Water vapor in the spectrum of the extrasolar planet HD 189733b: 2. The eclipse
Nicolas Crouzet, Peter R. McCullough, Drake Deming, Nikku Madhusudhan
aa r X i v : . [ a s t r o - ph . E P ] S e p Draft version October 8, 2018
Preprint typeset using L A TEX style emulateapj v. 5/2/11
WATER VAPOR IN THE SPECTRUM OF THE EXTRASOLAR PLANET HD 189733b: 2. THE ECLIPSE
Nicolas Crouzet , Peter R. McCullough , Drake Deming , and Nikku Madhusudhan Dunlap Institute for Astronomy & Astrophysics, University of Toronto, 50 St. George Street, Toronto, Ontario, Canada M5S 3H4 Space Telescope Science Institute, Baltimore, MD 21218, USA Department of Physics and Astronomy, Johns Hopkins University, 3400 North Charles Street, Baltimore, MD 21218, USA Department of Astronomy, University of Maryland, College Park, MD 20742, USA NASA Astrobiology Institute’s Virtual Planetary Laboratory, USA and Institute of Astronomy, University of Cambridge, Madingley Road, Cambridge, CB3 0HA, UK
Draft version October 8, 2018
ABSTRACTSpectroscopic observations of exoplanets are crucial to infer the composition and properties of theiratmospheres. HD 189733b is one of the most extensively studied exoplanets and is a corner stone forhot Jupiter models. In this paper, we report the day-side emission spectrum of HD 189733b in thewavelength range 1.1 to 1.7 µ m obtained with the Hubble Space Telescope
Wide Field Camera 3 inspatial scan mode. The quality of the data is such that even a straightforward analysis yields a highprecision Poisson noise limited spectrum: the median 1- σ uncertainty is 57 ppm per 0.02 µ m bin. Wealso build a white-light curve correcting for systematic effects and derive an absolute eclipse depth of96 ±
39 ppm. The resulting spectrum shows marginal evidence for water vapor absorption, but canalso be well explained by a blackbody spectrum. However, the combination of these WFC3 data withprevious
Spitzer photometric observations is best explained by a day-side atmosphere of HD 189733bwith no thermal inversion and a nearly solar or sub-solar H O abundance in a cloud-free atmosphere.Alternatively, this apparent sub-solar abundance may be the result of clouds or hazes which futurestudies need to investigate.
Subject headings:
Planets and satellites: atmospheres — Planets and satellites: individual (HD189733b) — Methods: observational — Techniques: spectroscopic INTRODUCTION
Spectroscopic observations of exoplanets are crucialto infer the composition and properties of their atmo-spheres. The eclipse, when the planet passes behindthe star, probes the day-side emission spectrum of theplanet’s atmosphere. The expected signal for molecularsignatures is a few 100 ppm; this extreme precision is cur-rently achievable only for planets orbiting bright stars.As a result, only a handful of exoplanets have been char-acterized spectroscopically. With a host star of magni-tude J=6.07, HD 189733b is one of the most extensivelystudied exoplanets along with HD 209458b. HD 189733bis a 1.144 M jup , 1.138 R jup gas giant planet orbiting anactive K dwarf in 2.22 days (Bouchy et al. 2005). Spec-tra and phase curves obtained at several wavelengths de-livered quantities of information about its atmosphere.Grillmair et al. (2007, 2008) obtained spectra in emissionwith
Spitzer revealing strong water vapor absorption andthe presence of an extra-absorber in the day-side upperatmosphere. A hot spot was detected eastward of thesubstellar point revealing an equatorial super-rotatingjet (Knutson et al. 2007), as anticipated by atmosphericcirculation models of hot Jupiters (Showman & Guillot2002). Brightness temperature measurements con-strained the pressure-temperature profile of both sidesof the atmosphere and the day-night heat redistributionefficiency (Knutson et al. 2012; D´esert et al. 2011). Inaddition, signatures of non-equilibrium chemistry werereported (Knutson et al. 2012; Swain et al. 2010). Sev-eral molecules were identified in the planet’s trans- [email protected] mission and emission spectra with
Hubble Space Tele-scope (HST)
NICMOS (Swain et al. 2008, 2009, 2014).However, subsequent studies of NICMOS data con-cluded that instrumental systematics may contribute sig-nificantly to the observed spectra (Gibson et al. 2011;Crouzet et al. 2012). More recently, CO and H O weredetected in the day-side of the planet’s atmosphere us-ing ground-based high-resolution spectroscopy with KeckII NIRSPEC and VLT CRIRES (Rodler et al. 2013;de Kok et al. 2013; Birkby et al. 2013). A new era be-gan for exoplanet spectroscopy with the installation ofthe Wide Field Camera 3 (WFC3) onboard
