Natasha E. Batalha

Radiative transfer calculations are essential for modeling planetary atmospheres. However, standard methods are computationally demanding and impose accuracy-speed trade-offs. High computational costs force numerical simplifications in large models (e.g., general circulation models) that degrade the accuracy of the simulation. Radiative transfer calculations are an ideal candidate for machine learning emulation; fundamentally, radiative transfer is a well-defined physical mapping from a static atmospheric profile to the resulting fluxes, and high-fidelity training data can be created from first-principles calculations. We developed a radiative transfer emulator using an encoder-only transformer neural network architecture, trained on 1D profiles representative of solar-composition hot Jupiter atmospheres. Our emulator reproduced bolometric two-stream layer fluxes with mean test set errors of ∼1% compared to the traditional method and achieved speedups of more than 100× in batched GPU inference. In matched comparisons against a long short-term memory model (LSTM) baseline, a transformer was consistently the more accurate and more robust architecture when training data were limited, while the LSTM achieved better accuracy at full scale. Emulating radiative transfer with machine learning opens up the possibility for faster and more accurate routines within planetary atmospheric models.

JWST has provided unprecedented access to ultracool brown dwarfs and has pushed the boundaries of directly imaging temperate giant planets. As we continue to push toward detecting sub-Saturn and Neptune-like planets, it is crucial to develop atmospheric and evolutionary models that better capture the complexity and diversity of planetary atmospheres similar to the gas and ice giants in our solar system. We present Sonora Flame Skimmer , the next suite of cloud-free 1D atmospheric and evolutionary models in chemical equilibrium and disequilibrium probing colder temperatures (down to 50 K), smaller objects (down to log( g ) = 2), and a wide range of metallicities (10x subsolar to 100x supersolar) and C/O ratios (solar to 2.5x solar). Beyond expanding the physical parameter space of previous Sonora models, we update the opacities and evolutionary model framework from Sonora Bobcat , as well as the chemical treatment of volatiles (H ₂ O, CH ₄ , NH ₃ ) and carbon species such as CO ₂ from Sonora Elf Owl . For the evolution of these substellar objects, we find that high-metallicity atmospheres lead to slower cooling compared to solar metallicity, while the strength of vertical mixing ( K _zz ) has a negligible impact on the evolutionary tracks. At the highest metallicity explored here (100× solar), the deuterium-burning and hydrogen-burning minimum masses fall to 5.39 and 45.03 M _J , respectively. All the models presented here, including the atmospheric structure, chemical profiles, spectra, synthetic photometry, and evolutionary models, are publicly available.

WISE J0855–07 is the coldest known brown dwarf (estimated T _eff ∼ 250 K), offering a rare opportunity to probe atmospheric physics in the temperature and mass regime that overlaps directly with exoplanets. At these temperatures, modest perturbations to the thermal structure and chemical abundances can produce measurable changes in the emergent mid-infrared spectrum, making time-resolved spectroscopy a powerful way to test whether the atmosphere is longitudinally and temporally homogeneous. Using James Webb Space Telescope NIRSpec/G395M time-series spectra spanning ∼3 to 5.2 μ m over 11 hr, we analyze every third spectrum from the dataset with a series of time-resolved atmospheric retrievals using the Brewster retrieval framework. Using this approach, we model epoch-to-epoch changes in the atmospheric state that drive the observed variability, including the thermal structure and key molecular abundances. From the time-series retrievals, we find that the primary source of the observed spectral variability is driven by the variations in the thermal structure, which range from ∼1 to 3 and CO ₂ , further modulate the spectrum within specific molecular features. These measurements provide a retrieval-based view of atmospheric variability in an ultracold, planet-like object, directly constraining how thermal and chemical structure evolve on observable timescales.

