The Active Asteroids Citizen Science Program: Overview and First Results

We present the Citizen Science program Active Asteroids and describe discoveries stemming from our ongoing project. Our NASA Partner program is hosted on the Zooniverse online platform and launched on 2021 August 31, with the goal of engaging the community in the search for active asteroids—asteroids with comet-like tails or comae. We also set out to identify other unusual active solar system objects, such as active Centaurs, active quasi-Hilda asteroids (QHAs), and Jupiter-family comets (JFCs). Active objects are rare in large part because they are difficult to identify, so we ask volunteers to assist us in searching for active bodies in our collection of millions of images of known minor planets. We produced these cutout images with our project pipeline that makes use of publicly available Dark Energy Camera data. Since the project launch, roughly 8300 volunteers have scrutinized some 430,000 images to great effect, which we describe in this work. In total, we have identified previously unknown activity on 15 asteroids, plus one Centaur, that were thought to be asteroidal (i.e., inactive). Of the asteroids, we classify four as active QHAs, seven as JFCs, and four as active asteroids, consisting of one main-belt comet (MBC) and three MBC candidates. We also include our findings concerning known active objects that our program facilitated, an unanticipated avenue of scientific discovery. These include discovering activity occurring during an orbital epoch for which objects were not known to be active, and the reclassification of objects based on our dynamical analyses.


ABSTRACT
We present the Citizen Science program Active Asteroids and describe discoveries stemming from our ongoing project.Our NASA Partner program is hosted on the Zooniverse online platform and launched on 2021 August 31, with the goal of engaging the community in the search for active asteroidsasteroids with comet-like tails or comae.We also set out to identify other unusual active solar system objects, such as active Centaurs, active quasi-Hilda asteroids, and Jupiter Family Comets (JFCs).Active objects are rare in large part because they are difficult to identify, so we ask volunteers to assist us in searching for active bodies in our collection of millions of images of known minor planets.We produced these cutout images with our project pipeline that makes use of publicly available Dark Energy Camera (DECam) data.Since the project launch, roughly 8,300 volunteers have scrutinized some 430,000 images to great effect, which we describe in this work.In total we have identified previously unknown activity on 15 asteroids, plus one Centaur, that were thought to be asteroidal (i.e., inactive).Of the asteroids, we classify four as active quasi-Hilda asteroids, seven as JFCs, and four as active asteroids, consisting of one Main-belt Comet (MBC) and three MBC candidates.We also include our findings concerning known active objects that our program facilitated, an unanticipated avenue of scientific discovery.These include discovering activity occurring during an orbital epoch for In 1949 comets ceased to be the only solar system objects known to display activity when near-Earth asteroid (4015) Wilson-Harrington was observed with a pronounced tail (Cunningham 1950).In the seven intervening decades, fewer than 60 asteroids have been found to be active, a tiny fraction of the ∼1.3 million known minor planets, and the vast majority of discoveries have taken place in just the last 25 years (see Table 1 of Chandler et al. 2018).Nevertheless, these objects have provided a wealth of knowledge (Hsieh & Jewitt 2006a;Jewitt 2012), ranging from informing us about the volatile distribution in the solar system and possible origins of terrestrial water (Hsieh & Jewitt 2006b), to further insight into astrophysical processes such as the Yarkovsky-O'Keefe-Radzievskii-Paddack (YORP) effect (e.g., (6478) Gault; Kleyna et al. 2019).Roughly half of the observed activity in apparently asteroidal bodies has been attributed to stochastic events, such as impacts (including the Double Asteroid Redirection Test (DART) impact), with the remainder seen to be recurrently active, a characteristic potentially diagnostic of volatile sublimation.Before our program, fewer than 15 of the known active asteroids were classified as Mainbelt Comets (MBCs), recurrently active, sublimationdriven active asteroids that orbit exclusively within the main asteroid belt (Hsieh & Jewitt 2006b).
A similar story applies to the Centaurs, bodies thought to originate in the Kuiper belt that are now found between the orbits of Jupiter and Neptune (see review, Morbidelli 2008).Unlike the active asteroids, the first active Centaur, 29P/Schwassmann-Wachmann 1 (Schwassmann & Wachmann 1927), was identified retroactively after Centaurs were realized as a class following the discovery of (2060) Chiron in 1977 (Kowal & Gehrels 1977).Notably, these bodies are too cold for water ice to sublimate, so other species (e.g., CO 2 ) or processes must be involved (Jewitt 2009;Snodgrass et al. 2017;Chandler et al. 2020).
As with active asteroids, few (< 20) active Centaurs have been found, so finding more of these objects will significantly further our knowledge about this minor planet population.
Another group of active objects not typically associated with comets are the active quasi-Hilda aster-oids, sometimes referred to as quasi-Hilda comets, active quasi-Hilda objects, or active quasi-Hildas.This dynamical class shares a name with the Hilda asteroids, a small body population bound in stable 3:2 interior mean-motion-resonance with Jupiter, and span a region from the outer asteroid belt to the Jupiter Trojans (Szabó et al. 2020).However, quasi-Hildas are not in true resonance with Jupiter, though their orbits are reminiscent of the Hildas when observed in the Jupiter co-rotating reference frame (Chandler et al. 2022) as discussed later in Section 6.Consequently, these objects are challenging to identify, with dynamical modeling requisite to confirm a quasi-Hilda orbit confidently.Roughly 3,000 quasi-Hildas have been loosely identified (Toth 2006;Gil-Hutton & Garcia-Migani 2016), with fewer than ∼15 observed to be active.
All of the aforementioned classes (active asteroids, active Centaurs, and active quasi-Hilda asteroids) remain largely mysterious, with so few objects known that it is difficult to draw statistically robust conclusions about these populations.The clear remedy, then, is to find more of these objects.There are numerous astronomical archives containing vast numbers of images in which minor planets can be seen, but they have not been examined because of the overwhelming numbers involved.We set out to do this with the help of online volunteers through Citizen Science, a paradigm that simultaneously achieves outreach and scientific goals.Here, we (1) briefly introduce the Citizen Science project Active Asteroids and the underlying system that produces the images we show to volunteers, (2) describe a broadly applicable technique we created to improve the quality of classification analyses, and (3) present results stemming from the first two years of the Active Asteroids program, including objects previously unknown to be active.

