Yet more MDS testing!

(This page is linked from Durham_MDS_Testing, Durham_MDS_Testing2 and Durham_MDS_Testing4 )

These are outstanding issues from the improved MD08 stacks provided by Paul and Heather in October 2009 (md200910-2). You can click on the images to see large versions:

Background problems due to removing the correlated read noise

The above is a portion of MD08.skycell.041.stk.29871, Heather's stack of the 41 best i-band images using the latest version of IPP. The vertical dark streaking around all the bright images is due to the background polynomial fits used in removing the correlated read noise. The displayed image is about 12 arcmin from top to bottom.

Clipping of stars at the stacking stage

Above is one of Paul's nightly stacks of 8 i-band images, MD08.skycell.041.stk.25532 . This is the best seeing (~1.0" FWHM) of any of them. Masked pixels are shown in red. Note how most of the centres of the stars are masked. This masking happens at the stacking stage, and is not in the warps. The scale here is about 6 arcmin from top to bottom.

Another of Paul's nightly stacks of 8 i-band images, MD08.skycell.041.stk.26482 . The seeing is worse here (~1.4" FWHM) but the clipping is also worse. There are virtually no unmasked stars on this image, and several galaxies are also masked.

Photometry on these new stacks

Here we show the comparison with SDSS for four stacks of MD08 skycell 041 i-band. The original 200907 release; Paul's nightly stack 25532 (8 frames); Heather's new stack of 41 frames; Paul's version of 25532 fixed to remove the problem of clipped stars. All except the original have the convolution correlated noise problems fixed. The x-axis is SDSS magnitude - blue crosses are SDSS classified stars, red circles SDSS classified galaxies. We have used the Panstarrs PSF mags for this comparison, so don't be surprised that the galaxies don't agree.

Original 200907 stack (20 frames. 1.3" FWHM)	Nightly stack 25532 (8 frames, 1.0" FWHM)

Heather's stack 29871 (41 frames, 1.4" FWHM)	Nightly stack 25532 after Paul's fix for clipped stars

Points to note: the stars on Heather's stack show a scale error wrt SDSS. Pauls nightly stack before the clipping fix was problematic. After the fix the photometry is much improved, but still shows more scatter than the original data.

Warp-warp comparisons: We examine the input warps to Paul's nightly stack 25532. We use 14462(1), 14464(3), 14465(2), 14466(6), 14467(7), 14468(5), 14469(8), 14471(4) - the numbers in brackets indicate those used in the table below. Unfortunately these are not the identical warps used by Paul, although they are from the same original exposures. Paul has a later reprocessed set which have not been magiced, so we cannot access them. In the table below we compare IPP aperture magnitudes (AP_MAG) between pairs of warps. The RMS is probably the more interesting figure, as the Mean just tells you whether it was photometric or not (or whether the seeing varied). The comparisons are restrictred to AP_MAG magnitudes between -14 and -10 and the differences are clipped at +/-0.20 to remove outliers.

Pair	Mean	RMS
1v2	-0.047	0.030
1v3	-0.028	0.030
1v4	-0.028	0.032
1v5	-0.031	0.029
1v6	-0.019	0.029
1v7	-0.015	0.028
1v8	-0.022	0.031

The RMS of 0.03 implies that the scatter internal to each warp should be about 0.02 (sqrt(2) of the warp-warp scatter).

We then did the same comparison between each warp and the 25532 stack. These gave a consistent RMS of +/-0.040 mags, i.e. there is a larger scatter between stack and warp than there is between warp and warp!

Clipping: Below is an example of a remaining problem with the clipping. This is the weight map round a star on Paul's fixed 25532 8-exposure stack. Successive gradations in the levels indicate how many warps are contributing to a particular pixel (with at most 8 warps it is easy to see the difference in levels) - brighter means fewer warps. Notice how, as one approaches the star radially, successively fewer and fewer warps appears to contribute (note: in the very centre the star is no longer sky noise dominated so the quantisation is no longer seen - I have set this area to be white).

More investigations can be found on Durham_MDS_Testing4.