Exoplanet atmospheres with GIANO. I. Water in the transmission spectrum of HD 189733b
M. Brogi, P. Giacobbe, G. Guilluy, R. J. de Kok, A. Sozzetti, L. Mancini, A. S. Bonomo
AAstronomy & Astrophysics manuscript no. 32189_arxiv_v2 c (cid:13)
ESO 2018March 1, 2018
Exoplanet atmospheres with GIANO
I. Water in the transmission spectrum of HD 189 733 b
M. Brogi , , P. Giacobbe , G. Guilluy , , R. J. de Kok , , , A. Sozzetti , L. Mancini , , , and A. S. Bonomo Department of Physics, University of Warwick, Coventry CV4 7AL, UKe-mail: [email protected] Center for Astrophysics and Space Astronomy (CASA), University of Colorado Boulder, Boulder, CO 80309, USA INAF-Osservatorio Astrofisico di Torino, via Osservatorio 20, 10025, Pino Torinese, Italy Dipartimento di Fisica, Università di Torino, via P. Giuria 1, I-10125 Torino, Italy SRON Netherlands Institute for Space Research, Sorbonnelaan 2, 3584 CA Utrecht, The Netherlands Leiden Observatory, Leiden University, Postbus 9513, 2300 RA, Leiden, The Netherlands Utrecht University, Department of Physical Geography, PO Box 80115, 3508 TC, Utrecht, The Netherlands Dipartimento di Fisica, Università di Roma Tor Vergata, via della Ricerca Scientifica 1, 00133 Roma, Italy Max Planck Institute for Astronomy, Königstuhl 17, 69117 Heidelberg, Germany
ABSTRACT
Context.
High-resolution spectroscopy ( R ≥
20 000) at near-infrared wavelengths can be used to investigate the composition, structure,and circulation patterns of exoplanet atmospheres. However, up to now it has been the exclusive dominion of the biggest telescopefacilities on the ground, due to the large amount of photons necessary to measure a signal in high-dispersion spectra.
Aims.
Here we show that spectrographs with a novel design - in particular a large spectral range - can open exoplanet characterisationto smaller telescope facilities too. We aim to demonstrate the concept on a series of spectra of the exoplanet HD 189 733 b taken atthe Telescopio Nazionale Galileo with the near-infrared spectrograph GIANO during two transits of the planet.
Methods.
In contrast to absorption in the Earth’s atmosphere (telluric absorption), the planet transmission spectrum shifts in radialvelocity during transit due to the changing orbital motion of the planet. This allows us to remove the telluric spectrum while preservingthe signal of the exoplanet. The latter is then extracted by cross-correlating the residual spectra with template models of the planetatmosphere computed through line-by-line radiative transfer calculations, and containing molecular absorption lines from water andmethane.
Results.
By combining the signal of many thousands of planet molecular lines, we confirm the presence of water vapour in theatmosphere of HD 189 733 b at the 5.5- σ level. This signal was measured only in the first of the two observing nights. By injectingand retrieving artificial signals, we show that the non-detection on the second night is likely due to an inferior quality of the data.The measured strength of the planet transmission spectrum is fully consistent with past CRIRES observations at the VLT, excludinga strong variability in the depth of molecular absorption lines. Key words.
Planets and satellites: atmospheres – Planets and satellites: individual (HD 189733 b) – Techniques: spectroscopic
1. Introduction
In the past few years, spectroscopy at resolving powers above20,000 has become a valuable tool to unravel the composition,structure, and dynamics of exoplanet atmospheres. The key as-pect of this technique is its ability to resolve molecular bandsinto the individual lines. This enables line-by-line comparisonof model spectra to observations, typically obtained by cross-correlation. Hundreds or thousands of molecular lines are co-added into a single cross-correlation function (CCF), enhanc-ing their faint signal by a factor of approximately √ N lines . Highspectral resolution also allows us to resolve the change in the or-bital radial velocity of the planet. For exoplanets orbiting theirparent star in just a few days, these changes are in the orderof tens of km s − across a few hours. Thanks to its unique andhighly variable Doppler signature, the spectrum of the planet canbe disentangled from the main contaminants (i.e. the spectrumof the Earth’s atmosphere and the stellar spectrum), which areinstead stationary or quasi-stationary. Thus, detections at highspectral resolution are particularly robust against contamination from spurious sources. Furthermore, they are self-calibrated,eliminating the necessity of a reference star of similar brightnessand spectral type in the same field of view.After more than a decade of inconclusive attempts at us-ing high-resolution spectroscopy for investigating exoplanet at-mospheres (Collier Cameron et al. 1999; Charbonneau et al.1999; Wiedemann et al. 2001; Leigh et al. 2003; Lucas et al.2009; Barnes et al. 2010; Rodler et al. 2010), Snellen et al.(2010) devised a successful observational strategy and detectedCO absorption lines in the transmission spectrum of exoplanetHD 209458 b via near-infrared (NIR) spectroscopy with theCryogenic InfraRed Echelle Spectrograph (CRIRES, Kaeuflet al. 2004) at the ESO Very Large Telescope (VLT). Two yearslater, Brogi et al. (2012) and Rodler et al. (2012) demonstratedthat the same measurements could be extended to non-transitingplanets and used to estimate their orbital inclination and hencetrue mass. While CRIRES began to routinely detect CO andH O in the atmospheres of transiting (de Kok et al. 2013; Birkbyet al. 2013) and non-transiting (Brogi et al. 2013, 2014; Birkbyet al. 2017) planets, Lockwood et al. (2014) showed that simi-
Article number, page 1 of 10 a r X i v : . [ a s t r o - ph . E P ] F e b & A proofs: manuscript no. 32189_arxiv_v2 lar measurements were possible even at a lower resolving powerof 25 000, obtained with NIRSPEC at Keck. They designed analternative analysis where the planet’s Doppler shift no longerneeds to change during an observing night (Piskorz et al. 2016,2017), potentially enabling the study of longer-period bodies.Besides detecting molecular species, determining their relativeabundances, and constraining the overall thermal structure ofthe atmosphere, high-resolution spectroscopy is also capable ofmeasuring planet rotation and even global winds (Schwarz et al.2016; Brogi et al. 2016) thanks to the broadening and asymmetryin the cross-correlation function.So far the major downside of the method has been the enor-mous request in terms of signal-to-noise. Only the biggest tele-scopes on the ground, namely VLT and Keck, have succeeded inproviding an adequate collective area for securing solid detec-tions in the near-infrared, and only for the brightest known ex-oplanet systems (stellar K magnitudes below 7.5). This has lim-ited the sample to a handful of hot Jupiters orbiting very brightstars with spectral types ranging from late F to early K. This lim-itation is, however, tied to the technology of near-infrared spec-trographs that were available in the past, and can be overcomeby the recent development of instruments with bigger through-put and / or larger spectral range. The high resolution spectro-graph GIANO (Oliva et al. 2006; Origlia et al. 2014), mounted atthe Nasmyth-A focus of the 3.6-m Telescopio Nazionale Galileo(TNG), belongs in the latter category. With its nearly contiguouscoverage of the entire Y, J, H, and K bands, it provides a 21-foldincrease in spectral range compared to CRIRES at the VLT.In this work we have successfully tested the capabilities ofGIANO for characterising exoplanet atmospheres. We analysedspectra of HD 189 733 b (Bouchy et al. 2005) taken in andaround transit and extract the signal of the transmission spectrumof the planet. HD 189 733 b is considered one of the best-studiedexoplanets to date, due to the brightness of its parent star (K =
2. The GIANO spectra and their calibration
Two transits of HD 189 733 b were observed on July 11 andJuly 30, 2015, with GIANO at the TNG. The data consist in asequence of nodded observations following an ABAB pattern,each exposed for 60 s. At each nodding position, a fibre of 1 (cid:48)(cid:48) in diameter (on-sky) feeds the spectrograph. A second fibre, lo-cated at the other nodding position (3 (cid:48)(cid:48) away) is instead lookingat the sky, providing an accurate reference for subtracting thethermal background and telluric emission lines. Each time thetelescope nods, the fibres swap, i.e. the object fibre becomes thesky fibre and vice-versa.The fibres were re-imaged onto a 2 × slicer, resulting in twospectral tracks per nodding position, or four tracks per order. Theentire NIR spectrum (0.95–2.45 µ m) was cross-dispersed at a re-solving power of R =
50 000 and orders 32 to 80 are imaged ona 2k ×
2k HAWAII-2 detector. Contiguous wavelength cover-age is ensured for orders higher than 45 (bluer than ∼ . µ m),whereas for lower orders (redder wavelengths) the coverage isincomplete, down to 75% for order 32. In this work, we haverenumbered the orders so that order 0 is the reddest and order 46the bluest. We skipped the extraction of the two bluest orders inthe cross-dispersed spectrum for lack of photons.Both observations consist of spectra taken before, during,and after the transit of the planet (Figure 1). We collected a totalof 90 and 110 spectra in the night of July 11 and 30, respectively.We measured a signal-to-noise ratio (S / N) of 50-60 per spectrum A i r m a ss Fig. 1.
