JWST observations of K2-18b can be explained by a gas-rich mini-Neptune with no habitable surface

To simulate a Hycean K2-18b, we first modeled a pressure-temperature (P-T) profile using the climate code contained within the Photochem software package (Wogan et al., 2023). The climate model uses correlated-k radiative transfer with opacities detailed in Appendix D of Wogan et al. (2023). The code constructs P-T profiles assuming the lower atmosphere follows a moist-pseudo adiabat connected to an isothermal stratosphere. For K2-18b, we assume a 215 K stratosphere following Hu (2021). Our approach can consider any number of condensing species (e.g., Graham et al., 2021), but H₂O is the most important condensable for a habitable K2-18b.

For a Hycean K2-18b, we nominally assume a 1-bar H₂-dominated atmosphere with a water-saturated troposphere to facilitate comparison with previous work (Hu et al., 2021; Madhusudhan et al., 2023a; Innes et al., 2023). For such a composition, our cloud-free climate model predicts that K2-18b would not be habitable because the surface temperature would exceed the critical point of H₂O (Figure 1a), consistent with past studies (Innes et al., 2023; Scheucher et al., 2020). However, it has been suggested that high-altitude clouds or hazes could potentially reflect short wave radiation allowing for a cooler climate (Madhusudhan et al., 2021; Piette & Madhusudhan, 2020). To approximate the cooling effects of clouds in our cloud-free climate simulations, following Hu et al. (2021), we arbitrarily reduce the incoming solar radiation by 30% which permits a $\sim 320$ K surface (Figure 1a). We adopt this habitable P-T profile, shown in Figure 1b, for all Hycean scenarios. Our photochemical simulations include up to %-level CH₄ and CO₂, but the Figure 1b P-T profile ignores their greenhouse contribution. For this analysis, this is justified because the climate of a Hycean K2-18b is uncertain, and our climate model predicts the surface temperature would increase by $\lesssim 10$ K when accounting for CH₄ and CO₂.

With our estimated P-T profile (Figure 1b), we then simulate steady-state photochemistry using Photochem (Wogan, 2024a). The photochemical model contained in Photochem solves a system of partial differential equations approximating molecular transport in the vertical direction and the effect of chemical reactions, photolysis and condensation. We have made several updates to the reaction network and thermodynamic data originally published in Wogan et al. (2023). We improved the kinetics and thermodynamics of $\mathrm{CH_{3}O}$, $\mathrm{H_{2}COH}$ and related species, all of which are detailed in Appendix Table A1. For key reactions, we nominally adopt new kinetics following Xu et al. (2015), but we also consider alternative rates from Klippenstein (2023). Theses updates are important for estimating photochemical methane production on Hycean worlds as discussed in Section 3.1. We have also updated our H₂O and H₂ photolysis data (Appendix Table A1). The updated network is available on Zenodo (see the “.yaml” files in the “input/” folder of Wogan (2024b)). Because the UV spectrum of K2-18 has not been measured, we instead use the UV spectrum of GJ 176 measured by the MUSCLES survey for our photochemical calculations (France et al., 2016) following the Hu et al. (2021) analysis.

We additionally model K2-18b as a gas-giant mini-Neptune with no habitable surface. We take the same approach as Hu (2021) and simulate the massive hydrogen atmosphere over two stages: The first considers the deep atmosphere (500 to 1 bar), and the second simulates the upper atmosphere (1 to $10^{-8}$ bar). The lower atmosphere stage captures the equilibrium-to-disequilibrium transition (i.e., gas quenching) that occurs deep in a gas-giant atmosphere (e.g., Zahnle et al., 2016), while the upper atmosphere model approximates the impact of UV photolysis and gas condensation on composition.

