Content uploaded by Giuseppe Longo
Author content
All content in this area was uploaded by Giuseppe Longo on Jul 20, 2013
Content may be subject to copyright.
arXiv:astro-ph/9809187v1 15 Sep 1998
The Palomar Digital Sky Survey (DPOSS) 1
S. G. Djorgovski1, R. R. Gal1, S. C. Odewahn1, R. R. de Carvalho2, R.Brunner1,
G. Longo3, R. Scaramella4
1Palomar Observatory, Caltech, Pasadena, CA 91125, USA
2Observatorio Nacional, CNPq, 20921 Rio de Janeiro, Brasil
3Osservatorio Astronomico di Capodimonte, I-80131 Napoli, Italy
4Osservatorio Astronomico di Roma, I-00040 Monteporzio, Italy
Abstract. We describe DPOSS, a new digital survey of the northern sky, based on the
POSS-II photographic sky atlas. The survey covers the entire sky north of δ=−3◦in 3
bands, calibrated to the Gunn gri system, reaching to equivalent limiting magnitude of
Blim ∼22m. As a result of the state-of-the-art digitisation of the plates, detailed processing
of the scans, and a very extensive CCD calibration program, the data quality exceedes
that of the previous photographically-based efforts. The end product of the survey will
be the Palomar-Norris Sky Catalog, anticipated to contain >50 million galaxies and >2
billion stars, down to the survey classification limit, ∼1mabove the flux detection limit.
Numerous scientific projects utilising these data have been started, and we describe briefly
some of them; they illustrate the scientific potential of the data, and serve as the scientific
verification tests of the survey. Finally, we discuss some general issues posed by the advent
of multi-terabyte data sets in astronomy.
1 Introduction
The Palomar Digital Sky Survey (DPOSS) represents a digital version of the POSS-
II photographic sky atlas [19]. It is based on the scans of the original plates, done
at the Space Telescope Science Institute [14], [16]. The final result of this effort will
be a catalog of all objects detected down to the survey limit, the Palomar-Norris
Sky Catalog (PNSC). For more details, see, e.g., [5].
The goal of this project is to provide a modern, uniform digital data set covering
the entire northern sky in 3 survey bands (photographic J F N, calibrated to Gunn
gri), with a photometric and object-classification accuracy of sufficient quality to
enable a wide range of scientific follow-up studies.
The survey processing and calibration were designed to extract all of the in-
formation present in the plates. Our tests indicate that the resulting DPOSS and
PNSC data are superior to other, photographically based sky surveys in the past
in terms of the photometric quality, uniformity, depth, and object classification ac-
curacy. We believe that this improvement is due to a combination of factors: (a)
a superior scanning process, which minimizes scattered light problems, while main-
taining the full angular resolution and dynamical range present in the plate data;
1To appear in Wide Field Surveys in Cosmology, proc. XIV IAP Colloq., eds. S. Colombi and
Y. Mellier, in press.
(b) an unprecedented amount of CCD calibrations and object classification training
data sets; (c) detailed processing which extracts all of the information present in
the plate scans, including various advanced techniques for object classification.
While the future fully digital sky surveys (e.g., the Sloan DSS; [8]) will provide
photometric data of superior quality and depth than what is possible with photo-
graphic technology, such data will not be generally available over a substantial area
on the sky for some years to come. In the meantime, DPOSS and PNSC can provide
to the astronomical community data of adequate quality for many scientific studies,
e.g., as envisioned for the SDSS. Moreover, DPOSS will also cover the lower Galactic
latitudes, where SDSS data will not exist.
We have already started a number of scientific follow-up studies using DPOSS
data, some of which are described below. We find that in order to create scientif-
ically useful data, it is absolutely essential to try to use the data for some actual
scientific studies right away, and thus discover any possible problems which may be
affecting the data. No other kind of a priori constructed tests can really provide
the necessary feedback. Scientifically viable catalogs cannot be constructed without
such validation tests. Yet, this is a lesson which is not always appreciated by science
managers or funding agencies.
