Exoplanet Biosignatures: Understanding Oxygen as a Biosignature in the Context of Its Environment

1.1. Reliability

It is the volatile byproduct of oxygenic photosynthesis—the metabolism driven by sunlight, which is the dominant source of energy on our planet's surface—and so clearly has a biological origin. Until recently, O₂ was considered to have little or no abiotic planetary sources, as the Sun's UV spectrum drives photochemistry that produces only trace amounts in Earth's atmosphere (e.g., Walker, 1977; Hu et al., 2012; Domagal-Goldman et al., 2014; Harman et al., 2015). O₂ also has no significant geological sources—unlike methane (CH₄) and carbon dioxide (CO₂), which are known to be produced by geological processes (Etiope and Sherwood-Lollar, 2013). Very early work considered O₂ to be an even stronger biosignature when seen along with gases such as CH₄ and nitrous oxide (N₂O) in chemical thermodynamic disequilibrium (e.g., Lederberg, 1965; Lovelock, 1965; Hitchcock and Lovelock, 1967; Lovelock and Kaplan, 1975), although given Earth's negligible abiotic O₂ sources, its high abundance alone indicates a biological origin.

1.2. Survivability

With a strong, planet-wide photosynthetic source, and relatively weak sinks compared with early periods of Earth's history (and different stellar hosts), O₂ has risen over time to become the second most abundant gas in our atmosphere (at around 21% by volume). The atmosphere's more dominant gas, N₂—which may also be biologically mediated (Catling and Kasting, 2007; Johnson and Goldblatt, 2015)—is considered a much poorer biosignature because N₂ only exhibits strong absorption in the extreme ultraviolet (XUV; 0.01 < λ < 0.121 μm). In this wavelength region, several other common molecules also absorb H₂O, CO₂, CO, O₂, CH₄ and other hydrocarbons (Rothman et al., 2013), and stars are much less luminous in the UV than in the visible regions, reducing the UV reflection or transmission signal from the planetary atmosphere. Sinks for O₂ include outgassing of reduced gases such as CH₄ and H₂, crustal oxidation, and photolytic destruction in the stratosphere.

1.3. Detectability

These factors lead to a high abundance and even mixing throughout the atmospheric column, making O₂ accessible to observation by transit spectroscopy. Depending on the observing geometry and atmospheric composition of a terrestrial (predominantly composed of silicate rocks and metals) planet, transmission spectroscopy may not reach the near-surface environment, instead probing the stratosphere and upper troposphere (Bétrémieux and Kaltenegger, 2013, 2014; Misra et al., 2014b; Meadows et al., 2018). This gives O₂ a significant advantage over other proposed, often more complex biosignature molecules, including methanethiol and dimethyl disulfide (Pilcher, 2003; Domagal-Goldman et al., 2011; Seager and Bains, 2015), which are more susceptible to photolysis by UV radiation and so are confined largely to the lower troposphere, with significantly smaller concentrations in the stratosphere (Domagal-Goldman et al., 2011).

O₂ is also one of the very few biogenic gases that absorbs strongly in the visible and near-infrared (NIR), likely wavelength regions to be covered by upcoming NASA missions and ground-based telescopes seeking to characterize extrasolar planets. O₂ has strong features at wavelengths <0.2 μm, and a γ band at 0.628 μm, B-band at 0.688 μm, a strong A-band at 0.762 μm, and the a ¹Δ_g band at 1.269 μm (Rothman et al., 2013). O₂ could also be inferred from detection of its photochemical byproduct, ozone (O₃) (Ratner and Walker, 1972), which has strong bands in the UV (0.2–0.3 μm), visible (0.5–0.7 μm) and midinfrared (MIR) (9.6 μm) (Rothman et al., 2013). O₂ at high abundance could be identified by collisionally induced absorption as O₂ molecules collide with each other or form the O₄ complex, which has strong bands in the visible (0.34–0.7 μm; Hermans et al., 1999; Thalman and Volkamer, 2013) and in the NIR (at 1.06 μm, Greenblatt et al., 1990; and at 1.27 μm, Maté et al., 2000).

However, although O₂ is clearly a strong biosignature, recent research has indicated that its availability in a planetary atmosphere and its unique interpretation as a biological product may not be as straightforward as originally thought. These insights have come from exploration of the processes governing the long-term evolution of atmospheric O₂ on the early Earth, and advances in theoretical modeling of star–planet interactions for planets orbiting other stars. This research indicates that the picture is far more complex, with both significant false negatives—planetary processes that suppress the buildup of atmospheric O₂, despite the presence of biological production (Lyons et al., 2014; Planavsky et al., 2014b; Reinhard et al., 2017), and false positives—planetary processes that generate large (modern Earth-like) abundances of O₂ abiotically (e.g., Wordsworth and Pierrehumbert, 2014; Luger and Barnes, 2015; Gao et al., 2015; Harman et al., 2015; Tian, 2015; see Meadows, 2017 for a review), being possible.

On the early Earth, it is likely that oxygenic photosynthesis was developed well in advance of the rise of O₂ to significant levels in our atmosphere, constituting a false negative. Isotope measurements point to transient low levels of O₂ in the early Earth's environment 3.0–2.65 Ga, which were likely generated by oxygenic photosynthesis (Czaja et al., 2012; Crowe et al., 2013; Planavsky et al., 2014a; Riding et al., 2014). Yet the irreversible, global accumulation of O₂ in the atmosphere, which was likely mediated by burial and removal of organic carbon from Earth's surface environment (Kasting, 2001; Lyons et al., 2014), evidently occurred somewhat later, between 2.45 and 2.2 billion years ago (Ga; Farquhar, 2000; Bekker et al., 2004; Canfield, 2005). Recent sulfur isotope data (Luo et al., 2016) place this transition at 2.33 Ga. However, recent studies of Earth's carbon isotope record suggest that the record does not constrain the history of organic burial well enough to evaluate whether changes in fractional organic burial since 3.6 Ga can explain the rise of O₂ (Krissansen-Totton et al., 2015). Other proposed mechanisms for oxygen's rise include hydrogen escape from the upper atmosphere (Catling et al., 2001) and the long-term evolution of volcanic–tectonic processes at Earth's surface (Kump and Barley, 2007; Kasting, 2013; Lee et al., 2016). Additional data suggest that the permanent rise of O₂ to high levels in Earth's atmosphere (>1% of the present atmospheric level (PAL)) may have been delayed even further, to 0.8 Ga (Planavsky et al., 2014b), long after the advent of oxygenic photosynthesis.

This delay between the evolution of the biological source of O₂, photosynthesis, and the subsequent rise of abundant O₂ in our atmosphere is due to diverse geological sinks for O₂ in the environment effectively countering its biological production—ultimately resulting in a false negative. This example points to the importance of understanding a planetary environment to identify characteristics that may potentially reduce our ability to see a biosignature, even if a strong biological source exists.

Environmental context will also be extremely important for identifying the likelihood for false positives in a planetary environment—abiotic processes that could mimic a biosignature. O₂ was long considered the most robust biosignature possible, because in Earth's modern environment, there were no known false positives, such as geological processes and photochemistry, that could generate it. For example, on Earth, the abiotic production of O₂, principally by photolysis of water vapor, is at least a million times less than that produced by photosynthesis (Walker, 1977; Harman et al., 2015).

However, recent modeling studies examining star–planet interactions, especially for exoplanets orbiting M dwarfs, have identified several mechanisms that could potentially lead to the abiotic production of significant amounts of O₂. These mechanisms ultimately rely on conditions that lead to the photolysis of CO₂ or H₂O, two O-bearing gases that are likely to be common on terrestrial planets in the habitable zone (HZ). One mechanism for O₂ buildup requires the vaporization of oceans, and the subsequent photolysis of water vapor and loss of H, for planets orbiting early, super-luminous M dwarf stars (Luger and Barnes, 2015). A low noncondensable gas inventory in a planetary atmosphere can drive high stratospheric H₂O abundances by increasing the altitude of the tropospheric “cold trap,” leading to the photolysis of H₂O and the loss of H to space. Because the water loss is a strong function of the planet's atmospheric composition, this potential abiotic O₂ generation mechanism can affect planets orbiting any type of star (Wordsworth and Pierrehumbert, 2014). Finally, photolysis of CO₂ as a source of atmospheric O₂ may occur on planets orbiting M dwarfs. Whether photolytic O₂ is produced is dependent on both how well the UV spectrum of the star photolyzes CO₂ and the suppression of catalytic or other processes that would allow the CO and O to rapidly recombine (e.g., Domagal-Goldman et al., 2014; Tian et al., 2014; Gao et al., 2015; Harman et al., 2015). Each of these mechanisms leaves tell-tale signs in the planetary environment that can point to the abiotic source of the O₂ (e.g., Schwieterman et al., 2015b, 2016).

