NIRSpec/G395H: High-Res Exoplanet Spectroscopy

• NIRSpec/G395H is a high-resolution grating mode of JWST's NIRSpec that provides time-series spectroscopy over 2.87–5.14 μm, enabling precise exoplanet atmospheric studies.
• It employs a dual-detector setup with a SUB2048 subarray in BOTS mode, achieving spectral resolutions of R ~2700 and precisions of 18–36 ppm per bin.
• Data processing leverages multi-pipeline reductions and robust systematic noise corrections to extract molecular features such as CO, CO₂, and H₂O with high confidence.

NIRSpec/G395H is a high-resolution grating mode of the Near-Infrared Spectrograph (NIRSpec) on JWST, optimized for time-series spectroscopy in the 2.87–5.14 μm wavelength range. G395H underpins a substantial fraction of current advances in the atmospheric characterization of exoplanets and substellar objects, especially those in the critical 2–10 R⊕ regime. The unique spectral coverage and resolving power ($R \sim 2700$) of G395H allow the detection, quantification, and modeling of molecular features such as CO, CO₂, H₂O, SO₂, and SiO, while also placing strong upper limits on metal enrichment, cloud decks, and atmospheric mean molecular weight for atmospheres lacking robust spectroscopic features.

1. Instrument Configuration and Capabilities

NIRSpec/G395H operates in Bright Object Time Series (BOTS) mode with the F290LP long-pass filter, dispersing light onto two detectors: NRS1 (2.87–3.714 μm) and NRS2 (3.82–5.14 μm), separated by a characteristic ~0.1 μm gap. Standard slit configuration is S1600A1, typically paired with the SUB2048 subarray for time-resolved applications. The native resolving power is $R \approx 2700$, equivalent to $\Delta\lambda \sim 1.5$–$1.8 \times 10^{-3}$ μm per spectral element and a velocity resolution of $\Delta v \sim 110$ km s⁻¹.

Table 1. Representative G395H System Settings | Property | Typical Value | Detail | |-----------------------------|------------------------|----------------------------| | Wavelength range | 2.87–5.14 μm | NRS1: 2.87–3.72 μm; NRS2: 3.82–5.14 μm | | Resolving power ($R$) | ~2700 | Native—binned to lower $R$ as needed | | Readout pattern | NRSRAPID, 3–7 groups/integration | S/N and red noise performance scale with groups (see below) | | Slit | S1600A1 | 1.6″×1.6″ | | Subarray | SUB2048 | 32 rows × 2048 columns |

For bright targets, exposures employ 3–7 groups per integration to prevent saturation while retaining a high duty cycle. Photon-limited noise floors are typically achievable, with measured precisions of $\sigma \sim$18–33 ppm per 0.02 μm bin for super-Earth and sub-Neptune transits (Teske et al., 27 Feb 2025, Redai et al., 9 Jul 2025, Alderson et al., 29 Mar 2024, Wallack et al., 1 Apr 2024).

2. Data Reduction Methodologies and Systematic Noise

All contemporary analyses employ multi-pipeline data reduction for G395H, leveraging custom routines atop the JWST pipeline, including Eureka!, Tiberius, ExoTiC-JEDI, FIREFLy, and JexoPipe (Sarkar et al., 10 May 2024, Gordon et al., 22 Nov 2025, Wallack et al., 1 Apr 2024). Core steps include:

• Detector-level corrections: reference pixel subtraction, superbias, dark subtraction, linearity correction, cosmic-ray (jump) detection (thresholds 10–15σ), 1/f noise removal via column-median subtraction.
• Spectral extraction: aperture (box or optimal) extraction after trace location (Gaussian or polynomial fits), local background subtraction, masking and interpolation of flagged pixels.
• Wavelength calibration: 2D mapping via assign_wcs/extract_2d; accuracy $\sim$0.01 px.
• Systematics modeling: regression against trace centroids, time polynomials, engineering mnemonics, and, for longer visits, PCA decomposition to remove residuals from flexure, trace rotation, heater cycles, and alternating-column noise.

Systematic noise is mitigated by favoring higher group numbers per integration ($G \geq 4$), which reduces the amplitudes of principal component systematics (trace flexure, heater-cycle astigmatism) by up to 2–3x (Gordon et al., 22 Nov 2025). Red noise dominates for $G\leq 3$, especially in the blue-most region (2.8–3.5 μm). PandExo predictions typically underestimate measured error bars by 5–12% (Gordon et al., 22 Nov 2025). Inter-detector offsets ($\Delta_{\rm NRS1-NRS2} \sim 20$–70 ppm) are nearly ubiquitous and must be modeled in the final transmission spectrum (Sarkar et al., 10 May 2024, Teske et al., 27 Feb 2025).

