The Digitized Second Palomar Observatory Sky Survey (DPOSS) II: Photometric Calibration

Abstract

We present the photometric calibration technique for the Digitized Second Palomar Observatory Sky Survey (DPOSS), used to create seamless catalogs of calibrated objects over large sky areas. After applying a correction for telescope vignetting, the extensive plate overlap regions are used to transform sets of plates onto a common instrumental photometric system. Photometric transformations to the Gunn gri system for each plate, for stars and galaxies, are derived using these contiguous stitched areas and an extensive CCD imaging library obtained for this purpose. We discuss the resulting photometric accuracy, survey depth, and possible systematic errors.

Full text

Page 1

arXiv:astro-ph/0210298v1 14 Oct 2002
The Digitized Second Palomar Observatory Sky Survey (DPOSS) II:
Photometric Calibration
R. R. Gal1, R. R. de Carvalho2,4, S. C. Odewahn3, S. G. Djorgovski, A. Mahabal, R. J. Brunner,
P. A. A. Lopes2
Palomar Observatory, Caltech, MC105-24, Pasadena, CA 91125
ABSTRACT
We present the photometric calibration technique for the Digitized Second Palomar
Observatory Sky Survey (DPOSS), used to create seamless catalogs of calibrated objects
over large sky areas. After applying a correction for telescope vignetting, the extensive
plate overlap regions are used to transform sets of plates onto a common instrumental
photometric system. Photometric transformations to the Gunn gri system for each
plate, for stars and galaxies, are derived using these contiguous stitched areas and an
extensive CCD imaging library obtained for this purpose. We discuss the resulting
photometric accuracy, survey depth, and possible systematic errors.
Subject headings: catalogs — surveys — techniques: photometric
1.Introduction
The answers to many important cosmological questions require large sky surveys, encompassing
hundreds or thousands of square degrees. Large scale structure in the universe, the power spectrum
at large angular scales, and the distribution of stars in our own galaxy all rely on homogeneous,
moderately deep imaging with small (and well understood) systematic errors, as well as enormous
sky coverage. Until very recently, such data was available only from photographic plates, which,
with their large physical size, fill the focal plane of Schmidt telescopes. Unfortunately, these plates
are notoriously difficult to calibrate, with large sensitivity variations among plates and even within
a given plate. Nevertheless, they provide the only current source of all-sky imaging data. Various
projects, such as the Automated Plate Scanner (APS, Odewahn & Aldering 1995, hereafter OA95),
have used digitized scans of the first Palomar Sky Survey, to generate large area catalogs. The more
1Johns Hopkins University, Center for Astrophysical Sciences, 3701 San Martin Dr., Baltimore, MD 21218
Email: rrg@pha.jhu.edu
2Observat´ orio Nacional, Rua Gal. Jos´ e Cristino, 77 - 20921-400, Rio de Janeiro, RJ, Brazil
3Arizona State University, Dept. of Physics & Astronomy, Tempe, AZ 85287
4Currently at Osservatorio Astronomico di Brera, Via Brera 28, 20121 - Milano, Italy

Page 2

– 2 –
recent efforts by the Automatic Plate Measuring team (APM, Maddox, Efstathiou & Sutherland
1990, hereafter MES ) and other groups have utilized the Southern Sky Survey plates or scans of the
more recent Second Palomar Sky Survey for this purpose. Some of these projects have used only
a fraction of the available data, with limited CCD calibration, which makes assessing large-scale
photometric errors difficult, if not impossible. For instance, MES used a scheme to transform all of
their plates onto a common photometric system, but only had CCD calibration data for about one
third of their plates. In the Northern sky, the GSC-II (McLean, Lasker, & Lattanzi 1998) and the
USNO (Monet 1998) both provide catalogs (based on some of the same plate material), but they
lack accurate photometric calibration and are not targeting faint objects. Only the recently begun
Sloan Digital Sky Survey (SDSS, York et al. 2000), will provide improved data in the Northern sky,
and only at galactic latitudes |b| > 40◦.
To address these issues, and provide the community with a large catalog of sources with
accurate photometry, we have produced the Digitized Second Palomar Observatory Sky Survey
(Djorgovski et al. 1999, 2002a). Prior papers have discussed the object detection and classification
techniques, using the SKICAT software package (Weir, Fayyad & Djorgovski 1995; Weir, Fayyad,
Djorgovski & Roden 1995), and a companion paper (Odewahn et al. 2002) presents more recent
Artificial Neural Network (ANN) and Decision Tree (DT) classifiers employed in this survey. Gal
et al. (2000) has presented the vast CCD imaging sample obtained at the Palomar 60-inch for the
purpose of calibrating DPOSS. Here we discuss the techniques used to derive the plate photometric
calibration, and generate a seamless catalog of objects over the high-galactic-latitude (|b| > 30◦)
Northern sky. We briefly review the salient details of the DPOSS and CCD object detection and
photometry schemes in §2. The use of plate overlap regions to transform sets of plates on a uniform
instrumental magnitude system is described in §3. The derivation of the photometric calibration
using CCD data is discussed in §4, including our final photometric errors. We conclude with a
brief discussion of potential applications and future developments in §5. We note that this paper
supersedes the earlier discussion of DPOSS calibration presented in Weir, Djorgovski & Fayyad
(1995).
2. DPOSS and CCD Data
2.1. DPOSS
The POSS-II photographic survey (Reid et al. 1991) covers the entire Northern sky (δ > −3◦)
with 897 overlapping fields (each 6.5◦, with 5◦spacings), and, unlike the old POSS-I, has no gaps
in the coverage. Approximately half of the survey area is covered at least twice in each band, due
to plate overlaps. Plates were taken at the Palomar 48-inch Oschin Schmidt telescope in three
bands: blue-green, IIIa-J + GG395, λeff∼ 480 nm; red, IIIa-F + RG610, λeff∼ 650 nm; and very
near-IR, IV-N + RG9, λeff∼ 850 nm. The bandpasses are illustrated in Figure 1. Typical limiting
magnitudes reached are gJ ∼ 21.5, rF ∼ 21.0, and iN ∼ 20.3, i.e., ∼ 1m− 1.5mdeeper than the

