A Unified Catalog of Radio Objects Detected by NVSS, FIRST, WENSS, GB6, and SDSS

1.6 MB·English

by  Amy E. Kimball

A Unified Catalog of Radio Objects Detected by NVSS, FIRST,
WENSS, GB6, and SDSS
Amy E. Kimball1
9
0 Zˇeljko Ivezi´c1
0
2

n
a
J
ABSTRACT

6
1 We construct a catalog of radio sources detected by GB6 (6cm), FIRST and NVSS (20cm),
and WENSS (92cm) radio surveys, and the SDSS optical survey. The 2.7million entries in the

]
publicly-available master catalog are comprised of the closest three FIRST to NVSS matches
h
′′
p (within 30 ) andvice-versa,andunmatched sourcesfromeachsurvey. Entries aresupplemented
- by data from the other radio and optical surveys, where available. All objects with even a small
o
probability of physical association are included, such that catalog users can easily implement
r
t theirownselectioncriteriafordataanalysis. Weperformdataanalysisinthe∼3000deg2 region
s
a of sky where the surveys overlap,which contains 140,000 NVSS-FIRST sources, of which 64,000
[ are detected by WENSS and 12,000 by GB6. About one third of each sample is detected by

2 SDSS. An automated classification method based on 20cm fluxes defines three radio morphol-
v ogy classes: complex, resolved, and compact. Radio color-magnitude-morphology diagrams for
0 these classes show structure suggestive of strong underlying physical correlations. Complex and
5
resolved sources tend to have a steep spectral slope (α ∼ −0.8) that is nearly constant from 6
6 ′′
to 92cm, while the compact class (unresolved on ∼ 5 scale by FIRST) contains a significant
0
. number of flat-spectrum (α ∼ 0) sources. In the optically-detected sample, quasars dominate
6
the flat-spectrum compact sources while steep-spectrum and resolved objects contain substan-
0
tial numbers of both quasars and galaxies. Differential radio counts of quasars and galaxies are
8
0 similar at bright flux levels (> 100mJy at 20cm), while at fainter levels the quasar counts are
: significantly reduced below galaxy counts. The optically-undetected sample is strongly biased
v
towardsteep-spectrumsources. InsamplesofquasarsandgalaxieswithSDSSspectra(2,885and
i
X
1,288 respectively), we find that radio properties such as spectral slope, morphology, and radio
r loudness are correlated with optical color and luminosity.
a
Subject headings: catalogs — galaxies: active — radio continuum: general
1. INTRODUCTION dio galaxies and quasars (Urry & Padovani 1995;
Jackson & Wall 1999) attempts to explain much
Quasars and powerful radio galaxies dominate
ofthis richvarietyofobservationaldataasarising
the observed counts of continuum radio sources
from essentially the same anisotropic processes
above milliJansky flux levels, and display spec-
whichappear verydifferent to us because ofvary-
tacular morphological variety that is correlated
ing viewing angles to the radio jets. This con-
with other properties such as spectral slope and
jecture has fundamental implications for our un-
luminosity. The unification paradigm for ra-
derstanding of quasars and galaxies, but for a
strong test of the unification paradigm one essen-
1Department of Astronomy, University of Washing- tially needs a large statistical sample with well-
ton, Box 351580, Seattle, WA 98195, USA; akim-
controlled selection criteria and robust estimates
[email protected],[email protected]
1
of the source morphology, as well as appropriate dominate radio counts down to 5mJy at 20cm.
models to interpret the data. The more recent work of Mauch & Sadler (2007)
Statistical studies of radio emission from ex- includes the largest radio-selected galaxy sample
tragalactic sources are entering a new era, result- available from a single radio survey, combining
ingfromthe availabilityoflargeskyradiosurveys the NVSS with the 6 degree Field Galaxy Survey
that are sensitive to milliJansky flux levels (e.g., (Jones et al. 2004); they confirm that radio-loud
Becker et al.1995;Condon et al.1998;De Breuck AGN and star-forming galaxies have quite differ-
2000). The catalogs based on these surveys con- ent distributions in the plane of radio power ver-
tainlargenumbersofsources,havehighcomplete- sus absolute K band (infrared) magnitude. The
ness and low contamination, and are available in study of radio and optical properties of quasars
digitalform. Thewidewavelengthregionspanned was extended by de Vries et al. (2006) who found
by these surveys, from 6cm for GB6 to 92cm for that 10% of SDSS quasars have detectable radio
WENSS, and detailed morphological information cores at 1.4GHz (> 0.75mJy), and 1.7% have
at20cmprovidedbyFIRSTandNVSS,allowsig- double-lobedmorphology,i.e., areassociatedwith
nificant quantitative and qualitative advances in multiple FIRST components.
studies of radio sources. In addition, the optical In this paper, we describe the construction of
catalog obtained by the Sloan Digital Sky Sur- a unified catalog of radio sources detected in the
vey (SDSS; York et al.2000) can be usedto sepa- 6cm GB6, 20cm FIRST and NVSS, and 92cm
rate quasarsfromgalaxies,andthe redshifts mea- WENSS radio surveys, and the SDSS optical sur-
suredbySDSSallowacomprehensivestudyofthe vey (Data Release 6, hereafter DR6). We began
optical-radio correlationfor quasars and galaxies. by merging the FIRST and NVSS surveys into
FIRST and NVSS, conducted separately at a single catalog containing over 2million sources
the Very Large Array (VLA), were the first ra- detected by at least one survey; about 500,000
dio surveys with sufficiently high angular resolu- sources are detected by both FIRST and NVSS.
