Probing the haze in the atmosphere of HD 189733b with HST/WFC3 transmission spectroscopy

1.7 MB·English

by  N. P. Gibson

Mon.Not.R.Astron.Soc.000,1–9(2002) Printed1February2012 (MNLATEXstylefilev2.2)
Probing the haze in the atmosphere of HD 189733b with
HST/WFC3 transmission spectroscopy
N. P. Gibson1(cid:63), S. Aigrain1, F. Pont2, D. Sing2, J.-M. D´esert3, T. M. Evans1,
G. Henry4, N. Husnoo2 and H. Knutson5
1Department of Physics, University of Oxford, Denys Wilkinson Building, Keble Road, Oxford OX1 3RH, UK
2School of Physics, University of Exeter, Exeter, EX4 4QL, UK
3Harvard-Smithsonian Center for Astrophysics, Cambridge, MA 02138, USA
2 4Tennessee State University, 3500 John A. Merritt Blvd., PO Box 9501, Nashville, TN 37209, USA
1 5Division of Geological and Planetary Sciences, Caltech, 1200 E California Blvd, Pasadena CA 91125, USA
0
2

n Accepted2012January30.Received2012January27;inoriginalform2012January12
a
J

1 ABSTRACT
3
WepresentHubble Space Telescopenear-infraredtransmissionspectroscopyofthe

] transiting exoplanet HD 189733b, using Wide Field Camera 3. This consists of time-
P
series spectra of two transits, used to measure the wavelength dependence of the
E
planetaryradius.TheseobservationsaimtotestwhethertheRayleighscatteringhaze
.
h detected at optical wavelengths extends into the near-infrared, or if it becomes trans-
p parentleavingmolecularfeaturestodominatethetransmissionspectrum.Duetosat-
- urationandnon-linearityaffectingthebrightest(central)pixelsofthespectrum,light
o
curveswereextractedfromtheblueandredendsofthespectraonly,correspondingto
r
t wavelengthranges of 1.099−1.168µm and 1.521−1.693µm, respectively,for the first
s
a visit,and1.082−1.128µmand1.514−1.671µmforthesecond.Thelightcurveswere
[ fitted using a Gaussian process model to account for instrumental systematics whilst

simultaneouslyfittingforthetransitparameters.Thisgivesvaluesoftheplanet-to-star
1
radiusratiofortheblueandredlightcurvesof0.15650±0.00048and0.15634±0.00032,
v
respectively,forvisitoneand0.15716±0.00078and0.15630±0.00037forvisit2(using
3
7 a quadratic limb darkening law). The planet-to-star radius ratios measured in both
5 visitsareconsistent,andweseenoevidenceforthedropinabsorptionexpectedifthe
6 haze that is observed in the optical becomes transparent in the infrared. This tenta-
. tivelysuggeststhatthehazedominatesthetransmissionspectrumofHD189733binto
1
near-infrared wavelengths, although more robust observations are required to provide
0
2 conclusive evidence.
1
Key words: methods: data analysis, stars: individual (HD 189733), planetary sys-
:
v tems, techniques: spectroscopic, techniques: Gaussian Processes
i
X
r
a
1 INTRODUCTION Transitingplanetsalsoprovideanopportunitytoprobe
atmospheric structure and composition, through transmis-
The study of transiting planets is revolutionising our un-
sionandemission(eclipse)spectroscopy.Transmissionspec-
derstanding of planets beyond our Solar System. Transit-
troscopyisameasurementoftheplanet-to-starradiusratio
ing planets are those which pass between us and their host
as a function of wavelength. As the optical depth in the at-
stars, periodically blocking starlight and producing a char-
mosphereiswavelengthdependent,sotooisthealtitudeat
acteristiclightcurve.Thesearevitaltostudiesofplanetary
which the planet becomes opaque to starlight. Wavelength
structure,astheradiusofaplanetanditsorbitalinclination
dependent measurements of the planet-to-star radius ratio
can be measured from the transit, and coupled with radial
are therefore sensitive to atomic and molecular species in
velocity data the mass, bulk density and structure may be
theatmosphere(e.g. Seager&Sasselov2000;Brown2001).
inferred.
Observations with the Hubble Space Telescope (HST)
and Spitzer Space Telescope have provided the most de-
tailed observations of exoplanet atmospheres to date (see
(cid:63) E-mail:[email protected] e.g. Charbonneau et al. 2002; Vidal-Madjar et al. 2003;
(cid:13)c 2002RAS
2 N. P. Gibson et al.
Charbonneauetal.2005;Demingetal.2005).Inparticular, byBertaetal.(2011)toobservethetransmissionspectrum
HD189733bisoneofthemoststudiedexoplanets,beingone of GJ1214b, resulting in near-photon limited observations,
of the brightest known transiting systems. At optical wave- andshowingtheWFC3willproveapowerfulinstrumentfor
lengthsitstransmissionspectrumislargelyfeatureless,dom- such measurements. However, here we are using the WFC3
inatedbyRayleighscatteringfromahighaltitudehazelayer in a noticeably different regime to observe a much brighter
from ∼0.3–1.0µm first detected by Pont et al. (2008) us- star. Sect. 2 describes the observations and data reduction,
ingtheACSinstrument,andrecentlyconfirmedusingSTIS andSect.3describestheanalysisofthelightcurves.Finally
(Sing et al. 2011). The haze is thought to consist of con- Sects. 4 and 5 present our results and conclusions.