Orbital phase of exoplanet HD 189 733 b and corresponding air-mass during the two nights of observations analysed in this work. Plusand star symbols denote out-of-transit and in-transit data, respectively. per pixel averaged across the entire spectral range. Peak valuesof S / N = Each science spectrum is processed with our implementationof the IDL algorithm fixpix with 10 iterations to correct forbad pixels. Di ff erence images are created by subtracting each B-image (i.e. the spectrum taken in nodding position B) from theA-image, after dividing each image through the master flat field.This subtraction removes the thermal background and sky emis-sion lines.Di ff erently from common practice, we did not merge the Aand B spectra into a combined AB spectrum. Conversely, weextracted the spectra separately at the A and B positions to in-crease the time resolution of our observations. This choice alsoovercomes the necessity of measuring the shift between the dis-persion solution in A and B spectra at this stage. This step wouldbe redundant since we re-align the spectra to the telluric refer-ence frame in Section 2.2.We used the master flat-field to identify the apertures corre-sponding to each order, each nodding position, and each slicedspectral trace. On the (A–B) flat-fielded science images, we re-fined the position of the apertures by fitting the spectral trailswith four Gaussian profiles per order simultaneously (two slicedimages and two nodding positions) at 32 equally-spaced posi-tions along the dispersion direction. These sampled values werethen fitted with a second-order polynomial to the full 2048 pix-els of the dispersion direction and for each pixel the spectraltraces are rectified in the spatial direction via spline interpo-lation. Lastly, the extraction of the one-dimensional spectra isperformed by optimal extraction (Horne 1986) on the rectifiedimages. B spectra need to be multiplied by − / N estimates, see Section 5. Article number, page 2 of 10. Brogi et al.: Exoplanet atmospheres with GIANO R e l a t i v e p o s i t i o n o f A a n d B s p e c t r a ( p i x e l s ) Night 1B positionsA positions D r i f t r e l a t i v e t o a v e r a g e p o s i t i o n ( p i x e l s ) Temporal drifts are limited to < 0.1 pixels R e l a t i v e p o s i t i o n o f A a n d B s p e c t r a ( p i x e l s ) Night 2B positionsA positions D r i f t r e l a t i v e t o a v e r a g e p o s i t i o n ( p i x e l s ) Temporal drifts are > 0.1 pixels
Fig. 2.
Stability of the GIANO spectrograph. The dispersion solution of the spectra taken at A and B position di ff ers as a function of the spectralorder considered (left panel, averaged over the observing night). In addition, we measure temporal variations in the global dispersion solution(right panel). Their amplitude di ff ers between the first and the second night of observations (top-right and bottom-right panels, respectively).These measurements suggest that although A and B spectra move coherently (i.e. their relative shift stays the same), their absolute position isvariable on a timescale of minutes. Furthermore, the overall stability of GIANO di ff ers between observing nights. The high spectral resolution of GIANO allows us to resolvethe orbital motion of the planet HD 189 733 b during transit.Its radial velocity indeed changes by approximately 32 km s − from ingress to egress. In contrast, the absorption spectrumof the Earth’s atmosphere (telluric spectrum) is stationary inwavelength. We intend to exploit this intrinsic di ff erence in theDoppler signature of the planetary and telluric signals to disen-tangle the former from the latter. As part of this strategy, we alignall the observed spectra to the telluric reference frame.We measured the position of a set of telluric lines in every or-der and every spectrum in the sequence. The left-hand panels inFigure 2 show that there is a clear trend in wavelength betweenthe dispersion solution for the A positions (in red) and that ofthe B positions (in blue). This trend seems consistent betweenthe two observing nights (top and bottom row). The right-handpanels show the temporal variations in the wavelength solutiononce the overall trend is removed. Not only does the wavelengthrecorded on a certain pixel of the detector vary as a function oftime, but also the amplitude of this variation di ff ers between thetwo observing nights. Whereas on July 11 the dispersion solu- tion of GIANO is stable to 1 /
20 of a pixel (1 pixel equals to 2.7km s − ), on July 30 it varies by at least 0.2 pixels. We further dis-cuss the stability of the night of July 30 in Section 4.2. For bothnights, we proceeded to re-align each spectrum in the temporalsequence by spline interpolating it based on the measured shifts. Once the spectra are all in the telluric reference frame, we cali-brated their wavelengths. We note that the set of arc frames takenafter the observing night are not suitable for calibration at thelevel of precision required by our analysis. Since we showed inSection 2.2 the instability of the dispersion solution in time, anycalibrations non-simultaneous to the science observations couldresult in a biased wavelength solution. In the NIR telluric ab-sorption lines provide instead an excellent simultaneous cali-bration source, except for some orders of the J band where theEarth’s atmosphere is particularly transparent. However, the starHD 189 733 is a K1-2V dwarf and it shows significant absorp-tion lines at near-infrared wavelengths. These can also be usedfor calibration. Article number, page 3 of 10 & A proofs: manuscript no. 32189_arxiv_v2
We built a template for wavelength calibration by multiply-ing a Phoenix stellar spectrum (Allard et al. 2011) matching theproperties of HD 189 733 by the atmospheric transmission spec-trum generated via the ESO Sky Model Calculator . The stellarspectrum is Doppler-shifted based on the barycentric and sys-temic velocities at the time of the observations. For each order,we visually paired absorption lines in the template and in the av-eraged observed spectrum. We then computed the wavelengthsof the line centroids in the template spectrum, and the pixel valueof the corresponding lines in the averaged spectrum. The cen-troids were determined by computing a super-sampled versionof each line via spline interpolation. Super-sampling is achievedvia spline interpolation at the 1 /
20 of a pixel. We recorded thefractional pixel at which the flux in the super-sampled observedline reaches a minimum, and the corresponding wavelength inthe super-sampled template spectrum. We exclude from the anal-ysis lines that appear saturated, close to saturation, or blended.The (pixel, wavelength) relation obtained from the above is fit-ted with a fourth-order polynomial, and the fit is assumed as thewavelength solution. For most of the orders we achieve a resid-ual scatter per line well below 1 km s − . Figure 3 shows theresiduals as a function of wavelength, and its + σ (0.75 km s − )and − σ (1.25 km s − ) dispersion on the global sample. Some ofthe orders corresponding to spectral windows where the Earth’satmosphere is particularly opaque do not have enough flux or un-saturated spectral lines to calibrate in wavelength. Furthermore,for a few orders in the J band, we notice an evident mismatch be-tween modelled and observed spectral lines, which also preventsus from obtaining a reliable calibration. Due to these challenges,we exclude six orders from the night of July 11 (orders 8, 9,10, 23, 24, 25, corresponding to the ranges 983.06-983.30 nm,1352.7-1408.7 nm, and 1804.8-1937.8 nm) and six orders fromthe night of July 30 (orders 8, 23, 24, 43, 45, 46, correspondingto 970.8-983.4 nm, 1009.7-1032.3 nm, 1352.8-1408.7 nm, and1895.1-1937.8 nm).
