The DAFT/FADA survey

15 Pages·2010·2 MB·English

A&A523,A21(2010) Astronomy
DOI:10.1051/0004-6361/201015174 &
(cid:2)c ESO2010 Astrophysics
The DAFT/FADA survey
I. Photometric redshifts along lines of sight to clusters
(cid:2)
in the z=[0.4,0.9] interval
L.Guennou1,2,C.Adami1,M.P.Ulmer2,1,V.LeBrun1,F.Durret3,4,D.Johnston6,O.Ilbert1,D.Clowe5,13,
R.Gavazzi3,4,K.Murphy5,T.Schrabback10,S.Allam6,J.Annis6,S.Basa1,C.Benoist7,A.Biviano8,A.Cappi9,
J.M.Kubo6,P.Marshall11,12,A.Mazure1,F.Rostagni7,D.Russeil1,andE.Slezak7
1 LAM,OAMP,Pôledel’EtoileSitedeChâteau-Gombert,38rueFrédéricJoliot-Curie,13388MarseilleCedex13,France
e-mail:[email protected]
2 DepartmentPhysics&Astronomy,NorthwesternUniversity,Evanston,IL60208-2900,USA
3 UPMCUniversitéParis06,UMR7095,Institutd’AstrophysiquedeParis,75014Paris,France
4 CNRS,UMR7095,Institutd’AstrophysiquedeParis,75014Paris,France
5 DepartmentofPhysicsandAstronomy,OhioUniversity,251BClippingerLab,Athens,OH45701,USA
6 FermiNationalAcceleratorLaboratory,POBox500,Batavia,IL60510,USA
7 OCA,Cassiopée,Boulevarddel’Observatoire,BP4229,06304NiceCedex4,France
8 INAF/OsservatorioAstronomicodiTrieste,viaG.B.Tiepolo11,34143Trieste,Italy
9 INAF-OsservatorioAstronomicodiBologna,viaRanzani1,40127Bologna,Italy
10 LeidenObservatory,LeidenUniversity,NielsBohrweg2,2333CALeiden,TheNetherlands
11 KavliInstituteforParticleAstrophysicsandCosmology,StanfordUniversity,2575SandHillRoad,MenloPark,CA94025,USA
12 PhysicsDepartment,UniversityofCalifornia,SantaBarbara,CA93601,USA
13 AlfredP.SloanFellow
Received8June2010/Accepted9August2010
ABSTRACT
Context. As acontribution to the understanding of the dark energy concept, the Dark energy American French Team(DAFT, in
French FADA) has started a large project to characterize statistically high redshift galaxy clusters, infer cosmological constraints
fromweaklensingtomography,andunderstandbiasesrelevantforconstrainingdarkenergyandclusterphysicsinfutureclusterand
cosmologicalexperiments.
Aims. Thepurposeofthispaperistoestablishthebasisofreferenceforthephoto-zdeterminationusedinalloursubsequentpapers,
includingweaklensingtomographystudies.
Methods.Thisprojectisbasedonasampleof91highredshift(z≥0.4),massive(>∼3×1014 M(cid:5))clusterswithexistingHSTimaging,
forwhichwearepresentlyperformingcomplementary multi-wavelengthimaging. Thisallowsusinparticulartoestimatespectral
types and determine accurate photometric redshifts for galaxies along the lines of sight to the first ten clusters for which all the
requireddataareavailabledowntoalimitofI = 24./24.5withtheLePharesoftware.Theaccuracyinredshiftisoftheorderof
AB
0.05fortherange0.2≤z≤1.5.
Results. Weverifiedthatthetechniqueappliedtoobtainphotometricredshiftsworkswellbycomparingourresultstowithprevious
works. In clusters, photo-z accuracy is degraded for bright absolute magnitudes and for the latest and earliest type galaxies. The
photo-zaccuracyalsoonlyslightlyvariesasafunctionofthespectraltypeforfieldgalaxies.Asaconsequence,wefindevidencefor
anenvironmentaldependence ofthephoto-zaccuracy, interpretedasthestandardusedspectralenergydistributionsbeingnotvery
wellsuitedtoclustergalaxies.Finally,wemodeledtheLCDCS0504masswiththestrongarcsdetectedalongthislineofsight.
Keywords.surveys–galaxies:clusters:general–galaxies:distancesandredshifts–cosmologicalparameters–darkenergy
1. Introduction
Thediscoverytenyearsagooftheaccelerationoftheexpansion
(cid:2) Based on observations made with the NASA/ESA Hubble Space oftheUniverse(Riessetal. 1998)whichistypicallyexplained
Telescope, obtained from the data archive at the Space Telescope byassumingthatmostofitsenergyisintheformofanunknown
Institute and the Space Telescope European Coordinating Facility. darkenergy(DE),isoneofthemostpuzzlingissuesofmodern
STScI is operated by the association of Universities for Research in cosmology.Effortshavethereforebeenundertaken,suchasthe
Astronomy, Inc. under the NASA contract NAS 5-26555. Also based
DarkEnergyTaskForce(Albrechtetal.2006)ortheESA-ESO
on observations made with ESO Telescopes at Paranal and La Silla
workinggrouponfundamentalphysics(Peacocketal.2006)to
Observatories under programme ESO LP 166.A-0162. Also based
designprojectstomeasureDEanddetermineitsnature.Ashigh-
on visiting astronomer observations, at Cerro Tololo Inter-American
Observatory, National Optical Astronomy Observatory, which is op- lightedbythesereports,understandingDErequiresbigsurveys
erated by the Association of Universities for Research in Astronomy, toovercomecosmicvarianceandshotnoiseaswellasnewex-
undercontractwiththeNationalScienceFoundation. perimentstocontroltheunknownsystematicuncertainties.
