Gaia Data Release 1: Validation of the photometry

10.5 MB·

by  D. W. Evans

Astronomy&Astrophysicsmanuscriptno.GAIA-CS-CP-IOA-DWE-050Final (cid:13)cESO2017
January23,2017
Gaia Data Release 1
Validation of the photometry
D.W.Evans1,M.Riello1,F.DeAngeli1,G.Busso1,F.vanLeeuwen1,C.Jordi2,C.Fabricius2,A.G.A.Brown4,
J.M.Carrasco2,H.Voss2,M.Weiler2,P.Montegriffo3,C.Cacciari3,P.Burgess1,andP.Osborne1
1 InstituteofAstronomy,UniversityofCambridge,MadingleyRoad,CambridgeCB30HA,UK
e-mail:[email protected]
2 InstitutdeCiènciesdelCosmos,UniversitatdeBarcelona(IEEC-UB),MartíFranquès1,E-08028Barcelona,Spain
3 INAF–OsservatorioAstronomicodiBologna,viaRanzani1,40127Bologna,Italy
4 SterrewachtLeiden,LeidenUniversity,POBox9513,2300RALeiden,theNetherlands
7
1
ReceivedMonthDay,201X;acceptedMonthDay,201X
0
2
ABSTRACT
n
a
Aims.ThephotometricvalidationoftheGaiaDR1releaseoftheESAGaiamissionisdescribedandthequalityofthedatashown.
J
Methods.Thisiscarriedoutviaaninternalanalysisofthephotometryusingthemostconstantsources.Comparisonswithexternal
0
photometriccataloguesarealsomade,butarelimitedbytheaccuraciesandsystematicspresentinthesecatalogues.Ananalysisof
2 thequotederrorsisalsodescribed.Investigationsofthecalibrationcoefficientsrevealsomeofthesystematiceffectsthataffectthe

fluxes.
]
Results.Theanalysisoftheconstantsourcesshowsthattheearly-stagephotometriccalibrationscanreachanaccuracyaslowas3
M
mmag.
I
. Keywords. Astronomicaldatabases;Catalogues;Surveys;Instrumentation:photometers;Techniques:photometric;Galaxy:gen-
h eral;
p
-
o
r1. Introduction photometric data, the photometric residuals, and an analysis of
t
s theepochphotometryofconstantsourcesinSects.7-9.Exter-
aThephotometriccalibrationofthefirstdatareleaseoftheGaia nalcomparisonsarethendescribedinSect.10.Finally,Sect.11
[catalogue (Gaia Collaboration et al. 2016a) aims to achieve summarises the conclusions. Appendix A contains a list of the
1 mmag-level precision (van Leeuwen et al. 2016). This is car- acronymsusedinthispaper.
vried out via an internal, self-calibrating method as detailed in Further validation of the overall catalogue can be found in
3Carrasco et al. (2016). No comparison with a set of standards Arenouetal.(2016).
7wouldbesufficienttoconfirmthattheaccuraciesquotedforthe
Thefollowingsubsectionprovidesabriefdescriptionofthe
8photometrywerevalidbecausetheprecisionaimedforisbetter
Gaiainstrumentsanddata.Manymoredetailsareavailablefrom
5thantheprecisionofothercurrentlyavailablelargecataloguesof
GaiaCollaborationetal.(2016b).
0photometricstandards.Also,unexpectedsystematiceffectshave
.
1beenfoundintheGaiadatathatrequiredadditionalcalibrations
0to be carried out with respect to the initial plan (Riello et al.
2. Inputdata
72016),suchaslineartrendswithtimeandincreasedbackground
1level. It is necessary to check that these calibrations have re-
Gaiaisascanningsatellite.Thefullskyisexpectedtobecov-
v:movedallthesystematiceffectsandthattheaccuraciesachieved eredinaboutsixmonthsofobservations,butthenumberofob-
iareclosetothephotonnoiselevel.Itisexpectedthatfuturere- servationspersourcelargelydependsontheastrophysicalcoor-
X
leases of the Gaia catalogue will have improved accuracies as
dinatesofthesource.
rfurthercalibrationsareintroducedintothedataprocessing.
a The main input data for the photometric processing comes
Although no colour nor spectral information is included in fromtheAstrometricField(AF)CCDs.Thisisanarrayofseven
this data release, some validation is given to the processing of rows (parallel to the along-scan (AL) direction) and nine strips
thespectralcalibrationsofblueandredphotometers(BPandRP, (parallel to the across-scan (AC) direction) of CCDs collecting
respectively).Thisisduetotheuseofcolourinformationinthe light in the Gaia G broad band. Colour information for each
calibration of the G-band photometry, which is itself internally source (also a fundamental ingredient for the photometric cali-
calibrated.Wealsopresentsomeresultsofthevalidationofthe brations)isderivedfromthelow-resolutionspectracollectedby
epochphotometryavailableintheGaiaEPSLrelease(Eyeretal. theBPandRPinstruments.Thelightisdispersedinthealong-
2016). scandirection.
Sections 3 to 6 of this paper cover the direct validations of In the following we will refer to field-of-view (FoV) and
thecalibrationscarriedoutinthephotometricprocessing.Thisis CCDtransits:aFoVtransitincludesseveralCCDtransits(usu-
followed by internal consistency checks using the accumulated ally nine AF, one BP, and one RP CCD transit). It should be
Articlenumber,page1of15
A&Aproofs:manuscriptno.GAIA-CS-CP-IOA-DWE-050Final
noted that the two FoVs are simultaneously projected onto the thiscalibrationandofthenominaldispersionfunctionbringsall
focalplanearray. thedataontothesamewavelengthsystem.
Only small windows centred on the detected sources are The BP/RP windowsthat are assigned onboard will gener-
downloaded from the satellite. The size of these windows de- allynotbewellcentredonthesource.Thisisexpected,giventhe
pendsonthemagnitudeestimatedonboard;onlybrightsources designoftheinstrument,andisduetovariousfactors:theloca-
areobservedwith2Dwindows.Differentconfigurationsarere- tion of the centroid from the SM observation may not be very
ferred to as “window classes”. The shape of the windows (nor- accurate, and sources may have a non-negligible motion along
mallyrectangular)canbecomplicatedbyconflictsbetweenad- or across scan. This implies that considering an arbitrary refer-
jacentsourcesincrowdedregionsinthesky.Thesenon-nominal ence wavelength, the location will not correspond to the same
cases have not been treated yet and have not contributed to the locationinsamplespaceeveninspectraofthesamesource(for
photometrypublishedinGaiaDR1. adefinitionofsamplespaceseeGaiaCollaborationetal.2016b).
