A public dataset of Ariel simulated observations for developing exoplanetary atmosphere data reduction pipelines
A public dataset of Ariel simulated observations for developing exoplanetary atmosphere data reduction pipelines
Authors: Lorenzo V. Mugnai, Kai Hou Yip, Andrea Bocchieri, Ανδρέας Παπαγεωργίου, V. Batista, Orphée Faucoz, Angèle Syty, Tara Tahseen, Enzo Pascale, Ingo Waldmann Date: 2026-05-05 Paper ID: openalex:2605.03719
Summary
This paper presents a comprehensive, high-fidelity synthetic dataset for the ESA Ariel mission, designed to facilitate the development and benchmarking of exoplanetary atmosphere data reduction pipelines. By providing ground-truth labeled time-series spectroscopy, the authors enable researchers to stress-test data-driven detrending methods against instrumental and astrophysical systematics. Additionally, the paper establishes a deep learning baseline and critically assesses the impact of dataset shift on the performance of ML-based detrending in real-world scenarios.
Key Contributions
- Provides a comprehensive public dataset of simulated Ariel exoplanet atmosphere observations for benchmarking data reduction and detrending algorithms.
- Establishes a deep neural network baseline for time-series spectroscopic data reduction to enable method comparison.
- Identifies and quantifies the critical challenge of dataset shift in ML-based detrending when training and operational distributions diverge.
Open Questions & Future Work
Archivist Review
I have approved the simulated dataset as a valuable, reusable resource for benchmarking detrending algorithms. I also approved the open question regarding generalization under domain shift, as it specifically highlights the systemic risk to scientific inference in survey-class missions. The uncertainty quantification question was rejected as a general ML topic rather than a central, unique challenge specific to the scientific domain presented.
Approved Open Questions
- Generalization under domain shift: Scientific inference in survey missions depends on algorithms that generalize across diverse populations; failure to handle distribution shift directly compromises the homogeneity and reliability of the survey results.
Rejected Candidates
- [open_question] Reliable uncertainty quantification (
reliable-uncertainty-quantification) - duplicate_existing: Uncertainty quantification calibration is a general ML problem already well-represented by existing literature and vault themes.
Datasets
Links
Metadata & Links
- url
- https://arxiv.org/abs/2605.03719
- paper_id
- 2605.03719
- paper_source
- openalex
- domain
- computer-vision
- tags
- computer-vision
- architectures
-
- datasets
- ariel-simulated-exoplanet-dataset
- concept_slugs
-
- dataset_slugs
- ariel-simulated-exoplanet-dataset
- skill
- TimeSeriesSkill
- created_at
- 2026-05-08T05:44:24Z