3. Transmission Spectra and Performance Benchmarks

G395H transmission spectra for small planets frequently exhibit sub-50 ppm bin-to-bin scatter when binned to $\Delta\lambda\sim0.02$ μm or $R\sim200$–400 (Teske et al., 27 Feb 2025, Redai et al., 9 Jul 2025, Alderson et al., 29 Mar 2024, Wallack et al., 1 Apr 2024). Absolute transit depth uncertainties per bin routinely reach 18–36 ppm, with systematics-limited floors for bright host stars. Empirical performance on well-behaved targets compares favorably to photon-noise expectations.

• Planet properties: $\sim$2 R⊕, $T_{\text{eq}} \sim 400$–900 K, $M \sim 7$–10 M⊕
• Spectroscopic precision: 18–36 ppm per bin (NRS1/NRS2), 0.02 μm bins
• White-light residuals: 130–170 ppm post-detrending
• Transmission feature threshold: $\gtrsim$100 ppm at 3$\sigma$ significance
• Limits on mean molecular weight ($\mu$): $\gtrsim$6–8 g/mol; atmospheric metallicity required for detectability frequently $>$175–300$\times$ solar for clear atmospheres (Teske et al., 27 Feb 2025, Redai et al., 9 Jul 2025, Alderson et al., 29 Mar 2024, Wallack et al., 1 Apr 2024).

For hot/warm Jupiters and Neptunes, G395H yields robust molecular feature detections—H₂O, CO₂, CO, SO₂—with significance up to $\Delta \ln B \sim$ 11–85 (%%%%24$G\leq 3$25%%%%), and percent-level abundances (Ahrer et al., 16 May 2025, Gressier et al., 19 Sep 2025, Alderson et al., 2022, Grant et al., 2023).

4. Signal Extraction: Model Fits, Cross-Correlation, and Atmospheric Retrieval

Transmission light curves are modeled as $F(t,\lambda) = S(t) \cdot T(t,\lambda)$, with $S(t)$ encompassing systematics and $T(t,\lambda)$ the astrophysical transit (batman, quadratic limb darkening). Limb darkening is drawn from tailored grids (MPS-ATLAS, Stagger). Fits combine MCMC (emcee) and Levenberg-Marquardt; "prayer-bead" error inflation is standard for phase-curve and highly correlated datasets (Luque et al., 4 Dec 2024, Gordon et al., 22 Nov 2025).

Atmospheric constraints are extracted through forward modeling and nested sampling retrievals (UltraNest, PyMultiNest, TauREx, POSEIDON, HyDRA, CHIMERA). Physical grid fits (PICASO, Line et al.; (Teske et al., 27 Feb 2025, Redai et al., 9 Jul 2025, Ahrer et al., 15 Sep 2025, Lew et al., 8 Feb 2024, Ahrer et al., 16 May 2025)) span metallicity ($Z/Z_\odot$ = 1–1000), cloud-top pressure (1–10⁻⁶ bar), and mean molecular weight ($\mu$). Non-physical fits include "feature search" (flat line, slope, step, Gaussian) to statistically evaluate possible spectral structure.

Cross-correlation techniques (petitRADTRANS, HITEMP/HITRAN templates) enable robust molecular confirmations by searching over optimized wavelength intervals. Gaussian normalization of data/templates and narrow spectral windows prime molecular detection to SNR $\gtrsim$9 (H₂O), $\gtrsim$7.5 (CO), $\gtrsim$5 (CO₂). This technique efficiently detects minor species and isotopologues (e.g., $^{13}$CO) (Esparza-Borges et al., 2023, Esparza-Borges et al., 29 Sep 2025).

5. Key Science Results and Constraints

Featureless G395H spectra for the majority of studied sub-Neptunes and super-Earths strongly rule out low-metallicity, H₂-dominated atmospheres. For instance:

• TOI-776c: $M/H < 180$–$240\times$ solar ruled out for $P_{\text{cloud}} \gtrsim 10^{-3}$ bar; $\mu \gtrsim 6$–$8$ g/mol (Teske et al., 27 Feb 2025).
• GJ 357b: $M/H < 300\times$ solar and $\mu < 8$ g/mol excluded at $3\sigma$ (Redai et al., 9 Jul 2025).
• TOI-836b,c: $M/H < 175$–$250\times$ solar ruled out for clear atmospheres down to $\sim0.1$–$10^{-4}$ bar; flat spectra consistent with high metallicity and/or high-altitude aerosols (Alderson et al., 29 Mar 2024, Wallack et al., 1 Apr 2024).
• GJ 1132b: Two-visit repeatability analyses highlight noise-dominated spectral slopes and emphasize the necessity of multi-transit confirmation for $<$50 ppm features (May et al., 2023).

