The Red-sequence Cluster Survey-2 (RCS-2): survey details and photometric catalog construction

0.6 MB·English

by  David G. Gilbank

AJ accepted
PreprinttypesetusingLaTEXstyleemulateapjv.11/10/09
THE RED-SEQUENCE CLUSTER SURVEY-2 (RCS-2): SURVEY DETAILS AND PHOTOMETRIC CATALOG
CONSTRUCTION
David G. Gilbank
DepartmentofPhysicsandAstronomy,UniversityofWaterloo, Waterloo,Ontario,N2L3G1,Canadaand
DepartmentofAstronomyandAstrophysics,UniversityofToronto,,50StGeorgeStreet,Toronto,Ontario,M5S3H4,Canada
M. D. Gladders
1
DepartmentofAstronomyandAstrophysics,UniversityofChicago,5640S.EllisAve.,Chicago,IL,60637, USA
1
0
2 H. K. C. Yee
DepartmentofAstronomyandAstrophysics,UniversityofToronto,50StGeorgeStreet, Toronto,Ontario,M5S3H4,Canada
n
a and
J
B. C. Hsieh

9 InstituteofAstrophysicsandAstronomy,AcademiaSinica,P.O.Box23-141,Taipei106,Taiwan
1 AJ accepted

] ABSTRACT
O
The second Red-sequence Cluster Survey (RCS-2) is a ∼1000 square degree, multi-color imaging
C survey using the square-degree imager, MegaCam, on the Canada-France-HawaiiTelescope (CFHT).
. It is designed to detect clusters of galaxies over the redshift range 0.1 < z <1. The primary aim is
h ∼ ∼
p to build a statistically complete, large (∼104) sample of clusters, covering a sufficiently long redshift
- baseline to be able to place constraints on cosmological parameters via the evolution of the cluster
o mass function. Other main science goals include building a large sample of high surface brightness,
r strongly gravitationally-lensed arcs associated with these clusters, and an unprecedented sample of
t
s several tens of thousands of galaxy clusters and groups, spanning a large range of halo mass, with
a
which to study the properties and evolution of their member galaxies.
[
This paper describes the design of the survey and the methodology for acquiring, reducing and

2 calibratingthedatafortheproductionofhigh-precisionphotometriccatalogs. Wedescribethemethod
v forcalibratingourgriz imagingdatausingthecolorsofthestellarlocusandoverlappingTwo-Micron
0 All-SkySurvey(2MASS)photometry. Thisyieldsanabsoluteaccuracyof<0.03magonanycolorand
7 ≈0.05 mag in the r-band magnitude, verified with respect to the Sloan Digital Sky Survey (SDSS).
4 Our astrometric calibration is accurate to ≪ 0.3′′ from comparison with SDSS positions. RCS-2
3 reaches average 5σ point source limiting magnitudes of griz = [24.4,24.3,23.7,22.8], approximately
2. 1-2 magnitudes deeper than the SDSS. Due to the queue-scheduled nature of the observations, the
1 dataarehighlyuniformandtakeninexcellentseeing,mostlyFWHM∼<0.7′′ inther-band. Inaddition
0 to the main science goals just described, these data form the basis for a number of other planned
1 and ongoing projects (including the WiggleZ survey), making RCS-2 an important next-generation
: imaging survey.
v
i Subjectheadings: surveys—techniques: photometric—galaxies: clusters: general—galaxies: general
X — cosmology: observations
r
a
1. INTRODUCTION references therein).
Constructinglarge,well-definedsamplesofgalaxyclus-
Clustersofgalaxiesareidealtracersofthelargestden-
ters has a long and varied history. The first system-
sity fluctuations in the Universe, and their abundance
atic searches involved visual identification of overden-
(and its evolution with cosmic time) may be used to
sities of optical galaxies on photographic plates (Abell
place constraints on cosmology (e.g., Eke et al. 1996).
1958; Abell et al. 1989). In the 1970s, with the advent
They also provide ideal laboratories for studying galaxy
of X-ray telescopes above the Earth’s atmosphere, selec-
evolution. Originally used in this way since they con-
tionofclustersfromtheirextendedX-rayemissionfound
tain a large number of galaxies all in the same location,
favour(Mitchelletal.1976;Serlemitsosetal.1977). Re-
it has since become clear that the properties of their
cently,acombinationoflargeformatCCDdetectorsand
member galaxies are markedly different from galaxies in
objectivealgorithmstosearchefficientlyforsignaturesof
the general field (e.g., Dressler 1980, Balogh et al. 1999,
galaxy clusters has led to a revival in the use of optical
Ellingson et al. 2001), implying that mechanisms which
selectioninclustersurveys(Postmanetal.1996;Kepner
truncate star formation and transform galaxy morphol-
etal.1999;Galetal.2000;Gladders&Yee2000;Gilbank
ogy operate on cluster scales (e.g., Treu et al. 2003 and
etal.2004). Avarietyoftechniqueshavebeensuggested
toexploitthe expectedluminosityand/orcolordistribu-
[email protected]
2 Gilbank et al.
tion of galaxies in clusters. The big advantage of these et al. 2007, 2010), X-ray (Hicks et al. 2008), strong
surveys compared with the older visual searches is that and weak-lensing (from an ACS snapshot programme,
the detection method could be automated and charac- PI:Loh;ACSSNeCosmologyprojectPI:Perlmutter)and
terised,meaningthatthesurveyselectionfunctioncould SZ observations. In this way, the relationship between
bequantified. Arguablythemostefficientmethodisthat our mass proxy (optical richness from the survey data)
of Gladders & Yee (2000) which uses the fact that the and cluster mass can be understood.
cores of galaxy clusters are dominated by galaxies with One of the primary goals of RCS-1 was to place con-
old stellar populations, forming a tight red-sequence in straints on cosmological parameters (Ω , σ , Gladders
M 8
color magnitude space (Visvanathan 1978; Bower et al. et al. 2007) via the growth of the cluster mass function.
