Table of Contents
Fetching ...

Gaia DR3 supervised classification of asteroid reflectance spectra

Marco Delbo, Thomas Dyer, Ullas Bhat, Chrysa Avdellidou, Laurent Galluccio, Amelia Minton

Abstract

We present a supervised, probabilistic taxonomic classification of asteroid reflectance spectra from Gaia Data Release 3 (DR3). Using high-quality Gaia DR3 spectra and a reference set of spectra from the literature consisting exclusively of asteroids with robust spectroscopic taxonomic types, we construct a principal-component (PC) representation of the Gaia reflectances. For each major spectral complex (C, S, X) and several end-member classes (B, D, A, L, K, V), we model the distribution of reference objects in PC space using multivariate kernel density estimation (KDE). This yields likelihoods for each spectral class and provides a quantitative measure of classification confidence. Validation against a sample of objects with known spectral classes demonstrates good performance for classes with distinctive reflectance signatures, including the S-complex, D, V, and A types. Spectrally continuous classes (B-C-complex, K-L-S-complex, and X-complex) show the expected degrees of mixing given the limited wavelength range of Gaia's spectrophotometry. We further explore the compositional structure of six major asteroid collisional families using our Gaia-derived spectral classes, finding excellent agreement with ground-based spectroscopy and revealing enhanced detections of olivine-rich A type material in the Flora and Eunomia families, as well as new insights into the spectral diversity of the Tirela family. The resulting catalogue constitutes a fully probabilistic taxonomic classification for the full Gaia DR3 asteroid sample. It offers a resource for studying the compositional structure of the main belt, identifying family interlopers, and linking asteroid populations to meteorite groups, and establishes a methodological framework for future Gaia releases, in particular for the validation of the Gaia DR4, expected by the end of 2026.

Gaia DR3 supervised classification of asteroid reflectance spectra

Abstract

We present a supervised, probabilistic taxonomic classification of asteroid reflectance spectra from Gaia Data Release 3 (DR3). Using high-quality Gaia DR3 spectra and a reference set of spectra from the literature consisting exclusively of asteroids with robust spectroscopic taxonomic types, we construct a principal-component (PC) representation of the Gaia reflectances. For each major spectral complex (C, S, X) and several end-member classes (B, D, A, L, K, V), we model the distribution of reference objects in PC space using multivariate kernel density estimation (KDE). This yields likelihoods for each spectral class and provides a quantitative measure of classification confidence. Validation against a sample of objects with known spectral classes demonstrates good performance for classes with distinctive reflectance signatures, including the S-complex, D, V, and A types. Spectrally continuous classes (B-C-complex, K-L-S-complex, and X-complex) show the expected degrees of mixing given the limited wavelength range of Gaia's spectrophotometry. We further explore the compositional structure of six major asteroid collisional families using our Gaia-derived spectral classes, finding excellent agreement with ground-based spectroscopy and revealing enhanced detections of olivine-rich A type material in the Flora and Eunomia families, as well as new insights into the spectral diversity of the Tirela family. The resulting catalogue constitutes a fully probabilistic taxonomic classification for the full Gaia DR3 asteroid sample. It offers a resource for studying the compositional structure of the main belt, identifying family interlopers, and linking asteroid populations to meteorite groups, and establishes a methodological framework for future Gaia releases, in particular for the validation of the Gaia DR4, expected by the end of 2026.
Paper Structure (12 sections, 3 equations, 5 figures, 3 tables)

This paper contains 12 sections, 3 equations, 5 figures, 3 tables.

Figures (5)

  • Figure 1: Kernel density estimate (KDE) contours in the PC1-PC2 space for different spectral classes indicated by the letters in each plot. Contours represent the estimated density from a Gaussian KDE, providing a smoothed view of the clustering of high-quality (S/N$>$50) reference set asteroids in principal component space for the given spectral class (scatter points). The background grey points represent the positions in PC1, PC2 space of asteroid with DR3 reflectance spectra with (S/N$>$50).
  • Figure 2: Cumulative distribution of the highest predicted class probability for all objects in the sample. Objects are sorted in descending order of their maximum class probability, and the cumulative number of objects is shown as a function of probability. This representation illustrates the confidence level of the classification, highlighting the fraction of objects with highly secure assignments and the gradual transition toward lower-confidence classifications.
  • Figure 3: Confusion matrix showing the performance of our supervised taxonomic classifier applied to Gaia DR3 asteroid spectra. The matrix displays, for each predicted class (rows), the fraction of classifications that correspond to each true taxonomic type (columns). Values are row-normalized so that each row sums to unity, providing a “precision-style” measure of class purity. Strong diagonal dominance is observed for B, D, S, A, and V types, while expected mixing appears within transitional or spectrally continuous groups such as B-C and K-L-S. This representation highlights both the robustness of the classifier for spectrally distinctive classes and the intrinsic overlaps among intermediate taxonomic types in the Gaia DR3 wavelength range.
  • Figure 4: Comparison of taxonomic class distributions for members of the asteroid families discussed in this work (see text). Each pie chart shows the fractional contribution of major spectral classes among family members with Gaia DR3 reflectance spectra with S/N$\geq$30. Left column: Gaia DR3 supervised classifications. Right column: Literature classifications based solely on spectroscopic observations (SPEC only), providing the highest-fidelity reference for comparison. Plots labelled A, B, C, D, E, and F correspond to the Flora, Eunomia, Adeona, Hungaria, Vesta and Tirela families, respectively
  • Figure 5: Distribution of geometric visible albedo for members of the Tirela asteroid family. Open histograms represent family members with previously known albedo values, while shaded histograms show members for which both albedo and taxonomic classes are derived from Gaia DR3 in this work. High-albedo objects are predominantly L type asteroids, whereas low-albedo objects are mostly D types.