TheCCDsareoperatedinatime-delayedintegration(TDI) The location of the centre of the source, or more precisely
mode whereby charges are integrated while they move across ofareferencewavelengthinthedispersedimageofthesource,
theCCD.TheeffectiveexposuretimeoveroneCCDisapprox- canbepredictedbyextrapolationfromtheseriesofsourcecen-
imately 4.5 seconds, but this can be reduced by activating the troids of each of the AF observations that precede the BP/RP
“gates”. Several different gate configurations are defined and a observationsinaFoVtransit.However,thisrequiresanaccurate
particulargateisassignedtoeachCCDtransitdependingonthe knowledgeofthegeometriccalibrationoftheBP/RPCCDswith
magnitudeofthesourceasestimatedfromthestripofCCDpre- respecttotheAFCCDs(andofthesatelliteattitude).
ceding the AF (known as the “star mapper” or SM CCDs) and To calibrate the geometry of the instrument, this prediction
theACpositionofthesourceinthefocalplane.OveraFoVtran- needs to be compared to the actual location of the reference
sit,differentgateconfigurationscanbeusedonindividualCCD wavelength in the observed spectra. This is done by selecting
observations.Theactivationofagatetriggeredbythetransitof a small fraction of the observed spectra based on a filter in
abrightsource,willaffectallothersourcesobservedsimultane- colour and magnitude, so that we can be confident that we are
ouslyinthesameregionofaCCD.Itmayalsoaffectonlypart usingspectraofsourceswithasimilarspectralenergydistribu-
ofawindow,thuscreatingcomplexgatecasesthathavenotyet tion, where the sample position corresponding to the reference
beentreatedbythephotometricprocessing. wavelengthwillbethesame(exceptfortheeffectofnon-perfect
Different gate and window class configurations effectively centringofthewindowmentionedabove)andthatwearefilter-
definedifferentinstruments(referredtoascalibrationunits)that ing observations with a high S/N. The colour range adopted is
need to be calibrated to form one consistent reference system [0.3,0.6] for the calibration of the BP instrument and [1.3,1.6]
(formoredetailsseeCarrascoetal.2016). for the calibration of the RP one. These correspond approxi-
matelytospectraltypesFandK(basedonnominalknowledge
Theinputdatatothephotometricprocessingconsistsofim-
oftheinstrumentandpre-launchsimulations).Thiscolourselec-
age parameters (such as fluxes, centroids, and goodness of fit
tion,inadditiontootherfiltersdesignedtoselectisolatedspec-
measurements) for the SM and AF CCD transits, and raw BP
traandtoavoidspectrathatareaffectedbycosmicrays,yields
and RP spectral data. Errors on the G-band flux measurements
a sufficient number of calibration spectra. This is of the order
are estimated in the image parameter determination (IPD) pro-
of several hundreds for each calibration unit of the large-scale
cess;formoredetailsrefertoFabriciusetal.(2016).
componentoftheALgeometriccalibrationmodelwhichisthe
Two significant and unexpected features were discovered
one that is updated most often (every 20 OBMT revolutions or
during the commissioning period and required the introduction
about5daysforGaiaDR1).Theselectedspectraarealignedand
of ad hoc calibrations. One is the presence of stray light scat-
usedtogenerateareferencespectrum,whichisthenfittedback
tered by the solar shield, causing the background level to be
toeachspectrumtoevaluatethesamplepositionofthereference
uptotwoordersofmagnitudehigherthanexpected(withlarge
wavelength within the actual sampling. Two reference spectra
variationsdependingontherotationphaseofthesatellite).The
aredefinedfortheentiredataset,oneforBPandoneforRP.
additionalstraylightcomponentofthebackgroundcanbecali-
brated,buttheassociatednoisewillaffectperformanceforfaint At this stage, the processing has concentrated on differen-
tialcalibrationsofthevariousinstrumentconfigurationsontothe
objects.Theotherfeatureisadecreaseovertimeofthethrough-
sameinternalreferencesystem.Thisisthentiedtotheabsolute
putoftheinstrumentsduetocontinuedcontaminationbywater
ice.Thewavelength-dependenttransmissionlossisdifferentfor system adopting the nominal pre-launch knowledge of the in-
strument. In this simplified schema, it is acceptable to adopt as
thetwoFoVsandvariesacrossthefocalplane.Thisaddsasys-
tematiceffecttothephotometricdatathatisordersofmagnitude the reference wavelength the nominal value which corresponds
larger than expected and in particular affects our ability to cre- to the central sample of a perfectly centred window and to as-
sume that the reference spectrum (being the result of an accu-
ate a consistent reference catalogue for the internal calibration.
mulation over an extremely large number of observed spectra)
An additional calibration of this strong time dependency in the
willberepresentativeofaperfectlycentredspectrum.
transmissionhadtobeincludedtosolvethisproblem.