HARVEST: THE IMAGE CUTOUT PIPELINE
With the goal of discovering previously unknown minor planet cometary activity we created a pipeline to extract small images of known minor planets from publicly available archival astronomical images; these extracted small images are interchangeably known as cutouts, thumbnails, or "subjects" in Zooniverse terminology.We initially created the Hunting for Activity in Repositories with Vetting-Enhanced Search Tech-niques (HARVEST) pipeline for our proof-of-concept work Searching Asteroids For Activity Revealing Indicators (SAFARI; Chandler et al. 2018).Since then we have substantially improved upon and optimized HARVEST (Chandler et al. 2019(Chandler et al. , 2020(Chandler et al. , 2021(Chandler et al. , 2022;;Chandler 2022), so we provide here a comprehensive description of the complete system.

Pipeline Overview
HARVEST runs as a series of steps that are composed of constituent tasks; tasks are executed in series or, when possible, in parallel.Tasks are primarily written in Python 3 code, with some compiled programs called as specified in the subsections below.We optimized the pipeline for execution on high performance computing clusters that employ the Slurm task scheduler (Yoo et al. 2003), so the top-level pipeline steps are conducted via Bash shell scripts.Key concepts needed to understand HARVEST are provided here.Advanced technical considerations are discussed in Chandler (2022).
Throughout HARVEST we implement an "Exclusion" system that is essential to optimizing the chances of success that volunteers will identify activity in an image they examine.For example, we do not want to submit images to volunteers for classification that we determine (via automated algorithm) contain no source at the center of the frame.These are described in the corresponding pipeline subsections below.

Database
HARVEST makes use of a custom MySQL relational database of our own design.The database is composed of numerous tables to optimize memory usage.Here we described the key elements essential to the pipeline.
Observations are records holding the UT observation date and time, as well as the identity of the telescope, instrument, broadband filter, Principal Investigator (PI) name, and proposal ID.Each Observation record can have one or more associated Field records, each containing airmass, angular separation from the pointing center to the Moon's center, and Right Ascension and declination (RA, Dec) sky coordinates.
Datafiles are the records specific to a particular version of a produced data file, such as exposure time and release date (when the data became or will become publicly available).We store our computed depth estimate here (discussed further in Section 2.4).
Each Datafile record may have many Thumbnail records, one for each of the individual cutouts centered on a known minor planet we produce.We strive to keep only one thumbnail per unique combination of Observation and Solar System Object, despite the necessity to download different versions of datafiles in cases where the archive-provided datafile was corrupted.
Solar System Objects are records containing compiled information about individual bodies of the solar system, including orbital elements and discovery circumstances.
Skybot Results are the tabular data returned by the Institut de Mécanique Céleste et de Calcul des Éphémérides (IMCCE) Skybot Service (Section 2.5), such as computed sky position and apparent V -band magnitude, geocentric and heliocentric distances, phase angle, and solar elongation.

HARVEST Step 1: Catalog Queries
In this step we query astronomical image archives for metadata pertaining to observations.The essential elements include sky coordinates, exposure UT date/time, exposure time, broadband filter selection, release date (when the data becomes public), and data location (URL).We primarily query instrument archives that hold calibrated data with well-calibrated World Coordinate System (WCS) header information.The two archives we query are the National Optical and Infrared Laboratory (NOIRLab) AstroArchive and the Canadian Astronomy Data Centre (CADC) data archive.Our pipeline produces thumbnail images for several instruments, with Dark Energy Camera (DECam) the sole data source we have made use of thus far for our Active Asteroids Citizen Science program.
We query external resources for information about known minor planets, including orbital elements (e.g., semi-major axis a, eccentricity e, inclination i), identity information (e.g., minor planet numbers, provisional designations), and discovery circumstances (e.g., date, site).These include the Minor Planet Center (MPC), JPL Small Body Database, and Lowell Observatory's AstOrb database (Moskovitz et al. 2022).The Ondrêjov webpage 1 lists objects discovered at Ondrejov Observatory (site code 557), and includes identifiers sometimes not found at the MPC, but that may be returned by Sky-Bot (Section 2.5).We note the late Kazuo Kinoshita's comet page 2 is no longer being updated, but it is included here as we have incorporated his work.
We exclude observations (1) taken at an airmass greater than 3.0, (2) we calculate to have a pointing center < 4 • from the Moon's center, (3) with invalid pointing coordinates (e.g., RA >360 • ), (4) acquired with broadband filters typically unfavorable to activity detection.We exclude data files that (1) are uncalibrated 1 http://www.asu.cas.cz/∼ asteroid/news/numbered.htm 2 https://jcometobs.web.fc2.com(i.e., raw) as activity is harder to detect and the embedded WCS is likely insufficient to place the object at the center of our cutouts, and (2) stacked (co-added) images that typically eliminate moving objects.

HARVEST Step 2: New Data Handling
Magnitude Estimates -We compute a rough estimate of image depth, a value not necessarily provided with the archival data.We employ functions that are instrumentspecific and, wherever possible, are based upon an observatory-supplied exposure time calculator (ETC).In cases where no ETC was available, we applied our DECam-derived estimator, adjusting the mirror area as needed.We estimate the magnitude limit achievable for a minimum detection at a 10:1 signal to noise ratio (SNR).To compare the depth estimate with object-specific magnitudes computed by ephemeris services (e.g., JPL Horizons), which are always provided in Johnson V -band, we apply a rudimentary apparent magnitude offset from measured apparent Vega magnitudes of the Sun (Willmer 2018).The difference between ephemeris magnitude and depth we call delta magnitude (∆m, Section 2.5).This allows us to exclude thumbnails for which an object and potential activity is fainter than the detection limit of a given exposure.
Version Selection -Archives may provide multiple datafile versions for an observation, such as InstCal and Resampled images via AstroArchive.We choose a single datafile to work with and exclude all others.If we later encounter a problematic (i.e., corrupt) data file, we can select a different version.
NASA JPL Object Data -We maintain an internal table of NASA JPL-provided minor planet parameters (e.g., semi-major axis) that may not be provided by other services we utilize.We query both JPL Horizons and the JPL Small Body Database (Giorgini et al. 1996).
Solar System Object Parameters -Here we assemble a consolidated set of dynamical elements (semi-major axis a, inclination i, eccentricity e, perihelion distance q, and aphelion distance Q) and compute the Tisserand parameter with respect to Jupiter, T J needed for dynamically classifying objects.The classes are listed below, and the methods are discussed in Section 6.
Object Classification -Each minor planet in our database is labeled with a single associated class from the following: Comet, Amor, Apollo, Aten, Mars-crosser, inner Main-belt (IMB), middle Main-belt (MMB), outer Main-belt (OMB), Cybele, Hungaria, Jupiter Family Comet (JFC), Hilda, Trojan, Centaur, Damo-cloid, Trans-Neptunian object (TNO)/Kuiper Belt object (KBO), Phocaea, or Interstellar Object.These are dynamical classes, with the notable exception of comets, which are classified as such when visible activity has been reported.We use the class name provided by the IMCCE Quaero Service as these are included with Sky-Bot results -but we intervene to reclassify some objects as long as they are not labeled as Trojan asteroids.Specifically, following the procedures described in Section 6, minor planets with (1) a Tisserand parameter with respect to Jupiter (Section 6) 2 ≤ T J < 3 we reclassify as a JFC, (2) T J < 2 we reclassify as a Damocloid, or (3) a a J < a < a N (a semi-major axis a between those of Jupiter and Neptune, a J and a N , respectively) are labeled a Centaur.We note that we treat classifications in HARVEST as approximate as they are not based upon custom dynamical simulations, however the rough fit is adequate for our purposes (e.g., selecting images for a subject set; Section 3.3).