densate particles; the most likely candidate being MgSiO
3
grains (Lecavelier Des Etangs et al. 2008a,b). However, Na
hasbeendetectedusinghigherresolutionobservations(Red-
2 WFC3 OBSERVATIONS AND DATA
fieldetal.2008;Huitson2011).TransitsatUVwavelengths
REDUCTION
have shown the planet to have an escaping hydrogen atmo-
sphere (Lecavelier Des Etangs et al. 2010). The situation TwotransitsofHD189733bwereobservedwiththeIRchan-
in the near-infrared is more debated; whilst the transmis- nelofHST/WFC3on2010September4and2010November
sion spectrum obtained with NICMOS from 1.4–2.5µm ap- 10,aspartofprogramGP11740(P.I.F.Pont).Observations
parently showed significant H O and CH features (Swain used the G141 grism, and we obtained slitless spectra from
2 4
et al. 2008), narrow band photometry with NICMOS (Sing ∼ 1.08−1.69µm with a resolution of λ/∆λ ∼ 130 at the
etal.2009)provedtobeinconsistentwiththisyetconsistent central wavelength. For each visit, 592 spectra were taken,
with extrapolation of the optical haze. Spitzer observations precededbyanimagetakenwiththeF167Nfilterinorderto
at3.6µmareagainconsistentwiththehaze,althoughlonger set the wavelength calibration. As HD 189733 is not in the
wavelength observations suggest the presence of molecular continuous viewing zone of HST, each transit was observed
absorption (D´esert et al. 2009, 2011). over 4 half-orbits (∼48 minute blocks). For both visits, the
The interpretation of molecular features in the NIC- first, second and fourth orbits cover the out-of-transit part
MOStransmissionspectroscopyrequiresthehazetobecome of the light curve, and consist of 142, 150 and 150 spectra,
largelytransparentsomewherebetween∼1–1.4µm,andthe respectively,andthethirdorbitcoversthein-transitpartof
transmission needs to show significant temporal variation the observation, consisting of 150 spectra.
for the spectroscopic and narrow-band observations to be As HD 189733 is a particularly bright star (V ∼
consistent. A more likely explanation is that the systemat- 7.67,K ∼ 5.54), we used exposure times of only 0.225 sec-
ics in the NICMOS transmission spectroscopy dataset were onds1. These were taken in multiaccum mode with 2 non-
not fully accounted for as suggested by Sing et al. (2009). destructive reads (rapid, nsamp=2). This means that the
In Gibson et al. (2011a) we reanalysed the NICMOS trans- detector is read out three times per exposure, once at the
missionspectroscopy,andclaimedthatstandardmethodsto start, once mid-way, and once at the end of the exposure.
model instrumental systematic based on linear basis func- Hereafterwerefertotheseimagesasthezeroth,firstandsec-
tions were not reliable in this case, and that the detections ond readouts. The overheads for bright targets with WFC3
ofmolecularspeciesreliedonunjustifiedassumptionsabout are dominated by reading out the detector, and dumping
theformofthesystematicsmodel.Thereforetheuncertain- thetemporarystoragebufferofWFC3totheHSTsolidsate
ties are largely underestimated and not precise enough to recorder.Thereforethecamerawasusedinsubarraymode,
probeforthepresenceofmolecularspecies.Thisinterpreta- reading out only the central 128×128 pixels (irsub128),
tion was largely confirmed in Gibson et al. (2011b), where whichcontainsmost(butnotall)ofthefirstorderspectrum,
wedevelopedaGaussianprocess(GP)modeltoextractthe but excludes the zeroth order image and the second order
transmissionspectrumrobustlyinthepresenceofsuchsys- spectrum. This reduced the readout time and increased the
tematics, although it is important to note that this view is interval between buffer dumps, eliminating the need for a
still disputed by the original authors (Swain et al. 2011). long buffer dump during each orbit. We used the shortest
GPsallowtheuseofnon-parametricmodelsforthein- possible exposure time in this readout mode. This resulted
strumental systematics, and infers the transit parameters in a cadence of ∼18 seconds, the minimum allowed by the
inafullyBayesianframework.Thereforefewerassumptions WFC3 electronics.
aremadeabouttheformofthesystematicsmodel,providing The data were reduced using the latest version of
more robust transmission spectra. The NICMOS transmis- the calwf3 data-pipeline (v2.3). This removes bias-drifts,
sion spectrum of HD 189733b inferred using GPs does not subtracts the zeroth readout from each subsequent non-
exclude the presence of molecular species beyond ∼1.5µm; destructive read, subtracts dark images, and corrects each
however,iftheyarepresent,wecannotconstrainthemvery pixelfornon-linearity.AstheWFC3hasnoshutter,asmany
well with current data. Either way, the question remains of as∼60000e−/pixelwererecordedinthezerothreadoutim-
how far the haze observed in the optical extends into the age. These are accumulated between the final reset of the
infrared.Observationsbridgingthegapbetweentheoptical detector before an exposure (the detector is continuously
and NICMOS data should help resolve the situation.
Herewepresentobservationsdesignedforthatpurpose;
1 It is not possible to defocus WFC3, and at the time of phase
HST transmission spectroscopy of HD 189733 using the
IIpreparation,‘driftscan’mode(McCullough&MacKenty2011)
Wide-Field Camera 3 (WFC3) from 1.08−1.69µm. These
wasnotavailable.Thisinvolvesslowlymovingthetelescopedur-
observations were originally planned with NICMOS as part ing exposures to smear out the cross-dispersion PSF, therefore
ofprogramGP11740(P.I.F.Pont),beforethefailureofthe avoiding saturation of the brightest pixels and allowing longer
NICMOScoolingsystemtorestart.WFC3wasrecentlyused exposures.