HST in 2009.Gibson et al. (2012) observed two transits of HD 189733bwith WFC3. Unfortunately, the central part of thespectrum saturated the detector; only the edges wereused to derive the planetary radius in two wavelengthranges. The recently implemented spatial scanning mode(McCullough & MacKenty 2012) now allows WFC3 toobserve HD 189733 in much better conditions: this modeis designed to obtain high sensitivity on bright stars forexoplanet spectroscopy by reducing overheads and avoid-ing detector saturation. Observations of HD 189733with WFC3 in spatial scanning mode (HST program12881, PI: McCullough) were obtained during a plane-tary transit (McCullough et al. 2014) and during a plan-etary eclipse (this paper).Although clear atmosphere models are often used to in-terpret exoplanetary spectra, recent results suggest thatclouds and hazes may play an important role in the at-mosphere of exoplanets. Atmospheric clouds would re-sult in an increased opacity which may obscure indi-vidual molecular spectral features. As examples, spec- Crouzet et al.tra of the hot Jupiters HD 209458b and XO-1b with
HST
WFC3 in spatial scanning mode showed a wa-ter vapor absorption feature at 1.4 µ m with an ampli-tude much smaller than expected from clear atmospheremodels (Deming et al. 2013). The extreme case of anabsence of molecular features would indicate an atmo-sphere completely dominated by opaque clouds. Us-ing 2-channel near-infrared photometry with NICMOSduring transits of HD 189733b, Sing et al. (2009) foundresults consistent with Rayleigh scattering from haze.No water vapor absorption was detected. Similarly,Gibson et al. (2012) suggested that haze may dominateits near-infrared transmission spectrum. Higher lay-ers of the atmosphere of HD 189733b have also beenprobed. High altitude haze was inferred by a UVtransmission spectrum dominated by Rayleigh scatter-ing (Lecavelier Des Etangs et al. 2008; Sing et al. 2011;Huitson et al. 2012), which would explain the narrow-ness of alkali features and the increased planetary ra-dius in the UV compared to the infrared. After re-analyzing HD 189733 data over a wide wavelength range,Pont et al. (2013) proposed a general explanation for theatmosphere of HD 189733b at its terminator: Rayleighscattering by clouds dominates in the UV, and settlingdust and a cloud deck yield a featureless spectrum inthe near infrared and in the infrared. A marginal de-tection of a decreasing geometric albedo in the visi-ble may also indicate optically thick reflective cloudson the day-side of the planet, although with a lowalbedo overall (Evans et al. 2013). Beyond the caseof hot Jupiters, examples of featureless transmissionspectra are GJ1214b (Berta et al. 2012; Kreidberg et al.2014), GJ436b (Knutson et al. 2014a), and HD 97658b(Knutson et al. 2014b), although for low-mass planets,flat transmission spectra could alternatively be explainedby high mean molecular weight atmospheres. In allcases, the inferences of hazes/clouds are motivated by thenon-detection of spectral features of expected molecules,and a blueward rise in the optical part of the transmis-sion spectrum. However, observational constraints onthe chemical composition of the haze/cloud material arenon-existent. In this context, measuring the transit andeclipse spectra of HD 189733b in the near infrared, wheremolecular features are expected, is crucial. The observa-tions presented in this paper and in McCullough et al.(2014) bring new constraints on the cloud hypothesis forthe atmosphere of HD 189733b.For tidally locked exoplanets such as HD 189733b,one might expect different atmospheres on the day side,the nightside, and along the terminator. However, verystrong equatorial winds would tend to homogenize bothsides by increasing the heat redistribution and mix-ing the composition. In the case of HD 189733b, theheat redistribution was found to be relatively small(Grillmair et al. 2008; Knutson et al. 2012). Measuringthe transit and eclipse spectra of HD 189733b in thesame wavelength range with the high precision of WFC3spatial scans offers a unique opportunity to consistentlyprobe these two regions of the atmosphere.We present the observations in section 2, the data re-duction in section 3, the resulting spectrum in section 4.We perform a whitelight analysis in section 5. Resultsare then discussed in section 6, followed by a summaryin section 7. OBSERVATIONS
We used
HST
WFC3 with the newly implemented spa-tial scanning mode, developed in part to enable obser-vations such as these (McCullough & MacKenty 2012).In this mode, a controlled scan is applied to the tele-scope during the exposure in a direction perpendicularto the wavelength dispersion direction (Figure 1). Thistechnique is particularly efficient for bright stars such asHD 189733 (see McCullough et al. 2014, for more de-tails). Several programs have already benefited fromthis technique (Deming et al. 2013; Wakeford et al. 2013;Kreidberg et al. 2014; Knutson et al. 2014a,b).
Figure 1.