ArticlepublishedbyEDPSciences Page1of15
A&A523,A21(2010)
Table1.DecimaldegreesJ2000coordinates,observedbands,CTIOexposuretimes,andredshiftforeachofthetenconsideredclusters.
Clustername RA Dec Band Exposuretime z 90%completenesslevel
deg deg s F814Wmagnitude
LCDCS0110 159.464 −12.724 B 11×600 0.58 26.2
LCDCS0130 160.168 −11.934 B 11×600 0.70 26.2
LCDCS0172 163.601 −11.772 B 11×600 0.70 26.2
LCDCS0173 163.681 −12.764 B 11×600 0.75 26.2
CLJ1103.7-1245a 165.895 −12.780 B 11×600 0.63 26.2
LCDCS0340 174.542 −11.560 B 11×600 0.48 26.2
LCDCS0504 184.189 −12.022 B 11×600 0.79 26.2
LCDCS0531 186.995 −11.587 B 11×600 0.64 26.2
LCDCS0541 188.126 −12.8434 z(cid:7) 18×500 0.54 25.8
R 8×600 26.6
LCDCS0853 208.541 −12.517 B 11×600 0.76 26.2
In this context, galaxy clusters, together with several other immediategoalofthispaperisthentodescribethesephotomet-
probes, are expected to play a major role (e.g. Nichol 2007). ric redshift measurementson the first 10 completed clusters in
These objects have indeed long held a place of importance in oursample.Thiswillallowusinthefuturetocombinephoto-zs
astronomy and cosmology. (e.g. Zwicky 1933) inferred from withweaklensingshearmeasurementsbothtocarryouttomog-
observations of the Coma cluster that the matter in our uni- raphyandtobuildmassmodelsforclusters.Thispaperwillalso
versecouldbeintheformofadarkcomponent(thiscomponent provide the foundationfor other future works that will use the
was first supposed to be low surface brightness diffuse light). photo-zsproducedbytheprocessdescribedheretostudycluster
The measurementof the baryonfractionin X-rayclusters (e.g. galaxypopulations.
Lubin et al. 1996; Cruddace et al. 1997), combined with big- Throughout the paper we assume H = 71 kms−1Mpc−1,
0
bang nucleosynthesisconstraints allowed to put an upper limit Ωm = 0.27, and ΩΛ = 0.73. All magnitudes are in the
on the matter density of the Universe. The resulting value was ABsystem.
considerablylessthan the theoretically-favoredcriticaldensity.
Cluster number counts (e.g. Evrard 1989) and cluster correla-
tion functions (e.g. Bahcall & Soneira 1983) have been used 2. Observations
to constrain the amplitude of mass fluctuations and strengthen
support for the Cold Dark Matter (CDM) structure formation The 10 clusters for which we produced photometric red-
paradigm.Galaxyclusterscanalso be used to testthe redshift- shifts (hereafter photo-zs) were originally observed as part of
distance relation (e.g. Supernovaeas standard candles, Baryon the EDisCS program (e.g. White et al. 2005), but new data
AcousticOscillationsorweaklensingtomographywithclusters, (mostly B, but also R and z(cid:7), see Table 1) were obtained and
e.g.Hu1999)orthegrowthofstructuresthroughweaklensing, some data were also collected from the literature (V, R, I, z(cid:7),
clusternumbercounts,orintegratedSachs-Wolfeeffect.Clusters F814W,SpitzerIRAC) tocompletethedatasetinordertocal-
are also intrinsically interesting in many aspects, including the culate more accurate photo-zs. We thus created a full data set
influence of environment on galaxy formation and evolution. with BVRIz(cid:7), HST ACS F814W,andSpitzer IRAC 3.6μm and
Building a detailed picture of galaxy and large-scale structure 4.5μm(channels1and2).Figure1showsthespectralcoverage
growth (e.g. clusters) is therefore necessary to understandhow achievedwiththissetoffilters.
theUniversehasevolved.
TheDarkenergyAmericanFrenchTeam(DAFT,inFrench
2.1.HSTACSdata
FADA) has started a large project to characterize statistically
high redshift galaxy clusters, infer cosmological constraints We have retrieved from the HST archives data for 10 EDisCS
fromweak lensing tomography,andunderstandbiases relevant clustersobservedwiththeACSintheF814Wfilter,eachimage
for constraining DE and cluster physics in future cluster and including4tiles(2×2mosaic)of2ksandacentraltileof8ks
cosmological experiments. This work is based on a sample of (Desaietal.2007).Theachieveddepthforpointsourcesatthe
91highredshift(z = [0.4;0.9]),massive(>3×1014 M(cid:5)) clus- 90%levelisoftheorderofF814W ∼ 28forthedeeppartsand
terswithexistingHSTimaging,forwhichwearepresentlyper- F814W ∼ 26 for the shallow parts (see Fig. 4)1. The full data
formingcomplementarymulti-wavelengthimaging.Thiswillal- reductiontechniqueisdescribedinSchrabbacketal.(2010)and
low us in particular to estimate accurate photometric redshifts willbeexpandedinacompanionpaper(Cloweetal.,inprepa-
for as many galaxies as possible. The requested accuracy de- ration), but we give here the salient points. The data were re-
pends on both our ability to discriminate between cluster and ducedusingamodifiedversionoftheHAGGLeSpipeline,with
backgroundfieldgalaxieswithoutloosingtoomanyobjectsand careful background subtraction, improved bad pixel masking,
on the weak lensing tomography method internal parameters. and proper image registration. Stacking and cosmic ray rejec-
Catalogs of cluster galaxies (e.g. Adami et al. 2008) typically tionweredonewithMultidrizzle(MD)(Koekemoeretal.2002),
show photometric redshifts spanning a total (∼3σ) interval of takingthetime-dependentfield-distortionmodelfromAnderson
±0.15 in photo-z. This means that the goal of our survey is to et al. (2007) into account. The pixelscale was 0.05arcsec and
havephotometricredshiftswitha1σprecisionbetterthan0.05. weusedaLanczos3kernel.Afteraligningtheexposuresofeach
With such a precision, the photo-z uncertainties would not be
the expected dominant source of errors in our method, except 1 Consideringtotalmagnitudesandanalysisbeingperformedatthe1.8
whenconsideringlensing andlensed objectsat redshiftgreater andthe3sigmaSextractorlevelrespectivelyforthedeepandshallow
than 0.8 and closer than 0.4 along the redshift direction. The partstolimitfakeobjectdetections.