3. Extracting the planet signal
At this stage of the analysis the planet signal is outshone byorders of magnitude by the stellar and telluric spectra. How-ever, the planet’s orbital velocity has a non-zero radial compo-nent during transit, which amounts to −
16 km s − in ingress and +
16 km s − in egress. Consequently, while telluric and stellarlines are stationary or quasi-stationary (the stellar barycentricvelocity changes by 0.2 km s − during transit) in wavelength,the planet spectrum experiences a detectable change in Dopplershift during the 110 minutes of transit. The analysis describedin this section aims to make use of this peculiar Doppler signa-ture to remove the telluric and stellar spectra while preservingthe planet signal as much as possible. We have treated each order separately, as a bidimensional ma-trix in which wavelength ( λ ) is on the x -axis and time ( t ) on the y -axis. Due to the possible di ff erences in the properties of thedetector (quantum e ffi ciency and bad pixels) between the twonodding positions, A and B spectra (i.e. odd and even spectrain the temporal sequence) are also treated separately. Lastly, wereport a clear discontinuity between the right and left quadrantof the detector. Hence, we also separated x -pixels 0–1023 from × × =
164 matrices ofspectra for each observing night. We note here that we alreadydiscarded the six orders with inaccurate wavelength calibration(Section 2.3).The analysis begins by taking the natural logarithm of eachmatrix. This choice reflects our expectation that at first order thedepth of telluric lines depends on the exponential of the geomet-ric airmass ( a ( t ), see below). Before taking the logarithm, all thepixels below a linear count level of 0.1 are masked to avoid di-vergent or infinite values. Residual bad columns of data (i.e. cer-tain wavelengths falling on damaged portions of the detector),previously identified by visual inspection, are also masked.With the data in logarithmic space, the median of the 300brightest pixels is subtracted from each observed spectrum to re-move variations in throughput due to pointing, seeing, and skytransparency. A temporal average of each matrix is then taken toconstruct the mean observed spectrum for the night. A robust lin-ear fit between each observed spectrum and the mean spectrumis computed and subtracted out, removing most of the spectralfeatures. However, an additional correction is required to modelthe change with time of the depth of telluric lines. For each λ i inthe data, this is done by subtracting the fit of the observed fluxlog [ F ( λ i , t )] with the geometric airmass a ( t ). If all the species inthe Earth’s atmosphere were vertically mixed and had constantabundances, a linear fit with airmass would su ffi ce. However, thewater vapour content is likely to change during the night, andin addition water is confined to the Earth’s troposphere. Conse-quently, the amount of water vapour along the line of sight willchange in a non-trivial fashion. In these data, we fitted the func-tional dependence by airmass aslog [ F ( λ i , t )] = c i + c i a ( t ) + c i a ( t ) , (1)which we find far superior to a linear fit in modelling the depth oftelluric lines. The fit is computed and subtracted out for each ofthe spectral channels in the data (each of the columns in the ( λ, t )matrix). Although we still detected residual telluric signal in ourdata (see Section 3.2 below), we verify that it is not possibleto improve the correction by just increasing the degree of thepolynomial in a ( t ). We suspect that this happens because even ona photometric night the water vapour content changes irregularlyon timescales of minutes, and these variations cannot be easilymodelled with a smooth function of time.With telluric lines removed as above, the previous mask forbad columns and / or low-count pixels is applied again, settingthe corresponding pixels to zero. An additional sigma-clippingis applied with threshold of eight times the standard deviation ofeach matrix to ensure that no residual strong outlier is present.Lastly, the exponential of the data is taken to restore linear fluxvalues.Since we removed only the time-averaged quantities fromthe data, or we modelled each spectral channel as function oftime separately, we preserved most of the planet signal in theprocess. This is due to the fact that absorption lines in the planetspectrum will shift wavelength during the night, and thereforewill fall on a certain spectral channel only for a limited amountof time. Even if preserved, the planet spectrum is still below the noisethreshold at this stage. The deepest absorption lines in the planettransmission spectrum barely reach a depth of 1% comparedto the stellar continuum, while our typical S / N is 50-60 per
Article number, page 4 of 10. Brogi et al.: Exoplanet atmospheres with GIANO C a li b r a t i o n e rr o r s ( k m s − ) Fig. 3.
Residuals of the wavelength calibration (in km s − ) versus wavelength for the full spectral coverage of GIANO. The dashed lines denotethe ± σ in the distribution of the residuals. For most of the orders, a sub-km / s per line wavelength solution is achievable. spectrum per spectral channel. However, there are thousandsof molecular lines in the spectral range covered by GIANO.Their signal can be co-added by cross-correlating the residualdata after Section 3.1 with template spectra for the planet atmo-sphere. These are computed via plane-parallel, line-by-line ra-diative transfer calculations following the prescriptions of Brogiet al. (2016) for what concerns T - p profile and H –H collision-induced absorption. In this work CO is not included as trace gas.Given that this molecule mainly absorbs at selected wavelengthintervals (longer than 2290 nm, and much more weakly in therange 1560-1600 nm), that GIANO data is a ff ected by strongmodal noise (i.e. noise related to fibre modes) in the K band, andthat the stellar Rossiter-McLaughlin e ff ect would likely domi-nate the CO signal without an appropriate correction of the stel-lar spectrum (Brogi et al. 2016), we opt to explore H O andCH as trace gases in this analysis, with opacities taken fromHITEMP2010 (Rothman et al. 2010) and HITRAN2012 (Roth-man et al. 2013) respectively. Not only are these two speciesstrong tracers of chemistry in hot-Jupiter atmospheres (Moseset al. 2011; Madhusudhan 2012; Moses et al. 2013), but theyalso show a dense forest of spectral lines across most of the NIR.Hence, they are well suited to exploit the large spectral range ofGIANO.In our calculations we first set the volume mixing ratio(VMR) of methane to 10 − and explore VMRs of 10 − , 10 − ,and 10 − for water. We then set VMR(H O) = × − , i.e.the chemical equilibrium value for T = / O, andsolar metallicity (Madhusudhan 2012), and add CH at VMRsbetween 10 − and 10 − , in increments of one order of magni-tude. We kept the T - p profile in the lower part of the atmosphere( p > . ff erent temperatures (500 Kand 1500 K) for the upper part of the atmosphere. Althoughthese would produce completely di ff erent emission spectra (i.e.absorption and emission lines, respectively), they have a muchsmaller impact on the transmission spectrum. The main e ff ect ison the overall depth of absorption lines, with the hotter upperatmosphere producing models with up to 4 × the contrast ratioof the colder models. The influence on the relative strength ofabsorption lines is instead minimal. In conclusion, for this studywe test a set of 14 models (seven relative abundances and two T - p profiles).Each model has the average transit depth subtracted be-fore cross-correlation. In this way both the models and the data have their continuum set to zero. The cross correlationis computed on a fixed grid of radial velocity lags between −
225 and +
225 km s − , in steps of 2.7 km s − , which approx-imately matches the pixel scale for most of the GIANO or-ders. This results in a lag vector of 167 radial velocity values.For each spectrum, each order, and each radial-velocity lag, theDoppler-shifted model spectrum is computed by spline interpo-lation and cross-correlated with the data. The output of the cross-correlation is a set of 47 matrices (one per order), each with di-mensions (167 × n spec ), where n spec is the number of spectra oneach night (90 and 110 on July 11 and 30, respectively).
