The Final Calibration Pipeline (nfcalib)

Description

Figure 1: Sketch of the photometric calibration process.

Extinction is measured approximately every three hours by the PT telescope. Before calibration can begin, these values are interpolated to create approximate field by field extinction values, as shown at the top of figure 1.

The pipeline then examines patches overlapping the 2.5m images one by one. (See middle of figure 1.) For each star (saturated stars are rejected) which matches between the imaging 2.5m data and the PT patche, the following equation, indicating a zeropoint, is derived in each of 5 filters (u' g' r' i' z'):

where:

• The mag_PT is the calibrated magnitude of the star as determined from the Photometric Telescope (PT) system.

• k_PT is the extinction co-efficient, as determined from PT observations of a set of primary standard stars obtained the same night as the 2.5m imaging run was scanned. These measurements are typically made every three hours, and are interpolated between these points. As a function of filter, these co-efficients are typically:
  • k_u' = 0.55 +/- 0.10
  • k_g' = 0.15 +/- 0.05
  • k_r' = 0.10 +/- 0.04
  • k_i' = 0.08 +/- 0.02
  • k_z' = 0.06 +/- 0.02

• Color_term coefficients are solved for once or twice per year and held constant in between. They are of order a few percent, and allow for slight differences in the filter bandpasses between the imaging 2.5m telescope and the PT.

This set of approximately 100 star-by-star zeropoints is averaged with weights depending primarily on the errors in the star magnitudes to give a single zeropoint for each filter in each PT patch overlap. Because PT patches overlap just a few fields every hour or more, the solution must be extended from these fields to cover the remaining fields. Solutions are extrapolated as a constant to the beginning and ending of the run, and interpolated in fields between solutions, as shown in the bottom of figure 1.

Requirements

11. Time variation of photometricity along a stripe:

• Requirement: The rms variation in the photometric zero-point shall not exceed 0.6% over an arbitrary time span for any camera column.
• Basis: This test is an end-to-end test of the transfer of calibration from the PT to the 2.5m; additional errors may come from atmospheric extinction variations and other uncalibrated sources.
• How to test: Observe PT patches overlapping 2.5m scans with 1/2 hour separation. PT patches shall have been measured in close proximity in time and reduced in a single excal run with one extinction coefficient. Compute the rms of extinction-corrected zero points. Use data with minimal change in seeing to eliminate it as a variable. Use big aperture magnitudes on the 2.5m data to eliminate the effect of the PSF varying with seeing and position on the 2.5m camera.
• Current status: An ideal data set for this test does not yet exist, so this requirement has not yet been rigorously tested. Using existing photometric data sets, typical peak-to-peak errors of 1% or less are seen in g' and r', with occasional exceptions, indicating we are close to meeting the requirement. This requirement is currently being actively pursued with highest priority and an updated status will be presented at the review.

Results and Status

We have examined calibrations along three runs (745, 752, 756) which all at least partially overlap using available PT patches and tests comparing filters and adjacent columns.

• 756: From comparisons with the other runs and PT patches, run 756 is believed to be photometric to approximately 1%. Zero points determined for this run by nfcalib are constant to within ~1% (with a few exceptions) and well correlated across columns and filters, and thus are (mostly) within spec.
• 752: This run appears to be photometric to within 2%, and the determined zero points vary no more than this.
• 745: This run is believed to have a real extinction change of 4% across the run. This change is tracked by the zero point solutions.

Top priority is meeting the requirements listed above. While dispersion of matching any one PT patch to a 2.5m imaging run appears to yield rms errors within tolerance, patch to patch matching systematics are occasionally several times the requirement value. We are currently working to identify the source of these zero point variations between patches.

Problems we have examined or uncovered in this work include:

• The method of solution doesn't seem to be a problem; median, sigma clipped weighted mean, and least squares methods of determining the value all give similar solutions. The distribution of zero point values generally appears to be a well behaved Gaussian (sigma ~ 0.03) with a small number of outliers.
• PT patches appear to suffer from edge effects -- solutions which come from overlaps with only the edge of a patch are not reliable. This is believed to be related to sensitivity problems in the corners of the PT images and is being worked on. The current patch positions are not idea, and provide poor coverage for some columns. This has been fixed and will no longer be a problem as more patches are observed. Finally, due to historic problems with the PT, the number of good patches available is smaller than nfcalib would like. This will also improve as more patches are observed.
• We discovered that variations in seeing can cause systematic offsets if a large enough aperture is not used in photo for the variable PSF aperture corrections for 2.5m counts. (To confirm the stability of calibrations, nfcalib currently generates its own large aperture 2.5m magnitudes. Photo is being fixed now.)
• Extinction varying significantly on timescales shorter than 3 hrs would introduce additional errors, but they would be correlated across filters and columns.

Additionally, numerous cross checks of the PT patches against themselves and against repeated 2.5m imaging scans are underway.

Improvements

A suite of Q/A integrated tests which examine nearby SDSS imaging stripes for systematic photometric offsets is desired.

In some cases patches have been observed multiple times, or slightly offset patches may overlap the same fields. The ability to combine overlapping patches for a better solution should be added to nfcalib.

Automated code to run every six months to one year to determine color term co-efficients does not yet exist.

Needed:

Improved 2.5m aperture magnitudes from the photo pipeline. This is being fixed now.