Sloan Digital Sky Survey
Review of Data Processing Operations
Status of Photometric Calibrations
as of July 2000
July 5, 2000
The top level requirements for photometric calibrations is to achieve
rms errors of 0.02 mag in r', 0.02 in g'-r', r'-i',
0.03 in u'-g', i'-z'
for any randomly selected group of stars. The requirement on colors is
written with the expectation that there is a correlation of errors
among different filters, so errors of 0.02 mag on, say g' and r'
individual combine so as to yield an error of 0.02 mag on g'-r'.
No constraints on correlations on a single area of the sky.
The calibration of the SDSS data is a multistage process:
- The fundamental reference star is BD+17 4708. Its SDSS magnitudes
are derived from accurate spectrophotometry.
- A set of ~150 primary stars around the sky have been calibrated
relative to the fundamental star using observations obtained during
1998-2000 using the USNO 40 inch telescope, a UV thinned CCD, and a set of
SDSS filters. The effective wavelengths of this system define the SDSS
- A set of secondary patches have been defined that cover the 2.5m
survey area. These are spaced so as to cover all scanlines on each stripe
approximately once per hour of scanning. The photometric telescope (PT)
observes a mix of primary stars and secondary patches on each
photometric night. These data are processed by the MT pipeline.
The outputs are calibrated patches and extinction doefficients night by night.
- Each scanline from the 2.5m camera is calibrated using
data from one or more patches using the Final Calibration pipeline.
Summary of known problems
- We currently use twilight flats for the u' band and dome
flats for the other bands. Flatfielding of the PT patches has been
difficult due to several problems. These include:
- Oil contaminant that collects on the CCD. This problem is thought
to have been due to use of an improper vacuum grease (which evaporated
in the heat of the PT dome during the summer) and is thought to
be solved, but continuous monitoring is needed.
- Incomplete baffling of the PT. Three rounds of filter masking and
baffle additions have been made to the telescope since the beginning of
- Improper filtering of the flatfield lamp, producing the wrong
color balance in the g' band.
- Two glass correctors near the focal plane have curved
surfaces; these can lead to scattered light problems. No effects have been
discerned so far; not are expected if antireflection coatings are working as
As a consequence, we have much data in hand with varying degrees of
- 3 bands (g', r', i') use interference filters to define the
long wavelength limit. The filters on the 2.5m (and perhaps on the PT)
have shifted effective wavelength, likely due to the fact that the filters
on the 2.5m are in vacuum, while on the PT they are in dry air.
This complicates our ability to compute color transformations between
the different telescopes. At present, no color terms are used, which
introduces small calibration errors, but more importantly means that it
is possible that the 2.5m photometry will not be on the same systems as
the standard stars.
- The MT pipeline code has not been tested rigorously end-to-end to
determine if its outputs are correct. (Casual testing has been done).
Known problems include:
- Fringing in the i' and z' bands is severe enough
to need correcting; nothing is done at present.
- Errors on quantities are not computed correctly.
- There is a sense that scatter in computed magnitudes
is larger than it ought to be.
- The final calibration pipeline is known to not compute zero points
efficiently (using medians rather than weighted averages), although
a recent change in the pipeline produced no significant change in the
- The photometric pipeline corrects PSF magntitudes to a
standard aperture, and until recently the aperture size was too small,
introducing seeing-dependent variations in the zero point (up to 5% in bad
Detailed Requirements and Tests
Each detailed requirement is quoted for just the r' filter;
requirements for other filters or colors scale the same as for the top level
- Internal error of primary stars < .01 mag
- Status: Met. 1+ years of USNO data have been reduced.
The rms errors per observation are 0.01 mag in g', r', i', z', 0.03
mag in u'. Each star is observed 10+ times, so errors in mean
are must smaller.
- QA check needed: None
- Future work: Monitor primaries for variability;
monitor additional stars in each field.
- USNO to PT transformation: No more than 0.6% error per star.
- Status: Not verified.
We have many solutions PT vs USNO, but b terms are not solved
for yet. Any transformation errors will likely show up as
correlations of residuals with color. Additional tests are to
search for correlation of residuals with time, azimuth, etc.
Hogg finds nonzero color terms, but it is unclear if he
used the latest standard star values; he certainly did not
use c terms, which will affect b.
- Known problems: There is a suspected problem with changes in the
filter transmissions curves due to environmental effects.
- What needs to be done: Measure the transmission
curves of the PT and USNO filters. Concoct a "super-excal" that reduced a
large body of PT data against USNO; use proper transformation equations.