Page 3

– 3 –
POSS-I. The image quality is improved relative to the POSS-I, and is comparable to the southern
photographic sky surveys.
The original survey plates are digitized at STScI, using modified PDS scanners (Lasker et al.
1996). 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). The preliminary
astrometric solutions provided by GSC-II (McLean, Lasker, & Lattanzi 1998) are good to r.m.s.
∼ 0.5′′, and are in the process of being improved substantially.
An extensive effort, centered at Caltech, and with sites in Italy (Osservatorio Astronomico
di Capodimonte and Osservatorio Astronomico di Roma) and Brazil (Observat´ orio Nacional), has
resulted in the processing, calibration, and cataloging of nearly all scans at |b| > 10◦, with the
detection of all objects down to the survey limit, and star/galaxy classifications accurate to 90% or
better down to ∼ 1mabove the detection limit. Object detection and photometry is performed by
SKICAT, a novel software system developed for this purpose (Weir, Djorgovski & Fayyad 1995). It
incorporates some standard astronomical image processing packages, commercial Sybase database,
as well as a number of artificial intelligence (AI) and machine learning (ML) modules. We measure
∼ 60 attributes per object on each plate in each filter. A subset of these are used for classification,
as described in Odewahn et al. (2002).
2.1.1. Vignetting Correction
Large area detectors have always faced the problem of nonuniform illumination across the
detector. The resulting effect, primarily due to placing a square detector in a telescope with a round
aperture, is called vignetting. The vignetting pattern must be removed if one is to use the data
at larger radii from the plate center. A simple (but incorrect) approach is to assume a circularly
symmetric function and apply it to the data. However, though largely radially symmetric, the
pattern has finer structure, due to the bending of the plates, the filters used, etc. Here we describe
the procedure we have used to remove the vignetting pattern from the DPOSS data. This procedure
is performed separately for the three filters used.
Basically, we stack and combine multiple DPOSS plate images to obtain a “master vignetting
correction” field, in a manner analogous to generating a flat field for CCD data. Because this
procedure was developed after most plates were processed into catalogs, the vignetting correction
image is used to derive corrections to the object magnitudes post-processing, as well as correct the
actual pixel data. The following procedure was followed, using one hundred plates for each band:
• Bin each plate image 8 × 8, resulting in a 2880 × 2880 pixel image.
• Normalize each image by the median of its central 720 × 720 region.
• Stack the resulting images. This results in a 2800 × 2800 × 100 array.

