The James Webb Space Telescope (JWST) has introduced a significant era for exoplanetary science, specifically through the Near-Infrared Spectrograph (NIRSpec) and the Mid-Infrared Instrument (MIRI). Researchers utilizing the Seek Algorithm framework have begun applying probabilistic latent semantic indexing to differentiate between chemical species in the atmospheres of gas giants. This methodology, termed Exo-Atmospheric Semantic Mapping (EASM), treats spectral signatures not merely as isolated data points but as elements within a high-dimensional latent space. By analyzing data from the Transiting Exoplanet Community Early Release Science (ERS) Program, EASM facilitates a comparative assessment of the NIRSpec G395H and MIRI Low-Resolution Spectrometer (LRS) gratings.

Primary investigations focus on the detection and quantification of water vapor (H₂O) and methane (CH₄). The G395H medium-resolution grating provides a spectral range of approximately 2.87 to 5.14 microns, while the MIRI LRS covers the 5 to 12 micron regime. The integration of these disparate data sets allows for a detailed mapping of molecular absorption across the infrared spectrum. EASM employs Bayesian inference models to assign statistical probabilities to molecular concentrations, effectively filtering instrumental noise from true atmospheric signals by identifying correlated spectral motifs across multiple wavelength bands.

By the numbers

• Spectral Resolution (R):NIRSpec G395H operates at a resolution of approximately R~2700, whereas MIRI LRS provides a lower resolution of approximately R~100.
• Wavelength Coverage:Combined observations span from 0.6 microns (via NIRISS/NIRSpec) to 12 microns (via MIRI), covering critical vibrational-rotational transitions for volatile species.
• Noise Floor:The NIRSpec G395H instrument has demonstrated a noise floor in the range of 20 to 50 parts per million (ppm) per spectral bin for bright targets.
• Methane Bands:EASM specifically identifies the 3.3 μm and 7.7 μm bands as primary latent features for CH₄ detection.
• Water Vapor Features:Probabilistic mapping utilizes the 1.4, 1.8, and 6.1 μm bands to constrain H₂O abundance across diverse pressure levels.

Background

Traditional exoplanet atmospheric retrieval relies on forward-modeling software that compares observed spectra to pre-calculated chemical grids. While effective, these methods often struggle with the degeneracy of parameters, where different combinations of temperature, pressure, and abundance yield similar spectral profiles. Exo-Atmospheric Semantic Mapping (EASM) addresses this by incorporating probabilistic latent semantic indexing (PLSI), a technique originally developed for natural language processing to identify hidden topics within text documents. In the context of spectroscopy, "topics" are replaced by physical atmospheric components, such as specific molecular species or aerosol scattering properties.

The application of EASM to JWST data was prompted by the need to manage the unprecedented sensitivity and precision of the observatory's instruments. The Early Release Science Program provided a benchmark for testing these algorithms, particularly on the hot Saturn-mass planet WASP-39b. By mapping spectral features into high-dimensional latent spaces, researchers can identify latent variables that correspond to physical phenomena, such as thermal inversions or non-equilibrium chemistry, which might otherwise remain obscured by instrumental systematic errors.

The NIRSpec G395H Latent Space

The NIRSpec G395H grating is considered a flagship tool for EASM due to its high resolution and coverage of the carbon dioxide (CO₂) and carbon monoxide (CO) regions. However, its utility in mapping H₂O and CH₄ is equally significant. At a resolution of R~2700, the G395H configuration allows the Seek Algorithm to resolve individual line wings, which are critical for determining the vertical pressure-temperature profile of the planet. In the latent space, the 3.3 μm methane band manifests as a distinct cluster of high-probability vectors that must be disentangled from water vapor features occurring in the same region.

The EASM methodology uses non-parametric density estimation to assess the likelihood of these features. For NIRSpec data, the algorithm must account for detector-level artifacts, such as 1/f noise and the "snowball" effect of cosmic ray hits. By identifying spectral motifs that persist across multiple dither positions and exposures, the latent semantic indexer can assign a lower probability to transient instrumental features, thereby refining the uncertainty estimates for molecular abundances.

MIRI LRS and Mid-Infrared Correlations

MIRI LRS provides the long-wavelength counterpart to NIRSpec observations. Between 5 and 12 microns, the atmosphere transitions into a regime dominated by the 6.1 μm water band and the 7.7 μm methane band. The lower resolution of MIRI LRS (R~100) presents a challenge for semantic mapping, as the broader spectral features are more prone to overlap. To compensate, EASM utilizes kernel-based density estimation to find correlations between the high-resolution NIRSpec data and the lower-resolution MIRI data.

When a signal is detected at 3.3 μm in NIRSpec, the EASM algorithm searches for a corresponding signature at 7.7 μm in the MIRI data. If both features are present and statistically consistent within the latent space, the probability of methane presence is significantly upgraded. This cross-instrument correlation is vital for identifying biosignatures or trace gases like phosphine (PH₃), where a single-band detection might be dismissed as noise. The MIRI LRS noise floor is typically higher than that of NIRSpec, often influenced by the thermal background of the telescope itself, which EASM must model as a latent bias during the indexing phase.

Statistical Identification of Spectral Motifs

A core component of EASM is the differentiation between true atmospheric signals and stellar contamination. Transiting planets are observed against the backdrop of their host stars, and stellar activity—such as starspots or faculae—can introduce wavelength-dependent features that mimic atmospheric absorption. EASM addresses this by incorporating stellar spectral models into the latent space as a set of "noise motifs."

The algorithm utilizes Bayesian inference to perform a simultaneous fit of the stellar and planetary components. If a spectral feature at a specific wavelength correlates more strongly with the temporal variation of the star's brightness than with the planet's transit depth, the EASM system downgrades its significance in the planetary atmospheric profile. This strong quantification of uncertainty is what differentiates EASM from standard retrieval techniques, as it provides a mathematically rigorous way to account for the systematic errors inherent in high-resolution spectroscopy.

Implications for Planetary Formation Models

The ultimate goal of using the Seek Algorithm and EASM on JWST data is to refine models of planetary formation and evolution. The ratio of carbon to oxygen (C/O ratio) is a key indicator of where a planet formed within its protoplanetary disk. By accurately mapping H₂O, CO₂, CO, and CH₄ across the 0.6 to 12 micron range, EASM provides a high-fidelity measurement of this ratio.

Comparative latent semantic mapping has revealed that different instruments can lead to slightly different C/O estimates if analyzed in isolation. NIRSpec G395H data may lean toward one value due to its sensitivity to CO₂, while MIRI LRS might suggest another based on CH₄ features. EASM harmonizes these findings by constructing a unified high-dimensional latent space that accounts for the instrument-specific noise floors. This complete approach ensures that the resulting atmospheric parameters are based on the totality of the observed spectral fingerprints, leading to more strong conclusions about the habitability and origin of these distant worlds.