- PT linearity - Peak uncorrected nonlinearity error shall be < 0.3%.
- Status: Satisfied
- Check: Tests are reported in "Verification of the
SDSS Photometric Telescope I. Electronics" by Fukugita, Ichikawa, Watanabe,
and Yasuda Peak-peak nonlinearity is 1%. Corrections are applied in mtpipe.
Without correction, worst error is 0.5%. Errors in corrections have not been
- Nightly PT extinction & zero-points - rms per star/sqrt(# degrees of freedom)
- Status: Satisfied on good nights. Main source of error is variations
in extinction and uncorrected flatfielding errors.
- Test: Look at rms solutions of standard stars on a typical night:
We get rms mag per star:
rms of solutions are above / sqrt(23); .007 mag or less.
To do: Repeat with latest standard star values; repeat
with fringing corrections; repeat i' and z' with grid corrections or remove
observations not exactly centered on CCD. Need to compute rms properly in
- Flatfielding of PT patches - rms error of 3x3 grid test patches should
be no worse than 0.6% for a single star.
- Status: One set of grid tests has been fully processed in 5 filters
with PT in a usable state. rms errors are as follows:
u' .008 mag (MJD 51614)
g' .020 mag (MJD 51614)
r' .016 mag (MJD 51614)
i' .010 mag (MJD 51667)
z' .018 mag (MJD 51667)
We are not yet meeting spec except at u'.
- Known problems: The flat field illumination pattern has changed
several times due to configuration changes in the PT (addition of baffles,
changes in FF lamp filterse, etc). An additional problem was introduced
when the CCD warmed up just before the cooling system expired.
- Plans to rectify: Continue to obtain more grid
tests, now in a 3x3 pattern. If the flatfields remain stable and repeatable,
it will be possible to construct a correction frame based on the measured
pattern of photometric variations. At present the amount of data is
inadequate to do this. If other contributions to the error budget are
smaller than budgeted, the spec on flatfielding could be relaxed.
- Zero-point differences for bright and faint stars on PT patches
shall not exceed 0.6%.
- Status: Not tested. Data are in hand - they just need someone to
- Available test data: Short and long exposures of patch 10d5.
Many patches that overlap 2.5m scans. A few patches that, by
accident, overlap by 1/2 CCD width.
- How to rectify: Perhaps the Hogg task force can take
- Photometric mismatch between PT and 2.5m after applying a linear
color term shall be less than 0.6% rms.
- Status: Unresolved. It is now apparent that the filters on the
2.5m have shifted effective wavelength since they have been in a vacuum.
No one has recomputed the shift in colors with the new data.
- Tests data: This requirement was to be tested using synthetic
photometry alone. However, it can be partially tested using
PT vs 2.5m data, looking for residuals vs. color.
- No QA tests are needed during the survey.
- Known problems: Getting traces of the 2.5m and PT
filters is tedious.
- Variations in the aperture correction vs. column number within a single
2.5m CCD shall not exceed 0.6%.
- Status: Spot checks show that we are meeting this in the camcols
with the worst PSF variation. A full set of tests remains to be done.
- Test data: Comparison of camcol 6, r', psf mag with synthesized
- QA tests: Plots of 2.5m minus PT residuals vs. CCD columns are
a standard QA plot.
- Known problems: At present the 2.5m aperture
corrections use too small an aperture.
- Spatial variations in 2.5m - PT photometric calibration vs. CCD column
shall be less than 0.6% rms.
- Status: Spot checks show that we are meeting this requirement.
- Test data: Numerous PT vs 2.5m overlaps.
- Same QA plots as in the previous requirement provide a cross-check
- Known problems: None
- Time variation of aperture correction along a column shall not exceed
- Status: Not met.
- Test data: Runs 745, 752, 756.
- QA tests: These are standard QA plots produced by final calibration.
- Known problems: Photo uses too small an aperture. A
big one may require adding wing stars. Some tests with synthetic aperture
photometry indicate that a big aperture can meet this requirement.
- Time variation of photometrity along a stripe shall not exceed 0.6%.
- Status: No known problems, but only spot checks have been done.
- Test data: Runs 745, 752, 756. PT-2.5m zero points change by
1-2% peak-peak. A known event where the extinction increased during
run 745 was properly tracked by the PT.
- QA tests: On long runs with multiple PT patches observed on the same
night, track the variation in zero point of the photometric transfer.
This is done in final calibration.
- Known problems: We have much experience with
photometricity variations at the USNO (Flagstaff AZ, elev. 7000'), which
show that variations up to 4% can occur on some nights.
Review of SDSS Data Processing and Distribution