Rather than demonstrating that O₂ is not suitable as a biosignature, our new and evolving understanding of false positives for O₂ increases its robustness (Meadows, 2017). The study of O₂ as a biosignature has taught us that by attempting to understand the environmental context, including the possibility for false positives and negatives, we will ultimately improve our interpretation of exoplanet observations and increase our confidence that they do indeed indicate the presence of life. Knowledge gained from the delayed rise of O₂ during the history of our own planet helps us identify environmental contexts for false negatives for O₂, and interdisciplinary modeling studies of exoplanets have helped us to explore and identify the stellar and planetary characteristics that are more likely to lead to O₂ false negatives. This knowledge can guide target selection for habitable planets, and can be used to design observations to search for the necessary environmental context to increase our confidence in the interpretation. Moreover, these studies serve as a template for future exoplanet biosignature development, and the lessons learned from our study of O₂ can help guide the development of a generalized framework for biosignature development and detection (see e.g., Catling et al., 2018).

In Section 2, we describe in more detail the evolution of O₂ throughout Earth's history, highlighting the ways in which the environment initially suppressed and then finally permitted its development as a detectable biosignature. In Section 3, we discuss the known potential false positive mechanisms for terrestrial exoplanets, and their likely observable environmental impact. In Section 4, we discuss the specific observational implementation for O₂ detection and how confidence in its biogenicity can be enhanced, including (1) characterization of planetary, stellar, and planetary system environmental context; (2) the search for O₂; and (3) the specific environmental characteristics that could be sought to help discriminate false negatives and positives. We describe the photometric and spectroscopic measurements required to execute these three steps for different near-term exoplanet observing facilities, including the James Webb Space Telescope (JWST), extremely large ground-based telescopes, and for proposed space-based direct imaging coronographic/starshade missions. In Section 5, we discuss and prioritize the most important discriminating measurements, and then discuss the robustness of any detection as a function of accessible wavelength range and other telescope characteristics. In Section 6, we present our conclusions.

2.1. The evolution of oxygenic photosynthesis

Much research has been directed at understanding the evolutionary transition between anoxygenic and oxygenic photosynthesis, and this article is not intended to be an exhaustive review of the various schools of thought. There is a difference in the complexity of photosynthetic machinery between oxygenic and anoxygenic phototrophs. The light reactions of oxygenic photosynthesis in cyanobacteria, algae, and plants use two linked photosystems: photosystems I (PSI) and II (PSII). PSII and the oxygen-evolving complex allow for the extraction of electrons from water molecules, which have a very positive redox potential (O₂/H₂O pair; E₀′ = +0.87 V) (see overview in Schwieterman et al., 2018). The photosynthetic machinery of anoxygenic phototrophs is different. These primitive phototrophs use a single photosystem, which can be classified into one of two families—type I reaction centers, including PSI in chloroplasts, cyanobacteria, green sulfur bacteria, heliobacteria, and phototrophic acidobacteria, and type II reaction centers, such as PSII in chloroplasts, cyanobacteria, purple bacteria, and green nonsulfur bacteria (Allen and Williams, 1998; Bryant et al., 2007). Rather than using chlorophyll pigments that absorb light at wavelengths in the visible range, anoxygenic phototrophs utilize a diversity of bacteriochlorophyll pigments that absorb in the NIR (Senge and Smith, 1995).

The evolution of photosynthesis may have been driven, in part, by the redox potential of electron donors or reductants within an environmental context, with a step-wise increase in the redox potentials of reductants donating electrons to PSII (Pierson and Olson, 1989; Olson, 2006). The most ancient form of anoxygenic photosynthesis uses strong reductants such as H₂ or H₂S to donate electrons to a single photosystem (Olson and Pierson, 1986), but the limited availability of these volcanogenic compounds may have provided evolutionary pressures to develop a second photosystem, which could use weaker but more abundant reductants, such as Fe(II) and water (Olson, 1970, 1978).

Modern cyanobacteria produce O₂ resulting from the extraction of electrons from water, but there is a large difference in the redox potentials between the O₂/H₂O pair (E₀′ = +0.87 V) of oxygenic photosynthesis and the S₀/HS⁻ pair (E₀′ = −0.27 V) commonly used by anoxygenic photosynthesis. Researchers have speculated that an intermediate reductant, such as Fe(II) (E₀′ = +0.30 V at circumneutral pH) or Mn(II) (E₀′ = ∼+0.60 V), could have bridged the gap and acted as a transitional electron donor before water (Pierson and Olson, 1989; Olson, 2006). The widespread abundance of reduced iron (and potentially reduced manganese) on the early Earth would have made it particularly suitable as an electron donor for photosynthesis. Pierson et al. (1993) pointed out that the oxidized iron products of this type of photosynthesis would have provided substantial protection from UV radiation for surface-dwelling phototrophs before the development of an O₃ shield.

Due, in part, to the wide availability of water as an electron donor compared with the limited availability of volcanogenic reductants, oxygenic phototrophs spread easily and dominated aquatic and terrestrial habitats. Indeed, the evolution of oxygenic photosynthesis would ultimately result in a radical restructuring of Earth surface environments, with significant impacts on all major biogeochemical cycles. The so-called Great Oxidation Event, which occurred at ∼2.3 Ga (Bekker et al., 2004; Luo et al., 2016), set the stage for the evolution of more complex life-forms dependent on O₂. Consequently, O₂ is not only a promising biosignature, but it may also indicate an environment in which multicellularity, and even intelligence, can be supported (e.g., Catling et al., 2005).

It is of course difficult to estimate the likelihood that oxygenic photosynthesis would evolve on an exoplanet. The availability of carbon sources (CO₂ and/or reduced carbon) must be considered, and in the case of CO₂ fixation, the availability of reductants or electrons must also be factored in. In speculating about the evolution of photoautotrophy on an exoplanet, volcanism—a common geological process on terrestrial habitable planets—can supply CO₂ and other reduced species such as H₂. The possibility may exist that the same evolutionary selective pressures based on electron donor availability that gave rise to oxygenic photosynthesis on the early Earth could play out on an exoplanet, leading to the same metabolic outcome.

If photosynthesis did evolve on an exoplanet, it would likely coevolve with its environment to develop light-harvesting pigments that make use of the available stellar spectrum at the surface of the planet. Previous work suggested that chlorphyll a evolved to absorb the peak solar photon flux onto the planetary surface (Kiang et al., 2007a, 2007b), which could potentially be predicted for different star–planet combinations. The peak photon flux will be the result of the interplay between the incident solar spectrum and subsequent absorption by molecules in the atmosphere, as well as efficiency considerations for NIR radiation, which may be too low in energy to drive photosynthesis efficiently beyond a limiting wavelength (Kiang et al., 2007a, 2007b). On Earth, O₃ absorption in our atmosphere drives the Sun's incident peak photon flux near 600 nm redward. The peak photon flux at the surface is at 688 nm, where the principal photosynthetic pigment, chlorophyll a, absorbs (Kiang et al., 2007b). However, on the early Earth, without abundant O₃, the peak photon flux at the surface may have been different, but potentially also pushed redward by strong blue absorption from hydrocarbon haze. Such hazes may have been present during early periods of Earth's history when the atmosphere was more reducing (Arney et al., 2016).

In the case of exoplanets, the spectrum of the host star will be well known in the visible range, and the planetary atmospheric composition may be constrained by spectral observations. These observations will allow radiative transfer models to predict photon fluxes at the planetary surface, and potentially help identify the most likely spectral region for photosynthetic pigment absorption. However, it should be noted that the Kiang et al. (2007a, 2007b) supposition that phototrophs evolved to absorb the maximum photon flux (which was developed when considering vegetation) does not hold up when applied to cyanobacteria and anoxygenic phototrophs. Photosynthesis in these microbes is often inhibited by high light levels, and maximum photosynthetic activity can sometimes be observed at only 10–20% of the maximum solar irradiation (e.g., Murata et al., 2007; Trouwborst et al., 2007).

2.2. The suppression, and ultimate rise, of O₂ on the early Earth

Although the evolution of oxygenic photosynthesis represented a critical evolutionary innovation, it was not a sufficient condition for appreciable and persistent planetary oxygenation, which may not have happened until as recently as 0.8 Ga (Lyons et al., 2014). Understanding oxygen's late rise in our atmosphere requires a basic discussion of the factors modulating Earth's global O₂ cycle (Fig. 1). Oxygenic photosynthesis on the modern Earth fixes carbon (i.e., converts atmospheric CO₂ into organic carbon) at a net rate of ∼100 Pg (10¹⁵ g) of carbon per year, distributed roughly evenly between the marine and terrestrial photosynthetic biospheres (Field et al., 1998). For every mole of carbon fixed into biomass by oxygenic photosynthesis, one mole of free O₂ is released to the environment.

However, the vast majority of this O₂ is consumed very rapidly through aerobic respiration in surface environments, with only a very small fraction of the organic carbon initially produced by oxygenic photosynthesis (less than 1%) escaping respiration and becoming buried in marine and terrestrial sediments (Berner, 1982). This relatively slow “leak” of organic carbon into Earth's upper crust represents a net source of O₂ to Earth's atmosphere. In addition, a fraction of photosynthetically produced organic matter that escapes burial will be consumed by microbial sulfate reduction, which essentially transfers electrons from organic matter to sedimentary pyrite (FeS₂), the burial of which similarly results in a net release of O₂ to surface environments. These fluxes are augmented by the irreversible escape of H to space, which scales with the total abundance of H-bearing reduced gases at the homopause (e.g., CH₄ and H₂; Hunten, 1973). Although this represents a very small flux on the modern Earth, it is critical to consider as a mechanism of secular oxidation on timescales of planetary evolution (Catling et al., 2001). Input fluxes of O₂ are ultimately balanced on long timescales by O₂ consumption during the uplift and oxidative weathering of reduced mineral phases in Earth's crust (including previously buried organic carbon and pyrite sulfur) and through reaction with volcanic and metamorphic reductants (Kasting and Canfield, 2012).