75 50 25 0 25 50 75Radial velocity (km/s)0.030.020.010.000.010.02 P l a n e t o r b i t a l p h a s e All orders - no masking
75 50 25 0 25 50 75Radial velocity (km/s)0.030.020.010.000.010.02 P l a n e t o r b i t a l p h a s e Selected orders + masking
Fig. 4.
Cross-correlation between the sequence of spectra ofHD 189 733 observed with GIANO on July 11, 2015, and a model spec-trum for the planet HD 189 733 b containing H O at VMR = − ,shown as function of radial velocity and planet orbital phase. Ingressand egress of HD 189 733 b are marked with horizontal dashed lines. Inthe left panel, we co-added all the GIANO orders after removing telluriclines similarly to past CRIRES observations (Brogi et al. 2014, 2016).In the right panel, we applied an additional mask at the position of tel-luric H O lines. The latter step removes the majority of time-correlatedresidual telluric noise, leaving a faint trail of positive correlation alongthe expected planet radial velocity (marked by slanted dashed lines).Details can be found in Section 3.1.
Even at this stage the planet signal is not expected to bedetectable yet. We therefore proceeded to co-add the 47 cross-correlation matrices with equal weights, even though the sig-nal in each order will depend on i) the transparency of theEarth’s atmosphere, ii) the density and depth of absorption linesin the spectrum of HD 189 733 b, and iii) the position of ex-oplanetary lines compared to strong telluric lines. Factors ii)
Article number, page 5 of 10 & A proofs: manuscript no. 32189_arxiv_v2 and iii) could be computed from our set of model spectra usedfor cross-correlation, but this would make the weighting model-dependent. In order to avoid this bias, we therefore assumedequal weighting.The remainder of this Section describes the analysis tunedspecifically for the night of July 11, 2015. This is due to the factthat on the night of July 30, 2015, no signal is detected (see Sec-tion 4 and more specifically Section 4.2). However, the qualita-tive behaviour of the cross-correlation signal and the consequentstrategy, discussed below, apply to both nights. As illustrated inthe left panel of Figure 4, the co-added cross-correlation sig-nal shows a prominent feature spanning the entire sequence andsmoothly shifting in time from anti-correlation (lighter coloursin the plot) to correlation (darker colours). We identified this fea-ture with spurious residual telluric absorption because it is cen-tred at zero radial-velocity lag in the telluric frame, it appearsregardless of the planet orbital phase (hence it is not correlatedwith the transit of the planet), and it is enhanced when cross-correlating the data with a model for the Earth’s transmissionspectrum (not shown here), containing the cold H O spectrumrather than the more complex and rich hot H O spectrum of theplanet HD 189 733 b.In past CRIRES data where we de-trended telluric lines byairmass (Brogi et al. 2012, 2013, 2014), residuals were clearlyvisible in the processed spectra prior to cross-correlation (i.e. atthe end of the analysis in Section 3.1). Therefore, the strongestmeasured residuals at the position of telluric lines could be usedto de-trend the rest of the spectral channels, generally achiev-ing a near photon noise correction. In these GIANO data, how-ever, no residual is clearly visible prior to cross-correlation. Wetherefore took a more straightforward approach where we maskthe data prior to cross-correlation by assigning zero value tothose spectral channels corresponding to telluric lines deeperthan 5-30%, with the actual threshold varying based on the or-der but not significantly influencing the outcome. We repeatedthe cross correlation, and visually inspect the matrices for eachorder. For 24 of the 41 orders with good wavelength calibra-tion, any residual telluric signal is suppressed below the cross-correlation noise. For the remaining 17 orders, we still identi-fied some cross-correlation noise by visual inspection, althoughin broader patterns spanning tens of km s − in radial velocity.These noise structures do not resemble the signature of residualtelluric lines. We therefore also discarded these additional ordersfrom the analysis, ending up utilising approximately 50% of theavailable spectral range of GIANO (orders 2, 7, 12-17, 19, 27-30, 33-35, 38-44, and 46). We note that some of the discardedorders are a ff ected by either strong telluric absorption (orders8-10, 22-25) or modal noise (orders 0-6). They are both likelycauses of this extra cross-correlation noise.With the 24 good orders selected as above, we achieved a co-added matrix of CCFs relatively free from telluric contamination(Figure 4, right panel). A slanted trail of darker cross-correlationvalues (higher correlation in the adopted colour scheme) isbarely visible at the expected planet radial velocity (along theslanted guidelines) and only in-transit. This is the CCF of theplanet transmission spectrum, which we then proceeded to co-add in the rest-frame of HD 189 733 b to obtain the total signalfrom the planet.Co-adding requires computing the planet radial velocity v P in the telluric reference system. Assuming a circular orbit, this isgiven by: v P ( t , K P ) = v bary ( t ) + v sys + K P sin[2 πϕ ( t )] , (2) where v bary is the velocity of the barycentre of the solar sys-tem with respect to the observer, v sys is the systemic velocityof HD 189 733 ( − . − , Triaud et al. 2009), K P is theplanet orbital radial velocity semi-amplitude, and ϕ the planetorbital phase. The latter is zero at the centre of the transit, 0.5at mid-secondary eclipse, and 0.25-0.75 when the planet is inquadrature. It is computed from the time t of the observations,the time of mid-transit T and the planet orbital period P (Agolet al. 2010) as the fractional part of ϕ ( t ) = t − T P . (3)For each value of K P we shifted each cross correlation functionat time t via linear interpolation so that it is centred around v P .We then co-added the shifted cross-correlations in time to ob-tain the total cross-correlation signal from the planet as a func-tion of rest-frame velocity v rest and planet radial velocity semi-amplitude K P (Figure 5, left panels). If the planet transmissionspectrum is detected, we measure a signal at v rest = . + . − . km s − and K P = . + . − . km s − . The latter being computedfrom the semi-major axis, orbital period, and orbital inclinationin the literature (Triaud et al. 2009; Agol et al. 2010), and byerror propagation. Since the uncertainty in v rest varies with or-bital phase via Eq. 2, we computed it for every value of ϕ andadopted the mean value of the series. This simulates the shiftingand co-adding of the CCF illustrated above.We note that although K P is generally well known for transit-ing exoplanets, we still explore a full range of values. This allowsus to verify that no other spurious signals produce a significantdetection near the planet’s position. In Figure 5 (top panels), forinstance, residual uncorrected signal is still present at very low K P . This is typically due to either stellar residuals (if at zerorest-frame velocity) or telluric residuals (at non-zero rest-framevelocity, as in this case). Exploring a su ffi ciently large parame-ter space o ff ers a strong diagnostic power on all sources of noiseand we strongly recommend its implementation every time thesignal from the planet can be separated in Doppler space fromstellar and telluric lines. Following Brogi et al. (2012, 2013, 2014), we estimated the sig-nificance of the detection with two di ff erent methods. Firstly, wedivided the peak value of the cross-correlation function at each K P by the standard deviation of the noise away from the peak.In this way we computed the S / N of the detection, which is agood proxy for its significance in the absence of significant cor-related noise. The latter hypothesis is verified by studying thedistribution of the cross-correlation values more than 10 km s − away from the planet radial velocity (hereafter denoted as ‘out-of-trail’ values). In Figure 6, we compare it to a Gaussian curvewith the same mean and standard deviation as the sample. We didnot measure any deviations from a Gaussian distribution downto approximately 4 σ , the limit being imposed by the number ofavailable cross-correlation values. In the same Figure, we showin red the distribution of the cross-correlation values within 4km s − from the planet radial velocity (‘in-trail’ values). It is al-ready evident even from visual inspection that while for the firstnight (top panel) the in-trail distribution has a higher mean thatthe out-of-trail distribution (consistent with a higher degree ofcorrelation), in the second night (bottom panel) the two distribu-tions broadly overlap.We quantified the above statement statistically by perform-ing a generalised t-test on the data. The null hypothesis H is that Article number, page 6 of 10. Brogi et al.: Exoplanet atmospheres with GIANO