Thegeometriccalibrationmodelisdefinedbythefollowing
components(formoredetailsseeCarrascoetal.2016):
3. ValidationofBP/RPspectralcalibrations – a large-scale component, computed over a short timescale,
definedbyalinearcombinationofshiftedLegendrepolyno-
EventhoughcolourinformationisnotincludedinGaiaDR1,BP mials describing overall effects of translation, rotation, and
andRPdataareprocessedtoproducethecolourinformationre-
curvature;
quiredtocalibratetheG-bandphotometry.Thissectionfocusses – anoffsetfordifferentgateconfigurations(relativetotheun-
onthevalidationofthealong-scangeometriccalibrationofthe gatedcase)computedonalongertimescale,takingintocon-
spectrophotometric data. This is a fundamental element in the siderationtheresidualeffects;
computationofcolourinformationintheformofspectrumshape – anoffsetfordifferentCCDACstitchblocks,alsocomputed
coefficients (SSC; see Carrasco et al. 2016). The application of on a long timescale, taking into account the effects due to
Articlenumber,page2of15
D.W.Evansetal.:GaiaDataRelease1
Fig. 1: Evolution in time of the zeroth-order coefficients of the Fig.2:Evolutionintimeofthefirst-(distributedbetween4.5and
large-scalecomponentoftheBPgeometriccalibration.Theunits 6 ms) and second-order coefficients (distributed between −0.5
onthe ordinateaxisarems (theTDIperiodis 1ms).The units and 0.5 ms) of the large-scale component of the BP geometric
ontheabscissaaxisareOBMTrevolutions(onerevolutioncor- calibration.Theunitsontheordinateaxisarems. Theunitson
responds to approximately 6 hours). The OBMT range covers the abscissa axis are OBMT revolutions (one revolution corre-
theentirescienceacquisitionperiodforGaiaDR1,i.e.between sponds to about 6 hours). The OBMT range covers the entire
25July2014and16September2015.Differentcoloursareused scienceacquisitionperiodforGaiaDR1.Colour-codingisasin
toindicatedifferentCCDrows(CCDrows1to7fromredtovio- Fig.1.
let,lightercoloursfortheprecedingfieldofview,darkeronesfor
thefollowingfieldofview).Eachlarge-scalecalibrationcovers
atimerangeofabout20revolutions(5days). Figure 2 shows the time evolution of the first- and second-
order large-scale coefficients. The colour-coding is the same as
inFig.1.Thesecond-ordercoefficientsarealwaysverycloseto
thephotolithographyprocessusedtomanufacturetheCCDs 0.Bothsetsofcoefficientsarequitestable.
(for more details on the definition of the stitch blocks see Finally, Fig. 3 shows the offset calibrated for different gate
GaiaCollaborationetal.2016b). configurations.Theonlygatesthatcouldbecalibratedoverthe
wholeperiodareGate09,Gate11,Gate07,andGate05.Theco-
Suddenvariationsinthevaluesofthecalibrationcoefficients efficients for Gate05 and Gate07 are quite noisy (the width of
overtimeshouldonlytakeplacecorrespondingtoparticularand thedistributionoftheparametervaluesforthesetwoconfigura-
knownsatelliteeventsorfeatures/changesintheinputdatapro- tionsis0.22and0.14pixelstobecomparedwith0.02obtained
duced by the upstream systems. Therefore, the main validation forbothGate09andGate11).Thisisduetothesmallamountof
analysis is based on the temporal evolution of the calibrations. dataavailableforthesecalibrations.Timerangescoveringabout
Figure1showstheevolutionversustime(inOBMTrevolutions, 160revolutions(40days)wereusedforthisrunofthegateoffset
whereonerevolutionlastsapproximately6hours;seeGaiaCol- calibration.Longertimerangescouldbeadoptedinfutureruns
laboration et al. 2016b) of the zeroth order coefficients in the ifthecalibrationsaresufficientlystable.
large-scale component. Some known events are marked in the The final set of coefficients, calibrating small-scale effects
plotusingverticallines(twodecontaminationactivitiesindark on the scale of CCD AC stitch block, produces offsets that are
green and two refocus activities in blue). As expected, these always below 0.05 pixel in absolute value and are quite stable
significantly affect the calibrations. Decontamination activities (the width of the distribution over time of these parameters is
wereintroducedtomitigatetheproblemofcontaminationaffect- lowerthan0.02pixelsineverycalibrationunit).
ing mirrors and CCDs. During these activities, the mirrors and The validation of the geometric calibrations is primarily
CCDs were heated. The decontamination and refocus activities basedontheanalysisofthestandarddeviationofthesinglecal-
mainlyaffectthebasicangle,whiletheyseemtohaveanalmost ibrations. The standard deviation of the zeroth-order parameter
negligibleeffectontherelativegeometryofAFandBP/RP. foreachsinglecalibrationunit,overthe6monthsfollowingthe
Itisinterestingtonotethatthevariationsfollowingadecon- first refocus event, is of the order of 0.06 ms for BP (equiva-
taminationseemtotakeplacewithsomedelay.Thisislikelydue lent to 0.06 pixel or 0.5 nm in terms of wavelength) and 0.15
tothefactthatdatacollectedjustafteradecontaminationevent ms for RP (i.e. 0.15 pixel or 1.65 nm in terms of wavelength).
arenotofsufficientqualitytogenerateanupdatedAFgeomet- Errors of this size are negligible when computing the spectrum
ric calibration, and therefore this update takes place only once shapecoefficientsusedforthephotometriccalibrations.Thisis
theentireinstrumenthascooleddownsufficiently.Variationsin the only relevant quantity for Gaia DR1 as spectral data is not
theleveloftheBP/RPgeometriccalibrationcoefficientsareex- yetincludedintherelease.
pected whenever a new geometric calibration for the AF field It should be noted that systematic errors on the geometric
comesinplace.ThisoccursbecausetheextrapolationoftheAF calibrationparameterswouldnotaffectthephotometriccalibra-
centroidstotheBP/RPCCDsdependsontheAFgeometriccal- tionsastheywillsimplyresultinaslightlydifferentsetofSSC
ibration. bandsbeingusedforthedefinitionofthecolourinformation.
AscanbeseenfromFig.1,thelarge-scalecomponentisvery TheRPresults(notshowninthispaper)areequivalenttothe
stableoverstretchesofnominaloperations. BPones.
