Transmission Spectroscopy of WASP-79b from 0.6 to 5.0 μm

We analyzed WASP-79b Wide Field Camera 3 (WFC3) data from the Panchromatic Exoplanet Treasury (PanCET) program (HST GO-14767: PI: Sing & López-Morales). During its primary transit in March of 2017, HST observed WASP-79b in spatial scan mode, which slews the telescope during the exposure and moves the spectrum perpendicularly to the dispersion direction on the detector (Kreidberg et al. 2014). This mode allows for longer integration times by distributing the incoming energy over multiple pixels. The WFC3 instrument utilized its G141 GRISM to acquire spectra from 1.1 to 1.7 μm over five HST orbits, during which we collected 65 science frames using 138 s integrations. We provide an overview of the data analysis process below, and a detailed description of the process can be found in Stevenson et al. (2014).

The Transit Reduction, Extraction, and Calibration Software (T-RECS) pipeline produces multiwavelength, systematics-corrected light curves from which we derive wavelength-dependent transit depths with uncertainties (Stevenson et al. 2014). The bias correction is performed using a series of bias frames stacked to form a single master bias frame that is applied uniformly to all of the science frames. We extract a pixel window centered on the spectrum that includes pixels along the spatial direction that are used in the optimal spectral extraction as well as in the background subtraction (Stevenson et al. 2014). We modeled the spectroscopic flat field using the coefficients provided in the updated flat-field file sedFFcube-both.fits.

Because the background for HST is consistent over time, areas outside of the spectrum can be used to interpolate the background values for the region within the spectrum by computing the median of each column. We perform 5σ outlier detection by stacking the images in time and evaluating each pixel along the time axis for outliers. To account for imprecision in the instrument pointing during data collection, each spectrum is cross-correlated with the first spectrum to measure and correct for the pointing drift over time (Stevenson et al. 2014).

The raw transit light curves for WASP-79b exhibit ramp-like systematics comparable to those seen in previous WFC3 data. Following standard procedure for HST transit light curves, we did not include data from the first orbit in our analysis (Kreidberg et al. 2014). We corrected for systematics in the remaining orbits by modeling the systematics as a function of time, which includes an exponential ramp term fitted to each orbit, a linear trend term, and a quadratic term for limb darkening.

We modeled the band-integrated light curve in order to identify and remove systematics, most of which are wavelength-independent with WFC3, and to establish the absolute transit depth when comparing transmission spectra from different instruments using non-overlapping wavelengths (Stevenson et al. 2014). We created this white light curve (WLC) by summing the flux values over the entire wavelength range. We used the Bayesian Information Criterion (BIC) to select the best systematics model component, and our final analytic model for the HST/WFC3 data takes the form

$$\begin{eqnarray}&&F(t)={F}_{s}T(t)L(t)H(t)\end{eqnarray} \tag{ 1 }$$

where F(t) is the measured flux at time t, F_s is the out-of-transit system flux, T(t) is the primary-transit model component with unity out-of-transit flux (Mandel & Agol 2002), $L(t)=a(t-{t}_{0})+1$ is the time-dependent linear model component with a fixed offset, t₀, and free parameter, a, and $H(t)=1-\exp (-a\times P+b)+c\times P$ fits the HST hook using a rising exponential with free parameters a, b, and c, where P represent the number of HST orbits since the beginning of the transit. The WLC extraction for the HST/WFC3 data resulted in a transit depth of 1.1282% ± 0.0032% (see Figure 2).

Zoom In Zoom Out Reset image size

We use the Divide–White method described by Stevenson et al. (2014) to model the wavelength-dependent (i.e., spectroscopic) light curves, without making any prior assumptions about the form of the systematics, by utilizing information within the wavelength-independent (white) light curves. This can be done for an arbitrary number of wavelength bins, though 10 to 15 bins provide sufficient resolution to reveal features of interest while maintaining sufficient signal to noise in each bin.

To construct a spectrum, we are interested only in the relative transit depths of the different wavelength bins. We can therefore estimate uncertainties with our differential evolution Markov Chain Monte Carlo (DE-MCMC) algorithm, assuming fixed parameters for a/R* and cos i (Stevenson et al. 2014). For the HST, LDSS-3C (Section 2.3), and Spitzer (Section 2.4), we assumed a fixed a/R* of 7.2900 and a cosi of 0.070993, based on an analysis of the TESS data for WASP-79b (Section 2.1). The transit midpoint was carried as a free parameter and estimated in the WLC analyses and then fixed for the spectroscopic analyses, as it is wavelength-independent. Figure 2 shows results for the HST WLC extraction as well as results for the 15 wavelength bins from the spectroscopic light curve extraction, and Figure 3 shows the associated 1σ residuals.