For the first stage, we use the PICASO climate model (Mukherjee et al., 2023) to generate a P-T profile with opacities appropriate for a $100\times$ solar metallicity with a solar C/O at chemical equilibrium assuming a geothermal heat flow consistent with an intrinsic temperature ($T_{\mathrm{int}}$) of 60 K (Hu, 2021). Note that the intrinsic temperature affects the upper atmosphere abundance of gases such as CH₄ (Fortney et al., 2020). We discuss this $T_{\mathrm{int}}$ dependence in Section 3.2, but leave a full parameter space exploration for future work. Next, using the P-T profile, we do a full kinetics simulation with the Photochem model between 500 bar and 1 bar using our network of $\sim 600$ reversible reactions described previously (Section 2.1). We fix the lower boundary to chemical equilibrium composition, and allow the kinetics model to predict the chemical equilibrium-to-disequilibrium transition. The deep atmosphere adopts an altitude-independent eddy diffusion coefficient of $K_{zz}=10^{8}$ cm² s^-1 following Hu (2021).

In the second stage we simulate K2-18b’s upper atmosphere using results from the first stage as lower boundary conditions. We do not use the PICASO P-T profile above 1 bar because PICASO assumes the entire atmospheric profile is at chemical equilibrium which would not be the case for the cool upper atmosphere of K2-18b. The chemical equilibrium assumption creates a stratospheric inversion in the P-T profile from greenhouse gases such as CH₄. Furthermore, PICASO assumes a dry convective lapse rate but the P-T profile in much of the upper troposphere should follow a moist-pseudo adiabat because of water condensation. As an alternative to PICASO, we extrapolate the P-T profile above the 1 bar level by drawing an adiabat upwards using the Photochem climate model until it intersects an isothermal 215 K stratosphere. Appendix Figure A1a compares the PICASO profile to the modified profile that we adopt. Finally, using the modified P-T profile, we compute the photochemical steady-state of the upper atmosphere (1 to $10^{-8}$ bar) to predict its composition. At the lower boundary, we fix all gas concentrations to the values predicted at the 1 bar level of the lower atmosphere kinetics simulation described in the previous paragraph. The upper atmosphere simulation assumes a Jupiter-like eddy diffusion profile as used in Hu (2021) (Appendix Figure A1b).

We use the PICASO code (Batalha et al., 2019) to compute the transmission spectra of simulated Hycean and mini-Neptune atmospheres adopting the R=60,000 resampled opacities archived on Zenodo (Batalha et al., 2022). The main text presents clear-sky spectra because the JWST data do not favor high altitude clouds. The Appendix shows the spectral effects of water, elemental sulfur (S₂ and S₈) and hydrocarbon clouds and hazes.

To investigate K2-18b as a Hycean world we first consider an uninhabited planet. Our nominal simulation, called “model 1”, assumes a 1 bar H₂-dominated atmosphere, 0.8% CO₂ fixed at the surface, and other settings and boundary conditions detailed in Table 1. We choose 0.8% CO₂ because it is the median concentration implied by the JWST spectrum (Madhusudhan et al., 2023b). Methane has a zero-flux lower boundary condition, therefore all accumulated CH₄ is the result of the photochemical reduction of CO₂ to CH₄. Figure 2a shows the steady-state composition of the model 1 atmosphere as a function of pressure revealing that only 0.4 part-per-billion (ppb) CH₄ is photochemically stable. Methane is slowly produced by the following sequence of reactions:

By analyzing column-integrated reaction rates we have determined that Reaction (1d) is the rate-limiting step. Another important path with the same net reaction replaces both reactions involving CH₃O with alternatives that depend on the isomer H₂COH:

Methane is effectively destroyed by photolysis followed by several oxidizing reactions:

