A Population Study of Gaseous Exoplanets

Our analysis started from the raw spatially scanned spectroscopic images, using our specialized software for the analysis of WFC3, spatially scanned spectroscopic images (Tsiaras et al. 2016a, 2016b). The reduction process included the following steps: zero-read subtraction, reference pixels correction, nonlinearity correction, dark current subtraction, gain conversion, sky background subtraction, calibration, flat-field correction, and bad pixels/cosmic rays correction. In a broad sample like the current one, the possibility of observing additional sources in the field of view is high. Hence, we could not define the sky-area prior to the analysis, and the use of an automatic tool was necessary. The selected sky-area pixels were those with a flux level below a certain threshold—twice the flux median absolute deviation (mad) from the median flux—in all non-destructive reads. In cases where multiple transit observations were available (see Table 1), we calculated the position shifts by comparison with the first spectroscopic image of the first observation. This approach was followed to eliminate any systematic position shifts between the direct images of the different observations. While absolute calibration using the direct image has an uncertainty of ±0.5 pixels (Kuntschner et al. 2009), relative calibration can provide uncertainties of ±0.005 pixels (Figure 14 in Varley et al. 2017).

HD 189733 b During the spatial scans of HD 189733 b the spectrum was shifted above the upper edge of the detector. Hence, only the first three non-destructive reads were used from the forward scans and only the last five from the reverse scans. Due to the different exposure times, forward and reverse scans were processed independently as two different transit observations.

Following the reduction process, we extracted the flux from the spatially scanned spectroscopic images to create the final transit light curves per wavelength band. We considered one broad band (white) covering the whole wavelength range in which the G141 grism is sensitive (1.088–1.68 μm), and two different sets of narrow bands (spectral). The resolving power of each set of narrow bands at 1.4 μm was 50 (low) and 70 (high), respectively. In both sets, the widths of the narrow bands were varying between 0.0188 and 0.0390 μm, in a way that the flux of a Sun-like star would be equal in all the bands. The choice of the narrow bands sizes ensured an approximately uniform S/N across the planetary spectrum. We extracted our final light curves from the differential non-destructive reads, a commonly used technique (Deming et al. 2013). In this way we also avoid any potential overlap of different spectra in cases where close companions exist.

We modeled the stellar limb-darkening effect using the nonlinear formula proposed by Claret (2000). The coefficients were fitted on the specific intensity profiles, evaluated at 100 angles, directly computed from the ATLAS model (Howarth 2011), for stars with effective temperatures higher than 4000 K, or PHOENIX (Allard et al. 2012) model, for stars with effective temperature lower than 4000 K, convoluted with the throughput of the G141 grism of the WFC3 camera. The stellar parameters used can be found in Table 2.

Fitting the limb-darkening coefficients directly to the light curves (together with the other transit and instrumental parameters) is not an option, because of the many parameter degeneracies. This limitation applies particularly valid to HST observations, as they present periodic gaps during the transit events.

A detailed study by Morello et al. (2017) shows that uncertainties in the stellar models do not significantly affect the atmospheric spectra in the WFC3 passband. For a subset of planets, where fitting a linear limb-darkening coefficient was possible, we tested this option and found that it is not affecting the shape of the final spectrum but may introduce only a vertical offset. The only exception was WASP-43 b (see Figure 1).

Zoom In Zoom Out Reset image size

As in previous observations with WFC3 (e.g., Kreidberg et al. 2015; Evans et al. 2016; Line et al. 2016; Wakeford et al. 2017), our extracted raw white light curves were affected by two kinds of time-dependent systematics: the long-term and short-term "ramps." The first affects each HST visit and has a linear behavior, while the second affects each orbit and has an exponential behavior. Additional systematics that cannot be described by the above functional forms are also very common (Wakeford et al. 2016). To account for these effects, we fitted a model for the systematics simultaneously with the transit model (Kreidberg et al. 2014b; Tsiaras et al. 2016b). We varied the parameters of the short-term ramp for the first orbit in the analyzed time-series, as in many cases the first orbit was affected in a different way compared to the other orbits. In addition, the parameters of the exponential short-term ramp also varied for the mid-orbit ramps caused by buffer dumps during an HST orbit. Finally, forward and reverse scans were combined together by using a different normalization factor to account for the shift between them.