Zoom In Zoom Out Reset image size

The results of the HST/WFC3 analysis, which indicate the presence of water in WASP-79b's atmosphere, are discussed in later sections.

We observed the primary transit of WASP-79b on the night of 2016 December 20 for nearly 8 hr (00:31-08:14 UT, airmass = (1.1–1.0–1.8) (4)), collecting 1230 science frames using 7 s integrations. We utilized LDSS-3C's turbo read mode with low gain and applied 2×2 pixel binning to minimize readout times, overall achieving a duty cycle of 31%. The most recent upgrade of the instrument to LDSS-3C constituted an upgrade to a deep-well detector that eliminated the fringing issues seen previously (Stevenson et al. 2016b).

Our science masks utilized three, 12''-wide slits for observations of our target star (WASP-79, V = 10.1) and the two comparison stars (V = 10.8, 12.7). The brighter comparison star is a G dwarf star with a T_eff of 5834 K. The spectra from the dimmer comparison star were too noisy to provide reliable atmospheric corrections, so we relied strictly on the brighter reference star. Unfortunately, the brighter reference star was sufficiently displaced from the target star on the detector (1467) that the resulting atmospheric corrections are not necessarily consistent. This results in relatively large error bars on the transit depth estimates for the LDSS-3C data.

As described in Stevenson et al. (2016a), we correct for the observed flux variations caused by fluctuations in Earth's atmosphere by dividing the WASP-79b light curve by the comparison star. We start by fitting the WLC (0.625–1.025 μm) to maximize the signal-to-noise ratio (S/N), using both transit and systematics model components. The first utilizes a Mandel & Agol (2002) transit model with selected free parameters and fixed quadratic limb-darkening parameters derived from stellar Kurucz models (Castelli & Kurucz 2004) assuming a stellar temperature of 6500 K and log g of 4.2. We found early in the analysis that there was a shift of the illuminated pixels on the detector in the middle of the transit that was caused by the telescope rotating as it passed through zenith (Figure 4). For the systematics component, we tested various combinations of linear and quadratic models in combination with rotation and intrapixel functions to account for the aforementioned rotation and pixel shift to determine which combination of models provided the best fit, based on the BIC and χ² values. Our final analytical model takes the form

$$\begin{eqnarray}&&F(t)={F}_{s}T(t)R(t)Q(t)I(t)\end{eqnarray} \tag{ 2 }$$

where F(t) is the measured flux at time t, F_s is the out-of-transit system flux, T(t) is the primary-transit model component with unity out-of-transit flux, $R(t)=1+{aA}+b\cos$(π/180 × (θ(t) + θ₀)) is the time-dependent instrument rotation model component with free parameters a, b, and θ₀, where A = airmass, Q(t) uses a quadratic polynomial to fit a pixel response ramp in the data, and I(y) fits the pixel shift using a linear function in the dispersion direction. The WLC for the 2016 December LDSS-3C data resulted in a transit depth of 1.1626% ± 0.0152%.

Zoom In Zoom Out Reset image size

As with the HST/WFC3 data, we apply the Divide–White technique (Stevenson et al. 2014) to remove the wavelength-independent systematics. To account for the wavelength-dependent systematics, each spectroscopic channel requires a rotation correction with airmass, a quadratic function in time, and an intrapixel response shift correction. Due to unfavorable weather effects during the night of the LDSS-3C observation, the displaced reference star, and the telescope rotation, we found the data to be very noisy with significant numbers of outliers in most channels. To remove these outliers, we performed the following iterative outlier rejection process.

• 1.  
  We ran the simulation with no masking or outlier rejection so that we could visually determine whether there were any sections of the data that should be removed entirely. Based on the results of this run and the weather information for the observation timeframe, we removed times 02:16:14 UT–02:47:25 UT and times 03:24:58 UT–04:11:56 UT for all channels. Additionally, based on the normalized flux values (see Figure 5), the 6540–6590 and 7570–7700 channels were masked to remove them from the light curve analysis, as they showed atmospheric absorption that could not be accounted for using the reference star, which was artificially increasing the transit depths in those channels. There were significant changes in the local humidity over the course of the night, particularly between ∼05:00 UTC and ∼08:00 UTC that may have contributed to the noise in the data.
• 2.  
  We then ran two consecutive boxcar median masks with 3σ rejection on the photon flux data.
• 3.  
  We re-ran the simulation on the results from step 2, and ran three consecutive 3σ outlier rejection masks on the residuals for the resulting transit models.
• 4.  
  Finally, we re-ran the simulation on the results from step 3, masking the outliers identified in steps 2 and 3.