Example of image acquired with WFC3 using thespatial scanning mode. The wavelengths are spread horizontallywhereas the spatial scan occurs vertically. The intensity is inlog(ADU) s − as indicated at the top. The first order spectrum ofHD 189733A, used is in this work, is located in the upper left quad-rant. Also visible are the second order spectrum of HD 189733Ain the upper right quadrant, the first and second order spectrumof the companion star HD 189733B below and slightly overlappingwith those of HD 189733A, and the spectra of fainter stars. One eclipse of HD 189733b was observed on June 24,2013. The observations are divided into five
HST orbits,the planetary eclipse occurring during the fourth orbit.In total, 159 exposures of 5.97 s each were acquired, cor-responding to 32 exposures per orbit (except for the firstorbit in which the first image is a direct image). Weused the G141 grism covering a spectral range from 1.1to 1.7 µ m, the 512 ×
512 pixel subarray, and a scan rateof 2 arcsec s − . The spectral trace is spread over 111detector rows by the spatial scan and 150 columns bythe dispersive element. The resulting spectral resolutionis R = λ/ ∆ λ = 130 (Dressel 2014). The detector is readout in MULTIACCUM mode with the RAPID samplesequence and NSAMP = 7 (each image is composed of7 successive readouts after the initial read). The firstand second order spectra of HD 189733 are visible inthe images, as well as the spectra of the companion starHD 189733B. The latter slightly overlap with the former,but our data reduction method eliminates this overlap.he eclipse of HD 189733b 3 Table 1
Summary of
HST
WFC3 observations
HST
Program (P.I.) 12881 (McCullough)Number of
HST orbits 5Number of scans per orbit 16 Forward, 16 ReverseDuration of scan (s) 5.97Scan rate (arcsec s − )[pixels s − ] (2.00)[16.5]Peak signal on detector (e − px − ) 4 . × Grism ( λ ) G141 (1 . − . µ m)Detector subarray size (pixels) 512x512Sample sequence RAPIDSamples per scan 8Start of first scan (HJD) 2456467.855665Corresponding planetary orbital phase 0.40717Start of last scan (HJD) 2456468.144658Corresponding planetary orbital phase 0.53743Notes. Forward and reverse scans were interleaved. The planetaryorbital phase is defined to be 0.5 at mid-eclipse. The spatial scans occur alternatively in two directions,which we will designate as “forward” and “reverse” inthis paper. The observation parameters are summarizedin Table 1. DATA REDUCTION
Spectrum extraction
We intentionally keep the data reduction as simple aspossible in order to emphasis the data quality obtainedwith
HST
WFC3 in spatial scanning mode. This is instark contrast with previous similar observations such asthose obtained with
HST
NICMOS ( e.g.
Crouzet et al.2012), in which complex data reduction methods werenecessary. The improvement of the data quality of WFC3compared to NICMOS is already evident after a first lookat the images: variations of the spectrum position on thedetector are nearly invisible to the eye and the individualpixel values are much more stable.We start with the Intermediate MultiAccum images(ima.fits), which are corrected for dark current, non-linearity, and other calibrations by the CALWFC3pipeline at STScI. We separate the images accordingto the scanning direction, resulting in two sets of data.Then, for each image, we build seven differential imagesfrom the eight readouts. Indeed, the WFC3 pixels areread non-destructively while the signal is building up andwhile the vertical scanning occurs, and eight intermedi-ate readouts are recorded during each exposure. The firstreadout occurs quickly after the exposure starts and isknown as the zeroth read, whereas the last readout cor-responds to the full scan image. The signal recorded be-tween two consecutive readouts is retrieved by subtract-ing them, i.e. D i = R i − R i − , where R i are the readouts, D i are the “differential images”, and i = 1 , ...,
7. Fig-ure 2 shows an example of a differential image. Its sizealong the scan direction is now reduced as it correspondsto 1/7th of the full scan. Considering all the observa-tions, this leads to seven sets of differential images foreach scan direction, i.e. fourteen sets in total. Withineach set, the spectral trace is always at the same posi-tion on the detector. We build a 1-D spectrum for eachset independently. This method eliminates the overlapof the companion star with the target, minimizes the skybackground contribution, and allows consistency checksbetween the spectra obtained from the fourteen sets.We calibrate each differential image ( D i ) by the F139Mflat-field. We consider the pixel response is not wave- Figure 2.
Example of a differential image obtained after sub-traction of two consecutive readouts of the image shown in Figure1. The extension of the various spectral traces along the scanningdirection is now reduced, as the full scan is now split between 7differential images (see text). The intensity is in log(ADU) s − as indicated at the top. The stellar and sky regions are shown ingreen and blue respectively. length dependent at the level required to extract theplanetary spectrum, and we simply divide each imageby this two-dimensional flat-field. Further tests using awavelength dependent flat-field or no flat-field calibrationdo not result in significant differences in the final spec-trum. Furthermore, flat-fielding errors should in princi-ple be largely removed by dividing the in- by the out-of-eclipse spectra.We search for bad pixels and cosmic rays located in thespectral trace region on a row-by-row basis. We computea median-smoothed version of the row using a 10-pixelkernel. We calculate the median of the absolute devia-tion of the row with respect to its smoothed version andidentify pixels deviating by more than 15 times this me-dian. This yields typically 2 to 3 deviant pixels over thespectral region in each differential image, which are infact already apparent by eye. We replace them by thevalue of the median-smoothed row at this location. Wefind good agreement between the bad pixels identified byour method and those flagged by the CALWFC3 pipelineas bad pixels in the data quality arrays, i.e. with a flagvalue of 4: “Bad detector pixel” (Rajan 2010) in exten-sion 3 of the ima.fits files.As commonly done with HST , we do not use the firstorbit in which the telescope is known to settle to its newthermal environment resulting in an unstable behavior.As a result, the “hook” observed in the whitelight curveat the beginning of each orbit, a WFC3 feature alreadyreported by Berta et al. (2012) and Deming et al. (2013),is much stronger in the first orbit. All images from thefour remaining orbits are used in the analysis.We define rectangular regions containing the stellarsignal and the sky background. A stellar region is de-fined for each set of differential images. The regions are154 columns wide and 20 rows tall for forward scans Crouzet et al.(22 rows tall for reverse scans). The reverse directionyields slightly taller scans due to the longer exposuretime (McCullough & MacKenty 2012) . The sky regionis common to all images and is 96 columns wide, 166rows tall, and located in the upper left corner.Spectra are built in a very straightforward manner.The sky background is calculated on the sky region usingthe SKY procedure in IDL and subtracted to the image.The sky values range from 0 . × − to 1 . × − relatively to the maximum flux per pixel in the spectraltrace of HD 189733. Then, we sum each column in thestellar region individually; this creates a one-dimensionalspectrum for each differential image D i (figure 3). Foreach set of differential images, we average these spectrafor each orbit, and average the three out-of-eclipse orbitstogether (only one orbit is in-eclipse). Then we dividethe in-eclipse spectrum by the out-of-eclipse spectrum toextract the planetary spectrum. We build 14 planetaryspectra from the 14 sets of differential images, which weaverage together. For all these calculations, we use themost simple possible average (function MEAN in IDL),with e.g. no calls for outliers nor medians . Figure 3.
