Identifying Candidate Atmospheres on Rocky M Dwarf Planets via Eclipse Photometry

The ability of rocky planets orbiting M dwarfs to form and retain atmospheres is a major question in the field of exoplanets because of the forthcoming opportunity to observe these worlds for evidence of habitability and life (Shields et al. 2016; National Academies of Sciences, Engineering, and Medicine 2018). M dwarfs undergo a long pre-main-sequence phase that exposes planets that would later be in the nominal liquid water habitable zone to strong irradiation (Chabrier & Baraffe 2000) as well as high EUV and solar wind fluxes (Dong et al. 2018). Even once on the main sequence, M dwarfs can still exhibit strong flaring events (e.g., Davenport et al. 2012) and the ratio of their high energy luminosity to bolometric luminosity is substantially larger than for Sun-like stars (e.g., Ribas et al. 2017; Peacock et al. 2019). Atmospheric escape on planets orbiting M dwarfs could therefore be extremely high and sustained, raising the possibility that the worlds orbiting in these stars' habitable zones might be predominantly bare rocks with little chance of hosting a surface biosphere (Zahnle & Catling 2017, and references therein).

Although volatile loss could be prevalent on M dwarf planets, there are also reasons to be hopeful about the presence of atmospheres on these worlds. These planets might accumulate massive atmospheres in the first place (e.g., Barnes et al. 2016; Ribas et al. 2016), could have magnetic fields that would guard against some loss mechanisms (e.g., Segura et al. 2010), could outgas secondary atmospheres from their interiors, could have atmospheres with high mean molecular weight gases and thermospheric coolants that suppress atmospheric escape, or could be resupplied with volatiles from an external source, such as through cometary bombardment. The relative efficiency of these processes remains highly uncertain, however, so the final say on whether atmospheres are common on rocky planets around M dwarfs will have to be obtained empirically.

The three main techniques for detecting atmospheres on exoplanets are transmission spectroscopy during a planet's transit (transit spectroscopy), emission spectroscopy during a planet's secondary eclipse (eclipse spectroscopy), and thermal phase curves over the course of a planet's orbit. Transit and eclipse spectroscopy have been discussed extensively elsewhere (Miller-Ricci et al. 2009; Bean et al. 2010; Barstow & Irwin 2016; Morley et al. 2017; Batalha et al. 2018; Louie et al. 2018; Lustig-Yaeger et al. 2019). These techniques rely on inferring the spectral signature of atmospheric gases. For a transit that means ruling out a flat transmission spectrum. For an eclipse that means ruling out a blackbody spectrum and detecting spectral features that are consistent with gas phase molecules in the planet's atmosphere (although in practice the interpretation can be subtle, see Section 3). Thermal phase curves as a means of detecting atmospheres were proposed by Seager & Deming (2009). This technique relies on the signature of an atmosphere's heat redistribution. As long as the planet can be assumed to be tidally locked into synchronous rotation with permanent day- and nightsides, a bare rock planet would exhibit a large day–night temperature difference in its thermal phase curve, whereas an atmosphere would tend to reduce this temperature difference (Seager & Deming 2009; Selsis et al. 2011; Koll & Abbot 2016; Kreidberg & Loeb 2016).

Unfortunately, all three techniques will likely require substantial investments of observing time (Deming et al. 2009; Kaltenegger & Traub 2009; Rauer et al. 2011; Snellen et al. 2013; Rodler & López-Morales 2014; Serindag & Snellen 2019). For transit and eclipse spectroscopy, estimates suggest that atmospheric detection will require anywhere from multiple to more than a dozen repeat observations with the James Webb Space Telescope (JWST; Morley et al. 2017; Batalha et al. 2018; Louie et al. 2018). Phase curves are inherently costly, because they need to span at least half a planet's orbit, and in some cases the observation might have to be repeated to attain the desired signal-to-noise, thus also requiring long observation periods.

Finally, observations would ideally not just detect the presence of an atmosphere but characterize it in detail. Doing so will be even more expensive than the above estimates suggest because any single technique suffers from a number of degeneracies and false positive scenarios. For example, transit spectroscopy can be limited by the presence of hazes and clouds (e.g., Kreidberg et al. 2014; Sing et al. 2016); determining composition from eclipse spectroscopy requires simultaneously determining the atmosphere's thermal structure (e.g., Madhusudhan & Seager 2009; Line et al. 2016); and inferring an atmosphere's thickness from its thermal phase curve requires simultaneous knowledge about its composition (Koll & Abbot 2015). Any effort to move beyond atmospheric detection to detailed characterization will thus likely have to combine multiple techniques, increasing the observational effort even more.

