Galaxy And Mass Assembly (GAMA): Spectroscopic analysis

4.3 MB·English

by  A. M. Hopkins

GAMAspectroscopicanalysis 1
Galaxy And Mass Assembly (GAMA): Spectroscopic analysis
A. M. Hopkins1⋆, S. P. Driver2,3, S. Brough1, M. S. Owers1, A. E. Bauer1,
M. L. P. Gunawardhana1,4, M. E. Cluver1, M. Colless1, C. Foster5, M. A. Lara-Lo´pez1,
6 7 8 8 9 10 11
I. Roseboom , R. Sharp , O. Steele , D. Thomas , I. K. Baldry , M. J. I. Brown , J. Liske ,
P. Norberg12, A. S. G. Robotham2,3, S. Bamford13, J. Bland-Hawthorn4, M. J. Drinkwater14,
3
1 J. Loveday15, M. Meyer2, J. A. Peacock6, R. Tuffs16, N. Agius17, M. Alpaslan2,3, E. Andrae16,
0
16 12 4 15 13 4
2 E. Cameron , S. Cole , J. H. Y. Ching , L. Christodoulou , C. Conselice , S. Croom ,
n N. J. G. Cross6, R. De Propris18, J. Delhaize2, L. Dunne19, S. Eales20, S. Ellis1, C. S. Frenk12,
a
21 16 13 6 3 22 23
J A. Graham , M. W. Grootes , B. Ha¨ußler , C. Heymans , D. Hill , B. Hoyle , M. Hudson ,

0 M. Jarvis24,25, J. Johansson26, D. H. Jones10, E. van Kampen11, L. Kelvin2,3, K. Kuijken27,
3
A´. Lo´pez-Sa´nchez1,28, S. Maddox19, B. Madore29, C. Maraston8, T. McNaught-Roberts12,

]
O R. C. Nichol8, S. Oliver15, H. Parkinson6, S. Penny10, S. Phillipps30, K. A. Pimbblet,10,
C 31 17 25 32 4 17
T. Ponman , C. C. Popescu , M. Prescott , R. Proctor , E. M. Sadler , A. E. Sansom ,
.
h 29 2 33 4 34
M. Seibert , L. Staveley-Smith , W. Sutherland , E. Taylor , L. Van Waerbeke ,
p
15 35 4 3 24
- J. A. Va´zquez-Mata , S. Warren , D. B. Wijesinghe , V. Wild , S. Wilkins
o
r 1AustralianAstronomicalObservatory,POBox915,NorthRyde,NSW1670,Australia
t 2InternationalCentreforRadioAstronomyResearch(ICRAR),UniversityofWesternAustralia,Crawley,WA6009,Australia
s
a 3SchoolofPhysics&Astronomy,UniversityofStAndrews,NorthHaugh,StAndrews,KY169SS,UK
[ 4SydneyInstituteforAstronomy,SchoolofPhysics,UniversityofSydney,NSW2006,Australia
5EuropeanSouthernObservatory,AlonsodeCordova3107,Vitacura,Santiago,Chile
1 6InstituteforAstronomy,UniversityofEdinburgh,RoyalObservatory,BlackfordHill,Edinburgh,EH93HJ,UK
v 7ResearchSchoolofAstronomy&Astrophysics,AustralianNationalUniversity,CotterRoadWestonCreek,ACT2611,Australia
7 8InstituteofCosmologyandGravitation(ICG),UniversityofPortsmouth,DennisSciamaBuilding,BurnabyRoad,PortsmouthPO13FX,UK
2 9AstrophysicsResearchInstitute,LiverpoolJohnMooresUniversity,TwelveQuaysHouse,EgertonWharf,Birkenhead,CH411LD,UK
1 10SchoolofPhysics,MonashUniversity,Clayton,Victoria3800,Australia
7 11EuropeanSouthernObservatory,Karl-Schwarzschild-Str.2,85748,Garching,Germany
1. 12InstituteforComputationalCosmology,DepartmentofPhysics,DurhamUniversity,SouthRoad,Durham,DH13LE,UK
0 13SchoolofPhysics&Astronomy,UniversityofNottingham,UniversityPark,NottinghamNG72RD,UK
3 14SchoolofPhysics,UniversityofQueensland,BrisbaneQLD4072,Australia
1 15AstronomyCentre,UniversityofSussex,Falmer,BrightonBN19QH,UK
: 16MaxPlanckInstituteforNuclearPhysics(MPIK),Saupfercheckweg1,69117,Heidelberg,Germany
v 17JeremiahHorrocksInstitute,UniversityofCentralLancashire,PrestonPR12HE,UK
i 18CerroTololoInter-AmericanObservatory,LaSerena,Chile
X
19DepartmentofPhysicsandAstronomy,UniversityofCanterbury,PrivateBag4800,Christchurch,NZ
r 20SchoolofPhysicsandAstronomy,CardiffUniversity,TheParade,CardiffCF243AA,UK
a
21CentreforAstrophysics&Supercomputing,SwinburneUniversity,Hawthorn,VIC3122,Australia
22InstitutdeCie`nciesdelCosmos,FacultatdeF´ısica,Mart´ıiFranque`s1,E-08028Barcelona,Spain
23DepartmentofPhysics&Astronomy,UniversityofWaterloo,Waterloo,ONN2L1Z5,Canada
24UniversityofOxford,DepartmentofPhysics,DenysWilkinsonBuilding,KebleRoad,Oxford,OX13RH,UK
25PhysicsDepartment,UniversityoftheWesternCape,PrivateBagX17,Bellville7535,SouthAfrica
26Max-PlanckInstitutfuerAstrophysik,Karl-Schwarzschild-Str.1,85741Garching,Germany
27LeidenObservatory,LeidenUniversity,POBox9513,2300RALeiden,TheNetherlands
28DepartmentofPhysicsandAstronomy,MacquarieUniversity,NSW2109,Australia
29CarnegieInstitutionforScience,813,SantaBarbaraStreet,Pasadena,California,91101,USA
30DepartmentofPhysics,UniversityofBristol,TyndallAvenue,BristolBS81TL,UK
31SchoolofPhysicsandAstronomy,UniversityofBirmingham,Edgbaston,Birmingham,B152TT,UK
32UniversidadedeSa˜oPaulo,IAG,RuadoMato1226,Sa˜oPaulo05508-900,Brazil
33AstronomyUnit,QueenMaryUniversityLondon,MileEndRd,LondonE14NS,UK
34UniversityofBritishColumbia,Vancouver,BCV6T2C2,Canada
35AstrophysicsGroup,ImperialCollegeLondon,BlackettLaboratory,PrinceConsortRoad,LondonSW72AZ,UK
Accepted2013January4
(cid:13)c 2013RAS,MNRAS000,2–19
Mon.Not.R.Astron.Soc.000,2–19(2013) Printed31January2013 (MNLATEXstylefilev2.2)
ABSTRACT
TheGalaxyAndMassAssembly(GAMA)surveyisamultiwavelengthphotometricandspec-
troscopicsurvey,usingtheAAOmegaspectrographontheAnglo-AustralianTelescopetoob-
tainspectraforupto∼ 300000galaxiesover280squaredegrees,toalimitingmagnitudeof
r <19.8mag.Thetargetgalaxiesaredistributedover0<z .0.5withamedianredshift
pet
of z ≈ 0.2, although the redshift distribution includes a small number of systems, primar-
ily quasars,athigherredshifts,upto andbeyond z = 1.The redshiftaccuracyrangesfrom
σv ≈ 50kms−1 toσv ≈ 100kms−1 dependingonthesignal-to-noiseofthespectrum.Here
we describetheGAMA spectroscopicreductionandanalysispipeline.We presentthesteps
involvedin takingthe raw two-dimensionalspectroscopicimagesthroughto flux-calibrated
one-dimensionalspectra.TheresultingGAMA spectracoveran observedwavelengthrange
of 3750 . λ . 8850A˚ at a resolutionof R ≈ 1300.The final flux calibrationis typically
accurateto 10−20%,althoughthereliabilityisworseattheextremewavelengthends,and
poorerinthebluethanthered.Wepresentdetailsofthemeasurementofemissionandabsorp-
tionfeaturesintheGAMAspectra.Thesemeasurementsarecharacterisedthroughavariety
of quality controlanalyses detailingthe robustnessand reliability of the measurements.We
illustratethequalityofthemeasurementswithabriefexplorationofelementaryemissionline
propertiesofthegalaxiesintheGAMAsample.Wedemonstratetheluminositydependence
of the Balmer decrement, consistent with previously published results, and explore further
how Balmer decrement varies with galaxy mass and redshift. We also investigate the mass
and redshift dependencies of the [NII]/Hα vs [OIII]/Hβ spectral diagnostic diagram, com-
monlyusedtodiscriminatebetweenstarformingandnuclearactivityingalaxies.
