This page is linked from the PS1_Science_Processing_Status_20090705 page and the KPAT:_Commissioning_Data_Test page.

Note there are three other Durham MDS pages, Durham_MDS_Testing2, Durham_MDS_Testing3 and Durham_MDS_Testing4.


Testing of MDS image quality - Durham (Jim Geach)

I have been looking at fields MD07 and MD08, checking out image quality, and trying to highlight potential issues that will affect extragalactic science.

General image quality

There are some significant flat-fielding issues, especially near the edges of the field. Zooming in, there are some strange re-sampling effects in the stack - these are going to seriously mess up faint-source science, for example in this zoom-in of a g-band stack from MD07 (cell 056):

I’m quite worried about the ‘chunky’ effect seen in (e.g.) the abrupt level changes, and the mottling effect. I’m guessing that all of these are a combination of convolution, dealing with NaN-ed regions in the input frames, and how the sky is dealt with. I’m trying to use the deep CFHTLS Groth Strip and EN1 (INT) images to figure out how much of that mottling is noise, and how much is contributed by faint galaxies - I hope to come up with something quantitative. We should be worried about that that.

UPDATE: Here's something we've just noticed in some of the warps - trailing of stars. Here is an example from g-band MD08 (cell 077):

More worrying: here is an example of a comparison of g and r-band stacks of skycell 030. Something seems to have gone quite badly wrong - especially with g. I suspect this is a masking issue - looks like something from a bright star has infected the image?

Finally, just looking at a few of the z-band warped images, there might still be an issue with correlated readnoise:

Not yet clear to me how much of an issue this is with the other filters, until we look at all of the data. Will do that in the next few days.

Object detection

The .cmf catalogue seems only to contain fairly bright (and mainly point-like sources). I ran Sextractor on the frame with a very simple 3-sigma / 5-pixel threshold (quite low, I admit - but I wanted to push it to see what came up). I got the following (red circles are Sextracted and green boxes are from the supplied cmf catalogue). Note - should the cmf catalogue have extended / faint sources in it? There might be some de-blending going on, because some of the big bright things have been split into two.

Depth/photometry and comparison to existing deep optical images

Here is an example of a side-by-side view of an r-band image of MD07, compared with the CFHTLS (Megacam) image. I've overlaid the positions and photometry from the CFHTLS so you can compare depths:

... and the same for MD08, compared to Subaru Suprimecam (i-band):

Of course, we don't expect the current MDS data to be as deep as these comparison frames yet (in fact, it looks like we're hitting r~24 at 5-sigma)... but the image quality should be something to aim for in order to be competitive in extragalactic studies.

You can see that (roughly) a 5-sigma detection is at about r=24 mag. Here is a plot showing the MDS instrumental mag (I’ve bootstrapped the z.p. from the CFHTLS deep catalogue) compared to the Megacam magnitude (note that the CFHTLS catalogue should be going down to r=28 mag). You can see where the scatter blows up at about r = 24 mags (I’ve shown two versions of the plot). This was only rough, I need to do something a bit more rigourous to work out the true depth. Also, this is for just one cell:

More to add...

Stacking tests

I had a go at re-stacking skycell 077 (g-band) for MD08. I had to chuck away frames with bad readout problems, and then applied a simple background subtraction (sans objects). Below is the comparison to the PS stack -- it seems cosmetically better, but I'm not sure about the depth yet. Note this was the stack of the warped frames - I applied a median combination with 3-sigma clipping. I also smoothed with a gaussian, not the kernels that went into the PS stack.

Here is a set of images that show what regions have to be masked in the 19 input frames... in fact, I didn't use any of the frames with the blocky readout problems, as this tends to affect neighbouring chips, and I think is the cause of the background variations in the PS stack (I can re-produce something similar if I leave them in). On the right is the 'total' mask, where the colour corresponds to how many frames have bad pixels at a given x,y location.