1992). A number of other realisations of red-sequence This demonstrated for the first time the feasibility of
basedclusterfindingalgorithmsexist(e.g.,Koesteretal. suchanapproachusinganoptically-selectedclustersam-
2007) differing in some details but all relying on accu- ple. This approach used the measured relation between
rate colors from imaging in two or more filters. The massandrichness,butalsoshowedthatmeaningfulcon-
observedcolorof this sequence provides anaccuratedis- straints could be obtained using a self-calibration tech-
tance estimate. The application of this method led to nique (Majumdar & Mohr 2004) to estimate the form
the construction of the first Red-sequence Cluster Sur- of this relation from the survey data themselves. These
vey (RCS-1, Gladders & Yee 2005), a 72 square degree authors demonstrate that the best constraints are ob-
imaging survey in two bands (R and z′) designed to tained when accurate mass estimates are available for a
C
locate galaxy clusters from 0.2<z <1.1 using the tech- subsample of clusters within the survey. It is worth em-
∼ ∼
nique of Gladders & Yee (2000). phasising that even if there is significant scatter in the
Not only is optical selection of galaxy clusters under- relation between mass and the proxy (as we have found
going a revival, but astronomy in general is entering an for optical richness), it is only important that the size
eraof‘surveyscience’whereanunprecedentednumberof of the scatter be well understood. With an order-of-
wide-fieldoptical(andNIR)surveysarecurrentlyunder- magnitudelargersurveythanRCS-1,itbecomesfeasible
way or planned, such as LSST (LSST Science Collabo- to alsoconstrainthe equationofstate ofdark energy,w,
rations et al. 2009); Pan-STARRS1; UKIDSS (Lawrence (Majumdar & Mohr 2004) and this is in part the mo-
etal.2007);andDES2. Inaddition, many ofthese wide- tivation for RCS-2. RCS-1 also produced a significant
sampleofstronglygravitationallylensedarcsaroundthe
field optical surveys are specifically targeted at areas
clusters found. The number and redshift distribution of
surveyed for clusters using other methods, such as the
theselensingclusterswereusedtoargueaboutthephys-
BlancoCosmologySurvey(Highetal.2010)oftheSouth
icalpropertiesoftheclustersresponsiblefortheirlensing
Pole Telescope (SPT, Carlstrom et al. 2009) Sunyaev
cross-section and the relevance of such systems to con-
Zel’dovicheffect(SZ)-selectedclustersurvey. Forsurveys
strainingcosmology(Gladders et al.2003). The identifi-
using other methods (such as SZ selection), the optical
cation of such high surface brightness, strongly-lensed
data are critical for the verification of the cluster can-
galaxies is another primary science driver for RCS-2.
didates found and for the determination of photometric
The massive clusters can be used as gravitational tele-
redshifts. Furthermore, surveying the same areas with
scopes for studying high-redshift galaxies (e.g., Pettini
multiple techniques allowsimportantcomparisonsofthe
etal.2000;Wuyts etal.2010)whichwouldotherwise be
different selection effects and the resulting properties of
too faint to observe in detail. The giant arcs can also
the clusters found (e.g., Donahue et al. 2001; Gilbank
be used as probes of the properties of the cluster lenses
et al. 2004; Rasmussen et al. 2006).
themselves.
In this paper we describe the second Red-sequence
With a statisticalsample of galaxyclusters, suchas in
Cluster Survey (RCS-2), the largest survey of this new
RCS-1,itispossibletostudythepropertiesoftheirmem-
generation for which imaging has already been com-
ber galaxies (e.g., their luminosity functions and blue
pleted. This builds on the methodology of RCS-1. The
fractions) by stacking subsamples built from the survey
RCS collaboration has invested a large amount of work
data themselves (Gilbank et al. 2008; Loh et al. 2008).
in attempting to characterize the selection function and
With the order-of-magnitude larger RCS-2, it becomes
the properties of clusters selected with this technique.
feasible to measure much weaker trends and push mea-
Many of these results are directly applicable to RCS-2
surements of cluster galaxies down to much lower over-
(suchasmass–richnesscalibrations)andsoitisusefulto
densities. The addition of photometric redshifts (e.g.,
summarize some of the RCS work to date.
Hsieh et al. 2005) will allow these techniques to be ex-
Theefficiencyoftheselectionmethodemployedbythe
tended to the field environment. Such galaxy evolution
red-sequence surveys is that it can locate and estimate
studies will be explored in future work with RCS-2.
the redshifts of clusters using only one color (two filter)
The outline for this paper is as follows. In §2 we give
data,giventheappropriatechoiceoffilters. Itisimprac-
an overview of the survey design and observations; §§3
tical to obtain mass estimates from follow-up observa-
&4dealwithCCDpre-processing,reduction,objectde-
tionsofthe∼30000clusterswhichwillbefoundinRCS-
tection and photometry; §5 describes the photometric
2,sothesurveydatathemselvesmustbeusedtoproduce
calibrationvia accuratefits to the star colors in our sur-
aproxyforclustermass. Significant,representativesam-
vey fields; §6 outlines the procedure for and accuracy
ples of clusters from RCS-1 have been followed up using
of the astrometric calibration. §§7 & 8 describe the in-
a varietyof massestimatorssuchas dynamical(Gilbank
corporation of additional data into our primary RCS-2
catalogs: i-band data which covers a large (∼70%) sub-
1 seehttp://pan-starrs.ifa.hawaii.edu/public/ sample of the primary g, r, z survey area; and public
2 seehttps://www.darkenergysurvey.org/ imaging data from the CFHTLS-Wide survey, respec-
RCS2 survey construction 3
tively. §9 describes the final cleaning of the photometric
catalogs: stitchingintocontiguouspatches,removingdu-
plicatedatabetweenoverlappingpointings,andmasking
of artefacts. §10 summarises and describes ongoing and
future work for the survey.
2. SURVEYOVERVIEW
The wide-field imaging capability of the square degree
imager, MegaCam (Boulade et al. 2003), on the 3.6-m
CFHT makes it feasible to carry out a survey covering
a significant fraction of the sky in a modest amount of
observingtime. Coupledwiththeexquisiteseeingcondi-
tions attainable from the summit of Mauna Kea, such a
survey can achieve impressive depths/resolution within
this time.
The main science drivers for RCS-2 dictate a survey
area of ∼1000 deg2 with the ability to detect galaxy
clusters out to z∼1 using the method of Gladders & Yee
(2000). FollowingthestrategyofRCS-1(Gladders&Yee
2005), which used R - and z-band imaging, r′- and z′- Fig.1.—ExampleofacatalogproducedforoneMegaCampoint-
C ing,illustratingthelayoutoftheCCDs. Pointsshowr<24objects
band3 filters arechosento coverthe bulk ofthis redshift classifiedasgalaxiesinoneexamplepointing(0133A0). The9×4
range. In order to better distinguish lower-redshift clus- grid of the individual detectors is clearly visible, as are the large
gapsbetweenthetopandbottomrowsandtheothers,describedin
ters, g-bandisaddedto accuratelymeasurethe colorsof
thetext. Circularregionsabsentofgalaxiesshowwhereourmasks
galaxies at z∼<0.4, once the 4000˚A break begins to move forbrightstarhaloes(describedin§9.1)havebeenapplied.
blueward of the r-filter. The g-band filter also helps to
identify stronglygravitationally-lensedarcsaroundclus- the lowest richness clusters will not be detected in the
ters, since the former tend to be relatively blue. Thus, worse seeing imaging (due to the reduced depth) all the
thesurveycomprisesthree-color,g,r,z,imagingoverthe wayouttoz∼1,butricherclusterswill. Sincetheformer
whole area. i-band imaging is obtained for the majority are exponentially more numerous, this should have neg-
ofthisareaviaadata-exchangewiththeCanada-France ligibleimpactoncosmologicalconstraints. g-bandimag-
High-z Quasar Survey (Willott et al. 2005). ing was always performed in the better-seeing bracket
in order to preserve the low surface brightness require-
2.1. Observations ment for the detection of strongly-lensed arcs. Our final
measured depths are detailed in §5.3.