Page2of15
L.Guennouetal.:TheDAFT/FADAsurvey.I.
Fig.1.Transmissioncurvesoftheavailablesetsoffilters.Upperfigure:
infraredfilters(Irac1:cyan,Irac2:blue).Lowerfigure: visiblefilters Fig.2.ResultingastrometricdistortionmapforoneofourBbandobser-
(fromlefttoright:B,V,R,Iingreen,F814Winyellow,andz(cid:7)). vations.LargeblackrectanglesaretheindividualMOSAICCCDs.Blue
toredcolorsshow0.25to0.27arcsecastrometryresidualuncertainties.
0.54440
tileseparately,shiftsandrotationsbetweenthetilesweredeter-
minedfromseparatestacksbymeasuringthepositionsofobjects
intheoverlapregions.Asfinalstep,mosaicstacksincludingall 0.39150
tiles of onecluster were created. We set these ACS mosaicsas
0.95340
astrometric references for ground-based data. Our first results
basedonweaklensingmeasurementsandweaklensingtomog-
0.54960 0.54190
raphywillbedescribedinacompanionpapers(Cloweetal.,in
preparation).
0.54240
0.53440
(cid:7)
2.2.B,Randz groundbaseddata 0.39130
ThenewB,Randz(cid:7)bandobservationspresentedherewerecon- 0.54700
ducted at the CTIO Blanco telescope using the multi-CCD de-
viceMOSAIC(seeTable1).Exposuretimeswerecomputedto
reachanexpecteddepthofF814W ∼24.5(AB)atthe10σlevel. Fig.3.LCDCS0541tricolorimage(50arcsec×30arcsec)madewith
The seeing was on average about 1.1 arcsec, 0.7 arcsec and theB(CTIO:shownasblue),F814(HST:shownasgreen)andz(cid:7)(VLT:
0.9 arcsec for the B, R and z bands respectively. During the shownasred)filters.Knownspectroscopicredshiftsarealsoshown.
observations, we followed a regular dithering pattern with an
amplitudeof 5−10arcsec to improvecosmetics(e.g.inter-chip
separations)of the finalimages.Thestandardstar fieldsSA98,
2.4.Groundbaseddatareduction
SA104,SA107aandSA107bwerealsoregularlyobserveddur-
ing the nights (3 standard stars per night). This allowed us to Aftertheclassicalreductionscheme(offsets,flatfields,etc.),we
deriveextinctioncurvesfortheBlancositeandperformphoto- realized that the gains between the different MOSAIC CCDs
metriccalibrations.
were notinitially verywell constrained.To solve thisproblem,
weobservedaSDSSfieldinallthebands,whichallowedusto
adjustthesegains.Thesecorrectionswereinmostcasessmaller
2.3.Tools
than10%.
For the image reduction, we used the MIDAS, SCAMP and The SCAMP and SWarp tools were used to perform the
SWarp (e.g. Bertin et al. 2002; Bertin 2006) packages to pro- astrometry and homogenize internal photometry in order to
duceimageswithcosmicraysandotherimagedefectsremoved, put each of the individual images on a common grid to cre-
andtoproducefinalcalibratedandalignedimages;thesenewly ate a merged image without cosmic rays, with inter-chip gaps
acquireddata havethenbeen combinedwith the previouslyre- filled and CCD defaults erased. This is a commonly employed
duced data. Descriptions of the MissFits, SCAMP and SWarp technique for the CFHT Megacam and CFH12K images (e.g.
software are given in http://www.astromatic.net.We co- McCracken et al. 2003). We usually found it necessary to use
aligned the FITS images in different bands by using SCAMP a third-order polynomial to model the astrometric distortions.
andthencombinedthemto generatepanchromaticimages(see We then generated weight maps which took into account bad
e.g.Fig.3). pixels, overscans and low efficiency areas. Figure 2 shows the
Page3of15
A&A523,A21(2010)
resulting astrometric distortion map obtained for one of the
Bbandobservations.
We then used standard stars observed during the nights to
buildextinctioncurves2.
The CCDs from MOSAIC are affected by some cross talk.
Wecorrectedforthiseffectbysubtractingfromacontaminated
(receiving)CCDthecontaminating(sending)CCDweightedby
a factor given on the web page of MOSAIC (http://www.
lsstmail.org/noao/mosaic/calibs.html). In any case,
thisonlyaffectedbrightmagnitudeobjects.However,consider-
ingthatthenumberofsuchobjectsperCCDwasnon-negligible,
wehadtotakecross-talkintoaccount.