tion to allow unambiguous matching with deep Where available, we supplement this survey with
optical surveys, providing identifications for a data obtained by the GB6 and WENSS surveys
large number of radio sources. The two sur- that enable the computation of radio spectral
veys have the same radio frequency, but FIRST slopes; about 30,000 sources are detected by all
goes slightly deeper with higher resolution and four radio surveys. The radio sources were also
smaller sky coverage (§2). Machalski & Condon cross-correlated with the optical SDSS catalogs;
(1999) and Sadler et al. (1999, 2002) measured nearly 92,000FIRST sources have an SDSS coun-
′′
the radio luminosity function of radio-loud active terpart within 2 .
galactic nuclei (AGN) and star-forming galax- The radio and optical surveys chosen for in-
ies by cross-correlating the NVSS with spectro- clusion in the unified catalog primarily cover the
scopic galaxy surveys. Magliocchetti et al. (2002) northern celestial hemisphere. We have opted to
matched FIRST to the 2 degree Field Galaxy usethesesurveysinordertotakeadvantageofthe
Redshift Survey (Colless et al. 2001), using spec- high astrometric accuracy of FIRST, which, de-
tra to classify galaxies as “classical” radio galax- signedtocoverthesameregionofskyastheSDSS,
ies, starburst and late-type galaxies, and Seyfert is limited to the northern Galactic cap. Several
galaxies. Ivezi´c et al. (2002) cross-correlated the recent large-area radio surveys covering a wide
FIRST surveywith the SDSSphotometricsurvey, rangeoffrequenciesarealsoavailableinthesouth.
resulting in a much larger sample of optical iden- These include the Sydney University Molonglo
tifications (about one third of FIRST sources are SkySurveyat36cm(SUMSS;Mauch et al.2003),
matched to an SDSS source, and 0.16% of SDSS the Parkes-MIT-NRAO survey at 6cm (PMN;
sources are matched to a FIRST source, with a Gregory et al. 1994), and the Australia Telescope
′′
matching radius of 2 ), with galaxies outnum- 20-GHz (1.5cm) survey (AT20G; Massardi et al.
bering quasars 5:1, and a negligible fraction of 2008).
radiostarsinthesample. Best et al.(2005)cross- Themainadvantagesoftheunifiedcatalogpre-
correlatedtheSDSSspectroscopicgalaxiessample sented in this paper are the multi-wavelength ra-
withbothFIRSTandNVSS,andfoundthatAGN dio data (92cm and 6cm in addition to 20cm),
2
and the increased number of optical identifica- spectra, redshift and luminosity. We also discuss
tions from the SDSS DR6. By combining these some catalog applications. We summarize our re-
five extensive surveys, we have assembled precise sults, and discuss the suitability of the catalogfor
astrometric measurements, flux at multiple wave- comparison with radio evolution models, in §5.
lengths,spectralindex,morphologicalinformation
(size, shape, and orientation of resolved objects), 2. SOURCE SURVEYS
and optical identifications into a single compre-
Beforeproceedingwithadescriptionofmerging
hensive catalog of radio objects. At optical wave-
procedure for the FIRST and NVSS catalogs, we
lengths, the SDSS provides colors and classifica-
briefly describe each survey used in creating the
tion(i.e.,intoquasarsandgalaxies)forthe≈30%
multi-wavelength radio catalog, including the sky
of sources detected optically. The unified cata-
coverage, wavelength, astrometric accuracy, and
log is primarily a resource of low-redshift (z . 2)
fluxlimit(seeTable1forsummary). Skycoverage
quasars and radio galaxies with AGN, but may
ofeachsurveyisshowninFigure1. The regionof
also prove to be an important source of rarer ob-
sky observed by all of the contributing surveys is
jectssuchasradiostars,high-redshiftquasars,and
indicated by the cyan solid line.
high-redshift galaxies. The area observed by all
five surveys is nearly 3,000deg2. The unified cat-
2.1. Radio surveys
alogprovidescomprehensivemulti-wavelengthob-
servationsatgreaterdepthandforalargernumber 2.1.1. FIRST
ofsourcesthananypreviouslyavailablecatalogof
The FIRST survey (Faint Images of the Ra-
radio objects.
dio Sky at Twenty centimeters; Becker et al.
The limits in sky coverage of the catalog are
1995) used the VLA to observe the sky at 20cm
defined by the FIRST and NVSS sky coverage
′′
(1.4GHz) with a beam size of 5.4 and an rms
(Fig.1). Theunifiedcatalogthusincludessources
sensitivity of about 0.15mJy beam−1. Designed
in the Galactic plane which are imaged by the
to cover the same region of the sky as the SDSS,
NVSS.Galacticsourcesmustbestudiedwithcare:
FIRST observed 9,000deg2 at the north Galactic
due to the nature of interferometry, radio surveys
◦
cap and a smaller ∼2.5 wide strip along the Ce-
(depending on their angular resolution) generally
lestial Equator. It is 95% complete to 2mJy and
do a poor job of imaging highly-extended sources
80% complete to the survey limit of 1mJy. The
such as Galactic HII regions and supernova rem-
survey contains over 800,000unique sources, with
nants, whose angular sizes often reach several ar-
astrometric uncertainty of . 1′′. FIRST includes
cminormore. Theanalysispresentedinthispaper
twomeasuresof20cmcontinuumfluxdensity: the
is limited to the sky covered by FIRST, which is
greater than 30◦ from the Galactic plane. peak value,Fpeak,andthe integratedflux density,
F ,measuredbyfitting atwo-dimensionalGaus-
int
In this paper, we discuss scientific applications
sian to co-added images of each source.