Articlenumber,page3of15
A&Aproofs:manuscriptno.GAIA-CS-CP-IOA-DWE-050Final
300
BP
200
n
100
0
RP
100
n
0
0 0.15 0.3 0.45
error[e/pix/sec]
BP ROW1 BP ROW2 BP ROW3 BP ROW4 BP ROW5 BP ROW6 BP ROW7
Fig. 3: Evolution in time of the gate offset coefficients of the Fig. 6: Error distribution (in electron/pixel/s) for the stray light
large-scalecomponentoftheBPgeometriccalibration.Theunits maps built with 1D observations (colour-coded by CCD row).
on the ordinate axis are ms. The units on the abscissa axis Top:BP.Bottom:RP.
are OBMT revolutions (one revolution corresponds to about 6
hours). The OBMT range covers the entire science acquisition
periodforGaiaDR1.Differentcoloursareusedtoindicatedif- tron/pixel/s).Forthemapsobtainedwiththe1Dtransits,thedis-
ferentgateconfigurationsasindicatedbythelabels.
tributions are broadly similar for all CCDs. However, while in
RPthereisaclosesimilaritybetweentherows,inBPthereare
moredifferences;thisisexplainedbythefactthatinRPthestray
4. ValidationofBP/RPstraylightcalibration
lightfeaturesaresimilarforallCCDs,whileinBPthefeatures
canchangesignificantlyinshape,position,andstrength.Forthe
AsdescribedinRielloetal.(2016),thecurrentimplementation
maps obtained with the 2D transits, it is evident that there are
of the background correction takes into account only the stray
two distributions: the first between 0 and 0.05, very similar to
light calibration as it is the most important contribution to the
that for 1D transits and the second between 1.35 and 1.4. The
background. As Carrasco et al. (2016) have noted, this correc-
latter is due to the stray light bins with only one measurement,
tion of the stray light also includes the smoother component of
so that the error in that case is not the error on the median but
the astrophysical background. The stray light is modelled as a
theerroronthatsinglemeasurement.
discrete2Dmap,obtainedbyaccumulatingeightrevolutionsof
data(correspondingtoroughly2days).Themapcoordinatesare Theresidualsobtainedbysubtractingthemapfromthesame
theheliotropiccoordinatespinphaseandtheACcoordinate.The data used to calculate it have also been analysed. Figure 9
1Dand2Dtransitsareprocessedseparatelybecauseanalysisof shows examples of residual histograms. The histograms were
thedatashowsthatwhilethestructureofthemapisverysimi- normalised to the same area to allow a better comparison. The
lar,thereisasmalloffset(stillunderinvestigation)betweenthe distribution is well centred around zero, with a different width
two.Sincetherearemanymore1Dtransitsthan2Dtransits,they for 1D and 2D transits, showing that the model is correct and
have a much higher weight in the determination of the bin val- thatthereisnoresidualtrend.
ues,andusingamapbuiltwithbothkindsofobservationsleads An additional check is the calculation of the scatter of the
toanovercorrectionwhenremovingthebackgroundforthe2D residuals, made using an interquartile method used in the Hip-
transits. The amplitude of this offset depends on the CCD and parcos mission (ESA 1997) which uses the percentile values at
variesbetween0and2electron/pixel/s.Thisdoesnotaffectthe 15.85and84.15%torobustlyestimatethestandarddeviationof
photometrybecauseitiscalibratedoutbycreatingseparatemaps the distribution (see also Sect. 9). The scatter is lower for 1D
for 1D and 2D windows. In addition, the grid used to build the than2Dobservations,asshowninFig.9(leftpanel)wherethe
mapsfor1Dand2Dtransitsaredifferent,with360binsinphase valueis∼ 0.053electron/pixel/sforresidualsfrom1Dobserva-
and 20 in AC coordinate in the former and 180 and 15 in the tions,whileitis∼ 0.131electron/pixel/sforresidualsfrom2D
latter.Thisallowsfeweremptybinsforthemapsbuiltwith2D observations.Thisisexpected,sincefor2Dwindowsthenumber
transits. ofobservationsismuchlower(about10%ofthenumberof1D
Examples of stray light maps obtained with 1D transits for windows)andthereforethemodelislessaccurateandtheresid-
BP and RP are shown in Fig. 4, while maps obtained with 2D ualsarebigger.ThesameresultsapplytoRPaswell.Theworst
transitsareshowninFig.5. case,shownin therightpanelofFig.9,iswhenthevariationsin
Afirstvalidationisdonedirectlybyinspectingthemapand theACandALdirectionsarequitelarge,butthisisexpectedas
looking at the distribution of the errors. The value in a bin is wellsincetheresolutionofthemapisnotsufficienttoreproduce
themedianvalueobtainedfromallobservationscontributingto the rapid variations. Unfortunately, increasing the resolution of
the bin. The median was chosen instead of the mean to reduce themapisnotanoptionbecausethereissimplynotenoughdata
theeffectofoutlierscausedforinstancebycosmicraysorcon- available to robustly measure the background level. This scat-
tamination from stars. The error for a bin is calculated as the tertranslatesintoanerrorinmagnitudewhichiswellbelowthe
median absolute deviation (MAD) associated with the median expectedend-of-missionerror:intheworstcase,thevaluesare
value. Figures 6, 7, and 8 show the histograms for the stray comparablebutitshouldbenotedthattheend-of-missionerror
light map bin errors, with a histogram bin size of 0.001 (elec- is calculated based on the accumulation of the measurements,
Articlenumber,page4of15
D.W.Evansetal.:GaiaDataRelease1
1D Values BP ROW5 rev 1086.0 1D Values RP ROW5 rev 1086.0
10 10
1750 1750
c] c]
1500 x/se7.5 1500 x/se7.5
pi pi
AC[pix]11020500 erBin [e-/ 5 AC[pix]11020500 erBin [e-/ 5
750 eP 750 eP
u u
al al
500 V 500 V
n2.5 n2.5
a a
250 Me 250 Me
0 0
0 45 90 135 180 225 270 315 360 0 0 45 90 135 180 225 270 315 360 0
phase[deg] phase[deg]
Fig. 4: Stray light discrete maps for BP and RP in the left and right plots, respectively, built with 1D observations (see text for
details).InabscissaisthespinphaseandinordinatetheACcoordinate.Thebinvaluesinelectron/pixel/sarecolour-codedasinthe
bartotheleft.Revtimeinthefigurelabelindicatesthestartofthetimerangewhenthedatawereacquired,inthiscaserevolution
1086.
2D Values BP ROW5 rev 1086.0 2D Values RP ROW5 rev 1086.0
10 10
1750 1750
c] c]
1500 x/se7.5 1500 x/se7.5
pi pi
AC[pix]11020500 erBin [e-/ 5 AC[pix]11020500 erBin [e-/ 5
750 eP 750 eP
u u
al al
500 V 500 V
n2.5 n2.5
a a
250 Me 250 Me
0 0
0 45 90 135 180 225 270 315 360 0 0 45 90 135 180 225 270 315 360 0
phase[deg] phase[deg]
Fig.5:SameasinFig.4for2Dobservations.Itshouldbenotedthattheblacksquaresindicatethatthereisnoinformationforthat
bin.Interpolationisdoneforthesecases.SeeRielloetal.(2016)formoredetails.