In the process of analysing DPOSS and creating the PNSC, we have encoun-
tered the problems posed by the handling, management and manipulation of multi-
Terabyte data sets, which are now appearing in astronomy and many other fields.
Beyond the simple handling and maintenance of such data sets, a much more inter-
esting set of problems arises on how to explore or data-mine them effectively – how
to convert this great abundance of information into actual scientific results. Tradi-
tional data processing techniques are simply inadequate for these tasks. We describe
below some of our experiences and ideas on how to approach these problems.
2 Survey Specifics
The POSS-II covers the entire northern sky (δ > −3◦) with 894 overlaping fields
(6.5◦square each with 5◦spacings), and, unlike the old POSS-I, with no gaps in
the coverage. Approximately half of the survey area is covered at least twice in
each band, due to plate overlaps. Plates are taken in three bands: blue-green,
IIIa-J + GG395, λeff ∼480 nm; red, IIIa-F + RG610, λef f ∼650 nm; and very
near-IR, IV-N + RG9, λef f ∼850 nm. Typical limiting magnitudes reached are
BJ∼22.5, RF∼20.8, and IN∼19.5, i.e., ∼1m−1.5mdeeper than the POSS-I.
The image quality is improved relative to the POSS-I, and is comparable to the
southern photographic sky surveys. As of the summer of 1998, the plate taking is
100% complete in J, ∼97% complete in F, and ∼83% complete in N, and it should
be finished by the spring of 1999.
The original survey plates are digitized at STScI, using modified PDS scanners
[14]. The scanning is closely following the plate taking in terms of the completeness.
The STScI group is using these data and an independent processing software to
generate the second HST guide-star catalog, GSC-2. The plates are scanned with
15-micron (1.0 arcsec) pixels, in rasters of 23,040 square, giving ∼1 GB/plate,
or ∼3 TB of pixel data total for the entire digital survey (DPOSS). Preliminary
astrometric solutions are good to r.m.s. ∼0.5 arcsec, and should eventually reach
r.m.s. ∼0.3 arcsec. An independent digitisation of these plates is planned at
USNOFS, by Monet et al.
These data are superior to the widely available DSS scans, for several reasons:
(a) the original plates are scanned, rather than second copies; (b) the plates have
a finer grain, better image quality, and reach ∼1mdeeper than POSS-I; (c) the
scans have an improved dynamical range and finer pixels, viz., 1.0 arcsec instead
of 1.73 arcsec used in the DSS. All of the cataloging work at STScI and Caltech is
done on the original scans, rather than the compressed images available through the
appropriate web servers and in the anticipated future digital media distributions.
Our ongoing effort at Caltech and the sites in Italy (OAC, OAR) and Brasil
(ON/CNPq) is to process, calibrate, and catalog the scans, with the detection of all
objects down to the survey limit (approximately to the equivalent limiting magnitude
of BJ∼22m), and star/galaxy classifications accurate to 90% or better down to
∼1mabove the detection limit.
We use SKICAT, a novel software system developed for this purpose [28], [25].
It incorporates some standard astronomical image processing packages, commercial
Sybase DBMS, as well as a number of artificial intelligence (AI) and machine learning
(ML) modules. We measure ∼60 attributes per object on each plate, a subset of
which is used for object classifications.
An essential part of this effort is the extensive CCD calibration program, con-
ducted mainly at the Palomar 60-inch telescope (with approximately 40 nights per
year allocated to this program), and some additional data on the equatorial fields
obtained at CTIO and ESO. The data are calibrated in the Gunn gri system; this
will also make a tie-in with the future Sloan DSS easier. There is a good bandpass
match (and thus small color terms) for the F and N plates and the rand ibands,
but the J plates extend considerably bluer than the gband. These CCD data serve
a dual purpose: for magnitude calibrations, and as the training data sets for our au-
tomated object classifications, described below. Only the best-seeing data are used
for the latter purpose. The typical limiting magnitudes of calibrated plate data are
glim ≈21.5m,rlim ≈20.5m, and ilim ≈19.8m(the CCD images reach ∼1m−2m
deeper).