60 40 20 0 20 40 60Planet rest-frame velocity (km/s)050100150200 P l a n e t m a x . R V ( k m / s ) S/N map - Selected orders P l a n e t m a x . R V ( k m / s ) Significance (T-test) σ )60 40 20 0 20 40 60Planet rest-frame velocity (km/s)050100150200 P l a n e t m a x . R V ( k m / s ) S/N map - Selected orders P l a n e t m a x . R V ( k m / s ) Significance (T-test) σ ) Fig. 5.
Detection of H O in the transmission spectrum of HD 189 733 bobserved on July 11, 2015 (top panels) and the non detection on thenight of July 30, 2015 (bottom panels). The total cross correlation sig-nal is shown as function of rest-frame velocity v rest and planet radialvelocity semi-amplitude K P . The measured S / N and corresponding sig-nificances are shown in the left and right panels, respectively. These arecomputed as explained in Section 3.3. The cross-correlation shown hereis obtained with a model spectrum containing H O at VMR = − andCH at VMR = − . out-of-trail and in-trail values have the same mean. The test isperformed for the same range of v rest and K P as before. For eachvalue of the two parameters, di ff erent sets of cross-correlationvalues populate the in-trail or out-of-trail sample, and we trans-lated the corresponding t value into a significance at which H isrejected (Figure 5, right panels), which is by definition the sig-nificance of our detection.
4. Results
Despite the exclusion of a fraction of the data due to telluricresiduals, we clearly detect the absorption of water vapour in thetransmission spectrum of HD 189 733 b taken on July 11, 2015.When cross correlating with a model containing H O at VMR = − and CH at VMR = − , we measure a S / N = . σ (Figure 5, top panels). Varying the VMR ofH O, as well as including non-negligible VMRs of CH in ourmodels result in a marginal decrease in S / N for the entire rangeof VMRs tested. This suggests that CH is not detected in theplanet’s transmission spectrum, in line with previous measure-ments in the K band (Brogi et al. 2016). We do not detect thetransmission spectrum of HD 189 733 b on the second availabletransit (July 30, see Figure 5, bottom panels). In Section 4.2 wefurther discuss this non-detection.By averaging the maxima in the CCF and t-test matrices(Fig. 5, top panels), we derived a planet maximum radial veloc- N o r m a li s e d o cc u rr e n c e Distribution of cross-correlation values - Night 1
Gaussian distributionOut-of-trailIn-trail0.3 0.2 0.1 0.0 0.1 0.2 0.3Cross-correlation values10 N o r m a li s e d o cc u rr e n c e Distribution of cross-correlation values - Night 2
Gaussian distributionOut-of-trailIn-trail
Fig. 6.
Distribution of cross-cross correlation values away from the ra-dial velocity of HD 189 733 b (out-of-trail values, in blue), comparedto a Gaussian distribution with the same sample mean and variance(dashed line, light blue). The top panel is for the first night of obser-vations, the bottom panel for the second. In both datasets, the sampledistribution shows no deviation down to approximately four times thestandard deviation. For comparison, the corresponding distribution ofthe cross-correlation values around the planet radial velocity (in-trail)is shown in red. In the data from the first night, the two distributionsappear shifted as one would expect if the planet transmission spectrumis detected. Conversely, no evident shift is detected in the data fromthe second night. The statistical significance of these shifts is quantifiedthrough a generalised t-test as explained in Section 3.3. ity of K P = + − km s − . The 1- σ uncertainty corresponds toa drop of one in either S / N or significance. The measured K P isfully compatible with the literature value reported in Section 3.2( K P = . + . − . km s − ), and also with the value measured inBrogi et al. (2016), that is K P = + − km s − . We note that thelarge uncertainty in K P is due to the relatively small change inplanet redial velocity during transit (a few tens of km s − ), in-su ffi cient to tightly constrain the orbit at this level of S / N. Pastdayside observations, particularly those co-adding spectra span-ning a wide range orbital phases (Brogi et al. 2012, 2014) pro-vided instead much tighter constraints on K P , with error barsof a few km s − . We measured a net blue-shift of the planetCCF ( − . + . − . km s − ). Although very marginal, this is in linewith previous measurements by Brogi et al. (2016) and Louden& Wheatley (2015) regarding the strength of day-to-night sidewinds at the planet’s terminator. We note that high-altitude windsflowing from the dayside to the night-side hemisphere of hot Article number, page 7 of 10 & A proofs: manuscript no. 32189_arxiv_v2
Jupiters are predicted by global circulation models, and expectedto produce a radial-velocity anomaly of 1-2 km s − in the peakposition of the CCF (Miller-Ricci Kempton & Rauscher 2012;Showman et al. 2013; Kempton et al. 2014). This is in line withthe measurement in this paper, although the signal detected withGIANO is both significantly broader and at lower significancethan the combined CO + H O detection of the transmission spec-trum of HD 189 733 b made in the K band with CRIRES (7.6 σ ,see Brogi et al. 2016). This contributes to the bigger uncertain-ties in estimating the net blue-shift from the peak position of theco-added planet signal.In Figure 7 we compare the width of our detection on thefirst night of observations to a simulated profile. The latter isthe convolution of the autocorrelation function of the best-fittingmodel, a Gaussian with FWHM = − approximating theinstrument profile of GIANO, and a rotational profile obtainedwith the rigid-rotation model of Brogi et al. (2016) for an equa-torial rotation of 3.4 km s − (i.e. their best-fitting value). Theagreement between the measured and the theoretical widths ofthe signal is excellent. This further suggests that no other spu-rious source of correlated noise contribute to the data. We notethat if we increase the level of telluric noise by progressively re-including those discarded orders where the masking of telluriclines did not work, and / or by reducing the threshold of the mask(see details in Section 3.1), we observe a progressive broadeningof the planet signal especially in the red wing, which is coinci-dent with the position of telluric lines (red-shifted) relative to theplanet signal on July 11 (see Figure 8). The correspondence be-tween the expected and the measured width of the signal is hencean additional indicator for an adequate correction of residual tel-luric signal.The marginal detection of a global blue shift on these datasuggests that the lower spectral resolution of GIANO comparedto CRIRES is unlikely to provide meaningful constraints on theplanet rotational rate and / or equatorial winds, which were al-ready hard to constrain with CRIRES at a significance of 7.6 σ .We therefore refrain from applying a more sophisticated modelfor the planet rotation in this study.