Given how difficult it is to detect and characterize atmospheres on small exoplanets, a fast screening technique is needed to identify those planets that are most promising for follow-up campaigns. An efficient test for the presence or absence of an atmosphere will also enable exploration of a larger number of planets than can be studied in detail, which is crucial for developing statistical insight into the formation and evolution of planetary atmospheres.

Here we propose such a test, by considering how the planet's atmospheric heat transport affects its dayside thermal emission, which can be measured through its broadband secondary eclipse depth. Our proposal is similar to that of Seager & Deming (2009), but we focus solely on the observable dayside signature.

The energy budget of the planet's dayside can be written as (Burrows 2014)

$$\begin{eqnarray}&&{T}_{\mathrm{day}}={T}_{* }\sqrt{\displaystyle \frac{{R}_{* }}{d}}{\left(1-{\alpha }_{B}\right)}^{1/4}{f}^{1/4}.\end{eqnarray} \tag{ 1 }$$

Here T_day is the observed dayside brightness temperature, T_* is the stellar temperature, R_* is the stellar radius, d is the planet's semimajor axis, α_B is the planet's Bond albedo, and f is the so-called heat redistribution factor. There are two limits for f:

$$\begin{eqnarray}f=\left\{\begin{array}{ll}2/3 & \ \mathrm{instant}\ \mathrm{reradiation}\\ 1/4 & \ \mathrm{uniform}\ \mathrm{redistribution}\end{array}\right..\end{eqnarray} \tag{ 2 }$$

If a planet has no or a sufficiently thin atmosphere then we consider it effectively a bare rock and $f\to 2/3$ (Hansen 2008). Conversely, if the planet has a thick enough atmosphere that its winds redistribute heat between day- and nightside then f will be reduced. In the limit in which atmospheric heat transport becomes highly efficient $f\to 1/4$. Our proposal amounts to observing T_day, to infer whether f is significantly smaller than the bare rock limit.

The main promise of atmospheric detection via eclipse photometry is that it does not require high spectral resolution, so it should require less observation time than spectroscopy or phase curves. Of course, like every other technique, eclipse photometry also suffers from false negatives and false positives. For example, Equation (1) shows a degeneracy between f and the albedo α_B. Physically, this means a bare rock with high albedo can mimic the dayside thermal emission of a low-albedo planet with a thick atmosphere. As we discuss in detail in Section 6, we do not believe that this and other degeneracies will greatly affect our proposal, based on both physical modeling and the empirical observation that bare rocks in the solar system have low albedos.

Nevertheless, because false positives are possible and in analogy to the Kepler and TESS missions, we consider planets whose dayside brightness temperature strongly deviates from that of a bare rock as "candidate atmospheres," but whose atmospheric nature should be confirmed via follow-up.

The goal for the rest of this paper is to quantify how much time is required to infer an atmosphere via eclipse photometry, how this effort compares to the effort needed with other atmospheric detection techniques, and how many planets exist that could potentially be studied with this technique. To do so we use atmospheric models to simulate the atmospheres of three nearby rocky planets that are among the best known targets for atmospheric characterization: TRAPPIST-1b (Gillon et al. 2016; Delrez et al. 2018), GJ1132b (Berta-Thompson et al. 2015), and LHS3844b (Vanderspek et al. 2019). Common to all three is that they are too hot to be habitable, which makes them easier to characterize than habitable-zone planets and also decreases the likelihood of false positives for our proposed technique (see discussion).

Our models are described in Section 2. We use these models to generate simulated JWST observations and compare our results against previous work in Section 3. We then quantify the observational effort required for detecting an atmosphere using a wide range of techniques, which we present in Section 4. For all three modeled planets, we find that a single-eclipse observation with JWST will be able to confidently detect the atmospheric heat redistribution signal of a thick atmosphere. In contrast, most other techniques will require more observation time. Eclipse photometry is therefore a quick and viable way of inferring atmospheres on rocky exoplanets. In Section 5 we then estimate how many other rocky planets exist for which eclipse photometry might be feasible. We find that TESS should detect more than 100 rocky planets that this technique could be applied to, which opens up the possibility of a future statistical survey of atmospheres on rocky exoplanets. We discuss our results in Section 6 and conclude in Section 7.