We obtain a median of 2 CCD fields per plate, in all 3 bands. This is an
unprecedented amount of CCD calibration for a photographically based sky survey.
Yet, it is a minimum necessary in order to achieve the photometric uniformity of
∼5−10% in magnitude zero-points, both across the individual plates, and between
the plates, as demonstrated in our early tests [27]. We also use spatially filtered
measurements of the local sky intensity to determine the “flatfield” corrections for
the plates; these are partly due to the vignetting of the telescope optics, and partly
to the individual plate sensitivity variations. Many projects, e.g., studies of the
large-scale structure, require magnitude accuracy and uniformity at this level of
accuracy or better.
Particular attention was paid to star-galaxy classification. SKICAT uses artificial
induction decision tree techniques [26]. We are now introducing additional neural-
net (NN) based classifications [17], [18]. By using these methods and superior CCD
data to train the AI object classifiers, we are able to achieve classification accuracy of
90% or better (aiming for ∼95% or better) down to ∼1mabove the plate detection
limit; traditional techniques achieve comparable accuracy typically only ∼2mabove
the detection limit. This effectively triples the number of usable objects for most
scientific applications of these data, since in most cases one wants only either stellar
objects or galaxies.
Moreover, since the surface density of stars varies greatly over the sky, while the
surface density of galaxies (corrected for the extinction) remains roughly constant,
the contamination signal (e.g., stars misclassified as galaxies) can vary greatly across
the sky. This is a major, but solvable problem for many projects, and for any sky
survey, yet it is seldom addressed with the care it requires. It is essential to calibrate
and understand well the statistical accuracy of object classifications in order to draw
meaningful conclusions from ostensible “star” or “galaxy” catalogs.
Classification problems are present at all magnitudes, limited by the S/N at
the faint end, by saturation at the bright end, and by crowding at any flux level.
We are thus investigating the use of complementary object classifiers optimized for
different signal levels, optimal combining of object parameters measured in different
bandpasses, etc.
The final result of this effort will be the Palomar-Norris Sky Catalog (PNSC),
which will contain all objects down to the survey limit (equivalent to Blim ∼22m),
with classifications and their statistically estimated accuracy available down to ∼1m
above the detection threshold. The catalog will be confusion limited at low Galactic
latitudes, where the surface density of sources exceedes ∼20 million per plate. We
estimate that the catalog will contain >50 million galaxies, and >2 billion stars,
including ∼105quasars. The expected median redshift for the galaxies is z∼0.2,
reaching out to z∼0.5.
The catalog and its derived data products will be published electronically as soon
as the validation tests are complete, and our funding allows it, probably starting in
the early 1999. We note that the size of the DPOSS data set, in terms of the bits,
numbers of sources, and resolution elements, is ≈1,000 ×the entire IRAS data set,
and is ≈0.1×the anticipated SDSS data set.
3 Some Initial Scientific Applications
This large new data set can fuel numerous scientific studies in the years to come.
While the survey is not very deep by modern large-telescope standards, it does cover
2πsterad, and it does so reasonably uniformly, with a good wavelength baseline.
This enables several types of investigations:
•Optical identifications of sources detected in other wavelengths, e.g., in sur-
veys ranging from radio through IR to x-ray, and the resulting statistical
multi-wavelength studies. This could also lead to detections of astrophysically
interesting objects with extreme flux ratios (e.g., brown dwarfs, ultraluminous
IR galaxies, etc.).
•Statistical studies, such as the measurements of the large-scale structure or
Galactic structure, where the large numbers of sources can enable meaningful
fits of models with a large number of parameters and with small Poissonian
error-bars. The essential requirement here is that the survey calibrations are
uniform and well understood, both in the terms of flux calibration, and object
classification.
•Searches for rare, or even previously unknown types of objects, as defined by
clustering in the parameter space, e.g., objects with unusual colors and/or
morphological structure, etc. This may be the most intriguing type of appli-
cation, as it can lead to some really novel discoveries.
