Towards the Habitable Worlds Observatory: Retrieval of Reflection Spectra from Evolving Earth Analogues using 1D CNNs

Sarah G. A. Barbosa; Raissa Estrela; Paulo C. F. da Silva Filho; Daniel B. de Freitas; Lorenzo V. Mugnai

paper

Towards the Habitable Worlds Observatory: Retrieval of Reflection Spectra from Evolving Earth Analogues using 1D CNNs

arXiv:2508.00076

Abstract

The Habitable Worlds Observatory (HWO) will characterize Earth-like exoplanets from low-resolution reflected-light spectra. To assess how atmospheric evolution shapes what can be recovered, we trained a convolutional neural network on 109,945 spectra generated with the Planetary Spectrum Generator for Archean, Proterozoic, and Modern atmospheres under LUVOIR-B and HabEx/SS observing assumptions, separating the contributions of observational noise and model variability. We find that recovery depends jointly on abundance and spectral structure. Across eras, O $_{2}$ recovery weakens in Proterozoic atmospheres and CH $_{4}$ becomes reliable only when abundant, while an isolated UV band yields robust Modern O $_{3}$ recovery ( $R^{2} ≃ 0.98$ ) despite poor Modern CH $_{4}$ recovery ( $R^{2} = 0.32$ - $0.35$ ). Without being supplied feature locations, the network assigns $58$ - $71%$ of total attribution to visible wavelengths, while the ultraviolet remains decisive for O $_{3}$ and the near-infrared for CH $_{4}$ and CO $_{2}$ . Aggregate retrieval leverage declines beyond approximately $1.7 μ$ m, supporting a practical long-wavelength boundary in this setup. These wavelength dependencies also shape detectability. At baseline noise, $83$ - $85%$ of Modern O $_{2}$ cases satisfy a $3 σ$ criterion, while Proterozoic O $_{2}$ remains below $50%$ when noise is halved and Modern CH $_{4}$ remains below $3 σ$ as observational noise approaches zero because model uncertainty caps its significance. Together, these results imply molecule-specific priorities for HWO trade studies. Low-abundance O $_{2}$ retains considerable leverage from improved observational quality while motivating better inference, whereas at the 1,000-hour benchmark O $_{3}$ and Modern CH $_{4}$ show predominantly model-associated uncertainty in these networks, motivating broader training coverage and improved uncertainty calibration.

arXiv comment29 pages, 15 figures, 5 tables. Accepted for publication in RAS Techniques and Instruments. Data and trained models: https://doi.org/10.5281/zenodo.21959726; code: https://github.com/SarahBarbosa/geexhp