Keywords: galaxies:evolution–galaxies:formation–galaxies:general
1 INTRODUCTION separation, and target prioritisation is presented by Baldryetal.
(2010), with the tiling algorithm described by Robothametal.
Galaxysurveysthathavemoderateresolution(R∼1000−2000)
(2010) and the photometric analysis by Hilletal. (2011). Stel-
opticalspectroscopicmeasurements,suchastheTwo-degreeField
lar masses for the GAMA galaxies have been quantified by
Galaxy Redshift Survey (2dF; Collessetal. 2001) and the Sloan
Tayloretal. (2011), and the low redshift stellar mass function is
DigitalSkySurvey(SDSS;Yorketal. 2000)areamong themost
detailed in Baldryetal. (2012). The broadband luminosity func-
productiveresourcesforunderstandinggalaxyformationandevo-
tions are derived by Lovedayetal. (2012), and the Hα lumi-
lution.Theastrophysicalinformationencodedintherest-frameop-
nosity functions and evolution presented in Gunawardhanaetal.
tical region of the spectrum is among the most well-understood
(2013). Galaxy nebular metallicity measurements are detailed in
and well-calibrated aspect of galaxy evolution studies, and pro-
Fosteretal. (2012) and Lara-Lo´pezetal. (2013). Galaxy groups
videsawealthofdetailregardingthephysicalprocessesoccurring
in GAMA have been quantified by Robothametal. (2011), and
withingalaxies.Thisrangesfromquantitativemeasurementsofstar
galaxystructuralparametersmeasuredbyKelvinetal.(2012).
formationrate(SFR),metallicity,velocitydispersion,obscuration,
andmore,throughtodiagnosticsdistinguishingbetweenthepres-
TheGAMA survey used 68 nights of observing timeon the
enceof starformationoranaccretingcentral supermassive black
Anglo-AustralianTelescope(AAT)over2008–2010.Thistimewas
hole(anactivegalacticnucleus,AGN).Incombinationwithbroad-
usedtoconductahighlycompletesurveyinthreeEquatorialfields
bandphotometricmeasurementsspanningultravioletthroughtora- torpet <19.4magoveratotalof144deg2,48deg2ofwhichwas
diowavelengths,theredshiftandotherphysicalinformationfrom
observedtothedeeperlimitofrpet <19.8mag.Thisinitialphase
galaxyspectraprovidesapowerfultoolforexploringthedetailsof
ofthesurvey,usuallyreferredtoas“GAMAI,”allowedtheacqui-
galaxyevolution.
sitionofover112000newgalaxyspectraandredshifts,foratotal
The Galaxy And Mass Assembly (GAMA)1 survey is a
ofover130000redshiftsintheoriginalGAMAsurveyarea.Subse-
large multiwavelength photometric and spectroscopic galaxy sur-
quentlythesurveyhasbeenextended,withtheawardof110nights
vey (Driveretal. 2009, 2011) that provides exactly this compre-
of AATtimeover2010–2012, toexpand the surveyby including
hensive selectionof photometric and spectroscopic data.Thekey
twoSouthernfieldsandbroadeningthethreeEquatorialfields(re-
scientific goals are to use the galaxy distribution to conduct a
ferred to as “GAMA II”). This expands the total survey area to
series of tests of the cold dark matter (CDM) paradigm, in ad- 280deg2,whileachievingauniformdepthofrpet <19.8magover
dition to carrying out detailed studies of the internal structure
the full survey region. The goal is to compile ∼ 300000 galaxy
and evolution of the galaxies themselves. The scientific moti-
spectraoverthisarea.Atthetimeofwriting,wehavealreadyob-
vation for GAMA, the survey footprint, data processing, cata-
tainedover220000spectra.Indetail,todateGAMAhasobserved
logueconstructionandqualitycontrolaredescribedbyDriveretal.
224465 spectraof which222294 aregalaxy targets.Thesenum-
(2011). The target selection, including survey masks, star-galaxy
bers include repeat observations, and not all are main survey tar-
gets, as they include “filler” targets that take advantage of fibres
unabletobeallocatedtomainsurveytargetsonanygivenobser-
⋆ E-mail:[email protected] vationplate(seeBaldryetal.2010).Includingspectrafromother
1 http://www.gama-survey.org/ surveyswithintheGAMAregions,(SDSS,2dFGRSandothers,see
(cid:13)c 2013RAS
GAMAspectroscopicanalysis 3
Baldryetal.2010,fordetails),wehave299980spectra,ofwhich dispersionof1A˚/pixelandgivesacoverageof2100A˚.The385R
297067areofgalaxytargets(notallaremainsurveytargets).We gratingisusedintheredarm,centredat7250A˚.Thisgratinghas
have233777uniquegalaxytargets,ofwhich215458(92.2%)have a dispersion of 1.6A˚/pixel and gives a coverage of 3200A˚. This
redshift quality nQ > 3 (seeDriveretal. 2011, fordefinition of leadstospectrawitharesolutionthatvariesasafunctionofwave-
nQ,butinbriefnQ = 3or nQ = 4correspond toreliablered- length,fromR≈ 1000attheblueenduptoR ≈1600atthered
shifts). end.