We process our simulations to wavelength ranges that will be observable with JWST. For transit spectroscopy we consider NIRSpec/G235M between 1.66 and 3.07 μm. For eclipse spectroscopy we consider MIRI/LRS between 5 and 12 μm.

To evaluate whether JWST can detect an atmosphere we use a simple χ² metric. Qualitatively, if the reduced χ² value, ${\chi }_{\nu }^{2}={\chi }^{2}/\nu$, is much bigger than ${\chi }_{\nu }^{2}\sim 1+\sqrt{2/\nu }$, which is of order unity, we can reject the null hypothesis that the planet is a bare rock and consider this an atmospheric detection. Here ν is the number of degrees of freedom in a given spectrum. For example, with ν = 3, we might have some confidence in an atmospheric detection once ${\chi }_{\nu }^{2}\gg 1+\sqrt{2/3}=1.8$. More formally, the probability that ${\chi }_{\nu }^{2}\gt 2$ is 11% and the probability that ${\chi }_{\nu }^{2}\gt 3$ is 3%. The null hypothesis, and thus the χ² value, as well as ν have to be defined differently for each technique as follows.

For transit spectroscopy we compute χ² from the fit between the observed wavelength-dependent transit depth and a flat line, where the flat line is simply the average transit depth in the NIRSpec wavelength range. Because the flat line is derived from the observations, ν is equal to the number of observed data points minus one.

For eclipse spectroscopy we require observations to rule out a blackbody to count as an atmospheric detection. We note that this definition is susceptible to false positives: bare rocks can have spectral features (see discussion) and a planet's emission spectrum can be contaminated by reflected stellar light. For cool stars this could impart molecular features in emission that are due to molecules in the star's, not the planet's, atmosphere. There are also possible false negatives: an atmosphere with very thick clouds could hypothetically resemble a blackbody. Such an atmosphere would be undetectable via spectroscopy, and detection would instead need to rely on eclipse photometry or thermal phase curves. Our detection metric for eclipse spectroscopy is therefore overconfident, and in practice atmospheres might be more difficult to detect using this technique.

The temperature of the null hypothesis blackbody, which physically corresponds to the planet's dayside brightness temperature, is a priori unknown so we use the same two-step procedure as one would follow with actual observations. First, we fit a blackbody to the observed spectrum using scipy.optimize.curve_fit, optimizing for the blackbody's temperature. Second, we compute ${\chi }^{2}$ from the fit between the observed emission spectrum and the best-fit blackbody spectrum. Because one degree of freedom is used to fit the blackbody's temperature, ν is again equal to the number of observed data points minus one.

For eclipse photometry we compute χ² from the difference between the observed emission spectrum and a blackbody spectrum that assumes no heat redistribution. We assume that the surface albedo of the no-heat-redistribution blackbody is known and equal to 0.1 (see Section 6). Although we are computing a photometric signal, we use the same spectral resolution as for eclipse spectroscopy. In theory we could bin even further, to a single photometric data point, but in practice we find that increased binning leads to little increase in statistical significance for many cases. Because the eclipse depth of the no-heat-redistribution blackbody is defined independently of any observed data points, ν is equal to the number of observed data points.

For phase curves we compute χ² from the phase curve amplitude, i.e., the day–night emission difference. To do so we first generate a nightside emission spectrum by rescaling the emitted dayside flux via a spectrally uniform factor that depends on the atmosphere's heat redistribution, ${F}_{\mathrm{night}}=3/5\times (2/3-f)/f\times {F}_{\mathrm{day}}$. This expression guarantees the correct nightside fluxes in the thick and thin atmosphere limits. We then compare the phase amplitude (i.e., the day–night flux difference) of the planet with an atmosphere to the phase amplitude of a bare rock, and set ν equal to the number of observed data points. We note that phase curves contain additional information that can be used to infer the presence of an atmosphere, such as hot spot offsets. Here we only focus on the phase curve amplitude because global climate models suggest that hot spot offsets become negligible on rocky planets with relatively thin atmospheres (Koll & Abbot 2015, 2016).