In H₂-rich solar system atmospheres (e.g., Saturn’s), methane has a long lifetime to destruction because, after photolysis (Reaction (4a)), it recombines: $\mathrm{CH_{3}}+\mathrm{H}+\mathrm{M}\rightarrow\mathrm{CH_{4}}+\mathrm{M}$ (Moses et al., 2000). The same recombination is inefficient in model 1 because, unlike the gas-giants in the solar system, model 1 has substantially more oxidizing gases like H₂O. In particular, water vapor photolysis at Ly-$\alpha$ wavelengths ($\lambda=126.56$ nm) produces oxygen atoms (Reaction (4b), Slanger & Black, 1982) that rapidly oxidize CH₃ before CH₄ is reformed. Methyl is also destroyed by atomic oxygen sourced from a sequence of reactions involving CO₂ photolysis: $\mathrm{CO_{2}}+h\nu\rightarrow\mathrm{CO}+\mathrm{O}$, $\mathrm{H_{2}O}+h\nu\rightarrow\mathrm{H}+\mathrm{OH}$, and $\mathrm{CO}+\mathrm{OH}\rightarrow\mathrm{CO_{2}}+\mathrm{H}$ which has the net reaction $\mathrm{H_{2}O}+h\nu\rightarrow\mathrm{O}+\mathrm{H}+\mathrm{H}$. Overall, efficient methyl oxidation in additional to slow CH₄ production (Reaction path (1)), results in only trace amounts of atmospheric CH₄.

Our result that CH₄ cannot accumulate in model 1 is not sensitive to many model assumptions. For example, we have recomputed model 1 with vertically constant $K_{zz}$ between $10^{4}$ and $10^{6}$ cm² s^-1, N₂ concentrations ($f_{\mathrm{N_{2}}}$) between $\sim 1$ ppm and 1%, and troposphere relative humidities ($\phi$) spanning 0.1 to 1. Within this parameter space, our photochemical code predicts the maximum stable CH₄ concentration is only 4 ppb for $K_{zz}=10^{4}$ cm² s^-1, $f_{\mathrm{N_{2}}}=1$ ppm, and $\phi=1$. As an additional test, we recomputed model 1 using alternative rates for Reactions 1d and 2 derived by Klippenstein (2023) using ab initio methods. Klippenstein (2023) predicts that these important rate-limiting reactions are faster than the Xu et al. (2015) rates we nominally assume (Appendix Table A1) at the temperatures and pressures relevant to Hycean atmospheres. Despite this difference, our model using the Klippenstein (2023) rates predicts only 32 ppb CH₄.

Up to this point, we have modeled K2-18b as a habitable, yet uninhabited planet. Now we consider an inhabited case, which we refer to as model 2. Model 1 imposes the surface concentration of H₂, CO₂ and N₂, but most all other gases, including CH₄ and CO, are dictated by photochemistry. If K2-18b is a Hycean world inhabited by microbial life then CH₄ and CO could be biologically modulated gases like they were on the anoxic Archean Earth (Kharecha et al., 2005; Wogan & Catling, 2020; Thompson et al., 2022). Chemosynthetic methanogens can consume H₂ and CO₂ for energy, producing methane as a waste gas:

CO is also food for acetogenic microbes:

The produced $\mathrm{CH_{3}COOH}$ could have been food for acetotrophic methanotrophs ($\mathrm{CH_{3}COOH}\rightarrow\mathrm{CH_{4}}+\mathrm{CO_{2}}$). Model 2 simulates K2-18b as a Hycean world with boundary conditions representing biological influence from these early Archean metabolisms (Table 1). To model methanogenic life, we impose a surface CH₄ flux needed to replicate the %-level concentration implied by the JWST data, which ended up being half the modern Earth’s biological methane flux ($5\times 10^{10}$ molecules cm^-2 s^-1, Jackson et al., 2020). We also add a CO deposition velocity of $1.2\times 10^{-4}$ cm s^-1 to approximate the influence of CO-consuming acetogens (Reaction (6), Kharecha et al., 2005). At photochemical steady-state, model 2 has 2% CH₄ and a $\sim 10^{-5}$ CO mixing ratio at the surface (Figure 2b).