We have already started a number of scientific projects along these lines using
DPOSS data. They represent both our scientific motivation for doing the work, and
also serve as scientific verification tests of the data, helping us catch and correct
processing errors and improve and control the data quality.
Galaxy counts and colors in 3 bands from DPOSS can serve as a baseline for
deeper galaxy counts and a consistency check for galaxy evolution models. Our
initial results [27] show a good agreement with simple models of weak galaxy evo-
lution [13] at low redshifts, z∼0.1−0.3. We are now expanding this work to a
much larger area, to average over the local large-scale structure variations. We are
also planning to cross-correlate our galaxy counts and colors with the new Galactic
extinction map [21]; this should lead to both an improved DPOSS galaxy catalog,
and a better extinction map.
Another anticipated data product is a catalog of ∼5×105brightest galaxies
on the northern sky (down to B∼17m), with automated, objective morphologi-
cal classifications and at least some surface photometry information. We have also
started automated searches for low surface brightness galaxies (in collaboration with
J. Schombert), and for very compact, high surface brightness galaxies, i.e., probing
regions of the parameter space where the selection effects tend to hide galaxies.
Follow-up redshift surveys of these objects are now uder way at Arecibo and Palo-
mar.
Our galaxy catalogs have been used as input for redshift surveys down to ∼21m,
e.g., in the Palomar-Norris survey [22], and several other groups plan to use our
catalogs for their own redshift surveys.
Our preliminary investigation [4] of the galaxy two-point correlation function in
10 fields near the north Galactic pole, where the corrections due to extinction and
object misclassifications are minimal, indicate less power at large scales than was
found in the APM survey [15]. We suspect that the differences may be due in part to
an order of magnitude difference in the amount of CCD calibrations used in the two
surveys, and to the differences in star-galaxy classification accuracy. Just about any
instrumental effect or calibration nonuniformity would add spurious power at large
angular scales. This is a very important test for the scenarios of large scale structure
formation, and we want to be sure that we understand our data fully before drawing
any far-reaching cosmological conclusions. The fact that we have data in 3 bands in
DPOSS should be of great utility here.
We are using DPOSS data to create an objectively and statistically well defined
catalog of rich clusters of galaxies [7]. There are many cosmological uses for such
clusters, and while the subjective nature of the Abell catalog has been widely rec-
ognized as its major limitation, many far-reaching cosmological conclusions have
been drawn from it. There is thus a real need to generate well-defined, objective
catalogs of galaxy clusters and groups, with well understood selection criteria and
completeness.
Our approach is to use color selection of galaxies which are more likely to define
denser environments, followed with the adaptive kernel smoothing of the resulting
surface density map. Statistically significant peaks are then found using a bootstrap
technique. We typically find ∼1−1.5 cluster candidates per deg2, and recover all
of the known Abell clusters in the same fields. Our cluster sample reaches about a
factor of 2 deeper than the Abell sample, and extends to lower richness, thanks to
the greater depth of the plates and the automated selection of overdense regions.
We are also completing a redshift survey of about 100 of the newly selected clusters,
in order to independently quantify our selection and completeness criteria.
We estimate that eventually we will have a catalog of as many as 20,000 rich
clusters of galaxies at high Galactic latitudes in the northern sky, with a median
redshift hzi ∼ 0.2, and perhaps reaching as high as z∼0.5. We plan to use this
cluster catalog for a number of follow-up studies, including cluster clustering, cross-
identifications with x-ray selected samples, etc. We are also conducting detailed
studies of the known galaxy clusters, e.g., their galaxy luminosity functions, mor-
phology, etc.
Another ongoing project is a survey for luminous quasars at z > 4. Quasars at
z > 4 are valuable probes of the early universe, galaxy formation, and the physics
and evolution of the intergalactic medium at large redshifts. The continuum drop
across the Lyαline gives these objects a distinctive color signature: extremely red
in (g−r), yet blue in (r−i), thus standing away from the stellar sequence in the
color space. Traditionally, the major contaminants in this type of work are red
galaxies. Our superior star-galaxy classification leads to a manageable number of
color-selected candidates, and an efficient spectroscopic follow-up. As of mid-1998,
over 40 new z > 4 quasars have been discovered. We make them available to other
astronomers for their studies as soon as the data are reduced.