(cid:13)c 2002RAS,MNRAS000,1–9
WFC3 transmission spectroscopy of HD 189733b 3
reset when not recording photons or being read out) and
thetimeofzerothread,andmayalsobeinfluencedbyper-
sistence effects. The detector exhibits significant non-linear 120
behaviour, which is correctable at low to intermediate lev-
100
els, and takes the zeroth read counts into account. At high
levels, (cid:38)78000e−/pixel, the response becomes highly non- 80 30
linear and not correctable (Dressel 2011)2. Unfortunately, 1
60 ×
for both the first and second readouts, and for both vis- −
e
40
its, a strip of pixels is saturated along the central region of
the spectra. Fig. 1 shows a plot of the raw images for one 20
spectra,withthezeroth,firstandsecondreadsshown,after
0
subtraction of the bias levels imposed during each detector
reset3. It therefore shows the raw electron counts per pixel,
that contribute to non-linearity. The non-linearity limit of 120
100
(cid:38)78000e−/pixel is surpassed by a significant region in the 80
60 y
cthenerterefooref tthheesfiersrtegreioandsoucta,nannodtmproosvtidoef tuhseefsuelcotnradnrseitadloiguhtt, 0 20 40 60x 80 100 120 02040
curves.Wetriedtoextractusefuldatafromthezerothread
spectra,buttheyresultedinunphysicaltransitdepths,pre-
Figure1.Wireframeplotoftherawimagesthatmakeupatypi-
sumably related to persistence effects in the detector. Sub-
calexposure,withthezeroth(red),first(green)andsecond(blue)
tracting the zeroth read is important to remove this effect reads shown. The bias levels imposed during each detector reset
fromthesubsequentreads.Thepipelinedoesnotapplythe havebeensubtracted,thereforeshowingtherawelectroncounts
non-linearitycorrectiontothezerothreadpriortothis.This perpixelthatcontributetonon-linearity.Thenon-linearitylimit
is a potential source of error in using the current pipeline, of (cid:38) 78000e−/pixel is surpassed by a significant region in the
buttherecordedcountsinthezerothreadarewithinthelin- centreofthefirstreadout,andmostofthesecondreadout,there-
earregime(atleastintheregionsofthespectrumwefinally foretheseregionscannotprovideusefultransitlightcurves.
use), so we assume this effect is negligible on our final light
curves. The pipeline usually only applies the non-linearity
correctiontonon-saturatedpixels;nonethelessweforcedthe over much of the wavelength range in both the first and
pipelinetoapplythecorrectiontoallpixels,andproceeded second readout the flux is non-linear.
to extract spectra for both first and second readouts. Transits for each pixel channel were then constructed
To extract the spectra, we used a technique similar to from time series of the spectra. The transits suffer from in-
that used in Gibson et al. (2011a) for HST/NICMOS spec- strumental systematic noise, which is typical of HST obser-
troscopy, using a custom pipeline written in python. The vations of transit light curves (e.g. ACS: Pont et al. 2008,
spectral trace was first determined for each image by fit- NICMOS:Swainetal.2008,Gibsonetal.2011a,STIS:Sing
ting a Gaussian profile in the cross-dispersion axis (y) for et al. 2011). These may arise due to thermal breathing and
each of 128 columns along the dispersion direction (x). A temperaturevariations,pluspointingdriftswhichaggravate
straight line was fitted to the centres of the profiles to de- pixel-to-pixelsensitivityvariations.Transitsinthesaturated
fine the spectral trace. A sum of 10 pixels was taken along regionsofthespectrapredictablyshowedshallowerthanex-
the cross-dispersion direction centred on the spectral trace pected transit depths (often unphysical). We explored var-
todetermine thefluxforeach pixelchannel, aftersubtract- ious methods to extract useful information from the satu-
ing a background value. The extraction width was chosen rated regions of the spectra. Indeed, this is why we chose
to minimise the out-of-transit RMS after corrections for to use a custom pipeline rather than standard extraction
instrumental systematics were applied (see Sect. 3). The tools, as it allows much more control over the data reduc-
final light curves were not particularly sensitive to vary- tion process, and the benefits of more sophisticated meth-
ing the extraction width between ∼7–15 pixels. The back- ods such as optimal extraction over aperture extraction are
groundwasdeterminedfromaregiononthesamepixelcol- marginalforhighS/Ndata.Onesuchmethodwastosimply
umn above the spectrum, but we also tested a global back- maskthesaturatedpixels(thesamepixelsineveryimage),
ground correction. The final light curves were not sensitive and sum over the remaining pixels. As the FWHM in the
tothechoiceofbackgroundregion,giventhelowbackground cross-dispersion direction is rather narrow (∼2 pixels), the
counts ((cid:39)35e−/pixel) and lack of spatial variation. Fig. 2 saturated pixel contained much of the signal. More impor-
showsspectraextractedforbothrunsandforboththefirst tantly, the systematics resultant from pointing drifts were
and second readouts. The dashed line is first order spectra greatlyenhancedwhenthecentralpixelismasked,assmall
scaled to match the exposure time of the second readout. variations in position cause a large change in the response.
The saturated region is obvious in the second readout, but After further tests, reluctantly we concluded that useful in-
formationcouldnotbeobtainedwherethecentralpixelwas
saturated, and decided to extract two light curves at either
end of the wavelength range in the first readout spectra,
within the unsaturated regions.
2 This is defined as where the detector counts deviate by more
For the first visit, we summed the first 15 pixel chan-
than5%fromlinearity.
