Testing the performance of cross-correlation techniques to search for molecular features in JWST NIRSpec G395H observations of transiting exoplanets

TL;DR

This work extends a cross-correlation framework to search for multiple molecules in JWST NIRSpec/G395H transmission spectra of WASP-39b, using template spectra from petitRADTRANS and three normalization strategies. It demonstrates robust detections of H$_2$O, CO$_2$, and CO—most strongly with Gaussian normalization—and confirms CO in prior analyses, while reporting non-detections for several other species. The approach includes a velocity-grid cross-correlation, baseline definition from template-template CCFs, and an MCMC-based significance assessment to account for transmission-spectrum uncertainties. Overall, the study establishes cross-correlation as a robust, computationally efficient tool for identifying molecular features in exoplanet atmospheres and provides a framework for future abundance inferences and targeted molecule searches.

Abstract

Cross-correlations techniques offer an alternative method to search for molecular species in JWST observations of exoplanet atmospheres. In a previous article, we applied cross-correlation functions for the first time to JWST NIRSpec/G395H observations of exoplanet atmospheres, resulting in a detection of CO in the transmission spectrum of WASP-39b and a tentative detection of CO isotopologues. Here we present an improved version of our cross-correlation technique and an investigation into how efficient the technique is when searching for other molecules in JWST NIRSpec/G395H data. Our search results in the detection of more molecules via cross-correlations in the atmosphere of WASP-39b, including $\rm H_{2}O$ and $\rm CO_{2}$, and confirms the CO detection. This result proves that cross-correlations are a robust and computationally cheap alternative method to search for molecular species in transmission spectra observed with JWST. We also searched for other molecules ($\rm CH_{4}$, $\rm NH_{3}$, $\rm SO_{2}$, $\rm N_{2}O$, $\rm H_{2}S$, $\rm PH_{3}$, $\rm O_{3}$ and $\rm C_{2}H_{2}$) that were not detected, for which we provide the definition of their cross-correlation baselines for future searches of those molecules in other targets. We find that that the cross-correlation search of each molecule is more efficient over limited wavelength regions of the spectrum, where the signal for that molecule dominates over other molecules, than over broad wavelength ranges. In general we also find that Gaussian normalization is the most efficient normalization mode for the generation of the molecular templates.

Figures (17)

• Figure 1: Transmission spectrum of WASP-39b obtained from the JTEC ERS Program NIRSpec G395H observations across NRS1 and NRS2 detectors extracted at pixel-level by the Tiberius pipeline. The right axis shows the relative opacity contributions of $\rm H_{2}O$, $\rm SO_{2}$, $\rm CO_{2}$ and $\rm CO$, which are expected to be the dominant spectral features in this spectral range in WASP-39b.
• Figure 2: Examples of the template extraction from the atmospheric models at JWST's spectral resolution (top panels) for three molecules, $\rm H_{2}O$ (left column panels), $\rm CH_{4}$ (middle column panels) and $\rm CO$ (right column panels), using three different normalization approaches. The second, third, and final rows show the resulting normalized templates using the polynomial fitting, Gaussian, and high-pass frequency filtering approaches, respectively.
• Figure 3: Definition of the baseline regions (shaded regions, see Table \ref{['tab:baselines']}) with respect to the template-template CCF within the optimal wavelength ranges using the Gaussian normalization on NRS1 (top panels), NRS2 (middle panels) and both ranges combined (bottom panels) for each detected molecule, i.e, $\rm H_{2}O$ (left column), $\rm CO_{2}$ (central column) and $\rm CO$ (right column). In Appendix \ref{['Appendix:baselines_freq']} we show an analogous plot considering the frequency filter normalization instead.
• Figure 4: Result of the cross-correlation search of $\rm H_{2}O$ along the NRS1 detector ($\rm 2.71 - 3.72~\mu$m) using the Gaussian normalization mode (top) and the frequency-filter normalization mode (bottom). The solid-dotted line shows the resulting data-template CCF and the dashed line shows the template-template CCF shifted towards the expected radial velocity offset produced by the barycentric velocities of WASP-39 system and JWST at the time of the observations.
• Figure 5: Results of the cross-correlation search of $\rm CO_{2}$ along the optimal wavelength range of NRS2 detector using the Gaussian normalization mode (top) and along the wavelength ranges of NRS1 and NRS2 detectors using frequency-filter normalization (bottom). The solid-dotted line shows the data-template CCF and the dashed line shows the template-template CCF.