Our initial results [10], [11], are the best estimates to date of the bright end of
the quasar luminosity function (QLF) at z > 4, and are in excellent agreement with
the fainter QLF evaluated in a completely independent survey [20]. We confirm the
decline in the comoving number density of bright quasars at z > 4. We find intrigu-
ing hints of possible primordial large-scale structure as marked by these quasars [6],
but more data and tests are needed to check this result. Our follow-up projects
include a search for protogalaxies and possible protoclusters in these quasar fields,
a new survey for high-redshift DLA absorbers, etc.
We have also started optical identifications of thousands of radio sources, e.g.,
the VLA FIRST sources [3]. Our preliminary results indicate that there are ∼400
compact radio source ID’s per DPOSS field, and we expect a comparable number
of resolved source ID’s. Eventually, we expect to have >105ID’s for the VLA
FIRST sources, plus many more from other surveys. Our primary goal is to select
radio-loud quasars at z > 4; to date, 2 such objects have been found [23]. We
have also obtained DPOSS IDs for a sample of several hundred flat-spectrum radio
sources, selected from the GB, TEX, and PKS samples. For sources with the spectral
index α > −0.3, which should be mostly quasars, the DPOSS identification rate
approaches 97%. We thus find that at most a few percent of such radio sources may
be completely obscured by dust [12], in contrast to some other claims [24].
In the area of statistical gravitational lensing studies, we have explored the possi-
bility of microlensing of quasars, by looking for a possible excess of foreground galax-
ies near lines of sight to apparently bright, high-zQSOs from flux-limited samples.
We find at most a modest excess. We are also planning to use our galaxy counts to
explore the possible lensing magnification of background AGN by foreground large
scale structure [2].
Much remains to be done in the area of Galactic astronomy. Star counts as a
function of magnitude, color, position, and eventually proper motion as well, fitted
over the entire northern sky at once, would provide unprecedented discrimination
between different Galactic structure models, and constraints on their parameters.
With ∼2×109stars, such studies would present a major advance over similar efforts
done in the past. With the inclusion of IR data such studies can be made much more
powerful. A combination of optical data from DPOSS and IR data from 2MASS
can be very efficient in a search for brown dwarfs, or stars with unusual colors or
variability.
We are now applying the same techniques we use to search for galaxy clusters
to our star catalogs, in an objective and automated search for sparse globulars in
the Galactic halo, tidal disruption tails of former clusters, and possibly even new
dwarf spheroidals in the Local Group (recall the Sextans dwarf, found using similar
data [9]). We are also using DPOSS data to map the tidal cutoff regions of selected
Galactic globulars [30].
This is just a modest sampler of the scientific uses of DPOSS, which are already
under way. We can expect much more in the years to come.
4 Technology Issues: How to Harvest the Abundance of
Multi-Terabyte Astronomical Data Sets
The advent of multi-terabyte astronomical data sets will change profoundly the
face of astronomy. We are facing not only terabytes of raw data (or pixels), but
terabytes of reduced data (or catalogs): archives containing ∼109−1010 objects with
≥102measured parameters each. At this time, several large digital sky surveys are
planning to produce data sets or archives of this size, but such data volumes will
become more common or even standard for individual projects or experiments. This
new wealth of information will enable:
1. New astronomy: Doing statistical astronomy “right”, not limited by poisso-
nian noise or data poverty; searching for rare or new types of astronomical objects
or phenomena; asking new kinds of questions as we characterize the sky as a whole,
rather than work with small samples of objects; and so on. Combining giga-object
surveys from different wavelengths should be especially valuable. This is a quanti-
tatively and qualitatively different enterprise, with a different set of requirements
and goals from the existing (and very useful) astronomical data centers and web
tools like Simbad, NED, SkyView, etc., which are mainly intended to provide data
services for individual objects or small samples of objects or fields.