3 The bias level does not result from physical charges in each nels, and final 37 pixel channels to form what we will refer
pixelthereforedoesnotcontributetothedetectornon-linearity. to as ‘blue’ and ‘red’ channels, respectively. Similarly, for
(cid:13)c 2002RAS,MNRAS000,1–9
4 N. P. Gibson et al.
age (first readout), ∆Y and ψ from the straight line fits
140 to the spectral trace, and W from the average FWHM of
) theGaussianprofilesfittedtoeachextractioncolumninthe
310×120 cross-dispersion direction. The phase (θHST) of HST was
el (100 also calculated for each time. Unfortunately no useful tem-
n perature information was available in the image headers.
n
a 80 Berta et al. (2011) provide a detailed description of WFC3
h
c
pixel 60 sayttsetemmpattticosrfeoprrotrdauncsemhisesrieo.nHsopweecvtreor,scwoepyn,owtehitchhatwtehdeofonromt
er 40 of the systematics in our light curves are slightly different
p giventhedifferentobservingstrategiesused;mostnotewor-
−
e 20 thy we do not see a prominent ‘ramp’ effect. This is likely
becausewedonotrequirebufferdumpsduringorbits,which
0
0 20 40 60 80 100 120 140 leadstoregularcadence,thereforestablepersistenceeffects
x pixel
for each exposure.
140
)
30120
1× 3 ANALYSIS
el (100
nn 3.1 Gaussian process model
a 80
h
c The light curves were modelled using the GP model de-
xel 60 scribedinGibsonetal.(2011b),whichwebrieflyrecaphere.
pi
er 40 Each light curve is modelled as a GP with a transit mean
p function:

−
e 20
f(t,x)∼GP(T(t,φ),Σ(x,θ)),
0
0 20 40 60 80 100 120 140
where f is the flux, t the time, x the vector containing
x pixel
K auxiliary measurements, T is the transit (mean) func-
tionwithtransit(hyper-)parametervectorφ(includingthe
Figure 2.Examplespectraextractedforoneimageforthefirst
planet-to-star radius ratio, ρ), modelled using the analytic
(top) and second (bottom) visits, showing spectra for both the
first and the second readouts. The dashed green line is the first equations of Mandel & Agol (2002), and Σ is the covari-
readout scaled to the exposure time of the second. Clearly the ance matrix with hyperparameter vector θ. The covariance
spectra of the second readout are saturated, but much of the between two points xn and xm is given by the squared ex-
wavelength range in both the first and second readout is non- ponential kernel function:
linearandnotcorrectabletoascientificallyusefullevel(seetext).
We therefore only extracted light curves from either end of the k(x ,x )=ξexp(cid:32)−(cid:88)K η (x −x )2(cid:33)+δ σ2,
wavelengthrange.Theblueandreddashed-dottedlinesshowthe n m i n,i m,i nm
extractionregionsusedforeachvisit.Notethatthezerothreadis i=1
subtractedpriortoextractingthesespectra,andtheyaresummed whereξisahyperparameterthatspecifiesthemaximumco-
alongthey-axis,hencedonotcorrespondinasimplewaytothe
variance, η are the inverse scale hyperparameters for each
detectorsaturationlimit. i
auxiliary measurement vector, σ2 is the variance due to
white noise, and δ is the Kronecker delta. The kernel
nm
the second visit we extracted the first 11 and final 35 pixel accounts for the instrumental systematics, and this one de-
channelsintotheblueandredchannels.Theseextractionre- scribesasmoothfunctionoftheinputparameters,withthe
gions are marked in Fig. 2. Fig. 3 shows the extracted blue addition of white noise. In other words, the instrumental
and red channel light curves for the first and second visits. systematicsmodelassumesasimilarvaluewhenallrelevant
The wavelength dispersion of the WFC3/IR G141 grism is auxiliary measurements are close, with the ‘closeness’ spec-
position dependent, and the position of HD 189733 in the ified by the inverse length scales. Similarly to other HST
directimagewasusedtocalibratethewavelengths.Theblue analyses, we excluded orbit 1 from the fits. This has the
and red channels in the first visit correspond to wavelength added advantage of significantly speeding up inversion of
rangesof1.099−1.168µmand1.521−1.693µm,respectively, thecovariancematrixandthereforeinferenceoftheplanet-
and1.082−1.128µmand1.514−1.671µm,respectively,for to-star radius ratio for each light curve.
the second transit. ThedefinitionofaGPstatesthatthelikelihoodismul-
Instrumentalsystematicsareevidentinthelightcurves, tivariate Gaussian. We multiply the likelihood by gamma
andmustbeaccountedfortomeasuretheplanet-to-starra- hyperpriors of shape parameter unity for ξ and the η hy-
i
diusratio.Wethereforeextractedauxiliaryparametersfrom perparameters to form the posterior distribution (implying
the data in a similar way to Gibson et al. (2011a) in order uniform,improperpriorsfortheremainingparameters).The
to explore systematics models. These were the shift in x- scale length of the hyperpriors were set to large values so
position (∆X), y-position (∆Y), and average width (W) they have little influence over the inference, other than to
and angle (ψ) of the spectral trace. ∆X was determined ensure the hyperparameters remain positive. The posterior
from cross-correlation of the extracted spectra in each im- distributionoftheplanet-to-starradiusratio(ρ)isfoundby
(cid:13)c 2002RAS,MNRAS000,1–9
WFC3 transmission spectroscopy of HD 189733b 5
1.01 ‘Blue’channels,raw Run1:1.134µm 1.01 ‘Red’channels,raw Run1:1.607µm
1.00 1.00
flux0.99 Run2:1.105µm flux0.99 Run2:1.592µm
ve0.98 ve0.98
relati0.97 relati0.97
0.96 0.96
0.95 0.95
1.01 GPsubtracted 1.01 GPsubtracted
1.00 1.00
ux0.99 ux0.99
fl fl
ve0.98 ve0.98
relati0.97 relati0.97
0.96 0.96
0.95 0.95
1.01 Combined 1.01 Combined
1.00 1.00
ux0.99 ux0.99
fl fl
ve0.98 ve0.98
relati0.97 relati0.97
0.96 0.96
0.95 0.95
0.08 0.06 0.04 0.02 0.00 0.02 0.04 0.08 0.06 0.04 0.02 0.00 0.02 0.04
− − − − − − − −
orbitalphase orbitalphase
Figure 3. Light curves for the blue (left) and red (right) channels. Top: Raw extracted light curves, showing systematics that need
accounting for when inferring the planet-to-star radius ratio. Middle: ‘Cleaned’ light curves after subtracting the Gaussian process
systematics model. Bottom: Combined light curves from both visits. The cleaned light curves are shown for illustrative purposes only;
theplanet-to-starradiusratioisinferredsimultaneouslywiththeGPsystematicsmodel.Thefirstorbitwasexcludedfortheinference,
buttheprojectedGPmodelissubtractednonetheless.
