The major issues identified during the last 2MASS Science Team meeting relate
to the overall uniformity of point source photometry, and the ability
to meet the Level 1 Specifications. They include:
- Photometric response variations of the detector/focal plane, their
impact on photometric uniformity and how they may be minimized.
- Improved photometry algorithms, including the use of adaptive psf's
within KAMPHOT, and the hybrid aperture/KAMPHOT magnitudes.
- Photometric calibration issues, including how well can night-to-night
calibration (zero-point, extinction and color-terms) be evaluated using
photometric standard observations, and how uniform is photometry over
nightly and greater timescales.
DATA PROCESSING STATUS
Initial processing of the April/May 1995 Protocamera run data with
the operational (ops) pipeline was completed at 2:15am on 20 July 1995.
Coadded images, point and extended source lists with photometry and ancillary
files are available on-line to Team members. The location of data for
each scan is listed on the WWW 2MASS Protocamera Processing Status page
which can be reached via
the IPAC/2MASS information page.
All Team members are welcome and encouraged to review the data products.
You will need an active ipac computer account to view or retrieve files,
so please contact Jack Lampley at IPAC (818-397-9551) to set up an
account if necessary.
Point source extraction, position reconstruction, aperture and psf-fit
(KAMPHOT) photometry, extended source identification and photometry,
and source list cleaning and merging was carried out separately out for
each scan. All scans that displayed difficulty in any stage of
processing are flagged in the Bad Scan Log for each night. The most
common processing problem was position reconstruction; reconstruction failed
for 5-7% of the scans, primarily in very high source density and/or high
extinction regions because of difficulties in matching visual wavelength
guide stars with highly confused near infrared star fields. No effort
will be made at this time to complete position reconstruction for these
fields. Fields that lack proper position reconstruction still have source
extractions, photometry and coadded images in most cases. New tools that
were developed for the position reconstruction of IRTS data have been
tested on selected confused 2MASS regions and successful position
reconstruction has been achieved. These tools will be further tested and
incorporated in the final version of 2MAPPS.
DETECTOR SPATIAL PHOTOMETRIC RESPONSE VARIATIONS
A number of special observations were made during the April/May run to
examine how the photometric response varies across the 2MASS Protocamera
detector focal plane. The observations consisted of 10 repeated one degree
scans across M67, with the telescope stepped 40" (20 camera pixels) in the
cross-scan direction (east) between each scan. Since each point on the
sky is sampled six times during a scan, a star could fall on up to 60
different positions on the array during the 10 scans. The photometric
response around the focal plane is mapped by measuring how the
instrumental brightness of a star varies with position on the array.
Factors other than detector response can contribute to brightness variations
in a source, such as pixelization, seeing and atmospheric transparency.
To minimize the impact of these other effects, and to improve the statistical
accuracy of the measurements, we make use of large numbers of stars observed
in the 10 scans. In addition, we evaluate how the brightness of a star
varies in each apparition relative to its average brightness for all
apparitions.
The steps used to create a photometric response "map" are as follows:
- The photometric response test scans are processed using the ops
pipeline and standard dark-sky flat-fielding techniques.
- Simple aperture photometry is extracted for sources in each individual
frame (i.e. not the coadds).
- The average brightness of each source for all apparitions (at all
positions on the array) is calculated.
- The ratio of the brightness of a source in each apparition to its
average brightness is calculated.
- The ratios for all sources that fall in 25x42 pixel subsections of
the array are then averaged together. That is, the array is subdivided
into 10x6 positional "bins" and the ratios of all sources that fall within
those bins are averaged together.
The relative photometric response map is the 10x6 array of average
brightness ratios. In practice, a brightness threshold is imposed on
the sources used to generate the map to ensure that the ratios
calculated are not dominated by measurement errors for individual
stars. Response maps have been generated for J, H and Ks data obtained on
95-04-22, and they may be viewed from
the IPAC/2MASS information page.
The Prototype Camera was cooled with liquid argon during the April/May
run, which resulted in a slightly warmer array temperature than
previously (approx. 80K versus 72K). Laboratory tests suggested
that this could help suppress the detector reset decay pattern
observed in the bias frames that seemed to dominate the earlier
responsivity tests. The new responsivity pattern is similar to that
seen in the 1-dimensional ROUND analyses of the 94-06-01 repeated scans
of M92. The largest variations occur in the in-scan direction with
response change up to 10% peak-to-peak in a pattern that follows
approximately the reset decay pattern. The amplitude of "step" that
occurs at mid-array is similar to the earlier data, although the full
peak-to-peak scatter appears to be slightly lower. The general shapes
of the response patterns in J, H and Ks are similar, but there are response
"features" that differ slightly between each that appear to be
statistically significant.
While the in-scan response variations appear drastic, the 2MASS
technique of obtaining six samples during a scan helps to smooth
them out. However, if one or more of the six samples for a source
is missed for any reason (i.e. bad pixels, cosmic-ray hits, etc...) the
net response along a column will be altered contributing to overall
photometric non-uniformity. Since a sample is likely to be missed for
a non-negligible percentage of sources (e.g. 3% of sources for 0.5% bad
pixels on the array), there is good motivation to smooth the in-scan
response function. This affects aperture photometry primarily since it
is derived from the coadded data. KAMPHOT is not as sensitive to
missed samples because it operates on individual frames and can exclude
"bad" samples.