Example of spectrum obtained from a differential image( D i ). Wavelength calibration
To calibrate the detector columns in wavelength, wematch our stellar spectrum to a known spectrum of sim-ilar spectral type. In keeping with our objective of sim-plicity, the G141 grism wavelength dispersion can beapproximated by a linear function. First, we calibratethe out-of-eclipse spectrum by the known G141 grismresponse. This requires an initial estimate of the wave-length solution, obtained by matching the sharp edgesof the grism response to that of our spectrum. Sec-ond, we compare a spectrum of the K1V star 107 Psc (Rayner et al. 2009) to our data. We remove the lowfrequency variations by subtracting a smoothed versionfrom each of these spectra. Finally, we match themin the wavelength range 1 . − . µ m using their com-mon spectral lines (Figure 4). We derive the relation However, recall that we identified and replaced outlier pixelvalues in the 2-D images. http://irtfweb.ifa.hawaii.edu/ ∼ spex/IRTF Spectral Library/index.html λ ( x ) = a × ( x − x ) + b , where x is the detector columnindex, x = 77, a = 0 . µ m px − , and b = 1 . µ m. Figure 4.
Wavelength calibration. Black: library spectrum ofthe K1V star 107 Psc. Red: stellar spectrum extracted from theWFC3 data. The large spikes in the WFC3 data at the ends of thebandpass are due to the low response of the grism at the edges.
Spectrum shifts
We investigate the variable shifts in the x direction(along the rows), which may be present in WFC3 spatialscan data (Deming et al. 2013). To calculate these, were-create full scan images by summing the calibrated for-ward differences, and build a template spectrum for eachscan direction from the 40th such image. We then cal-culate the shift of each row with respect to the templatespectrum. The row is shifted by steps of 1/100 pixel upto ± Uncertainties
We estimate the uncertainties of the exoplanetary spec-trum from the standard deviation of the lightcurve ofeach spectral channel (a detector column or group ofbinned columns corresponding to a given wavelength).These lightcurves are first corrected from whitelight fluxvariations. To this end, we extract a spectrum foreach calibrated full scan image excluding 20 columnsat each edge of the spectral trace, and scale its ampli-tude to that of the template. The best scaling factor isfound using the downhill simplex minimization procedure(AMOEBA). We build the whitelight corrected channelhe eclipse of HD 189733b 5lightcurves from these amplitude corrected spectra. Theuncertainty σ c,d for a given channel c and a given scandirection d is: σ c,d = s σ out,d m out,d N out,d + σ in,d m in,d N in,d (1)where σ in,d , σ out,d are the standard deviations of the in-eclipse and out-of-eclipse parts of the channel c lightcurverespectively, m in,d , m out,d their respective mean, and N in,d , N out,d their respective number of points ( N in,d =16, N out,d = 48). We find a median uncertainty of 160ppm and 146 ppm for the forward ( d = f ) and reverse( d = r ) scan directions respectively, for an expected min-imum Poisson noise of 162 and 153 ppm. Our final spec-trum is the average of the forward and reverse scan di-rection spectra, thus our final uncertainties are: σ c = 12 q σ c,f + σ c,r (2)The final median uncertainty is 110 ppm compared to aPoisson noise estimate of 111 ppm. Because the planetpasses behind the star during the eclipse, the effect ofstar spots on the emission spectrum can be neglected. RESULTS
The planetary emission spectrum is binned over 4columns using a boxcar average, resulting in indepen-dent 0.02 µ m bins. This yields a high precision Pois-son noise limited spectrum: the median uncertainty is57 ppm. This spectrum is reported in Table 2 columns1 to 4, with an average depth arbitrarily set to zero, andis shown in Figure 5. The absolute eclipse depth is cal-culated from the whitelight analysis in Section 5. Thespectra obtained from the 2 scan directions and for the14 sets of differential images are consistent within theiruncertainty and do not show any systematic differences.An independent analysis of these data was conductedusing the method prescribed in Deming et al. (2013).Both planetary spectra are largely consistent, well withintheir 1- σ uncertainties. This second spectrum as wellas the difference between both are reported in Table 2,columns 5 to 7. The smaller uncertainties of this secondspectrum are due to the convolution with a Gaussian ker-nel of Full Width Half Maximum of 4 columns used tobalance the effect of under-sampling (see Deming et al.2013), which is equivalent to an effective binning over ∼ WHITELIGHT ANALYSIS
We build a whitelight curve by summing the flux overthe stellar region and subtracting the sky background,using the same stellar and sky regions as for the spec-tral analysis. The fluxes measured at all 7 positionsare summed together to recover the flux collected dur-ing a full scan. We build a lightcurve for each scan di-rection and eliminate the first
HST orbit. The result-ing lightcurves are reported in Table 3, and are shownin Figure 6 as a function of time and of
HST orbitalphase. From the second to the fifth orbit, observationswithin each orbit are arranged similarly with respect tothe
HST orbital phase. Preparing the observations in
Figure 5.
