There is good documentation on how to do the basic reductions of data obtained with the Mosaic CCD camera; however, little has been written about how to preserve the photometric integrity of the data, or what sort of photometric accuracy can be achieved. As part of a modern study of the h and Per clusters, we investigated the issue of Mosaic photometry this summer.
First, the good news: standard 1-2% CCD absolute photometry is readily achieved by performing aperture photometry on the "stacked" Mosaic images, despite modest variations in the color-terms and point-spread-functions (PSFs) from one CCD to the next. PSF-fitting, such as DAOPHOT, fared considerably less well on our data obtained at the 0.9-m: the results of PSF-fitting on the stacked images were unacceptable to us (10% errors). If PSF-fitting is absolutely needed, we demonstrated that it is possible to do it on the individual exposures of the individual CCDs (a considerable task!); even so, PSF-variations within a single chip limit the precision to 3-5%. Details can be found on our Web page: http://www.physics.nau.edu/~pmassey/Mosaicphot.html . Here we summarize our tests and what we learned.
The Mosaic camera consists of eight 2048 × 4096 SITe CCDs arrayed in two rows of four chips each so that the full field is equivalent to an 8192 × 8192 image. Typically, program objects are observed in a dithered pattern of five exposures per filter to fill in the gaps between the CCDs, and to compensate for bad columns. Usually, standard stars are observed in a single exposure per filter; i.e., in an undithered manner. Extensive IRAF routines have been developed by Frank Valdes in collaboration with other staff members, to produce a single, "stacked" image for each sequence of dithered exposures.
While this technique is extremely useful for faint detections of objects near the sky limit, it was not clear to us that the photometric integrity of the data would necessarily be preserved. For instance, in this composite, stacked image, the light from a star may be the average of anywhere from three to five of the dithered exposures, and may have been on each of as many as four different CCDs. Concerns that come to mind include the following. (a) Each CCD has its own unique spectral response, and it is clear from the Mosaic manual that significant variations exist between the various chips. How does this translate to differences in the color terms between the various chips? (b) The PSFs are bound to be somewhat different on each exposure: what damage is done when one combines the images using any sort of bad-pixel rejection algorithm. (Does the center of a star look like a cosmic ray?) (c) Do the aperture corrections vary from the center of the field to the edge? These considerations will affect both aperture photometry and PSF-fitting. In addition, success in PSF-fitting on the stacked images would require testing whether or not a single PSF would suffice from CCD to CCD. Thus, when we tackled this problem, it was unclear if we would be able to get away with doing photometry on three frames (U, B, and V, say), or on 120 frames (3 colors × 8 chips × 5 ditherings).
Our data set consisted of short, medium, and long exposures on our program field. We decided to do the photometry both the hard way and easy way -- and so we did aperture and PSF fitting on the 120+120+8 = 248 CCD frames plus the nine "stacked" images, and inter-compared the agreement between the different exposure levels and methods. In addition, we observed Landolt standard star fields, and we spent a tedious couple of hours moving a group of standards of different colors from one chip to the next chip to the next chip, in all three colors.
For the "basic" reductions, we followed Frank Valdes' excellent "Guide to the NOAO Mosaic Data Handling Software,"available within IRAF by typing "help mscguide." We differed from the standard procedures in the following fashion: (a) We used bright twilight flats at the 0.9-m and found they worked excellently in flattening our data (< 1%) -- the differences between the twilight flats and the dome flats were quite severe (> 10%). Dark-sky flats constructed as advised in the document from one's program frames might work even better, but with short exposures of crowded fields, this was not practical to try. (b) Because the plate-scale changes as a function of position in the field, frames that have been flat-fielded in the normal manner must be fixed in order for photometry to be location independent. The scale change results in a 2% (area) change on frames obtained at the 0.9-m, and by 8% (area) at the 4-m. For dithered images, one normally uses "mscimage" to reconstruct images at a uniform scale in preparation for stacking. For undithered images, one could instead simply multiply by the relevant area changes, and indeed the IRAF task "mscpixarea" is designed to do just this. We discovered a normalization difference between the two tasks that led to zero-point errors between our dithered (program) and undithered (standard star) data. This has now been fixed, but when in doubt, it is far better to treat the standard star data and the program data identically, including the choice of interpolation scheme.
We found excellent fits to our standard star solutions (residuals 1-2%), and investigated in detail the variations of the color-terms from chip to chip. Indeed, the average color terms are very similar to other NOAO CCD systems. Our results translate to the following: over a limited color range (0.5 in B-V or U-B) using the stacked images is safe at the 1% level, at B and V, and 2% at U. The im8 chip is the worst (particularly at U). The actual color terms can be found on the Web page referred to above.
What about chip-to-chip differences in the aperture corrections? Although the PSF variations at the 0.9-m are relatively modest, we can see distinct differences in the shape of star images as we approach the corners of the field, and the FWHM changes on our images by about 30% at the extreme. We measured our standard stars through an aperture of radius 10 pixels, and set the zero-point of our photometry on the program fields using a 5 pixel radius aperture. We thus needed the aperture corrections from 5 pixels to 10, and we found that that was quite stable chip to chip, with a sigma of 0.017 mag.
We next found, to our comfort, that aperture photometry of the stacked images agrees perfectly well (better than 1%) with the average of the five dithered exposures, as long as one is careful that the effective airmass used for the stacked image is correct. By default, it is not.
Point-spread-function fitting fared less well. Inter-comparison shows that PSF-fitting on the stacked images introduces 1 sigma errors of 0.05 mags, for bright isolated stars. Furthermore, the errors are not dominated by photon noise, but rather are dominated by location-dependent variations in the PSF. Residuals show marked differences from chip-to-chip. Connected with these are systematic errors at the 0.10 mag level. We rejected PSF-fitting on the stacked images as a viable means of photometry for our purposes.
PSF fitting on the individual Mosaic CCDs was considerably better, with 1 errors of 0.03 introduced for bright, isolated stars, but with no signs of any systematic problems. We believe that PSF variations within the large area covered by even a single chip are responsible. We expect that observers at the 4-m would fare better, as the image quality is more uniform there.
Here are our conclusions:
1) Aperture photometry on the stacked images works very well, with absolute photometry of 2% readily achieved, and with errors in the colors of order 1% or so (many of the pitfalls we have discussed will be second order effects for color information, as systematic errors in the aperture corrections, and systematic deviations in the color terms, will tend to cancel.). We do, however, advise using a generous size aperture, i.e., a radius that is somewhat larger than the largest FWHM measured on the frame.
2) At the 0.9-m, PSF fitting on the individual CCDs gives considerably worse results than simple aperture photometry; PSF fitting may, of course, be necessary in crowded fields. PSF photometry is unlikely to have an internal precision of 3% (1 ), no matter how many counts in a star image. In addition, one has the same 2% absolute problem inherent in aperture photometry. PSF fitting on the stacked images simply did not work, with occasional zero-point problems of 0.10 mag.
We thank George Jacoby, who obtained these frames with one of us (PM), and gratefully acknowledge useful correspondence with Frank Valdes and Linsey Davis on the photometry issues discussed above.
Phil Massey, Cathy Slesnick (REU student)