The survey has already led to a number of published re- The“Two-degreeField”(2dF)instrument(Lewisetal.2002)
sults making use of the detailed emission and absorption line consistsofawidefieldcorrector,anatmosphericdispersioncom-
measurementsfromtheGAMAspectroscopicdata.Theseinclude pensator(ADC),andarobotgantrywhichpositionsopticalfibres
identification of the lowest-mass star forming galaxy population toanaccuracyof0′.′3onthesky.Thefibreshavea2′′diameterpro-
(Broughetal.2011),aself-consistentapproachtogalaxystarfor- jectedonthesky(Lewisetal.2002).Atumblingmechanismwith
mationrateestimatesandtheroleofobscuration(Wijesingheetal. two field plates allows the next field to be configured while the
2011a,b), and evidence for a star formation rate dependence currentfieldisbeingobserved.The392targetfibresfrom2dF are
in the high-mass slope of the stellar initial mass function fed to the AAOmega spectrograph, and eight guide fibre-bundles
(Gunawardhanaetal.2011),amongmanyotherGAMAteampub- areusedtoensureaccuratetelescopepositioningoverthecourseof
lications(forafullandcurrentlistseetheteamwebpage), along eachexposure.ForGAMAobservations,the2dFrobotfibreposi-
withadditionalworkledbycollaboratingsurveyssuchasHerschel- tionerisusedtoconfiguretypically345fibrestoobservegalaxies
ATLAS2. withinatwodegreefieldonthesky.Duetoavarying numberof
Inordertobestfacilitatesubsequentscientificanalysesofpub- damagedorunusablefibres(typicallyaround20),theactualnum-
licsurveydata,itiscrucialtoprovidefulldetailsof theobserva- beroffibresabletobeusedforgalaxytargetsisnotconstant.To
tions, processing and data product derivations (e.g., Boltonetal. quantifythis,thefirst-quartile/median/third-quartile numberoffi-
2012). Here we describe the spectroscopic pipeline process- bresongalaxytargetsforGAMAIwas332/345/348. Forthefull
ing for the GAMA survey. This encompasses an overview of survey to date, these numbers are 324/341/348. This distribution
the observations (§2), the steps involved in processing the raw has remained fairlysteady overthe duration of thesurvey so far.
two-dimensional spectroscopic images through to extracted one- Around25additionalfibresareusedtomeasuretheskyspectrumin
dimensional spectra and the initial redshift measurement process eachfield.Skypositionswereidentifiedusingaskymask,detailed
(§3),andfluxcalibrationoftheone-dimensionalspectra(§4).We inBaldryetal.(2010).Threefibresareallocatedtospectroscopic
alsopresenttheprocessesusedinmeasuringtheemissionandab- standardstars.
sorptionfeaturesofthespectra(§5)thatarerecordedintheGAMA TheintegrationtimeforeachGAMAfieldistypically60min,
database and made publicly available through the staged data re- splitintothree1200sexposures.Accountingfortheread-outtime
leases.NotethattheGAMAspectroscopicreductionandanalysis oftheCCDs(2min)andtheacquisitionofthecalibrationframes
pipelineisstillevolvingaswecontinuetoimprovesomeaspects. comprising flat-fieldsand arc-line exposures for wavelength cali-
Herewedescribethepipelinethatwasusedtoconstruct thefinal bration,thetimespentoneachfieldiswell-matchedtothetimere-
GAMAIdataset.Itwasthisdatasetthathasbeenusedintheinthe quiredforthe2dFpositionertoconfigurethefollowingobserving
variousinvestigationscitedabove.Thespectraandassociatedmea- plate,ensuringanefficientoverallsurveystrategy.Between20and
surements that will be available in GAMA DR2, the public data 30biasframesaretakenduringeachobservingsession.Startingin
release due in January 2013, also rely on the pipeline described 2011, we also began to use dark frames to refine the calibration,
here.Forthepurposesofthispaperweusethedataassociatedwith with from 10 to 30 dark frames, each of 1200s exposure, being
GAMASpecCatv08. acquiredeachobservingsession.
Throughout, allmagnitudesaregivenintheABsystem,and
weassumeacosmologywithH0 =70kms−1Mpc−1,ΩM =0.3
andΩΛ =0.7.
3 DATAPROCESSINGANDREDSHIFT
MEASUREMENTS
2 OBSERVATIONSATTHEAAT 3.1 Obtaining1Dspectra
TheGAMAspectroscopicobservationsusetheAAOmegaspectro- TherawdataareprocessedusingsoftwaredevelopedattheAAO
graph (Saundersetal. 2004; Smithetal. 2004; Sharpetal. 2006) called 2DFDR, (Croometal. 2004; Sharp&Birchall 2010). The
on the 3.9m Anglo-Australian Telescope (Siding Spring Obser- 2DFDR processing applies the standard sequence of tasks for 1D
vatory, NSW, Australia) for measuring the spectra of the target spectralextractionfrom2Dimages.Thisincludesbiassubtraction,
galaxies.Thisspectrographisstationedinthethermallystableen- flat-fielding,fibretrace(or“tramline”)fitting,andwavelengthcal-
vironment of one of the telescope Coude´ rooms. AAOmega pos- ibration.First,amasterbias(andforGAMAIIdataamasterdark)
sesses a dual beam system which allows coverage of the wave- are created using the available bias (and dark) frames. For each
lengthrangefrom3750A˚ to8850A˚ withthe5700A˚ dichroicused new plate configuration observed, the raw AAOmega frames are
by GAMA, in a single observation. Each arm of the AAOmega runthrough2DFDRatthetelescopetoprovidetheprocessedspec-
systemisequippedwitha2k×4kE2VCCDdetectorandanAAO2 tra.
CCDcontroller. Theblue armCCDisthinnedforimproved blue Thestandardparametersareusedinrunning2DFDR,withthe
response. The red arm CCD is a low fringing type. The grating followingmodifications.Weconsidernotonlythemasterbias(and
used intheblue armisthe580V, centred at4800A˚, whichhasa master dark), but also an overscan correction using a ninth-order
polynomial fit for the blue spectra, and second-order for the red
spectra.Thehighorderisnecessaryintheblueduetoaverystrong
2 http://www.h-atlas.org/ gradientinthefirst∼100pixelsthatisnotwell-modelledbylower
(cid:13)c 2013RAS,MNRAS000,2–19
4 A. M. Hopkinsetal.
Figure1.ExampleGAMAspectraforaselectionofgalaxies,illustratinghighqualityspectraandspectraaffectedbyinstrumentalandprocessingartifacts.The
spectrahavebeensmoothedwithafive-pixelrunningboxcaraveragetoaidinclarityofdisplay.Fromtoptobottom:Starforminggalaxyspectrum;Absorption
linegalaxyspectrum;Spectrumaffectedbyfringing;Spectrumaffectedbybadsplicing.EachspectrumincludesaninsetshowingtheSDSScolourgalaxy
image,aswellasidentifying commonemissionorabsorption features.EachgalaxyhasitsGAMAID,redshiftandstellarmass(fromTayloretal.2011)
listed,alongwithBalmerdecrement,Hαluminosityandstarformationrateforstarforminggalaxyspectra.Onlyabout3%ofGAMAspectraareaffectedby
fringingorbadsplicing.