Page 4

– 4 –
• For each pixel in the array, discard the ten brightest and ten faintest members, and obtain
the median of the remaining eighty. This produces a single 2880 × 2880 image, where each
pixel is now a clipped median of the one hundred contributing images.
• Normalize the resulting image such that the maximum in the central 720 × 720 pixel region
is 1.0.
Thus, all corrections are relative to the plate center, which is the best exposed part of the
plate. Contours of the master vignetting correction image in the F and J bands are shown in
Figure 2; the N map is similar. Further details of this procedure can be found in Mahabal et al.
(2002).
Nearly all plates at |b| > 10◦have been processed into catalogs.
processing is the Palomar-Norris Sky Catalog (PNSC), expected to contain ∼ 50×106galaxies and
> 2 × 106stars.
The final result of this
2.2. CCD Data
Details of the acquisition, processing, and calibration of the CCD data used to calibrate DPOSS
are provided in Gal et al. (2000). Here we provide a brief overview of these data.
We have imaged nearly 900 independent CCD fields for DPOSS calibration. The imaging
targets are selected from the list of northern Abell Clusters (Abell, Corwin & Olowin 1989), with
priority given to the richest clusters closest to the field centers. This strategy increase the number
of objects for both star/galaxy separation and photometric calibration. For those plates with no
Abell clusters, we image the plate center and two other pointings within the plate. All data are
obtained at the Palomar Observatory 60-inch telescope with the CCD imaging cameras placed at
Cassegrain focus. The Gunn gri filters (Thuan & Gunn 1976; Wade, Hoessel, Elias & Huchra 1979;
Schneider, Gunn, & Hoessel 1983) are used, which are well matched to the DPOSS bandpasses (see
Figure 1). Data are taken only on photometric nights with seeing better than 2′′. The mean seeing
for our data is ∼ 1.5′′, with limiting magnitudes of approximately mlim= 21.5m,21.5m,21.2min g,
r, and i, respectively ( ∼ 0.5m− 1mdeeper than the plate detection limits). For every night that
is deemed photometric, we observe a set of Gunn standards (Kent 1985).
Data are processed in the usual way using the IRAF data reduction package (Tody 1986).
All frames are bias subtracted and flat fielded, using a combination of dome and twilight flats.
Additionally, a dark sky flat field correction is generated by median filtering all of the unregistered
target frames in each filter on a given night. Finally, we apply a fringe correction to the i images.
The photometric standard stars observed on each night are photometered using the apphot package
in IRAF, and the resulting collection of between five and twelve measured instrumental magnitudes
used to determine the zero-point offset, airmass term, and color term in each filter.

Page 5

– 5 –
Object detection on the target frames is performed using the FOCAS package (Jarvis & Tyson
1981; Valdes 1982). The g, r, and i frames are processed independently, using detection parameters
of 2.5σ per pixel, a 25 pixel minimum area, and a sky value estimated individually for each image.
Object classification is also performed by FOCAS, and the classifications visually inspected. Bright
objects with incorrect classifications (usually due to saturated pixels) are corrected by hand. The
photometric coefficients (zero point, color term, and airmass term) derived from the standard star
observations are used to determine object magnitudes.
3. Plate Overlap Analysis
A great advantage of POSS-II over earlier surveys is the extensive overlap between neighboring
plates. Adjacent plates overlap each other by 1.5◦along edges covering 6.5◦, providing ∼ 92◦of
duplicated area for each plate pair. This compares favorably to the APM survey, whose scans
typically had 1◦overlaps (Maddox, Efstathiou & Sutherland 1990); with the strong effect of vi-
gnetting at plate edges, the extra 0.5◦of overlap in DPOSS is significant. These areas each contain
thousands of objects which can be used to derive the transformation of one plate’s instrumental
magnitude system to that of its neighbor. These transformations can be propagated across multiple
boundaries to tie a contiguous set of plates onto a common photometric system.
In practice, this instrumental magnitude transformation must be mapped independently for
stars and galaxies, especially for brighter magnitudes (minst < 19).
intensity (D-I) transformation has been linearized using the densitometry spots on each plate, stars
and galaxies nevertheless display distinct photometric behaviors. This is because the brighter pixels
in stellar images populate the highly nonlinear part of the D-I relation, where small errors in the
polynomial fit (mostly due to the small number of points characterizing the “shoulder” of the curve)
result in larger variations in the derived intensity. Pixels within galaxies occupy the more linear
part of the D-I transformation, resulting in more stable and linear photometry.
Although the density-to-
Additionally, propagating the photometric transformations across a large number of plates (as
would be necessary to knit the whole sky) can result in the unacceptable accumulation of (possibly
systematic) errors. This issue was addressed by Groth & Peebles (1986), and their technique utilized
by MES to estimate the extent of such errors in the APM. A similar approach was also taken by
OA95 in an analysis of galaxy counts at the North Galactic Pole, but that study was limited by
the small overlap regions and poorer quality of POSS-I. An early attempt to address the problem
of plate inhomogeneities in POSS-II is discussed in Picard (1991).
To avoid this issue entirely, we have elected to calibrate each plate using a sliding boxcar
technique, described below. We transform each plate’s eight surrounding neighbors onto the in-
strumental system of the center plate, match all the CCD data available for the resulting nine-plate
area, and derive the photometric calibration for the center plate. The individually calibrated plates
can then be quilted with no additional offsets because they are already on a calibrated, rather