After an initial fit, we scaled-up the uncertainties on the individual data points, in order for their median to match the standard deviation of the residuals, and fitted again. In this way, we adopted more conservative values for the uncertainties of the fitted parameters, taking into account the systematics that were not described by our functional form.

All of the white light curves were fitted for the R_p/R_* and T₀ parameters, using fixed values for the P, e and ω parameters, as reported in the literature (see Table 2). Concerning the i and a/R_* parameters, the planets in our sample can be divided in three categories:

(a) Successfully fitted⁴ with literature values for i and a/R_*: this category includes the majority of the white light curves (GJ 3470 b, HAT-P-11 b, HAT-P-26 b, HAT-P-38 b, HAT-P-41 b, HD 149026 b, WASP-29 b, WASP-31 b, WASP-43 b, WASP-67 b, WASP-69 b, WASP-74 b, WASP-80 b, WASP-101 b).

(b) Successfully fitted with i and a/R_* as free parameters: this category includes those light curves that showed additional systematics when the literature values for i and a/R_* were used, but corrected by fitting for i and a/R_* (HAT-P-1 b, HAT-P-3 b, HAT-P-12 b, HAT-P-18 b, WASP-39 b, and XO-1 b).

(c) Other effects: this category includes those light curves that showed additional systematics when the literature values for i and a/R_* were used, but could not be corrected by fitting for i and a/R_*. For these planets, we finally decided to adopt the literature values for i and a/R_* if either the transit ingress or egress was not observed (HAT-P-32 b shown in Figure 2, HD 189733 b, HD 209458 b, WASP-12 b, WASP-52 b, WASP-76 b and WASP-121 b) or the fitted values for i and a/R_* if both the transit ingress and egress were observed (GJ 436 b, HAT-P-17 b, WASP-63 b).

Zoom In Zoom Out Reset image size

The higher systematic residuals in the third category of light curves (above 1.5 times the expected photon-noise limited residuals, Figure 3) could be due to non-optimal sets of stellar limb-darkening coefficients. The most likely causes of discrepancy between predicted and observed limb-darkening coefficients are, for the cooler stars, stellar activity (e.g., Csizmadia et al. 2013) or inaccurate chemical models (e.g., Allard et al. 2012), and, for the hotter stars, the use of a plane-parallel approximation rather than full spherical geometry (Hayek et al. 2012; Morello et al. 2017). We tested two different approaches to reduce the systematic noise in the residuals by changing the limb-darkening coefficients, similar to those suggested by Howarth & Morello (2017): (1) fitting for a linear limb-darkening coefficient; (2) calculating the coefficients from stellar models with different temperatures. In this way, the resulting transit depths may vary by ∼100 ppm (2–3σ).

Zoom In Zoom Out Reset image size

Finally, we fitted the spectral light curves using the divide-white method introduced by Kreidberg et al. (2014b), where the white light curve was used as a comparison source, with the addition of a normalization factor and a wavelength-dependent slope, linear with time. In the same way as for the white light curves, we performed an initial fit and then scaled-up the uncertainties on the individual data points based on the standard deviation of the residuals, and fitted again. The i and a/R_* parameters are fixed to the literature values or to the best-fit values obtained for the relevant white light curves. Concerning the fitting of the spectral light curves, the wavelength-dependent slope was not correlated with the R_p/R_* parameter, despite the strength of the slope. The only exception was HAT-P-17 b, as no observations after the transit were included in this data set. However, the strength of the slope was insignificant throughout the spectrum of HAT-P-17 b (<1σ). For each planet, two final spectra were extracted at different resolutions (high and low, from the two sets of narrow bands). For the cases where multiple transit observations were available, the final spectra were the weighted average of the individual spectra, corrected for potential offsets in the white light curve depth from one transit to another.

In Figure 3, we can see that for the spectral light curves, the standard deviation of the residuals is on average 1.05 (1.17 in the worst case) times the the expected photon noise. In addition, the residuals autocorrelation is below 0.2. These metrics indicate that while the white light curves are subjected to remaining signals in their residuals, the divide-white method used here is efficient in removing those signals from the spectral light curves. Also, the tests with different limb-darkening coefficients show that the effect of the signals not fitted to the white light curves can cause arbitrary offsets in the final spectra, but not change their shape. Hence, the uncertainties reported for the spectra are referring to the relative transit depths, not including the uncertainties on the white light-curve depth.