of the unified radio catalog and present a pre-
liminary data analysis. In a companion paper 2.1.2. NVSS
(A. Kimball et al. in preparation), we will ex-
pand upon and refine this analysis by comparing TheNVSS(NRAO-VLASkySurvey;Condon et al.
the sourcedistributionin radiomorphology,radio 1998) was also carried out using the VLA radio
color, and optical classification space to the mod- telescope to observe the sky at 20cm (1.4GHz),
els of Barai & Wiita (2006, 2007). the same wavelength as FIRST. However, the
NVSS used a different antenna configuration, re-
The remainder of the paper is laid out as fol-
sultinginalowerspatialresolution(45′′ beam−1).
lows. In§2,wedescribethesurveysusedtocreate
Lower resolution radio surveys provide more ac-
theunifiedradiocatalog. In§3wediscussthecre-
curate flux measurements for extended sources,
ation of the catalog, including completeness and
where high-resolution surveys can miss a signifi-
efficiency of the matching algorithms. In §4 we
cantfractionoftheflux. Theastrometricaccuracy
present a preliminary analysis of the radio source
′′
ranges from 1 for the brightest NVSS detections
distribution according to radio morphology, radio
′′
toabout7 forthefaintestdetections. Thesurvey
color, optical identification, and, for sources with
3
covers the entire sky north of δ = −40◦ and con- 2.2.1. The SDSS Photometric Survey
tains over 1.8million unique detections brighter
The SDSS photometric survey contains flux
than 2.5mJy.
densities of detected objects measured nearly
For the analysis of this paper, we adopt in-
simultaneously in five wavelength bands (u, g,
tegrated NVSS flux densities rather than the
r, i, and z; Fukugita et al. 1996) with effec-
peak flux densities reported in the NVSS cata-
tive wavelengths of 3551, 4686, 6165, 7481, and
log. Condon et al. (1998) provides formulas for 8931˚A (Gunn et al. 1998). The catalog is 95%
converting peak flux densities to integrated flux
complete for point sources to limiting AB mag-
densities, as well as formulas for errors in NVSS
nitudes of 22.0, 22.2, 22.2, 21.3, and 20.5 respec-
measuredandcalculatedvalues. Theunifiedcata- ′′
tively. Typicalseeing is about 1.4 and positional
logincludes the integratedNVSS flux density, the ′′
uncertainty is less than about 0.1 (rms per co-
corrected peak flux density, and the deconvolved
ordinate for sources with r < 20.5, Pier et al.
major and minor axis sizes.
2003). The photometry is repeatable to 0.02 mag
(rms for sources not limited by photon statistics,
2.1.3. WENSS
Ivezi´c et al. 2003) and with a zeropoint uncer-
WENSS (Westerbork Northern Sky Survey; tainty of ∼0.02-0.03 mag (Ivezi´c et al. 2004a). A
Rengelink et al.1997)isa92cm(325MHz)survey compendiumofothertechnicaldetailsaboutSDSS
completed with the Westerbork Synthesis Radio can be found on the SDSS web site1, which also
Telescope. It maps the radio sky north of δ =29◦ provides an interface for public data access.
to a limiting flux of 18mJy, with a beam size of OpticalmagnitudesinthispaperrefertoSDSS
′′ ′′
54 ×54 cosec(δ). The complete survey contains model magnitudes, measured using a weighting
almost 220,000 sources, with a positional uncer- function determined from the object’s r band
tainty of . 1.5′′ for bright sources and . 5′′ for image (see Stoughton et al. 2002). The weight-
faint sources. ing function represents the better fitting of an
exponential profile and a de Vaucouleurs pro-
2.1.4. GB6 file. The chosen model is then used to deter-
mine the magnitude in all five bands. We cor-
TheGB6surveyat4.85GHz(GreenBank6cm
rected all magnitudes for Galactic extinction fol-
survey;Gregory et al.1996)wasexecutedwiththe
lowingSchlegel et al.(1998). When selecting can-
(now defunct) 91m Green Bank telescope in 1986
didatematchesfromtheSDSS,werequiredunique
November and 1987 October. Data from both
sourcesthatarebrighterthanr =22.2orbrighter
epochs were assembled into a survey covering the
◦ ◦ than z =21.2.
0 < δ < 75 sky down to a limiting flux of
18mJy, with 3.5′ resolution. GB6 contains over The morphologicalinformation from SDSS im-
75,000 sources, and has a positional uncertainty ages allows reliable star-galaxy separation to r ∼
of about 10′′ at the bright end and about 50′′ for 21.5(Lupton et al.2002;Scranton et al.2002). In
faint sources. brief, sources are classified as resolved or unre-
solved according to a measure of light concentra-
2.2. Optical Surveys: SDSS tion thatrepresentshow wellthe flux distribution
matches that of a point source (Stoughton et al.
