First taxonomic classification of Gaia DR3 asteroid reflectance spectra

The first clustering analysis of asteroids following the shape of their reflectance photo-spectra was carried out by Chapman et al. (1975) using 24 narrow filters between 0.3 to 1 µm over 110 asteroids. They obtained two groups, the C- and S-types, as they were “similar to Carbonaceous and Stony-metallic meteorites”, respectively. Ten years later, David Tholen, a PhD student at the University of Arizona (USA), increased the sample up to 600 asteroids using albedo and 8 narrow filters, ranging from 0.3 to 1.1 µm, later called the Eight Color Asteroid Survey (ECAS, Zellner et al., 1985). In Tholen’s taxonomy, Chapman’s S-group was further subdivided into five classes: S-, Q-, A-, R-, and V-types. The C-group was divided into nine classes, all having low albedo (< 0.15) visible spectra ranging from blue, neutral, to flat-red: B, C, F, G, D, T, and P classes. The NUV (below 0.45 µm) became critical to distinguish between some of those taxonomies, such as B and F types. In the last decades of the 20th century, the Charge Coupled Devices (CCDs) appeared in planetary sciences, with their higher sensitivity in visible and near infrared wavelengths, but lower sensitivity in the blue region. Bus & Binzel (2002b, a) doubled Tholen’s sample with their second phase of the Small Main-Belt Asteroid Spectroscopic Survey (SMASSII), a set of low-resolution spectral observations of 1,341 asteroids in wavelengths between 0.43 to 0.92 µm. Although they increased the sample of asteroids, they lost albedo, critical to disentangle classes such as E-, M-, P-type, as the information below 0.43 µm, needed to identify Tholen’s classes among the C-complex. More recently, machine learning and probabilistic methods have been dealing with classifying small bodies based on their reflectance spectra: Mahlke et al. (2022) developed a classification method that allows to use partial spectra over visible and near-infrared (NIR) and obtain the different probabilities of a given spectrum to belong to a certain taxonomy. Interestingly, they incorporate the albedo as Tholen did in the 80s. 

We used the reflectance spectra of asteroids provided by Gaia DR3 (Gaia Collaboration 2022). The reflectance spectra were obtained by dividing each epoch spectrum by the mean of the solar analog stars selected and then averaging over the set of epochs. The resulting spectrum consists of 16 reflectance values covering a wavelength range from 0.374 to 1.034 µm, with a resolution of 0.044 µm, and normalized to unity at 0.550 ± 0.025 µm. However, since its publication in June 2022, no taxonomy has been published yet. In this abstract, we present a summary of our recent effort to classify a large number (>10,000) of asteroids in the Gaia DR3.

The first step to obtain this classification was to study, understand, try to correct, and/or clean the effects of some artifacts present in the spectra. Summarizing, Tinaut-Ruano et al (2023) found and corrected an artificial reddening on NUV wavelengths due to the use of Solar Analogs (SAs) that were valid in the visible but not in the NUV. Gallinier et al (2023) found a systematic reddening in wavelengths between 0.7 to 0.9 µm. Tinaut-Ruano et al (2024) have found a correlation between the NUV slope with the signal-to-noise ratio (SNR), and in this work, we found the same effect in the NIR. We have corrected the effect of the SA and cleaned the sample using thresholds in the SNR, limiting the sample with complete spectra. We use albedo to combine both Tholen (the last one with NUV information) and Mahlke (the one with the larger number of objects) taxonomies. After the cleaning and crossmatch with albedo databases, we ended with a sample larger than 15,000 asteroids. 

Following an iterative clustering process, we were able to classify more than 11,000 asteroids in 13 different taxons: S, V, A, K, E, M, P, D, C, G (Ch), B, F, and L-types. This is three times larger than any previous spectral taxonomic classification. We will show in the presentation how those taxons are distributed among different families and dynamic populations, as well as the typical features and physical properties of each taxon. Our taxonomic classification has the following advantages: i) the use of the NUV and albedo; ii) a large number of objects classified systematically in different dynamical populations and families; iii) some taxonomies with a limited number of known objects have increased their members substantially. However, this taxonomy can not be used to compare spectra from different sources, as some of the artifacts remain in the Gaia spectra. See the consistent reddening in the median Gaia spectra between 0.7 and 0.9 µm in the following figure. There, we compare the median spectra of the objects classified from Gaia (blue lines) and the references from Mahlke 2022 (red dotted line). In the title of each subplot is the number of spectra classified for each taxonomy.

 

Some examples of achievements of this taxonomy are: i) we found differences in the NUV between F-types and B-types as expected. Their classification is supported by their presence in different families, F are located mainly in Nysa-Polana, B in Themis, Iduna and Pallas; ii) we have classified almost 700 objects from Eos family as K type with a high confidence, this is 70 times more than were previously classified from spectra; iii) We have classified almost 200 objects as L types increasing the previously classified spectra by a factor 10 of this taxon. Furthermore, we will show some preliminary results derived from this taxonomic classification, such as the composition of the families and the link to the parent bodies.

All in all, this classification provides a unique opportunity for the community to explore the composition of different families, dynamic populations, or, in the other way around, to explore how different taxons distribute in the solar system.