The only exception was WASP-43 b, for which different spectral slopes were obtained with different sets of limb-darkening coefficients. For this planet, our original spectrum obtained showed a decreasing trend toward longer wavelengths. We found that, when fitting for a linear limb-darkening coefficient, the trend in the spectrum is less strong (Figure 1) and in agreement with the literature (Kreidberg et al. 2014a). We report this as the final spectrum of WASP-43 b.

WASP-80 b: From the spectrum of WASP-80 b, one data point at 1.4 μm was excluded as it was contaminated by the zeroth order of the spectrum of a nearby source.

WASP-12 b: The spectrum of WASP-12 b was contaminated by a very close companion. To correct for this effect, we used the starring-mode spectroscopic images included in the data set. From those images, we calculated a dilution factor, which we then used to correct the spectra (Kreidberg et al. 2015).

The atmospheres of the planets analyzed here were simulated to range from 10⁻⁴ to 10⁶ Pa and sampled uniformly in log-space by 100 atmospheric layers. We tested for potential under-sampling of the atmosphere by running test retrievals at 250 and 50 layers. No significant degradation of retrieval accuracy for HST/WFC3 data could be found. Each trace-gas abundance was allowed to vary from 10⁻⁸ to 10⁻¹ in volume mixing ratios (log-uniform prior) for hot Jupiters and 10⁻⁸–1.0 for Neptunes (i.e., HAT-P-11 b). From here forth, all priors are assumed to be uniform unless specified otherwise. We calculated planetary equilibrium temperatures T_p assuming geometric albedos varying from 0.6 to zero and emissivity from 0.5 to 1 to calculate the temperature prior range (as shown in Equation (1)):

$$\begin{eqnarray}&&{T}_{p}={T}_{* }\,{\left(\displaystyle \frac{{R}_{* }}{2a}\right)}^{1/2}{\left(\displaystyle \frac{1-A}{\varepsilon }\right)}^{1/4}\end{eqnarray} \tag{ 1 }$$

where R_* is the stellar radius, a is the semimajor axis, A is the geometric albedo and ε is the planetary emissivity. For our temperature priors we used the $[{T}_{p}(A=0.6,\varepsilon =1)-500\,{\rm{K}}$, ${T}_{p}(A=0,\varepsilon =0.5)+500\,{\rm{K}}]$ range. We adopted a wide temperature prior to allow for significantly cooler terminator temperatures compared to the expected equilibrium temperatures. Due to the short wavelength coverage of the HST/WFC3 instrument, we typically only probe a very restricted range of the planet's temperature–pressure profile.

An isothermal temperature–pressure profile was assumed. While this is an oversimplification and can lead to retrieval biases (Rocchetto et al. 2016), the restrictive wavelength range of 1.1–1.8 μm does not allow the differentiation of an isothermal from a more complex profile. We adopted the planetary radius uncertainties reported in the literature as prior bounds and corrected them if needed.

Initially, exploratory retrievals were run to include a wide range of molecular opacities: H₂O, HCN, NH₃, CH₄, CO₂, CO, NO, SiO, TiO, VO, H₂S, and C₂H₂. No significant contributions were found but for H₂O, TiO, and VO. We hence restricted further retrievals to a smaller set of molecules: H₂O (Barber et al. 2006), CO (Rothman et al. 2010), CO₂ (Rothman et al. 2010), CH₄ (Yurchenko & Tennyson 2014), and NH₃ (Yurchenko et al. 2011). VO (McKemmish et al. 2016) and TiO (L. K. McKemmish 2018, in preparation) were added to the mix for planets with equilibrium temperatures exceeding 1400 K. Tau-REx is designed to operate with either absorption cross-sections or correlated-k coefficients. Both cross-sections and k-tables were computed from very high-resolution (R > 10⁶) cross-sections, which in turn were calculated from molecular line lists obtained from ExoMol (Tennyson et al. 2016), HITEMP (Rothman et al. 2010), and HITRAN (Rothman et al. 2013). Temperature and pressure dependent line-broadening was included, taking into account J-dependence where available (Pine 1992). The absorption cross-sections were then binned to a constant resolution of R = 15000 and the transmission forward models were calculated at this resolution before binning to the resolution of the data. Given the resolutions, wavelength range and uncertainties of the data at hand, we find no differences between the use of cross-section and k-tables in the final retrieval results. Rayleigh scattering and collision induced absorption of H₂–H₂ and H₂–He was also included (Borysow et al. 2001; Borysow 2002; Rothman et al. 2013).