Whether a planet develops a biogenic O₂-rich atmosphere will thus depend not only on the evolution of oxygenic photosynthesis but also on the long-term balance between sources and sinks of O₂ at the surface (e.g., Gebauer et al., 2017). The accumulation and stability of large biogenic O₂ inventories in a planetary atmosphere will, in turn, depend on how input/output fluxes scale with atmospheric O₂ levels (if at all) and on secular changes in planetary degassing, differentiation, and atmospheric escape. An important consequence of this is that the evolution of oxygenic photosynthesis cannot, on its own, be expected to yield a remotely detectable oxidizing atmosphere. The planetary environment also plays a crucial role.

Earth's history provides an instructive example. Constraints from stable isotopes and trace element proxies indicate the evolution of oxygenic photosynthesis on Earth by at least ∼3.0 Ga (Planavsky et al., 2014a) and perhaps much earlier (Rosing and Frei, 2004). However, complementary isotopic constraints indicate that Earth's atmosphere was pervasively reducing until ∼2.5 Ga (Farquhar et al., 2001; Pavlov and Kasting, 2002; Zahnle et al., 2006). There thus appears to have been a very significant period on Earth during which oxygenic photosynthesis was present but large amounts of O₂ did not accumulate in Earth's atmosphere. In addition, there is some evidence that after the initial accumulation of O₂ in Earth's atmosphere at ∼2.3 Ga, atmospheric O₂ levels remained relatively low for much of the subsequent ∼2 billion years (Planavsky et al., 2014b; Cole et al., 2016; Tang et al., 2016), during which biogenic O₂, although clearly present, may have been challenging to detect remotely given current technology (Reinhard et al., 2017).

In sum, the emergence of a biogenic O₂-rich atmosphere will depend on both the evolution of oxygenic photosynthesis and geochemical dynamics at the planetary surface that are favorable for the long-term accumulation of a large atmospheric O₂ inventory; if planetary conditions are not favorable, then a false negative will occur. These dynamics will, in turn, depend on a series of planetary factors that may be challenging to constrain observationally or from first principles. For example, heat flux from a planetary interior (as constrained by radiogenic element inventory and planet size), O₂ fugacity of the planetary mantle (as constrained by both initial chemistry and long-term recycling of materials from the surface), the degree of crustal differentiation (as constrained by both overall heat fluxes and planetary rheology), and ocean chemistry can interact to buffer atmospheric O₂ to low levels despite the presence of oxygenic photosynthesis. The ability to constrain these contextual variables through observations of the planet and star, or through modeling, may ultimately form a critical component of diagnosing false negatives for O₂ on living planets. Similarly, they lay the foundation for critical thinking on the nature and use of contextual information for other proposed biosignatures as well.

2.3. Earth's history of oxygen's companion biosignatures CH₄ and N₂O

The role of the environment in enhancing or suppressing the atmospheric accumulation of biogenic gases also applies to other species, such as CH₄ and N₂O, and the detection of these gases could be used to strengthen confidence that the observed O₂ in an exoplanet atmosphere is indeed biogenic (e.g., Hitchcock and Lovelock, 1967). Atmospheric levels of these gases will be a function of production at the surface (whether biological or abiotic), but will also depend strongly on sink fluxes due to biological recycling, geological processes, and photochemistry in the atmosphere. For example, N₂O is photochemically unstable in reducing atmospheres (Roberson et al., 2011), as these atmospheres lack the O₂, O₃, or CO₂ of more oxidizing atmospheres, which shields N₂O from UV photolysis to N₂. Consequently, in more reducing atmospheres, it is difficult to sustain high atmospheric levels of N₂O without large surface fluxes. The abundance of N₂O in planetary atmospheres also depends on the spectrum of the host star, and many M dwarfs are less able to directly photolyze N₂O, allowing longer lifetimes and higher abundances for the same surface flux (Segura et al., 2005; Rauer et al., 2011; Rugheimer et al., 2013, 2015). However, in general, unless the atmosphere is already oxidizing, whether due to the action of oxygenic photosynthesis, or stellar-driven atmospheric evolution to high-O₂ or high-CO₂ states (e.g., Meadows et al., 2018; Luger and Barnes, 2015), relatively high N₂O fluxes will be required to maintain elevated atmospheric N₂O levels. These kinetic constraints on biogenic gas abundance in planetary atmospheres have led to approaches that emphasize the magnitude of chemical disequilibrium between species (e.g., O₂/O₃ and CH₄), and this approach can potentially provide more specific information about the magnitude of production/consumption fluxes compared with the abundance of a single biogenic gas (Lovelock and Kaplan, 1975; Kasting et al., 2013; Krissansen-Totton et al., 2016a).

The evolution of atmospheric chemistry on Earth, as described hereunder, illustrates how searching for secondary gases to strengthen confidence in oxygen's biogenic origin may be both challenging and illuminating (Fig. 3).

2.4. Secondary photosynthetic biosignatures

Earth also demonstrates alternative ways in which our photosynthetic biosphere has impacted the environment beyond the abundant O₂ that eventually rose in our atmosphere. These secondary biosignatures include surface reflectivity signatures from light harvesting (Kiang et al., 2007a, 2007b) and other pigments developed for nonphotosynthetic purposes by phototrophs (Hegde et al., 2015; Schwieterman et al., 2015a), as well as the strong jump in reflectivity longward of chlorophyll a at 0.7 μm, from the red edge of vegetation (Gates et al., 1965; Kiang et al., 2007a). Another secondary biosignature of photosynthesis in globally averaged spectra of our planet is the seasonal variations in the abundance of CO₂ (Meadows, 2008) due to growth and decay of vegetation on land, which imparts a ∼2% seasonal variation in CO₂ abundance (Keeling, 1960; Hall et al., 1976; Keeling et al., 1976). Earlier in Earth's history before the permanent rise of O₂, small amounts of biogenic O₂ may also have shown seasonal variation (see Schwieterman et al., 2018, for a review; Reinhard et al., 2016).

Either a corroborative surface or temporal signature may be sought as a secondary confirmation of a photosynthetic origin for any detected O₂ in a planetary atmosphere. The vegetation red edge has likely been widespread on Earth since about 0.46 Ga (Carroll, 2001; Igamberdiev and Lea, 2006), corresponding with the rise of land plants, and may be coincident with the rise of O₂ to near-modern levels. However, in false negative situations wherein planetary processes are suppressing the rise of photosynthetically generated O₂, surface reflectivity biosignatures from photosynthetic and nonphotosynthetic pigments produced by phototrophs may have been the only way to detect photosynthesis (Schwieterman et al., 2015a).

3.1. Low noncondensable gas inventories

The possible abiotic O₂ generation mechanism that will likely be the most difficult to observationally recognize and preclude involves photolysis of water—and subsequent hydrogen loss—from terrestrial atmospheres that are depleted in noncondensable gases such as N₂ (Wordsworth and Pierrehumbert, 2014). This mechanism may produce Earth-like quantities of O₂ on an ocean-bearing world. To work, the effectiveness of the “cold trap”—the rapid reduction in temperature with altitude that causes rising water vapor on Earth to condense and thereby remain trapped in the troposphere—must be weakened. This occurs when the atmospheric temperatures are high, or when the total inventory of noncondensable gases—for example, N₂, O₂, or CO₂—is low (Wordsworth and Pierrehumbert, 2014). Because this mechanism to allow water into the stratosphere relies primarily on a planetary property, a low inventory of gases that do not condense at typical habitable zone terrestrial planetary temperatures and pressures, it may work for planets orbiting stars of any spectral type, including Sun-like stars. With a low inventory of noncondensable gases, water is able to rise higher into the atmosphere before it condenses, and it is more vulnerable to photolysis by incident stellar UV radiation at these higher altitudes. The water's H atoms escape, leaving abiotic O₂ to build up in the atmosphere. This continues until O₂, itself a noncondensable gas, overwhelms surface sinks to reach a sufficiently high atmospheric abundance that it can establish a cold trap, and halt the loss of water vapor. For an abiotic planet that is largely Earth-like except for the N₂ inventory, Wordsworth and Pierrehumbert (2014) calculate that this proposed mechanism could result in planets with a very Earth-like O₂ partial pressure (∼0.15 bars), while losing significant amounts of water.