30 20 10 0 10 20 30Rest-frame velocity (km s )0123456 C o - a dd e d p l a n e t s i g n a l ( S / N ) MeasuredExpected
Fig. 7.
Shape of the measured planet signal from the first night of ob-servations (black line) compared to the theoretical analogue (cyan line)computed by convolving the autocorrelation function of the best-fittingmodel, a Gaussian instrumental profile with FWHM = − ,and a rotational profile (rigid-body rotation, from Brogi et al. (2016))with equatorial velocity of 3.4 km s − and day-to-night wind speed of − . − . The agreement between the two profiles is remarkable. In Section 3.2 we showed that telluric lines need to be maskedeven after they are de-trended by airmass to achieve a cross-correlation signal free from telluric contamination. We explainedthat this is likely due to rapid modulations in the flux of telluriclines due to changes in the water vapour content and possibly in-strumental profile, which we cannot model with low-order poly-nomials. However, these variations are largely common-modebetween all the spectral channels a ff ected by telluric lines. Thisopens up to the possibility of using de-trending algorithms suchas principal component analysis (PCA) to model and removethese correlated (in time and in wavelength) signals.A PCA algorithm was successfully applied by de Kok et al.(2013) to CRIRES data, and more recently Piskorz et al. (2016,2017) have also utilised it - although along the spectral direc-tion rather than the temporal direction - to clean their data fromtelluric contamination. We have developed our own version ofPCA for de-trending GIANO data. As for the standard pipeline,we process each order separately and then co-add their CCFsat a later stage. So far our results have been sub-par comparedto the masking technique, mainly due to the fact that the algo-rithm struggles in isolating telluric residuals from other time-dependent noise components: – For an unweighted PCA (i.e. where each spectral channel isequally weighted), the most significant eigenvectors do notcontain telluric residuals, but rather trends in spectral chan-nels with very low flux (for instance at the centre of a satu-rated telluric line) and broad-band variations of the through-put (for instance due to the instrument modal noise). – Even when deep absorption lines are masked and the signal-to-noise of each channel is used as weighting, the eigenvec-tors still model broad-band sources of correlated noise. Themain reason for this is that broad-band variations a ff ect alarger portion of the data than narrow telluric lines. – If more eigenvectors are linearly combined and removedfrom the data, telluric residuals are eventually suppressed,but so is the planet signal, resulting in a marginal detectionaround S / N = Although the weather and the seeing were good on both nights,several indicators point to an inferior quality of the data takenon July 30. Firstly, the bottom-right panel of Figure 2 showsthat the shifts in the dispersion solution as recorded on the GI-ANO detector are one order of magnitude bigger than on the firstnight, pointing to a lower stability of the spectrograph. Counter-intuitively, we suspect that this is due to a particularly good see-ing. This enhances the non-uniform illumination a ff ecting thefibres of GIANO, and indeed we also measure an enhancedamount of modal noise in the K-band orders (0-7 in our num-bering). Secondly, on July 30 the observations were interruptedtwice for technical problems with the instrument and its inter-face. It resulted in a loss of 20% of the planetary transit (seegaps in Figures 1 and 8) and it potentially impacted the overallstability of the instrument. Finally, Figure 8 shows the radial ve-locity of HD 189 733 b in the observer’s reference frame. Thisalso equals the relative velocity shift between telluric and planetspectra. Ideally one would want to have this shift be as large aspossible to minimise the contamination between the two spec-tra. During the first night of observations (dark blue) the rela-tive radial velocity between HD 189 733 b and the Earth was Article number, page 8 of 10. Brogi et al.: Exoplanet atmospheres with GIANO largely non-zero, except for part of the egress. However, duringthe second observing night (light blue) the planet radial velocityis zero exactly at mid-transit, which is when the strength of theplanet’s transmission spectrum is maximum. As a consequence,the masking applied to telluric lines prior to cross-correlationwill also potentially a ff ect planet lines, as they are superimposedat mid-transit. The spectra on the second night were also taken athigher airmass on average (Figure 1), requiring a more aggres-sive masking. If this choice succeeded in minimising correlatednoise at low K P (Figure 5, bottom panels), it also eroded a largerfraction of the planet spectrum, decreasing our sensitivity. R V p l a n e t - t e ll u r i c , ( k m / s ) Fig. 8.
Radial velocity of HD 189 733 b as a function of planet orbitalphase, relative to the observer. This is the sum of the systemic velocityof the system, and the velocity of the barycentre of the solar systemcompared to Earth. Symbols and colours are the same as in Figure 1.It shows that on the second observing night (light blue) the planet haszero radial velocity close to mid-transit, i.e. when the strength of thetransmission spectrum of HD 189 733 b is maximum. We comment onthe di ff erences between the two observing nights in Section 4.2. We substantiate the claim of an inferior data quality on July30 by injecting the best-fitting planet model in the spectra fromboth nights, at three times the nominal level, and at the planet ra-dial velocity obtained from Eq. 2 with the literature values listedin Section 3.2. After running the injected spectra through the fulldata-analysis pipeline, we recover a signal-to-noise ratio of 9.4and 4.3 on the first and second night, respectively. Given that thereal planet signal is detected at S / N = / N of 2.2 on the second night,well below detectability. Based on the evidence presented in thisSection, we conclude that the absence of signal on July 30 is inline with the expectations.
It is useful to assess whether the detection presented here iscompatible with the previous detection of water at 4.8 σ withCRIRES (Brogi et al. 2016). For GIANO, we measure 2 200 e − per 60 s integration per resolution element (1 pixel = − )in the K band. With CRIRES, we measure 6 000 e − per 10 s in-tegration, per resolution element (1 pixel = − ). Scalingthe latter signal based on telescope size and pixel width whileassuming equal throughput, we would expect from GIANO ap-proximately 12 500 e − per 60 s integration per pixel. ObservedGIANO spectra are thus deficient by a factor of (12.5 / = / N ofCRIRES for the same exposure time, insu ffi cient to detect the transmission spectrum of HD 189 733 b. However, we shouldalso consider the increased spectral range of GIANO. We esti-mated this additional factor by adding random noise to the bestfitting model for the planet transmission spectrum, cutting it ac-cordingly to the wavelength range of GIANO and CRIRES, andcross-correlating it with the noiseless model. For each of the GI-ANO orders, the noise is rescaled to properly account for thewavelength-dependent transparency of the Earth’s atmosphere.We repeat the simulations 100 times, and we compare the dis-tributions of the retrieved S / N for CRIRES and GIANO. On av-erage, GIANO delivers 4.5 × the S / N of CRIRES based on thespectral range alone. However, GIANO has half of the spectralresolution, producing a loss of 30% in S / N for unresolved plane-tary lines. Putting all the scaling factors together, H O shouldbe detectable with GIANO in one transit at a significance of4 . σ × . × . × . = σ . This is consistent with our re-ported detection at 5.5 σ . Not only does this show that a de-tection at high significance is possible with just 110 minutes ofTNG time, but also that the overall strength of the planet’s trans-mission spectrum has not changed significantly since the previ-ous CRIRES observations from July 2012. Future monitoring ofthe planet via the same technique will allow us to put constraintson the level of variability of the NIR transmission spectrum duefor instance to aerosols at the planet terminator.