2.5.Completenesslevel
Although completeness per se is not important for our goal of
obtainingphoto-zsofthebackgroundsearchedgalaxies,itisin-
teresting to discuss this issue because net completeness deter-
mines how many galaxieswe will have in the end to carry out
weaklensingtomographyandclusterstudies.Thefirstimportant
parametertoestimateisthecompletenesslevelofourimagesin
eachband.Basedonthisinformationwecanthendeterminethe
magnituderangewhereallourbandscancontribute.Toestimate
thislevel,weransimulationsfortheHSTACSF814Wimages
as in Adamiet al. (2006). Inbrief,the simulationmethodadds
100 artificial stars of different magnitudes to the CCD images
andthenattemptstorecoverthembyrunningSExtractoragain
withthesameparametersusedforobjectdetectionontheorig-
Fig.4. Detection levels for point sources in the LCDCS 0541 fields.
inal images. In this way, the completeness is measured on the
ContinuouslinescorrespondtothedeeppartsoftheHSTACSF814W
original images. We investigated the catalog completeness for
imagesanddottedlinestotheshallowparts.Topfigure:blue: Bband,
point-likesourcesonly.Thecompletenesslevelsinmagnitudes
cyan:Vband,green:Rband,red:Iband,black:F814Wband,magenta:
arethereforean uppervalueforthe realcompletenesslevelfor z(cid:7) band.Bottomfigure:blue:Irac1band,red:Irac2band.Eachcurve
galaxiesofdifferenttypes. isgivenasafunctionoftheF814Wmagnitude.
For the other bands, we computed how many objects de-
tected in the HST ACS F814W images also gave a success-
ful magnitude measurement in the other bands when extracted
90%complete whateverthe consideredband. In the HST ACS
in SExtractor double-imagemode (to increase the depth of the
F814W shallow parts, the same limit ensures that our data are
catalogs). We note that the HST ACS F814W images are the
more than 90% complete for the ground based plus HST ACS
deepestamongallourbandsbyalargefactor.Multiplyingthese
F814W images, and more than 80% complete for the Spitzer
percentages (not always 100% because objects are sometimes data.F814W =24.5thereforeseemstobeareasonablecompro-
so close to the background in a given band that the fluxes
mise between depth and completenesslevel. This value is also
are not significantly positive) by the completeness of the HST
closeto theexpectedvaluegiventhechosenexposuretimes.If
ACSF814Wimagesthemselves,wethereforeestimatethecom-
we acceptlowercompletenesslevelsof theorderof 50%inall
plete detectionlevel of the consideredband (as functionof the thebands,wecanconsidermagnitudesasfaintasF814W =26.
F814Wmagnitude).
Sincewecancarryoutshearmeasurementsdowntoevenfainter
This method also has the following consequence: we must
magnitudes, this completeness result means we will be able to
take into account the varying exposure times in the HST ACS
useasizablefractionofthoseshearmeasurements.
F814Wimages(thecentershavelongerexposuretimesthanthe
Detection levels are also interresting to compute for future
edges).ThisresultsinavaryingmeansignaltonoiseintheHST
uses for a given band in itself, detecting and measuring the
ACS F814W images and we had therefore to adapt the extrac-
fluxes in this band without considering the HST ACS F814W
tionandmeasureSExtractorsignaltonoiseratios.Wetherefore
imageasadetectionimage.Wenotehoweverthatresultingcat-
estimated the pointsource completenessseparately in the deep
alogswouldthenbeshallowerthanpreviously.Wethereforeper-
(10ks)andshallow(2ks)portionsoftheHSTACSF814Wim-
formedthe same simulations we did for the HST ACS F814W
ages.Thisalsogavetwodifferentcompletedetectionlevelsfor imagesfortheB,V,R,I,z(cid:7),Irac1,andIrac2bands(Sextractor
theotherbands.
detection level of 1.8, pointsources). Figure5 gives the corre-
The results are displayed in Fig. 4 for LCDCS 0541 (other
spondingresults.
clusters show very similar results). This shows that consider-
ing point sources brighter than F814W = 24.5 in the HST
ACS F814W deep parts ensures that our data are more than 2.6.Star/galaxyseparation
2 TherespectiveextinctioncoefficientsKare0.202,0.095and0.04for The second step in our processing was to carry out a star-
the B,Randz(cid:7) bands,andthecorrespondingerrorsonzeropointsare galaxy separation. We only applied this task to the HST ACS
0.09inBand0.07maginRandz(cid:7).Observedmagnitudeshavethento F814Wimagesbecausetheyhavebyfarthebestseeing.Forthis
bediminishedbyKtimestheairmass. taskweemployedtheclassicalmethodconsistinginseparating
Page4of15
L.Guennouetal.:TheDAFT/FADAsurvey.I.
Fig.6. Central surface brightness versus total magnitude for the HST
ACSF814WLCDCS0541image.Bluedotsareconsideredasgalaxies,
reddotsasstars,andgreendotsasdefects.
Fig.5. Topfigure: detection levelsfor point sources inthe B,V,R, I,
z(cid:7) bands.ContinuouslinescorrespondtotheEDISCSdataanddashed
linescorrespondtotheCTIOdata.blue: Bband,cyan:V band,green:
Rband,red:Iband,magenta:z(cid:7)band.Bottomfigure:blue:Irac1band,
red:Irac2band.
Fig.7.Percentageofsaturatedstarsclassifiedasgalaxiesbyourmethod
asafunctionofmagnitude(seetext).
starsfromgalaxies(andfromdefects)incentralsurfacebright-
ness versus total magnitude plots. We show in Fig. 6 that 3.1.Generalstrategy
the star/galaxy discrimination is efficient up to F814W ∼ 27
(fainter than the initially expected depth for our multi wave- In order to calculate photo-zs, the data set had to be combined
lengthsurvey). with data available in the literature. We first re-sampled with
SCAMPandSWarpallimagestothepixelsizeoftheHSTim-
Wealsohadtodealwithbrightsaturatedstars.Theseobjects
wereusuallyclassifiedasgalaxiesbyclassicalstar/galaxysepa- ages and the image astrometry was similarly homogenized to
produceanalignmentprecisionoftheorderofonepixelbetween
rationmethods.Sinceweweremainlyinterestedinfaintobjects,
differentbands(1pixel=0.05arcsecafterrealignment).
these saturated stars were notreally a problemin our analysis.