HARVEST Step 3: Field Analysis
Here we perform tasks specific to a unique combination of telescope pointing and UT date/time, internally stored as Field records.As noted earlier, multiple records can exist for a single Observation record because different process types (e.g., InstCal) can result in slightly different WCS information, though our database should only maintain one non-excluded Field per Observation as a result of Version Selection (Section 2.4).
SkyBot -The IMCCE SkyBot (Berthier et al. 2006) service returns a table listing the solar system objects that may be found within a given combination of sky coordinates, UT date/time, and field of view.We construct each query as a "cone" (circular field) or "polygon" (rectangular field), depending on the instrument field of view, and query SkyBot for all new fields (Section 2.3) added to our database during the daily HARVEST schedule.For computational and service call efficiency (i.e., to avoid excessive queries to the SkyBot service), instead of querying all fields via SkyBot daily, we only periodically (every ∼ 90 days) resubmit fields to SkyBot to search for minor planets discovered since we previously queried the field via SkyBot.
Delta Magnitudes -During the SkyBot phase we calculate a metric to estimate how many magnitudes brighter (or fainter) an object will appear in a field, by where V JPL is the object's apparent V -band magnitude as computed by the JPL Horizons ephemeris service (typically Johnson V ), and V ITC is our computed Vband depth (Section 2.4).Objects with ∆ mag < 0 are above a SNR of 10 and should be detectable, whereas ∆ mag > 0 would likely not be detectable.We exclude SkyBot results from our database that have ∆ mag > −1 because our goal is to detect activity, and our experience has been that at least one additional magnitude of depth is necessary for this task.We acknowledge that activity outbursts could result in a significantly brighter apparent magnitude than our estimate, but maintain this threshold to eliminate low-probability detection events among a high volume of extraneous images volunteers will examine.Roughly 57% (∼21 million) of SkyBot results in HARVEST have been excluded because of our chosen ∆ mag threshold.Additional considerations for adjusting this threshold are discussed in Section 8.2.1.
Trail Length -Making use of the ephemeris-supplied apparent rate of motion on the sky, we estimate trail lengths for each object given the exposure time.We used this measurement for constructing the project Field Guide and, as of 2023 August 15, we have suspended submitting images with trails > 15 pixels for examination, as these have proven to be a common source of false-positive detections by volunteers.

HARVEST Step 4: Thumbnail Preparations
Data Download -Here we generate scripts to download data from astronomical archives.Downloads occur when new data has become publicly available, or a new object was found in an existing field.The transfer of data is handled by daemons we constructed for this purpose, each dedicated to downloading data from a single archive (e.g., AstroArchive, CADC).As the process can take days, HARVEST continues operations without waiting for these tasks to finish; these data can be processed during a subsequent execution of the pipeline.
Chip Corners -For every image we download we record the sky coordinates of each corner of all camera chips.For DECam there are > 60 chips that make up a mosaic that covers a roughly circular area on the sky.This step eliminates the need to check every chip corner for each thumbnail image to be produced, thereby enabling an order-of-magnitude compute time savings during thumbnail extraction (Section 2.7).This also allows us to determine if an object falls outside of any detector area (e.g., chip gap), negating the need to redownload a file from an archive, such as when a new object is discovered and in a Field (Section 2.5).pixel image, each requiring ∼ 1 Mb of disk space.We preserve WCS in thumbnail images, as well as primary headers and the headers for the specific chip from which the cutout was extracted.

HARVEST Step 5: Thumbnail Extraction
PNG Thumbnails -Here we convert the FITS thumbnail images to Portable Network Graphics (PNG) format for submission to Zooniverse or examination by our team.
We employ an iterative rejection contrast enhancement scheme (Chandler et al. 2018) to facilitate activity detection.Each PNG thumbnail requires ∼512 Kb of storage.

HARVEST Step 6: Thumbnail Analysis
Source Analysis -We produce tables of sources found within each cutout with SExtractor (Bertin & Arnouts 2010) and apply exclusions based on our analysis of these data.The following exclusions are applied, with representative statistics derived 10 July 2022, when HARVEST contained 22,004,739 non-excluded thumbnail records: (1) no source was detected within the center 20×20 pixel region; 16% (4,248,133 thumbnails) excluded.(2) > 150 sources were found in the center 270×270 pixel region; 4% (952,289) met this criteria.(3) > 5 blended (overlapping) sources were detected at the cutout center; 0.4% (84,697 thumbnails) were affected.
Source Tallying -All tasks that perform exclusions have concluded.We perform tallying to optimize reporting, and consistency checks: (1) Objects per Field: the number of non-excluded solar system objects in each field, and (2) SkyBot Source Density: the tally of nonexcluded SkyBot results associated with each field.

HARVEST Step 7: Reporting
SkyBot Reports -We generate plots and tables describing how recently each field has been submitted to Sky-Bot, primarily for diagnostic purposes.
Objects per Field -This diagnostic aid quantifies valid objects in each pointing that are passed on for processing.This helps us, for example, project our future Citizen Science project completeness (Section 8.2.1).

HARVEST Step 8: Maintenance
Datafile Checks -We check image files we have downloaded for integrity by querying the HARVEST database for images that have been marked as "bad datafiles" by other tasks, typically those failing the AstroPy FITS verification process.Files we diagnose as corrupt go through a process where we download the data again and, upon a second failure, we identify a replacement if another version is available (Section 2.4).
Datafile Exclusion by Property -Here we exclude from the HARVEST database all datafiles with invalid properties, such as exposure times < 1 s or NULL values.While this screening is also done during the Catalog step (Section 2.3), we routinely repeat screening as a safety measure.
Purge Datafiles -Once all thumbnail images have been extracted from a downloaded image and all subsequent analysis processes have completed, we purge the file from disk as we do not have the requisite storage necessary to keep all of the downloaded image data.