(cid:13)c 2013RAS,MNRAS000,2–19
GAMAspectroscopicanalysis 5
orderfits.Thewavelengthsolutionisdeterminedbyathird-order isthenimplementedbydividingeachindividual spectrumbythis
polynomial fit to the arc-lines, with thesolution tweaked using a correctionspectrum.
first-order(blue) or third-order (red) polynomial to the sky lines. Finally, 2DFDR splices together the blue and red spectra for
The higher order is required for the red CCD as there are more eachgalaxybydoingafirst-passfluxcalibrationtobestmatchthe
sky lines and the sky is brighter at the red end of the spectrum. spectraatthesplice-wavelength(5700A˚).Thepixelscaleis1A˚ in
The wavelength calibration is referenced to arc line wavelengths theblueand1.6A˚ inthered,althoughduringthesplicingstepthe
in air rather than vacuum, contrary to the convention adopted by red spectra are resampled to the same pixel scale as the blue, so
SDSS,andisaccuratetobetterthan0.1A˚,asmeasuredfromkey thefinalpixelscaleis∼ 1A˚ pix−1.Thisisdonewithaquadratic
strong sky line features. Cosmic rays are cleaned in each object interpolationensuringthatfluxisconserved,andwithappropriate
frameusinganimplementationofLaplaciancosmicrayidentifica- treatmentofmaskedorotherwisebadvalues.Thesameresampling
tion(vanDokkum2001),andapplyingclippingthresholdsof10σ isappliedtothevariancearraystocorrectlypropagatethe errors.
in the blue and 5σ in the red. The scattered light is subtracted Theoverlapregionbetweentheblueandredspectrais250A˚,from
assuming a first-order polynomial fit. The throughput calibration typically5650A˚ to5900A˚,althoughthisvariesslightlyfromspec-
methodconsidersafluxweightedvalueofthenightskyemission trumtospectrumdependingonthelocationonthedetector.
linestonormalisethefibrethroughputs. As2DFDRiscontinuingtobedevelopedandimproved,andto
mitigateagainstreductionmistakesatthetelescopeduringobserv-
Extraction of spectra is performed by first identifying fibres
ing,theentireGAMAdatasetisperiodicallyre-reduced,toensure
in a flat field frame and then fitting their locations as a function
thatthefinalspectroscopicdataproductsarehomogeneousandin-
ofCCDpositionusingamodelofthespectrographopticaldistor-
tion.Oncethetramlinesarelocated,aminimumvarianceGaussian ternally self-consistent. The version of 2DFDR used in producing
weightedextraction(Sharp&Birchall2010;Horne1986)isused thefinalGAMAIspectrawas2DFDRv4.42.
Fig. 1 shows examples of GAMA spectra after the 1D ex-
toobtainthefluxineachfibreperspectralpixel.Whileoptimally
tractionandflux-calibrationprocess(see§4below)arecompleted.
weighted,thisdoesnottakeintoaccountcross-talkbetweenfibres,
ThisFigureshowstwohighqualityspectra,andtwopoorquality
but given the restricted dynamic range of the GAMA targets (a
spectraillustratingsome of theinstrumental andprocessing limi-
rangeoflessthanthreemagnitudes)thelevelofcross-talkhasbeen
tationsinthesurvey.Thetwohighqualityspectraareanemission
showntobenegligible(Sharp&Birchall2010,theirFigure2).For
lineobject andanabsorption lineobject, bothwithredshift qual-
each 1D spectrum there exists a variance array determined from
ity of nQ = 4 (see Driveretal. 2011, for the definition of red-
the Poisson noise in the bias corrected 2D frame, the read noise
shiftqualityflagsandconventions).Thefirstpoorqualityspectrum
and the gain, and propagated through the reduction pipeline. Ex-
showngivesanexampleoffringing,visibleasthehigh-frequency
amination of repeat spectrashows that theuncertainties perpixel
oscillation in the continuum level, and accompanied by poor re-
arewellcharacterisedbythemeasurementsinthevariancearray,
exceptaroundthestrong5577A˚ skyline,andattheextremeends moval oftheskyfeatures(Sharpetal.2006,2013).Thefringing,
which is time-variant, is only present for some fibres, and arises
ofthewavelengthrange.Wheretherearedifferencesbetweenthe
duetoairgapsinthegluebetweentheprismandtheferrule.Over
repeatmeasurementsandthevariancearraysduetosuchsystemat-
timethesefibreshavebeenre-terminatedwithnew glueand new
ics,thedifferencesarealwayslessthanafactorof1.4.
ferrules.Whileitdoesn’tresolvetheproblemforexisting GAMA
Theinitialskysubtractionisperformedusingthe∼25fibres spectrathatareaffected,theAAOhasrecentlycompletedatotalre-
allocatedineachplatetoskypositions.Acombinedskyspectrum placementofall2dFfibreswithoptimalglueandferrulesthathas
ismadebytakingthemedianofthecorresponding pixelsineach now eliminatedthisproblem. Inthisparticularexamplespectrum
ofthenormalisedskyfibres,discardingthetwobrightestskyfibres aredshiftisstillabletobereliablymeasured,with nQ = 4.The
toavoidpotentialproblemsintheeventofinadvertentnon-skyflux secondpoorqualityspectrumshownisanexampleofabadsplice,
(due to an asteroid, passing satellite, orother moving object per- characterisedbyadramaticchangeincontinuumlevelatthesplice
haps),attheskyfibrelocation.Thecontinuumskysubtractionac- wavelengthof5700A˚.Thisfeatureisaconsequenceofpoorcon-
curacy istypically 2-3% of the sky level, although for especially tinuumlevelestimationduetopoorflat-fieldinginoneorbothof
strongskylinessuchasthatat5577A˚,theresidualscanbeworse theblueandredarms,inestimatinghowtoscalethetwocompo-
(for details see Sharp&Parkinson 2010). This is then followed nents for the splice. The reliability of the splicing of AAOmega
by an improvement to the sky subtraction based on subtracting a spectraisanareathatisthesubject ofongoing work, bothat the
combinationofprincipalcomponenttemplates(Sharp&Parkinson AAO withcontinued development of 2DFDR, and within GAMA
2010).Thisreducestheamplitudeoftheskysubtractionemission throughinvestigationofindependentfluxcalibrationprocessesfor
lineresidualstobelow1%inmostcases. theblueandredarmsseparately.Forthisspectrum,againaredshift
Strongatmospherictelluricabsorptionfeaturesintheredpart isstillable tobe measured withhigh reliability, nQ = 4. These
ofthespectrumneedtobecorrectedfor.Thetelluriccorrectionin- examples illustrate a key point, that a high quality redshift mea-
volves constructing a flux and variance weighted combination of surementcanbeobtainedfromapoorqualityspectrum,although
allthespectrainagivenfield,whichistheniterativelyclippedto subsequent measurements of emission or absorption features for
remove residual emission or absorption features (such as galaxy thatspectrummaynotbereliable.
