<p>In thirty-one years of observations, the Hubble Space Telescope (HST) has produced a vast archive of thousands of targeted observations. This includes galaxies,  clusters of galaxies, and gravitational lenses. Occasionally, closer objects such as Solar System bodies or artificial satellites cross the telescope's field of view during the observations, leaving trails in the images. On one hand, these trails can impact the observations. The standard data processing pipeline (DrizzlePac) cleans cosmic rays artifacts (Hoffmann et al., 2021), also removing asteroid trails, but it leaves residuals in the combined images. On the other hand, this is a great opportunity for the Solar System small bodies science, considering the already existing images from the huge HST Archive, containing more than 100 Tb of data and spanning three decades. </p><p>Our project is focused on studying serendipitous asteroid trails appearing in archival HST images. We used images from two instruments, namely the Advanced Camera for Surveys and Wide Field Camera 3, the ultraviolet and visible channels. These images were acquired between 2002 and 2021. We built an online citizen science project on the Zooniverse platform, Hubble Asteroid Hunter (www.asteroidhunter.org), launched on International Asteroid Day 2019, to identify the asteroid trails in the images (Kruk et al., in prep.). This project involved more than 11,000 people in search for asteroids, providing 2 million classifications for 150,000 images over a period of one year. The labels provided by the volunteers were used to train an automated classifier based on a deep learning algorithm, Google Cloud AutoML Vision. We recovered 2,400 trails in the HST images in total. </p><p>The asteroid trails appear curved as viewed by HST, because of Hubble’s motion around the Earth every 90 minutes. One example is asteroid 2001 SE101 passing in front of the Crab Nebula in a rare cosmic coincidence, discovered by citizen scientist Melina Thévenot and shown in Figure 1. The project also contributed to other serendipitous discoveries, such as new strong gravitational lenses in the background of some famous HST targets.</p><p><img src="https://contentmanager.copernicus.org/fileStorageProxy.php?f=gepj.ec984bd41ea069331202261/sdaolpUECMynit/1202CSPE&app=m&a=0&c=4d0a7574074b80ef68437340d6523257&ct=x&pn=gepj.elif&d=1" alt="" width="700" height="549"></p><p>Figure 1: Trail of asteroid 2001 SE101 passing in front of the Crab Nebula, M1 in 2005. The trail appears curved because of the motion of HST around the Earth. ESA Image of the Week: http://www.esa.int/ESA_Multimedia/Images/2019/10/Foreground_asteroid_passing_the_Crab_Nebula. Credit: ESA/Hubble & NASA, M. Thévenot. </p><p>We further analysed the asteroid trails in order to obtain their astrometry and photometry with a customised algorithm. We validated the trails visually, finding 1,700 trails presumably of Solar System objects. Their distribution in the sky is shown in Figure 2. We extracted each trail from the images by using a fixed-width aperture, which was moved along the trail.  The position and corresponding flux were obtained for each point along the trail. The calibration was performed using the WCS (World Coordinate System) information stored in the header. As a by-product of this algorithm, we were able to derive partial light curves. The apparent magnitude of the corresponding Solar System object was obtained by integrating all the flux along the trail.</p><p>We used the SkyBoT service provided by IMCCE/Paris Observatory and the JPL HORIZONS online solar system data and ephemeris for identifying the known objects. We computed the ephemerides taking into account the position of HST. Despite using the largest databases of minor bodies, we only matched 300 trails with already known asteroids, taking into account the orbital uncertainties and their apparent motion. Therefore, our data contains 1,400 unknown objects or objects with very large orbital uncertainties.  This is not surprising,  since most of the apparent magnitudes of our trails (Figure 3) are fainter than magnitude 22, which is the approximate limit for the asteroid discovery surveys performed with ground-based telescopes. Most of these objects will correspond to main-belt objects with sizes <1 km, thus will help us characterise the distribution of small size asteroids in the Main Belt, a population poorly explored by current studies. </p><p>This project demonstrates the power of combining novel tools such as citizen science and artificial intelligence to efficiently explore archival images and obtain important results, with the invaluable help of Zooniverse volunteers, beyond the original scope of the Hubble observations. </p><p><img src="https://contentmanager.copernicus.org/fileStorageProxy.php?f=gnp.666737e61ea063681202261/sdaolpUECMynit/1202CSPE&app=m&a=0&c=ad4fa7d50feb3749bfee5cd931b1625c&ct=x&pn=gnp.elif&d=1" alt="" width="700" height="539"></p><p>Figure  2: Sky distribution