CITIZEN SCIENCE PROJECT
We produced millions of thumbnail images (Section 2) to search for active objects.This task was impractical for our team to accomplish on our own, so we sought to engage the public in our endeavor.The paradigm we selected, Citizen Science, is known for (1) addressing tasks that are too numerous for individuals and/or too complex for computers to handle, and (2) volunteers can be trained to effectively accomplish the task with minimal training.Citizen Science programs engage the public in scientific inquiry, and thus serve as important outreach avenues and provide education opportunities.
The core approach of our project is to show the thumbnail images of known minor planets to volunteers and ask them whether or not they see evidence of activity (i.e., a tail or coma) coming from the object.As described in Section 2, these images originate from the pipeline we created for this purpose, HARVEST, that extracts images from publicly available archival images from the DECam instrument on the 4 m Blanco telescope at the Cerro Tololo Inter-American Observatory (CTIO) in Chile.Critically, before and during project preparations we carried out work that served as proofs-of-concept and validations that justified construction of this Citizen Science project (Chandler et al. 2018(Chandler et al. , 2019(Chandler et al. , 2020(Chandler et al. , 2021)).These results are described in Section 9.

Project Foundation
We chose to host our project, Active Asteroids, on the online Citizen Science platform Zooniverse because of their proven track record of supporting successful astronomy-related projects.Their team also provides developmental support for project customization, which is important for our project workflow (Section 3.4).
The overall process for Active Asteroids, from launch to ongoing operations, is as follows.(1) Prepare Zooniverse project (see sections below).( 2  of images, (d) notify volunteers of new data and other news, (e) investigate activity candidates.
We formally launched Active Asteroids3 on 31 August 2021.Since then more than 8,300 volunteers have examined over 430,000 images, carrying out a total of some 6,700,000 classifications (including both sample and training data).

Project Components
The project "workflow" is the task volunteers are asked to perform.At present, we have one concise workflow where we ask volunteers if they see activity (i.e., a tail or coma) coming from the central object, marked by a green reticle like that shown in Figure 1.The first time participants begin classifying images they are shown a tutorial we produced that demonstrates images of activity along with tips for avoiding activity lookalikes (e.g., background galaxies).During the classifying process, users may return to the tutorial at any time.
Also available during the classification process is our comprehensive Field Guide which discusses phenomena participants may encounter, such as cosmic rays.
The Zooniverse web structure includes several other areas important to project success.An "About" section includes pages describing (1) our research and science justification, (2) project team members, (3) a listing of results (e.g., publications) stemming from the project, and (4) a frequently asked questions (FAQ) page.The Talk discussion boards (forums) provide a place for participants and the science team to interact and build relationships.Surprisingly, we have made discoveries that first come to light on the Talk pages, well before the subject set was fully retired (discussed below).

Subject Sets
A "subject set" is a collection of images and associated metadata (e.g., image names, object designations).We try to select a subject set size (i.e., number of images) that balances preparation overhead with turnaround time to complete subject set retirement.Smaller subject sets are fully examined by volunteers in fewer days, but each batch requires significant overhead -both effort and time -for our team to (1) prepare each batch (described below) and ( 2) analyze classification data (Section 4).Conversely, large batches take longer to complete.We found a good balance to be a subject set size of ∼22,000 images, typically needed four to eight weeks for volunteers to examine (Section 8.2).
To create subject sets we (1) assign images from HARVEST based upon selection criteria (described below), and (2) gather images and prepare them for upload to Zooniverse by adding a green reticle (Figure 1).
The ability to select objects by criteria is motivated by the need to show volunteers a variety of images, and to optimize the discovery of activity.We fully recognize that our choices impart biases, but err on the side of making the best use of volunteer efforts.
We assemble each batch as a collection of members from different dynamical classes.As of 2023 August 17, we have submitted an 19 subject sets for examination, typically containing ∼22,000 images.The composition has changed too as we have exhausted the images of some minor planet classes, and have de-emphasized others (e.g., near-Earth objects (NEOs)) that have proven problematic for activity identification.
To improve chances for identifying activity we prioritize selecting images of objects closer to their perihelion passage, with the assumption that activity is more likely to be present around this point in an object's orbit.We achieve this effect by sorting HARVEST images by our simple metric, "percentage to perihelion" (Chandler et al. 2018), given by % where d, q, and Q are its orbital, perihelion, and aphelion distances, respectively.This metric is more efficiently sorted than the more familiar true anomaly angle, though % T →q does not describe the direction (i.e., inbound to, or outbound from, perihelion) of the object.By default, we only show one image of an object in a given batch, such that volunteers are examining the maximum number of individual minor planets.In cases where we have few object images remaining (e.g., Centaurs), we do increase this number.Conversely, as of 2022 October 22, we have had the option to skip objects entirely that have already been examined by volunteers at least once.This is especially useful for populations like the main-belt asteroids, where we have in our collection tens of thousands of images of unique minor planets, all essentially at perihelion.
As of 2023 August 18, we always apply further delta magnitude limits (Section 2.5).We typically require ≥ 2 magnitudes brighter than our computed exposure depth (i.e., ∆ mag ≤ −2).We consider this threshold reasonable given the project's current classification rate and projected completeness timescales (Section 8.2).

Training Set and Expert Scoring
The training system implemented by Zooniverse for Active Asteroids is designed to teach volunteers how to identify activity.We show in Section 4 that this system measurably improves activity detection ability for the vast majority of participants.The system also served to validate that the project was functioning as intended during the launch phase, and the training system continues to serve that function today.
For training purposes we created a subject set consisting of images known to show an object displaying cometary activity at the center.To achieve this we manually examined ∼10,000 images of known active bodies produced by the HARVEST pipeline and assigned a score to each image.The subjective scoring system, introduced in Chandler (2022), is as follows: (0) unidentifiable/missing, (1) point-source appearance, (2) vaguely fuzzy, (3) fuzzy; activity unlikely; (4) inconclusive activity indicators (coma and/or tail), (5) likely active; some ambiguity remains, (6) activity, not very ambiguous, but faint, (7) definitely active; medium-strength indicators, (8) definitely active; strong activity evidence, (9) definitely active; overwhelming activity indicators.All training images in Active Asteroids are derived from these images for which we applied a score of ≥ 5, our minimum threshold for which we consider the activity to be highly likely.
Active Asteroids is configured with two training features.The first is a system that periodically shows the user a training image, at an interval that decays with user experience (determined by the number of images N they have classified), given by the probability (3) The second feature is a feedback system, wherein immediate feedback is given to the user about their training image classification, whether their classification was "correct" or not.While this serves as a direct training mechanism for new participants, it also serves to reinforce the abilities of experienced users and helps keep volunteers engaged in the classification process.