emission lines). This process relies on the fact that in any one
fieldthereareabroadrangeofgalaxyredshiftssofeaturesarenot
3.2 Redshiftmeasurements
presentatthesamewavelengthinmanyspectra.Theresultingaver-
agespectrumisfitbyaloworderpolynomialintheregionaround Redshifts are measured from the one-dimensional galaxy spectra
thetelluricfeatures,whileexcludingtheregionswheretheabsorp- assoonaseachfieldisfullyreducedusingtheaboveprocess(i.e.,
tion ispresent. Dividing through by thispolynomial fit resultsin typicallyonthenightofobservationorthefollowingday).Thisis
atelluriccorrectionspectrumwhichissettobeequaltounityev- doneusingtheGAMA-specificversionofRUNZ,originallydevel-
erywhere outside of the telluric absorption bands. The correction opedbyWillSutherlandforthe2dFGRS,andnowmaintainedby
(cid:13)c 2013RAS,MNRAS000,2–19
6 A. M. Hopkinsetal.
Figure2.Thefluxcorrectionvectorsfromeachstandardforone2dFplate,
shownhereasanexample.Thegreylinesshowtheindividualfluxcorrec-
tionvectorsforeachstandardstar.TheblacklineshowstheB-splinefitto
themeanvector.
ScottCroom.ThisprocessisdescribedinDriveretal.(2011),and
furtherdetailswillbepresentedbyLiskeetal.(inprep),including
there-redshiftinganalysis,whichquantifiesthereliabilityofeach
measured redshift by having multiple team members re-measure
redshiftsforalllow-qualityflaggedmeasurements,andforasubset
ofhigh-qualityredshifts.
TheRUNZcodeusesacross-correlationapproachtoidentify-
inganautomatedredshift,butallowstheusertomanuallyidentify
aredshift intheevent of apoorresult fromthecross-correlation,
beforeallocatedaredshiftqualityflag(Q),from0to4,with4be-
ingacertainredshift,3beingprobablycorrect,2indicatingapos-
sibleredshift needing independent confirmation, 1indicatingthat
noredshiftcouldbeidentified,and0meaningthatthespectrumis
somehowflawedandneedstobereobserved.Followingthisinitial
inspection,theprocessisrepeatedbymultipleteammembers,inor-
dertodefinearobust,probabilisticallydefined“normalised”qual-
Figure3.(a)Afterdividingouttheaveragefluxcorrectionvector(Fig.2)
ityscale,nQ(Driveretal.2011).Detailsoftheerrorestimateson toremovethelowordershape,thelow orderresiduals foreachstandard
theredshiftsareprovidedbyDriveretal.(2011).There-redshifting starareshowningrey.Theresidualsarethenfitwitha4thorderLegendre
processandderivationoftheprobabilitiesassociatedwiththenQ polynomialandthemedianofthesepolynomialsisshownbytheblackline.
qualitywillbepresentedinLiskeetal,(inprep.). ThedottedlineistheB-splinefit.(b)Thehighordertermsarefitbyfirst
A small fraction (∼ 3%) of theGAMA spectraare affected dividingouttheloworder(grey)andthentakingthemedianoftheresult
byfringing,asdeterminedmanuallyduringtheredshiftingandre- (black).
redshifting processes. Of these ∼ 50% still yield a good quality
redshift,althoughotherspectroscopicmeasurementssuchasemis-
sionlineproperties(see§5below)arelikelytobeunreliable.
spectrophotometryfortheGAMAsurveypresentsachallengedue
to the 2′′ optical fibres used for spectroscopy, in addition to ob-
servinginconditionsthatarenotalwaysphotometric.Startingwith
4 SPECTROPHOTOMETRICCALIBRATIONAND thetwo-dimensional spectraloutputfrom2DFDR,wespectropho-
QUALITY tometricallycalibratethedatafollowingtheidlspec2dpipeline
usedfortheSDSSDR6(Adelman-McCarthyetal.2008).Wede-
4.1 Fluxcalibration
termineacurvaturecorrectionandrelativefluxcalibrationforeach
The main purpose of flux calibration is to first correct the platefromthestandardstarsobservedoneachplate.Theabsolute
wavelength-dependence of the system throughput (atmosphere, spectrophotometric calibrationisdeterminedsuchthatthefluxof
residual wavelength dependence of fibreentrance losses afterthe eachobjectspectrumintegratedovertheSDSSfiltercurvematches
ADC, optics, and CCD quantum efficiency), and second to pro- thepetrosianmagnitudeoftheSDSSphotometryforthatobject.
videanapproximate absolutefluxcalibration. Obtainingaccurate We typically assign three of the fibres on each 2dF plate
(cid:13)c 2013RAS,MNRAS000,2–19
GAMAspectroscopicanalysis 7
thefinalfluxcalibrationvectorhasbeendetermined,itisappliedto
eachindividual spectrumontheplate,resultingineachspectrum
beingcorrectlyfluxcalibrated,inarelativesense.
The final step in obtaining an absolute flux calibration then
involves tying the spectrophotometry directly to the r-band Pet-
rosianmagnitudesmeasuredbytheSDSSphotometry.Thisisac-
complishedbymultiplyingeachindividualspectrumbytheSDSS
r-band filter response. The SDSS magnitudes are based on pho-
toncounting,sothiscalculationisdonebyintegratingfν/ν times
thefiltertransmissionfunction.Wethendeterminetheratioofthis
fluxwiththatcorrespondingtotheSDSSPetrosianr-bandmagni-
tudeforeach object. TheGAMA spectraarethenlinearly scaled
accordingtothisvalue.Itisstraightforwardtomodifythisscaling
factorifalternativephotometricreferencesarepreferred.Inpartic-
ular, we are exploring the utility of Sersic r-band magnitudes as
an alternative, although for the present the direct comparisons to
SDSSmeasurementsprovideimportantconsistencychecks.
Figure4.Themedian ratio ofcommonGAMAandSDSSspectra. This To test how well the applied flux calibration method agrees
resultisderivedfrom574objectsoutofthe637objectsthatwereobserved
withthemethodsappliedtoSDSSfibrespectroscopy, welookto
bybothsurveys. The63spectra excluded wereduetoparticularly noisy
objectsobservedbybothsurveys.Thereare 637objectsobserved
GAMAspectra,orredshiftmismatchesbetweenthesurveys.Thespectra
bybothSDSSandGAMA,whicharemostlygalaxies,butalsoin-
arefirstnormalisedbythemedianfluxvalueoftheflux-calibratedspectrum,
cluding120standardstars.Afterremovingasmallnumberofspec-
(sincetheabsolutefluxcalibrationisscaleddifferentlybetweenGAMAand
SDSS).Thespectraarethenmedianfilteredby7A˚ andinterpolatedtothe tratoexcludeparticularlynoisyspectraorthosewithmismatched
samewavelengthscalebeforetheratiowastaken.Thedarklineshowsthe redshifts,Fig.4showsthemedianand68thpercentilesoftheratio
medianofthefluxratios,andtheouter,greylinesshowthe68thpercentile ofthenormalisedGAMAandSDSSspectraasafunctionofwave-
rangeofthedistributionofratiosforindividualobjects. length, afterthefluxcalibrationisapplied totheGAMA spectra.
