Even more MDS testing!

(This page is linked from the other Durham testing pages Durham_MDS_Testing and Durham_MDS_Testing2 and Durham_MDS_Testing3)

Here we look at stacks from the ps1-md release on November 18th 2009. Unless otherwise noted, we are dealing with the convolved stacks.

Photometry

NB - looks like variable PSF fitting was switched on between 29871 and 30204.

Shown (left) is the SDSS comparison for PSF mags on the ps1-md MD08 skycell 041 i-band stack 30204 (left) and Heather's previous stack (29871) on the same cell. Blue points are SDSS classified stars. Both have nominally the same exposure time (equivalent to 41 240sec exposures) and we assume come from the same set of exposures. However, the seeing is 2.0" on 30204 but only 1.4" on 29871, and the zero points differ by about 0.4 mags (note: the zero point for 29871 has the expected value). Clearly there are problems with the stellar photometry being non-linear. ADDED LATER: the image on the right shows the same comparison but for aperture mags (for SDSS stars only). We used the SDSS Ap7 values, which are rather large at ~15" diameter, hence the high scatter. The comparisons both seem linear.

However - above we show a direct IPP aperture magnitude (AP_MAG) comparison between 30204 and 29871. This is gratifyingly linear, from which we deduce there must be some problem in fitting the PSFs.

* The 0.4 mag offset is actually in the data. Curiously it is not seen in the unconvolved stack of 30204.

3 Dec 2009 - Paul/Gene have identified problems in the stacking code which means that the convolved data from the Novemeber release was incorrectly scaled. This may also have caused problems with the PSF mags.

One potential problem is demonstrated here. The above shows the difference in PSF magnitude as a function of x position on the stack for two reductions of the same stack (30204). One is the provided cmf file, the other is our own run of psphot (code from SVN as of Nov 25th 2009) on 30204. Note the shift in difference with x. It must be said that there is no such shift between 29871 and 30204, so this does not explain the non-linearity seen there.

Astrometry

Show above are the x and y position offsets between two reductions of the same stack (30204). One is the from the standard cmf file provided, the other is from our own run of psphot on 30204 (from code taken from the SVN on Nov 25th). The rms is about 0.4 pixels, so as both methods used identical datasets this is a measure of the reproducability of psphot. The offset is of order 0.1-0.2 pixels. NOTE: although not shown, 29871 v 30204 shows a similar scatter, athough with a different overall offset.

Noise in unconvolved stacks

These figures show aperture magnitudes for a selected list of objects, with a local sky estimate (a mode) in an annulus around each object for the 30204 stack. The diameter of the aperture is 40 pixels and the inner and outer diameters of the annulus at 60 and 80 pixels.

This first figure shows the magnitude against an error estimate based on the variation of the data in the annulus, the sky noise. Each pixel in the aperture is assumed to have at least the noise due to the sky, so these represent a lower limit on the true error in the aperture.

To check IPP stacking we also show the results of stacking the same warps using an independent software package (CCDPACK). The amount of clipping used in CCDPACK was greater, which explains the higher noise.

Repeating the same apertures, but using the variances generated by the IPP analysis gives the following plot. Reassuringly what should be better estimates of the error lie above the sky noise estimate.

Just for fun …

MD08 skycell 041 - left: grz colour image from the Pan-STARRS complete stack (200910-2 release), right: SDSS colour image of the same area of sky. Please note that the PS image is a MUCH longer exposure than SDSS!