I've run column 2, run 200005, fields 351-385 again. This time I've changex the images to be in units of picomaggies. This has the twin benefits of a) being closer to the natural image zeropoints when the exposure time is incorporated into the zeropoints (now -30.0, and flux20 = 10,000), and b) fixing the sky sigma being less then 1 and related setting fluxes of objects less than 1/2 to 0 by the integer photo bug. Now a photo run typically has:
Field 200005-2-381 u sky +- err. PSP: 0.00 +- 0.00 Frames: 373.68 +- 8.91 Field 200005-2-381 g sky +- err. PSP: 0.00 +- 0.00 Frames: 333.75 +- 3.39 Field 200005-2-381 r sky +- err. PSP: 0.00 +- 0.00 Frames: 822.80 +- 5.55 Field 200005-2-381 i sky +- err. PSP: 0.00 +- 0.00 Frames: 2260.13 +- 9.03 Field 200005-2-381 z sky +- err. PSP: 0.00 +- 0.00 Frames: 5003.81 +- 31.89
I ran sextractor on field 200005-2-358 in all 5 filters with the correct
zeropoint to find the following plot, where u is blue, g is green, r is red,
i is orange, and z is black. The histograms in 0.25 mag bins peak at
u,g,r,i,z = 24.75, 25.25, 24.25, 23.75, 23.25. This tells us that the
images are deep and that we in principle understand the zeropoints.
The psf error vs. psf magnitude plot looks better. It goes deeper
before the error starts rising, but one has to wonder about the magnitude
of the error as it never reaches above 15%.
The g-r vs. r psf magnitude plot for 200005, col 2, fields 351-385 (except 376).
All objects are in red, clean stars are in black. This looks understandable.
(clean stars = quasar ts cuts, without the cut involving psferr).
The color-color plots look promising:
The star galaxy separator probably will have to be changed.
I have as the LRG s/g separator: psf<2>-model<2> lt 0.24 at model<2> lt 19.5, else psf<2>-model<2> lt 0.40.
The plot below suggests we could use psf<2>-model<2> lt 0.04 at all model<2>.
This gives 13 galaxies/sq-arcminute. If we stay with 0.24 but extend it to all model<2>,
there are 7 galaxies/sq-arcminute.
The reductions are no deeper- in fact there is another problem: black is coadd, the red a single pass frame. The coadd errors pass 20% brighter then the single pass. This happens suspicously close to 2.5log(5) mags brighter, and 5 is the fpC_scale parameter that scales the image by a factor of 5 before processing.
Just to reassure ourselves, Jeff Kubo and Huan Lin ran sextractor on the 1, 6, and 35 image coadd and found the following number counts: The zeropoints are arbitrary but constant, and we see the expected 0, 1, and 2 mag shift of the peak of the number counts.
The i-band images, at least, look deeper to me: A 35-run version of fpC-200005-i3-0660.fit should about 2 mags deeper, and a 6-run version should be 1 mag deeper, then the one-run coadd. All stretched to the same limits:
35-run coadd
6-run coadd
1-run coadd
Since these are flux scaled, the coadd should cause the noise to recede- it looks like it does to me. Right at the moment I cannot quantify it, but Jeff says he might be able to run sextractor on these images in time for the phone con and we can double check the depth that way.