Day-side emission spectrum of HD 189733b obtainedwith WFC3 (blue), and theoretical model spectra for a clear plan-etary atmosphere of solar composition (cyan), and close to solarcomposition (red), with their averages at the WFC3 wavelengthbins (filled circles). The best-fit blackbody spectrum for the WFC3and
Spitzer data together (see Section 6), at 1295 K, is also shown(brown). The dashed lines are blackbody spectra at 1100 K (bot-tom) and 1500 K (top). The vertical axis is the ratio of planetaryto stellar flux. a repeatable manner is helpful to correct for systematiceffects such as the “hook”, as described next. A goodpractice is to use a fixed cadence with no interruptionsexcept the Earth occultation, if practical. In our case, abuffer dump interrupts each orbit, with 8 pairs of scansbefore and 8 pairs after the dump. In these lightcurves,three main features appear. First, the white-light flux islarger by 11% in the reverse scan direction with respectto the forward scan direction. This is consistent withthe longer exposure time in the reverse scan direction,which collects more photons (McCullough & MacKenty2012). As a result, we analyze each scan direction sep-arately. Second, the flux ramps upward by ∼ ∼ Figure 6.
Whitelight curve as a function of time (top) and of
HST orbital phase (bottom) for the forward (left) and reverse (right)scan direction for the 2nd (magenta), 3rd (green), 4th (blue), and5th (red) orbit. The flux is 11% greater in the reverse scan directionthan in the forward scan direction, increases by ∼ ∼ To derive the eclipse depth, we model the flux’s de-crease in time first using a second order polynomial. We Crouzet et al.
Table 2
Day-side emission spectrum of HD 189733b λ ∆ R p /R s σ Column ∆ R p /R s σ ∆∆( µ m) (ppm) (ppm) (ppm) (ppm) (ppm)1.1279 -21 81 98.5 -96 47 751.1467 -127 61 102.5 -18 50 -1091.1655 36 70 106.5 28 45 81.1843 -6 47 110.5 -3 44 -31.2031 -15 68 114.5 -7 43 -91.2218 -65 55 118.5 -45 50 -201.2406 -23 52 122.5 -32 42 91.2594 2 45 126.5 3 42 -11.2782 -66 50 130.5 -16 42 -511.2969 6 53 134.5 53 41 -471.3157 6 57 138.5 31 41 -241.3345 19 58 142.5 12 41 71.3533 -48 47 146.5 -21 41 -271.3721 -59 61 150.5 -27 41 -321.3908 -90 43 154.5 -40 42 -501.4096 34 51 158.5 8 42 261.4284 16 55 162.5 -2 42 181.4472 4 53 166.5 -29 42 331.4660 -38 68 170.5 -64 43 261.4848 -19 50 174.5 9 43 -291.5035 161 61 178.5 132 43 291.5223 129 62 182.5 115 43 141.5411 23 53 186.5 47 44 -231.5599 -6 50 190.5 3 45 -81.5786 77 60 194.5 39 45 381.5974 85 58 198.5 -16 46 1011.6162 -30 74 202.5 -64 46 351.6350 14 74 206.5 – – –Notes. Units are as indicated; parts per million is abbreviated ppm. The tabulated uncertainties apply to the differential eclipse depths;an additional uncertainty applies to the overall depth (see text). The first three columns refer to the analysis of N. C.; columns 5 and 6refer to the analysis of D. D.; the last column contains the difference of the differential spectra, column 2 minus column 5. Table 3
Whitelight curveEXPSTART MJD EXPSTART HJD Orbit Scan Photo-electrons Normalized flux56467.413080 56467.916605 2 1 391466756 0.9992756467.413739 56467.917265 2 -1 435391212 0.9998556467.414399 56467.917924 2 1 391697800 0.9998656467.415059 56467.918584 2 -1 435546624 1.0002056467.415718 56467.919244 2 1 391859136 1.0002756467.416378 56467.919904 2 -1 435641776 1.0004256467.417038 56467.920563 2 1 391938520 1.0004756467.417698 56467.921223 2 -1 435661104 1.0004756467.418357 56467.921883 2 1 391973876 1.0005656467.419017 56467.922543 2 -1 435716676 1.00059Notes. The printed table is a truncated version of the electronic table, to illustrate the format. Columns, left to right, are modified Juliandate of the start of the exposure, the associated heliocentric Julian date, the
HST orbit in the visit, the scan direction (1 = forward; -1 =reverse), the total number of photoelectrons from HD 189733, and the associated normalized flux before detrending. do not directly model the hook. Instead, we fit the samepolynomial function to all the out-of-eclipse points, withthe zero flux level as a free parameter for each set ofpoints located at the same position in the orbital se-quence (set 1: 1st point of orbits 2, 3, and 5; set 2:2nd point of orbits 2, 3, and 5; etc..., see Figure 7). Thein-eclipse points are excluded from the fit (orbit 4). Thisfit is then subtracted to the data points. The residualsfor both scan directions are normalized separately andput together. These residuals are close to the Poissonnoise. We calculate the eclipse depth δ p as the mean ofthe in-eclipse residuals, and its 1- σ uncertainty as: σ = s σ in N in + σ out N out (3)where σ in , σ out is the standard deviation of the in-eclipseand out-of-eclipse residuals, and N in , N out is the numberof in-eclipse and out-of-eclipse points ( N in = 32, N out =96). We find δ p = 68 ±
12 ppm.Because the choice of a second order polynomial is ar-bitrary, we perform the same analysis using other func-tions: an exponential function from the second to thefifth orbit, a linear function from the second to the fifthhe eclipse of HD 189733b 7orbit, and a linear function from the third to the fifth or-bit (orbits surrounding the eclipse). Inspection of resid-uals show that the linear function from the second to thefifth orbit is clearly a poor fit; we exclude this function .For the exponential fit, and the linear fit from the thirdto the fifth orbit, we find an eclipse depth δ e = 79 ± δ l = 122 ±