Determining whether temperate rocky exoplanets orbiting M stars retain atmospheres is currently a central goal of exoplanet astronomy. To this end, the James Webb Space Telescope has begun searching for atmospheres on these worlds with MIRI secondary eclipse spectroscopy and photometry. Here, we develop a novel climate-constrained Bayesian inference framework that yields atmospheric pressure and composition constraints from these datasets, while accounting for planetary, stellar, and model uncertainties. Our approach fits observations with model spectra derived from self-consistent pressure-temperature profiles at radiative–convective equilibrium, thus maximizing the information extracted from the data and providing more robust inferences than retrievals that use parameterized pressure–temperature profiles. We demonstrate the framework on the existing MIRI LRS eclipse spectrum of LTT 1445A b (1.34 R _⊕ and T _eq ≈ 431 K). An atmosphere does not need to be invoked to explain the data, meaning a bare rock model produces an adequate fit. If the planet has an atmosphere, the 2 σ upper limits on surface partial pressures are ≲1 bar for an optically thin gas like O ₂ , N ₂ , or CO, ≲0.1 bar for CO ₂ , ≲10 ⁻³ bar for H ₂ O, and ≲10 ⁻⁴ bar for SO ₂ . Scheduled MIRI F1500W observations could detect one of the thicker atmospheres permitted by the existing data (1 bar O ₂ and 0.01 bar CO ₂ ), if a precision of 20 ppm or better is achieved. This case study demonstrates that climate-constrained Bayesian inversion can turn rocky-planet eclipse spectra into the quantitative constraints necessary to test population-level atmospheric retention hypotheses, like the cosmic shoreline.

HD 209458b is the canonical hot Jupiter: the first to have its atmosphere measured and the first to hint at the role of aerosols in exoplanet atmospheres through the muting of Na absorption signatures in the optical. Here, we present JWST Mid-InfraRed Instrument (MIRI) Low Resolution Spectrograph (LRS) transmission observations of HD 209458b from 5 to 12 μ m, directly measuring the absorption signatures of its clouds for the first time. The observations indicate the presence of magnesium silicates, most likely Mg ₂ SiO ₄ or a mixture of Mg ₂ SiO ₄ +MgSiO ₃ . We also present a new methodology to reduce observational data, whereby the analysis is formulated as a rule-based model with a tree structure, enabling key decisions to be identified and uncertain decisions to be incorporated into subsequent modeling. With this data reduction, and using a combination of ARCiS free retrievals and PICASO+Virga self-consistent forward models, we are able to show that amorphous Mg ₂ SiO ₄ clouds explain the LRS data to high significance over either a clear (Δ ln ( Z ) = 16.63) or gray cloud atmosphere (Δ ln ( Z ) = 22.26). By combining the LRS dataset with archival JWST NIRCam and Hubble Space Telescope optical and near-infrared observations, we are able to more robustly constrain the properties of the magnesium silicate condensates, finding particle sizes of approximately 0.1 μ m and atmospheric pressures of the clouds of roughly 1–10 mbar. Our results add to the growing detections of silicate clouds as a dominant atmospheric component of hot Jupiters, with the exact silicate species contextualizing the atmospheric chemistry and potentially the formation conditions of these planets.

Sub-Neptunes are among the most common type of close-in planets found in our Galaxy, yet their bulk composition remains largely uncertain; H-rich envelopes overlaying rocky cores, volatile-rich planets, and carbon-rich interiors all remain viable configurations for members of this population. Atmospheric characterization has been proposed as a means of distinguishing between these scenarios, but growing evidence suggests that sub-Neptunes may host molten atmosphere–mantle interfaces that could alter the composition of their atmosphere. We use the PICASO 1D climate model, coupled to interior-structure and magma–atmosphere chemistry frameworks to quantify how clouds alter the atmospheric and interior structure of sub-Neptunes. For temperate sub-Neptunes like TOI-270 d, we find that clouds can lead to ≥1000 K heating at depth (∼10 ⁴ bar) and ∼600 K cooling at shallow pressures (∼1 bar). This heating is very sensitive to the cloud sedimentation efficiency and, to a lesser extent, to metallicity. Most sub-Neptunes in our sample should have a molten atmosphere–mantle interface, except TOI-1231 b and GJ 1214 b. For these two planets, cloudy models have a molten interface whereas clear models can allow for a solid boundary. Clouds can heat the atmosphere–mantle interfaces by a temperature difference between ∼1400 and 2600 K for most sub-Neptunes in our sample. Such cloud-driven heating can substantially change the composition of the interface with abundances of O ₂ , SiH ₄ , and SiO showing a ≥36% increase between cloudy and clear models of TOI-270 d. We discuss the implications of our results for the thermal evolution and measurements of intrinsic heat flux for this population.