The cross-scan response variations are smaller than we had feared a few
months ago, running about 5% peak-to-peak. Even these small variations
affect the photometric uniformity, though, as can be seen when comparing
repeatibility tests of stars measured during the response tests with those
measured in normal calibration scans that used very little cross-scan stepping
(M67 was also a calibration field used throughout the run). Photometry for
brighter stars (Ks<12) in the normal calibration scans have characteristic
dispersions of 1-2% in multiple measurements, which approaches the theoretical
limits imposed by the large pixels and pixelization errors. The dispersion in
the photometry of high SNR stars from the cross-stepped scans is 3-4%,
which should be more representative of the true photometric uniformity of
the survey.
If the repeatibility of the photometry from non-cross-stepped scans
represents the best that can be achieved, then the goal is to determine
how to carry out or correct the cross-stepped photometry to bring it to
the same accuracy. New analysis tasks to investigate this are now underway.
These include:
- Measure the photometric response maps for the nights of 95-04-26
and 95-05-10. Note any variations with time.
- Measure the photometric response pattern of the Protocamera using
data flat-fielded with twilight sky measurements.
- Examine photometric consistency after correcting the "normal" dark-sky
flattened images using the photometric response map described above.
IMPROVED PHOTOMETRY ALGORITHMS
- Adaptive PSF --
As proposed at the February 1995 Science Team meeting, two refinements
to the photometry algorithms were included in the processing of the
April/May Protocamera data. The first was to use an "adaptive" psf
in KAMPHOT instead of the single psf used to process the June 1994
data. Rather than determine the local "true" psf for every scan,
which would require prohibitive amounts of cpu time for the survey
and would be difficult in highly confused or very sparse regions, the
psf used to extract photometry from a scan was selected from a pre-defined
grid of psf's. Selection of the psf was based on a single seeing estimator
for the scan, PFRAC. PFRAC is a measure of the ratio of the energy containing
in the peak brightness pixel on a source to the total energy for an ensemble
of sources, and it correlates well with image FWHM. Thus, for low values of
PFRAC, the broadest psf was used, and sharper psf's were selected for larger
PFRAC's. Because of time limitations, a sparse grid of psf's containing only
three per band was generated for the initial processing of the
Protocamera data.
- Hybrid Magnitude --
Repeatibility studies of photometry from the June 1994 Protocamera data
indicated that aperture photometry yields slightly lower dispersion
for bright point sources and KAMPHOT performed better for fainter
sources. The cross-over point occurs near Ks=12.5. Therefore, to
exploit the strengths of each, an algorithm to derive a hybrid magnitude
that is a simple linear combination of the aperture and KAMPHOT
magnitudes was developed for use in the current Protocamera processing.
This algorithm defaults to pure aperture photometry for Ks < 11.5
and pure KAMPHOT magnitude for Ks > 13.5. The crossover and limits
occur approximately 0.5 magnitudes fainter for H and 1.0 magnitudes for J.
- Performance - Internal Repeatibility --
During the April/May Protocamera run, calibration fields normally were
observed by making six one-degree scans in rapid succession at each
wavelength. Thus, we have data to monitor the repeatibility of photometry
on timescales ranging from a few minutes to several nights. The internal
(short timescale) repeatibility of point source photometry produced with both
aperture and KAMPHOT algorithms from the April/May run is excellent,
as good if not better than that observed in the June 1994 data. The
new data show that the J and H repeatibility is as good as that at Ks.
Stars brighter than J=13, H=12.5 and Ks=12.0 show characteristic dispersions
of only 1-2% in their photometry. As in the June 1994 data, aperture
photometry provides slightly better repeatibility for brighter stars,
while KAMPHOT is better for faint stars. However, KAMPHOT with its
adaptive psf grid containing only 3 psf's per band, is nearly as good
as aperture photometry for bright stars, having characteristic dispersions
of 2-3%. Note this is as good as measured for the twilight-sky flattened
KAMPHOT results from June 1994. Using a more extensive grid of
psf's and optimum flat-fielding may further improve KAMPHOT to the point
where it can be used at all brightness levels, eliminating the need for
the hybrid magnitude. The hybrid magnitude tracks aperture KAMPHOT
photometry precisely as designed, following the minimum of the two.
The dispersion is less than or equal to 10% (i.e. measurement at SNR=10)
out to J=15.8, H=15.2 and Ks=14.3, in excellent agreement with the Level 1
Specification.
We also conclude that cooling the detector using LAr has had no adverse
impact on the photometric uniformity. Analyses are currently underway
to determine the effects of the warmer operating temperature on net
completeness and reliability.