marginalising over all the other transit parameters and co- in Sing et al. (2011). This was to enable easy comparison
variancehyperparametersusingMarkov-ChainMonte-Carlo with ACS and NICMOS transmission spectroscopy, but fu-
(MCMC),wherethefixedtransitparameters4 aresettothe ture studies will use the more reliable 3D limb darkening
valuesofPontetal.(2007,2008)excepttheephemeriswhich parameters. Inference of the planet-to-star radius ratio was
is calculated from the Agol et al. (2010) values5. The light performedforbothlawsholdingthelimbdarkeningparam-
curve function is multiplied by a normalisation (baseline) eters fixed.
functionduringfitting,inthiscasealinearfunctionoftime, We decided to take a different approach than for the
whoseparameterswealsomarginaliseoverintheMCMC.In NICMOSdata;ratherthanusealloftheauxiliaryinforma-
all,wefitfortheplanet-to-starradiusratio,theparameters tion,wewerecarefultouseonlytheinputsthatareneeded
of the normalisation function, and all the GP hyperparam- to model the shape of the systematics per orbit, but impor-
eters including the white noise. For each light curve we ran tantly do not change the flux offsets between orbits. There-
four chains and tested for convergence using the Gelman forewefirsttriedusingonlytheorbitalphaseofHST(θ )
HST
& Rubin (GR) statistic (Gelman & Rubin 1992). For more as input to the GP. This did not provide a satisfactory cor-
details on the GP model and MCMC routine, see Gibson rection for all orbits, particularly for orbits where a discon-
et al. (2011b). The limb-darkening parameters were calcu- tinuity is evident mid way through each orbit in the light
lated for each light curve using the methods described in curves6. We therefore included the cross-dispersion width
Sing (2010) for both quadratic and non-linear limb darken- (W) of the spectrum as an additional GP input. Plots of
ing laws. These were calculated using the 1D stellar atmo- θ andW forbothvisitsareshowninFig.4.Thesewere
HST
sphere models rather than the 3D model atmospheres used normalised to zero mean and variance of 1 prior to their
use as GP inputs to ensure the hyperpriors provide similar
constraints. This GP model provided a satisfactory correc-
4 We fix the scale length a/R(cid:63) and inclination i; these param- tion for the systematics, and the two visits gave consistent
eters are correlated with the planet-to-star radius ratio, but in results. The MCMC chains were well behaved in this case,
transmission spectroscopy we are interested in the posterior dis- andtheGRstatisticwaswithin1%ofunity.Theseareused
tribution of ρ conditioned on these parameters, i.e. the relative asourfinalvaluesforρ.Wealsotestedthismodelincluding
valuesforρ.
5 ThemoreuptodatephysicalparametersofAgoletal.(2010)
were not used here to enable direct comparison with NICMOS 6 The discontinuity is caused by HST crossing the Earth’s
andACSdata. shadow.
(cid:13)c 2002RAS,MNRAS000,1–9
6 N. P. Gibson et al.
by shifts in X, Y and the angle should have little or no
3.0 Visit1 impact. The second is that recent WFC3 analysis by Berta
2.0
H et al. (2011) did not require modelling any flux offsets be-
θ1.0
tweenorbitstofullyaccountforthesystematics.Indeedthe
0.0
‘divide-oot’methodusedshouldprovideasimilarcorrection
(pixels)222...333468 Visit1 tsohotwherGegPulmarordeepleuastiinnggθsHtruacntdurWe faosrienapcuhtos,rbgiitv.enthatthey
W
2.32
3.0 Visit2 3.2 Linear basis function model
2.0
H
θ1.0 As an independent check we also tried using the ‘standard’
0.0 method to model the systematics (see e.g. Brown et al.
xels)22..3324 Visit2 22000018;),Gi.iell.iluasnidng&thAerreixbtarsa2ct0e0d3;aPuoxniltiaertyapl.a2ra0m07e;tSerwsaainsebtasails.