We note that the χ² metric is overly conservative because it does not capture spectral correlations. For example, a high-resolution transit spectrum could have a small χ² value relative to a flat line, yet still show clear correlation between nearby points that are part of a spectral band. A full retrieval model would be able to detect this band structure, and thus infer an atmosphere, whereas a simple χ² test might miss it. To account for this effect we downsample all simulated spectra to low spectral resolution, so each spectral point corresponds to a single spectral band. To downsample we weight simulated data by the inverse variance at each wavelength, so points with smaller error bars contribute more to the spectral mean than points with larger error bars. In practice this mostly affects the MIRI/LRS bandpass, where detector efficiency as well as stellar photon count decrease notably between 5 and 12 μm. We select the low-resolution spectral bands by hand for each atmospheric composition. A retrieval algorithm would have to infer these bands from the data, so by giving ourselves this information we are increasing the likelihood of detecting a spectral signature (i.e., a real spectral retrieval would be less confident in detecting an atmosphere than our hand-tailored approach).

Figure 2 shows what our synthetic JWST transit spectra look like, with error bars representing a single transit. For both H₂O and CO₂ atmospheres we bin the transit down to just four spectral points, which capture the dominant bands and windows of each gas in the NIRSpec wavelength range. Figure 2 also shows the ${\chi }_{\nu }^{2}$ values relative to a flat line; for ease of viewing we only show ${\chi }_{\nu }^{2}$ for the simulations with 1 bar surface pressure. For H₂O atmospheres we find ${\chi }_{\nu }\gt 1$ in all cases. The best transit target is TRAPPIST-1b with χ_ν = 12.2 but even the worst target, GJ1132b, has χ_ν = 3.9. A single transit spectrum should thus be sufficient to infer an atmosphere. The ${\chi }_{\nu }^{2}$ values are smaller for CO₂, due to CO₂'s higher mean-molecular-weight (MMW) and smaller scale height, but even here we find ${\chi }_{\nu }=3.0$ for TRAPPIST-1b and ${\chi }_{\nu }=2.6$ for LHS3844b. The only scenario in which a single transit is not sufficient to detect an atmosphere is GJ1132b with a CO₂ atmosphere for which ${\chi }_{\nu }^{2}=0.9$. We note, however, that the detectability of these transit spectra is optimistic because they do not include any atmospheric aerosols. We explore the possible impact of clouds and hazes on transit spectroscopy in the next section.

Zoom In Zoom Out Reset image size

Figure 3 shows what our synthetic JWST emission spectra look like, with error bars representing a single eclipse. For H₂O we bin the data down to just two spectral points in the MIRI/LRS bandpass, for CO₂ we use five spectral points. Figure 3 shows the ${\chi }_{\nu }^{2}$ values for an observed emission spectrum relative to a best-fit blackbody and relative to a bare rock blackbody. We find that it is difficult to detect spectral features via eclipse spectroscopy for cool planets, while warm planets are feasible targets. For example, we find ${\chi }_{\nu }^{2}=0.3$ for TRAPPIST-1b with an H₂O atmosphere whereas ${\chi }_{\nu }^{2}=3.5$ for LHS3844b with a CO₂ atmosphere.

Zoom In Zoom Out Reset image size

In contrast to eclipse spectroscopy, we find that eclipse photometry can detect atmospheres with heat redistribution in the vast majority of cases. Even for a cool planet like TRAPPIST-1b with a H₂O atmosphere, we find that 1 bar of atmosphere leads to a notable difference between the dayside's broadband emission and a bare rock's (χ_ν = 13.7). The only case in which a single eclipse is not sufficient for a confident detection is TRAPPIST-1b with a CO₂ atmosphere, for which χ_ν = 1.3. The ease of detection via eclipse photometry strongly increases with temperature, and LHS3844b with 1 bar of CO₂ deviates very strongly from a bare rock (χ_ν = 10.3).

Figure 4 illustrates how we compute the JWST phase curve signal. Here we show LHS3844b with a CO₂ atmosphere. First, we rescale the simulated dayside to get a nightside emission spectrum. We then compute the day–night flux difference, and compare this difference to the day–night difference of a bare rock. The error bars are the same as in Figure 3, which amounts to binning the observed phase curve into bins of 31 minutes (the duration of a transit or eclipse for LHS3844b). We find that, similar to eclipse photometry, phase curves should be able to infer thick atmospheres with high confidence and a 1 bar atmosphere on LHS3844b would be ruled out with χ_ν = 30. The high confidence of this atmospheric detection, however, has to be weighed against its (potentially high) observational cost, which we consider in the next section.