A variety of cloud parameterizations of varying complexity exist in the context of atmospheric retrieval studies (e.g., Benneke & Seager 2012; Barstow et al. 2013; Griffith 2014; Line & Parmentier 2016). Here, we adopted the parameterization of Lee et al. (2013), which also finds implementation in an atmospheric retrieval context in Lavie et al. (2017). In transmission spectroscopy, the cloud optical depth as function of wavelength, ${\tau }_{c1,\lambda }$, is given by:

$$\begin{eqnarray}&&{\tau }_{c1,\lambda }={\int }_{0}^{l(z)}{Q}_{\mathrm{ext},\lambda }\pi {\alpha }^{2}{\chi }_{c}(z^{\prime} ){\rho }_{N}(z^{\prime} ){dl}\end{eqnarray} \tag{ 2 }$$

where z is the height in the atmosphere, α is the particle size of the cloud/haze, dl is the path length through the atmosphere, χ_c is the cloud mixing-ratio, ρ_N is the atmospheric number density, and Q_ext,λ is the cloud extinction coefficient given by:

$$\begin{eqnarray}&&{Q}_{\mathrm{ext},\lambda }=\displaystyle \frac{5}{{Q}_{0}{x}^{-4}+{x}^{0.2}}\end{eqnarray} \tag{ 3 }$$

where x = 2πα/λ and Q₀ determines the peak of Q_{ext, λ}. This can be understood as a cloud compositional parameter (Lee et al. 2013). For α ≪ λ, the formalism reduces to pure Rayleigh scattering. In addition to the above, we implemented an optically thick gray-cloud cover, parameterized as follows:

$$\begin{eqnarray}&&{\tau }_{c2}=\left\{\begin{array}{l}1,\mathrm{if}\ P\lt {P}_{\mathrm{cloud} \mbox{-} \mathrm{top}}\\ 0,\mathrm{otherwise}\end{array}\right.\end{eqnarray} \tag{ 4 }$$

where P_cloud-top is the cloud-top pressure. This dual parameterization allowed us to model optically thick cloud decks with a semi-transparent, hazy, atmosphere above P_cloud-top.

We initially kept Q₀, χ_c, α (called R_cloud in our retrieval corner plots), and P_cloud-top as free cloud parameters but found HST/WFC3 data to be insufficient to constrain Q₀. In initial tests, we varied Q₀ with a prior range from 0 to 100 but found Q₀ to be unconstrained by the data. We have therefore fixed Q₀ to the median value of 50 henceforth. We have found that uncertainty induced by either varying or fixing Q₀ is negligible given the quality of the data at hand. We set a log-uniform prior of χ_c ranging from 10⁻⁴⁰ to 10⁻¹⁰, particle size from 10⁻⁵ to 10 μm and cloud-top pressure from 10⁻⁴ to 10⁶ Pa (Lee et al. 2013).

In the end, we had 10–12 free parameters: five molecular abundances (seven when TiO and VO were included), temperature, planet radius, and three cloud-deck parameters. Each one of the two spectra per planet at different resolutions was retrieved, yielding 60 retrievals in total. However, we found no difference between the information retrieved from the two spectra at different resolution. The results reported are from the low-resolution spectra.

In order to quantify the detection significance of an atmosphere, we devised the ADI. The ADI is the positively defined Bayes Factor between the nominal atmospheric model (M_N) and a flat-line model (M_F). As stated above, the nominal model contains molecular opacities, cloud/haze opacities (${\tau }_{c1,\lambda },{\tau }_{c2}$) collision induced absorption of H₂–H₂/H₂–He and Rayleigh scattering. Other free parameters are the planet radius, R_p, and the temperature of the isothermal TP-profile, T_iso. The flat-line model contains only gray-cloud opacities, τ_c2, R_p, and T_iso. This parameterization always results in a flat-line spectrum but includes the model degeneracies found between cloud-top pressure, planet radius, and temperature. This way we capture both cloudy as well as clear sky scenarios. As the ADI is a fully Bayesian model selection metric, we naturally impose Occam's razor to our atmosphere detection significance.