Zoom In Zoom Out Reset image size

In addition to the expected drift in the dispersion direction of the LDSS-3C spectrum over the course of the observation, Diamond-Lowe et al. (2018) found a stretching of the spectrum equal to approximately 4 pixels for the target star and 2 pixels for the comparison star. To account for this effect, we calculated the stretch and the drift by optimizing a cubic spline fit of the target spectrum normalized to the reference spectrum. Figure 6 shows the calculated spectrum drift and stretch over time for both the target and reference stars, and it can be seen that the spectral drift was in excess of 1 pixel for both the target and reference stars.

Zoom In Zoom Out Reset image size

The results of the spectroscopic light curve extraction for the LDSS-3C data are shown in Figure 7, and Figure 8 shows the residuals for each waveband. Due to the large amount of noise in the data, we restricted the spectroscopic LDSS-3C analysis to eight channels to increase the S/N.

Zoom In Zoom Out Reset image size

Because the opacity of the exoplanet atmosphere varies with wavelength, the apparent size of the planet, and therefore the depth of the transit, also vary with wavelength. Having performed the spectroscopic light curve extraction and the systematics normalization via the Divide–White method, we can construct a spectrum from the relative transit depths of the selected wavelength bins.

Zoom In Zoom Out Reset image size

Figure 9 shows the relative transit depths of the WASP-79b HST data for 15 wavelength bins for the light curve extraction using the Divide–White normalization method. In this figure, the positive y-axis represents increasing transit depth, i.e., more absorption by the WASP-79b atmosphere. The resulting spectrum displays a noticeable peak centered at 1.4 μm, which represents a water feature. This feature is consistent with water features found in the spectra of other hot Jupiters (Sing et al. 2016), and an atmospheric retrieval corroborates this feature.

Zoom In Zoom Out Reset image size

Figure 10 shows the relative transit depths of the WASP-79b LDSS-3C data for eight wavelength bins for the light curve extraction using the Divide–White normalization method. It should be noted that the transit depth estimate for the 0.65 μm channels is likely somewhat low due to detector cutoff at the blue edge. Rackham et al. (2017) also found decreased transit depths at bluer wavelengths for GJ 1214b, a sub-Neptune orbiting a M4.5 dwarf star, which they attribute to the presence of faculae on the unocculted stellar disk. However, observations of WASP-79 indicate that its stellar activity is low. We collected XMM-Newton observations of WASP-79 on 2017 July 18, with S/N = 3.4. Its X-ray emission, L_X = 5.7 × 10²⁸ ergs/s (for a d = 248 pc; see Gaia DR2), yields a ratio of log L_X/L_bol = −5.5, indicating a low activity level, as expected for an early F star (J. Sanz-Forcada et al. 2019, in preparation.). The TESS data baseline varies within 1σ < 0.1%, so these data do not show evidence of short-term stellar activity variations in WASP-79. Furthermore, photometric observations of WASP-79 with the Tennessee State University C14 Automated Imaging Telescope (AIT) at Fairborn Observatory (see, e.g., Sing et al. 2015 for a description of AIT operations) show no significant brightness variability within the years 2017, 2018, and 2019. Nor does AIT see significant variability from year to year over the same interval to a limit of ∼0.005 mag, confirming the absence of longer-term activity variations. The photometric stability of WASP-79 suggests that the decreased transit depth at shorter wavelengths is not likely to be due to inhomogeneities in the stellar photosphere (Rackham et al. 2019). Given the low resolution of the LDSS it is not obvious what is causing the positive slope in the spectrum at bluer wavelengths.

Zoom In Zoom Out Reset image size

As discussed in Section 2.3.1, the atmospheric corrections likely do not fully account for the atmospheric dynamics during the observation, and the very deep transit depth at 0.95 μm is likely exaggerated by interference from H₂O in Earth's atmosphere. To account for red noise in the data, the uncertainties in the LDSS transit depth estimates are multiplied by the maximum correlated noise factor for each light curve.

Transit data for WASP-79b from the HST Space Telescope Imaging Spectrograph (STIS) instrument are currently being analyzed. STIS provides data from 0.3 to 1.0 μm, and these data should have smaller uncertainties than the LDSS-3C data, providing more insight into the atmospheric structure of this hot Jupiter.

The observations analyzed here are part of Program ID 13044 (PI: Drake Deming). The target was observed during transit with IRAC channel 1 (3.6 μm) and channel 2 (4.5 μm) (Fazio et al. 2004). The astronomical observing requests (AOR) are 62173184 and 62173696 for channels 1 and 2 respectively. All of these observations were carried out in subarray mode (32 × 32 pixels, 39'' × 39'') with a 30 minute peak-up observation preceding them. The use of a peak-up observation allows the instrument to stabilize the image on the detector "sweet spot" and decreases the likelihood of a ramp in the data (Ingalls et al. 2012). The frame time for both observations was 2 s.