3.2. Enhanced M dwarf premain sequence stellar luminosity

Another mechanism that may be capable of generating thousands of bars of abiotic O₂ explores the impact of the premain sequence, super-luminous phase of young M dwarf stars on terrestrial planet environments (Luger and Barnes, 2015). The extended contraction phase of premain sequence young stars makes them significantly more luminous (e.g., Baraffe et al., 1998) than they will be when they enter the main sequence hydrogen burning phase. Consequently, planets that form in what will become the main sequence habitable zone are subjected early on to very high levels of radiation (Lissauer, 2007). This super-luminous phase is longer for lower mass M dwarfs than other stellar spectral types, and can extend for up to 1 billion years (Gyr: Baraffe et al., 1998). Modeling suggests that this super-luminous phase can drive the loss of up to several Earth ocean equivalents of water for an M dwarf HZ terrestrial planet (Luger and Barnes, 2015; Tian, 2015). The resultant water-rich atmosphere will be susceptible to photolysis and hydrogen escape, and may produce hundreds or thousands of bars of O₂ (Luger and Barnes, 2015; photolysis of an Earth ocean of water can produce ∼240 bars of O₂ on an Earth-like planet, Kasting, 1997).

The amount of O₂ generated in this scenario is a strong function of stellar spectral type, XUV flux, original water inventory, planetary mass, and the position of the planet in the habitable zone. Planets in the outer regions of the habitable zone of M0–M3V stars are less likely to generate significant amounts of abiotic O₂ (Luger and Barnes, 2015), as are planets orbiting M0–M3V dwarfs with lower stellar XUV, as slower water loss may preclude O₂ generation and buildup (Tian, 2015). However, for planets orbiting later type M dwarfs (M4V and later) depending on the planet's original water inventory, mass, stellar parameters, and the strength of surface sinks (Rosenqvist and Chassefière, 1995; Schaefer et al., 2016), up to several hundreds of bars of photolytically produced O₂ could build up in the atmosphere (Luger and Barnes, 2015). For the recently discovered Proxima Centauri b, orbiting an M5.5V star, estimates of ocean loss during the 169 ± 13 Myr (Barnes et al., 2018) before the planet enters the habitable zone are sensitive to the assumed form of the XUV evolution. Results for global water loss range from less than one ocean (<250 bars O₂ generated), assuming a constant XUV flux for the host star for the first 3 Gyr (Ribas et al., 2016), up to 3–10 Earth ocean equivalents lost (750–2500 bars of O₂ generated), assuming that the XUV scales with the decreasing bolometric luminosity (Barnes et al., 2018).

For terrestrial planets, if substantial CO₂ is also outgassed over time, and the planet has little or no surface ocean in which to sequester CO₂, then these types of worlds could sustain atmospheres with large quantities of both CO₂ and O₂ (Meadows et al., in press). Over time the O₂ may be lost and CO₂ may continue to accumulate. Although we have evidence that Venus lost a global “ocean” that was at least 3 m deep (de Bergh et al., 1991), any abiotic O₂ generated was subsequently lost to surface or interior sinks (Rosenqvist and Chassefieré et al., 1995; Hamano et al., 2013; Schaefer et al., 2016) or top-of-atmosphere loss processes. The latter may include the recently discovered “electric wind” surrounding Venus, which can strip O₂ ions from the planet (Collinson et al., 2016).

3.3. Stellar spectrum driven photochemical production

Abiotic O₂ and its proxy, O₃, can also be formed through planetary photochemistry of CO₂. Each photochemical reaction requires photons that exceed a threshold energy level—or equivalently, photons at wavelengths shorter than a threshold wavelength—to be absorbed by and split the molecule undergoing photolysis. Planetary photochemistry is strongly sensitive to the UV spectral energy distribution of the parent star (e.g., Segura et al., 2003, 2005, 2010; Grenfell et al., 2007, 2014; Rugheimer et al., 2013, 2015), and in particular the ratio of shorter to longer wavelength UV radiation, which can split molecules and drive reactions that change the composition of the atmosphere, without relying on atmospheric escape.

The ultimate composition of the planetary atmosphere will then depend on the sources and sinks for photochemical reactions provided by the planetary environment. For example, photochemical abiotic O₂ production will depend strongly on the source of O atoms, which is primarily controlled by the abundance and photolysis rates of atmospheric CO₂, SO_2, H₂O, and other O-bearing gases. The atmospheric sink will depend on the availability of H atoms in the atmosphere from H₂, H₂S, and hydrocarbons. Although water vapor can be sourced from a planetary ocean, the availability of other gases, such as CO₂, SO₂, and reducing (H-bearing) gases, is governed by the planet's volcanic outgassing rates, which can be sustainable on long geological timescales.

In particular, CO₂ is likely a common atmospheric gas on terrestrial planets. In the Solar System, CO₂ dominates the atmospheric composition of Venus and Mars, and was likely a significant component of the early Earth's atmosphere as well (Kasting et al., 1993; Sheldon et al., 2006; Sleep, 2010; Driese et al., 2011). The key reactions to produce O₂ from CO₂ photolysis include CO₂+hν→CO+O, and the collisional recombination reaction O+O+M→O₂+M, where M is a third molecule that carries away excess energy to stabilize the collisional product. The yield of O₂ from this process depends on how efficiently the back reaction to regenerate CO₂ can occur. This, in turn, depends on the atmospheric abundance of catalysts such a HO_x (e.g., from water vapor; Tian et al., 2014) or NO_x (e.g., from cosmic rays; Grenfell et al., 2012) (c.f. Yung and DeMore, 1999; Stock, et al., 2017).

Aqueous reactions may also remove atmospheric O₂, especially if the planet supports a surface ocean, and these reactions are also important for our understanding of abiotic O₂ generation. For example, the rate of reaction of dissolved CO and O₂ to reform CO_2, and thereby draw down atmospheric O₂, is poorly understood, but is crucial to understanding the final balance of O₂ in the atmosphere (Harman et al., 2015). Similarly, weathering of surface crust (e.g., Anbar et al., 2007), and the sequestration of O₂ into the planetary mantle (e.g., Hamano et al., 2013; Schaefer et al., 2016) are key processes that control O₂ draw down, and could result in abiotic O₂ buildup if slow. Conversely, if these processes are aggressive enough, they could produce a false negative for biologically produced O₂ by scrubbing photosynthetically generated O₂ from a planetary atmosphere, as likely happened over Earth's history, and discussed in Section 2.

Several groups have identified potential photochemical mechanisms to generate abiotic O₂ and O₃ on terrestrial planets in the habitable zone (Domagal-Goldman and Meadows, 2010; Hu et al., 2012; Domagal-Goldman et al., 2014; Tian et al., 2014; Gao et al., 2015; Harman et al., 2015). However, the amounts of abiotic O₂ and O₃ generated in these simulations differ, in part, due to different assumptions about the abundance of available catalysts to drive recombination of CO_2, or destruction of O₂ and O₃, which can be affected by the availability of H₂O and the spectrum of the star, as well as the efficiency of the CO and O₂ reaction in seawater. The mechanism that could produce the largest signal from CO₂ photolysis requires a desiccated, cold, H-poor atmosphere. This water-poor atmosphere cannot support photolytic generation of the OH catalyst that accelerates CO₂ recombination (Gao et al., 2015). In this scenario, a catalytic cycle feedback with O₃ formation combines with the suppression of CO₂ recombination, and photochemical models predict stable atmospheric fractions of O₂ near 15%.

In another proposed mechanism, recombination of photolyzed CO₂ can be slowed by a parent star (typically an M dwarf) with a higher far-ultraviolet (FUV; 0.122 < λ <0.2 μm) to mid-ultraviolet (MUV; 0.2 < λ < 0.3 μm) and near-ultraviolet (NUV; 0.3 < λ < 0.44 μm) ratio when compared with the Sun. The higher FUV photolyzes CO₂, but the lower MUV–NUV radiation inhibits the photolysis of water and other HO_x chemistry that would drive recombination (Tian et al., 2014; Harman et al., 2015). For the cases considered for this mechanism, atmospheric O₂ abundances as high as 0.2% to 6% are predicted, with higher values corresponding to little or no O₂ sinks in the planetary environment (Harman et al., 2015). When more realistic sinks are included, abiotic O₂ abundances are reduced by many orders of magnitude (e.g., Harman et al., 2015).

As another possible abiotic mechanism for O₂ buildup, Léger et al. (2011) discussed the efficacy of known surface catalysts for abiotic photogeneration of O₂ from water splitting, but concluded that it was highly unlikely that sufficient catalysis could occur on a habitable planet to generate a false positive. However, a subsequent study argued that this may be possible for a planet with shallow oceans, strong NUV, and significantly large areas of surface TiO to produce Earth-like quantities of O₂ from the splitting of water >1 billion years (Narita et al., 2015).

O₃ can often serve as a proxy for O₂, especially in the MIR where O₂ does not have spectral features (e.g., Des Marais et al., 2002). However, abiotic O₃ could also be formed through photolysis of the abiotic O₂ generated by ocean loss, with calculated values ranging from 1% of PALs of O₃ up to values comparable with Earth's current O₃ abundance for planets orbiting in the HZ of M dwarfs, depending on whether liquid water remains (Meadows et al., 2018). However, the spectral slope of the UV radiation of the star could also photochemically generate O₃, even without buildup of O₂. FUV radiation can favor the generation of O₃ through photolysis of CO₂ (and O₂), whereas MUV or NUV radiation photolytically destroys O₃. Consequently, abiotic O₃ could accumulate—even without significant generation of O₂—for stars with the highest FUV to MUV ratios. O₃ was produced at 10% of Earth's current O₃ column abundance in the simulations of Domagal-Goldman et al. (2014) for M dwarf planets, without appreciable buildup of abiotic O₂.