PHOTOMETRIC CALIBRATION - External Repeatibility - Seeing Effects
Testing the photometric uniformity of data acquired over longer time
scales requires calibration to common photometric zero points at
each band. The first attempts at this using the April/May Protocamera
data consisted of deriving simple photometric transformations for the
standard stars within each calibration field. The transformations
have the traditional form:
Mtrue = Minst + k1*X
where k1 is the slope of the extinction correction, X is airmass and Minst
is the instrumental magnitude of the star. Note that we currently ignore
color terms. It was immediately apparent that the transformations using
both aperture, KAMPHOT and hybrid photometry were unphysical, exhibiting
much scatter and frequently having extinction terms much larger than the
canonical 0.1 mag/airmass typical for near-infrared wavelengths.
The cause of this problem is apparently linked to seeing variations.
For example, comparison of measurements of stars in the P221-C calibration
field observed at different times on 95-04-22, but at the same airmass,
shows a systematic zero point offset of nearly 0.1mag. The seeing estimator,
PFRAC, differs between the two indicating a seeing variation of nearly 0.5
arc-seconds FWHM. Examination of the point source curves of growth
derived from multiaperture photometry in these fields reveals that the
8" diameter aperture used in the current pipeline misses considerable
light, up to 15-18%, in the rather poor seeing we encountered for much
of the early part of the run. Moreover, the fraction of light falling
outside of the aperture varies considerably with seeing, leading to apparent
photometric offsets.
- Aperture Photometry Corrections --
To correct for the effects of seeing in the aperture photometry, we derived
mean aperture corrections (the difference between the 8" diameter aperture
magnitude and the "infinite" aperture magnitude) using all of the
brighter stars in each scan, and applied them to the photometry.
The corrected aperture photometry for the two P221-C scans shows no net
offset. Similarly, aperture corrections were derived for most of the
calibrations scans made during the run, and the corrected standard star
aperture magnitudes produce much more reasonable photometric transformations
for each night.
- KAMPHOT Corrections --
The variations in the KAMPHOT photometry are more difficult to understand.
One can imagine that psf-fitting, if performed properly, should
be "seeing-independent". One deficiency in the current implementation
of KAMPHOT is that only 3 psf's per band are used, and the selection of
psf is seeing driven (based on PFRAC). If the psf that is utilized for a
given measurement does not match the true psf there will be some
systematic offset in the photometry. Note that in the final version of
2MAPPS the KAMPHOT psf grid will be much better populated. However, we find
photometric inconsistencies even in tests where KAMPHOT is allowed to
measure the "true" local psf within scans.
Currently, KAMPHOT uses only a 2x2 pixel (4"x4") "kernel" to fit the
image profile. If the seeing degrades considerably, there may not be
sufficient information to achieve proper fitting and that may result in
an incorrect brightness estimate. Furthermore, as currently implemented,
KAMPHOT normalizes its psf-fit magnitudes to a constant internal zero-point
rather than to internal aperture photometry. That zero-point is the
proper value gauged from earlier Protocamera data obtained under fairly
unform seeing conditions. However, the psf zero-point normalization
should really be measured locally if significant seeing variations occur.
We are currently testing the effect of enlarging the KAMPHOT input
array size and its impact on processing time and signal-to-noise.
In addition, we propose that the most direct method to ensure
proper normalization of the KAMPHOT photometry will be to adopt the
traditional CCD photometry technique of normalizing to aperture photometry.
To implement this, both aperture and KAMPHOT photometry would
be extracted for a scan. KAMPHOT magnitudes would then be adjusted by
the zero-point correction necessary to drive the median KAMPHOT-Aperture
magnitude difference to be zero for a given scan. This solution assumes
that aperture photometry produces the right answer, as we believe it
will do when aperture corrections are measured and applied. This will
be difficult if not impossible in crowded fields. Consequently, it may
be preferable to deduce aperture corrections from a more robust
seeing estimator, such as PFRAC, that can be trusted in crowded
fields, or interpolated from nearby scans.
- Ongoing Analysis Tasks --
The following analysis tasks related to photometric transformation and
external uniformity are underway or planned for the near future:
- Derive aperture corrections and apply to aperture photometry for a
large number of scans to examine the uniformity of photometrically
calibrated data from different nights.
- Use all stars in calibration fields observed at different airmasses
to determine extinction terms for different nights and note variations
with time.
- Normalize KAMPHOT magnitudes to aperture-corrected aperture magnitudes
for repeated calibrations scans, and examine both internal and
external repeatibility of corrected KAMPHOT photometry. Determine
whether the hybrid magnitude is still necessary.
- Test performance of KAMPHOT with better populated psf grids.
- Test KAMPHOT using larger psf-fitting kernels and understand
impact on processing time and SNR.
CONCLUSIONS
Preliminary analysis of the point source photometry from the April/May
Protocamera has shown that the refined algorithms are producing results
with excellent internal uniformity. For the first time we have data
to study the effects of focal plane response variations and seeing
variations on global photometric uniformity. The preliminary
analysis indicates that these factors degrade the uniformity in
reasonably well understood ways, and we are further characterizing the
effects of seeing on psf-fit photometry. Efforts are now underway to
develop and test refinements to the flat-fielding procedures and photometry
algorithms that will minimize these effects.
We welcome any suggestions you might have regarding the observed effects
and planned corrections and analysis.