Zoom In Zoom Out Reset image size

Our results qualitatively agree with previous JWST calculations, even though we employ a number of different modeling assumptions and we simulate different instrument modes.

Batalha et al. (2018) computed signal-to-noise (S/N) for transit observations of cool habitable-zone planets and found that about 10 repeated transits with the NIRSpec Prism mode are needed to detect spectral features. This number is much larger than the single transit we find here, but Batalha et al. (2018) focused on cooler planets and included high-altitude clouds in their calculations. Indeed, as we show in the next section, clouds and hazes greatly increase the observational effort to detect an atmosphere via transit spectroscopy.

Louie et al. (2018) computed S/N for transit observations of warm super-Earths with H₂O atmospheres using the NIRISS instrument and found that 10 hr of telescope time (about 2–3 transits) are sufficient to detect spectral features on TRAPPIST-1b and GJ1132b with a high S/N of about 20–40. These numbers suggest that a single transit would be sufficient for atmospheric detection, in agreement with our results. Moreover, our χ² calculation is similar to their S/N metric, and we find that the two metrics agree to within a factor of 3 or better once we account for the different observation lengths.

Mollière et al. (2017) simulated transit spectra of GJ 1214b with a cloudy, relatively high mean molecular weight atmosphere and found that about 10 transits with NIRSpec could rule out a flat line with 95% probability. Although their result is again strongly affected by clouds and hazes and considered a different planet, it is comparable to an order of magnitude with estimates we present in the next section for how clouds and hazes can impact transmission spectroscopy.

Morley et al. (2017) estimated the amount of time required to characterize an atmosphere for both transit and eclipse spectroscopy using the same set of instrument modes as we do. They found that for a favorable transit target like TRAPPIST-1b with a CO₂ atmosphere, about six transits are needed to rule out a flat line at 5σ confidence, while other targets would require longer observations. Similarly, for a favorable emission target like GJ1132b, about 2–3 eclipses are needed to detect the secondary eclipse at 25σ confidence. We will show in the next section that these results are comparable to our own calculations, even though our estimates are slightly more optimistic. For example, we estimate that a 5σ detection of CO₂ on TRAPPIST-1b will require about four transits, compared to Morley et al.'s six transits.

We note that even though our detectability estimates are comparable to those of previous groups, we use different physical assumptions. For example, we simulate temperature profiles in self-consistent radiative–convective equilibrium. In contrast, Batalha et al. (2018) assumed a parameterized temperature profile that was based on analytic gray calculations, and Morley et al. (2017) assumed that all atmospheres are convective up to a pressure of 0.1 bar and are capped by an isothermal stratosphere whose temperature is equal to the skin temperature.

The different assumptions about temperature profiles should only have a small effect on transit spectra, but they will affect emission spectra. We find that convection is generally suppressed due to the red host star spectra and atmospheric shortwave absorption (Figure 1).

The clear majority of our simulations also do not show a transition between convective and radiative zones at around 0.1 bar, which has been proposed based on solar system atmospheres (Robinson & Catling 2014). This means vertical temperature gradients on M dwarf planets should generally be smaller, and signals for emission spectra lower, than one might predict with parameterized convective temperature profiles.

Bare rocks with high Bond albedos are an important false positive scenario for our proposal because, just like a thick atmosphere, a high albedo would also reduce a planet's dayside thermal emission (see Equation (1)).

We consider this false positive scenario unlikely. First, in a companion paper we compile geological and laboratory evidence which suggests that the surface albedos of rocky exoplanets with equilibrium temperatures in the range of the planets we consider here, 300 K < T_eq < 880 K, should be low (Mansfield et al. 2019). The underlying reason is that many geologic surfaces with high albedo (e.g., granites, clays) are either water-rich or require liquid water to form, which is unlikely for planets with T_eq > 300 K because these planets are located inside the inner edge of the M dwarf habitable zone. At the same time, planets with T_eq > 880 K are hot enough to vaporize substantial amounts of rock on their dayside over geologic timescales. This partial vaporization would preferentially remove more volatile species, and so could leave behind a low-volatile residue rich in aluminum and calcium compounds that have high albedos. By focusing on planets with 300 K ≤ T_eq ≤ 880 K, we minimize the possibility of either false positive scenario occurring.