2D BP 2D BP
1,250
120
BP
1,000 100
80
750
n n 60
500
40
250
20
0 0
0 0.5 1 1.5 2 0 0.1 0.2 0.3 0.4 0.5
error[e/pix/sec] error[e/pix/sec]
BP ROW1 BP ROW2 BP ROW3 BP ROW4 BP ROW5 BP ROW6 BP ROW7 BP ROW1 BP ROW2 BP ROW3 BP ROW4 BP ROW5 BP ROW6 BP ROW7
Fig.7:Errordistribution(inelectron/pixel/s)forthestraylightmapsbuiltwith2DobservationsforBP(colour-codedbyCCDrow).
Theplotontheleftshowsalldata,whiletheoneontherightcontainsonlytheerrorvaluesforthemapbinswithmorethanone
measurement.Seetextfordetails.
while the scatter is calculated on single measurements and will two ways. The first is used in the validation of the calibrations
decrease. themselves,andthesecondisusedinthevalidationofthepho-
tometryasawholeinthedetectionofanomalies.
When the calibrations are carried out, the unit-weight stan-
5. StudyofLSandSScalibrationcoefficients
darddeviationofthesolutioniscalculated.Thisisdefinedasthe
As described in Carrasco et al. (2016), two of the main pho- square root of the normalised chi-square (van Leeuwen 2007).
tometric calibrations are referred to as the large-scale (LS) and Thisgivesanindicationofhowwellthesolutionmodelisableto
small-scale(SS)calibrations.Theycanbeusedforvalidationin removeanysystematiceffects.Intheidealcase,thisvalueshould
Articlenumber,page5of15
A&Aproofs:manuscriptno.GAIA-CS-CP-IOA-DWE-050Final
2D RP 2D RP
1,250
120
RP
1,000 100
80
750
n n 60
500
40
250
20
0 0
0 0.5 1 1.5 2 0 0.1 0.2 0.3 0.4 0.5
error[e/pix/sec] error[e/pix/sec]
RP ROW1 RP ROW2 RP ROW3 RP ROW4 RP ROW5 RP ROW6 RP ROW7 RP ROW1 RP ROW2 RP ROW3 RP ROW4 RP ROW5 RP ROW6 RP ROW7
Fig.8:AsinFig.7forRP.
BP ROW1 BP ROW5
20 7
18 6
15 5
12
4
n10 n
3
8
2
5
2 1
0 0
-0.50 -0.25 0.00 0.25 0.50 -0.50 -0.25 0.00 0.25 0.50
residuals (e-/pix/sec) residuals (e-/pix/sec)
Fig.9:Residualdistribution(normalisedtothesamearea)forthestraylightmaps.Inbluethedataobtainedfrom1Dtransits,in
yellowfrom2Dtransits.Theleftplotshowsthebestcase(forBPROW1withscatter∼0.031for1Ddataand∼0.057for2Ddata),
whiletherightplotshowstheworstcase(forBPROW5,withscatter∼0.075for1Ddataand∼0.161for2Ddata).
be around 1.0. However, in these early stages of the mission,
it is not expected that the values found in the solutions would
be close to ideal, either because the calibration model does not
represent the systematics very well or because the quoted er-
rors on the fluxes do not correctly represent the true error (or
both).Figures10and11showexampleplotsofthestandardde-
viation for the large- and small-scale calibrations, respectively.
Where the standard deviation varies from the average value, it
indicatesaregionwherethecalibrationmodelisworseatmod-
ellingthesystematiceffectsandthatapossibleimprovementor
additional calibration feature is required. In the example of the
large-scalecalibration(Fig.10),theaveragevalueof5.0implies
that the observed scatter in the data for this configuration will
be5timesworsethanthequotederrorsforthoseperiodswitha
standarddeviationof5.0.Thisonlyaffectstheindividualtransit
measurements.Itshouldbenotedthattheerrorontheweighted
meanwillnotbeaffectedbythissincethemeasuredscatterhas
been accounted for in its calculation (see Carrasco et al. 2016, Fig. 10: Unit-weight standard deviation of the large-scale cali-
formoredetails). brationasafunctionoftime(insatelliterevolutions)foranex-
amplecalibrationunit.Inthiscase,AF6,Row1,WindowClass
In the example shown for the large-scale calibration
1, No Gate. The black lines are for the preceding and red for
(Fig.10),thepeaksseencorrespondtoshortperiods,sometimes
thefollowingFoVcalibrationunits.Theverticallinesrepresent
individual calibrations, which indicate problems with the IPD
significant satellite events: scanning law change (magenta), de-
(seeFabriciusetal.2016,formoredetails),suchastheuseofan
contamination(green),andrefocussing(blue).
incorrect orsuboptimalLSF/PSFlibrary. Future processingcy-
cles will use redetermined IPD values for which many of these
features will have been corrected. The period immediately af-
terthefirstdecontamination(withintheperiodcoveredbyGaia having reached thermal stability. The quality of these few days
DR1)maybeproblematicowingtothefocalplanepossiblynot isbeinginvestigatedfurther.Alsoseeninthisplotisanindica-
Articlenumber,page6of15
D.W.Evansetal.:GaiaDataRelease1
AF9 ROW7 None AF:ClassOne N = 269414
obs
17.5
15.0
n
o
ati12.5
vi
e
D10.0
d
ar
d 7.5
n
a
St
5.0
2.5
0.0
0 250 500 750 1,000 1,250 1,500 1,750 2,000
AC position
Fig. 11: Unit-weight standard deviation of the small-scale cal- Fig.12:Effectivezeropointofthelarge-scalesensitivitycalibra-
ibration as a function of across-scan position on the CCD for tion as a function of time for an example calibration unit. The
anexamplecalibrationunit.Inthiscase,AF9,Row7,Window calibrationunitisthesameasinFig.10,asaretheverticallines.
