Data reduction & modelling
Every abundance or isotope ratio is only as trustworthy as the spectrum and the forward model behind it. Snellen (2025) is blunt about this: retrievals can return unrealistically small uncertainties when correlated noise, telluric residuals, or an inflexible temperature structure are ignored. HRS made molecular detections unambiguous by resolving lines (Snellen et al. 2010); the next challenge is making quantitative inference equally robust — especially for rare isotopologues that sit near the noise.
From photons to a spectrum
For VLT/CRIRES+ I use and contribute to excalibuhr (and pycrires): darks, flats, order tracing, nodding, and wavelength calibration. Close-in companions such as β Pic b and GQ Lup B need more than a standard extraction. After modelling residual starlight, I co-add spatial rows weighted by the planet’s cross-correlation strength so that faint lines are not drowned in scattered light. Archival SPIRou spectra for M dwarfs are quality-filtered and coadded with barycentric alignment, with error bars that reflect epoch-to-epoch variance rather than photon noise alone — the same spirit as the careful SPIRou modelling framework of Cristofari et al. (2022).
JWST high-contrast spectroscopy adds another contaminant: wavelength-dependent residual speckles. For HD 19467 B I fit a linear combination of IFU speckle templates together with the atmospheric model (A&A 2026), so continuum wiggles are not absorbed into molecular abundances.
Retrievals that can say “maybe not”
Forward modelling uses petitRADTRANS (Mollière et al. 2019) with flexible P–T nodes, FastChem equilibrium chemistry (or free abundances for cooler T dwarfs), and nested sampling. Bayesian evidence comparisons with and without $^{13}$CO or $\mathrm{C}^{18}\mathrm{O}$ are how weak detections get validated — or rejected. For young JWST targets with infrared excess, a blackbody disc ring and optically thin CO slab are fitted jointly with the photosphere (TWA 27A / TWA 28); otherwise disc emission can bias both C/O and isotopes.
Open tools
- excalibuhr — CRIRES+ reduction
- pyROX — opacities from Kurucz / ExoMol / HITRAN
- jaxcross — JAX cross-correlation
- iris — infrared isothermal slabs
- splinenorm — B-spline continuum normalisation
- retrieval_base — shared retrieval infrastructure