Although the mechanisms that generate abiotic O₂ and O₃ are driven primarily by the interaction of the incident stellar spectrum with the planetary atmosphere, they can be balanced by the destruction or sequestration of O₂ and O₃ in the planetary environment. These losses could be through photolysis, catalysis, catalytic recombination into CO₂, or interaction with the planetary surface and ocean, if present. The net atmospheric accumulation of O₂ and O₃ is highly sensitive to these boundary conditions, which include weathering rates and aqueous sinks for CO, all of which are currently poorly understood.

3.4. Summary of false positives

Recent research indicates that there are several mechanisms that could produce abiotic O₂ and O₃ in a planet's atmosphere, with each presenting a potential false positive to different degrees. A summary of key components of this information is presented in Figure 2. Two of the mechanisms allow water to enter a planet's stratosphere where it is photolyzed, and the H atoms lost to space, resulting in O₂ buildup in the planet's upper atmosphere. Water entering the stratosphere is either enabled by loss of an ocean in a runaway greenhouse process (Luger and Barnes, 2015)—a mechanism that is most effective for late-type (i.e., less massive) M dwarfs—or through lack of noncondensable gases in the planetary atmosphere, which could affect planets orbiting stars of any spectral type (Wordsworth and Pierrehumbert, 2014). The runaway mechanism could produce an O₂-dominated atmosphere of hundreds of bars, and the lack of noncondensable gases could potentially result in atmospheres that are ∼15% O₂. Earth-like quantities of O₂ have also been proposed to be generated by the splitting of liquid water by a surface TiO photocatalyst (Narita et al., 2015).

The other major class of processes that build up abiotic O₂ relies on the photolysis of CO₂ and circumstances that inhibit CO₂ recombination from CO and O₂ (Hu et al., 2012; Tian et al., 2014; Gao et al., 2015; Harman et al., 2015). For photochemical production without atmospheric escape, O₂ abundances as high as 0.2% to 6% are predicted, with higher values corresponding to little or no O₂ sinks in the planetary environment. More realistic modeling of sinks can reduce these estimates by orders of magnitude (e.g., Domagal-Goldman et al., 2014; Harman et al., 2015). Finally O₃ may be considered a proxy for O₂ in a planetary atmosphere, and large abundances of abiotic O₃ may build up in the massive O₂-rich atmospheres possible after ocean loss (Meadows et al., 2018), although in these cases large amounts of O₂ will also be present. Domagal-Goldman et al. (2014) were not able to generate large abundances of O₂ from CO₂ photolysis for habitable planets orbiting M dwarfs, but did produce potentially detectable O₃ column abundance as high as 10% of Earth's modern abundance.

In most cases, however, the mechanism for abiotic production of O₂ or O₃ leaves a “tell,” an impact on the planetary environment that may be detectable. These indications can range from the presence of collisionally induced absorption from O₂ molecules that collide more frequently in dense, O₂-rich postocean-loss atmospheres (Schwieterman et al., 2016; Meadows et al., 2018), CO from the photolysis of CO₂ (Schwieterman et al., 2016), lack of water vapor (Gao et al., 2015), lack of collisionally induced absorption from N₂ (Schwieterman et al., 2015b), to the absence of reducing gases (Domagal-Goldman et al., 2014). In the following section we describe the observations needed to search for O₂ in a terrestrial planetary atmosphere and to discriminate whether that O₂ is abiotic or biological in origin based on characteristics of the parent star and the planetary environment.

4.1. Target selection

Careful target selection may help reduce the likelihood of false positive O₂, although the stellar and planetary properties most likely to result in false positive O₂ are still being explored. Several of the proposed false positive mechanisms for O₂ production are more likely to occur for planets around later type M dwarfs, which may be more susceptible to early ocean loss and buildup of O₂ from subsequent H loss (Baraffe et al., 1998; Luger and Barnes, 2015). In contrast, planets in the habitable zones of F, G, and K stars will not experience a super-luminous host star for a significant period of time. M dwarfs may also have NUV/FUV flux ratios that are more favorable for photochemical production of O₂ or O₃ (Domagal-Goldman et al., 2014; Tian et al., 2014; Harman et al., 2015). Systems older than a few Gyr may be preferred, to allow time for biologically generated O₂ to overwhelm sinks and rise in the atmosphere, and to increase the probability that abiotic O₂ generated during the star's super-luminous premain sequence phase has been depleted through sequestration in the planetary environment (Schaefer et al., 2016) or lost to space (Collinson et al., 2016; Ribas et al., 2016; Airapetian et al., 2017; Dong et al., 2017; Garcia-Sage et al., 2017).

When searching for biosignatures, it would, therefore, be advantageous to select older F, G, K, or earlier type M dwarf targets (M0–M3), where the habitable zones are likely safer from a false positive for O₂ through premain sequence ocean loss (Luger and Barnes, 2015). It will also be important to have measurements, or proxies, to estimate the UV spectrum of the M dwarf host star. However, some of these false positive mechanisms are less dependent on the host star, and are instead tied to planetary properties such as a lack of noncondensable gas species in the planetary atmosphere (Wordsworth and Pierrehumbert, 2014) or an abundance of surface TiO (Narita et al., 2015). In these cases, abiotic O₂ production could occur for planets with suitable characteristics orbiting stars of any spectral type. Target selection for detailed follow-up will instead be informed by characterization of the planet's environment, and specifically a census of bulk gases such as N₂, O₂, and CO₂. More theoretical work is needed to better understand the factors that affect habitability so that the planets most likely to be able to support life are chosen for detailed follow-up.

We are now entering a new era wherein the first observations to constrain the nature of terrestrial exoplanet atmospheres are being taken (e.g., de Wit et al., 2016), potentially providing some of the first empirical information relevant to habitability and false positive processes. Early observations of exoplanets with high levels of incident starlight such as GJ1132b (Berta-Thompson et al., 2015) and the inner TRAPPIST-1 planets (Gillon et al., 2016, 2017) may reveal key characteristics of their planetary environments that shape abiotic O₂ generation and persistence, and even provide an observational test for the limits of the habitable zone.

4.2. Key measurements of environmental characteristics that provide context for O₂ detection

Should a promising candidate planet be identified, there are several key observations that can help place detection of atmospheric O₂ into context, and help discriminate between biological and abiological sources. The first of these is detailed knowledge of the parent star, including its spectral type, metallicity, age, observed or inferred UV to infrared spectrum, and a measure of its activity levels, including the amplitude and frequency of stellar flares. The spectral type and age help to identify the likely stellar evolutionary path of the star (Baraffe et al., 1998), which drives planetary evolution, including atmospheric loss processes, which may produce false positives. The spectrum of the star is needed as input to atmospheric climate and photochemistry models, which will be used to determine atmospheric temperature, and the efficiency of photochemical production and destruction of O₂ and O₃.

Basic properties of the planet are also important, such as radius, orbit, and mass. Radius is most readily determined for transiting exoplanets where the fractional dimming of the star during the transit is a direct consequence of the planet's size, or more explicitly, its cross-sectional area. However, only a small fraction of habitable zone planets will transit (from 0.5% to 1.5% for G to M dwarf host stars). For nontransiting planets, determining a planetary radius is more challenging, as a size–albedo degeneracy exists in reflected light, as a planet could have a given measured brightness either because it is small but reflective, or because it is larger and less reflective. The range of plausible albedos for a terrestrial planet could easily span 0.06 to 0.9, potentially generating a factor of 7 uncertainty in the estimated size of the planet. Observations at MIR wavelengths—wherein emissivities for most terrestrial planetary surfaces are close to 1—may be used to infer planetary radius, especially if a color temperature can be derived from the shape of the MIR spectrum. JWST MIR thermal observations of Proxima Centauri b at different points in its orbit (a thermal phase curve) may reveal day–night temperature differences (Kreidberg and Loeb, 2016; Turbet et al., 2016; Meadows et al., 2018), whereas future ELTs may have the ability to image terrestrial planets in the habitable zones of a small sample of M dwarfs at wavelengths near 10 μm (e.g., Quanz et al., 2015). If the planet can be directly imaged at multiple epochs in its orbit, then the relative astrometry of the star and planet can be used to derive the orbital parameters. This can be done in either reflected light or in thermal emission, but the better angular resolution available at shorter wavelengths favors the use of reflected light imaging.

If the planet is not directly imaged, then its orbit may be determined indirectly—by measuring its dynamical influence on the host star or other planets in the system. The planetary mass and orbit, including its inclination, can be determined with sufficiently precise stellar astrometry to measure the reflex motion of the star induced by the planet. Stellar radial velocity (RV) measurements can determine all the orbital elements except for inclination, the angle of orientation of the planet's orbital plane to the observer. Since RV measures only the radial component of the star's orbital velocity, without knowing the orbital plane orientation, these measurements provide only a lower limit on the planetary mass (e.g., Anglada-Escudé et al., 2016).