MegaCamcomprises36CCDsarrangedinafourrow×
Single exposures were taken at each pointing position,
ninecolumngridatthe primefocusofCFHT.Eachchip
′′ i.e., no dithering was used. This is because sufficient
is 2048×4612 pixels with a pixel scale of 0.187 , allow-
depth is achieved in only a single exposure of 4-8 min-
ing it to adequately sample the exquisite seeing possible
utes (depending on filter), so the overhead associated
fromMaunaKea. ThespacingbetweenCCDsisapprox-
′′ ′′ with reading out the CCD (≈2 minutes) becomes a sig-
imately 13 , with larger gaps (≈80 ) between the up-
nificant fraction of the integration time if this is divided
permost and lowermost rows and the other CCDs. This
into two or more exposures. The priority is to cover as
gives sky coverage of 0.96×0.94 deg2 (see Fig. 1).
largeanareaofskyaspossibletothesedepthsinagiven
Observations were carried out in queue-scheduled
amount of observing time. With this strategy the sur-
mode on CFHT on runs between semesters 2003A and
vey will contain chip gaps, but these can be dealt with
2007B inclusive. PI imaging time was granted through
(for cluster finding, etc.) via geometric correction fac-
requests to Canadian and Taiwanese agencies.
tors (discussed in §9.2). Gaps in the data also appear
Exposure times were set to 4, 8, and 6 minutes in g,
r, and z respectively. In 0.65′′ seeing, according to the duetothe occasionalfailureofaCCDwithinthemosaic
camera.5
MegaCam exposure time calculator, these should corre-
Individual pointings are arranged in a grid pattern in
spondto5-σpointsourcelimitsofg ≈25.3,r ≈24.8,and
discrete patches/fields (discussed in the next section).
z ≈22.5. The depth was set in the two reddest bands
Pointingsarenamed with the patchname, followedby a
(r and z) by the requirement to reach ≈ M⋆ +1 red-
letter designating the column, beginning at ’A’ running
sequence cluster galaxies at z∼1. Part way through our
′′ east to west. Next a digit specifies a pointing’s location
survey observations, it was found that 0.65 seeing was
in the declination direction, beginning at ’0’, running
not available on as many nights as required by our pro-
fromsouthtonorth. Thus,pointingswithinpatch’0133’
gram per semester4. So, a two-tier strategy for r and
are labelled ’0133A0’ to ’0133F5’, from the south-east
z-bandimagequalitywasadoptedinwhichonlyhalfthe
′′ to north-west corners across the patch. Each pointing
surveywouldbeconductedinthe0.55-0.75 bracketand
′′
half would be conducted in 0.75-0.90 . This means that 5Inasmallnumberofcases,problemswiththecameraelectron-
icshave caused halfthe mosaicto failand sothe queue observers
3Hereafterweusetheshorthandofomittingtheprimenotation have observed the northernand southern halfof asinglepointing
fromtheMegaCamfilternames. intwoseparateobservationswithMegaCamoperatinginthishalf
4 Thiswas somewhatmitigated bythe significantimprovement functionalmode. Inthesecaseswehaveprocessedthe‘halfpoint-
inMegacamimagequalityresultingfromtheL3lensbeingreplaced ings’asiftheywereseparatepointingswithan18chipcameraand
intheupside-downpositioninlate2004. combinedthem intothefinalcatalogatthestitchingstage(§9.2).
4 Gilbank et al.
overlaps with its neighbors by ≈1 arcmin, typically. See
Fig. 2. TABLE 1
RCS-2 primarysurvey (PI imaging)
The originalobservingstrategywasto observea given
pointing sequentially in all three filters (g, r, z), but
the overheads associated with filter changing made this Patcha R.A. Dec.b extentc commentsd
relatively inefficient and so, after the first semester, the
queue observers switched to observing several neighbor- RCS-2primarysurvey(PIimaging)
ing pointings in a single filter and then repeating these 0047 00:47.4 +00:45 9×9
in the next filter. i-band imaging is typically performed 0133 01:33.2 -00:10 6×6
many months later for a given pointing. This is poten- 0310 03:10.3 -14:11 9×9 Initial patch extended to
better cover the WMAP
tially a consideration for projects wishing to use RCS-2
coldspot
for time variability studies. 0357 03:57.2 -08:48 6×6
1040 10:40.9 +57:48 6×6 LockmanHole
2.2. Field descriptions 1111 11:12.0 -05:52 9×9
1514 15:14.6 +08:55 9×10
Whenconductingalargesurvey,itisobviouslyadvan- 1613 16:13.6 +55:00 6×6 ELAIS-N1
tageous to place fields such that they overlapwith other 1645 16:45.9 +39:35 6×6 ELAIS-N2
large surveys at other wavelengths, whenever possible. 2143 21:41.0 -00:07 10×10
2329 23:26.1 -01:14 8×6 DEEP2
The RCS-2 survey area is divided into discrete patches
2338 23:38.5 -09:07 9×9
(where ‘patch’ refers to a contiguous set of MegaCam
pointings). The patches are divided approximately into CFHTLegacySurveyWidecomponent
twocategories: thosedesignedtooverlapwithothersur- W1 02:18.0 -07:27 9×8 XMM-LSS
veys(suchasSWIRE,Lonsdaleetal.2003)aretypically W2 08:57.8 -03:18 5×5
a6×6grid(i.e.,6degreesonaside);andthosedesigned W3 14:17.9 +54:30 6×7
W4 22:11.4 +01:48 6×6 VVDS
tobe‘blank’pointingstypically9×9. Thelargerpatches .
aredesignedtobesufficientlylargethattheclusteringof
a patch name
galaxy clusters may be used to provide additional con-
b RA & Dec (J2000) of the patch centre
straints on cosmology (e.g., Majumdar & Mohr 2004).
c size (in numberof 1 deg2 pointings).This gives theapprox-
Theselarger‘blank’pointingsareplacedinthefootprint
imate extent of the patch, but the layout is not necessarily
oftheUKIDSS(Lawrenceetal.2007)whereverpossible,
rectangular. The exact geometry can beseen in Fig. 3.