The spectra are normalised by the median flux of each spectrum
before taking the ratio, due to the different absolute flux calibra-
forobservingstandardstars.Thespectroscopicstandards aretypi- tionsapplied(tothefibremagnitudeinSDSS,andtothePetrosian
cally colour-selected to be F8 subdwarfs, similar in spectral type magnitude inGAMA).Itisclearfromthesolidlinethattheflux
to the SDSS primary standard BD+17 4708. The spectrum of calibrationappliedtotheGAMA spectraresultsinagood agree-
each standard star is spectrally typed by comparing it to a set ment across the entire wavelength regime with high fidelity, and
of theoretical spectra generated from Kurucz model atmospheres bestinthemid-wavelengthrangeofthespectra.Theextremeblue
(Kurucz 1992), using the spectral synthesis code SPECTRUM andredendsofthespectrumarenoisier,andworstintheblue,but
(Gray&Corbally 1994; Grayetal. 2001). A flux correction vec- theoverallagreementresultingfromthefluxcalibrationappliedto
tor, a one-dimensional array of wavelength-dependent correction theAAOmegaspectrafromGAMAisstillrobust.Overall,wefind
factors tied to the wavelength scale, is derived for each standard anaccuracyofaround10%inthefluxcalibratedspectra,rangingto
starbytakingtheratioof itsspectrum(inunitsof countsandaf- somewhatworsethan20%attheextremewavelengthendsofthe
ter correcting for Galactic reddening, Schlegeletal. 1998) to its spectra.Thereremainsanunresolvedissueassociatedwiththelevel
best-fitmodel (inunits of ergss−1cm−2A˚−1). Thisisillustrated oftheresponseintheblueendofthespectra,whichisthefocusof
in Fig. 2. There are a small number of plates (2 out of a total of ongoing work. For this reason, we advise caution when working
392forv08oftheGAMAdata)thatincludednostandardstarob- withthebluestspectraldiagnostics(suchasthe[OII]emissionline
servations.Fortheseplates,weusedthestandardstarsobservedon andthe4000A˚ break)inthecurrentgenerationofmeasurements.
theplateobservedeitherjustbeforeorjustaftertheplate lacking Basedonthisanalysis,weestimatethatthefluxcalibratedGAMA
standardstars. spectraaretypicallyaccuratetobetween∼10−20%,althoughthe
Wefirstcalculateanaveragefluxcorrection vectorfromthe smallfractionofpoorlysplicedspectraandthoseaffectedbyfring-
standardstarsoneachplate.Weselecthighsignal-to-noiseregions ing are likely to be much worse. We note that this precision has
ofthestandardstarspectra,anddivideoutthisaveragecorrection been estimated using bright spectroscopic targets. For the fainter
vector(Fig.3a).Byremovingtheoverallaverageofthelow-order GAMA spectra it is likely that the spectrophotometric precision
shape in the standard star continuum in this way, we can fit the willnotachievethislevel,andwearealsocontinuingtoworkon
residualstoderiveaplate-specificaveragecurvaturecorrection,to quantifyingthedependencewithmagnitude.
account for the declining CCD response at the extreme blue and
red wavelengths. These residuals are fit with 4th order Legendre
4.2 Spectralsignal-to-noise
polynomials(Fig.3a),andtheaverageloworderresidualisfound
bytakingthemedianoftheLegendrecoefficients.Forspectralre- Here we detail the continuum signal-to-noise (S/N) distribution
gionswherethestandardstarspectrahaveS/N >12,higherorder fortheGAMAspectraasafunctionofapparentmagnitude(Fig.5).
fluctuationsarealsocorrected.Thesehigherordertermsarefound The S/N per pixel isquantified by measuring the median of the
fromthemedianLegendrecoefficientsafterthelowordertermsare ratio of the observed flux to the noise in a 200A˚ window at the
dividedout.Thefluxcalibrationvectorappropriateforeachplate centreoftheblueandredarms.Thenoiseisdeterminedfromthe
is constructed by multiplying the lower order residual fit and the variancearraysdescribedin§3.1.Intheblue,thewavelengthrange
highorderresidualfit.ThisvectoristhenfitwithaB-spline,and 4600−4800A˚ wasselected,and7200−7400A˚ inthered.Itcan
thecoefficients areused asthe final fluxcalibration vector. Once beseenthattheS/N propertiesvaryasexpected,withbrightertar-
(cid:13)c 2013RAS,MNRAS000,2–19
8 A. M. Hopkinsetal.
Figure5.TheS/Nintheblue(left)andred(right)armsoftheGAMAspectra,asafunctionofg-orr-bandPetrosianmagnitude,respectively.Thesolidand
dashedlinesshowthemedianand68thpercentiles.ItisclearthatbrightertargetshavehigherS/N,andalsothattheredarmoftheAAOmegaspectrograph
isrelativelymoresensitiveforagivenexposuretime.Themagnitudelimitofr ≈17.7magatthebrightendcorrespondstotheSDSSmaingalaxysample
limitofr=17.77mag.TheGAMAselectionlimitofr=19.8magatthefaintendisalsoapparent.Theobjectsoutsidetheselimitsincludebrighttargets
notobservedbySDSSduetofibre-collisionlimitations,andfainttargetsincludedthroughtheadditionalKandz-bandselectionlimits(Baldryetal.2010),
togetherwithspecificadditionalfillertargets(Chingetal.inprep.).
targetsshownthatextendfainterthanournominalsurveyselection
Table1.GAMAspectralemissionlinesmeasuredbyGANDALF.
limitofr=19.8mag.Theseenterthesurveyasaresultofoursup-
Emissionline wavelength(A˚) plementaryK-andz-bandselectionlimits(Baldryetal.2010),as
wellastheadditionofaselectionof“filler”targets.Duetogalaxy
HeII 3203 clusteringandthelimitationofhowcloselyfibrescanbepositioned
[NeV] 3345 with2dF,therearefrequentlyanumberoffibresonanygivenplate
[NeV] 3425
thatcannotbeassignedtoaprimarysurveytarget.Inthiscase,and
[OII] 3727
inordertomaximisethescientificreturnfromthesurvey,weallow
[NeIII] 3869
fibrestobeassigned tosupplementary targetsselectedtosupport
H5 3889
specific scientific programs (Driveretal. 2011). In particular, we
[NeIII] 3967
Hǫ 3970 allocatesuchfillerfibrestotargetsystemsidentifiedasinteresting
Hδ 4102 objectsinthefar-infraredbytheHerschel-ATLASsurvey,aswell
Hγ 4340 as to radio sources with carefully identified optical counterparts,
[OIII] 4363 thatotherwiselackredshiftorspectroscopicmeasurements(Ching
HeII 4686 etal.inprep.).