For nontransiting planets, this RV orbital inclination ambiguity can be resolved by combining RV with imaging detections at more than one epoch and over a significant fraction of the planet's orbit, or through observations of emitted or reflected light from the planet in high-resolution spectroscopy. With the latter technique, the spectral shift in planetary atmospheric atomic or molecular lines reveals the RV of the planet itself (e.g., Snellen et al., 2015), and since the planetary orbital velocity is known from the stellar RV period, the inclination can be derived. Terrestrial planets will be extremely challenging to observe this way, and may have to await second- or third-generation instruments on ground-based ELTs. However, searching for emission from visible light aurorae on planets orbiting M dwarfs may help improve the contrast for planetary RV measurements (Luger et al., 2017). In the special case of transiting planets in multiplanet systems, the perturbations of one planet by another can sometimes be measured as changes in the timing of successive transits (transit timing variations) and allow the planetary masses to be derived (e.g., Gillon et al., 2017).

Another key planetary characteristic needed to interpret an observation of O₂ in a planetary atmosphere is the presence of water, which could be in the form of atmospheric gas, condensed clouds, and/or a surface ocean. Although the possibility that the planet can support liquid water is more likely for an Earth-like planet within the surface liquid water habitable zone (Kasting et al., 1993; Kopparapu et al., 2013), it is not guaranteed, as water may not have been delivered to the planet during its formation (Lissauer, 2007; Raymond et al., 2007), or it may have been subsequently lost (Luger and Barnes, 2015; Airapetian et al., 2016; Ribas et al., 2016; Barnes et al., 2018).

If water is present, it can help rule out several of the false positive mechanisms that rely on water loss (Luger and Barnes, 2015) or photochemistry in an extremely water-poor atmosphere (Gao et al., 2015). Knowledge of the water vapor profile—and the detection of condensate clouds that form at the water vapor cold trap—would rule out another false positive mechanism associated with high-altitude photolysis in atmospheres without a cold trap (Wordsworth and Pierrehumbert, 2014). However, if residual water persists after water loss from either early ocean loss or lack of noncondensable gases, then abiotic O₂ and water vapor may both be seen in the spectrum (Meadows et al., 2018). Absorption features for water occur throughout the visible (0.65, 0.7, 0.73, 0.8, and 0.95 μm), NIR (1.1, 1.4, and 1.8–2.0 μm), and MIR (6.3 and 18–20 μm) (Rothman et al., 2013).

Observing liquid surface water itself may be challenging and will likely only be possible for reflected light observations of nearby planets with an optimal orbital plane orientation. Observing liquid surface water is a less ambiguous habitability indicator than observations of gaseous water spectral features, which could be present in uninhabitable steam atmospheres. One such method to detect liquid surface water is ocean glint, caused by specular reflection off a smooth surface (Cox and Munk, 1954). Glint has been used to detect the presence of hydrocarbon seas on the surface of Saturn's moon Titan (Stephan et al., 2010) and water oceans in spacecraft views of Earth (Robinson et al., 2014).

Previously, Robinson et al. (2010, 2014) used a sophisticated 3D spectral Earth model to show that the spectral behavior of Earth deviates strongly from Lambertian (i.e., isotropic) scattering behavior toward crescent phases when specular reflection would be most apparent. Forward scattering from glint may mimic forward scattering behavior of clouds, but Robinson et al. (2010) showed that ocean glint even in the presence of realistic clouds (covering ∼50% of the planet) is up to a factor of 2 brighter than forward scattering from clouds and a nonglinting (i.e., Lambertian) ocean. Glint is most readily observed at phase angles within about 60°of new phase and for orbital plane inclinations that are no more than 30° from edge on to the observer. Glint from Earth's surface is most detectable at wavelengths near 0.8–0.9 μm where Earth's atmosphere is relatively transparent (although it may also be observed in the continuum between the longer 1.1, 1.4, and 1.9 μm water vapor bands).

Terrestrial exoplanet glint measurements are not possible in transmission, and outside the realm of feasibility for JWST phase curve measurements, even for the exoplanet orbiting the closest star (Proxima Centauri b; Anglada-Escudé et al., 2016). In the case of Proxima, measurement precision for detecting glint in an NIR phase curve would need to be smaller than 10⁻⁸ (Meadows et al., 2018), but this is significantly smaller than typical JWST noise floor estimates, which are on the order of 10⁻⁵ (Greene et al., 2016), and so unachievable. However, these measurements may be achievable for future ELTs and for nearby habitable zone targets observed by large space-based exoplanet characterization missions such as LUVOIR (Meadows et al., 2018).

Tectonic activity is critical to the long-term habitability of planets, as it both recycles elements required for life (e.g., Berner, 2004; Wordsworth, 2016) and serves as a global thermostat that regulates climate over geological timescales (Walker et al., 1981). Direct observations of tectonic activity will be extremely challenging, but observable impacts of tectonics on the planetary environment may be more accessible. A heterogeneous distribution of surfaces seen in maps generated by using time-resolved photometry (Cowan et al., 2009) may suggest the presence of continents. Gases released from volcanic eruptions could also be sought (Kaltenegger and Sasselov, 2009; Kaltenegger et al., 2010), but the resulting formation and accumulation of aerosols in the upper atmosphere would likely provide larger, time-variable signals (Hu et al., 2013; Misra et al., 2015). The “ingredients” for tectonic activity could also be inferred based on the elemental composition of the system. Planetary mass and radius, informed by the Fe, Si, and Mg abundance of the host star, could be used to constrain the interior composition and structure and the likelihood that a planet is tectonically active (Young et al., 2014; Dorn et al., 2015).

Additional contextual information to guide interpretation of a planet's habitability can come from atmospheric pressure. Rayleigh scattering could be used to indicate pressure above the visible surface (Benneke and Seager, 2012; von Paris et al., 2013), whether that be the planetary surface or a cloud or haze layer, as is the case for Venus. However, interpretation of these measurements can be compromised by a surface that strongly absorbs or scatters in the blue, as is the case for Mars and Earth, respectively. Rayleigh scattering will be less useful for planets orbiting M dwarfs, as the host star produces little radiation at the blue end of the visible spectrum wherein Rayleigh scattering dominates, significantly lowering the S/N at these wavelengths (Meadows et al., 2018).

Collision-induced features that are more sharply dependent on atmospheric pressure may provide an even more sensitive measurement of atmospheric pressure. For example, N₄ or O₄ features (N₂–N₂ or O₂–O₂) seen in the NIR could be used to indicate higher pressures (Misra et al., 2014a; Schwieterman et al., 2015b). N₄ absorbs strongly near 4.15 μm, just shortward of the strong 4.3 μm CO₂ band, and creates a strong spectral effect for abundances of pN₂ > 0.5 bar (Schwieterman et al., 2015b). The location of this strongly diagnostic feature in the infrared creates challenges for direct-imaging missions currently under study due to IWA and thermal background limitations to much shorter wavelengths. However, it remains perhaps the only feature capable of directly confirming large quantities of N₂ in an atmosphere (Schwieterman et al., 2015b).

Finally, a census of atmospheric gases should be performed to inform planetary composition, chemistry, and climate. The composition of the atmosphere will be important for assessing the potential of the planet to generate photochemical false positives and for calculating surface fluxes required to maintain O₂ and O₃ in the planetary atmosphere. It will also help identify potential sources of nutrients and energy for life at the surface. The atmospheric composition, and in particular knowledge of greenhouse gas abundances and distributions, when combined with information on the incoming stellar flux, can also be used to model the climate of the planet and its potential surface habitability.

However, to determine the planet's atmospheric composition, we will need not only good observations, but also robust techniques for spectroscopic retrieval that can maximize information content from low spatial- and spectral-resolution data of poor sensitivity (e.g., Irwin et al., 2008; Madhusudhan and Seager, 2009; Benneke and Seager, 2012; Line et al., 2014; Waldmann et al., 2015). The observation type (e.g., transmission, emission, and direct), wavelength coverage, and spectral resolution (among other factors) affect how complete a spectral search for atmospheric gases will be.

Degeneracies between physical parameters can also complicate inferences of the atmospheric composition. Such degeneracies have been described between the following: gas mixing ratios and temperature in secondary eclipse (Waldmann et al., 2015); mixing ratios, planetary radius, and reference pressure in transit (Benneke and Seager, 2012; Heng and Kitzmann, 2017); mixing ratios and cloud coverage in transit (Barstow et al., 2013); spectrally inactive gases (Benneke and Seager, 2012); and patchy clouds and mean molecular weight in transit (Line and Parmentier, 2016) to name a few. These degeneracies tend to arise as a result of low S/N observations that cover a narrow wavelength range. Consequently, many degeneracies can be lifted by using multiple instruments that together cover a wide wavelength (von Paris et al., 2013b; Barstow et al., 2015).

However, more work is needed to assess biases that arise due to the choice of parameterization in simple forward models (c.f. Feng et al., 2016), as well as how additional information on the star, planet, and entire system of interest may help to enable a more complete understanding of the atmospheric composition in the event of continued exoplanet spectrum limitations.