so that we will obtain NIR data for our optical imaging, d other relevant comments, such as overlapping surveys
useful for deriving stellar masses of galaxies, for exam-
eral of our patches (0047, 0133, 2143, 2329, 2338) with
ple. Thepatchsizeisreducedto∼6×6wherewetarget
GALEX in order to select z>0.5 Lyman Break Galaxy
other surveys, if the field size of the survey with which ∼
weoverlapissmallerthan6×6deg2. Wechoosediscrete candidates using optical/UV colors to generate a large
patches distributed across the sky in order to improve spectroscopic redshift sample to measure cosmological
observing efficiency (by distributing fields in RA), and parameters via the signature of baryon acoustic oscilla-
to minimise the effects of cosmic variance (although this tions imprinted on their clustering. The Canada France
wouldonlybeanissuefortherarestsystems,suchasthe High-zQuasarSurvey(CFHQS,Willottetal.2005)uses
mostmassiveclusters,sinceourindividualpatchesareso ourz-bandimaginginconjunctionwiththeirowni-band
large - the largest patches in RCS-2 are already bigger imaging to search for high redshift (z∼7) quasars.
thantheentireRCS-1!). Mostofourfieldsareequatorial These fields total 785 one-square degree pointings of
in order to maximize follow-up from telescopes in both imagingdataforRCS-2. Tothisweaddthefourpatches
hemispheres. ofthe publicly-availableWide componentofthe CFHLS
All patches are chosen to be in regions of low Galac- which tallies 171 deg2 in u⋆, g′, r′, i′, z′. The reformat-
tic extinction (see Fig. 3). Another consideration is the ting and recalibration of these data to resemble the re-
avoidance of bright stars. Using the Bright Star Cat- duction of our PI imaging data is described in §8. Thus,
alogue (Hoffleit 1964), patches are shifted slightly to overall, RCS-2 comprises 955 one-square degree point-
minimise the number of stars they contain in the range ings.
2 < m < 4 wherever possible. Stars this bright (and
V 3. PRE-PROCESSING
indeedsomewhatfainter)cancauselargereflectionarte-
facts (see §9.1) which can render significant areas of the Standard CCD pre-processing (bias subtraction,
image unusable. Stars brighter than m < 2 are po- bad-pixel masking, flatfielding, etc.) is performed
V
tentially hazardous to the detectors and so pointings in by the Elixir Project6 at CFHT. For the inter-
ourgridwhichwouldoverlapwithsuchstarsarenotob- ested reader, an excellent description of the raw
served. MegaCam data, including example images at var-
AlistofthepatchesobservedisgiveninTable1. Patch ious stages of Elixir reduction, may be found at
1040coversthe LockmanHole,alsosurveyedbySWIRE http://www.cfht.hawaii.edu/Science/CFHTLS-DATA/rawdata.html
and UKIDSS-DEX (Deep Extragalactic survey), as well We use the default Elixir reduction to handle all these
as partially coveredby UKIDSS LAS; 1613 also coversa preliminarysteps,exceptforthez-banddefringingwhich
SWIRE (ELAIS-N1) and UKIDSS-DEX field; 1645 cov- we found could be improvedupon. Due to the large vol-
ers a SWIRE (ELAIS-N2) field. 2329 targets a DEEP2 ume of data generated by our observations, rather than
field. In addition to these targeting specific surveys, our starting from raw (non-defringed) data and attempting
surveyfields havebeen usedby othersurveys. The Wig-
gleZ survey (Drinkwater et al. 2010) has surveyed sev- 6 seehttp://www.cfht.hawaii.edu/Instruments/Elixir/
RCS2 survey construction 5
Fig. 2.—Layout of atypical patch (0133). Points show objects inthe photometric catalog brighter than r<20. Thechipgaps within
each pointingareclearlyvisible. Pointings arelabelledwiththeir twocharacter names, as describedinthe text. Overdensities ofobjects
arevisiblewherepointingsoverlap,astheseduplicateareashavenotyetbeenremovedinthestitchingprocess(see§9.2). Artefactssuchas
satellitetrailshavenotyetbeencleaned(see§9.1)andexamples ofthesecanbeseeninpointingsC5andE3. Additionalgapsarevisible
fromwhereoneoftheCCDs(lowerrightcorner)wasnon-functional whenE0andF0wereobserved.
Fig.3.—LayoutofRCS-2patchesontheskyinCartesianprojection,plottedasafunctionofRA(hours)andDec(degrees). Grey-scale
shows distribution of Galactic Extinction from Schlegel et al. (1998). Squares denote individual Megacam pointings. (Horizontal gaps
between pointings in high declination patches are an artefact of the projection and the pointings do in fact overlap, as described in the
text.) Patchesarelabelledwithpatchnames. Additional(unlabelled)pointingsaround13hoursequatorialdenoteapatchwhichwasleft
uncompleted (1303) and will not be considered as part of the survey proper (due to the non-contiguous nature of the observed regions),
butindividualpointings willstillbeincludedinsearches forstronglenses.
6 Gilbank et al.
to generate new master fringe frames from other z-band problems when we attempt to measure offsets between
observationstakenat similartimes to our observations7, the differentfiltersfor eachpointing. This quickcalibra-
weinvestigatedthe possibility ofstartingwith the Elixir tionfollows a similar method to that which providesour
(non-optimally)-defringed z-band images. It was found final,accurateastrometriccalibration,whichisdescribed
that running these images through defringing code de- in detail in §6. For each pointing, a WCS is generated
velopedby the CFHT SupernovaLegacySurvey(SNLS) for each of the three filters, g, r, z, and this is used to
Team (kindly provided by A. Conley) resulted in im- find the approximate offset between the different filters.
provementtothedefringingforallz-bandimages.8 How-
ever, for a small fraction of the z-band data (∼5% of all 4. OBJECTDETECTIONANDPHOTOMETRY
our images), significant fringing is still present. The ini- Objectfindingandphotometryareperformedusingan
tial fringe amplitude in the raw data was at a level of updated version of Picture Processing Package (PPP;
∼15%ofsky9. Thiswastypicallyreducedtobetterthan Yee 1991). While the basic methods are the same as
5% of sky by the initial Elixir defringing and reduced to those described in Yee (1991) and Yee et al. (1998), be-
negligible levels (∼<0.5% of sky) by the SNLS code for cause of the large amount of imaging data involved, a
>95% of all our data. For the remaining ∼5% of our z- considerable number of additional algorithms have been
∼
band images, the residual fringing was at worst ∼3% of implemented to allow the procedures to be carried out
sky. Thus, the most serious impact to our photometric as a completely automated pipeline. In the following,
measurementsfromtheresidualfringingwouldbeanad- we provide brief descriptions of the basic methodology
ditional photometric error of ≈0.03 mag in ∼5% of the andthe added features that allow the automation of the
z-band photometry added in quadrature to the intrin- process.
