Gaia DR3 supervised classification of asteroid reflectance spectra

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.