[ArIV] 4711
[ArIV] 4740
Hβ 4861
[OIII] 4959
[OIII] 5007 5 MEASUREMENTOFSPECTRALFEATURES
[NI] 5199
The flux-calibrated 1D spectra are those used for all subsequent
HeI 5876
measured spectroscopic properties. Both emission lines and ab-
[OI] 6300
sorptionfeaturesaremeasured.Avarietyofstandardgalaxyprop-
[OI] 6364
[NII] 6548 erties are catalogued based on these measurements, including
Hα 6563 Balmer decrements and star formation rates from the Hα and
[NII] 6583 Hβ emission lines (Gunawardhanaetal. 2013), spectral diagnos-
[SII] 6716 tics(Baldwinetal.1981)todiscriminateAGNfromstarforming
[SII] 6731 objects(Gunawardhanaetal.2013),metallicitymeasurementsfor
thenebulargasfromtheemissionlinefeatures(Fosteretal.2012,
Lara-Lo´pez et al., in prep.), stellar velocity dispersions from ab-
sorptionfeaturesandtheD4000 age-estimateparameterfromthe
getsshowinghigherS/N.TheredarmoftheGAMAspectradis- 4000A˚ break,andmore.Thissectiondetailsthemeasurementpro-
playsatypicalS/N ofafewatthefaintestobservedmagnitudes, cessesforquantifying thespectral featuresusedinderiving these
increasingtowellover10atthebrightest.InthebluearmtheS/N nowstandardpropertiesforGAMAgalaxies.Themeasurementof
is typically between 1−5 from the faintest to brightest sources. thespecificpropertiesthemselvesaredetailedinthevariouspapers
ThereareaverysmallfractionofspectrawherethemeasuredS/N presentingtheanalysisofthoseparameters,referencedabove.
intheblueisnegative.Thisisaconsequenceofthecontinuummea- Emissionlines aremeasured intwo ways. Wefirstfit Gaus-
surementbeingnegativeforthesespectra,andisanartifactarising sians to a selection of common emission lines at appropriate ob-
frompoorflat-fieldingorscatteredlightsubtraction.Therearealso served wavelengths, given the measured redshift of each object.
(cid:13)c 2013RAS,MNRAS000,2–19
GAMAspectroscopicanalysis 9
Figure6.Upperleft:Thedistributionoftheratioof[OIII]λ4956/[OIII]λ5007,showingconsistencywiththevaluefixed,foragivendensityandtemperature,
byquantummechanics.Thesolidlineisthemedian,andthedashedlinesarethe68thpercentiles.Theexpectedratioof1/2.98(Storey&Zeippen2000)is
shownasthedot-dashedline.Upperright:Thefluxratioshownnowasafunctionofthefluxinthebrighterline.Thedashedlinehereistheexpectedvalue.
Thebottompanelsreproducethisanalysisfor[SII]λ6716/[SII]λ6731,wheretheexpectedratiois1/1.4(Osterbrock1989).
Thisisperformedforasubsetofcommonemissionlinesviaasi- aticcases,includingfailedmeasurementsorlinesfallinginmasked
multaneousiterativeχ2 fittingofpositiveemissionpeakstothese spectralregions,linesthatareapparentlytoonarrowtobefit(usu-
emissionlinesinthreeindependentlinegroups,aroundthewave- allycausedbybadpixels),andlinesthatarefitbythemaximumal-
lengthsofthe[OII],Hβ,andHαlines.Thelocalcontinuumspan- lowedwidthofthefittingroutine(aGaussianσ=10A˚),typically
ningeachfittingregionisapproximatedwithalinearfit.Alllines duetothepresenceofanintrinsicallybroadline.Thisprovidesa
ineachgrouparefittedsimultaneously.Aswellasthecontinuum baseline set of emission line fluxes, equivalent widths, and S/N
coefficientsandpeakintensities,asmallvelocityoffset(fromthe estimates.Forthose(smallfractionof)objectssuchasbroad-line
underlyingGAMAredshift)isallowed,andalinewidthcommon activegalacticnuclei(AGN)andsourcesthatexhibitlinesplitting
to all lines in the group is also fit. Line width is constrained to (indicativeofeithermergingsystemsorstarburstwinds), thesim-
lie between 0.5-5 times the instrumental PSF (3.4 CCD pixels, plesingleGaussianemissionapproximationwillnotprovideuseful
3-5A˚). Each line group is fitted with independent values for the estimates.
freeparameterstoaccommodatesmalllocalvariationsinthespec-
Independently, we also pursue a more sophisticated spectral
tral sampling. Flux values for individual lines are rejected if in-
measurement approach, in order to self-consistently derive both
clusion of a lineinthe global fitfailsto improve thereduced χ2
stellarkinematicsandemissionlineproperties.Todothis,weuse
valuesignificantly(byafactorof 3).Typical GAMA sourcesex-
thepubliclyavailablecodespPXF(Cappellari&Emsellem2004)
hibit only marginally resolved emission lines at the resolution of
andGANDALF(Sarzietal.2006).Wefirstextractthestellarkine-
the AAOmega spectra. No attempt has been made to fit multiple
maticsusingpPXF,matchingtheobservedspectratoasetofstellar
emission components to composite emission linestructures, such
populationtemplatesfromMaraston&Stro¨mback(2011)basedon
asthe[OII]doublet,atthistime.Integratedlineflux,EWandthe
theMILESstellarlibrary(Sa´nchez-Bla´zquezetal.2006),masking
associatedstatisticalerrorsareestimatedfromthefittingprocess.
the regions which could potentially be affected by nebular emis-
TheerrorsarethoseassociatedwiththeformalGaussianfittingpro-
sion,giventheobservedredshift.Forthisstep,wedowngradethe
cess, and as shown in §6.2 below, the error estimates are robust
spectralresolutionoftheMaraston&Stro¨mback(2011)modelsto
although likely to be somewhat underestimated for fainter galax- matchtheGAMAspectral resolution, adopting avalueof 2.54A˚