sic Poissonian error due to photon statistics. Thus, this
inflates the photometric errors for very bright objects 4.1. Object finding
where the intrinsic error is ∼<0.03 mag; but has negligi- In RCS-1, object detection was performed on coadded
ble impact on the vast majority of our objects, which R +z images in order to increase the overall depth of
C
are fainter and thus have larger intrinsic errors. The the observations and to ensure the inclusion of the very
one aspect of our catalog construction impacted by the reddest objects which may have been missed if the R -
C
presenceoflow-levelresidualfringingisobjectdetection. bandalone(thedeeperband)hadbeenused. Asaresult
The implications for this are discussed in §4. of the low-level (<3%) residual fringing in a small frac-
∼
We note that our final images are identical to those tion (<5%) of the MegaCamz data, it is not possible to
∼
generatedby Elixirandstoredinthe CFHT archive(ex-
everywhere use co-added r+z images for object detec-
cept for the z-bandwhichhave been further de-fringed).
tion. (The z-band fringing means that spurious objects
However,for each image, we also generate an associated
aremorelikelytobedetectediftheysitnearthepeakof
badpixelmask(BPM)usingtheproceduresoutlinedbe-
afringe,whereassomerealobjectswillbemissedifthey
low,whichstoresinformationaboutpixelscontaminated
sitinthevalleyofthefringepattern. Thehighernoisein
with reflection haloes from bright stars, satellite trails,
the sky caused by fringing also artificially increases the
and cosmic ray hits, etc.
detection threshold, meaning that some genuine objects
maybemissedincleanareas.) So,forconsistency,andto
3.1. Image alignment
be conservative, object detection is performed solely on
An initial estimate of the world coordinate system the r-band images. Our depths in r and z are matched
(WCS) is taken from the Elixir solution written in the (by construction) for z∼1 red-sequence cluster galaxies,
image headers. In many cases this solution can be off- andsoeventhoughthecatalogisformallyr-selecteditis
setfromtherealpointingpositionbyseveralarcseconds, alsoz-limited. Onlyforgalaxiesredderthanz∼1cluster
andasignificantfractionoftheseshowanon-physicalin- members will the catalog begin to become incomplete10.
strumental solution (such that one or more of the chips Therefore,the only slightlimitationis thatsearchingfor
appeartobeoverlappingordisplacedfromtheirnominal extremely red objects (r-dropouts) is not possible with
positioninthecameragrid). So,weconstructacorrected this catalog.
WCS by running fast object-finding, with a high thresh- PPP performs object detection on each chip as de-
oldset to find only a subsample ofthe brightestobjects, scribed in Yee (1991), with additional algorithms to au-
andpattern-matchtheseagainstanastrometricreference tomaticallycomputethedetectionthresholdandtoblock
catalog. We use a model for the global placement of the out problem regions on the r-image. If artifacts (re-
chipswithin the cameraandsolveforthe locationsofall flection haloes and saturated columns) associated with
objects within the pointing simultaneously. This allevi- bright stars contaminate the chip being considered, the
ates the problem with displacement of individual chips, measurement of the sky noise level, which is used in the
from the Elixir solution, which would otherwise cause estimate of the object detection level, may be biased
high. Furthermore, these regions will also contain many
7duetothequeue-schedulednatureoftheobservations,itisnot
false object detections. So, a mask is applied to remove
alwayspossibletofindsufficientz-banddatatakenwithinashort
enoughtime-frametodothisanyway. contaminated areas from consideration when estimating
8 Briefly, this code uses principal component analysis (PCA), the sky level and the noise.
taking a set of principal components (the eigenvectors) of the z- To mask bright starsand their halos,the approximate
band fringe pattern from data taken on an early SNLS observing
WCS solution derived in §3.1 (typically still accurate to
run (using MegaCam). It then attempts to find the amplitude of
the fringe pattern which must be subtracted (the eigenvalues) in
ordertoremovethepattern. 10 this limit is tagged on a frame-by-frame basis by the PPP-
9seehttp://www.cfht.hawaii.edu/Science/CFHTLS-DATA/rawdata.htmmleasuredphotometriclimits.
RCS2 survey construction 7
better than 1′′) is used to send a query to Vizier11 and The object positions detected from the r image are
download a catalog of all bright stars from the Tycho taken as the ‘master’ position list. To perform photom-
2 catalog (Høg et al. 2000) within the pointing. This etry on the images of the other filters, the position list
bright star list is used to flag regions in which the sky has to be transformed to the pixel coordinates of these
is heavily contaminated by light from a bright star, and images to an accuracy of about 2 pixels, so that PPP
which must not be used when estimating the sky level isabletodeterminewithoutambiguitythe subpixelcen-
for object detection. At this stage, a simple empirical troidsofthe objects. Forimagesthatareoffsetfromthe
relation between the observed size of the star halo and parent r image by less than 10 pixels, or, images whose
the cataloged magnitude of the star is used to set the offset can be estimated to an accuracy of better than
mask size (which is set to be conservatively larger than 10 pixels from their approximateWCS solution(§3.1), a
mightbe needed). Later in the pipeline, a more detailed simpleandveryfastalgorithmisusedtoregisterthepo-
model of the properties of the reflection haloes around sitionfile. Here,35to70brightestnon-saturatedobjects
bright stars is used to delimit the area for object find- areidentifiedinther imageusingaquick-findalgorithm
ing and to set mask flags in the final catalog (see §9.1). and used as position reference objects. Local maxima
Themaskingofsaturatedcolumnsisdonebyidentifying in the daughter frame within 10 pixels of the expected
andconnectingpixelsabovethesaturationvaluestarting positions of the reference objects are located and used
fromthe positions ofthe brightstars. The maskedareas to derive average shifts in x and y. A rotation, if one
are used neither in estimating detection threshold, nor existsandissufficientlysmall(<∼1degree),isalsomea-
in object finding. sured by comparing vector angles between pairs of ob-
Objectdetectionotherwisefollowsthe samebasicpro- jects. This transformation is performed on the position
cedure as described in Gladders & Yee (2005). Peaks list, and photometry carried out on the daughter frame.