4.3. Detecting the O₂ signal from a photosynthetic biosphere

On our own Earth, O₂ is abundant and evenly mixed throughout the atmosphere. The principal means of detecting O₂ is through its spectral features in the visible and NIR spectra at 0.69 (B-band), 0.76 (A-band), and 1.27 μm. On Earth-like exoplanets, it may be accessible to both transmission spectroscopy, which will likely probe above most of the deep atmosphere, and direct imaging, which can potentially sample the atmosphere all the way to the planetary surface. A spectral resolution of 70 would be needed to resolve the A-band at Earth-like abundances (Robinson et al., 2016). The O₄ collisionally induced absorption features in the visible and NIR spectra may also be used as proxies for O₂, and may help constrain its concentration (Misra et al., 2014a). They are weakly present in Earth's direct imaging spectrum (Tinetti et al., 2006) but become much stronger when seen in transmission (Pallé et al., 2009) or in denser O₂ atmospheres (Misra et al., 2014a). O₂ is accessible to planned transmission spectroscopy and direct imaging at 0.76 and 1.27 μm, if the telescope and star will support it. But the 1.27 μm O₂/O₄ band may produce a stronger signal for planets orbiting late-type M dwarfs that have low output at wavelengths shortward of 0.8 μm, and so induce low planetary transmission or reflected light signals. Ground-based high-resolution spectroscopy may be able to detect O₂ on some habitable zone terrestrial planets orbiting M dwarfs by using the B- and A-bands in the visible spectrum (Snellen et al., 2015; Lovis et al., 2017).

Although obtaining well-constrained abundances for these O₂-bearing molecules may be challenging with first-generation observatories, observations of even the presence or absence of different bands of these molecules may help constrain O₂ abundances. The strong dependence of O₃ and O₄ features on O₂ concentrations could also be highly diagnostic of both chemistry and pressure. This combination of features can allow, for example, discrimination between atmospheres that contain 1% O₂, 20% O₂, and 90% O₂ (Section 4.4).

O₃ features in the UV could also point to the presence of O₂ in atmospheres with lower levels of O₂, when O₂ itself is not detectable. Figure 3 shows Proterozoic Earth with 0.01 bar of CO₂, 0.0003 bar of CH₄ and 0.1% the PAL of O₂, as suggested by recent studies of chromium isotopes in geological samples from the mid-Proterozoic (∼1.8–0.8 Ga; Planavsky et al., 2014b). O₂ itself is extremely weak in the spectrum, and is not discernable in this plot, but O₃ produces a relatively strong UV (Hartley) band from 0.2 to 0.3 μm that is not saturated, unlike the present Earth's, and so there is still a relatively high signal in the bottom of the band. Consequently, obtaining an accurate measurement of the flux level at the bottom of the band, needed to quantify abundance, is potentially easier than for current Earth-like conditions.

Searching for O₂ in the atmosphere of Proxima Centauri b with ground-based telescopes may be possible in the next 5 years. Perhaps the most rapid path to this goal was suggested by Lovis et al. (2017), who proposed upgrades and the development of a coupling interface between the SPHERE high-contrast imager and the new ESPRESSO high-resolution spectrograph on the 8.2 m very large telescope. The combined high-contrast/high-resolution technique would provide the desired angular separation and contrast sensitivity of 10⁻⁷ in reflected light. This would make it possible to probe the O₂ bands at 0.63, 0.69, and 0.76 μm, along with water vapor at 0.72 μm and CH₄ at 0.75 μm. Lovis et al. (2017) calculated that a ∼4-sigma detection of O₂ on an Earth-like Proxima Cen b could be made in about 60 nights of telescope time, spread over 3 years to observe the planet at maximum separation from the star. Even better results for the M dwarf habitable zone planets within 8 pc are likely possible with the ELTs slated for first light in the mid-to-late 2020s (e.g., Rodler and López-Morales, 2014).

Future large space telescopes are being designed to detect O₂ for exoplanets orbiting F, G, and K stars. The design requirements for the HabEx and LUVOIR observatory concepts include the ability to detect O₂ in the atmospheres of neighboring Earth-like exoplanets (e.g., Dalcanton et al., 2015; Mennesson et al., 2016). Whether a planet's O₂ spectral features are in practice detectable with these observatories will depend on several parameters, including the planet's O₂ abundance and the distance of the planet–star system to Earth. For example, an Earth-like exoplanet orbiting a Sun-like star at 20 pc would require integration times exceeding hundreds of hours to detect the O₂ A-band with sufficient S/N for even a 6.5 m (JWST-sized) optical telescope with 2 λ/D IWA coronagraphic capability. For planet–star systems farther than about 20 pc, this important spectral feature would be unobservable. A larger telescope (12.7 m) could detect O₂ out to about 40 pc given the same IWA—although, again, requiring very long integration times. Figure 4 shows the integration time as a function of wavelength centered on the O₂ A-band required for a 15 m LUVOIR-class telescope to obtain S/N = 10 for a planet identical to modern Earth orbiting a solar twin at a distance of 10 pc. The spectral resolution assumed here is 150. For these assumptions, 10 s of hours are required to obtain S/N = 10 across most of the wavelength ranges shown. Figure 5 shows the spectrum for the planet that could be obtained across the UV-VIS-NIR wavelength range in 30 h per coronagraphic bandpass for this same observatory. The significant increase in the size of the error bars at longer wavelengths is due to thermal emission from the telescope, which is assumed to be heated to 270 K. The spectral resolutions assumed for the UV-VIS-NIR channels for both Figures 4 and 5 are R = 20 for the UV (λ <0.4 μm), R = 150 for the visible (0.4 < λ < 0.85), and R = 100 in the NIR (λ > 0.85). Both were simulated by using a coronagraph noise model based on the one described in Robinson et al. (2016).

4.4. Discriminating false positives

Once O₂ is detected in a planetary spectrum, searching for additional observed environmental features to place the O₂ in context will aid in discriminating false positive O₂ from true biological O₂. Hereunder we describe currently known methods for observational discriminants for abiotic and biological O₂.

4.5. Considerations for super-Earths

The majority of the mentioned discussion assumed that the habitable planet being observed was close to Earth size, but terrestrial planets may have radii up to 1.5 times that of Earth (Weiss and Marcy, 2014; Rogers, 2015; Fulton et al., 2017). These super-Earth atmospheres may be less Earth-like, as the larger masses of terrestrial super-Earths increase their ability to retain lighter molecules, such as H₂, and/or to produce substantial outgassing of heavier gases such as H₂O and CO₂. However, if their atmospheres are more Earth-like, the increased radius and gravity of larger planets will affect the detectability of O₂ (c.f. Rauer et al., 2011) and false positive discriminators to different degrees. For direct imaging, a larger planet will require smaller integration times for detection of spectral features, roughly proportional to 1/R_p² such that a 1.5 Earth radius super-Earth needs a little less than half the exposure time of an Earth radius planet. Enhanced gravity will decrease the scale height of the planetary atmosphere, as scale height H is proportional to 1/g. This may allow a planet to emit to space at a higher temperature (Kopparapu et al., 2014) potentially increasing detectability at thermal wavelengths, but will not affect the observability of features at nonthermal wavelengths, assuming the overall surface pressure of the atmosphere remained constant.

For transmission, however, the decrease in scale height due to enhanced gravity would decrease the size of molecular features in the spectrum and, therefore, their detectability. However, the increase in planet radius partially acts to offset this effect. After the scaling relationship for the size of molecule features observed in transmission (Winn, 2010), it can be shown that the size of molecular features in the atmosphere of a super-Earth scales inversely with the ratio of the density of the super-Earth to Earth. If we again assume a super-Earth with a 1.5 Earth's radius, and a mass of 4 Earth masses, then the size of molecular features in the transmission spectrum would be 84% that of an Earth-sized planet of similar atmospheric composition.

5.1. Planet detection

As discussed, one of the first challenges will be to detect the planet and discriminate it from potential background sources, which include “exozodiacal dust” (“exozodi”) from the planetary system dust disk, and background stars and galaxies. This could be done either with multiepoch photometric observations—revisits to the planetary system spread out over a longer time period—or with multiwavelength imaging in a single visit. Multiepoch observations will acquire the position of the planet against the background sky, which will more robustly discriminate the planet from background sources. These observations will constrain the planet's orbit and help determine mass and orbital parameters such as semimajor axis and eccentricity (Stapelfeldt et al., 2015), which are crucial for determining whether the planet is in the habitable zone of its parent star. An alternative, potentially faster method of planet detection would be to use a single visit with multiwavelength imaging to look for point sources near the star. Additional information could then be used to determine whether these sources are more or less likely to be planets or background objects. For example, point sources detected within the exo-zodi are more likely to be planets than sources detected outside the disk. During this detection phase, photometric colors—or low-resolution spatially resolved spectra across the field of view—could enhance our ability to discriminate planets from astrophysical noise sources such as stars, brown dwarfs, and galaxies (Seager et al., 2015), but a definitive determination of whether the planet is in the HZ would not be possible with a single visit.

