Assessment of Photometric Calibration

 

The primary sources of error are uncorrected extinction variations that affect both the PT patches and the 2.5m scans, incidents of bad seeing in the 2.5m scans coinciding with the crossing of PT patches, and statistical errors due to the small number of matching stars in certain PT patches in areas of the sky with low star density. Consequently, after final calibration, the data are subjected to a suite of tests to assess the reproducibility of photometric calibrations. These tests are neither exhaustive nor precise, but they do catch egregious errors. They are designed to uncover systematic calibration errors in the scan direction and across the imaging camera.

The principal tests for photometric accuracy include:

• Comparison of photometry of objects in overlaps between adjacent runs. Figure 11 contains a typical set of plots, for the difference in our calibrated magnitudes for objects measured in the overlap region in one camera column (camCol=3) for one pair of runs (752/756) for the five filters.
• Calculation of fiducial points in the color-color diagrams of stars. Stars lie for the most part on a narrow one-dimensional locus in SDSS color-color space newberg97, fan99a, finlator00. This locus is not straight, but shows some bends, in particular between K and M stars in the plot of vs. We measure the location of this break in color to high accuracy with the stars in 10-frame intervals, and look for trends in each run as a function of frame number. Figure 12 shows that the location of the stellar locus is stable in the EDR, with systematic drifts less than 0.01 mag. The ri Position and gr Position are the location of the bend in the stellar locus in the vs. color-color diagram, Figure 13. The iz Position is the mean for stars near that bend.
• Comparison of calibration zero points with the mean sky brightness of a given run. Although the sky brightness varies in any given band as a function of time, it is close to constant at a given time across the camera. Thus variations of the measured sky brightness as a function of camera column are an indication of photometric zeropoint errors.
• Examination of systematics in the photometry of stars as measured by the 2.5m and the PT in regions of overlap (see Figure 10).

These tests consistently indicate that the photometric zeropoints for the EDR data are internally consistent to within % in all bands for 90% of the frames in the EDR data. The worst outliers are all in u, and are up to 10% peak-to-peak in the worst of our data; as we saw above, these are due to ghosting in the u chip.

Because errors in different bands are often correlated, the colors of objects have smaller errors than might otherwise be indicated; stellar locii from different columns are typically aligned to better than 1% (measured in the g r i z bands). The median stellar locii from different runs are also aligned to better than 1%.

We discussed in § 4.4.5 that the estimated PSF magnitude errors are accurate to 10-20%. The errors themselves are impressively small, as is manifest by the width of the stellar locus; the rms scatter in bright star PSF photometry (where photon statistics are negligible), after correcting for zero-point offsets, is 0.02 mag in , , and , and 0.03 mag in and . The separation between the stellar locus and quasars in Figure 13 is due to this excellent photometry.

In addition to errors on the internal consistency of the photometry, the tie of the 2.5m photometry to an AB system has additional errors.

• The transfer of zeropoints from the USNO to the 2.5m currently does not use any color terms, thereby introducing errors in two places (USNO to PT and PT to 2.5m, as described above). The error so introduced is of order , although it may be somewhat smaller in r, i and z.
• AB systems are based on absolute spectrophotometric measurements of standard stars, and knowledge of the broad-band response functions, as well as the overall spectrophotometric scale tied to Vega; these factors together produce uncertainties at the level of 5%.
• The zeropoints given in fukugita96 ignore atmospheric absorption in the bandpass shapes, introducing an error of order in u and 1% in g.

Thus the system zeropoints could differ from the AB zeropoints by as much as .