2. A new style of doing observational astronomy: These vast data sets can
sustain extended data-mining by numerous users, making uses of the data which
were not even conceived by the original data producers. Anyone, anywhere, with a
computer and an internet connection will be able to do some first-rate observational
astronomy, without a need to access expensive or exclusive telescopes or other fa-
cilities, without necessarily being associated with an elite astronomical institution.
This will enable a broader sampling of the community talents, change the sociology
of astronomy, require new kinds of technical skills, and change the way that astro-
nomical research is done. In a way, there is a parallel with analytical theory and
numerical simulations, which ideally can go hand-in-hand, but can be (and usually
are) practiced by different scientists with different skills. So we will have traditional
observers and data miners, and various species in between.
The opportunities are great, but so are the technical challenges. Traditional
astronomical data processing tools are completely inadequate for the tasks at hand.
We, as a community, need to develop or adopt from elsewhere a whole new set of tools
and skills for an effective exploration of multi-terabyte data sets. In some sense, this
is similar to the digital imaging revolution of ∼15 −20 years ago, when astronomy
changed from dealing with small 1-dimensional data sets to dealing effectively with
ever larger digital images or data cubes at any wavelength. Eventually several
major, standard packages evolved, but this was not always a fully rational or optimal
process. Back then we had to learn about things like PSF and isophote fitting, and
now we have to learn about querying and exploring large databases.
These vast data sets will be changing constantly, as new or better calibrations
or reprocessing algorithms are introduced. Different archives will be combined and
recombined, creating new data sets, permanent or transitory. The very concept
of an astronomical catalog is changing, from a fixed set of printed volumes, to a
permanently evolving database which has to be accessed and explored in some non-
trivial manner. The largest astronomical catalogs in the past contained O(105)
objects with O(101) parameters each, were never recalibrated, and required several
thick printed volumes. The new multi-terabyte catalogs will never be printed, but
rather, will reside in network-accessible archives, and would always have to come
along with the software tools necessary for their use and exploration. Specifically,
we need:
1. New data structuring standards and formats. This would enable an easier
interchange and matching of data sets, and commonality of software tools. In addi-
tion to the formulation of computing standards (e.g., relational vs. object-oriented
databases, data exchange formats, standard user interfaces and query languages,
etc.), we should think about new astronomical conventions, which would be opti-
mised for our computing needs, rather than be based on fossilized conventions from
the past; a prime example of this is the Hierarchical Triangulated Mesh for sky par-
titioning [1]. The point is that we should structure the data in a way which would
make the most common types of queries work fast, and enable the new software
tools to sift through the data quickly and effectively.
2. New tools and expertise to navigate and explore these data spaces.
This would include fast ways to query the data in multidimensional parameter spaces
(with tens or hundreds of data dimensions), and the associated novel data visual-
isation problems (including perhaps virtual reality walks through the parameter
spaces). On a more sophisticated level, we need tools for an automated clustering
analysis and classification, both supervised (e.g., where the program is trained on a
set of examples of what to look for) and unsupervised (where the program decides
in some autonomous, statistically justified fashion how many different kinds of ob-
jects are there in the data space, and what they are). We would probably want to
make use of intelligent software agents (a simple example are some of the better web
search engines) which would search through our data parameter space looking for
desired kinds of objects (e.g., clusters of galaxies defined in some objective manner),
or discrepant objects standing away from the bulk of the data points, and so on. We
should make a good use of AI and ML techniques, and other modern data-mining
methods for true machine-assisted discovery.
This long list of desiderata amounts to a new information infrastructure for
astronomy. Virtual observatories are coming, and we should try to design them
right from the start, avoiding costly mistakes and unnecessary replication of effort.
Many of the necessary tools already exist; similar problems are faced by just about
any other data-intensive discipline, and astronomers can benefit greatly in this task
from collaborations with computer scientists. One thing is certain: we are entering
a new era of information-rich astronomy, and we should be ready for the abundance
it has to offer. More is different.