Class1,NoGate.TheredlinesshowthelocationsoftheCCD For this plot the SSC terms of the calibration model have been
stitch blocks and the green dots show the location of detected combined to form an effective zeropoint using default colours.
badcolumns. Thisisnecessarysincethereisnozeropointterminthecalibra-
tionmodel.SeeCarrascoetal.(2016)formoredetails.
tionthattheperiodbetweenthechangeinthescanninglawand
the first decontamination is of a poorer quality for the preced- AF9 ROW7 None AF:ClassOne N = 269414
obs
ing FoV in comparison to the rest of the Gaia DR1 period. It
should be noted that for Gaia DR1, the calibrations are carried 0.015
outapproximatelyeveryday,whichishowtimevariationinthe
0.010
responsefunctioniscalibrated. nt
For the small-scale calibration (Fig. 11), the main features poi 0.005
o
seen in the standard deviation plots arise from bad columns. er
ManyoftheseareconfirmedintheCCDhealthcalibrations(see nt z 0.000
e
Fmaibssriicoinu,stheitsailn.f2o0r1m6a,tfioonrmwoilrlebdeeutasields)t.oInmtahsekltahteerasfftaegcetesdosfatmhe- uival-0.005
q
plesaspartofthePSFfitof2DwindowsperformedbytheIPD E-0.010
process(seeFabriciusetal.2016).
Plotting the various calibration coefficients from the solu- -0.015
tions as a function of time (LS) and AC position (SS) is also
-0.020
agoodwayto identifyanomaliesandtoindicatewhere further 0 250 500 750 1,000 1,250 1,500 1,750 2,000
AC position
investigationisrequired(seeFigs.12and13).
The main features seen in the large-scale calibration plots
(Fig. 12) are the changes in the response of the CCD due to Fig. 13: Zeropoint of the small-scale sensitivity calibration as
the varying levels of contamination on the mirrors and CCDs. a function of across-scan position on the CCD for an example
Asthemissionprogressed,morecontaminantwasdepositedon calibration unit. As defined in Carrasco et al. (2016), 1.0 has
themirrorsandCCDs,thusreducingtheefficiencyoftheoverall been subtracted from the zeropoint. The calibration unit is the
system.TheresponseisdifferentbetweenthetwoFoVs;themir- sameasinFig.11.TheredlinesshowthelocationsoftheCCD
rorsassociatedwiththefollowingFoVaremorehighlycontam- stitch blocks and the green dots show the location of detected
inated. As already mentioned in Sect. 3, two decontamination badcolumns.
campaigns were performed during the period covered by Gaia
DR1.Thissuccessfullyimprovedthephotometricthroughputas
can be seen from Fig. 12. However, the contamination was not asmallvariationintheresponse ataroundACposition300for
fullyremovedandcontinuedtoincreasewithtime,albeitatare- asmallnumberofcolumns.Inthiscase,thereisnocorrespond-
ducedrate.Itshouldbenotedthatsomeofthespikesseeninthe ingspikeinthestandarddeviationplotindicatingthatthemodel
standarddeviationplots(Fig.10)arealsoseeninthecoefficient is reasonably correct and that this does represent a genuine re-
plots(Fig.12). sponsevariation.
The main variation mapped out by the small-scale calibra- It should be noted that for Gaia DR1, only a single set of
tion is the response as a function of AC position on the CCD SScalibrationsspanningtheentiretimerangewascomputedin
(Fig. 13). This is effectively a 1D flat field. Again, the bad ordertoensureenoughcalibratorsatthebrightendofthemagni-
columnspresentontheGaiaCCDscanbeseeninthezeropoints tudescale.InordertoverifythattheSScalibrationsareindeed
ofthesmall-scalecalibrations.Whencombinedwiththematch- stable over the entire time range, the period was divided into
ingstandarddeviationplot(Fig.11),thisshowsthatthecurrent threeandasetofcalibrationswasderivedforeachone.Nosig-
modelisnotappropriateforthesecolumns.Thisplotshowsalso nificantvariationwasseenbetweenthethree setsofcalibrations
Articlenumber,page7of15
A&Aproofs:manuscriptno.GAIA-CS-CP-IOA-DWE-050Final
Fig. 14: Zeropoint of the small-scale sensitivity calibrations as Fig.16:DistributionofthenumberofG-bandCCDtransitsfor
a function of across-scan position on the CCD (same calibra- eachsourceanalysed.
tion configuration as Fig. 11) where the time range of Gaia
DR1hasbeendividedgivingthreesetsofcalibrations.Wenote
the change in the ordinate scale. This plot was derived from a ofconvergencemetricneedstobeused.Theonechosenwasan
preparatoryprocessingrun(OR5S3). L1normandwaschoseninpreferencetotheL2normsinceitis
morerobusttooutliers.ThegeneralformoftheL1normis
(cid:90)
L1 Norm metric
|pi(x)−pi+1(x)|dx, (1)
0.0250
0.0225
wherep correspondstothecalibrationfactorfortheithiteration
i
0.0200 andxthesingularparametersofasource.Ifthisintegraliscar-
etric0.0175 riedoutoverarepresentativerangeofparameterspace,thenorm
M0.0150 representsthetypicalchangeinthecalibrationfactorswhengo-
ned 0.0125 ingfromoneiterationtothenext.Inthisanalysis,thiswasdone
bi byusingthesingularparameters(e.g.colour)ofabout1000ran-
m0.0100
o domlyselectedsources.Theoverallmetricusedwasthemedian
C
0.0075 valueofthenormsforthecalibrationsconsidered.
0.0050 Figure15showstheconvergencemetricfortheG-bandWin-
dow Class 2 calibrations (G > 16). The final data point shows
0.0025
the difference between the final LS calibrations of the iteration
0.0000
0 1 2 3 4 5 6 7 8 9 10 11 stageandtheLScalibrationsperformedaftertheSScalibrations
Iteration have been carried out. It can be seen that the photometric sys-
temconvergesverywell.Afterfiveiterationswerecarriedout,it
Fig. 15: Convergence metric as a function of iteration for the wasdecidedtostoptheinitialisationprocessconsideringthatthe
G-bandWindowClass2large-scalecalibrations.Thesemetrics changeshadreachedthemmaglevel.Infuturereleases,further
comparethelarge-scalecalibrationsbetweentwoiterations,the iterationswillbecarriedouttoimproveonthisperformance.
onenumberedintheplotandthefollowingone.Theexception
isthefinalpointwhichcomparesthelarge-scalecalibrationsat
the end of the initial set of iterations and those done after the 7. Analysisofaccumulationdata
small-scalecalibrationshavebeencarriedout;seeCarrascoetal.