are identified as significant enhancements measured in Occasionally, registration of the object positions for
a 3×3 tapered box more than 2.6σ above the local sky framesfortheotherfiltersofapointingisnotsostraight-
(excludingmaskedregions,asdescribedabove). Twocri- forward due to a significant rotation and even a small
teriabasedonminimumnumberofconnectedpixelsina scale change. This could occur when MegaCam is re-
smoothed image and the ‘sharpness’ of the candidate in moved and re-installed, or the optics of the camera are
the unsmoothed image are used to eliminate cosmic ray changed (as in the case when one lens was inverted) or
and noise spike detections. a filter is replaced(as in the caseof the i-bandfilter, see
§7). In these cases, more CPU-intensive algorithms are
4.2. Photometric measurement used to deal with the registration of the position files.
Total photometry of an object is derived based on the The basic algorithm is a brute-force technique of find-
growth curve of the object which is measured in a se- ing the best χ-square fit of the two sets of bright po-
ries of concentric circular apertures around the object, sition reference objects, stepping through small incre-
maskingnearbyobjectsasrequired(seeYee 1991forde- ments of xy shifts, rotation, and scale. The algorithm
tails). The growthcurveis initially measuredto 8.5′′ for stepsthroughincreasinglymoreparametersintheabove
all objects. The center of the apertures is determined order, so as to minimize computational time. PPP per-
by an iterative procedure that is accurate to fractional forms the overall registration process by starting with
pixels (down to better than 1/10 of a pixel, depending the simplest procedure, tests the validity of the trans-
on the signal-to-noise ratio of the detection). To allevi- form (based on whether the transformed positions from
atethe problemofpixelationatsmallradii,eachpixelis the r frame match the actual objects on the other filter
subdividedinto7×7subpixelsbeforeintegratingtheflux frames close enough for PPP to compute an accurate
within anaperture. The photometric curve of growthso fractional pixel centroid), and if not, performs the next
produced is used to identify an optimal aperture within more complicated matching procedure. In most cases,it
whichto measurethe totalmagnitude followingthe pro- is the i-band images (discussed in §7), sometimes taken
cedure described in Yee (1991). If the optimal aperture a year or more apart from the others, and comprising
of the object is smaller than the standard aperture of imagestakenthroughtwo differentfilters overthe whole
8.5′′ diameter, the magnitude within the optimal aper- survey, that require the more time-consuming registra-
ture is extrapolated to the standard aperture using cor- tion procedures.
rections derived from the shape of the growth curves of With growth curves for each object measured in all
bright point-source objects. The standard aperture size filters, colors are measured using a smaller aperture to
is chosen to include close to 100% of the light for small improve the signal-to-noise ratio by minimizing the sky
′′
objects under all our seeing conditions. For brighter re- noise. The “color aperture” used is either a 3 diameter
solved objects, which normally would still have increas- aperture or the optimal aperture, whichever is smaller.
ing flux at the 8.5′′ aperture, growth curves extended The same aperture is used for each filter, taking the r-
to a maximum diameter of 25′′ are measured and used band as the reference. No correction is made for seeing
in determining the optimal aperture to make sure that differencesbetweenthe bands since the seeingvalues are
the bulk of the light is included. We note that objects typically comparable and the aperture is already rela-
withanoptimalaperture smallerthan1.5′′ areclassified tively large compared to the PSF. The total magnitude
as non-detections. The error on the total magnitude is for each of the other filters is then calculated by adding
then calculated as the sky noise within the photometric therelevantcolortothetotalmagnitudeestimatedinthe
aperture used, which is sky noise-limited for the faint referencefilter(r). Thecolorerroristhenthequadrature
galaxies of interest here. sumoftheerrorsinthetwocoloraperturesusedtocom-
pute the color. Both color errors and total magnitude
11 seehttp://vizier.hia.nrc.ca/ errors are propagated through to the final photometric
8 Gilbank et al.
Accuratephotometriccalibrationisoneofthemostim-
1500
g portantaspectsofanyimagingsurvey. ForRCS-2,given
r
i the color-selected nature of our galaxy cluster sample,
z
obtaining the highest possible accuracy in the calibra-
1000 tion of each color is particularly important. Here we
describe the technique we employ to calibrate the colors
N of galaxies as accurately as possible using the colors of
stars in our fields.
500 The photometric catalogs generated by PPP all as-
sume a single photometric zeropoint for all of the obser-
vations made in each of the three filters. Some of our
observations were made in mildly non-photometric con-
0 ditions,aconditionweallowedafterinitialtestswithour
0.2 0.4 0.6 0.8 1.0 1.2
FWHM (arcsec) color-calibration technique demonstrated that we could
achieve accurate calibrationwithout the use of standard
Fig.4.— The distribution of seeing values (one value for each
chip)ineachfilter. stars. Only mildly non-photometric conditions were al-
lowed so as to not unduly decrease the depth of the ob-
servations, but this still increases the number of usable
catalogs.
nights available to our project, since we do not require
A 5-σ reference limiting magnitude for eachchip is es-
photometric conditions.
timated by scaling the flux of a set of bright reference
InRCS-1,the(R −z′)colorcalibrationofeachpoint-
starsuntilthey reacha signal-to-noiseleveloffive. Typ- C
ingwasperformedbyrequiringthecolorsofbrightgalax-
ically for extended sources,the 100%completeness limit
ies to agree with a reference pointing. The R -band ze-
is about 0.6 to 0.8 magnitudes brighter than this limit C
ropoint was then set by requiring the number counts of
(Yee 1991). An estimate of the seeing is made by mea-
galaxiestoagreefrompointing-to-pointing. Inprincipal,
suring the average FWHM of the reference PSF stars
the uniformity of the colors of stars or galaxies may be
identified in each chip. The object catalogs typically
used. However, since we are primarily interested in ex-
comprise around 4000 objects per CCD chip. The val-
tragalactic studies with RCS-2, we find it preferable to
ues of seeing measured by PPP, one value per chip, are
force-fitthepropertiesofstarstoagreefrompointing-to-
shown in Fig. 4. The median values of the seeing are
[g,r,i,z] = [0.79′′,0.71′′,0.53′′,0.67′′], but note that the pointing in our survey,decoupling any calibrationerrors
fromthe propertiesofthe galaxieswewishto study. For
distributions are broad due to our two-tier strategy. A
example, forcing the number counts of galaxies to agree
significant fraction of our data possesses seeing better
than <0.6′′in r. fromfieldtofieldthenmakesitimpossibletostudyden-
∼
sityvariationsofgalaxiesonscalescomparabletothesize
ofeachfield,sincewewouldhave,byconstruction,forced
4.3. Star-Galaxy Classification
all these values to agree. Thus in RCS-2 we will use the
Star-galaxy separationis performed by comparing the colors of stars to accurately set the calibration of each
shapeofthegrowthcurveofeachobjecttothe weighted color, and then calibrate the value of the magnitudes in
average growth curve from a set of four to eight refer- one reference band by comparison with an overlapping
ence PSFs from nearby unsaturated stars (see Yee 1991 reference survey (2MASS).
and Yee et al. 1996) chosen from a list of automatically
identifiedreferencePSFsfromthe samechip. Thelistof 5.1. Fitting the stellar locus
identifiedreferencePSFsiscreatedfromallbrightunsat- We use the uniformity of the colors of Galactic stars
urated objects that do not have close neighbors and are in order to set the color calibration from pointing-to-
deemed‘notfuzzy’byanalgorithmthatiterativelyelim- pointing in RCS-2. This method is an extension of the
inates objects that are ‘fuzzier’ than the average. Each idea used by Hsieh et al. (2005) to calibrate multicolor
objectisthengivenaclassificationof0–artifact/spurious (B, V) follow-up observations of RCS-1 fields. Hsieh
object ; 1 or 2–galaxy; 3–star; 4–saturated (Yee 1991). et al. (2005) constructed histograms of objects classified
The subdivisions of types 1 and 2 for galaxies is for his- asstellarandrequiredthattheir(z′−R ),(V−R )and
C C
torical reasons related to the development of PPP and (B −R ) distributions agree from pointing-to-pointing
C
nodistinctionis madeinpractice. Anobjectisclassified (holding the R -band calibration fixed).