For hot Jupiters and Neptunes:

• WASP-94Ab: H₂O (4$\sigma$), CO₂ (11$\sigma$), tentative CO (3$\sigma$), H₂S (2.5$\sigma$), best-fit C/O $=0.49^{+0.08}_{-0.13}$, $Z=2.17\pm0.65\times$ solar (Ahrer et al., 16 May 2025).
• HAT-P-26b: H₂O (4$\sigma$), CO₂ (10$\sigma$), SO₂ (13.5$\sigma$); metallicity $=11.4^{+13.3}_{-8.1}\times$ solar, subsolar C/O (Gressier et al., 19 Sep 2025).
• WASP-121b: Thermal dissociation of H₂O/H₂ on dayside, robust SiO detection (5.2$\sigma$), validating 3D GCMs and hemispheric retrieval (Gapp et al., 2 Jun 2025).
• WASP-39b: CO₂ (28.5$\sigma$), H₂O (21.5$\sigma$), SO₂ (4.8$\sigma$), and, at native $R \sim 2700$, clear detection of CO sub-band structure (Grant et al., 2023, Alderson et al., 2022).

6. Best Practices and Limitations for G395H Exoplanet Spectroscopy

Instrument strengths include broad simultaneous 3–5 μm coverage, high spectral resolution ($R \sim 2700$), and negligible saturation risk in SUB2048. Stability-driven practices adopted by all major teams:

• Employ $\geq 4$ groups/integration for reduced red noise.
• Column-median 1/f noise subtraction performed at group or integration level (not pixel-level).
• Systematics modeling with trace position, rotation, PCA features, and, where available, engineering mnemonics for long-duration phase curves.
• Detrend and model inter-detector ($\Delta$NRS1–NRS2 $\sim$40–70 ppm) and visit-to-visit offsets.
• Multi-pipeline reductions and joint fits of light-curves across wavelengths and detector segments.
• Conservative error inflation via prayer-bead or GP methods for phase-curve and systematics-limited reductions.

Challenges for small planets center on:

• Morphological systematics primarily in NRS1 blue, $2.8$–$3.5$ μm, especially at low group numbers.
• Residual red noise (heater cycling, flexure, astigmatism) on minute-to-hour timescales, robust only after PCA correction (Gordon et al., 22 Nov 2025).
• Ambiguous degeneracies between high-metallicity atmospheres and high-altitude aerosols; breaking these requires multi-transit campaigns ($>5$ visits, depending on target SNR and systematics; (Gordon et al., 22 Nov 2025)).
• Calibration uncertainties, primarily flat-field residuals ($\sim$2–6%) and slit-loss/pathloss.
• Modeling limitations for inhomogeneous targets and 3D chemistry (e.g., dayside thermal dissociation in WASP-121b).

7. Scientific Impact and Future Prospects

NIRSpec/G395H is now the definitive mid-infrared transmission spectroscopy tool at high resolution, delivering robust molecular detections and strong compositional constraints for small and giant exoplanets. Its observational legacy includes:

• De facto exclusion of primordial H₂/He envelopes for a large sample of super-Earths and sub-Neptunes, establishing the prevalence of compact, high-metallicity, or aerosol-dominated atmospheres below $\sim$10 R⊕ (Teske et al., 27 Feb 2025, Alderson et al., 29 Mar 2024, Wallack et al., 1 Apr 2024, Redai et al., 9 Jul 2025).
• Precision quantification of C/O, metallicity, and isotopologue ratios for brown dwarfs and giant planets (percent-level constraints possible for C/O and $^{12}$CO/$^{13}$CO; (Lew et al., 8 Feb 2024, Esparza-Borges et al., 2023)).
• Molecular identification spanning CO, CO₂, H₂O, SO₂, SiO, H₂S, and more, underpinning migration, formation, and photochemical histories (Ahrer et al., 16 May 2025, Gressier et al., 19 Sep 2025, Alderson et al., 2022, Gapp et al., 2 Jun 2025).
• Phase-curve studies accessing both transmission and emission at sub-20 ppm precision, mapping “Cosmic Shoreline” boundaries for atmosphere retention in rocky planets (Luque et al., 4 Dec 2024).

Ongoing limitations include residual instrument systematics and degeneracies in interpretation for flat spectra. Future multi-transit programs (COMPASS, BOWIE-ALIGN, DREAMS, TST) will refine metallicity vs. cloud/aerosol constraints and further exploit G395H for atmospheric retrieval in an expanding exoplanet census (Gordon et al., 22 Nov 2025).