Date: Wed, 1 Oct 1997 17:50:00 -0700 (PDT)
Subject: 2MASS WG Mtg #131 Minutes

           IPAC 2MASS Working Group Meeting #131 Minutes

Attendees: R. Beck, T. Chester, R. Cutri, D. Engler, J. Fowler,
           L. Fullmer, T. Jarrett, D. Kirkpatrick, G. Kopan,
           H. McCallon, B. Nelson, S. Wheelock, J. White


1.) Optical-Catalog Associations
2.) Telescope Control Errors
3.) Early Data Release
4.) Policy Change Regarding Darks/Flats
5.) 14-Bit Tests


1.) Optical-Catalog Associations

    As discussed in last week's minutes, the optical catalog
association capability has been implemented in the POSMAN module
POSPTS, complete with final tweaking of the 2MASS point-source
positions (image position coordinate updating will be implemented
in the near future).
    Analysis of early results, however, has shown that many 2MASS
point sources are not associated with optical catalog objects
(currently USNO-A and Tycho catalogs), and that two different
reasons are obvious. First, many of the 2MASS objects are very
faint and probably fairly red, thereby not showing up in the POSS
plates from which the USNO-A catalog was made or in the Tycho
survey. Second, many of the USNO-A objects have accumulated
significant proper motion, e.g, in excess of one arcsecond; this
is enough to cause them to fail the matching test with its
initial thresholds, which were set to compromise completeness
somewhat for the sake of reliability.
    The proper-motion effect illuminates the question of how we
really want the thresholds to be set. On the one hand, it would
be nice to make the associations for most of the objects that
have moved, since this would facilitate proper motion studies.
But in order to achieve that, the thresholds would have to be
opened up enough to allow some spurious associations, especially
in dense regions. Another question is how much the term
"association" implies about the objects being related; clearly it
implies less than "identification", but perhaps more than
"happens to lie nearby".
    Some users of the association information will not be
interested in proper motion at all. Since the associations supply 
five-channel photometric information when the associated
measurements apply to the same object, extremely useful criteria
are available for selecting specific classes of objects for
further study. This usefulness would be significantly diluted if
a large fraction of the associations were spurious because of
wide-open thresholds meant to capture proper-motion information.
    R. Cutri will conduct a survey of the science team members
who have special interests in the associations and gather a
refined set of requirements which will be used to set the
operational thresholds. It is likely that the data structure will
change also, and the matching algorithm may require some

2.) Telescope Control Errors

    R. Cutri reported that telescope pointing control errors have
been observed to increase significantly recently. What appears to
be a hysteresis effect on the order of 20 arseconds is seen in
right ascension in the data of 970927. The source of this problem
is currently being sought.

     [NOTE ADDED IN PROOF: R. Stiening distributed a report on
     971001 in which it is stated that errors up to 50 arcsec in
     RA and 16 arcsec in Dec occurred on a long slew across the
     zenith; after considerable investigation and hardware
     adjustments (tightenings and realignments), slews were
     achieved which returned to their starting point to within
     less than 2 arcsec on both axes, so this problem seems to be

3.) Early Data Release

    R. Cutri reported on a telecon in which early data release
had been discussed. The request for early data release came from
NASA Headquarters. The project concern is the one usually
associated with early data release: assuring the quality of the
data without diverting resources needed elsewhere to conduct the
survey efficiently. It appears that an acceptable plan has been
developed that involves phased releases. In the near term, the
2MASS image gallery already on the web will be made available to
the scientific community; this image gallery has been around long
enough for any concerns about its quality to have dissipated. In
two to three months, several carefully selected coadded images
and corresponding source lists will be added. Sometime around
next June, about half a dozen survey scans will be made available
on line, accompanied by real access tools.

4.) Policy Change Regarding Darks/Flats

    R. Cutri and J. Fowler reported on a telecon with M.
Skrutskie in which darks/flats acceptance thresholds and data
utilization policy were discussed. Early in the design stage of
the DARKS subsystem, the project expectations were that the
hardware would be fairly stable in terms of dark response and
pixel responsivity; this led to a policy involving testing each
night's dark and flat sequences against "canonical" values with
fairly tight acceptance criteria. If the night's darks/flats
deviated significantly from the canonicals, then they were not to
be trusted, and the canonical darks/flats were to be used. If the
night's results were within threshold, then they were to be used
instead of the canonicals, in order to track small variations.
The assignment of any set of results to "canonical" status was to
be done only via careful offline analysis, not in an automated
    In practice, the nightly results almost always have failed
the acceptance tests. Every time any change to the hardware has
been made, the behavior has changed enough to cause rejection
when compared to canonical darks/flats obtained from data taken
prior to the hardware change. Even the amount by which the tests
are failed drifts significantly from night to night. The problem
is paradoxical in that the more the night's results differ from
the canonicals, the less they can be trusted, but the more
inapplicable the canonicals may be.
    Since most nights processed so far have used darks and flats
that were probably sub-optimal, one might wonder how the products
could be coming out of the pipeline with such apparently high
quality. This is most likely due to the following facts: (a.) the
responsivities do not seem to suffer as much from variation as
the darks; (b.) the flattening method in PIXCAL/DFLAT depends on
additive pixel corrections (not multiplicative) which apparently
are able to absorb practically all of the dark model error. So
the good news is that the processing seems to be rather robust.
    It was decided that we want to track slowly varying changes
in hardware behavior, and so two changes will be made: (a.) the
acceptance thresholds will be opened up enough to allow more
nightly-versus-canonical tests to be passed; (b.) the concept of
"canonical" darks and flats will be modified - for purposes of
the DARKS subsystem, the canonical data will be those used in the
most recent previous observation night, independently of whether
those were computed from that night's data. This allows each
night's darks/flats to become the new canonicals automatically if
they pass the tests relative to the previous canonicals, thereby
tracking real slowly varying changes. At the same time, a "grand
canonical" set of darks/flats/masks will be maintained under
human judgment for most quality analysis; these will be the first
acceptable set after known hardware configuration. For now, these
will be the set obtained from the night of 970903.

5.) 14-Bit Tests

    Most subsystem cognizant engineers have now reported to R.
Cutri on the effects of the 14-bit simulation (see last week's
minutes). The consensus was that we should avoid going to 14-bit
A/D converters if feasible. J. Fowler reported degadations in the
output of BANDMERGE at the 1% to 3% level. L. Fullmer reported
that direct source comparison of the same scans processed both
ways showed small mean position offsets, but position-discrepancy
standard deviations well above 0.1 arcsec for all S/N bins.