Simplified access of asteroid spectral data and metadata using classy

Remote-sensing spectroscopy is the most efficient observational technique to characterise the surface composition of asteroids within a reasonable timeframe. While photometry allows to characterise much fainter targets, the resolution of characteristic absorption features via spectroscopy makes this technique the method of choice for many analysis avenues. The main contributors to our spectral database of asteroids have been large surveys, such as ECAS (Eight Color Asteroid Survey, [1]), SMASS (Small Main-Belt Asteroid Spectroscopic Survey, [2]), PRIMASS (PRIMitive Asteroids Spectroscopic Survey, [3]), MITHNEOS (MIT-Hawaii Near-Earth Object Spectroscopic Survey, [4]), and ESA Gaia. A significant number of observations has further been produced by individual efforts, focusing e.g. on asteroid families (e.g. [5, 6], among many others) and particular populations (e.g. [7, 8, 9], among many others).  Currently, the number of visible / near-infrared / mid-infrared remote-sensing spectra of asteroids is around 70,000. This number will increase significantly in the years to come thanks to Gaia DR4 (expected in 2026) and the SPHEREx survey (to be launched in 2025, [10]). This is a positive development for the minor body community, and it has been recognised that, as the number of spectra continues to increase, we will require more sophisticated analysis methods to make best use of it. Numerous efforts have been undertaken to exploit these datasets with modern statistical treatment, for example, to identify clusters of asteroids (taxomic classes, [11, 12, 13, 14]), to invesitage asteroid-meteorite relationships ([15, 16, 17]), and for mineralogical characterization ([18, 19]). The recent strong increase in the number of asteroids with an observed spectrum further enables the independent confirmation of results obtained with other observables with literature spectra  (e.g. asteroid families from dynamical elements, [20]).

A necessary prior step for all aforementioned analyses is the collection of public spectra from survey databases for the targets under investigation. These databases include NASA’s Planetary Data System, the Centre de Données astronomiques de Strasbourg, and online repositories of Gaia, SMASS, and MITHNEOS. Achieving a complete literature look-up for any given target is therefore a tedious task, given the large number of repositories and the heterogeneous datasets, in particular when accounting for smaller repositories based on the individual observational efforts referenced above. With an increasing number of spectra, this effort will  increase. Two effects are visible: (1) Authors tend to visit one or two repositories and use an incomplete subset of the available spectra of a given asteroid. This hurts the scope of the analysis and therefore the results, as it is known that asteroid spectra can be subject to intrinsic (e.g. activity, [21]) and extrinsic (e.g. phase colouring, [22]) variability. (2) Important information like the metadata (for example, the epoch of observations and therefore phase angle of target) is commonly neglected in large-scale analyses given the large effort required to extract it.

The community could therefore benefit from a data-aggregator service focused on asteroid spectra. The classy tool aims to fill this gap by directly addressing the issue of accessibility of asteroid spectra. It itself is the product of a large, dedicated effort to build and homogenise a database of asteroid spectra, in preparation for a machine-learning appication to derive an asteroid taxonomy [12]. There are two main uses cases that classy aims to solve: (1) Search for and access of asteroid spectral observations based on properties of the spectra and the targets (e.g. retrieve all NIR spectra of asteroids in the Themis family with albedos < 0.05), and (2) to classify spectra in different taxonomic schemes (Tholen [23], Bus-DeMeo [11], Mahlke [12]). In addition, classy offers important functionality for common preprocessing of spectral data: truncating, interpolating, smoothing, and feature  parameterisation. These preprocessing parameters are stored in a database that can be shared among users, supporting consistent data treatment among collaboration members and the publishing of reproducible results. Users can further ingest private observations into their local classy database to work in unison with public data. classy is connected to numerous online repositories of asteroid spectra, enabling to retrieve and query among approximately 70,000 spectra. A large effort was spent to extract the observational metadata from articles to allow studies to make use of the epoch of observation and the phase angle of the target. classy is a python package with a command-line interface and a web interface.¹ It provides the important connection between repositories of spectra and users, greatly facilitating machine learning and other data-driven projects. It is actively developed, with a focus on stability and regarding the number of spectra that can be retrieved. Documentation is available online.²

¹ https://classy.streamlit.app/
² https://classy.readthedocs.io/

Acknowledgments: Without observations, there is not much to do. I thank all observers who choose to make their data available to the community.
Dr Benoit Carry provided important contributions to initial development of the classy database, in preparation of Mahlke et al. 2022.

References:
[1] Zellner, B. et al. (1985), Icarus, 61, 355-416
[2] Xu, S. et al. (1995), Icarus, 115, 1-35
[3] Pinilla-Alonso, N. et al. (2022), DPS, #504.09
[4] Binzel, R. et al. (2019), Icarus 324, 41-76
[5] Mothé-Diniz, T. and Carvano, J. M. (2005), A&A, 442, 727-729
[6] Marsset, M. et al. (2016), A&A, 586, A15
[7] Clark, B. E. et al. (2004), AAS, 128, 3070-3081
[8] de León, J. et al. (2012), Icarus, 218, 196-206
[9] Devogèle, M. et al. (2018), Icarus, 304, 31-57
[10] Ivezic, Z. et al. (2022), Icarus, 371, 114696
[11] DeMeo, F. et al. (2009), Icarus, 202, 160-180
[12] Mahlke, M. et al. (2022), A&A, 665, A26
[13] Penttilä, A. et al. (2022), A&A, 649, A46
[14] Klimczak, H. et al. (2022), A&A, 667, A10
[15] DeMeo, F. et al. (2022), Icarus, 380, 114971
[16] Mahlke, M. et al. (2023), A&A, 676, A94
[17] Galinier, M. et al. (2023), A&A, 671, A40
[18] Oszkiewicz, D. et al. (2022), MNRAS, 519, 2917-2928
[19] Korda, D. et al. (2023), A&A, 669, A101
[20] Carruba, V. et al. (2024), MNRAS, 528, 796-814
[21] Marsset, M. et al. (2019), ApJ Letters, 882, L2
[22] Alvarez-Candal, A. et al. (2024), A&A, 685, A29
[23] Tholen, D. (1984), PhD Thesis, MIT