For each AOR we began with basic calibrated data (BCD) available on the Spitzer Heritage Archive. Each BCD file contains a cube of 64 frames of 64 × 64 pixels. Each set of 64 images comes as a single FITS file with a time stamp corresponding to the start of the first image. We determine the time of each frame in the set by adding the appropriate multiple of the frame time to the time stamp of the first image. The photometric extraction was performed following the methods detailed in Kilpatrick et al. (2017, 2019) utilizing both fixed and variable apertures across a range of sizes. Background subtraction and determination of the stellar centroid and noise pixel parameter were performed in each case.

Each transit fit was based on the model of Mandel & Agol (2002) implemented in Python by the BATMAN package (Kreidberg et al. 2015). We assumed an orbital eccentricity of zero and used the a/R* and cosi values derived from the TESS data from sectors 4 and 5. Stellar limb-darkening parameters were derived from ATLAS models and interpolated bi-linearly from tables presented in Sing (2010). We choose to use the quadratic form and fix coefficients to [0.04735, 0.15251] and [0.0604, 0.11834] for channels 1 and 2 respectively. The intrapixel sensitivity variation (Ingalls et al. 2012), the change in measured flux as a function of stellar centroid position, and methods of correction, are well documented (e.g., Ingalls et al. 2016). Here, we employ the nearest neighbors method (NNBR), otherwise known as Gaussian kernel regression with data (Lewis et al. 2013; Kilpatrick et al. 2017).

For each AOR, the best-fit values for all free parameters were initially determined using matrix inversion. The standard deviation of the normalized residuals (SDNR) times the β_red factor (Gillon et al. 2010) was used as a metric for selecting the best fit out of the multiple apertures. The results from the best-fit aperture were passed to a MCMC implemented by emcee (Foreman-Mackey et al. 2013) to derive uncertainties of each free parameter. The uncertainty on each data point in the light curve is inflated by the β_red factor to account for the unresolved correlated noise. We use a number of walkers at least twice the number of free parameters and run for 10⁵ steps per walker before testing for convergence using Gelman Rubin statistics with a threshold for acceptance of 1.01 (Gelman & Rubin 1992). The initial 10% of steps for each walker are discarded to remove the burn-in period.

At 4.5 μm we find a transit depth of 1.1396% ± 0.0103%. The SDNR of this observation was 0.04875 with a β_red factor of 1.09. At 3.6 μm we find a transit depth of 1.1224% ± 0.0080% with an SDNR of 0.005505 and β_red factor of 1.41. We find the center of transit time to occur 0.009835 ± 0.0008 days (14.15 ± 1.15 minutes) later than the predicted transit time (Smalley et al. 2012) in channel 1 and 0.009743 ± 0.00035 days (14.0 ± 0.5 minutes) in channel 2.

Table 1 provides the wavebands, normalized transit depths, and 1σ transit depth uncertainties for the previously described data sets. Table 2 provides the transit ephemerides and uncertainties for the TESS, Spitzer, HST/WFC3, and LDSS-3C observations. We used these transit times in conjunction with the Smalley et al. (2012) ephemeris to re-compute a new ephemeris and period for WASP-79b. Figure 11 shows the observed-computed (O-C) transit times and uncertainties for the values in Table 2 compared to the new ephemeris.

Table 1.  Normalized Transit Depths and Uncertainties

Instrument	Waveband	(Rp/R*)2	${\sigma }_{{({R}_{p}/{R}_{* })}^{2}}$
	(μm)
TESS	0.586–1.031	1.1396	0.014
	0.625–0.67	1.0725	0.0316
	0.675–0.725	1.0955	0.0206
	0.725–0.757	1.1026	0.0101
LDSS-3C	0.770–0.825	1.1175	0.0073
	0.825–0.875	1.1209	0.0204
	0.875–0.925	1.1610	0.0205
	0.925–0.975	1.2071	0.0215
	0.975–1.025	1.1282	0.0332
	1.125–1.160	1.1486	0.0050
	1.160–1.195	1.1514	0.0053
	1.195–1.230	1.1398	0.0051
	1.230–1.265	1.1395	0.0047
	1.265–1.300	1.1385	0.0061
	1.300–1.335	1.1431	0.0052
	1.335–1.370	1.1418	0.0051
HST/WFC3	1.370–1.405	1.1634	0.0053
	1.405–1.440	1.1524	0.0051
	1.440–1.475	1.1533	0.0061
	1.475–1.510	1.1532	0.0053
	1.510–1.545	1.1412	0.0054
	1.545–1.580	1.1420	0.0065
	1.580–1.615	1.1287	0.0056
	1.615–1.650	1.1201	0.0072
Spitzer	3.18–3.94	1.1224	0.0080
	3.94–5.06	1.1396	0.0103

Download table as:  ASCIITypeset image