AsdescribedinCarrascoetal.(2016),thedataforeachsource
(2016)andRielloetal.(2016)formoredetails.
isaccumulatedandvariousstatisticsgathered.Figure16shows
the distribution of the number of G-band CCD transits for
each source analysed. To remove most of the spurious detec-
at the level that was required for Gaia DR1 (see Fig. 14). The
tions made by Gaia, which are mainly around bright sources,
variationinthisplot,typicalforWindowClass1(13< G <16),
the validation analysis has a lower cut-off of 30 CCD transits
was0.17mmagasmeasuredbytherobustwidthmentionedear-
(roughly corresponding to 3 FoV transits) as seen in this his-
lier.
togram.Becausetheyarespurious,suchdetectionsareunlikely
tobematchedwithotherobservations(seethesectiononcross-
matches in Fabricius et al. 2016) and such “sources” will thus
6. Convergenceofthelarge-scalecalibrations
have low numbers of CCD transits accumulated. The average
As described in Carrasco et al. (2016), the photometric system numberofG-bandCCDtransitsforGaiaDR1isjustunder100
needs to be established in the initial stages of calibration. This (withmeanandmedianequalto97and79,respectively)which
is done by iterating between the large-scale calibration and de- correspondstoabout10FoVtransits.Thespreadinthenumber
termining the reference fluxes using the latest iteration of cali- ofobservationsisduetothescanninglaw,andsomesourceswill
brations.Inordertoshowthatthesystemisconverging,aform havesignificantlymoreobservationsthantheaverage.
Articlenumber,page8of15
D.W.Evansetal.:GaiaDataRelease1
TheincreaseinerrorseenatG=16islinkedtothechangein
thesizeofthewindowconfiguration.Intherange16<G<17,a
limitisreachedintheaccuracy,probablycausedbyIPDissues.
At G=13, the window class changes from 1D to 2D win-
dows for the brighter transits and the IPD algorithm therefore
changes(Fabriciusetal.2016).ThegreatesteffectisthatanAC
LSFcomponentisneededinthefitting.Atthisearlystageofthe
mission, the best AC LSF to use is not very sophisticated and
doesnotincludecolourorACvelocitydependencies.Although
the colour will remain the same for each observation for most
sources,theACvelocitywillnot.Thismeansthatanadditional
noise is introduced into the flux determination. Moreover, at
thispointtheeffectoffluxlossaffectsobservationsfainterthan
G=13. When initialising the photometric system from raw ob-
servations,caremustbetakentomakesurethatdiscontinuities
are not introduced into the system. This is described further in
Sect.4inCarrascoetal.(2016).Ifthereareproblemswiththis
Fig. 17: Distribution of error on the weighted mean G-value as calibration,thenalargerscatterwillbeseenforsourcesaround
afunctionofmagnitude.Theorangelineshowsthemodeofthe thismagnitude.
distribution. This plot is restricted to all sources with between Atthebrightendsomeoftheincreasedscatteriscausedby
90 and 110 CCD transits. The green line shows the expected saturation.Settingthegateconfigurationatthetimeofobserva-
errorsforsourceswith100CCDtransitsandforanominalmis- tion should remove most of the saturation by changing the ef-
sionwithperfectcalibrations.Theredlineshowsthesameerror fectiveexposuretime;however,theaccuracyoftheon-boardde-
function,butwithacalibrationerrorof3mmagaddedinquadra- terminationofthesourcemagnitude,whichdeterminesthegate
turetotheindividualobservations.Thedashedblacklinehasa configuration, is poor (about 0.3 mag) at the bright end (Gaia
slope of 0.4 and indicates that the faint end is sky dominated. Collaborationetal.2016b).Thismeansthatsomeobservations
Thedistributionhasbeennormalisedalongthemagnitudeaxis, arecarriedoutwithagateconfigurationthatdoesnoteliminate
i.e. scaled so that each magnitude bin has the same number of saturation. While some masking of saturated pixels is carried
sources in order to show features along the whole magnitude out by the IPD, the calibration library used for this purpose is
range.Thegreyscaleislinear. anearlyversionfromthecommissioningperiodwhichonlyac-
counts for numerical saturation. Updates of this library will be
inplaceforthenextrelease.
Theothervariationsatthebrightendarealsocausedbythe
The distribution of quoted error on the weighted mean1 as differentgateconfigurationsbeingset.Thischangestheeffective
a function of magnitude shown in van Leeuwen et al. (2016)
exposure time for each observation which alters the amount of
cannotbecomparedwithexpectationsbecauseeachsourcehas
smearing caused by the AC velocity. The amount of additional
a different number of observations. Figure 17 shows the same
noiseseenwilldependontheACLSFselected.Atthisstageof
analysis, but is restricted to those sources with between 90 and
themission,novariationintheACLSFismadeasafunctionof
110CCDtransits.Theresultsforthesesourcescanthenbecom-
ACvelocity(Fabriciusetal.2016).
pared with predictions for N = 100 using the formulation
obs From the accumulated data for each source, a P-value can
giveninJordietal.(2010).Thelowerline(green)givestheex-
be calculated from the χ2 of the weighted mean flux calcula-
pected errors for a nominal mission and no calibration errors.
tion.Thisisdefinedastheprobabilitythatthetransitsthathave
Thezig-zagvariationatG<12showstheeffectofgatingwhich
beenusedinformingtheweightedmeanfluxforeachsourceare
changestheeffectiveexposuretimeoftheobservations.Adding
normally distributed about the mean according to their quoted
a3mmagcalibrationerrortothisformulationshowsthegeneral
errors, i.e. there is no additional source of noise, for example
levelofcalibrationthathasbeenachievedforGaiaDR1.
sourcevariability.
