AI-enhanced Citizen Science Discovery of an Active Asteroid: (410590) 2008 GB140

Citizen Science is a framework in which scientific goals and public engagement are achieved simultaneously. Typically, volunteers carry out tasks that would be overwhelming in number or complexity for scientists. Notably, tasks usually are not possible to be carried out by automated means (i.e., AI), though the classification data from Citizen Science programs may be employed in training such systems.

Our NASA Partner program Active Asteroids ¹⁷ is a Citizen Science project designed to locate rare asteroids that show cometary activity. These objects inform our understanding of the early solar system and how it evolved, and point to where volatiles may be found today (Jewitt & Hsieh 2022). With the help of some 9000 volunteers our project has yielded many results since it launched in 2021, but the volume of data is exceptionally challenging to analyze, even with automated analytic vetting applied (Chandler et al. 2024a, in press). Moreover, upcoming petascale surveys, like the Vera C. Rubin Observatory's Legacy Survey of Space and Time, will increase the volume of images available to scour by orders of magnitude (Ivezić et al. 2019; Vera C. Rubin Observatory LSST Solar System Science Collaboration et al. 2021; Schwamb et al. 2023).

In essence, the Active Asteroids program shows images of known asteroids (and comets) to volunteers. These images have already been pre-screened to filter out, for example, images with exposure times insufficient to yield visible activity indicators. Volunteers classify images as either showing cometary activity (a tail or coma), or not. We analyze these classifications and assign likelihood scores to images based on weighted factors like participant experience, Chandler et al. 2024a, in press).

We set out to create an AI assistant to augment our image screening process. To this end, and informed by our past big data AI applications (e.g., Sedaghat & Mahabal 2018; Sedaghat et al. 2021, 2023), we created a binary classifier, "TailNet," that leverages Convolutional Neural Networks (CNNs; Krizhevsky et al. 2017). Notably, CNNs have proven successful with astronomical applications involving both our underlying Dark Energy Camera data source and Zooniverse-based Citizen Science (e.g., Walmsley et al. 2022). When training TailNet, the aforementioned likelihood scores served as the input labels.

Despite TailNet still being a prototype, it identified images of 2008 GB₁₄₀ (as well as 2016 UU₁₂₁; Sedaghat et al. 2024) as highly unlikely to be inactive, which we interpreted as likely to show activity. Our subsequent archival investigation revealed eight images of 2008 GB₁₄₀ showing activity (Figure 1) spanning UT 2023 April 23 to UT 2023 July 3. Then, 2008 GB₁₄₀ was inbound at true anomaly angles 259° < f < 273° and heliocentric distances 2.956 > r_H  > 2.852 au.

Zoom In Zoom Out Reset image size

2008 GB₁₄₀ (semimajor axis a = 2.933 au, eccentricity e = 0.145, inclination i = 142, perihelion distance q = 2.507 au, aphelion distance Q = 3.360 au) is a Main-belt asteroid with a Tisserand Parameter with respect to Jupiter T_J = 3.214. Thus, along with the activity we uncovered, we classify 2008 GB₁₄₀ as an active asteroid. Moreover, 2008 GB₁₄₀ is a candidate Main-belt comet as its activity near perihelion is compatible with sublimation-driven activity. 2008 GB₁₄₀ is inbound to perihelion (f = 317° on UT 2023 January 14) and thus is likely still active; further observations would help characterize the nature of 2008 GB₁₄₀'s activity and provide other insights into its physical properties.

Many thanks to Arthur and Jeanie Chandler for their ongoing support.

We thank Elizabeth Baeten (Belgium) for moderating the Active Asteroids forums. A special thanks to the Active Asteroids Superclassifiers: Angelina A. Reese (Sequim, USA), Antonio Pasqua (Catanzaro, Italy), Carl L. King (Ithaca, USA), Dan Crowson (Dardenne Prairie, USA), @EEZuidema (Driezum, Netherlands), Eric Fabrigat (Velaux, France), @graham_d (Hemel Hempstead, UK), Henryk Krawczyk (Czeladż Poland), Marvin W. Huddleston (Mesquite, USA), Robert Zach Moseley (Worcester, USA), Thorsten Eschweiler (Übach-Palenberg, Germany), and Washington Kryzanowski (Montevideo, Uruguay). Thanks to Cliff Johnson (Zooniverse), Chris Lintott (Oxford), and Marc Kuchner (NASA) for ongoing Citizen Science guidance.

This material is based upon work supported by the NSF Graduate Research Fellowship Program under grant No. 2018258765 and grant No. 2020303693. C.O.C., H.H.H., and C.A.T. acknowledge support from the NASA Solar System Observations program (grant 80NSSC19K0869). W.J.O. acknowledges support from NASA grant 80NSSC21K0114. This work was supported in part by NSF awards 1950901. This research received support through Schmidt Sciences. Chandler and Sedaghat acknowledge support from the DiRAC Institute in the Department of Astronomy at the University of Washington. The DiRAC Institute is supported through generous gifts from the Charles and Lisa Simonyi Fund for Arts and Sciences, and the Washington Research Foundation.

Computational analyses were run on Northern Arizona University's Monsoon computing cluster, funded by Arizona's Technology and Research Initiative Fund.

This project used data obtained with the Dark Energy Camera (DECam), which was constructed by the Dark Energy Survey (DES) collaboration. This research uses services or data provided by the Astro Data Archive at NSF's NOIRLab. Based on observations at Cerro Tololo Inter-American Observatory, NSF's NOIRLab (NOIRLab Prop. ID 2014B-0404, PI: D. Schlegel; Prop. ID 2019A-0305, PI Drlica-Wagner).

Facility: CTIO:4m (DECam) - .

Software: astropy (Robitaille et al. 2013), astrometry.net (Lang et al. 2010), PyTorch (Paszke et al. 2019).