Weusephotometricandspectroscopiccoverage
2002). As sources that emit strongly in the radio
from the sixth data release (DR6) of the Sloan
are almost exclusively extra-galactic at the high
Digital Sky Survey (SDSS; see York et al. 2000;
latitudes observed by the SDSS (|b| > 30), this
Stoughton et al. 2002; Adelman-McCarthy et al.
classification effectively divides radio sources into
2007, and references therein). The survey, not
“galaxies” (resolved) and “quasars” (unresolved).
yet finished but nearing completion, will eventu-
The optically-unresolvedsourcesmay alsoinclude
allycover10,000deg2 inthenortherngalacticcap
a small number of galaxies that are unresolved in
andasmallerregiononthecelestialequator. DR6
the SDSSimages,aswellasasmallfractionofra-
coversroughly9,600deg2,andcontainsphotomet-
diostars. Fortheremainderofthispaper,werefer
ric observations for 287million unique objects, as
well as spectra for more than 1million sources. 1http://www.sdss.org
4
to the two sets of photometric optical sources (as like colors. In fact, nearly 30%of the targets turn
opposedtothosewithavailablespectra,seebelow) out to be stars.
as “galaxies” and “optical point sources”. To distinguish the spectroscopic SDSS sample
from the photometric SDSS sample, we refer to
2.2.2. The SDSS Spectroscopic Survey
sources from the former as “spectroscopic galax-
A subset of photometric sources are chosen for ies” and “spectroscopic quasars”.
spectroscopicobservationaccordingtotheSDSS’s
spectral target selection algorithms. Targeted ex- 3. THE UNIFIED CATALOG
tragalacticsourcesincludetheflux-limited“main”
The cross-identification of radio surveys at dif-
galaxy sample (r < 17.77; Strauss et al. 2002),
ferentwavelengths,and with differentresolutions,
the luminous red galaxy sample (Eisenstein et al.
is not a straightforward task (Becker et al. 1995;
2001), and quasars (Richards et al. 2002). DR6
Ivezi´c et al. 2002; de Vries et al. 2006; Lu et al.
contains spectra for about 100,000 quasars and
2007). However, the accurate astrometry of
790,000galaxies. Thespectralwavelengthrangeis
FIRST allows simple positional matching to the
3800−9200˚A, with a resolution of 1800 and red-
shift accuracy of 30kms−1 (estimated from the otherradiosurveys,andtotheopticalSDSS,with
high completeness and low contamination. Cata-
main galaxy sample).
log entries are defined by a detection in at least
The main galaxy sample includes nearly all
oneofthe two20cmsurveys(FIRST andNVSS).
(∼ 99%) galaxies brighter than r < 17.77, result-
Where available, the 20cm data is supplemented
ing in a sky density of ≈ 90deg−2. Some targets
with 92cm and 6cm radio data, from WENSS
are rejected on the basis of low surface bright-
and GB6, and with optical observations from the
ness (whereredshifts become unreliableortargets
SDSS.
arespurious)orhighflux (whichcancontaminate
In the remainder of this section, we describe
neighboring fibers). Due to the physical thickness
′′ the matching technique used to create the unified
ofthe spectralfibers, two galaxiescloserthan 55
radio catalog. The complicated procedure out-
cannot be observed at the same time, although
linedhereinis intendedto include allobjects with
overlapping spectral plates allow some initially-
even a small chance of being physically-real asso-
skipped galaxies to be picked up later. A sec-
ciations. Therefore, catalog users are not limited
ond galaxy target algorithm selects luminous red
to the matching techniques or matching distances
galaxies(LRGs; Eisenstein et al. 2001), which are
used by the authors in the subsequent analysis in
typicallythebrightestmembersofgalaxyclusters.
this paper. The generous matching radii used to
LRGtargetsareselectedprimarilybycolor,based
create the catalog can easily be restricted by fu-
on knownLRG spectra at different redshifts. The
ture users using the catalog parameters detailed
resulting sample is typically more luminous and
in Appendix A.
much redder than the main galaxy sample, and
extends to fainter apparent magnitudes than the
3.1. Defining catalog entries from the
main sample.
FIRST and NVSS surveys
Thequasartargetselectionalgorithm(Richards et al.
2002)selects targetsfromunresolvedobjects with FIRST and NVSS both observed the sky at
i<19.1andcolorssimilartoredshift.3quasars, 20cm (1400MHz). Of the four radio surveys
unresolved objects with i < 20.2 and colors sim- incorporated into the unified catalog, these two
ilar to higher redshift quasars, and unresolved have the faintest flux limits and the highest spa-
sources within 2′′ of a FIRST source (i < 19.1). tial resolution, with FIRST having even higher
The completeness of the sample is ≈ 95% and resolution than NVSS. Radio surveys necessar-
the selection efficiency is ≈ 66%. Contamination ily involve a trade-off between high resolution,
is much higher than for the galaxy sample: the which allows for accurate determination of posi-
latter consists of resolved sources, which are thus tions, and low resolution, which allows for the de-
nearlyalwaysgalaxies,whereasthequasarsample tectionoflowsurfacebrightnesssourcesandcom-
iscontaminatedbynon-quasarpointsources,such plete flux measurements for extended sources. By
as distant galaxies or Galactic stars with quasar- combining high-resolution FIRST with the lower-
5
resolutionNVSS, it is possible to achievethe best nate the sky position of each catalog entry, when
of both options. FIRST provides accurate posi- possible. For NVSS sources without a FIRST
tional measurements, and the NVSS provides ac- match, we retainthe NVSS coordinates. The des-
curate flux measurements for extended and low- ignated position is then used when searching for
surface brightnesssources,where FIRST underes- counterparts in the other surveys.