5.2. Preliminary characterization

If the planet is detected in a single visit, then planetary photometric colors could be used to not only discriminate planets from background noise sources but also to attempt to discriminate between different types of terrestrial or more volatile rich (e.g., mini-Neptune) planets. However, planetary environments are rich and complex, especially terrestrial planets that may display inhomogeneous surfaces, volatiles in multiple phases, and a broad diversity of atmospheric compositions. Colors would, therefore, provide only a possibly ambiguous initial classification (e.g., Crow et al., 2011; Livengood et al., 2011; Hegde and Kaltenegger, 2013), because strong degeneracies in environment types that fit a given region of the color–color diagram likely exist (Krissansen-Totton et al., 2016b), and the overlying atmosphere and clouds can modify surface colors (Schwieterman et al., 2015a). Therefore, a potentially more valuable initial method of discriminating between planetary targets would be to obtain low-resolution (R ∼ 10) spectra to search for molecules with broad absorption features, such as water vapor and O₃ in the visible and NIR spectra. Although neither the H₂O nor the O₃ Chappuis band would be uniquely diagnostic of oceans or life, respectively, the detection of either of these gases would be consistent with habitability and biology. Water vapor features are present near the O₂ A-band at 0.72 and 0.85 μm, and at increasing strength at 0.95, 1.14, and 1.4 μm.

If multiepoch photometric/astrometric observations are used for detection, then the initial characterization could include determining the planet's orbit and, therefore, semimajor axis and eccentricity. These parameters will help determine the planet's insolation, which will constrain climate models of the planet, and indicate whether or not the planet is in the habitable zone. Mass determination may be done by using or combining any follow-up or existing RV data with the observed position of the planet. The orientation of the dust disk may also help break the RV degeneracy between mass and inclination of the planetary system, although an orbital solution for the planet will do this also. At this stage, direct constraints on planetary radius will only be possible if the planet transits, or if ground- or space-based MIR observations are able to observe the target (e.g., Quanz et al., 2015). Size or mass may help indicate planets that are more likely to be terrestrial (Rogers, 2015; Fulton et al., 2017), and size and mass will constrain its density.

Time-resolved, multiwavelength photometry, even in the limited wavelength range of a single 20% coronagraphic bandpass, and for a single visit, could be used to map the planet and search for surface inhomogeneities that may indicate a terrestrial planet (Cowan et al., 2009; Kawahara and Fujii, 2010; Lustig-Yaeger et al., 2017). This mapping could also take advantage of time-resolved segments of a long-exposure spectroscopic observation whose primary purpose is to search for atmospheric constituents (Lustig-Yaeger et al., 2017). For multiepoch observations, photometry at NIR wavelengths (optimally at 0.8–0.9 μm, and/or between water bands in the 1.0–1.4 μm region) comparing brightness at different points in the planet's orbit near gibbous and crescent phases could be used to search for ocean glint (Robinson et al., 2010, 2014). Glint observations may be challenging; observing the planet when half-illuminated produces a maximum separation between star and planet, but the crescent phase occurs at a close apparent separation between star and planet, when the planet is, therefore, more likely to be interior to the IWA. In addition, the inclination of the orbit of the planet to the observer must be closer to edge-on than face-on, otherwise glint will not be observed. Even so, LUVOIR-class missions with 16 m mirrors may be able to search for glint for a subset of their terrestrial planets (Meadows et al., 2018). Detection of any of these planetary characteristics—terrestrial mass or size, interesting maps, glint, or hints at the presence of water or O₃—would be enough to motivate further observations at higher spectral resolutions (R ∼ 150).

5.3. The search for an oxygenic photosynthetic biosphere with high-resolution spectroscopy

Once a promising target has been identified, significant time could be invested in obtaining a high-resolution spectrum. At an R of ∼150, individual features from many molecules can be detected. O₂ should be sought, and if detected, deeper observations should be taken to constrain its concentration. The O₂ A-band at 0.76 μm is a narrow but deep feature at Earthlike concentrations, and the wavelength range and spectral resolution of potential future direct imaging observatories such as HabEx and LUVOIR will be designed to detect it in the atmospheres of our exoplanetary neighbors. Detecting the O₃ UV Hartley band (0.2–0.3 μm) may reveal the presence of O₂ when O₂ concentrations are otherwise too low to produce observable spectral features. This may be the case for planets like the Proterozoic Earth, where its 0.1% PAL O₂ is likely not detectable in reflected light in the visible–NIR spectra with currently anticipated instrumentation, but the O₃ Hartley band produces a deep feature for λ <0.3 μm (Fig. 3). As stars are fainter in the UV spectrum, longer integration times, wider bandpasses, or both, will be required to obtain high S/N observations of UV absorption features. The 30-h observation of an Earth twin planet at 10 pc with a 15 m telescope shown in Figure 6 returned adequate S/N in the UV, but for a spectral resolution of 20, significantly lower than the R = 150 assumed for the visible channel. If O₂ or O₃ is not detected, but H₂O is, this may not indicate a lifeless planet, but rather one with a different dominant metabolism (e.g., methane-producing), or an environment that leads to a false negative, as would have been the case for the early Earth.

5.4. Further characterization and elimination of false positives

Other molecules should be sought to characterize the planetary environment, including O₃, O₄, CO, CO₂, and CH₄, as well as clouds and aerosols. The presence of O₄, especially for a planet orbiting a late-type M dwarf, may indicate that any O₂ present is the result of ocean loss and H escape due to H₂O photolysis (Luger and Barnes, 2015; Schwieterman et al., 2016). In direct imaging, the strong O₄ collision-induced absorption features in the visible region (0.34–0.7 μm) would likely be detectable in the same observations required to detect the O₂ (0.76 μm), and could potentially be in the same coronagraphic bandpass. However, for transmission observations, O₄ bands in the NIR at 1.06 and 1.27 μm would be more detectable than those in the visible, as they are less affected by the increase in atmospheric opacity engendered by Rayleigh scattering at shorter wavelengths. Note also that ground-based high-resolution spectroscopy will not be able to observe the broad features of O₄ collisionally induced absorption, and so this false positive discriminator is not available to this technique.

If abundant CH₄ is observed with O₂, this would make most photochemical false positives less likely, as CH₄ is a sink for photochemically generated O₂. It is also less likely to build up to detectable levels in the massive O₂ atmospheres of ocean loss planets. Thus, detection of CH₄ in an atmosphere with O₂ features would make ocean loss and photochemical false positives less likely, and suggest a high O₂ replenishment rate and a likely biogenic source of the O₂. However, CH₄ features may be challenging to observe at modern Earth-like concentrations because methane's strongest features are longward of 2.5 μm, where telescopic thermal emission and IWA issues may preclude observation. Weaker bands in the visible region (0.7–0.9 μm) would require abundances of CH₄ in excess of those seen in Earth's atmosphere today, but are within the wavelength ranges anticipated for direct imaging spacecraft and ground-based high-resolution spectroscopy (Lovis et al., 2017). Indeed, as described in section 2.2, there may have been no period in Earth's history wherein both O₂ and CH₄ coexisted in high abundance (Olsen et al., 2016; Reinhard et al., 2017).

Photochemical generation of O₂ and/or O₃ from CO₂ photolysis (Hu et al., 2012; Domagal-Goldman et al., 2014; Gao et al., 2015; Harman et al., 2015) should be associated with high amounts of CO₂ and CO in the planetary atmosphere. CO₂ has a weak band near 1.6 μm, which is undetectable on modern Earth but would likely have a strong feature for atmospheres with enough CO₂ to produce O₂ or O₃ photochemically (Domagal-Goldman et al., 2014). CO₂ has a much stronger band at 4.2 μm, although the latter band will be difficult to access for warm telescopes (either ground-based or actively heated space-based telescopes) dominated by thermal radiation in this spectral region. CO also absorbs near 2.35 and 4.6 μm, where a warm (room temperature) mirror will also preclude observation. In addition, these longer wavelengths will require a large telescope diameter to avoid falling within the IWA for observatories cold enough to see them, or an extremely large starshade to provide suppression at those longer wavelengths. Transit transmission observations, however, from the ground and JWST may allow access to longer wavelengths in the near term, compared with direct imaging techniques, and ground-based observations may also be attempted for the 2.35 μm CO band (c.f. Snellen et al., 2013).

5.5. Detailed planet characterization and the search for secondary biosignatures

If all known false positive mechanisms for the presence of O₂ have been eliminated, and if photochemical models and the presence of sinks such as CH₄ suggest a high O₂ flux into the environment, then this would make a biogenic source more likely. This conclusion can be bolstered not only by ruling out false positives, but also by searching for secondary confirmation of the dominant metabolism that is believed to have been discovered. In the case of oxygenic photosynthesis, this could come in the form of secondary biosignatures such as detection of a red edge or other pigment-associated feature in multiwavelength, time-resolved photometric mapping (Lustig-Yaeger, 2017) or in seasonally varying concentrations of O₂, CO₂, or CH₄ (Meadows, 2008; Reinhard et al., 2016), or the detection of other biosignatures such as N₂O (Schwieterman et al., 2018). Eventually, other biosignatures and their false positives will be identified and sought, and the measurements obtained from any single planet will be placed in the broader context of similar measurements made on other worlds (Fujii et al., 2018; Walker et al., 2018), further informing our understanding of comparative planetology, and the prevalence of life in the Universe.