Second, solar system analogs similarly suggest that exoplanets without atmospheres will have low albedos (Madden & Kaltenegger 2018). Notable bare rocks in the solar system include Mercury, which has an albedo of less than 0.1, the Moon and Ceres, which have albedos of 0.1–0.15, and asteroids, the majority of which have an albedo less than 0.2 (Wright et al. 2016). There are some airless bodies in the solar system with high albedos, such as Europa with an albedo of ∼0.6, and Jupiter's moon Io with an albedo of ∼0.5. However, neither Europa nor Io are plausible analogs for short-period exoplanets, because their high albedos are caused by water ice and condensed sulfur species that are unstable inside the inner edge of the habitable zone. We note that sulfur can exist in liquid form inside the inner edge of the habitable zone (Theilig 1982), but any sulfur pools or oceans would again have a low albedo (Nelson et al. 1983).

Third, recent thermal phase curve observations with Spitzer indicate that LHS3844b likely has no significant atmosphere and a low dayside albedo (Kreidberg et al. 2019). The best model fits suggest a basaltic surface composition and an albedo less than 0.2. Based on these considerations, we believe it is justified to assume that other rocky exoplanets will have similarly low surface albedos.

Another false positive scenario is a planet that is not tidally locked. In this case the dayside emission temperature would be lower due to the planet's rotation instead of atmospheric heat redistribution. Based on theoretical arguments we consider this scenario unlikely. First, atmospheric models show that even nonsynchronous rotators have day–night temperature contrasts similar to tidally locked rotators if the planet is sufficiently hot and the atmosphere sufficiently thin, so that the atmosphere's radiative timescale is short compared to the planet's rotation period (Rauscher & Kempton 2014). Following this argument GJ1132b in a 3:2 spin–orbit resonance would appear effectively tidally locked if its atmosphere were thinner than ∼0.5 bar. If the dayside temperature were then observed to be much cooler than a tidally locked bare rock, this would still indicate a relatively thick atmosphere. Second, tidal models suggest that nonsynchronous rotation is unlikely for short-period planets orbiting small host stars (Leconte et al. 2015; Barnes 2017), which includes all three targets we consider above, even though it might become relevant for planets around late K and early M dwarfs that are located inside their host stars' habitable zone.

A final false positive scenario is dynamical heat redistribution by a lava ocean instead of an atmosphere. This scenario does not apply to planets like LHS3844b or GJ1132b, and is only feasible on planets like 55 Cancri e which are hot enough that their dayside is molten while simultaneously cool enough that rock vapor does not form a thick atmosphere. However, even for 55 Cancri e we do not consider heat redistribution by a lava ocean likely based on theoretical estimates that lava ocean currents are too slow to affect planetary heat redistribution in the absence of a wind-driven circulation (Kite et al. 2016).

Thin atmospheres with inefficient heat redistribution are a likely false negative scenario for our proposal. Such atmospheres would be easier to detect via transit and eclipse spectroscopy, or potentially by detecting the planet's nightside emission via thermal phase curves (Figure 6).

We note that some atmospheres might be thin but still have significant cloud cover, analogous to how Mars' atmosphere is thin but can produce reflective clouds as well as optically thick planet-encircling dust storms. Such thin atmospheres might not reveal their presence via the atmosphere's heat redistribution but could still be detectable in eclipse photometry through the clouds' effect on the planet's albedo, which is a possibility for identifying candidate atmospheres that we explore in a companion paper (Mansfield et al. 2019).

Our models do not include the impact of clouds on the dayside emission spectrum, nor do we consider the increased day–night latent heat transport in atmospheres with condensation. Both processes should tend to reduce the dayside brightness temperature, and thus could affect the quantitative interpretation of eclipse observations. However, given that both processes are atmospheric phenomena, a low observed brightness temperature would thus still be indicative of an atmosphere.

We also assume blackbody spectra for the planet surface, even though minerals can induce spectral features on airless bodies (Hu et al. 2012). Surface-induced spectral features are an important potential false positive for eclipse spectroscopy, which should be explored in future work. Nevertheless, we do not expect that surface spectral features would negatively affect atmospheric detection via eclipse photometry. The underlying reason is that the emissivity of many minerals tends to increase from shorter to longer wavelengths, so the brightness temperature an observer sees at relatively long wavelengths in the MIRI bandpass will be biased high (Mansfield et al. 2019). This bias works in the opposite direction of atmospheric heat transport, so an observed cool dayside would be even more indicative of an atmosphere if we accounted for surface spectral features than it is with a blackbody surface.