With a methanogenic biosphere, CH₄ can accumulate to the %-levels suggested by recent JWST observations (Madhusudhan et al., 2023b). In contrast, on an uninhabited Hycean K2-18b, CH₄ should be at only ppb-levels (model 1) because large concentrations can not accumulate photochemically and other non-biological sources of methane seem implausible (Section 4).

Figure 3 shows K2-18b modeled as a gas-giant mini-Neptune with no habitable surface (i.e., model 3 in Table 1). Deep in the atmosphere, at 500 bar and 1700 K, fast reactions enforce chemical equilibrium for our assumed composition of $100\times$ solar metallicity with a solar C/O ratio. As gases mix upward to lower pressures and temperatures, reactions slow, causing an equilibrium-to-disequilibrium transition (i.e., gas quenching). N₂ and NH₃ chemistry quenches near 200 bar and $\sim 1400$ K, and the CO₂-CO-CH₄ system fails to maintain equilibrium near 100 bar and $\sim 1250$ K. These quench points are broadly consistent with Hu (2021) who constructed similar mini-Neptune models of K2-18b.

Quenched gases from the deep atmosphere mix upward to Model 3’s stratosphere where they are relevant to transmission spectroscopy. The atmosphere has 4% CH₄ along with 0.06% CO₂ at 1 mbar, which is broadly consistent with recent JWST observations (Madhusudhan et al., 2023b). Water vapor condensation between 0.07 bar and 4 mbar reduces its concentration, causing only 0.3% of the gas to be present at 1 mbar. At the same pressure level, there is also 0.3% CO and 0.07% NH₃. Only trace photochemically produced SO₂ is present ($\sim 10^{-7}$ mixing ratio) as most all sulfur is photochemically processed to S₂ and S₈ in the lower atmosphere where it condenses out (Zahnle et al., 2016).

We additionally tested the sensitivity of Model 3 to the assumed intrinsic temperature ($T_{\mathrm{int}}=60$ K), as this parameter can impact deep atmosphere quenching and the resulting stratospheric abundances of CH₄, CO₂, and CO (Fortney et al., 2020; Tsai et al., 2021b). Larger $T_{\mathrm{int}}$ (e.g, 100 K) drives an increase in CO that is hard to reconcile with JWST observations. For lower $T_{\mathrm{int}}$ values (e.g., 30 K), our model does not produce enough CO₂ to explain the JWST data. Future abundance constraints from JWST offer an exciting avenue to study K2-18b’s internal temperature and thermal evolution. We leave detailed exploration of this topic to a future study.

Figure 4 shows the simulated clear-sky transmission spectra of three scenarios for K2-18b compared to JWST NIRISS and NIRSpec observations: A lifeless Hycean planet (model 1), a Hycean world inhabited by an Archean-like biosphere (model 2) and a $100\times$ solar metallicity mini-Neptune with no habitable surface (Model 3). In all cases, we allow the simulated spectra to have an offset between the NIRISS and NIRSpec data as to best fit the observations, motivated by Madhusudhan et al. (2023b).

JWST data rules out model 1 ($\chi_{r}^{2}=3.22$) because the lifeless Hycean world does not have enough methane ($\sim 0.8$ ppb, Figure 2) to explain the observed CH₄ absorption shortwards of $4$ $\mu$m. On the other hand, the data do not strongly exclude an inhabited Hycean world (model 2, $\chi_{r}^{2}=1.51$). Model 2 fits the CH₄ and CO₂ spectral features in the data because it has 2% of biologically-produced methane along with 0.8% CO₂.