timatestheradioflux(Becker et al.1995;Lu et al. To match the FIRST and NVSS surveys, we
2007). use a matching radius of 30′′. We use a larger ra-
′′
dius of 120 to match to WENSS and GB6, since
3.2. Matching algorithm for the radio sur-
they both have lower positional accuracies. We
veys
also recordthe total number of WENSS and GB6
′′
matches found within 120 . All of the match-
Duetodifferentspatialresolutionandfaintlim-
ing radiiused to create the catalogare generously
its, combined with the complex morphology of
large, to ensure that the majority of physically-
radio sources, the positional matching of FIRST
real matches are included. To select smaller,
and NVSS sources cannot be done in an a pri-
cleaner samples for analysis, the catalog user can
ori correct way. For this reason, we adopt a
choose matches based on distance, effectively ap-
flexible method that enables later sample refine-
plyingasmallermatchingradius. Thecontamina-
ment after the initial positional association. The
tion and completeness as a function of matching
unified catalog contains the three closest FIRST
′′ radius are discussed in §3.4.
matches to an NVSS source within 30 , and the
three closest NVSS matches to a FIRST source The surveys used in the creation of the unified
within 30′′. For each source, we also record the catalogarethemselvescatalogsofindividualradio
total number of matches found within 5′′, 10′′, components. Thus, large multi-component physi-
30′′, and 120′′, which will allow users to inves- cal sources can be resolved into separate detec-
tigate radio properties as a function of environ- tions in the high angular-resolution surveys, par-
′′
ment density. The complete catalog contains over ticularly in FIRST (5 resolution), but also in
′′
2.7million entries, including FIRST-NVSS asso- NVSSandWENSS(∼50 resolution). Best et al.
ciations, isolated2 FIRST or NVSS sources, and (2005, hereafter B05)developed a cleansample of
NVSS sources that lie outside the FIRST survey double-lobed radio-loud galaxies by matching ra-
coverage. Becausewe matchFIRST toNVSS and dio components to spectroscopic SDSS galaxies.
then NVSS to FIRST, there are necessarily du- B05 used the optical core position and the opti-
plicate catalog entries: for example, a very close cal galaxy alignment to eliminate unlikely lobe-
match will appear once as a FIRST to NVSS configurations, based on lobe opening angles and
matchandonceasaNVSStoFIRSTmatch. How- distances. The complexity of a matching algo-
ever,the catalogincludes parametersto easily ex- rithm based only on radio components, as pre-
tract specific data samples, including the elimina- sented in this paper, is necessarily much greater
tion of any duplicates. Similarly, the processing becausetheopticalpositionofthecoreisunknown
flags can be used to treat cases such as distinct for the majority of sources. We therefore settle
FIRST sourcesmatchedtothe sameNVSSsource onthe matching schemeoutlined above,which al-
(and vice-versa, although the latter case is rare lowssampleselectionwiththeuser’sownpreferred
due to the lower spatial resolution and brighter matching criteria, without the need to repeat the
limit of NVSS). For further details, we refer the workthathasgoneintodevelopingtheunifiedcat-
reader to Appendix A. alog presented here.
Measured FIRST and NVSS sky positions are
3.3. Matching algorithm for the SDSS
rarely exact even for the same source. As shown
inFigure2,FIRSThasmoreaccurateastrometric We correlate the matched radio sources with
measurements than NVSS, due to its higher spa- the SDSS photometric survey, including spectro-
tial resolution. We therefore use FIRST to desig- scopic data when available, using a matching ra-
′′
dius of 60 . In addition to the nearest neigh-
2Isolated refers here to sources withno neighbors from the bor, the catalog includes the brightest neighbor
other20cmsurveywithin30′′. ′′ ′′ ′′ ′′
within a pre-defined radius (3 , 10 , 30 , or 60 ;
6
see Appendix A for details). The total number of SDSS (photometric) surveys. For comparison, it
matcheswithinthesamepre-definedradiusisalso also shows the estimated level of backgroundcon-
recorded, indicating optical source sky density at tamination. The nearest-neighbor distributions
each entry’s location. peak at small distances, where associations are
We note that some lobe-dominated radio likely to be real, then decrease sharply; the width
sourcesmaybemissedwhenpositionallymatching of the peaks depends on the astrometric accu-
SDSSandFIRSTsourcepositions. Whileeachra- racy of the corresponding surveys. The distribu-
dio lobe will be included individually in the radio tions slowly rise again at large distances due to
sample, it will be excluded from the optically- increased background contamination. Efficiency
matchedsubsetforlobe-coredistanceslargerthan and completeness are estimated using a model fit
the SDSS matching radius. Lu et al. (2007) esti- to the nearest neighbor distributions. Estimated
mated that about 8.1% of radio quasars do not valuesofcompletenessandefficiencybasedonthe
show a radio core within 2′′ of their optical posi- modelfittingarelistedinTable2. Forthesamples
tion, and thus would be missed by this algorithm. discussedhere,typicalvaluesare>90%complete-
However, Ivezi´c et al. (2002) and de Vries et al. ness and >80% efficiency.
