50 measures appear shifted a bit from the old ones such that this single linear fit is better than the one above:

FWHM = 1.667 * R50 + 0.037

This fit has 0.013 arcsec residuals.