The source code and example notebook of Probabilistic Spectral Reconstruction of Trans-Neptunian Objects from Sparse Photometry: A Framework for Taxonomy, Survey Optimization, and Outlier Detection

JWST TNO Spectra Reconstructor from NIR Photometry A machine learning framework for reconstructing full Trans-Neptunian Object (TNO) spectra from sparse JWST NIRCam photometry. Overview This project provides tools to bridge the gap between sparse photometric observations and high-resolution spectral models. By leveraging a PCA-based manifold of known TNO spectra (from the DisCoTNO survey), we can reconstruct full reflectance spectra (0.75–5.1 µm) and estimate taxonomic classifications for objects with limited data, such as Neptune Trojans. Key Components Spectral Manifold: A PCA-based representation of spectral shapes that captures ~99% of variance in TNO reflectance. Reconstruction Model: Uses AutoGluon (TabularPredictor) with quantile regression to predict PCA components from filter photometry. Uncertainty Estimation: Employs correlated Monte Carlo sampling to generate spectral realizations and 95% confidence intervals. Taxonomic Classifier: An AutoGluon-based classifier that identifies object types (e.g., bowl, cliff, methanol-rich) from predicted PCA components. Filter Integration: Comprehensive handling of JWST NIRCam filter throughput curves and bandwidths.