At this step, the very large CTIO Blanco MOSAIC images
For completenessof our description of our work, however,we
(∼30(cid:7)×30(cid:7)),theVLTFORS2images,andtheSpitzerIRACim-
investigatedwhatwasthecontributionoftheseobjects.We ex-
aminedbyeyeallF814W ≤20LCDCS0541objectsanddeter- agesweretruncatedtothesizeoftheHSTACSF814Wmosaics.
mined a real star catalog. From this, we determined which ob- We homogeneouslydetectedandmeasuredobjectswiththe
jectsinthislistwereclassifiedasstarsbyourautomatedmethod SExtractor package in double image mode (Bertin & Arnouts
basedonFig.6andwethendeducedthepercentageofstarsin- 1996). Detection was made on the deepest image (HST ACS
correctlyclassifiedasgalaxies.Figure7givesthesepercentages F814W image), and measurements were made inside the HST
asafunctionofmagnitude.Weclearlyseethatlimitingouranal- KronaperturesonthegroundbasedB,V,R,I,z(cid:7)imagesandon
ysistoF814W ≥19.5probablyensuresthatourgalaxycatalogs theSpitzer Irac1andIrac2 images.Atthisstep, we onlykept
arenotpollutedbystars. objectsdetectedinalltheavailablebands.
We could have degraded the HST images to the resolution
ofthegroundbasedimagesbeforeextractingsources.However,
becausewearedealingwithclustersofgalaxies,thiswouldhave
3. Registrationwithpreviousimagesandresulting resulted in the loss of significant numbers of galaxies in the
photo-zcomputation densecores.Figure8showsthepercentageofrecoveredobjects
when degrading the LCDCS 0541 F814W image to a 1 arcsec
Wenowdescribehowournewdatawerecombinedwiththepre- resolution.We clearly see that ∼40 % of the objects are lost at
viouslyacquireddata,andhowthephoto-zswerecalculated.We amagnitudeofF814W =24.5andthiswouldstronglypenalize
alsodiscussthereliabilityofthesephoto-zs. oursurvey.
Page5of15
A&A523,A21(2010)
Fig.8. Percentage of recovered objects (detection threshold of 1.8 in Fig.9. Simulated magnitude shifts to apply in LePhare as a function
SExtractor)intheHSTACSLCDCS0541imagewhendegradingthe of object sizeand for thevarious bands. Thesolid vertical lineisthe
imagetoa1arcsecresolution,asafunctionoftheF814Wmagnitude. median value of the object sizesfor F814W ≤ 24.5in real data. The
verticaldashedlineshowstheminimalsizeof90%oftheobjects.
3.2.Imageresolution
contributionsto these valuesof the zero pointshifts and of the
shiftsduetothedifferentspatialresolutions.
DetectingobjectsinHSTimageswithaspatialresolutionsome-
times ∼10 times better than other imagesalso has non negligi- Afirsttesttoevaluatetheinfluenceofthespatialresolution
istocomputetheshiftstoapplyinLePharewhendegradingthe
ble consequences. For example, total magnitudes measured in
F814Wimages to a 1 arcsec resolution(close to the resolution
groundbased or in Spitzer imagesare probablynot correctes-
timates: fluxescomputedfrom∼1 arcsec spatial resolutionim- of the ground based images). In this case, only the Irac 1 and
2imagesshouldrequirelargeshiftsbecausetheyhavetheworst
agesinsidetheHSTKronapertureareobviouslyunderestimates
resolution, of the order of 2 arcsec. The values are equal (for
of the total fluxes. However, the main goal of this measure is
LCDCS0541)to0.01(Bband),−0.14(V),−0.18(R),−0.03(I),
not to estimate the total magnitude of the objects but to com-
−0.15(F814W),0.12(z(cid:7)),−0.39(Irac1),−0.44(Irac2).Inthis
putephoto-zs.Theproblemcanthereforebesolvedbyusingthe
case, shiftsare rathersmall andthe onlylargevaluesoccurfor
LePhare photo-z package (e.g. Ilbert et al. 2006). Briefly, the
thetwoIracbands,asexpected.Thismeansthatalargepartof
LePharepackageis ableto compareobservedmagnitudeswith
the shifts applied to magnitudesin LePhare is probablydue to
predicted ones created by templates from the literature as for
differentimageresolutions.
examplein HyperZ (Bolzonella et al. 2000) or e.g. in Rudnick
etal.(2003).Weselectedspectralenergydistributions(hereafter We can also evaluate the effect of different resolutions on
SEDs)fromPollettaetal.(2006,2007)withaCalzettietal.ex- the images with simulations. We generated artificial objects
tinctionlaw(e.g.Calzetti&Heckman1999).ThesearetheSEDs with FWHM varying from to 0.2 to 1.2 arsec and applied the
whichgivethe bestresults. The fittingthen allowsto constrain magnitude measurement process to these images as seen with
simultaneously the redshift and nature of each object (galaxy the HST F814W configuration and as seen in the other bands.