We obtained the Bayesian evidence of our nominal model, E_N, and of the pure-cloud/no-atmosphere model, E_F, and calculated the ADI as follows:

$$\begin{eqnarray}&&\mathrm{ADI}=\left\{\begin{array}{l}\mathrm{log}({E}_{{\rm{N}}})-\mathrm{log}({E}_{{\rm{F}}}),\mathrm{if}\ \mathrm{log}({E}_{{\rm{N}}})\gt \mathrm{log}({E}_{{\rm{F}}})\\ 0,\mathrm{otherwise}\end{array}\right..\end{eqnarray} \tag{ 5 }$$

The ADI is a positively defined metric and equivalent to the logarithmic Bayes Factor (Kass & Raftery 1995) where log(E_N) > log (E_F).

The low-resolution spectra obtained for all of the planets in our sample are included in the on-line database (see the following section). The ADI index has been reported for all of the planets in Figure 4 and Table 3. The spectra in Figure 4 are ordered by decreasing ADI.

Zoom In Zoom Out Reset image size

Given the definition of the ADI index in the previous section, an atmosphere is detected at 3σ and 5σ level for ADIs above 3 and 11, respectively. In our sample, we find that 16 out of 30 planets feature statistically significant atmospheres, with ADIs higher than 3. While parameter constraints of atmospheric models for many of the planets with ADIs lower than 3 can be significant, indicating the presence of water (WASP-80 b, WASP-43 b, HAT-P-12 b, HAT-P-38 b, WASP-31 b, WASP-63 b, GJ 3470 b, WASP-67 b, WASP-74 b), the model as a whole is not. Hence, ADIs below 3 signify atmospheric nondetections, as the spectral feature amplitudes are insufficient (given the uncertainties in the data) to favor the more complex atmospheric model, M_N over the lower dimensional flat-line model M_F. To verify the presence of water in these planets, additional observations are necessary. We have to note here that for WASP-43 b the presence of water has been confirmed using additional observations during the eclipse of the planet (Kreidberg et al. 2014a). By adopting the ADI, we were able to draw several important conclusions about this population of exoplanets and spectroscopic observations of exoplanets in general.

Previous population studies suggested that the observed spectra do not show the strong molecular features expected for a clear sky atmosphere (Iyer et al. 2016; Sing et al. 2016). Interestingly, even in this larger sample with all of the planets expected to feature some modulations given the precision of the observations, the ADI does not correlate with the pre-calculated S/N. To exclude any observational biases we repeated the S/N calculation using the median uncertainty of our final observed low-resolution spectra instead of the pre-calculated uncertainties, we will refer to this quantity as observationally corrected S/N (o.c. S/N). Interestingly, we find that for the planets with an o.c. S/N below 20, the ADI index is not correlated to the o.c. S/N (Figure 5). In this regime, we can find planets that scored highly on paper in terms of potential detections of atmospheric features but turned out to be difficult to interpret (e.g., WASP-101 b), and planets that appeared relatively challenging to observe on paper but delivered very solid detections (e.g., HAT-P-11 b). This absence of predictability showcases the need for exploratory observations prior to major time investments with large-scale facilities such as the JWST.

Zoom In Zoom Out Reset image size

Considering the warm and hot Jupiters in our sample (M > 0.16 M_Jup, i.e., excluding the Neptunes: GJ 436 b, GJ 3470 b, HAT-P-11 b and HAT-P-26 b), the Pearson correlation coefficient indicates that the ADI is more strongly correlated with the planetary radius (0.51, p-value = 0.7%) than the planetary temperature (0.43, p-value = 3%) but not correlated with the surface gravity (−0.28, p-value = 16%) or the planetary mass (0.20, p-value = 32%). These parameters are plotted against ADI in Figures 6–9. These results indicate that planetary surface gravity is a secondary factor in identifying inflated atmospheres (Laughlin et al. 2011; Spiegel & Burrows 2013; Weiss et al. 2013).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Very hot and highly irradiated planets, with atmospheric temperatures above 1800 K feature high ADI atmospheres. Our quantitative retrievals suggest that the cloud-top pressures in these planets are significantly high, meaning clouds are deep in the atmosphere, if present at all (Table 3), while retrieved water abundances are constant within the errors. While we cannot determine the absolute atmospheric water abundances, given the relative narrow wavelength range probed, we can exclude scenarios where water is significantly destroyed or depleted in the upper atmospheres of irradiated and inflated hot Jupiters. In addition, the spectra of HAT-P-41 b, WASP-12 b, and WASP-121 b show no contribution from photochemical hazes (Zahnle et al. 2009; Kopparapu et al. 2012; Miller-Ricci Kempton et al. 2012). We can conclude that planets with temperatures higher than 1800 K feature clear atmospheres in the terminator regions at HST/WFC3 wavelengths.