12 ppm respectively. In their analysisof WFC3 data of WASP-12b obtained in staring mode,Stevenson et al. (2014) also noticed that different rampmodels yield different values for the transit depth (notethat in our approach the ramp model is implicit).For each model, we estimate the error on the eclipsedepth caused by stellar variability, thought to originatemainly from star spots. We model this variability bya sinusoidal function of period 11.95 d (Henry & Winn2008), and rescale the amplitude of 1.5% in the b+y band(Knutson et al. 2012) to our wavelength of 1.4 µ m, us-ing a blackbody emission function with a stellar effectivetemperature of 4980 K and a spot temperature of 4250 K(Pont et al. 2008). We find an amplitude of 0.8% at 1.4 µ m. This stellar variation is added to the data at allpossible phases of the sinusoidal function prior to per-forming the function fit. The maximum difference on theeclipse depth with or without this variability is negligi-ble for the second order polynomial and the exponentialfits (0.1 ppm), and is 5 ppm for the linear function fitfrom the third to the fifth orbit, with a standard devi-ation of 1.7 ppm. We quadratically add this standarddeviation to the 1- σ uncertainty for this latter function(which result in a change of a fraction of ppm).The actual function being unknown, we include the fullrange spanned by the three models in our final estimateof the eclipse depth δ . This yields δ = 96 ±
39 ppm. Wederive an approximate brightness temperature T b for theplanet by: δ = R p B ( T b ) R ⋆ B ( T eff,⋆ ) (4)where R p and R ⋆ are the planet and stellar radii respec-tively, and B ( T b ) and B ( T eff,⋆ ) are the spectral radianceof blackbodies of temperature T b and T eff,⋆ respectively.For simplicity, we approximate the stellar spectrum bya blackbody emission of temperature T eff,⋆ = 4980 K(a more accurate model is used in Section 6). Wefind a brightness temperature T b = 1419 +71 − K in thewavelength range 1 . − . µ m for the day-side ofHD 189733b. DISCUSSION
We model the dayside atmosphere of HD 189733busing the atmospheric modeling and retrieval methodof Madhusudhan & Seager (2009) and Madhusudhan(2012). The emission spectrum is calculated by solvingfor 1D line-by-line radiative transfer in a plane paral-lel atmosphere, with constraints of local thermodynamicequilibrium (LTE), hydrostatic equilibrium and globalenergy balance. The model has twelve free parametersincluding a parametric temperature profile and molecu-lar abundances parameterized as uniform mixing ratios A similar decrease is also present in the transit data(McCullough et al. 2014) and in that case is well represented bya linear fit from the second to the fifth orbit.
Figure 7.
Top: white-light curve as a function of time for theforward (left) and reverse (right) scan direction for the 2nd (ma-genta), 3rd (green), 4th (blue), and 5th (red) orbit. The verticaldashed lines indicate the 1st, 2nd, 3rd, and 4th contact of the plan-etary eclipse, from left to right. We model the decrease in time byfitting the same function, here a second order polynomial, to eachset of points taken at the same
HST orbital phase using the out-of-eclipse orbits, with the flux offset as a free parameter to accountfor the hook (plain lines). Bottom: residuals after fitting a secondorder polynomial. in the atmosphere. The model includes opacity contri-butions due to all the dominant molecules expected inH -rich hot Jupiter atmospheres in oxygen-rich as wellas carbon-rich regimes (see e.g. Madhusudhan 2012),namely, line opacity due to H O, CO, CO , CH , HCN,and C H , and H -H collision induced opacity. As forthe temperature structure, the parametric temperatureprofile can model dayside temperature profiles with andwithout thermal inversions. Given a dataset, we ex-plore a wide range of inversion and non-inversion mod-els as well as varied molecular mixing ratios in searchof regions in parameter space that best match the data.In the present case, we explore models that can simul-taneously explain our WFC3 dataset along with previ-ously published Spitzer observations in six photometricbandpasses (Charbonneau et al. 2008; Agol et al. 2010;Knutson et al. 2012).Considering our present WFC3 observations alone(Figure 5), we find nominal evidence for water absorp-tion in the WFC3 bandpass (1 . − . µ m). Figure 5shows model spectra that provide a good match to thedata. In the temperature regime of HD 189733b, asshown by the P - T profile in Figure 8, several spectro-scopically strong molecules are expected to be prevalentin the atmosphere for a solar abundance composition,e.g. H O, CO, CH , and CO . However, H O is themost abundant of the molecules and dominates the ab-sorption in the WFC3 bandpass, followed by a weakercontribution from CH , CO, and CO which are uncon-strained by the WFC3 data. Our data can be explainedby a double-trough water absorption feature in the emis-sion spectrum of a solar composition model atmosphere.On the other hand, given the error bars in our data, afeatureless blackbody spectrum also provides a very goodfit to the data. The best-fit blackbody spectrum, corre-sponding to an isothermal atmosphere at T = 1435 K,has a χ of 24.8 for 26 degrees of freedom. Therefore, ourobservational uncertainties preclude a robust constrainton the water abundance using our WFC3 data alone.Combining our data with previously published Spitzer photometric observations from Charbonneau et al.