C
assaturatedifoneormorepixelsisabovethesaturation OurmethodisinspiredbyIvezi´cetal.(2004)whoused
ADU value as provide in the Elixir FITS header of the the uniformity of the stellar locus in high-Galactic lati-
image. tude fields inSDSS to assessthe accuracyoftheir (inde-
Basedontestsusingsimulatedimages,thestar-galaxy pendent)photometriccalibration(usingstandardstars).
classificationisrobustuptothe100%completenesslimit They showed that, for high latitude (|b| > 15 degrees)
under typical seeing conditions (Yee 1991). At magni- fields, the colors of faint stars are essentially identical
tudes fainter than 5σ, star-galaxy classification is per- and are dominated by stars in the Galactic halo (and
formed on a statistical basis using the “variable classi- thus behind the majority of the Milky Way dust). They
fier” procedure based on the relative sharpness (rather showedthatonecandefineprincipalcolorsalongthestel-
than absolute sharpness) of the objects as described in lar locus, and that constructing a histogram along these
Yee (1991). principal colors results in a Gaussian distribution where
thewidthisdominatedbytheintrinsicwidthofthelocus
5. PHOTOMETRICCALIBRATION (∼0.02mag)ifthephotometricerrorsaresmall. Thuswe
RCS2 survey construction 9
combine the idea of principal stellar colors for checking
3 200
the accuracy of photometry (Ivezi´c et al. 2004) with the
150
color histogram-matching technique (Hsieh et al. 2005)
to form a method for calibrating imaging surveys from 2 N100
the principal colors of stars.
50
Our procedureis as follows. We begin by constructing
a reference sample. Since our primary aim is a highly r−z 1 −01.5 (g−r)0−.(0g−r) 1.5
uniformcolorcalibrationacrossourentiresurvey,wede- 0
400
sign a calibration set using only MegaCam data to cal-
300
ibrate our filters internally to the native MegaCam sys- 0
tem. We can then transform our (uniformly-calibrated) N200
catalogs to any other (standard) system afterwards. To
100
builda referenceset, weselecta subsampleofaround20 −1
0.0 0.5 1.0 1.5 2.0 0
RCS-2pointingswhichoverlapwithSDSS.Bothdatasets g−r −0.5 0P.10 0.5
are corrected for Galactic Reddening using the map of
Fig.5.—Mainpanel: color–colordiagramofasubsampleofstars
(Schlegel et al. 1998), since the majority of the (|b|>15 used in the reference dataset. Dashed lines indicate the positions
degrees)starsliebehindtheMilkyWaydust(Ivezi´cetal. oftheprincipalcolorsused. Theverticallineisthe(g−r)principal
2004). (All our fields lie at|b|>30 degrees,with the ex- colorwhichissimply(g−r)−(g−r0)where(g−r)0isthereference
color. A histogram of this principal color is shown in the upper
ception of the CFHT Legacy W2 field which is |b| > 20
right panel. The lower right panel shows a histogram of the P1
degrees.) For each pointing we match individual objects principal color, described inthe text and indicated as the sloping
with their counterparts in SDSS and work out the me- dashedlineinthemainpanel. Seetextfordetails.
dian offset from SDSS for that pointing and filter. For
The (g−r) color cut ensures that only the blue branch
example, for the ith pointing in the r-band filter, the
of the stellar locus is used. This results in a very clean
natively calibrated magnitude, r , is related to the
i,CAL Gaussianinthehistogram,shownasthelowerrightpanel
instrumental magnitude produced by PPP, r , by
i,instr in Fig. 5.
the offset dr by
i
5.2. Application of the fitting procedure
r =r +dr (1)
i,CAL i,instr i
With the above reference set in-hand, each science
where
pointing may be examined in turn and its star colors
dri =median(rij,SDSS −rij,inst) (2) forced to agree with that of our reference set. All ob-
jects classified as unsaturated stars with magnitude er-
and r and r refer to the jth object matched
ij,SDSS ij,inst rors <0.2 mag are extracted and color histograms built
between SDSS and MegaCam respectively. Now, for the
inthe same wayas for the reference sample. The (g−r)
r-band, the colortermbetween MegaCamr andSDSS r
histograminthesciencepointingiscross-correlatedwith
isnegligible,sothistransformationeffectivelytransforms
the of the reference pointing, and the offset ∆(g −r)
1
onto the SDSS AB system. For the other filters there
found. This procedure is typically found to be accurate
are color terms. Adopting the above procedure for all
to a few hundredths of a magnitude (≈ 0.05), by com-
filters (i.e., neglecting to specifically apply a color term)
paringtheresultwithanindependentcalibrationsuchas
means that our reference set is calibrated onto a native
SDSS.Inordertoimprovethisaccuracyfurther,aGaus-
MegaCam system which agrees with SDSS for objects
sianprofileisfittediterativelyaroundthepositionofthe
with the colors of the median-colorobjects in SDSS. We
peak of the histogram, and the peak position and width
derive the color terms which may be applied to convert
recorded. The offset between the Gaussian-fitted peak
the RCS-2 native system to SDSS AB at the end of this
andzeroisarefinedestimateofthecoloroffset,∆(g−r) .
2
section.
We can now take each (uncalibrated) science point- ThewidthofthefittedGaussian,σ(g−r),isusedtocheck
for problems with the photometry/calibration method.
ing in turn and calibrate it to agree with the above set.
