Previous Article
Next Article
Table of Contents

CCD Photometry: Some Problems with Reductions (1Dec92)
(from CTIO, NOAO Newsletter No. 32, 1 December 1992)

We have recently received several comments about possible photometric
problems found from measurements based on CCD images from our 1.5-m and
0.9-m telescopes. We wish to point out that there are in principle
several possible causes of these problems, including detector faults,
telescope/dome misalignments, reduction procedures and algorithms,
shutter timing errors, and sky conditions.

The CCDs are regularly checked for linearity and charge transfer
performance in our laboratories. Our old controllers, in use for more
than a decade, have recurrent noise problems, and the analog-to-digital
converters (ADC) have degraded in performance.  Also, some of them have
only 14-bit resolution. These latter ADCs are connected to the (15 bit)
computer system so that the least significant bit of the latter is
always zero, which means that the raw data contain only even numbers.
Spectroscopic observers usually set their CCD gain near 1 electron/ADU
so that in low background conditions readout noise always dominates
quantization noise. However photometric observers commonly set the gain
to a few electrons/adu in order to achieve more dynamic range.
Consequently, with low noise CCDs the quantization noise will be
comparable to, or maybe even dominate, the readout noise. In the low
background case the data will then appear quantized near values 0, 2,
4, 6, 8, etc. (14 bit ADCs) or 1, 2, 3, 4, etc. (16 bit ADCs). Our old
CCD controllers are in the process of being replaced, and in the
meantime we have retro-fitted new ADCs which give the full 15 bits of
resolution (0-32767).  The new controllers will use all 16 bits
(0-65535). Notwithstanding, the current information shows that the
detectors have excellent photometric performance.  However, we
recommend that observers consult their staff contact if they are
uncertain which gain setting is appropriate for their program.

The shutter timing errors are constant for a given CCD and
shutter/preflash assembly and can be calibrated out of the data.
Typical errors (0.9-m) are additive constants of about 50 ms in the
center, falling to 0 ms at the corners of the Tek 2048.

Both the 0.9-m and 1.5-m suffer some image quality degradation, due to
problems with mirror supports and alignment (unfortunately typical of
old telescopes). These problems may affect PSF fitting routines. As we
have noted before, the 0.9-m is a classical Cassegrain and has field
coma, which becomes important for the larger CCDs.  Users should be
very careful when attempting to do PSF photometry on these telescopes,
and consult the relevant staff contact. We are currently evaluating
these effects, and hope to provide a corrector in the coming year for
the 0.9-m.

We have noted two problems that definitely do affect recent
photometry.  On rare occasions the autodome gets "lost" in the
telescope control system, and of course this can occult the aperture.
Observer Support on the mountain checks this routinely, and the night
assistants easily recognize this problem. Observers should go into the
dome at least once during the run to make sure they feel confident
about the dome positioning.

The second problem concerns reduction routines, and has recently been
(re-)discovered. Under some conditions several of the IRAF
APPHOT/DAOPHOT algorithms used for determining "sky" perform poorly.
Specifically, in the case where the background is only a few counts and
detector read noise is not dominant, the width of the histogram of sky
values is comparable to the quantization of the data, and thus methods
assuming a smooth histogram may be inappropriate. This includes mode
and median estimators (e.g., the routine QPHOT uses modal sky).
Experiments by Mario Hamuy show that for a star with peak counts of
500-1000 ADU and near-zero sky, random errors of 0.2-0.4 mag. (several
tenths) are possible. In these cases an estimator such as the mean,
perhaps with rejection of outlying values, is much superior. The
circumstance where sky background is small most often arises when
making observations of standard stars or observations with narrow-band
filters. Frequently, standard stars are exposed to have almost
full-scale counts at peak, and precise location of the sky is not
needed. However recently, particularly with the advent of large CCDs,
it has become common to observe fields containing standard stars with a
wide range in magnitude (e.g. Landolt, AJ, 104, 340, 1992). In order to
use all the available standards in these fields, correct placement of
the sky background is essential. Note that the above comments do not
apply to the original (Stetson) versions of DAOPHOT. Peter Stetson
informs us that presently there is no evidence to show that his method
of calculating the sky (if mean >  median, then sky = 3  median - 2
mean, or if mean <  median, then sky = mean, where the median is taken
as the average of the 5% of the pixels around the numerical median)
introduces serious errors.

Finally, atmospheric extinction is presently elevated due to volcanic
dust from Mt. Pinatubo. Kv is still in excess of 0.20. Although it
appears that the absorbing layers are now relatively uniform, observers
engaged in all-sky photometry should proceed with caution.

A. Walker, R. Schommer, N. Suntzeff,
S. Heathcote, M. Hamuy

Previous Article
Next Article
Table of Contents