(2008); Agol et al. (2010); Knutson et al. (2012) rules Crouzet et al.out an isothermal atmosphere or one with a thermal in-version (Figure 8). Even the best-fit blackbody spectrumrepresenting an isothermal atmosphere at T = 1295 Kis unable to explain all the existing data simultaneously.Furthermore, a thermal inversion, i.e. with temperatureincreasing with decreasing pressure, is conclusively ruledout by the data as such a model will predict even higherfluxes in all the Spitzer IRAC bands than observed. Onthe other hand, the sum-total of data can be explainedby a dayside atmosphere with no thermal inversion anda nearly-solar or sub-solar abundance H O mixing ratio.Two representative best-fitting models with nearlysolar-abundance H O mixing ratios are shown in Figure8 with their associated pressure-temperature profiles.The corresponding abundances are reported in Table4, as well as the statistical significance of each modelfit. In this table, the BIC is the Bayesian InformationCriterion, which takes into account the number of freeparameters in evaluating a model fit. It is calculated asfollows: BIC = χ + k ln N , where k is the number offree parameters (12 for the non-isothermal atmospheremodels and 1 for the blackbody), and N is the numberof data points (34, i.e.
28 from WFC3 and 6 from
Spitzer ). The comparison of these models to the data ina χ sense or using the BIC shows that the atmospheremodels with molecular features provide significantlybetter fits than the best blackbody model.A variety of non-isothermal atmosphere models witha range of compositions can also fit the data. The twomodels shown in Figure 8 and Table 4 are examples cho-sen because of their chemical compositions which comeclose to a solar-abundance H O mixing ratio while stillproviding good fits to the data; an exactly solar abun-dance composition would have a H O mixing ratio be-tween 5 × − and 10 − depending on the tempera-ture. More generally, however, the space of best-fittingsolutions prefer manifestly sub-solar H O abundances:the 1- σ range of H O abundance derived from the sum-total of data used is between 1 . × − and 2 . × − ,with a modal value of 7 . × − . Our H O abundancesare generally consistent, at ∼ σ , with other studies inthe past also reporting constraints that are consistentwith a sub-solar H O abundance on the dayside of HD189733b (e.g., Madhusudhan & Seager 2009; Swain et al.2009; Lee et al. 2012; Line et al. 2012, 2014) using pre-vious
Spitzer and
HST data, some of which have sincebeen revised. However, our 1- σ limits suggest lowerH O abundances than previously suggested. Note thatlow altitude, thin, or patchy clouds, and/or haze in theplanet’s atmosphere could also damp spectral features,as discussed below. Since our models are cloud-free, themolecular abundances derived here should be interpretedas lower limits if obscuring clouds are present.The lack of a thermal inversion in the dayside at-mosphere of HD 189733b is consistent with previ-ous studies, both observational and theoretical. Sev-eral studies using past data have suggested the ab-sence of a thermal inversion in HD 189733b ( e.g.
Burrows et al. 2008; Grillmair et al. 2008; Swain et al.2009; Madhusudhan & Seager 2009). Several theoreticaland empirical studies have also predicted that its daysideatmosphere would be unlikely to host a strong thermalinversion. With an equilibrium temperature of ∼ Table 4
Molecular abundances relative to H and statistical significancefor two fitting models and a blackbody model, for thecombination of WFC3 and Spitzer data.Model 1 Model 2 Best-fitblackbodyH O 1 . × − . × − -CO 9 . × − . × − -CH . × − . × − -CO . × − . × − - χ H and HCN are also present, but are negligible. χ arenon-reduced χ . d.f. indicates the number of degrees of freedom. HD 189733b is one of the less irradiated hot Jupiters.Based on the original TiO/VO hypothesis (Hubeny et al.2003; Fortney et al. 2008), cooler hot Jupiters such asHD 189733b would not be hot enough to host gaseousTiO and VO in their upper atmospheres which havebeen proposed as inversion-causing compounds. Evenif alternate inversion-causing compounds were possible,Knutson et al. (2010) suggested that the high chromo-spheric activity of the host star (HD 189733) might dis-sociate them in the planetary atmosphere.The ensemble of data obtained on HD 189733b intransmission and emission from
HST and
Spitzer is yetto be understood completely. Pont et al. (2013) inter-preted the combined transmission spectrum by an at-mosphere dominated by dust and clouds. They alsonoted that several puzzling features arising from the in-terpretation of the
Spitzer phase curves with a clear so-lar composition atmosphere model (Knutson et al. 2012)could be explained by clouds (see Table 8 of Pont et al.2013). However, McCullough et al. (2014) recently re-ported a robust detection of water vapor absorption inthe transmission spectrum of the planet, which rules outa featureless transmission spectrum, as well as an at-mosphere dominated by opaque a clouds at pressures
P > . λ < . µ m) than at longer wave-lengths ( λ > . µ m) was reported by (Evans et al.2013) who suggested an “intermediate cloud scenario”for the day-side of the planet in which clouds are presentand become optically thick at pressures correspondingto the Na absorption wings. Recent theoretical studieshave also suggested the possibility of clouds in the day-side atmospheres of hot Jupiters (e.g. Heng & Demory2013; Parmentier et al. 2013). However, as shown byBarstow et al. (2014), the reported albedo spectrum isinsufficient to conclusively constrain the presence ofclouds in the dayside atmosphere of HD 189733b, as acloud free atmosphere with Rayleigh scattering due toH and non-solar Na abundances could explain the dataequally well. Our detection of molecular features in theWFC3 bandpass places additional constraints on the pos-sibility of clouds on the dayside. With thick clouds onhe eclipse of HD 189733b 9 Figure 8.