(2006) found fewer than 5% and 2%, respectively, As discussed above, the unified catalog in-
of quasarsare double-lobedin FIRST, using opti- cludes individual radio components, and does not
cal core positions without assuming radio emis- correctly handle NVSS multi-component sources.
sion in the core. Because NVSS has a much Two lobes from a very large (&100′′) source may
larger spatial resolution than FIRST, even fewer appearastwo detectionsin NVSS andthus in the
double-lobed sources can be expected in NVSS. unifiedcatalogaswell. ThevaluesinTable2show
B05 found 6% of their spectroscopic SDSS galax- that when matching up individual radio compo-
ies to be double-lobed in NVSS. Note that while nents between different surveys, the completeness
the percentage is larger for galaxies, as expected is well over 90%. As stated in §3.3, investigations
from orientation effects, this value is an upper using optical and radio surveys suggest that only
limit for radio-optical samples using photometric afewpercentofradiosourcesaredouble-lobedin
optical data, as the spectroscopic SDSS sample FIRST; much smallerpercentagesare expected in
is comprised of the nearest sources, which thus the NVSS. Indeed, B05 note that the NVSS reso-
have large angular size. Since the effect is not lutionislargeenoughthat∼99%ofradiosources
overwhelming, the unified catalog only includes are contained in a single NVSS component. Al-
direct positional matches between SDSS and ra- though the missing double-lobed sources are not
dio (FIRST or NVSS) catalogs. However, catalog addressedfurtherinthispaper,wetackletheissue
users can easily repeat the procedures for finding inthecompanionpaper,wherewethoroughlydis-
double-lobed sources developed by Ivezi´c et al. cuss the classification of a WENSS-NVSS-FIRST
(2002); de Vries et al. (2006); Best et al. (2005), subsample (see §5.4.1 for discussion). It will in-
which would involve matching external optical clude the missing double-lobed sources, found by
samples to the radio catalog sources. determining the complete NVSS-FIRST environ-
ment around each WENSS object, requiring ex-
3.4. Completenessandefficiencyofmatched tensive visual analysis.
samples
3.5. FIRST matching statistics
A side effect of using large matching radii is
increased contamination by coincidental “line-of- Ofthefourradiosurveysusedtocreatetheuni-
sight” matches to physically unrelated objects. fied catalog, FIRST extends to the faintest flux
Using the nearest-neighbor distributions, we es- limit. We discuss here the likelihood of finding a
timate efficiency (fraction of matches which are FIRSTsourceinoneormoreoftheotherthreera-
physically real) and completeness (fraction of real dio surveys. In this section and for the remainder
matches that were found) as a function of match- ofthispaper,welimitouranalysistothecommon
ing radius. Figure 3 shows the distribution of region observed by all of the contributing surveys
distances between FIRST sources and the near- (see Fig. 1).
est neighbor from the NVSS, WENSS, GB6, and Thenumberofmatchesfoundbetweendifferent
7
surveys is a strong function of source flux. The 3.5.1. NVSS sources without a FIRST counter-
flux range of objects in the unified catalog covers part
severalordersofmagnitude. Thus,forconvenience
Because FIRST goes fainter than the NVSS by
weconvertradiofluxtoan“ABradiomagnitude”,
a factor of 2.5, there are many FIRST sources
following Ivezi´c et al. (2002): without an NVSS counterpart (∼ 38deg−2, see
f PanelAofFig.4;∼0.53deg−2 forFIRSTsources
int
tradio =−2.5log , (1) brighter than 10mJy). However, the catalog also
(cid:18)3631Jy(cid:19)
contains some NVSS sources without a FIRST
where f is the integrated flux density. This for- counterpart(∼7.8deg−2;∼0.56deg−2forsources
int
mulaplacestheradiomagnitudesontheAB sys- brighter than 10mJy). For an astronomical ob-
ν
temofOke & Gunn(1983). Anadvantageofthat ject to be detectable by NVSS but not FIRST, it
systemisthatthezeropoint(3631Jy)doesnotde- must be large and have a surface brightness too
pend on wavelength and thus enables convenient faint for the high-resolution FIRST survey. On
data comparison over a large wavelength range. the other hand, it is possible that NVSS sources
For example, a source with a radio flux of 1mJy without FIRST counterparts are simply spurious,
hast =16.4;ifithasconstantF (aflatspec- orhavebadlymeasuredNVSSpositions,i.e. their
radio ν
′′
trum), its visual magnitude is also V =16.4. FIRST counterpartsareoutsidethe 30 matching
radius. The toppanel of Figure 5 shows the near-
Figure 4 shows the magnitude distribution for
est neighbor distribution from the WENSS cat-
FIRST sources with and without radio matches
alog for NVSS sources lacking a FIRST counter-
from the other surveys. Each panel roughly indi-
′′
partwithin30 . Itpeaksatsmalldistancesdueto
cates the t magnitude corresponding to the
FIRST
physically-associated objects, demonstrating that
other survey’s faint limit. For example, the frac-
NVSS sources lacking a FIRST counterpart but
tionofFIRSTsourceswithacounterpartinNVSS
′′
withaWENSSmatch(within55 ),aredominated
is greater than 0.9 at t = 15 but drops to
FIRST
by real sources. If these are not spurious detec-
0.5 at t = 15.5 as shown in Panel A, in-
FIRST
tions, they belong to the category of extended,
dicating that the NVSS faint limit corresponds
low-surfacebrightnessobjectsmissedbythe high-
to t ≈ 15. The sensitivity limit of GB6,
FIRST
resolution FIRST.