However, an inhabited Hycean world is not required to explain the data. Our model of a gas-giant mini-Neptune (model 3) has a comparable fit ($\chi_{r}^{2}=1.51$) largely because of its 4% CH₄ and 0.06% CO₂ at $\sim 1$ mbar. The spectra show small H₂O absorption because water vapor is cold trapped at $\sim 4$ mbar (Figure 3). Also, NH₃ has small absorption features at 1.5, 2, and 3 $\mu$m. Madhusudhan et al. (2023b) used the JWST data to argue for an NH₃ upper bound of $\sim 3\times 10^{-5}$ at 95% confidence assuming a vertically constant NH₃ concentration. At 1 mbar, our mini-Neptune model has $7\times 10^{-4}$ NH₃, but photolysis rapidly diminishes the gas’s concentration towards lower pressures (see Figure 3). Using a transmission contribution function (Equation 8 in Mollière et al. (2019)) we find that the 3 $\mu$m NH₃ feature (Figure 4c) is sensitive to pressures between $\sim 10^{-3}$ and $\sim 10^{-5}$ bar where the ammonia concentration is between $7\times 10^{-4}$ and $10^{-14}$ mixing ratio. While a direct comparison is challenging, our modeled heterogeneous NH₃ profile appears broadly compatible with the vertically constant upper bound derived by Madhusudhan et al. (2023b).

Models 2 and 3, and the retrievals presented in Madhusudhan et al. (2023b), are not able to reproduce the apparent large absorption feature near $\sim 1$ $\mu$m (red points in Appendix Figure A2). Disregarding these six data points reduces $\chi_{r}^{2}$ values for both model 2 and 3 to about $1$. Future visits with NIRISS SOSS will be valuable to determine whether the scatter near $\sim 1$ $\mu$m is physical or instrumental.

Our conclusion that the data strongly rule out model 1 but not model 2 or 3 is unchanged when including various aerosol opacities (Appendix Figure A2). Accounting for aerosols, the model 1 simulated spectrum remains a poor fit ($\chi_{r}^{2}=2.33$) when compared to model 2 and 3 ($\chi_{r}^{2}\approx 1.4$).

Hu et al. (2021) pioneered the use of photochemical models to simulate K2-18b as a lifeless Hycean planet. They consider a 1 bar H₂ atmosphere, 1% N₂, a comparable P-T profile to ours (Figure 1b), and CO₂ concentrations between 400 ppm and 10%. In all scenarios, they predict the photochemical accumulation of %-levels CH₄, which contrasts with the ppb-level CH₄ we compute in very similar scenarios (e.g., model 1). There are two reasons our results differ. First, Hu et al. (2021) assumed that the photolysis of H₂O produces only OH + H. However, we demonstrate here that the seemingly minor channel that produces O + H + H (Reaction 4b) is important for CH₄ destruction in K2-18b’s atmosphere. To verify this insight, we have rerun the 400 ppm-CO₂ model of Hu et al. (2021) using their photochemical network and code, and with the sole inclusion of the O + H + H channel the steady-state CH₄ drops from 1% to $3\times 10^{-5}$ mixing ratio.

The second reason our results differ has to do with CH₄ production. Hu et al. (2021) modeled Reaction (1d), a critical step to methane formation, with its high-pressure limit rate constant. This approach can accurately estimate the rate at high pressures ($\sim 100$ bar) but substantially over predicts the rate at $<1$ bar where third body collisions are more scarce. We have also rerun the 400 ppm-CO₂ model of Hu et al. (2021) using their code, now also with the Xu et al. (2015) pressure-dependent rate constant in Appendix Table A1, and find that the steady-state CH₄ mixing ratio further drops to $2\times 10^{-7}$. This concentration is broadly consistent with model 1 shown in Section 3.1, given remaining subtle differences in the temperature, diffusivity, and radiative transfer.

To further test the above explanation, we have also used the Photochem code to reproduce the 400 ppm-CO₂ case in Hu et al. (2021). Adopting their boundary conditions, P-T profile, and eddy diffusion profile, our chemical network predicts $3\times 10^{-7}$ mixing ratio CH₄ at steady state. When we perform the same simulation but use the Hu et al. (2021) pressure-independent rate for Reaction (1d), our code predicts $10^{-4}$ mixing ratio CH₄ should accumulate. When Photochem also omits the O + H + H branch of H₂O photolysis, the CH₄ concentration further rises to $\sim 0.5\%$. These Photochem results are generally compatible with our Hu et al. (2021) code calculations for the same experiments.