Thewidthofthedistributionissetbytheintrinsicwidth
Wedothisforeverypointing,includingthosewhichwere
of the stellar locus convolved with our photometric er-
usedinconstructingthereferencesample,toensurecom-
rors, and thus should be approximately constant from
plete consistency across the survey. For calibrating g, r,
z data we will calibrate the (g −r) and (r −z) colors, pointing-to-pointing. If σ(g−r) > 0.15 mag then the
pointingisflaggedashavingpotentialproblemswiththe
initially holding the r-band magnitude fixed. We start
photometry. The inflated width of the Gaussian usu-
with the (g−r) color since the red branch of the stellar
ally means that the (g−r) colors are smeared out due
locus is almost independent of (r−z) color (Fig. 5, see
to the the g and r frames not having been registered
also fig. 2 of Ivezi´c et al. 2004).
correctly (occasionally due to the wrong pointing posi-
The (g−r) principal color is obtained by simply sub-
tion being observed by the queue observers). A similar
tracting the (g−r) color of the peak of the (g−r) his-
warning flag can be generated by flagging unusually low
torgam,whichwedenote(g−r) ,where(g−r) =1.289.
0 0 amplitudes of the Gaussian peak (since the number of
The resulting histogram, now centred on zero color, is
stars in each pointing is approximately constant). Such
shown as the upper right panel in Fig. 5.
calibration problems are later inspected manually.
The principal color used to calibrate (r−z) is defined
The calibration in (g−r) is thus
by fitting a linear relation to the blue part of the color–
color diagram (Fig. 5): (g−r) =(g−r) +∆(g−r) +∆(g−r) , (4)
CAL inst 1 2
P1=(r−z)−[0.779(g−r)−0.152]∩(g−r)<1.158. (3) where ∆(g − r) is the offset determined by cross-
1
10 Gilbank et al.
correlation of the color histograms, and ∆(g − r) is
2 6
the offset from the peak of the iteratively-fitted Gaus-
sian. Inthe caseofaGaussianwidthσ(g−r) >0.15mag, 5
∆(g−r) issettozeroandawarningflagset. Inasmall
2 4
minority of cases, this warning flag is caused by some
problem other than bad registrationof the images (such 3
as contamination by satellite trails). In this case, if vi- g−J
sual inspection of the calibration histograms reveals no 2
obvious problem, then the calibration is accepted using
1
just the cross-correlationoffset, ∆(g−r) .
1
Asimilarprocedureis appliedin(r−z),butusingthe 0
principal color P1 (Eqn. 3) when constructing the color
histogram. Note that the (g−r) color cut used with P1 −1
0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6
to select the blue branch of the stars gives a very sharp, g−r
clean histogram in (r−z). This can be done accurately Fig.6.— Thereference pointing used tomeasure the (g−J)−
since the (g −r) color has been calibrated in the pre- −(g−r)locusbycombining2MASSandRCS-2photometry. The
vious step. Again, a cross-correlation offset followed by blue branch in (g−r) exhibits the smallest scatter and so this is
fittedwiththefunctioninEqn.5(dashedline).
a Gaussian fit to the located peak is performed and the
same check, σ <0.15 mag, made of the results.
P1 filter, defining the r-band shift as zero, ∆g = ∆(g −
r) and ∆r = 0. The calibration offset measured from
5.3. Magnitude calibration
2MASS, ∆g , gives resulting overall corrections to
2MASS
The above procedure results in very accurate calibra- the instrumental magnitudes of
tion of the (g−r) and (r−z)colors (<0.03 mag). How-
∼
ever, the magnitude in any individual filter has not yet ∆gtot =∆g+∆g2MASS (6)
been accurately calibrated and may be considerably in
and
error by several tenths of a magnitude (the data may
∆r =∆r+∆g (7)
have been observed through thin cloud, and even if not, tot 2MASS
wehavemadenoexplicitadjustmentforatmosphericex- to preserve the calibrated (g −r) color, and where ∆r
tinction). Operationallywe held the r-band fixedabove, was above defined as zero.
but in order to calibrate the individual magnitudes it is Similarly, for z the offset would be
sufficient to calibrate any one of the individual filters,
∆z =∆z+∆g , (8)
providedwe hold the colors fixed to their calibrated val- Tot 2MASS
ues. To obtain a calibration, we need a survey which
since its calibration was performed in (r−z) and r has
overlaps with objects in our sample and possesses well-
just had ∆g added to it.
2MASS
calibratedphotometry. SDSSwouldbeideal,butitdoes
Just to reiterate the procedure, we take data, which
not overlap with all of our pointings (or even all of our
may have been acquired in non-photometric conditions,
patches). So, we use NIR data from 2MASS12. J-band and obtain an accurate photometric calibration simply
total magnitudes from the 2MASS point source catalog by forcing the colors of stars to agree with each other
are downloaded for objects in the RCS-2 footprint. A frompointing-to-pointing(whichensureshighly uniform
reference set is created again using the SDSS-calibrated colorcalibration)andobtaina magnitude calibrationby
pointings, as above. Thus, we can construct color–color forcingthesecolorstoagreewithareferencesetincluding
diagrams expected for stars in any combination of our J-band magnitudes from 2MASS in one of the colors.
(+2MASS)observedfilters. Weuse(g−J)versus(g−r) This procedure negates any need for separate correction
asshowninFig.6. Asmentionedabove,weonlyneedto ofobservationaleffects suchasexposure time differences
calibrate one filter. Although z-band is closer in wave- and atmospheric extinction. We note that the idea of
lengthtoobservedJ-band,itdoesnotmatterwhichcom- usingstarstoprovideortestaphotometriccalibrationis
bination is used, since the fit is performed in color–color notanewone(indeed,Highetal.2009recentlypresented
space and the locus is just as well defined in these col- animplementationofasimilartechnique),butwebelieve
ors.13 we are the first to implement such a method as the sole
For (g−J) the principal color used is: means of calibrating a survey of this size.
Itisworthemphasisingthatsincethestarsusedinthe
P2=(g−J)−[2.226(g−r)−0.764]∩(g−r)≤1.2. (5)
stellarlocus-fittingarebehindthemajorityoftheGalac-
The offsets, calculated as above, result in a g-band tic dust, and we have reddening-corrected our reference
shift, ∆g, which must be applied to each filter in such a dataset,the magnitude offsets measuredbetweeninstru-
wayastopreservethecalibratedcolors;i.e.,ifwere-write mentalmagnitudes and the reference set will correctthe
theoriginalcolorshift(Eqn.4)asshiftstotheindividual data to a system already corrected for Galactic Extinc-
tion. We note that in practice we prefer to both redden-
12 WethankourWiggleZcollaboratorsfororiginallysuggesting ingcorrectourinstrumentalsciencecatalogsandourref-
thisidea. erence dataset before performing the calibration. Thus,
13 Thereasonforchoosing g isthat thisisthelastofthethree the magnitude differences, ∆mag (e.g., Eqn. 4), can be
filtersonwhichPPPperformsphotometry, sousingitasthecali-
directly applied to the initial instrumental magnitude
bration filter is useful forcatching problems withthe photometry
– if the last filter (g) fails, then all the photometry fails for that catalogs(i.e., not correctedfor Galactic Extinction) and
pointing. thiscorrectsthecataloginitiallytoobservedratherthan