t =13.3,occursatt ≈12(panelB).The
GB6 FIRST
limitisbrighterinFIRSTthanGB6because,asis Assuming that realNVSS sourceshavea coun-
well known and as we confirm with a much larger terpart in at least one of FIRST and WENSS,
sample in §4.2, most radio sources are intrinsi- we estimate an upper limit for the fraction of
callyfainterat6cmthanat20cm(intermsofF ; spurious NVSS detections. Nearly 14% of NVSS
ν
i.e. corresponding faint limits depend on spectral objects in the region of survey overlap have no
′′
slope). RequiringaGB6detectionthereforebiases FIRSTcounterpartwithin30 ,and12.2%haveno
a sample to the brightest sources. Panel C shows FIRST or WENSS counterpart, the latter within
′′
that the WENSS faint limit, t = 13.3, cor- 55 . NVSS has a much fainter flux limit than
WENSS
responds to a limit of t ≈ 14, because radio WENSS (2.5mJy as opposed to 18mJy): sources
FIRST
sourcesare,againonaverage,intrinsicallybrighter near the NVSS faint limit are likely to fall below
at 92cm than at 20cm. the WENSS faint limit. For the sample of NVSS
detections brighter than 18mJy, only 0.43% lack
Roughly 60% of FIRST sources have an NVSS
′′ both FIRST and WENSS counterparts, which we
counterpartwithin30 . Ofthoseradiosourcesob-
served by both surveys (≈ 48deg−2), 50% have a interpretasanupperlimitonthefractionofspuri-
ousNVSSsources. Theselectiononwhichwebase
WENSS counterpart and 12% have a GB6 coun-
′′ this estimate is biased against sources brighter at
terpart within 120 , while 11% have both. The
20cm than at 92cm; however, this type of radio
sky density of sources observed by all four radio
surveys is ≈ 5.3deg−2, or ∼ 5% of the FIRST source is rare, as we later show (§4.2). Accord-
ing to de Vries et al. (2002), NVSS begins to lose
source density.
completeness below 12mJy. We find that 2.3%
of NVSS sources above this limit have no FIRST
match.
8
3.6. Catalog availability three morphology classes, for which we construct
radio “color-magnitude” diagrams. We also uti-
The full catalog and several pre-made subsets
lize optical photometric identification in the anal-
areavailablefordownloadonthecatalogwebsite3,
ysis,classifyingsourcesasSDSSpointsources,ex-
which also provides a list and description of the
tended sources, or faint sources (non-detections).
data parameters. The datasets are described in
Hereafterourdiscussioncoverssamplesselected
Appendix B. All 2,724,343 rows of the complete
usingtheconservativematchingradiilistedinTa-
catalog(sampleA)areavailableinatarredarchive
of fits files, each covering a 5◦ wide strip in right ble 2. We settle upon these values by looking for
an optimal balance between completeness and ef-
ascension. A small subset of the catalog selected
from ≈ 100deg2 of sky, with 16,453 rows (sam- ficiency of the matched samples, as illustrated in
Fig. 3. The matching radii used herein are as fol-
ple B), allowsusersto familiarize themselves with
lows, with the larger values for all catalog asso-
the data format. Several scientifically-useful sub-
′′
ciations given in parentheses: FIRST-NVSS, 25
sets of the catalog are pre-made for convenience;
′′ ′′ ′′
(30 ); FIRST-WENSS, 30 (120 ); FIRST-GB6,
someoftheseareanalyzedinthenextsection. The
′′ ′′ ′′ ′′
70 (120 ); FIRST-SDSS, 2 (60 ). Estimated
firstof these subsets containsradio fluxes andpo-
values of completeness and efficiency, calculated
sitions for NVSS-FIRST associations (sample C),
bycomparingwithrandommatching,arelistedin
including6and92cmdata,whileanothercontains
Table 2.
sources matched by all four radio surveys as well
as the SDSS (sample F). Two more subsets con- Our analysis is limited to the 2955deg2 region
sistofspectroscopicgalaxies(sampleG)andspec- where the contributing surveys overlap(Fig. 1) in
troscopic quasars (sample H), detected by NVSS, order to control the selection criteria of the ana-
FIRST, WENSS, and SDSS. Also provided are a lyzed samples. There are nearly 490,000 catalog
sample of isolated4 FIRST-NVSS sources (sam- entries in the overlap region, including multiple
ple I) and isolated FIRST-NVSS-SDSS sources FIRST components matched to the same NVSS
(sample J). Finally, a set of high-redshift galaxy source, as well as duplicate entries (see §3.2). To
candidates is available (sample K; see §5.2). A obtain a sample of physical FIRST-NVSS associ-
more detailed description of these data files can ations,it is sufficient to select catalogentries con-
be found in Appendix B. taining an NVSS source and its nearest FIRST
match: individual objects are typically not re-
4. PRELIMINARY CATALOG ANALY- solvedintomultiple componentsinNVSSbecause
SIS of the survey’s larger beam. This method selects
∼ 140,000 unique, likely physical NVSS-FIRST
In this section, we investigate the distribution associations. Of those, ∼ 64,000 are matched to
of sources in radio color-magnitude-morphology a WENSS source, ∼ 14,000 to a GB6 source and
parameter space. We define the relevant mor- ∼48,000 to an SDSS source.