In our retrievals, we considered a mixture of opaque cloud-deck and hazes, all planets but WASP-69 b are consistent with a gray, opaque cloud-deck. In this study, both opaque clouds and hazes were uniformly distributed along the terminator. Line & Parmentier (2016) showed that non-uniform cloud coverage can mimic high-molecular weight (hmw) atmospheres. While hmw atmospheres are not observed in our hot Jupiter retrievals, we note that HST/WFC3 data alone is not sufficient to differentiate between hmw, and low-molecular weight atmospheres with patchy cloud coverage (Line & Parmentier 2016). This is particularly relevant for the warm Neptune HAT-P-11 b, where a hmw atmosphere was postulated by Fraine et al. (2014). Asymmetric cloud coverage can be observed in ingress/egress signatures of the light curves (Line & Parmentier 2016; von Paris et al. 2016) but the incomplete phase coverage of the HST/WFC3 data is insufficient to confirm or reject patchy cloud coverage models.

The 16 spectra that show statistically significant atmospheres presented here are well described with a combination of gray-clouds, extended, particulate Rayleigh curves and water. Two notable exceptions are WASP-76 b (see Figures 10 and 11) and WASP-121 b. Both planets are hot Jupiters with equilibrium temperatures of ∼2000 K. The retrieval results show that the atmosphere is haze free (i.e., clear), and TiO, VO, and H₂O opacities determine the observed spectral shape. The TiO and VO model is favored with a Bayes Factor of 8.52 (4.44σ significance) when compared to a pure-water and haze dominated atmosphere for WASP-76 b. However, we would like to caution the reader that correlations between H₂O, TiO, and VO abundance, planet radius and cloud-top pressure exist in the retrieved posterior distributions. The retrieval features a high-H₂O (∼10^−2.0) and high-TiO (∼10^−2.5) mode, which is likely unphysical. More observations, in particular in the optical wavelengths, are required to fully distinguish between a TiO/VO abundant and high-altitude haze model. In the case of WASP-121 b, we find both models to be statistically indistinguishable from each other. As discussed above, in this analysis, we do not take into account effects due to patchy or non-uniform cloud covers (e.g., MacDonald & Madhusudhan 2017). In particular, Kempton et al. (2017) show that non-uniform clouds/hazes on WASP-121 b can cause observable spectral gradients in the HST/WFC3 wavelengths.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The sparse sampling of HST/WFC3 data and the short wavelength ranges do not allow us to conclusively exclude atmospheric haze models for these planets at this stage, though we note that the particulate extended Rayleigh curve would be unusually strong. Observations at longer wavelength ranges are required to conclusively determine the absolute abundance of molecular tracers.

The remaining 14 spectra without a statistically significant atmosphere can be explained by either opaque, high-altitude clouds or low water abundances, as no-atmosphere models are unlikely for gas-giant planets. Given the uncertainties in the observed spectra, we are sensitive to water mixing ratio higher than 10⁻⁸, for cloud-free atmospheres. We also note that combinations of water depletion and high-altitude clouds cannot be ruled out. Current space and ground-based data cannot constrain absolute abundances of trace gases beyond their detection. Future instrumentation such as the JWST or dedicated space missions probing a broader wavelength range will be able to break these degeneracies.

The spectra of 12 out of the 30 planets in our sample have been previously studied. These planets are: GJ 436 b (Knutson et al. 2014a), HAT-P-1 b (Wakeford et al. 2013), HAT-P-11 b (Fraine et al. 2014), HAT-P-32 b (Damiano et al. 2017), HD 209458 b (Deming et al. 2013), HD 189733 b (McCullough et al. 2014), WASP-12 b (Kreidberg et al. 2015), WASP-31 b (Sing et al. 2015), WASP-43 b (Kreidberg et al. 2014a), WASP-101 b (Wakeford et al. 2017), WASP-121 b (Evans et al. 2016) and XO-1 b (Deming et al. 2013). Figure 12 shows a comparison between the extracted spectra here and in the literature. The only noticeable difference is HD 209458 b, which we believe is due to the different calibration method used (Tsiaras et al. 2016b). We plan to further investigate this behavior is a future study. Concerning the detection of water vapour and other molecules (TiO, VO) and clouds, our results are consistent with previous results in the literature.

Zoom In Zoom Out Reset image size