(pi2.30 functions, and fitting for the coefficients using linear least
W2.28 squares(seeGibsonetal.2011a,bforadetaileddescription
0.08 0.06 0.04 0.02 0.00 0.02 0.04 ofthemethodusedhere).UsingW andθ asbasisfunc-
− − − − HST
orbitalphase tions, and higher order squared terms, provided consistent
results with the GP model (within 1σ), and the residual
Figure 4. Plots of θHST and W for both visits as a function permutationmethoddescribedinGibsonetal.(2011a)pro-
of orbital phase of HD 189733. These are the auxiliary inputs vided similar uncertainties. When we added ∆X, ∆Y and
usedasinputvectorsforthefinalGPmodeloftheinstrumental ψ as additional basis vectors the instrument model became
systematics.
unstableandthetwovisitsprovideddifferentplanet-to-star
radiusratiosfortheblueandredtransits,validatingouruse
of W and θ as the only inputs to the GP.
HST
orbit 1 in the fitting procedure, which did not significantly
changetheresults,nordidusing(GP)smoothedversionsof
theW auxiliaryparameter.Thelightcurvesproducedafter
4 RESULTS
subtractingtheGPsystematicsmodelsareshowninFig.3.
WefurthertestedtheGPmodelusing∆X,∆Y andψ 4.1 WFC3 light curve fits
as additional GP inputs. The MCMC did not converge in
The inferred planet-to-star radius ratio for both quadratic
this case, or give consistent results for the two runs. These and non-linear limb darkening laws are given in Tab. 17,
additionalinputsshowsimilarstructuretotheW inputbut
as both have been used in the literature for HD 189733.
are much noisier. This causes degeneracy in the GP hyper-
Theuncertaintiesarecalculatedfromthedistributionofthe
parameters thus demanding much longer MCMC chains to
MCMCchains,afterremovingthefirst20%ofeach,andare
achieve convergence. However, we also ran tests using GP
givenbythelimitswhichencompass68.2%oftheprobability
smoothed input parameters (see Gibson et al. 2011b), and
distribution.
the MCMC now converged, showing that the level of noise
The planet-to-star radius ratio obtained with the
ontheinputscanalsosignificantlyaffectconvergence.This
quadratic limb darkening law are plotted in Fig. 5, and
led to consistent values for ρ when compared to using only
the dashed line is a weighted mean of all four data points.
θ and W as inputs, although with larger uncertainties.
HST Clearlynoplanet-to-starradiusratiovariationisdetectedin
Given that these additional inputs are particularly noisy,
theWFC3data.Acomparisonwithotherdatarequirescor-
and more importantly that they show (small) offsets be-
rectionsforunoccultedstarspots,discussedinthefollowing
tween orbits, we chose not to use these parameters in our
section.
final results, as this can cause flux offsets to be fitted by
the GP. For ψ, the in-transit values are also outside the
range covered by the out-of-transit data, hence requiring 4.2 Spot corrections
an extrapolation to the in-transit data. In theory, these is-
As HD 189733 is an active star, we need to correct the
suesmightbeaddressedwithappropriatepriorwidthsinthe
transmission spectrum for unocculted star spots to make
gammahyperpriors,butthisbecomesrathersubjectivewith
a detailed comparison with data taken at different times
no robust way to set them. Of course, in an ideal Bayesian
and wavelengths. Here, we use the spot corrections as de-
framework we should be able to include all the (noisy) in-
scribed in Sing et al. (2011) in order to place the WFC3
puts.Moresophisticatedtoolstoevaluatetheposteriordis-
data in context alongside other transmission spectra of HD
tributionandmarginaliseoverthehyperparameterssuchas
189733b.However,giventhecomplexityofspotcorrections,
Bayesian Quadrature methods (e.g. O’Hagan 1991) might
thiswarrantsitsownpublication,andwillformalargepart
enable this in the future.
of a future paper (Pont et al, in prep), where we will anal-
Ourapproachofusingθ andW astheonlyinputsis
HST
yse the complete transmission spectra of HD 189733b from
supportedbytwofurtherpiecesofevidence.Thefirstisthat
the intrapixel sensitivity variations are much smaller than
for NICMOS, with no measurable variation found during 7 We have simply given the wavelength range of the extraction
ground-testing for WFC3, compared with ∼ 40% for NIC- region; more detailed analyses would need to take into account
MOS (Dressel 2011); therefore the variations in flux caused thespectralresponse,butisnotwarrantedhere.
(cid:13)c 2002RAS,MNRAS000,1–9
WFC3 transmission spectroscopy of HD 189733b 7
contrastbetweenthespottedandunspottedstellarsurface.
0.159 However, these may still influence the extracted transmis-
sion spectrum.
Fig.6plotsthecorrectedWFC3dataalongsidetheACS
0.158
datafromPontetal.(2008),theNICMOSphotometryfrom
Singetal.(2009),andthereanalysisoftheNICMOStrans-
0.157 mission spectroscopy from Gibson et al. (2011b). The ACS
ρ
data were already corrected for spots using a similar tech-
niqueandweretakendirectlyfromthepaper;theNICMOS
0.156 spectroscopy and photometry were corrected in the same
way as the WFC3 data. The WFC3 data are largely con-
sistent with ACS and NICMOS photometric data, but dis-
0.155
agrees with the red end of the NICMOS spectroscopy. This
1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 tentativelysuggeststhatthetransmissionspectrumisdom-
wavelength(µm) inatedbythehazeintotheNIR,andthefeatureat∼1.6µm
mayindeedbeofinstrumentalorigin(Gibsonetal.2011b).
Figure 5. Measured planet-to-star radius ratio of HD 189733b
However,giventhedifficultyinperformingspotcorrections
fromWFC3datausingquadraticlimbdarkeningpriortospotcor-
(particularlywiththelargebaselinebetweenthedatasets),
rections.Visit1and2areshownbythegreenandyellowpoints,
and in extracting the WFC3 data, it is not wise to rule out
respectively. The dashed horizontal line marks the weighted av-
erageofallfourpoints. the feature at ∼1.6µm from these data alone.