OPTIMIZING CLASSIFICATION ANALYSIS
Volunteers examine images we produce with the HARVEST pipeline (Section 2).Training images always show activity and are described in Section 3.4; images yet to be examined are referred to as "sample" images.A "classification" occurs when a volunteer clicks a "yes" or "no" button when asked if they see activity in the image.Images are randomly selected from the current subject set (Section 3.3) of images we have uploaded to Zooniverse.Each image is nominally classified by 15 unique participants before the image is "retired," with the exception of training images which are, by design, never retired.In some uncommon situations, > 15 classifications occur for sample images; in those cases, we make use of the first 15 classifications.
Classification data are not static because we regularly upload new subject sets to the project.We developed the techniques described herein with a snapshot from 2022 July.At that time, 6,609 unique volunteers had examined ∼170,000 images, with ∼5 million classifications in total, including training images.

Naïve Assessment Metric and Threshold
Initially, we computed, for each image i, a simple activity likelihood metric M 0 (i) as the ratio of "yes" classifications for the image, Y i , and the total number of classifications, i.e., the sum of yes and no (N i ) responses for that image, as For the development of the new metrics (discussed in the subsequent section) we validated underlying premises (e.g., users become more experienced through time) as we developed the methods.The exception was this naïve metric, which served as the starting point from which we set out to improve our classification analyses.
Here we also define the minimum "threshold" L min of a metric.This serves to differentiate between images likely to show activity -and thus qualify as candidates that our team will investigate -from those that are not.For the naïve (unjustified) threshold we chose L 0 ≥ 80%.From this initial combination of metric and threshold we set out to test and improve upon our initial selection of metric and threshold.

New Metrics
Weighting based upon assessment of volunteer trends has been employed by other Citizen Science programs.For example, Gollan et al. (2012) found Citizen Scientists may on average not perform as well as professional scientists when performing the same tasks, but also that some individuals do meet or exceed that same standard. (5) Metric 2: log10 Number of Classifications -As users classify more images, they generally become more experienced, so they should be given greater weight.This metric quantifies "experience" as log 10 of the total number of classifications for a given user.We placed an upper limit of 10,000 classifications, and scaled weights to span the range of 0 to 1 by dividing all weights by 4 (i.e., log 10 10, 000).
where T total is the total number of images the user had examined, including training images.
Metric 3: Optimism Debiasing -We found that some users identify activity much more often than would normally be expected, thus their weight needed to be lowered accordingly.
Noting that the activity occurrence rate among mainbelt asteroids is estimated to be roughly 1 in 10,000 (Je-witt et al. 2015;Hsieh et al. 2015;Chandler et al. 2018), we expect a low rate of "yes" classifications.Moreover, from our cursory examinations, we estimated no more than ∼1% of images should warrant being flagged as likely active.To search for bias we first described the fraction of classifications a user u submits as positive by where Y u and N u are the total number of "yes" and "no" classifications for that user, respectively.∼35% of users (2,400) clicked "yes" over 20% of the time, indicating optimism bias is present.Any weight-based purely upon training accuracy is skewed for users selecting "yes" to most images they classify, with a potential resultant weight of unity for training accuracy while incorrectly reflecting activity detection ability.For this metric, the more frequently a volunteer sees activity, the lower their weight becomes, via where Y sample is the number of times a user saw activity in sample images, and T sample is the total number of sample images the user classified.

Control List and Initial Threshold
In order to test the efficacy of each metric (or combination of metrics) we maintained a "control list" of images that our team vetted and labeled as strong activity candidates.We set a threshold L min for each metric (i.e., L 1 , L 2 , and L 3 ) by iteratively increasing or decreasing the threshold L in 10% increments until all control list images appeared in the final output list of candidates.We arrived at an L min = 40% threshold that resulted in control list completeness for all metrics; henceforth this served as our initial threshold when testing metrics and combinations thereof.Throughout, a secondary goal was to minimize the number of extraneous (inactive) images flagged as promising candidates, while still including those from the control list.

Incorporating Temporal Trends
Figure 2 shows combined user weights over time for 10 randomly selected users, where time here is measured only by the number of images classified.User weights typically improved, but not always (e.g., user #7 of Figure 2), indicating that one or both of the metrics not solely dependent on classification count (i.e., M 1 -training accuracy, or M 3 -optimism debiasing) must be significantly altering the weight.
This finding showed (1) a need to evaluate metrics temporally, (2) each metric may need a multiplier

User's Weight
Figure 2. The weight for the first 1,000 images for ten unique, randomly selected users (numbered by markers 0 -9) who classified between 1,000 and 10,000 images.Each number represents 20 images classified, and scores are cumulative.
(weight), and (3) we cannot assume user abilities improve over time.To capture this time-dependent weight we employed a 5 th order polynomial fit for each user's weight over time.
We tried combinations of metric weights, ranging from 0 to 10, for each metric (M 1 , M 2 , M 3 ), plus 100×, 1,000×, and 10,000× to test extrema.For computational efficiency, we (1) eliminated weight combinations that were integer multiples of each other that would yield identical scores, and (2) selected only even number weight multiples, thereby reducing the number of required compute tasks while still covering the full range of weights.We combined the weighted metrics via where W is the combined weight for a user, W 1 is the weight of M 1 (training image accuracy), W 2 the weight for M 2 (log 10 of classification count), and W 3 is the weight for M 3 (optimism debiasing).
We created an overall weighted likelihood score, L, computed as a ratio of the sum of an image's weighted user "yes" classifications, w Y , to the sum of all the users' weights w, given by, with m the number of users who classified the image as showing activity, and k is the total number of users who classified the image.