or star), as well as its characteristics such as photometric type Namely,these are B band (seeing of 1.05arcsec), V (0.68arc-
(hereafter T). With the selected class of SED, T varies from 1 sec),R(0.82arcsec),I (0.62arcsec),z(cid:7)(0.54arcsec),andIrac1
to31.Numbersbetween1and7correspondtoellipticalgalax- and Irac 2 (1.9 arcsec). We show in Fig. 9 the differences be-
ies,numbersbetween8and12toS0,Sa,andSbgalaxies(early tweentrueandmeasuredmagnitudesasafunctionofobjectsize
spiralgalaxies),numbersbetween13and19toSc,Sd,andSdm for the various filters. These shifts are most of the time lower
galaxies(latespiralgalaxies),andnumbersbetween20and31to than0.2mag,exceptfortheIrac1andIrac2bands.Attheme-
activegalaxies.LePhareisalsoabletoestimatepossibleshiftsin dian object size, Irac 1 and Irac 2 magnitude shifts are of the
photometricvalues,bycomparingthephotometricandspectro- order of 0.95, in perfect agreementwith the previously quoted
scopicredshiftsusedfortrainingsets,andalltheclustersconsid- shiftsestimatedwithLePhare.
eredinthispaperhavedeepspectroscopiccatalogs(seeFig.10)
of ∼100 redshifts per line of sight (Halliday et al. 2004; and
3.3.Blendedobjects
Milvang-Jensenetal.2008).Shiftsarecomputedfixingphoto-zs
to the spectroscopicvalues and averagingthe residualsin each Objects which are nearby (∼1(cid:7)(cid:7)) but separated in ACS im-
ofthebands. agesmaybeblendedingroundbasedorSpitzerimages.Using
This is useful to take into account internal photometry in- SExtractor in double image mode avoids the incorrectidentifi-
homogeneities between different bands, and also allows us to cation of faint ACS detected objects with incorrect objects in
take into accountthe differentspatial resolutionswith LePhare thegroundbasedorSpitzerimages.Wecalculatethefluxinside
by applying zero point shifts to our magnitudes. The mean theKronapertureinthepoorerspatialresolutionimagesasde-
(over the ten clusters) applied zero point shifts before photo-z terminedfromthe higherspatialresolutionimage(HST) atthe
computations in the present paper are 0.00 ± 0.14 (B band), exact place of the faint object. This procedure limits the cross
−0.22±0.06(V),−0.20±0.07(R),−0.18±0.10(I),−0.34±0.08 talkbetweenthefluxes.Furthermore,weflagtheseblendssowe
(F814W),0.37±0.11(z(cid:7)), −0.86±0.30 (Irac1), −0.92±0.31 candetermineif includingtheobjectsornotchangestheresult
(Irac 2). These values are not negligible, mainly for the Irac 1 wederivefromourphoto-zcatalogsinastatisticallysignificant
and Irac 2 bands. We will now try to estimate the relative manner.However,asweshowbelow,thequalityofthephoto-zs
Page6of15
L.Guennouetal.:TheDAFT/FADAsurvey.I.
LCDCS 0173). We will however quantify this possible effect
laterinthepaper.Wealsoremarkthatcatastrophicerrorsmainly
occurtowardsthehighphotometricredshifts(atz≥1.5)andwe
willdiscussthepossibleconsequencesonoursurveyinthefinal
section.
Thequestionisnowtoknowifphoto-zsofblendedobjects
arealsoacceptable.Wethereforeflaggedallsuchobjectsbyse-
lecting galaxies with a close neighbor (at less than 1.5 arcsec,
given the ground based seeings) and less than 0.5 mag fainter
than the primary object (enough to potentially bias the magni-
tude estimate). Such objects are potentiallypolluted by a com-
parableorbrighterobject(lessthan0.5magfainter,orbrighter).
We then generated Fig. 13 where all such objects from the ten
consideredclustersareshown.Thereducedσis0.08.Theper-
centageofcatastrophicerrorsis∼10%,higherthanforthewhole
sampleofgalaxies(blendedornot).Howeverevenifthesetwo
values(thereducedσandthepercentageofcatastrophicerrors)
arehigherthanforthewholesample,theyremainacceptablefor
ourpurposesandweconcludethatblendingisnotaredhibitory
problemforz≤1.05andF814W ≤23.5.
3.5.Photo-zaccuracyandenvironmentaldependence
Although the first goal for the FADA/DAFT project is to de-
termine the photo-zs of backgroundfield galaxies, it is also of
interesttodeterminetheachievedphoto-zaccuracyandthefrac-
tionofcatastrophicerrorsasafunctionofvariouscluster/galaxy
internalparameters.Inclusters,Adamietal.(2010)havealready
Fig.10. Redshift (top) and F814W magnitude (bottom) histograms of shownwiththeX-rayselectedXMM-LSSsamplethatlatetype
thespectroscopicsample. galaxiestendtoexhibitpoorerphoto-zprecisionthanearlytype
galaxies.Moreover,theXMM-LSSclustersarenotverymassive
structuresandenvironmentaleffectsareperhapsnotasstrongas
for the blends is close to those derived from the non-blended inthepresentlyconsideredclusters.
objects. Withthesamplepresentlyinhand,wedeterminedtheeffect
of galaxy type and magnitude on the photo-z accuracy in the
redshiftrangeforwhichspectroscopicdataexist.Similarly,we
3.4.Photo-zaccuracywiththespectroscopicsample
can investigate the possible effects of the environment(cluster
Given the results of Coupon et al. (2009), our data should be versusfieldregions).