5. Discussion
We have tested whether or not a 4-m telescope with a perform-ing high-resolution spectrograph can successfully study the at-mospheres of exoplanets at high spectral resolution. We haveshown that GIANO in its initial fiber-fed configuration, despitethe much smaller throughput, is capable of rivalling CRIRESat the VLT in detecting molecular absorbers with broad-bandopacity, such as water vapour. We have demonstrated that theanalysis devised for CRIRES still works at half of the spectralresolution, but residual telluric residuals need to be treated moreaggressively (e.g. masked) in order to avoid contamination ofthe planet signal. Smart scheduling of the observations in orderto maximise the modulus of the combined barycentric plus sys-temic velocity of the target is strongly advised to further reducetime-correlated noise coming from uncorrected telluric absorp-tion.GIANO has been recently updated to a slit-fed instrumentand coupled to HARPS-N to deliver quasi-contiguous wave-length coverage from 0.383 to 2.45 µ m (Claudi et al. 2017). Inthis configuration, the instrument will allow us to access in asingle observation much smaller atmospheric pressures throughsodium absorption (Wyttenbach et al. 2015; Louden & Wheatley2015). Furthermore, it will be possible to search for the signatureof the weaker absorption bands of water vapour (Allart et al.2017; Esteves et al. 2017), which are very sensitive to the pres-ence of clouds or hazes. On the instrumental side, this upgradeenhanced the throughput of GIANO by at least a factor of four.Furthermore, it eliminated the modal noise a ff ecting the ordersin the K band (approximately orders 0-7 in our analysis, whichwe largely discarded). When the simulations of Section 4.3 areupdated to account for these better performances, and neglectingdecreased performances in correcting for telluric lines, we con-clude that GIANO could potentially deliver 2.5-3 × the S / N ofCRIRES at the VLT in the same observing time, when lookingfor broad-band absorbers such as H O or CH .This advantagecould be used either to open up high-resolution spectroscopyto fainter systems or to start investigating the atmospheres ofsmaller and cooler exoplanets. Article number, page 9 of 10 & A proofs: manuscript no. 32189_arxiv_v2
The planned upgrade for CRIRES (CRIRES + , see Follertet al. 2014) will transform the instrument into a cross-dispersedspectrograph, bringing a 10 × increase in spectral range. Togetherwith state-of-the-art detectors, a refurbished AO system, andimproved calibration sources, CRIRES + is likely to overcomeany gaps in sensitivity with GIARPS. Whereas in the past high-dispersion observations have been limited to either the southernor the northern hemisphere, with both spectrographs online wewill be able to access the entire sky.Adding to the fleet of near-infrared, high-resolution spec-trographs, IGRINS (Park et al. 2014) and CARMENES (Quir-renbach et al. 2014) have recently come online, and SPIROU(Artigau et al. 2014) will be available soon. These instrumentsare revolutionary both in terms of spectral range and throughputcompared to the old CRIRES. If the NIR sky at their observato-ries is of comparable quality to La Palma or Cerro Paranal, and iftheir initial specifications will be confirmed (especially in termsof throughput), these instruments will be well suited to performstudies of exoplanet atmospheres such as the one presented here.Importantly for future comparative characterisation of exo-planets, spectra taken at high- and low-resolution contain highly-complementary information (Pino et al. 2017). Whereas at highresolution temperature and opacity in the planet’s atmosphereare encoded in the relative ratio between hundreds of thou-sands of spectral lines, at low resolution the same informa-tion translates into broad-band spectral variations. Brogi et al.(2017) recently demonstrated that constraints on abundances,temperature-pressure ( T - p ) profile, and consequently metallic-ity of exoplanets can be greatly enhanced when combining theconfidence intervals from space (HST / WFC3 and Spitzer / IRACdata) and ground observations (CRIRES). To enable this syn-ergy, especially in the era of TESS and JWST, it is necessaryto extend high-resolution spectroscopic observations of exoplan-ets to as many ground-based telescopes as possible. Based onthe results of this paper, and considering the increased ease ofscheduling, a 3-4-m telescope equipped with a state-of-the arthigh-resolution spectrograph could not only assist, but even out-perform the biggest ground-based telescopes for detecting keymolecular constituents such as H O and CH .The above observations make use of both spectral and tem-poral resolution to extract exoplanet signals from the over-whelming stellar and telluric contaminants. With a total inte-gration time of ten hours, core-to-wings line ratios as small as10 − (1 σ ) have been detected through their combined cross-correlation signal. Contrast ratios exceeding a billion could po-tentially be reached if the spatial separation is also used, i.e. bycoupling a high-resolution spectrograph to a coronographic orextreme-AO system (Snellen et al. 2015). Early demonstrationsof the principle were recently presented (Snellen et al. 2014;Schwarz et al. 2016), and the prospect of targeting biomarkersin terrestrial exoplanets orbiting in the habitable zones of M-dwarfs seems feasible when the next generation of extremelylarge telescopes come online (Snellen et al. 2013; Rodler &López-Morales 2014). Acknowledgements.