Metric Selection and Evaluation
For each set of weight combinations we (1) calculated a score for all sample (non-training) images using that set of weights, (2) determined the threshold L min needed to include the images of our control list (Section 4.3), and (3) recorded the number of images, I f , that received a score L ≥ L min for that set of weights, including images not part of the control list.
We evaluated each metric independently and in combination, and compared these to the naïive metric M 0 .Our newly crafted method for determining which images warrant further investigation performed markedly better than the nave method (Section 4.1).The naïve method resulted in a threshold of L min = 46.66%for I f = 2, 513 images (1.48% of the classified images), 795 more than our weighted method.
We selected a final weight combination of W 1 = 7, W 2 = 2, W 3 = 1, with a threshold L min = 47.3%.This combination resulted in 1,718 activity candidate images for our team to examine, ∼ 1% of the ∼170,000 images (discussed at the beginning of this section) used for the classification analysis improvements.
We evaluated the 15 image retirement criteria (Section 3.3) by varying the number of classifications required per image prior to subject retirement.We calculated the combined weighted score for every image, considering only the first n classifications for each image, for 1 ≤ n ≤ 15.
The number of extraneous images flagged for investigation declined through n = 14.While n = 14 unexpectedly performed marginally better (233 images) than n = 15, we interpret this result as indicating that 14 classifications per image represents a necessary minimum to achieve the best results for our project.We chose to keep n = 15 for the live project, in keeping with the original Zooniverse recommendation.

ACTIVITY CANDIDATE INVESTIGATION
Once we have scores produced by our classification analysis system (Section 4) we next produce a list of images that match the threshold, as justified in the pre-ceding section.We then examine each image and apply the same 0 -9 scoring system described in Section 3.4.We further investigate candidates with scores ≥ 3, through archival and -when appropriate -followup telescope observations.In 2023 January we decided to start announcing discoveries through Research Notes of the American Astronomical Society (RNAAS), especially for time-sensitive cases (e.g., the object is approaching perihelion and activity detection is useful for diagnosing the underlying mechanism).References to these publications are provided in Section 7.

Archival Investigation
Our archival investigation process typically involves first querying our internal HARVEST database for additional images of the object.This first pass enables us to quickly rule out some false positive candidates, for example, those with apparent activity that we recognize as a background source when the object is viewed in an image sequence.
For the remaining candidates we next query external services via three pipelines we have written for the purpose, and manually query two additional sources.There are significant drawbacks to these systems as compared to HARVEST, most notably a high fraction of "junk" images that are either too faint to see the candidate, or the candidate is not captured by a detector.Nonetheless, this in-depth search can yield images of activity that are not available through the HARVEST pipeline.
We query the following archives (see Acknowledgements for additional references) as part of the aforementioned pipelines: CADC SSOIS -We query the CADC Solar System Object Information Search (SSOIS) (Gwyn et al. 2012)  ZTF Alert Stream -We download ZTF alert stream data (Patterson et al. 2018) and keep only solar system data.
After downloading the relevant data, many sources require pre-and post-processing to, for example, perform astrometry to replace an inadequate (or absent) plate solution.We perform astrometry as needed via Astrometry.net(Lang et al. 2010), which makes use of multiple source catalogs, including Gaia (Collaboration et al. 2018) and SDSS (Ahn et al. 2012).We produce thumbnail images in FITS and PNG format, and record sky position angle information indicating the anti-solar and anti-motion vectors, as computed by JPL Horizons.
A member of the science team visually examines all of the thumbnail images produced by our follow-up pipelines and searches for activity indicators, such as tails and comae.Thus far we have examined over two million thumbnail images as Active Asteroids follow-up and in developing the HARVEST pipeline.These data are from myriad sources and vary greatly in image quality and character (e.g., chip gaps, image orientation), and the vast majority of these data do not contain any useful information.Thus we do not submit images from these secondary pipelines for volunteer examination.

Follow-up Observations
Objects we deem appropriate for follow-up observations are added to an internal list of candidates needing further telescope observations.Hereafter we refer to telescopes by their name or the instrument: Dark Energy Camera (DECam), Inamori-Magellan Areal Camera and Spectrograph (IMACS), Gemini Multi-Object Spectrograph (GMOS), Inamori-Magellan Areal Camera and Spectrograph (IMACS), the Large Binocular Telescope (LBT), Lowell Discovery Telescope (LDT), and the Vatican Advanced Technology Telescope (VATT).Table 1 lists the telescopes and facilities our team employs to carry out follow-up observations of activity candidates.We make use of ground-based facilities in both hemispheres to maximize declination coverage.For target selection, we prioritize objects that are near perihelion (i.e., true anomaly angles of f ≥ 290 • and f ≤ 70 • ).

DYNAMICAL CLASSIFICATION
To gain insight into the objects we are studying we classify them in a dynamical class, such as JFC or Centaur.A common tool employed to distinguish between different dynamical classes is the Tisserand parameter (Tisserand 1896) with respect to Jupiter, which conveys the relative influence of Jupiter on a given object's orbit, and is defined by where e and i are the orbital eccentricity and inclination of the body, and the semi-major axis of the body and Jupiter are a and a J , respectively.Objects with T J < 3 have historically been considered dynamically cometary (see e.g., Carusi et al. 1987Carusi et al. , 1995)), whereas objects with T J ≥ 3 have been considered dynamically asteroidal (Vaghi 1973a,b).Objects with 2 < T J < 3 are considered to be JFCs if active (Jewitt 2009), while objects with T J < 2 are considered Damocloids (e.g., the class namesake, 5335 Damocles; McNaught et al. 1991;Asher et al. 1994 if inactive, or Halley-type comets or long-period comets, such as the retrograde, T J = −0.395C/2014 UN 271 (Bernardinelli-Bernstein) (Bernardinelli et al. 2021), if active (Jewitt 2005).Importantly, objects with T J > 3 have orbits that do not cross the orbit of Jupiter (Levison 1996), i.e. the orbits are entirely interior, or exterior, to the orbit of Jupiter.
It is also important to note that objects may appear to be inactive upon their initial discovery, and consequently are referred to as asteroidal even though their dynamical properties (e.g., T J ) are suggestive of a cometary body.In the interim these objects may be referred to as an asteroid on a cometary orbit (ACO, Fernández et al. 2005;Licandro et al. 2006;Kim et al. 2014, a dormant comet (Ye et al. 2016), a comet nucleus (Lamy et al. 2004), an extinct comet (Fernández et al. 2001), or a Manx comet (Meech et al. 2014).
We adopt the Jewitt (2009) definition whereby Centaurs (1) have perihelia and semi-major axes between the semi-major axes of Jupiter (a J ≈ 5 au) and Neptune (a N ≈ 30 au), and (2) are not in 1:1 MMR with any planet.
Membership in the quasi-Hilda family cannot be established by orbital parameters alone, although rough Tisserand parameter constraints of 2.9 ≤ T J ≤ 3.1 have proven useful for locating candidate members (see Oldroyd 2022).To provide additional diagnostic information we examine Jupiter corotating reference frame orbital plots (Figure 3) to establish similarities to other established quasi-Hildas, as described in Chandler et al. (2022).Hilda asteroids are in stable 3:2 interior mean motion resonance with Jupiter (Murray & Dermott 1999), but the quasi-Hildas are near, not within, this resonance.Notably, Quasi-Hildas have a distinguished tri-lobal feature in the reference frame corotating with Jupiter (Figure 3e).We generate these plots by integrating the object of interest for 200 yr along with the Sun and the planets (excluding Mercury) using the REBOUND IAS15 N -body integrator (Rein & Liu 2012;Rein & Spiegel 2015) in python.