Furtherfeaturescanbeseeninthisfigure. TheP-valuescanbecalculatedusingtheequation
At the faint end, the main difference between the nominal (cid:32)(n−1) χ2(cid:33)
andcurrentmissionistheincreasedstraylightlevelwhichleads P= Q , , (2)
2 2
topoorerperformancethanexpected.Thiscannotbecalibrated
outsinceitispurelyanincreaseinthenoiselevel. where Q is an incomplete gamma function (Press et al. 1993)
The jumps at approximately G=13 and G=16 are due to andnthenumberofCCDtransits.
changesinthewindowclasswhichaffecttheIPDalgorithmand Ifthequotederrorsareaccurateandrepresentative,thisisa
verygoodwayofdetectingvariability.However,thisisnotthe
thenumberofpixelspresentintheimagewindow.Itshouldbe
casewiththecurrentG-banddataandthequotederrorsfromthe
noted that they do not occur exactly at these magnitudes since
theplotsaremadewithcalibratedphotometry,whichisdifferent IPDdonotaccountformodelinaccuracies,suchasusinganLSF
thatistoosimpleintheIPDfit.ThismeansthattheCCD-level
totheon-boardmagnitudeestimatesthatwereusedtodetermine
transitswouldbeseenashavinganunderestimatedquotederror.
theconfigurations(gateandwindowclass)foreachobservation.
Aconsequenceofthisunderestimationisthatalmostallsources
haveaG-bandP-valueof0.0andareseenasvariable.Although
1 Thequotederrorontheweightedmeanincludesacontributionfrom nodirectrescalingoftheindividualphotometricerrorsiscarried
themeasuredscatterandthusaccountsforanyunderestimationofthe out for Gaia DR1, the calculation of the error on the weighted
errorsontheindividualtransits.Seethesectiononthe“Referencepho- meanfluxdoestakeintoaccountthescatterofthedataandthus
tometryupdate”inCarrascoetal.(2016). thiserrorisrealistic.
Articlenumber,page9of15
A&Aproofs:manuscriptno.GAIA-CS-CP-IOA-DWE-050Final
2500000
Accumulations RP Full GDR1 2000000
1500000
350,000,000 1000000
500000
325,000,000 0
0.03
300,000,000
275,000,000 0.02
250,000,000
225,000,000 0.01
Number112570050,,,000000000,,,000000000 Residual[mag] 0.00
0.01
125,000,000 −
100,000,000 0.02
−
75,000,000
50,000,000 0.03
25,000,000 − 0 500 A1C0[0p0ix] 1500 196601.0E72.0E73.0E74.0E75.0E76.0E7
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 102 103 104 105
counts
P-Value
Fig.19:DistributionofphotometricresidualsagainsttheACco-
Fig.18:P-valuedistributionforG fortheGaiaDR1sources. ordinateforalldatainAF1CCDsandWindowClass1(assigned
RP
tosourceswithmagnitude13<G <16asestimatedonboard).
ThesituationisdifferentforG andG sincethefluxde- 3000000
BP RP 2500000
terminationiscarriedoutusingasimpleintegrationratherthan 12500000000000
1000000
amodelfit(Carrascoetal.2016).Theerrorsherehavecontribu- 500000
0
0.03
tionsfromphotonnoise,backgrounddetermination,andthege-
ometric and differential dispersion calibrations. As can be seen
0.02
from Fig. 18, the main feature in the P-value distribution for
G isthepeakat0.0,whicheitherindicatesvariabilityorthat 0.01
tshoReuPrccaelsib.rTahtieonsigmnoifidcealnitsflnaottdwisetrllibmutaitocnhebdetwtoetehne0d.0ataanfdor1.t0heinse- Residual[mag] 0.00
dicatesthatthequotederrorsarerealistic.Nosuchflatdistribu- 0.01
−
tionwasseenintheequivalentG-bandanalysis.
0.02
−
0.03
8. Analysisoftheresiduals − 1200 1400 1600 180O0BMT[r2e0v0]0 2200 2400 2600 01.0E72.0E73.0E74.0E75.0E76.0E7
101 102 103 104 105
counts
Adetailedanalysisoftheresidualsallowsustovalidatethecor-
rectness of the calibration models by showing that there are no Fig.20:Distributionofphotometricresidualsintimeforalldata
systematicdependenciesleftfromthecalibrationparametersaf- in AF1 CCDs and Window Class 1 (assigned to sources with
ter the application of the calibrations. In this case residuals are magnitude in the range 13 < G < 16 as estimated on board).
computedasthedifferencebetweenthecalibratedepochmagni- The time is given in OBMT revolutions (one revolution corre-
tudeandthereferencemagnitudeforeachsource. spondstoapproximately6hours).Verticalsolidredlinesmark
Each calibration unit is calibrated independently and there- theoccurrencesofdecontaminationactivities,whiledashedlines
forewillnaturallyhaveresidualscentredonzero.Wehaveanal- correspondtorefocusevents.
ysed residual distributions for all CCDs in various magnitude
ranges,andindeedcannotseesignificantdifferences.
In particular, residuals do not show any significant depen-
dency on the calibration parameter AC coordinate. Figure 19
shows one such distribution (for the case of the AF1 CCDs
and for the window class configuration nominally assigned to
sourceswithmagnitude13 <G < 16).Thisisrepresentativeof
similardistributionsinotherlocationsonthefocalplane.
In Fig. 20 the distribution in time of the residuals for the
sameCCDsandmagnituderangeusedinFig.19showsanon-
GaussiandistributionoftheresidualsfortheEPSLperiodwhere
the data was heavily affected by contamination and poor LSF
calibrations.Fromthefirstdecontamination(markedbythefirst
continuousverticalredline)onwardsthereisnosignofsystem- -0.005 0.005
aticproblemsintheresidualdistribution.
Fig. 21: Distribution of the median photometric residual in the
Askymapofthemedianphotometricresidual(asshownin
sky for all data in AF1 CCDs and Window Class 1 (assigned
Fig. 21 for the same set of observations used in other residual
to sources with magnitude in the range 13 < G < 16 as esti-
plots in this section) indicates that there are some areas of the
matedonboard).Themapisshownusingequatorialcoordinates
skyandinparticularsomesatellitescansthatwerenotproperly
inMollweideprojection.
calibratedatthe0.01maglevelintheworstcases.
Articlenumber,page10of15