ies.Inaddition,flagsareprovidedtoidentifyavarietyofproblem-
(FWHM)fortheMILESresolution(Beifiorietal.2011).Next,we
(cid:13)c 2013RAS,MNRAS000,2–19
10 A. M.Hopkinset al.
use GANDALF v1.5 to simultaneously fit both Gaussian emis- temperature, with the [OIII], [NII] and [SII] line pairs being ob-
sionlinetemplatesandthestellarpopulationtemplates,whichare vious choices. Due to the stellar absorption of Hα in the vicin-
broadenedtoaccountforthederivedstellarkinematics,tothedata, ity of the [NII]λ6548 line, and (to a lesser degree) its proxim-
whilealsocorrectingfordiffuse(stellarcontinuum)obscuration.It ity to the Hα emission line itself, this particular ratio is less ro-
isimportanttonotethatdiffuseobscuration,i.e.thatcausedbydif- bust. Using the measurements from the Gaussian fits, in Fig. 6
fusedustinthegalaxy,affectstheentirespectrum,andnoobscura- weshowthedistributionof[OIII]λ4956/[OIII]λ5007comparedto
tioncorrectionisappliedpurelytoemissionlinesduringthefitting theexpected ratioof 1/2.98 (Storey&Zeippen 2000). The same
process (even though this option is available). We made this de- is shown for [SII]λ6716/[SII]λ6731, where the expected ratio is
cisioninordertominimisethenumberofpoorspectralfitscaused 1/1.4(Osterbrock1989).Lineratiosareonlyincludedinthisanal-
bylowspectralS/Nthatotherwiseledtoambiguousemission-line ysisfor pairswhere the brighter of the twolines has aflux mea-
fluxes,whennoisewasamoresignificantfactorthanreddening.A surement above 3σ. In the right panels, the ratio is shown as a
Calzetti(2001)obscurationcurveisusedinestimatingtheobscura- functionofthefluxofthebrighterline.Atlowerflux(morespecif-
tioncorrectionstothestellarcontinuum,andisappliedthroughout icallylowerS/N),thefainterlineislesseasilydetected,andabias
theGANDALFmeasurementprocess.Thenebularemissionlines towardhighervaluesoftheratiocanbeseen.Thisresultdemon-
fromGANDALFaresubsequentlycorrectedforobscurationusing strates the robustness of the line fitting and measurement within
a Milky Way obscuration curve (Cardellietal. 1989), as recom- eachspectrum.
mendedbyCalzetti(2001).
GANDALF’ssimultaneousfittingmechanismallowsustoac-
6.2 Duplicatemeasurements
curately extract ionised gas emission from the stellar continuum,
minimisingcontaminationfromstellarabsorption,inordertocal- AsmallnumberofGAMAspectraduplicateobservationsofpartic-
culateemissionlinefluxesandgaskinematics(velocitiesandve- ulartargets,oftenduetothere-observationofobjectswherealow
locitydispersions)fromtheGaussianemissionlinetemplates.The redshift quality was initially obtained. This is typically a conse-
outputs from this analysis are line fluxes and equivalent widths, quenceoftheobjectsbeingatthefainterendoftheGAMAtarget
velocitydispersions (fromthestellarabsorption linesin thebest- selection, and the resultshere are consequently illustrativeof the
fitting SEDs), and associated derived products such as Balmer robustness of the measurements for the fainter population. There
decrement. A listof emissionlinesforwhichfluxand equivalent isnoattemptmadetocombinetheseduplicateGAMAspectra,or
widthmeasurementsareextractedisgiveninTable1.Thewave- themeasurements fromthem,due tothecomplexsystematicsin-
lengthsgivenfortheselinesarethewavelengthsasmeasuredinair, volved inthe flux calibration stepsdetailed above. Wecan, how-
asopposedtothevaccuumwavelengthsusedbySDSS(Yorketal. ever,takeadvantageoftheseduplicateobservationstounderstand
2000). theprecision towhichwecan measure ouremission lineequiva-
Our final set of measurements includes the flux, equivalent lentwidthsandfluxes.Todothis,weselectthoseduplicatemea-
width and signal-to-noise ratio for each emission line, as well as surementswherebothspectrahavebeenallocatedaredshiftquality
avelocitydispersioninferredfromthelinewidth.Wecalculatean 36nQ 64(Driveretal.2011).Weillustratethedifferencesbe-
emission line ratio diagnostic classification (Baldwinetal. 1981) tween these duplicate measurements, using both the GANDALF
for each galaxy based on these measurements. The stellar veloc- andtheGaussianfits,forlinefluxes(Fig.7)andequivalentwidths
itydispersion and E(B −V) values from theSED fits,for both (Fig.8)oftheHαand[OII]emissionlines.Thesetwolinespecies
diffuseandnebularobscuration (ifapplicable), arealsorecorded. were chosen for the following reasons. First, they represent the
While stellar velocity dispersions are measured for all spectra, best-andworst-casescenarioswithHαbeing(typically)astrong
thesearetypicallyonlyrobustforspectrahavinghigh S/N (e.g., line in a high S/N region of the spectrum, and [OII] being both
Proctoretal.2008;Shuetal.2011,Thomasetal.,inprep.).Inad- weak(inmanycases)andsituatedatthelowS/N endofthespec-
dition, the best-fit SED templateand associated χ2, along witha trum. Thus, these two lines give an idea of the spread expected
cleanemission-linefreeabsorptionspectrumareavailable. in the precision of the measurements across the full wavelength
andS/N range.Second,theHαlineisaffectedbyunderlyingab-
sorption due to the stellar continuum which is corrected for dur-
ingtheGANDALFfittingprocess.However,thiscorrectionrelies
6 QUANTIFYINGMEASUREMENTRELIABILITY criticallyonarobustmodelfittotheunderlyingstellarcontinuum
and errors in this fit may add systematic uncertainties to the flux
We explore the quality of our spectroscopic measurements thor-
andequivalent widthmeasurements.Suchsystematicsarequanti-
oughly. Here we detail the internal reliability for measurements
fied by comparing the differences in the duplicate measurements
withineachspectrum(§6.1),therepeatabilityofourmeasurements
usingduplicateobservations(§6.2),andtheself-consistencyofour to the quadrature sum of the uncertainties on the measurements
(Fig. 9 and Fig. 10). The 68th percentiles are typically 1.5 to 2
measurements between the two independent approaches we use
unitsfromthemedian,suggestingthattheformalstatisticalerrors
(§6.3).Wegoontooutlinetheimpactofstellarabsorptiononthe
GaussianfitBalmerlinemeasurements(§6.4),thereliabilityofve- onthelinemeasurementsaresomewhatunderestimatedfromboth
locitydispersionestimates(§6.5),andtheextentofapertureeffects theGANDALFandtheGaussianfits.Thisislikelyrelatedtothe
(§6.6). factthattheduplicatespectraaredominatedbyfaintergalaxytar-
gets.Forthefainterorlower S/N targets,thecontinuumlevelin
the spectra tends to be noisier, and more affected by systematics
such as poor scattered light subtraction or sky subtraction. This
6.1 Internalconsistency
leadsinturntogreatervariationbetweentherepeatmeasurements,
Asimpletestoftheinternalconsistencyofourlinemeasurements asthecontinuumlevelestimatedforeithertheGaussianfittingor
is to measure the ratios of emission lines from ionised species byGANDALFcanbemoreeasilyover-orunder-estimated.Theer-
that should befixed by quantum mechanics forfixed density and rorsonthelinefittingmaynotaccuratelyreflecttheuncertaintyin
(cid:13)c 2013RAS,MNRAS000,2–19