Correlated noise distorts exo-Earth retrievals

Nicole Wolff and a group of exoplanet researchers warned on September 2 that a common simplifying assumption in atmospheric retrieval work — that noise is independent from one wavelength bin to the next — can distort results for Earth-like planets. In simulated spectra for the planned Habitable Worlds Observatory, the team found that correlated noise can push retrieval codes toward two different “best” answers rather than one stable solution. One branch can look broadly Earth-like, while another can imply a denser atmosphere with different inferred properties. The paper, posted to arXiv as “Impacts of Correlated Noise on Retrievals of Exo-Earth Atmospheres,” lists Wolff, Bruce Macintosh, Tyler D. Robinson, Sarah Blunt, Jean-Baptiste Ruffio, Beck Dacus, Alex Madurowicz, Marshall Perrin and Laurent Pueyo as authors. The study focused on simulated ultraviolet, visible and near-infrared spectra and integrated a Gaussian-process model of correlated noise into the forward and inverse modeling tool `rfast`, according to the abstract. (astrobiology.com) ### What exactly is the problem the paper is flagging? The Habitable Worlds Observatory is being designed to directly image and characterize Earth analogs, but the authors said residual speckles and other instrumental effects can leave behind noise that is not random from pixel to pixel or wavelength to wavelength. In that case, the retrieval software can mistake structured noise for planetary information. Astrobiology’s September 2 summary said those correlations can hinder atmospheric characterization in high-contrast spectroscopy. (arxiv.org) Correlated noise matters because retrieval frameworks are built to infer atmospheric composition and planetary properties from faint spectra. The authors said many HWO retrieval studies still generate spectra without noise spectral correlations, and their paper set out to test what changes when those correlations are included explicitly. ### How did the researchers test it? The arXiv abstract said the team quantified the impact of correlated noise on inferred molecular abundances and planetary properties across different spectral resolutions, signal-to-noise ratios and noise correlation length scales. (astrobiology.com) The simulations were not observations of a newly discovered world; they were controlled tests designed to see how retrievals behave when the noise has known properties. (arxiv.org) The `rfast` retrieval framework was the main test bed. By injecting Gaussian-process noise with set characteristics, the researchers could compare the retrieved atmosphere against the known input atmosphere and measure when the fit drifted away from the truth. ### What does “bifurcated best-fit solutions” mean here? The Astrobiology summary said the retrievals can split into alternative fits that both appear plausible under the model. (arxiv.org) One solution can resemble an Earth-like atmosphere, while another can favor a thicker-atmosphere scenario. That is the central warning in the paper: the same noisy spectrum can support materially different atmospheric stories if the correlations are not handled correctly. A 2021 systematics study on exoplanet atmospheric retrievals reached a related conclusion in another context, finding that correlated noise can improve formal goodness-of-fit while degrading retrieval accuracy. That earlier work dealt with transmission spectra rather than HWO-style exo-Earth direct imaging, but it documented the same broad risk that structured noise can be overfit. ### Why does this matter for future exo-Earth missions? (astrobiology.com) The Habitable Worlds Observatory is meant to search for and characterize true Earth analogs, making retrieval reliability central to the mission’s science case. If correlated noise can generate multiple credible atmospheric solutions, then confidence in inferred gases, pressures or habitability-relevant properties depends not only on signal strength but also on whether the noise model is realistic. That is an inference from the paper’s setup and stated target mission. (arxiv.org) Another September 2 methods paper highlighted the same broader push for higher-fidelity simulation tools for HWO, describing the pyEDITH exposure-time calculator as support for next-generation habitable-planet searches. Together, the two papers point to a common near-term task: improving the realism of the modeling pipeline before mission data arrive. (astrobiology.com) ### What happens next? The September 2 preprint is now on arXiv, where the full methods, author list and simulation setup are available. Any follow-up will likely come through revised versions of the manuscript, conference presentations or future HWO retrieval studies that incorporate correlated-noise treatments into tools such as `rfast`. (arxiv.org) (astrobiology.com)