Table 1. Planet-to-star radius ratio calculated from the GP in-
5 DISCUSSION AND CONCLUSION
ference,givenforbothlimbdarkeninglaws.
We have presented HST/WFC3 spectroscopic observations
Wavelength(µm) ρ(Rp) visit of the transiting system HD 189733. The aim of this study
R(cid:63) wastobridgethegapbetweentheACSopticaldataofPont
(quadraticlimbdarkening) et al. (2008), near-infrared transmission spectra with NIC-
1.099−1.168 0.15650±0.00048 1
MOS(Swainetal.2008;Gibsonetal.2011b),andNICMOS
1.082−1.128 0.15716±0.00078 2
photometry of Sing et al. (2009). Unfortunately, the WFC3
1.521−1.693 0.15634±0.00032 1
spectra were saturated for much of the central region, and
1.514−1.671 0.15630±0.00037 2
we could only extract useful light curves at either extreme
(non-linearlimbdarkening)
1.099−1.168 0.15650±0.00047 1 of the G141 grism, corresponding to central wavelengths of
1.082−1.128 0.15735±0.00076 2 ∼1.134µmand1.607µmforvisit1and∼1.105µmand1.592
1.521−1.693 0.15644±0.00032 1 µmforvisit2.Theredandbluelightcurvesshowconsistent
1.514−1.671 0.15672±0.00043 2 transit depths for both runs when using either quadratic or
non-linearlimbdarkeninglaws.Wehavenotdetectedadrop
intheplanet-to-starradiusratioonewouldexpectiftheop-
UV to IR. In addition it is important to model the tran- ticalhazedoesindeedbecometransparentbetween∼1.1and
sit light curves with consistent stellar parameters and limb 1.4µm, as has been suggested to explain the different be-
darkening laws, which will also be addressed in this paper. haviour observed at optical and near-infrared wavelengths.
Correctingforunoccultedspotsrequireslongtermmon- Correction for unocculted star spots shows the WFC3
itoring of the stellar flux to obtain the variation in spot dataareconsistentwiththeACSdataofPontetal.(2008)
coverage between visits, and an estimate of the flux of andtheNICMOSphotometryofSingetal.(2009),butdis-
the unspotted surface. These were obtained by fitting long- agree with the red end of the NICMOS transmission spec-
baseline ground-based coverage from the T10 0.8 m Auto- troscopy of Swain et al. (2008) and the reanalysis of Gib-
matedPhotoelectricTelescope(APT)atFairbornObserva- sonetal.(2011b).Amoredetailedanalysisofthecomplete
toryinsouthernArizona,detailedinHenry&Winn(2008) transmission spectrum of HD 189733b is in preparation,
thatcoversalltheHSTvisitsconsideredhere.Thephotom- with more detailed spot corrections, consistent limb dark-
etry was fitted using GP regression with the quasi-periodic ening laws and stellar parameters. Nonetheless, this work
kernel given in Aigrain et al. (2011). The spot correction tentativelysuggeststhatRayleighscatteringfromthehigh-
was then performed in the same way as described in Sing altitude haze dominates the transmission spectrum up to
et al. (2011), assuming a spot temperature of 4250K mea- andincludingnear-infraredwavelengths.Wenotethatboth
sured from STIS and ACS transits. The spot corrections theWFC3dataandtheNICMOSphotometryofSingetal.
shifted the planet-to-star radius ratio for the blue and red (2009)haveshownconsistenttransitdepths,andhavefailed
light curves by −0.0010 and −0.0006, respectively, for visit to produce the deep feature at ∼1.6µm reported in Swain
one, and −0.0009 and −0.0005 for visit 2. et al. (2008). However, whilst WFC3 appears better be-
We note that no spot-crossing events are visible in our haved than NICMOS transmission spectroscopy, given that
data. It is highly likely that spots were occulted given that the data are near the non-linear range of the detector, and
theopticaltransitsalmostalwaysshowspot-crossingevents the difficulty in performing spot corrections, more evidence
(e.g. Pont et al. 2007; Sing et al. 2011), rather they are isrequiredtoconclusivelyresolvetheissueandprovidecon-
not visible in our data given the lower S/N and lower flux tinuous coverage from optical to near-infrared wavelengths.
(cid:13)c 2002RAS,MNRAS000,1–9
8 N. P. Gibson et al.
0.159
0.158
0.157
0.156
0.155
ρ
0.154
0.153 WFC3Visit1
WFC3Visit2
0.152 ACS(Pontetal. 2008)
NICMOS(Singetal. 2009)
0.151 NICMOS(Gibsonetal. 2011)
0.5 1.0 1.5 2.0 2.5
wavelength(µm)
Figure6.TransmissionspectraofHD189733showingtheWFC3datapresentedinthispaper,theACSdatafromPontetal.(2008),the
NICMOSphotometryfromSingetal.(2009),andthereanalysisoftheNICMOStransmissionspectroscopyfromGibsonetal.(2011b)
aftercorrectionforunoccultedspots(seetext).ThedashedlineistheRayleighscatteringhazefunctiongiveninLecavelierDesEtangs
etal.(2008a).TheWFC3blueandredlightcurvesshowconsistentdepth,providingnoevidencefora‘drop-off’inthehazeobservedat
opticalwavelengths.TheWFC3dataarealsoconsistentwiththeRayleighscatteringhazeextrapolatedintotheNIR;howevergiventhe
difficultiesextractingthelightcurvesandperformingthespotcorrectionswehesitatetodrawfirmconclusionsfromtheWFC3data.