RESULTS
The Active Asteroids project has prompted discoveries by our team before and after the project launch.As it is the goal of this manuscript to encapsulate all of the results stemming from our program to date, we briefly summarize all findings here and identify connections to new results introduced both in this manuscript and in the interim (Section 5).We first present our pre-launch discoveries (Section 7.1) in chronological order, and our post-launch discoveries (Section 7.2) by dynamical class, with constituent objects sorted by provisional designation (and thus original object discovery date).As discussed in Section 2, we first conducted a proofof-concept to demonstrate the viability of DECam data as a source of images for activity discovery (Chandler et al. 2018).We justified this determination in part by identifying one known active asteroid, (62412) 2000 SY 178 (Sheppard & Trujillo 2015), after searching the 35,640 images we produced with the initial version of the HARVEST pipeline (Section 2).These images consisted of 11,703 unique minor planets, allowing us to produce a rudimentary activity occurrence rate estimate of one in ∼12,000, in rough agreement with the existing 1:10,000 estimate (Jewitt et al. 2015;Hsieh et al. 2015).
A consideration for drawing statistically robust conclusions from our project is volunteer ability to detect activity, as discussed in Section 4. For example, Active Asteroids volunteers did not flag an image of (62412) 2000 SY 178 as an activity candidate as defined by our analysis system (Section 4).However, the image (Figure 1) does indeed show a faint tail and was in fact drawn from the same data in which Sheppard & Trujillo (2015) made the activity discovery.Yet we see many other instances where volunteers identified activity in known active objects that our team had difficulty spotting.With different individuals involved, both volunteer and science team, we do not find it surprising that outcomes are not entirely predictable, but we feel it is important to emphasize the point here.These considerations reinforce the need for many volunteers to examine a given image.

Active Asteroid (6478) Gault
In 2019 January, asteroid (6478) Gault (Figure 4a; Prop.ID 2012B-0001, PI Frieman, observers SK, DT, NFM) was reported to be displaying activity (Smith et al. 2019;Hui et al. 2019;Jewitt et al. 2019;Marsset et al. 2019;Moreno et al. 2019;Ye et al. 2019;Devogèle et al. 2021).For the first time, our team made use of the HARVEST pipeline, which was not yet complete, to identify images of Gault in DECam data.In Chandler et al. (2019) we reported our subsequent discovery that Gault had been active during multiple prior epochs.We found Gault's activity was not correlated with perihe- lion passage, and we postulated that Gault is recurrently active due to rotational spin-up, supported by Kleyna et al. (2019) findings of YORP-induced effects on Gault.
Our findings stemmed from tools we created to help us understand potential observational biases and correlation effects with perihelion passage and activity outbursts.Even with this strategy in place, both observability (the number of hours an object is above the horizon as observed from a given observatory) and perihelion passage must be coincident to maximize the chances an object will be shown to volunteers.For example, only the 2016 perihelion passage coincided with a peak in observability.At other times (e.g., 2013, 2019) Gault was highly observable, but it was not near perihelion, or Gault was minimally observable (or not observable) at perihelion, as was the case in 2012.

Active Centaur C/2014 OG392 (PANSTARRS)
Our team discovered activity emanating from Centaur 2014 OG 392 (Figure 4g; Prop.ID 2019A-0337, PI Trilling, observer C. Trujillo), now designated C/2014 OG 392 (PANSTARRS) following our discovery, while testing our project workflow in preparation for the Active Asteroids program.As part of this testing we treated the object as if it had been discovered by volunteers, first carrying out an archival investigation, then follow-up telescope observations, as described in Section 5.
We successfully confirmed the presence of activity during our own observations with DECam (UT 2019 August 30 250 s VR-band, Prop.ID 2019A-0337, PI Trilling, observer C. Trujillo) on UT 30 August 2019 (Chandler et al. 2020).Given the elapsed time between the archival activity and new observations, it is likely the object had been active for years.Additional observations we obtained with the 4.3 m LDT enabled us to classify C/2014 OG 392 as a red centaur (see review by Peixinho et al. 2020), to estimate a diameter of 20 km, and carry out mass loss estimates.We also introduced a novel technique to estimate the species likely responsible for sublimation at the experienced orbital distances, in this case carbon dioxide and/or ammonia.
Since project launch we have submitted all thumbnail images available of Centaurs for classification.Project volunteers did identify C/2014 OG 392 as active, including the original archival images that prompted our publication, as well as the new observations we conducted.Moreover, volunteers identified activity in images of other known active Centaurs.However, while we are actively investigating several leads stemming from Active Asteroids, C/2014 OG 392 remains the only active Centaur discovery by our program thus far.

Main-belt Comet 433P
Just prior to project launch, (248370) 2005 QN 173 , subsequently designated 433P, was discovered to be active (Fitzsimmons et al. 2021).In addition to the HARVEST pipeline we also debuted our secondary pipelines developed for archival investigation (Section 5.1).We successfully identified 81 images of the object, spanning 31 observations, in which we could confidently identify 433P.Of these we found a single image (Figure 4f), dated UT 2016 July 22 (Prop.ID 2016A-0190, PI Dey, observers Dustin Lang, Alistair Walker), that showed 433P unambiguously active with a long, thin tail oriented towards the coincident anti-solar and anti-motion vectors as projected on sky (Chandler et al. 2021).Our discovery of a previous activity epoch, that occurred near perihelion, along with 433P's probably C-type spectral class (Hsieh et al. 2021), allowed us to classify the object as a MBC.At the time, just ∼15 of these objects were known.
We introduced Wedge Photometry as an activity detection and measurement tool in Chandler et al. (2021).This tool, which shares similarities to one by Sonnett et al. (2011), measures flux in annular regions around a target, using different angular wedge sizes, to identify the presence of a tail and to measure its angle for comparison with ephemeris computed anti-solar and antimotion projected vectors.As with Ferellec et al. (2022), who developed a similar tool around the same time, we found background sources to significantly impede the practicality of this approach.In the future, especially for surveys with high-quality templates -as should be the case for the Legacy Survey of Space and Time (LSST) -the tool may be of practical use to filter out false positives and thus improve the overall quality of images we provide volunteers for classification.