We thank the referee M. Kuerster for his insightful com-ments, which contributed to improving the quality of the manuscript. We thankR. Claudi and S. Benatti for their insightful discussion about GIARPS. We thankF. Borsa, A. Maggio, N. Nikolov, and I. Pagano for their initial contribution to thedesign of these high resolution observations. M.B. acknowledges partial supportby NASA, through Hubble Fellowship grant HST-HF2-51336 awarded by theSpace Telescope Science Institute. P.G. gratefully acknowledges support fromINAF through the Progetti Premiali funding scheme of the Italian Ministry ofEducation, University, and Research. G.G. acknowledges the financial supportof the 2017 PhD fellowship programme of INAF. The research leading to these results has received funding from the European Union Seventh Framework Pro-gramme (FP7 / References
Agol, E., Cowan, N. B., Knutson, H. A., et al. 2010, ApJ, 721, 1861Allard, F., Homeier, D., & Freytag, B. 2011, in Astronomical Society of the Pa-cific Conference Series, Vol. 448, 16th Cambridge Workshop on Cool Stars,Stellar Systems, and the Sun, ed. C. Johns-Krull, M. K. Browning, & A. A.West, 91Allart, R., Lovis, C., Pino, L., et al. 2017, A&A, 606, A144Artigau, É., Kouach, D., Donati, J.-F., et al. 2014, in Proc. SPIE, Vol. 9147,Ground-based and Airborne Instrumentation for Astronomy V, 914715Barnes, J. R., Barman, T. S., Jones, H. R. A., et al. 2010, MNRAS, 401, 445Birkby, J. L., de Kok, R. J., Brogi, M., et al. 2013, MNRAS, 436, L35Birkby, J. L., de Kok, R. J., Brogi, M., Schwarz, H., & Snellen, I. A. G. 2017,AJ, 153, 138Bouchy, F., Udry, S., Mayor, M., et al. 2005, A&A, 444, L15Brogi, M., de Kok, R. J., Albrecht, S., et al. 2016, ApJ, 817, 106Brogi, M., de Kok, R. J., Birkby, J. L., Schwarz, H., & Snellen, I. A. G. 2014,A&A, 565, A124Brogi, M., Line, M., Bean, J., Désert, J.-M., & Schwarz, H. 2017, ApJ, 839, L2Brogi, M., Snellen, I. A. G., de Kok, R. J., et al. 2012, Nature, 486, 502Brogi, M., Snellen, I. A. G., de Kok, R. J., et al. 2013, ApJ, 767, 27Charbonneau, D., Noyes, R. W., Korzennik, S. G., et al. 1999, ApJ, 522, L145Claudi, R., Benatti, S., Carleo, I., et al. 2017, European Physical Journal Plus,132, 364Collier Cameron, A., Horne, K., Penny, A., & James, D. 1999, Nature, 402, 751de Kok, R. J., Brogi, M., Snellen, I. A. G., et al. 2013, A&A, 554, A82Esteves, L. J., de Mooij, E. J. W., Jayawardhana, R., Watson, C., & de Kok, R.2017, AJ, 153, 268Follert, R., Dorn, R. J., Oliva, E., et al. 2014, in Proc. SPIE, Vol. 9147, Ground-based and Airborne Instrumentation for Astronomy V, 914719Horne, K. 1986, PASP, 98, 609Kaeufl, H.-U., Ballester, P., Biereichel, P., et al. 2004, in Proc. SPIE, Vol. 5492,Ground-based Instrumentation for Astronomy, ed. A. F. M. Moorwood &M. Iye, 1218–1227Kempton, E. M.-R., Perna, R., & Heng, K. 2014, ApJ, 795, 24Leigh, C., Collier Cameron, A., Horne, K., Penny, A., & James, D. 2003, MN-RAS, 344, 1271Lockwood, A. C., Johnson, J. A., Bender, C. F., et al. 2014, ApJ, 783, L29Louden, T. & Wheatley, P. J. 2015, ApJ, 814, L24Lucas, P. W., Hough, J. H., Bailey, J. A., et al. 2009, MNRAS, 393, 229Madhusudhan, N. 2012, ApJ, 758, 36Madhusudhan, N. & Seager, S. 2009, ApJ, 707, 24Miller-Ricci Kempton, E. & Rauscher, E. 2012, ApJ, 751, 117Moses, J. I., Madhusudhan, N., Visscher, C., & Freedman, R. S. 2013, ApJ, 763,25Moses, J. I., Visscher, C., Fortney, J. J., et al. 2011, ApJ, 737, 15Oliva, E., Origlia, L., Ba ff a, C., et al. 2006, in Proc. SPIE, Vol. 6269, Society ofPhoto-Optical Instrumentation Engineers (SPIE) Conference Series, 626919Origlia, L., Oliva, E., Ba ff a, C., et al. 2014, in Proc. SPIE, Vol. 9147, Ground-based and Airborne Instrumentation for Astronomy V, 91471EPark, C., Ja ff e, D. T., Yuk, I.-S., et al. 2014, in Proc. SPIE, Vol. 9147, Ground-based and Airborne Instrumentation for Astronomy V, 91471DPino, L., Ehrenreich, D., Wyttenbach, A., et al. 2017, ArXiv e-prints[ arXiv:1709.09678 ]Piskorz, D., Benneke, B., Crockett, N. R., et al. 2016, ApJ, 832, 131Piskorz, D., Benneke, B., Crockett, N. R., et al. 2017, AJ, 154, 78Quirrenbach, A., Amado, P. J., Caballero, J. A., et al. 2014, in Proc. SPIE, Vol.9147, Ground-based and Airborne Instrumentation for Astronomy V, 91471FRodler, F., Kürster, M., & Henning, T. 2010, A&A, 514, A23Rodler, F. & López-Morales, M. 2014, ApJ, 781, 54Rodler, F., Lopez-Morales, M., & Ribas, I. 2012, ApJ, 753, L25Rothman, L. S., Gordon, I. E., Babikov, Y., et al. 2013,J. Quant. Spectr. Rad. Transf., 130, 4Rothman, L. S., Gordon, I. E., Barber, R. J., et al. 2010,J. Quant. Spectr. Rad. Transf., 111, 2139Schwarz, H., Ginski, C., de Kok, R. J., et al. 2016, A&A, 593, A74Showman, A. P., Fortney, J. J., Lewis, N. K., & Shabram, M. 2013, ApJ, 762, 24Snellen, I., de Kok, R., Birkby, J. L., et al. 2015, A&A, 576, A59Snellen, I. A. G., Brandl, B. R., de Kok, R. J., et al. 2014, Nature, 509, 63Snellen, I. A. G., de Kok, R. J., de Mooij, E. J. W., & Albrecht, S. 2010, Nature,465, 1049Snellen, I. A. G., de Kok, R. J., le Poole, R., Brogi, M., & Birkby, J. 2013, ApJ,764, 182Triaud, A. H. M. J., Queloz, D., Bouchy, F., et al. 2009, A&A, 506, 377Wiedemann, G., Deming, D., & Bjoraker, G. 2001, ApJ, 546, 1068Wyttenbach, A., Ehrenreich, D., Lovis, C., Udry, S., & Pepe, F. 2015, A&A, 577,A62]Piskorz, D., Benneke, B., Crockett, N. R., et al. 2016, ApJ, 832, 131Piskorz, D., Benneke, B., Crockett, N. R., et al. 2017, AJ, 154, 78Quirrenbach, A., Amado, P. J., Caballero, J. A., et al. 2014, in Proc. SPIE, Vol.9147, Ground-based and Airborne Instrumentation for Astronomy V, 91471FRodler, F., Kürster, M., & Henning, T. 2010, A&A, 514, A23Rodler, F. & López-Morales, M. 2014, ApJ, 781, 54Rodler, F., Lopez-Morales, M., & Ribas, I. 2012, ApJ, 753, L25Rothman, L. S., Gordon, I. E., Babikov, Y., et al. 2013,J. Quant. Spectr. Rad. Transf., 130, 4Rothman, L. S., Gordon, I. E., Barber, R. J., et al. 2010,J. Quant. Spectr. Rad. Transf., 111, 2139Schwarz, H., Ginski, C., de Kok, R. J., et al. 2016, A&A, 593, A74Showman, A. P., Fortney, J. J., Lewis, N. K., & Shabram, M. 2013, ApJ, 762, 24Snellen, I., de Kok, R., Birkby, J. L., et al. 2015, A&A, 576, A59Snellen, I. A. G., Brandl, B. R., de Kok, R. J., et al. 2014, Nature, 509, 63Snellen, I. A. G., de Kok, R. J., de Mooij, E. J. W., & Albrecht, S. 2010, Nature,465, 1049Snellen, I. A. G., de Kok, R. J., le Poole, R., Brogi, M., & Birkby, J. 2013, ApJ,764, 182Triaud, A. H. M. J., Queloz, D., Bouchy, F., et al. 2009, A&A, 506, 377Wiedemann, G., Deming, D., & Bjoraker, G. 2001, ApJ, 546, 1068Wyttenbach, A., Ehrenreich, D., Lovis, C., Udry, S., & Pepe, F. 2015, A&A, 577,A62