Furthermore, Madhusudhan et al. (2023a) was unable to reproduce Hu et al. (2021) using an independent photochemical model and network. “Case 11” in Madhusudhan et al. (2023a) is very similar to the 10%-CO₂ case in Hu et al. (2021), representing a 1 bar uninhabited Hycean world (i.e., zero-flux boundary conditions for CH₄ and CO). At photochemical steady-state, Madhusudhan et al. (2023a) finds that only 55 ppb CH₄ should persist, which aligns with our conclusion that methane should be a trace gas (e.g., $<1$ ppm) on such a planet.

Yu et al. (2021) and Tsai et al. (2021a) also simulated a 1 bar H₂-dominated atmosphere on K2-18b, but instead with a hot $\sim 600$ K rocky surface (i.e., no habitable ocean). They find that $\sim 0.1\%$ to 1% CH₄ can accumulate alongside $\sim 1\%$ of CO₂ and CO. We have done similar simulations with the Photochem code and found that larger CH₄ concentrations are stable in this case because the hot $\sim 600$ K surface breaks down the kinetic barriers to CH₄ production. For example, when temperature is increased from 320 K to 600 K, the rate of the reaction $\mathrm{HCO}+\mathrm{H_{2}}\rightarrow\mathrm{H_{2}CO}+\mathrm{H}$ increases by about six orders of magnitude while Reaction (2) increases by a factor of $\sim 30$. In contrast, methane production is far more kinetically inhibited on a Hycean planet with a habitable 320 K surface (e.g., model 1). Furthermore, the 1 bar scenarios in Yu et al. (2021) and Tsai et al. (2021a) with rocky surfaces can be ruled out because such a planet would be denser than K2-18b’s observed density. To explain the planet’s mass and radius with only a silicate interior and a H₂-rich envelope, interior modeling suggests the atmosphere needs to be $\gtrsim 1000$ bars thick (Madhusudhan et al., 2020, 2023b).

Substantial methane from non-photochemical abiotic processes is hard to sustain on a Hycean planet. To explain K2-18b’s density, a Hycean world needs a large high-pressure ice layer that separates deep rocky material from the surface water ocean (Madhusudhan et al., 2021). Water-rock reactions and subsequent transport of CH₄ is conceivable (Thompson et al., 2022), but improbable in this case since the high overburden pressure of water and ice inhibits the production of fresh crust to be hydrated (Krissansen-Totton et al., 2021). Moreover, the shutdown of melting of deep-subsurface silicates also precludes the possibility of volcanic CH₄ (Krissansen-Totton et al., 2021; Noack et al., 2016; Kite & Ford, 2018). Massive asteroid impacts on the early Earth may have made transient atmospheric methane (Wogan et al., 2023). However, ephemeral impact-induced CH₄ is unlikely on K2-18b because the planet is $\sim 2-3$ billion years old (Guinan & Engle, 2019), while substantial bombardment is expected to end within the first several hundred million years of planet formation (Lichtenberg & Clement, 2022).

Our results suggest that both an inhabited Hycean world (model 2) or a mini-Neptune with a massive H₂ atmosphere (model 3) are not strongly ruled out by the JWST data ($\chi_{r}^{2}\approx 1.5$). However, in addition to evaluating the fit to the data, we also must assess the relative complexity of each scenario. The inhabited Hycean world (model 2) requires a cool habitable surface, but models suggest that a cloud-free $1$-bar H₂-rich atmosphere should trigger a hot runway greenhouse (Figure 1, Innes et al., 2023; Pierrehumbert, 2023). A supercritical steam-dominated atmosphere would have a small scale height incompatible with JWST observations (Scheucher et al., 2020). For a temperate surface, climate codes need to assume the presence of high-altitude clouds or hazes that scatter away starlight (Madhusudhan et al., 2021; Piette & Madhusudhan, 2020).