Comparison with SDSS - (Nigel Metcalfe)

Here is a comparison of the PSFmags from the cmf files of four adjacent (arbitrarily chosen) MD07 stacks with SDSS model mags (r,g,i and z, as labelled). Blue indicates an object flagged as a star by SDSS. I haven't rejected any images flagged as defective in any way, so this probably explains the blue points offset by 2 mags. UPDATE: looks like these offset points are stars with masked centres - note that there are brighter stars which are not masked. Is this the 'removing intermediate brightness stars as cosmic rays' problem again? It appears particularly bad on r cells 40 and 52.

Clearly the zero-point varies from cell to cell (do we expected this? after all, no calibration has been done as far as I can tell - YES the zero point depends on the number of warps which went into the stack). The r-band stellar locus also appears tighter in cell 41 than in the other r-band cells. As the are no extended source magnitudes in the cmf files, all the galaxies (black open circles) are offset.

Update: I have now done the SDSS comparisons for all the MD08 stacked skycells. Here are the histograms of the instrumental zero-point distributions, corrected for the number of warps which went into each skycell. In z, one outlier has been clipped before determining the mean and rms.

g 33.41 +/- 0.08*33.63 +/- 0.05
r 34.00 +/- 0.0634.16 +/- 0.04
i 34.55 +/- 0.0733.93 +/- 0.06
z 33.27 +/- 0.0433.59 +/- 0.06**

* cells 044 and 093 have been excludeded with zp > 33.9
** cell 074 with a zp of 35.2 has been excluded

The file attachment:md08_stack_zeropoints lists the zero points for each MD08 skycell in g,r,i and z, both before and after correction for the number of warps. It also tabulates the number of warps for each skycell. The values for MD07 can be found in attachment:md07_stack_zeropoints .

zeropoints in stacked frames

Using the data above, here are the zeropoints per second from the MD08 stacks, presuming 19, 25, 20 & 24 warps went into the g,r,i & z stacks for MD08 and 16, 21, 36, 17 for MD07 (these numbers came from the dbinfo files for each stack).


Apart from the i-band, these are all slightly brighter (0.05-0.1) than Eddie Schlafly's values. There will be correction (of the right sign) for light missing outside the IPP PSF though (Eugene says this should affect both sets of measurements).


Detection threshold differences

Here is a image of part of the above r-band mosaic of four stacks, with sextractor detections in green. Interestingly, the skycells have gaps between them (I thought they would overlap slightly?), and one of the cells has many more detections. The reason for the latter is not clear. The pixel-pixel noise and background levels are indistinguishable from the other three - the only difference appears to be in the structure of the background (Jim's mottling effect). Note also the dark halos around bright objects where the background subtraction has been too aggressive.

Galaxy cores masked

Many of the brighter galaxies have had their centres masked out as if they were stars. This seems to affect stuff down to r~17, but surely these are not saturated? Below are some examples from the r-band:

UPDATE: please ignore the magnitudes listed on the plots - there are from the IPP but are, of course, trashed by the masking. In terms of SDSS r mags, masked galaxies go as faint as about 16.5 .

Stacking tests

To test the problems with the stack, Peter Draper and I have repeated the stacking of MD08 r-band skycell 041 using STARLINK software. This stack has 25 input warps - unfortunately for some unknown reason two were missing from Durham's data bundle, so we have only been able to stack 23. Shown below are crops from: TOP ROW - left: the original IPP stack (seeing 2.0" FWHM); right: a STARLINK median clipped stack where the warps have been smoothed with a Gaussian to 2.0" seeing before stacking; BOTTOM ROW - left: a STARLINK median clipped stack where image has only been smoothed after stacking to 2.0" with a Gaussian; right: the STARLINK median stack with no smoothing at all (seeing about 1.35"). All images are shown with the same intensity scale.

Note that we have NOT used the masks or the weights when combining these images using STARLINK.

Area loss

Here is a histogram of the number of warps contributing to each pixel in MD08 r-band skycell 041 (there is no weighting in this and the mask files have not been used - if a pixel in the warp image is 'blank' then it does not contribute, otherwise it does). If coverage were perfect this would be a delta function at 23. In fact the median is only 13-14. At the moment I have only done this cell, so I have no idea if this is typical or not.

This page is getting very long, so also see Durham_MDS_Testing2, Durham_MDS_Testing3 and Durham_MDS_Testing4 for further remarks.