abletoprovidegoodphoto-zsformagnitudesatleastasfaintas We chose to merge the 10 clusters in a single photometric
i(cid:7) = 24(ABmagnitudes)andforredshiftslowerthan∼1.5.We versusspectroscopicredshiftcatalog.Thefollowingresultswill
first estimate the qualityof ourphoto-zsbycomparingthem to thereforeapplyfortheconsideredclusterredshiftrange(z[0.4;
available spectroscopic redshifts. This gives us insight on the 0.9]).We show in Figs. 14 and 15 the variation of the reduced
photo-z quality in the magnitude and redshift ranges covered σ of the photo-zsas a function of photometrictype and of ab-
by the spectroscopic catalogs. We show in Fig. 10 the redshift solutemagnitudeforclustergalaxieswithinaprojectedcluster-
and F814W band magnitude histograms of the available spec- centricradiusof0.5Mpcand1Mpc,andfieldgalaxies.Galaxies
troscopic redshifts along the ten considered lines of sight. The were selected as members of the 1 Mpc radius region if their
photo-zqualityassessmentwillbevalidupto z ∼ 1.05andfor redshiftdifferedby less than 3 times the velocity dispersionof
magnitudesbrighterthan∼23.5andreallystronguptoz∼1and Clowe et al. (2006) comparedto the mean cluster redshift. For
formagnitudesbrighterthan∼23. the 0.5 Mpc region, the limit was set to less than one time the
Wethengeneratedthephoto-zversusspectroscopicredshift velocitydispersionofCloweetal.(2006).
(specz hereafter) plots shown in Figs. 11 and 12. These plots These figures show that we have a worse photo-z accu-
givethedispersionsaroundthemeanrelation(boththeNMAD racy for the brightest cluster galaxies (in absolute magnitude).
reduced sigma of Ilbert et al. (2006)3 and the regular sigma Differencesbetweenbestandworstvaluesrepresentmostofthe
value:second momentof the value distribution), and the mean timeafactoroftwo,ingoodagreementwiththeresultsofAdami
shiftbetweenphoto-zsand speczs.We clearlyseethatonaver- etal.(2010).Thetendencyisclearlydifferentinthefieldwhere
age,photo-zsand speczsareingoodagreementwithareduced thevariationisnotsignificant.Consideringnowthephotometric
σ of the orderof 0.05aroundthe mean relation(regularsigma typeT,weseeinFig.14theworsephoto-zaccuracyforthelat-
oftheorderof0.09).Thepercentageofcatastrophicerrors(ob- estand earliesttypeobjectsin cluster regionswhile againfield
jectswithadifferencebetweentrueredshiftandphoto-zofmore galaxiesdonotshowanysignificanttendency.
than 0.2×(1+specz)) is not negligible, but remains lower than Considering now catastrophic error percentages,in clusters
5%. It is also tempting to see an increased uncertainty in the sucherrorsoccurforbothbrightandlatetypegalaxies.Table2
photo-zestimateswhenconsideringclustergalaxies(exceptfor gives the spectral type and magnitude intervals for which we
have non null catastrophic error percentages in clusters. In the
3 1.48×median(|Δz|/(1+z)). field,thesepercentagesarenonnullwhateverthegalaxytypeor
Page7of15
A&A523,A21(2010)
Fig.11.Spectroscopicversusphotometricredshiftsfor4clusters.Wealsogivethedispersionsaroundthemeanrelation(reducedvalue,classical
valueexcludinggalaxiesforwhichthedifferencebetweenspectroscopicandphotometricredshiftsisgreaterthan0.5),andthemeanshiftbetween
photo-zsand speczs.Thesolidinclinedlinesgivetheperfectrelationwhilethedottedlinesgivethe±0.15relations.Theverticallinesgivethe
positionoftheclusteralongthelineofsight.
magnitudeandthereisnocleartrendtohavepreferablyhighor be fainter. From Fig. 17, we can also say thatSEDs for cluster
low catastrophic error percentages for specific galaxy types or galaxies would need to be (sligthly) fainter at red wavelengths
magnitudes. and brighter by ∼0.1 mag at blue wavelengths. The effect re-
How can these tendencies be understood? It is tempting to mainsmodest(lessthan0.1magmostofthetime)butitisnearly
saythatphotometricredshiftSEDsareenvironmentdependent. systematicoverthe10consideredlinesofsight.Themagnitude
This would not be surprising as the commonly used SEDs are shifts computed with the global training sets are therefore not
adaptedtolowdensityenvironmentsandresultingphoto-zaccu- perfectlyadaptedtoclustergalaxies,andusedSEDsarealsonot
racycouldbe degradedwhen consideringcluster galaxies.The verywelladaptedtoclustergalaxies.Thisprobablyexplainspart
spectral types showing in clusters the most atypical evolution ofthephoto-zaccuracydependenceontheenvironment.These
comparedto field objects are early type galaxies(cluster dom- resultsthereforeconfirmtheneedforhighdensityenvironment
inant galaxies) and very late type galaxies (galaxieswith short SEDswhenhighprecisionphoto-zsarerequiredforclusterstud-
bursts of star formation induced by the intracluster influence). ies. The main improvementshould essentially come from new
These are exactly the ones showing the worst photo-z accura- SEDsforbrightandveryearlyorverylatetypeclustergalaxies
cies in the previous tests. Moreover, during the process of the atvariousredshifts.
zero point shift estimates, we recall that we are comparingthe
photometric and spectroscopic redshifts used for training sets.