Observations and model spectra of day-side thermal emission from HD 189733b. The observations are show in blue circleswith error bars, and include data obtained using WFC3 in the 1.1-1.7 µ m range (this work) and Spitzer photometry at longer wavelengths(Charbonneau et al. 2008; Agol et al. 2010; Knutson et al. 2012). Two theoretical model spectra are shown in the red and cyan curvescorresponding to chemical compositions shown in Table 4 (model 1 and model 2 respectively) and non-inverted temperature profiles shownin the lower-right inset. The brown curve shows a model blackbody spectrum representing an isothermal atmosphere at 1295 K. Theband-pass integrated model points for each model are shown in the same colored circles. The dashed gray lines show blackbody spectra at1100 K (bottom) and 1500 K (top). The vertical axis is the ratio of planetary to stellar flux. The top left inset zooms in on the WFC3spectrum, and the bottom right inset shows the atmospheric pressure-temperature profile for the three models. the dayside, we would expect a blackbody spectrum tofit the observed infrared emission spectrum better than amodel spectrum with molecular features in the observedbandpasses; our retrieved model solutions favor the op-posite. Interestingly, a new analysis of the
Spitzer
IRSdata including data not included in the original spectrumfrom Grillmair et al. (2008) is also inconsistent with ablackbody spectrum (Todorov et al. 2014). In principle,alternate scenarios might be able to explain both the re-ported albedo spectrum and the molecular features weare observing, e.g., a layer of clouds at low altitudes pro-ducing the former and a clear atmosphere at higher al-titudes producing the latter, or Rayleigh scattering dueto H with non-solar Na abundances, which future re-trieval studies using clouds models could investigate (e.g.Barstow et al. 2014; Lee et al. 2014).The transit and eclipse spectra of HD 189733b ob-tained with the high precision of WFC3 in spatial scan-ning mode yield new constraints both on the limb andday-side regions of its atmosphere. These results sup-port a general picture of an atmosphere dominated bywater vapor in the near infrared, albeit with a lower H Oabundance than for a solar composition atmosphere. Ifclouds are present, these abundances would be degener-ate with the amount of cloudiness; however, the presenceand composition of such clouds are not constrained obser-vationally (see Madhusudhan et al. 2014, for a discussionon this low H O abundance). These results point towardsthe need for including the effects of clouds in hot Jupiteratmosphere models, although still poorly understood dueto their extreme complexity. Finally, these transit andeclipse spectra provide keys to further interpretation us- ing full 3-D atmosphere models such as developed by e.g.
Showman et al. (2009). Extensive characterizationof other hot gaseous giant planets would also improve ourunderstanding of their atmospheres. The TESS mission(Ricker et al. 2014) should detect new hot Jupiters andhot Neptunes around bright stars, providing ideal targetsfor such characterization. SUMMARY
We observed HD 189733b during a planetary eclipsewith
HST
WFC3 and extracted the emission spectrumof the planet in the wavelength range 1 . − . µ m. Usinga straightforward data reduction method of the spatiallyscanned spectra, the derived spectrum is Poisson noiselimited. A white-light analysis including a correction forWFC3 instrumental systematic effects yields the abso-lute eclipse depth in this wavelength range. The result-ing spectrum shows marginal evidence for water vaporabsorption, but can also be well explained by a black-body spectrum. However, the combination of our WFC3data with previous data from Spitzer reinforces a gen-eral picture of a day-side atmosphere in thermochemicalequilibrium with no thermal inversion, and dominatedby water vapor features in the near infrared.The authors gratefully acknowledge everyone who hascontributed to the
Hubble Space Telescope and theWFC3, and particularly those responsible for implement-ing the spatial scanning, which was critical to these ob-servations. We thank in particular John MacKenty andMerle Reinhart. We acknowledge conversations withSuzanne Hawley, Leslie Hebb, Veselin Kostov, Rachel0 Crouzet et al.Osten, and Neill Reid. This research used NASA’s As-trophysics Data System Bibliographic Services, the SIM-BAD database operated at CDS, Strasbourg, France, andwas funded in part by
HST grant GO-12881 and Originsof Solar Systems grant NNX10AG30G.grant GO-12881 and Originsof Solar Systems grant NNX10AG30G.