The search for an atmosphere on the closest rocky M dwarf planet, LTT 1445 A b, has been the subject of intense investigation from both the ground and space. Here, we present the first JWST transmission spectrum of LTT 1445 A b using a single visit spanning 3–5 μ m using NIRSpec/G395H. We conduct two independent reductions of the data using both the Eureka! and ExoTiC-JEDI pipelines. Overall, we measure the NRS1 transit depths to a median precision of ∼23 ppm in 42 spectroscopic channels with uniform widths of 30 pixels (∼0.02 μ m), and the NRS2 transit depths to ∼36 ppm precision in 64 spectroscopic channels, also with uniform widths of 30 pixels. We rule out any statistically significant spectral features at this precision and place limits on atmospheric metallicity using a grid of chemical equilibrium models with grey opaque clouds. Using NIRSpec/G395H alone, we can place limits on the atmospheric metallicity of ≳350 × solar when the opaque pressure level is greater than 0.01 bar. We also conduct a combined analysis of JWST/NIRSpec and Hubble Space Telescope WFC3 transmission data and find our atmospheric limits can be extended ≳500 × solar when considering both datasets. Future analyses both in transit and emission will uncover whether there are detectable atmospheric features.

The James Webb Space Telescope and the future Habitable Worlds Observatory aim to discover exoplanet atmospheric spectra that detect life. Currently, most existing spectral “retrieval” algorithms focus on inferring the abundances of biogenic gases from these spectra. However, abundances are hard to interpret as signatures of life because they are modified by photochemistry, climate, and atmospheric escape. To address this problem, we develop a method for inferring the fluxes of gases at a planetary surface by inverting a coupled photochemical-climate model. As a proof of concept, we apply the approach to a synthetic 10-transit JWST NIRSpec Prism spectrum of TRAPPIST-1 e, assuming it hosts a biosphere similar to that of the Archean Earth. The retrieval confidently detects CO ₂ and CH ₄ and can constrain the flux of CH ₄ into the atmosphere to within approximately 1.5 orders of magnitude (68% CI), provided that TRAPPIST-1’s near-UV spectrum is accurately known. We demonstrate how inferred surface gas fluxes naturally fold into a probabilistic assessment of life, finding that ∼80% of the surface gas flux posterior is consistent with a CH ₄ -producing metabolism for our nominal test case. As with any inverse problem, these results are conditional on a number of assumptions in our forward model. Overall, we argue that increasing the robustness of life detection on exoplanets requires moving beyond atmospheric abundances toward inference of the surface fluxes that sustain them.

We present a JWST/NIRSpec G395H transmission spectrum of TOI-260 b, a T _eq ∼ 490 K, R _p = 1.76 R _⊕ planet. The transmission spectrum is derived by combining two transit observations, collected as part of the JWST Compositions of Mini-Planet Atmospheres for Statistical Study program. We achieved the same median transit depth precision of 37 ppm in both visits, as well as a median precision of 26 ppm when combining the spectroscopic light curves from the two visits. Implementing a 30 pixel wide ( R ∼ 200) spectroscopic binning scheme, we find that the transmission spectrum is mostly featureless, with a possible feature around 3.17 μ m. We assess the significance of any features in the transmission spectrum with a suite of nonparametric models, which confirm the presence of a potential feature in the NRS1 bandpass and an offset between the NRS1 and NRS2 detectors. To investigate the atmospheric composition of TOI-260 b, we run a series of PLanetary Atmospheric Tool for Observer Noobs retrievals. We do not detect any clear molecular signatures, but the combined data from the two visits are sufficient to constrain the atmospheric metallicity to greater than 200 × solar, assuming no opaque deck ≲2.5 mbar. We also investigate causes of the potential feature near 3.17 μ m; while we find some compatible gaseous species and cannot fully discard the possibility of an astrophysical origin, we suspect a systematics origin owing to the variance in strength and position of the feature. Overall, this look at TOI-260 b adds to the small sample of radius valley planets, which already seem to show a diversity in their atmospheric compositions. Determining the true nature of these enigmatic planets will require a larger telescope time investment.

We present a major update to the open-source atmospheric modeling package PICASO , designed for simulating the thermal structure and spectra of hydrogen-rich atmospheres of brown dwarfs and exoplanets. This release, PICASO 4.0 , expands upon the existing radiative-convective equilibrium model framework by incorporating several new capabilities. Key additions include the integration of Virga for self-consistent cloud modeling, new flexible treatments for rainout and cold trapping of volatile species, and support for photochemistry. We also introduce a parameterized energy injection scheme to simulate additional external or internal heating processes. These features are motivated by lessons from recent JWST observations that reveal the prevalence of nonequilibrium chemistry and clouds. We benchmark the new functionalities against previously published results in the literature, including the Sonora Diamondback grid, energy injected atmospheres, patchy cloud models, and other photochemical models of WASP-39b. PICASO continues to be actively developed as an open-source package aimed at enabling reproducible, community-driven atmospheric modeling of all substellar objects.