3.6.Photo-zqualitychecks:beyondthespectroscopic
These training sets are dominatedby field galaxiesfor mostof
samplelimits
the presently considered lines of sight (∼60% of the available
redshifts are field objects), so photometric redshifts for clus- Wenowaskthequestionofthephoto-zqualitybeyondz=1.05
ter galaxies may well not be optimally computed. In order to aswellasforobjectsfainterthanF814W = 23.5.Theseranges
test these possibilities, we computed the shifts to apply to our cannotbetestedwiththespectroscopicdatainhand,sowechose
photometrywhenincludinginspectroscopictrainingsetsallthe the same approach as in Ilbert et al. (2009). This paper shows
availableredshiftsoronlythosebelongingtoclusters.Figure16 thatthepercentageofgalaxieswith anindividualphoto-zerror
shows the difference between these shifts as a function of the (estimated by LePhare) larger than a given value is an indica-
considered photometric band disregarding the cluster redshift. tor of the catastrophic error percentages while the mean value
Figure 17 shows the same shifts but at the rest frame wave- of the individualphoto-z errors (for a given object subsample)
length (only for optical bands). This last figure is sensitive to is a goodapproximationof the dispersionof the specz/photo-z
the generalSED shape. We clearly see that mostof the optical relation(forthesamesubsample).
magnitudesneedtobebrightertoreachthebestSEDwhenus- Beforedirectlyapplyingthisapproach,we firstneedtotest
ing cluster training sets, except for the B band which needs to itonourdata.Forthis, we selectgalaxieswith aspectroscopic
Page8of15
L.Guennouetal.:TheDAFT/FADAsurvey.I.
Fig.12.SameasFig.11forthe6otherclusters.
result is 67 ± 8%, in good agreement with the expected 68%
forthe1σinterval.Wecanthereforequantifytheglobalphoto-z
accuracybasedonthe1σerrorbarsofindividualphoto-zs.
Wethencomputedthepercentageofgalaxieswitha1σerror
barlessthan0.2(1+z)(similarlyasinAdamietal.2008).This
givesusanestimateofthepercentageofgalaxieswithphoto-zs
whicharenotcatastrophicerrorsasafunctionofmagnitudeand
asafunctionofredshift,evenbeyondthespectroscopiclimit.We
plotinFig.18thesepercentagesforthetenmergedlinesofsight
considered.We first confirm that F814W ≤ 23.5 and z ≤ 1.05
galaxiesprobablyhavereliablephoto-zs,assuggestedinthepre-
vious section. Moreover, percentages statistically remain glob-
allyhigherthan90%formagnitudesbrighterthanF814W ∼24
or24.5exceptinthez = [1.5;2.0]redshiftrange.Asexpected,
Fig.13.Photometricversusspectroscopicredshiftsfortheblendedob- theworsesituationappearsforthez = [1.5,2.0]range(seealso
jectsinthespectroscopicsample. Coupon et al. 2009). The presently available magnitude pass-
bandsarenotwelladaptedtothisrange,theBalmerbreakbeing
locatedredwardofthez(cid:7) band,andtheLymanbreakstillbeing
redshift and we compute how many times the spectroscopic bluer than the B band. Finally, galaxy fluxes contaminated by
value falls inside the 1σ interval given by the photo-zs. The brighterclose (blended)neighborsdo notseem to show values
Page9of15
A&A523,A21(2010)
Fig.14.Reducedσofphoto-zsversusgalaxyphotometrictypeT.From Fig.15. Reduced σ of photo-zs versus galaxy i(cid:7) absolute magni-
toptobottom:clustergalaxiesinsidea500kpcradius,insidea1Mpc tude. From top to bottom: cluster galaxies inside a 500 kpc ra-
radius,andfieldgalaxies.Errorbarsforthetypesaresimplythesecond dius, inside a 1 Mpc radius, and field galaxies. Error bars for the
order momentum of the galaxy type distribution in the selected type absolute magnitudes are simply the second order momentum of
bins ([1; 7], [8; 12], [13; 19], [20; 31]). Error bars for thereduced σ the galaxy magnitude distribution in the selected magnitude bins
arePoissonianerrorbarsandarethereforedirectlyproportionaltothe ([−24;−23],[−23;−22],[−22;−21],[−21;−20]).Errorbarsforthere-
inverseofthenumberofgalaxiesinsidetheconsideredbin. ducedσarePoissonianerrorbarsandarethereforedirectlyproportional
totheinverseofthenumberofgalaxiesinsidetheconsideredbin.
significantlydifferentfromisolatedobjects.Takenatfacevalue,
theseresultssuggestthatourphoto-zsarenotstronglypolluted thanF814W =24.5andz≤1.5.Wemayalsoconsidergalaxies
bycatastrophicerrorsdowntoF814W ∼24.5. brighterthanF814W =24andatz≥3.75.
WeplotinFig.19themeanindividualphoto-zerrorsperbin
ofmagnitudeasafunctionofmagnitudeandphoto-zforvarious 3.7.Photo-zqualitychecks:thez≥3domain
redshiftintervals.Thisgivesusanestimateofthe1σuncertainty
aroundthe specz/photo-zrelation in the consideredredshiftin- As an additional external test of the photo-z uncertainties for
terval.Weconfirmandextendtheresultsoftheprevioussection. distant and faint galaxies, we took advantage of the giant arcs
Galaxies brighterthan F814W ∼24.5 and in the z = [0.4;1.5] detected along the LCDCS 0504 cluster (see Fig. 20). These
or brighter than F814W ∼ 24 and in the z = [3.0; 6.0] range arcsarelikelytobemultipleimagesofalownumberofsources
have relatively low photo-zuncertainties. We also confirm that and should therefore have identical redshifts when they origi-
photo-zsinthez=[1.5;3.75]rangearepoorlyconstrained. natefromasingleobject.We computedphoto-zsforthese arcs
These tests therefore lead us to adopt a conservative andatleastfourofthemproducedvaluescloseinredshiftwith
approach, and to limit our catalogs to galaxies brighter similarspectraltypes(T in[21,31],allconsistentwithanactive
Page10of15