of the asteroids detected in HST observations. The vast majority of asteroids are in the Ecliptic plane (denoted with red). The two gaps are due to the lack of HST images in the Galactic Plane. </p><p> </p><p><img src="https://contentmanager.copernicus.org/fileStorageProxy.php?f=gnp.5b8160171ea064091202261/sdaolpUECMynit/1202CSPE&app=m&a=0&c=9cef2fc703d9f1aae3c9a489a03680d4&ct=x&pn=gnp.elif&d=1" alt="" width="700" height="516"></p><p>Figure 3: The apparent magnitude distribution of the Solar System objects identified in HST observations. The majority of the identified asteroids have magnitudes >22, fainter than the detection capabilities of many ground-based surveys. </p><p><strong>References:</strong></p><p>Hoffmann, S. L., Mack, J., et al., 2021, “The DrizzlePac Handbook”, Version 2.0, (Baltimore: STScI).</p>

<p>The Hubble Space Telescope (HST) archives hide many unexpected treasures, such as trails of asteroids, showing a characteristic curvature due to the parallax induced by the orbital motion of the spacecraft. We have explored two decades of HST data for serendipitously observed asteroid trails with a deep learning algorithm on Google Cloud, called AutoML, trained on classifications from the Hubble Asteroid Hunter (www.asteroidhunter.org) citizen science project. </p> <p>The project was set up as a collaboration between the ESAC Science Data Centre, Zooniverse, and engineers at Google as a proof of concept to valorize the rich data in the ESA archives. I will present the first results from the project, finding 1,700 asteroid trails in the HST archives (Kruk et al., 2022). Their distribution on the sky is shown in Figure 1.</p> <p>The majority of the asteroid trails (1,031) we found are faint (typically > 21 mag, see Figure 2) and do not match any entries in the Minor Planet Center database, thus likely  correspond to previously unidentified asteroids (see a few examples in Figure 3). We will argue that a combination of AI and crowdsourcing is an efficient way of exploring increasingly large datasets by taking full advantage of the intuition of the human brain and the processing power of machines. </p> <p>The second part of this project aims to analyze in detail these potentially new asteroids and use them to improve our current understanding of the size distribution of small-sized asteroids, and thus help constrain models of the evolution of our Solar System.</p> <p>Taking into account Hubble’s motion around the Earth, the parallax effect can be computed to obtain the distance to the asteroids by fitting  simulated trajectories to the observed trails and obtaining the best fit (Evans et al. 1998). We show one example of a curve fit to the observed trail in Figure 4. Once we know the distance to the asteroids, we are able to obtain their absolute magnitudes and, combined with an assumed albedo, we can obtain their sizes. This method is also able to estimate an envelope for the asteroid's main orbital parameters. </p> <p>Given Hubble’s resolution and capability of reaching faint magnitudes, we expect that many of the new asteroids to be small-sized Main Belt asteroids (diameter <1 km), a population that is too faint and thus difficult to image from ground-based observatories. In addition, with the typical 30 min exposures of Hubble, some of these unknown objects show lightcurves that could be used to infer the rotation and shape of these asteroids, which is helpful for better assessing their type and origin.</p> <p>This project may serve in the future as a “proof of concept” for an automated detection and analysis pipeline in large astronomical archives or surveys.</p> <p> </p> <p><img src="" alt="" width="607" height="301" /></p> <p><strong>Figure 1:</strong> Distribution on the sky of the Solar System Objects (SSOs) identified in the HST images in Mollweide projection. The blue stars show the identified, known asteroids. The orange circles show the location of objects for which we did not find any associations with SSOs. The ecliptic is shown with red. The two gaps in this plot correspond to the Galactic plane, which was not observed by HST.</p> <p> </p> <p><img src="" alt="" width="371" height="249" /></p> <p><strong>Figure 2:</strong> Distribution of apparent magnitudes for the SSOs identified in the HST images. The measured magnitudes for the identified objects (blue bars) and for the objects for which we did not find any associations with known SSOs (orange bars).</p> <p><img src="" alt="" width="642" height="361" /></p> <p><strong>Figure 3</strong>: Examples of unidentified trails in HST observations. The HST observation IDs, clockwise, from the top left, are: j8pv03020, jds47w010, j9bk75010, icphg2010, jdrz23010, and jcng06010.</p> <p><img src="" alt="" width="419" height="279" /></p> <p><strong>Figure 4</strong>: A trail fitting example using the parallax method. The distance solution yielding the best fit for a simulated parallax using HST trajectory is shown in red, in blue the observed trail.</p>