Acknowledgements. The DPOSS/PNSC cataloging effort at Caltech is supported
by a generous grant from the Norris Foundation. Some of the software technology de-
velopment has been supported by the grants from NASA, JPL, and Caltech. SGD also
wishes to acknowledge support from the Bressler Foundation. We are indebted to the
entire POSS-II photographic survey team, the scanning team at STScI, to Palomar Ob-
servatory for generous allocations of telescope time used for CCD calibrations, and to our
numerous collaborators and students whose work is making DPOSS and PNSC become
reality. This work is a part of the CRONARio collaboration. Finally, SGD wishes to thank
the conference organizers for their hospitality, and to acknowledge the valuable guidance
provided by [29].
References
[1] Szalay, A., Brunner, R., et al. 1998, in prep.
[2] Bartelmann, M., & Schneider, P. 1994, A&A, 284, 1
[3] Becker, R., White, R., & Helfand, D. 1995, ApJ, 450, 559
[4] Brainerd, T., de Carvalho, R., & Djorgovski, S. 1995, BAAS, 27, 1364
[5] Djorgovski, S.G., et al. 1998, in New Horizons From Multi-Wavelength Sky Surveys,
eds. B. McLean et al., IAU Symp. #179, p. 424, Dordrecht: Kluwer
[6] Djorgovski, S.G. 1998, in Fundamental Parameters in Cosmology, eds. Y. Giraud-
Heraud et al., Gif sur Yvette: Editions Fronti`eres, in press
[7] Gal, R.R., et al. 1997, BAAS, 29, 1380
[8] Gunn, J.E. & Knapp, G. 1993, in Sky Surveys: Protostars to Protogalaxies, ed. B.T.
Soifer, ASP Conf. Ser. 43, 267
[9] Irwin, M., et al. 1990, MNRAS, 244, 16P
[10] Kennefick, J.D., et al. 1995a, AJ, 110, 78
[11] Kennefick, J.D., Djorgovski, S.G. & de Carvalho, R. 1995b, AJ, 110, 2553
[12] Kollmeier, J., et al. 1998, BAPS, 43, 433
[13] Koo, D., Gronwall, C., & Bruzual, G. 1995, ApJ, 440, L1
[14] Lasker, B. 1994, in Astronomy fromWide-Field Imaging, eds. H. MacGillivray et al.,
IAU Symp. #161, p. 167, Dordrecht: Kluwer
[15] Maddox, S., et al. 1989, MNRAS, 242, 43P
[16] McLean, B., et al. 1998, in New Horizons From Multi-Wavelength Sky Surveys, eds.
B. McLean et al., IAU Symp. #179, p. 431, Dordrecht: Kluwer
[17] Odewahn, S.C., et al. 1992, AJ, 103, 318
[18] Odewahn, S.C. 1997, in Nonlinear Signal and Image Analysis, Ann. N.Y. Acad. Sci.
808, 184
[19] Reid, I.N., et al. 1987, PASP, 103, 661
[20] Schmidt, M., Schneider, D. & Gunn, J. 1995, AJ, 110, 68
[21] Schlegel, D., Finkbeiner, D., & Davis, M. 1998, ApJ, 500, 525
[22] Small, T., Sargent, W.L.W., & Hamilton, D. 1997, ApJS, 111, 1
[23] Stern, D., et al. 1998, BAAS, 30, 902
[24] Webster, R. et al. 1995, Nature, 375, 469
[25] Weir, N., et al. 1994, in Astronomy fromWide-Field Imaging, eds. H. MacGillivray et
al., IAU Symp. #161, p. 205, Dordrecht: Kluwer
[26] Weir, N., Fayyad, U., & Djorgovski, S.G. 1995a, AJ, 109, 2401
[27] Weir, N., Djorgovski, S.G. & Fayyad, U., 1995b, AJ, 110, 1
[28] Weir, N., Fayyad, U., Djorgovski, S.G. & Roden, J. 1995c, PASP, 107, 1243
[29] Wells, P. 1984, The Food Lover’s Guide to Paris, NY: Workman Publ.
[30] Zaggia, S., et al. 1998, in Galactic Halos, ed. D. Zaritsky, ASP Conf. Ser. in press