phology and spectral parameters using five ra-
dio fluxes: peak FIRST flux (20cm), integrated 4.1. Morphology classification of radio
FIRSTandNVSSfluxes(20cm),GB6flux(6cm), sources
and WENSS flux (92cm). While the unified cat-
4.1.1. “Simple” vs. “complex” FIRST-NVSS
alogcontainsmanymore usefulparametersinher-
sources
ited from the original catalogs, in this prelimi-
nary analysis we focus only on these few. We use FIRST is a high-resolution interferometric sur-
FIRSTandNVSSfluxmeasurementstodefinetwo vey, and thus underestimates the flux of ex-
morphologyestimators,anduseGB6andWENSS tended and lobe-dominated sources (Becker et al.
fluxes with NVSS flux to compute radio spectral 1995; Lu et al. 2007). Additionally, a multiple-
slopes. Morphology estimators are used to define component source with core and lobes may be
detected as three objects by FIRST but as only
3http://www.astro.washington.edu/akimball/radiocat/ a single object in the lower-resolutionNVSS. The
4Isolated herereferstoanNVSSsourcewithasingleFIRST difference between the two 20cm magnitudes is
counterpartwithin30′′. thus a measurement of source morphology that
9
indicates angular extent and complexity. We de- detections into three morphology classes: “com-
fine plex”, (simple) “resolved”, and (simple) unre-
∆t=t −t . (2) solved or “compact”.
FIRST NVSS
This assertion was verified by extensive vi-
Beforeapplying∆tasaclassifierweinvestigateits
sual inspection of FIRST images (Fig. 8). Over
dependenceont forNVSS-FIRSTpairs(sam-
NVSS 1000 FIRST stamps (2×2 arcmin2) of optically-
ple C) in Figure 6. The distribution in ∆t is bi-
identified radio quasars and radio galaxies were
modal,withthetwopeakscenteredat∆t=0and
classified both visually and automatically (as
∆t=0.7. Throughthevisualinspectionofathou-
above) into “complex”, “resolved”, and “com-
sandFIRSTimages(seebelow,§4.1.3),weverified
pact” categories. A comparison of the results
that sources in the ∆t ∼ locus are single compo-
is given in Table 4. The two methods are con-
nent sources, while those in the ∆t∼.7 locus are
sistent: over three quarters of all sources (81%
multiple-componentorextended. Thelowerpanel
of quasars and 76% of galaxies) receive identical
showsthe∆tdistributionforfourmagnitudebins.
classificationfromthe two methods. Twoinitially
We fit the sum of two Gaussians to each of the
surprisingresultsarethesignificantfractionofob-
four distributions; best-fit parameters are listed
jects classified as “complex” by one method and
in Table 3. The ∆t distribution is similar for all
“compact” by the other. In particular, 15% of
magnitudes,althoughtheoverlapbetweenthetwo
visually-complex quasarswere automatically clas-
peaksincreasesforfaintersourcesasfluxmeasure-
sified as “compact”. As it turns out, the majority
ment errors increase. We adopt ∆t = 0.35 (solid
ofthesesourcesappear(byeye)tobe asymmetric
horizontalline)astheseparatorbetween“simple”
double-lobed sources with a very high flux ratio.
(∆t<0.35) and “complex” (∆t>0.35) sources.
Their FIRST images show two lobes, hence the
4.1.2. “Unresolved” vs. “resolved ” FIRST visual “complex” classification. However, their
sources total flux is due primarily to the brighter lobe,
meaning FIRST and NVSS flux measurements
TheratioofthetwoFIRSTfluxmeasurements— are similar, resulting in an automatic classifica-
peak and integrated—yields a second measure of
tion of “compact”. Additionally, 26% of visually-
morphology: adimensionlesssourceconcentration compact galaxies were classified automatically as
′′
on ∼5 scale. We define “complex”. An inspection of flux values reveals
that many of these are borderline cases, with ∆t
F 1/2
θ = int . (3) value just above the ∆t = 0.35 cutoff. It is likely
(cid:18)F (cid:19)
peak that these galaxies have faint, diffuse emission
which is detected by the NVSS but is not visible
Sources with θ ∼ are highly concentrated (unre-
in the FIRST images.
solved), while sources with larger θ are extended
(resolved). We adopt log(θ2) = 0.05 (θ ≈ 1.06) Theautomaticmorphologyclassificationpresents
difficulties because of non-unique pair matching,
as the value separating resolved and unresolved
e.g. multiple FIRST detections matched to a sin-
sources. This choice is motivated by the distribu-
gle NVSS object. The resolved/unresolved clas-
tion of sources in the two-dimensional ∆t vs. θ
sification is based on FIRST fluxes and therefore
distribution, discussed below.
applies to FIRST components individually. In
4.1.3. Automatic morphology classification of choosing an analysis sample based on FIRST-
radio sources NVSS pair matching, we retain only the closest
FIRSTmatchtoanNVSSobject(seeAppendixA
The two-dimensional ∆t vs. θ distribution for
fordetails). ThefluxesoftheclosestFIRSTmatch
sources with t < 13.5 (the brightest 30%
NVSS areused,whilemoredistantFIRSTsourcesareig-
of NVSS-FIRST pairs) is illustrated in Figure 7,
nored,whenclassifyingaFIRST-NVSSpairasre-
along with the marginal distributions for the two
solved/unresolved. Approximately 17% of NVSS
morphology parameters. The distribution in the
detections in the FIRST footprint have multi-
lower panel suggests that 20cm radio fluxes can ′′
ple FIRST matches within 30 . Most of these
be used to automatically separate FIRST-NVSS
sources (88%) are classified as complex, presum-
10