FutureobservationsofHD189733in‘driftscan’modeshould Brown T. M., 2001, ApJ, 553, 1006
help provide this evidence. Brown T. M., Charbonneau D., Gilliland R. L., Noyes
R. W., Burrows A., 2001, ApJ, 552, 699
Charbonneau D., Allen L. E., Megeath S. T., Torres G.,
ACKNOWLEDGMENTS Alonso R., Brown T. M., Gilliland R. L., Latham D. W.,
MandushevG.,O’DonovanF.T.,SozzettiA.,2005,ApJ,
All of the data presented in this paper were obtained
626, 523
from the Multimission Archive at the Space Telescope Sci-
Charbonneau D., Brown T. M., Noyes R. W., Gilliland
ence Institute (MAST). STScI is operated by the Associa-
R. L., 2002, ApJ, 568, 377
tion of Universities for Research in Astronomy, Inc., under
Deming D., Seager S., Richardson L. J., Harrington J.,
NASA contract NAS5-26555. Support for MAST for non-
2005, Nature, 434, 740
HST data is provided by the NASA Office of Space Sci-
D´esertJ.,LecavelierdesEtangsA.,H´ebrardG.,SingD.K.,
enceviagrantNNX09AF08Gandbyothergrantsandcon-
EhrenreichD.,FerletR.,Vidal-MadjarA.,2009,ApJ,699,
tracts. N. P. G and S. A. acknowledge support from STFC
478
grantST/G002266/2.Weareextremelygratefulforthesup-
D´esert J.-M., Sing D., Vidal-Madjar A., H´ebrard G.,
port provided the WFC3 instrument team, in particular H.
Ehrenreich D., Lecavelier Des Etangs A., Parmentier V.,
Bushouse, and discussions with P. McCullough. Finally, we
Ferlet R., Henry G. W., 2011, A&A, 526, A12
thankthereferee,R.Gilliland,forhiscarefulreadingofthe
Dressel L., 2011
manuscript and helpful suggestions.
Gelman A., Rubin D. B., 1992, Stat. Sci., 7, 457
GibsonN.P.,AigrainS.,RobertsS.,EvansT.M.,Osborne
M., Pont F., 2011b, ArXiv e-prints
REFERENCES
Gibson N. P., Pont F., Aigrain S., 2011a, MNRAS, 411,
AgolE.,CowanN.B.,KnutsonH.A.,DemingD.,Steffen 2199
J. H., Henry G. W., Charbonneau D., 2010, ApJ, 721, Gilliland R. L., Arribas S., 2003, Instrument Science Re-
1861 port NICMOS 2003-001
Aigrain S., Pont F., Zucker S., 2011, MNRAS, 1898 Henry G. W., Winn J. N., 2008, AJ, 135, 68
Berta Z. K., Charbonneau D., D´esert J.-M., Miller-Ricci Huitson C., 2011, submitted
KemptonE.,McCulloughP.R.,BurkeC.J.,FortneyJ.J., LecavelierDesEtangsA.,EhrenreichD.,Vidal-MadjarA.,
Irwin J., Nutzman P., Homeier D., 2011, ArXiv e-prints BallesterG.E.,D´esertJ.-M.,FerletR.,H´ebrardG.,Sing
(cid:13)c 2002RAS,MNRAS000,1–9
WFC3 transmission spectroscopy of HD 189733b 9
D. K., Tchakoumegni K.-O., Udry S., 2010, A&A, 514,
A72
Lecavelier Des Etangs A., Pont F., Vidal-Madjar A., Sing
D., 2008a, A&A, 481, L83
Lecavelier Des Etangs A., Vidal-Madjar A., D´esert J.-M.,
Sing D., 2008b, A&A, 485, 865
Mandel K., Agol E., 2002, ApJL, 580, L171
McCulloughP.,MacKentyJ.,2011,WFCSpaceTelescope
Analysis Newletter 6
O’Hagan A., 1991, Journal of Statistical Planning and In-
ference, 29, 245
Pont F., Gilliland R. L., Moutou C., Charbonneau D.,
Bouchy F., Brown T. M., Mayor M., Queloz D., Santos
N., Udry S., 2007, A&A, 476, 1347
PontF.,KnutsonH.,GillilandR.L.,MoutouC.,Charbon-
neau D., 2008, MNRAS, 385, 109
Redfield S., Endl M., Cochran W. D., Koesterke L., 2008,
ApJL, 673, L87
Seager S., Sasselov D. D., 2000, ApJ, 537, 916
Sing D. K., 2010, A&A, 510, A21+
Sing D. K., D´esert J., Lecavelier Des Etangs A., Ballester
G. E., Vidal-Madjar A., Parmentier V., Hebrard G.,
Henry G. W., 2009, A&A, 505, 891
SingD.K.,PontF.,AigrainS.,CharbonneauD.,D´esertJ.-
M.,GibsonN.,GillilandR.,HayekW.,HenryG.,Knutson
H.,LecavelierDesEtangsA.,MazehT.,ShporerA.,2011,
MNRAS, 416, 1443
Swain M. R., Deroo P., Vasisht G., 2011, in IAU Sympo-
sium, Vol. 276, IAU Symposium, A. Sozzetti, M. G. Lat-
tanzi, & A. P. Boss, ed., pp. 148–153
SwainM.R.,VasishtG.,TinettiG.,2008,Nature,452,329
Vidal-Madjar A., Lecavelier des Etangs A., D´esert J.-M.,
Ballester G. E., Ferlet R., H´ebrard G., Mayor M., 2003,
Nature, 422, 143
(cid:13)c 2002RAS,MNRAS000,1–9