Beyond this climate paradox, $\sim 1$ bar of H₂ may also be susceptible to rapid escape driven by extreme-ultraviolet radiation (i.e., XUV, Hu et al., 2023). Even if a 1 bar H₂ atmosphere could withstand modern XUV radiation, K2-18b likely experienced exceptionally high XUV fluxes during the host M star’s pre-main sequence, potentially driving hundreds of bars of H₂ loss (Luger et al., 2015), as so a remnant thin $\sim 1$ bar atmosphere would be highly fortuitous. As noted previously, replenishing H₂ with volcanism would be unlikely on a Hycean K2-18b because rocky material in the deep subsurface would be at pressures too high for melting and outgassing (Kite & Ford, 2018; Noack et al., 2016).

In contrast, our model of a gas-giant mini-Neptune (model 3) is relatively straightforward. For a $100\times$ solar composition, solar C/O, and $T_{\mathrm{int}}=60$ K, which are physically plausible given K2-18b’s mass, a spectrum broadly consistent with the JWST data falls out of our model. Unlike a Hycean world, a mini-Neptune does not require a biosphere to explain the disequilibrium combination of atmospheric CH₄ and CO₂. Instead, these gases emerge in model 3 from deep-atmosphere quenching (Figure 3). Even though both an inhabited Hycean world and a mini-Neptune are allowed by JWST data, the climate of a Hycean world and the atmosphere’s resilience to escape is hard to explain, so we favor the mini-Neptune model for its simplicity.

While our mini-Neptune simulation is relatively simple, it makes assumptions that should be investigated with more sophisticated modeling. Namely, we simulate the planet’s climate (Figure 3) using a two-stage approach that is not fully self-consistent with photochemistry (Section 2.2), yet an accurate tropopause temperature is important for predicting whether H₂O cold-trapping can reproduce the JWST non-detection of water vapor. The water vapor cold trap would be better approximated by a model that is self-consistent with photochemistry and accounts for the possibility of convection inhibition (Innes et al., 2023). The need for a H₂O cold trap is not unique to a mini-Neptune planet. A Hycean world would also need substantial water condensation to explain the H₂O non-detection. An additional shortcoming of this study is that we only consider one mini-Neptune scenario with a composition of $100\times$ solar metallicity, and solar C/O. Further modeling could tune metallicity and the C/O ratio to get an even better fit to the JWST observations.

Clearly distinguishing between the inhabited Hycean and mini-Neptune interpretations with future JWST observations will be challenging. Ammonia should be unique to a mini-Neptune K2-18b (Figure 3, Yu et al., 2021; Tsai et al., 2021a; Madhusudhan et al., 2023a; Hu et al., 2021). However, even if future observations are unable to detect NH₃, this would not necessarily prove the inhabited Hycean case. In our mini-Neptune model, the NH₃ features act to fill CH₄ spectral windows (Figure 4). This small ammonia absorption is difficult to distinguish from clouds which have a similar effect on the spectrum (Figure A2). Additionally, there are several reasons why ammonia could be less abundant on a mini-Neptune K2-18b than we have estimated, making a detection even more challenging. Hu (2021) predicted stratospheric NH₃ could be photochemically depleted to undetectable concentrations ($<1$ ppm) on a gas-rich K2-18b if tropospheric mixing is slow ($\lesssim 10^{3}$ cm² s^-1). Also, nitrogen could dissolve into a magma ocean at the base of a thick H₂-rich envelope, preventing a large observable NH₃ abundance in the upper atmosphere (Shorttle et al., 2024).

An inhabited Hycean world could be identified with the detection of a biogenic gas. Madhusudhan et al. (2023b) found weak evidence for dimethyl sulfide (DMS) in K2-18b’s transmission spectrum, a gas almost exclusively produced by life on Earth (Catling et al., 2018). If DMS is detected with statistical significance, it might be difficult to account for its presence without a biosphere on a Hycean planet.