Post-launch Discoveries
For the remainder of this Section, we discuss 16 objects, all classified as active by Active Asteroids volunteers and brought to our team's attention as a result.A representative thumbnail showing activity for each object is provided in Figure 4.
We classify each object into a dynamical class, and in the process refer to (1) object-specific properties (e.g., inclination, perihelion distance), and (2) the gallery of Jupiter corotating reference frame plots (Figure 5).
Table 3 provides a unified collection of data pertaining to the observed activity, most notably the date ranges of activity along with corresponding heliocentric distances r H and true anomaly angles f .With the exception of 282P/(323137) 2003 BM 80 , all objects are referred to by their primary provisional designation, with full and al-ternate designations for each object listed in corresponding subsections.In the tables and subsections provided we generally group objects by dynamical class first, then sort objects by provisional designation within each dynamical class.

Active Asteroids
The Active Asteroids program has thus far led us to discover four new active asteroids: 2007 FZ 18 , 2010 LH 15 (seen to be active at 2 apparitions), 2015 FW 412 , and 2015 VA 108 .They have T J values ranging from T J = 3.160 to T J = 3.351, placing them all firmly outside of the Jupiter Family Comet or quasi-Hilda regimes.As indicated in Table 3, all activity took place near perihelion passages, with the earliest activity at a true anomaly angle f = 320 • , and the latest f = 33 • .For all active asteroids we identified, this behavior is consistent with sublimation-driven activity, and thus these objects are all MBC candidates.The recurrent activity we found for 2010 LH 15 is additional evidence supporting sublimation-driven activity as the underlying mechanism, thus it is likely an MBC.
2017 QN84 -2017 QN 84 activity (Figure 4r; UT 2017 December 23, Prop.ID 2017B-0307, PI Sheppard) was identified by Active Asteroids participants and initially reported on project forums.While we only identified a single image of 2017 QN 84 with activity, we produced a comparison image that clearly demonstrates there were no background sources that could be mistaken as activity (Chandler 2022).Moreover, the activity extends from 2017 QN 84 towards the coincident anti-solar and anti-motion directions (as projected on sky), approximately 2 o'clock (PA ∼ 300 • East of North), suggesting a physical phenomenon rather than an image artifact.On the date we see the activity, 2017 QN 84 was outbound at r H = 2.62 au and f = 38 • .We classify 2017 QN 84 (T J = 2.944, a = 3.77 au, e = 0.34, i = 12.1 • , q = 2.48 au, Q = 5.06 au) as a JFC.

Classification Metrics
We describe here a brief preliminary analysis of the Active Asteroids classifications and results.We caution that the inferences herein (1) have not been debiased in any way, and we impart significant biases in our subject selection process (e.g., we sort objects submitted for classification by distance from perihelion; Section 3.3), (2) classifications are incomplete (e.g., ∼241,000 of ∼1.1 million main-belt asteroids have been examined by the project thus far), (3) our investigation into newfound activity epochs is ongoing, and (4) some dynamical classifications require dynamical simulations (Section 5) and thus may have been incorrectly labeled in the past.While we primarily make use of the object class returned by the Quaero service (Berthier et al. 2006), some classes contain objects with ambiguous membership, for example, JFCs that also qualify as NEOs.
Table 2 shows metrics by object class, with analyses considering subjects (images) and unique objects.Overall, Active Asteroids volunteers classified 1.3% of the images as showing activity, with our team concurring 33.3% of the time (i.e., 0.04% of all images examined).We investigated these candidates (Section 5) except for training images (Section 3.4), which are included in Table 2 to indicate, for example, volunteer expertise (Section 4).
Despite the disclaimers mentioned above, especially regarding biases and classification incompleteness, we can still make some rough inferences.Of the 240,989 unique asteroids examined by volunteers, they labeled 25 (0.010%) as showing activity, consistent with prior activity occurrence rate estimates, roughly 1 in 10,000 (Hsieh et al. 2015;Jewitt et al. 2015;Chandler et al. 2018), or 0.01%.
Active Asteroids volunteers examined 193 Centaurs, of which 11 (5.7%) qualified as activity candidates by our enhanced classification analysis (Section 4).To assess the Centaur activity occurrence rate in the context of the broader Centaur population we first needed a list of Centaurs from which we shall make our comparisons.(Oldroyd et al. 2023c).
We also must define Centaur as these objects are described by multiple definitions in the literature.Derived from Jewitt (2009), we define a Centaur as an object (1) with a semi-major axis a and perihelion distance q between the aphelion distances of Jupiter and Neptune (i.e., , and (2) not in 1:1 mean-motion-resonance with a giant planet.This latter requirement excludes the 24 Neptune Trojans and 2 Uranus Trojans known as of UT 2023 December 9, and the a and q constraints exclude objects that cross the orbit of Jupiter.We queried the MPC list of Centaurs and Scattered-Disk Objects5 and the JPL Small Body Database (via their query tool6 for objects that match our a and q requirements.The results largely overlapped, noting that (1) the MPC did not include objects with comet designations in their list, and (2) in two cases (2010 HM 23 and 2015 FZ 397 ), orbital element disagreement between the two services (e.g., 2010 HM 23 a = 32.35au via JPL, and a = 27.90 via the MPC) caused an object to appear on one list but not the other.In both cases, we included the objects on our final list.Subsequently, we removed the known Trojans.
Of the 346 Centaurs on our list, 31 are active Centaurs, indicating an activity occurrence rate among the Centaurs of about 9%.This figure is in agreement with Peixinho et al. (2020), and lower that 13% rate of Jewitt (2009) that was measured when only 92 Centaurs were known (of which 12 were active).Because Active Asteroids volunteers only examined images of about half of the known Centaurs, it is unsurprising that the 5.7% identification rate differs from our 9% rate.
Similarly, we consider the ratio of JFCs to ACO, where the former have shown activity, and the latter have not (Licandro et al. 2006(Licandro et al. , 2016)).We queried the JPL Small Body Database for objects with Tisserand parameter's with respect to Jupiter 2 < T J < 3, the canonical range for JFCs, and excluded the Jupiter Trojans as well as all objects from our Centaur list.We note that we only counted one object per parent designation (i.e., we excluded fragments except for the primary designation).We flagged each object on our list as either active or inactive, based on their cometary designation or lack thereof, and we also flagged the 13 qualifying objects included in this work as active, plus another established active object, 2008 GO 98 (García-Migani & Gil-Hutton 2018).Of the 14,407 minor planets on our ACO + JFC list, 668 have been observed to be active.Thus, we find the apparent occurrence rate (i.e., observed fraction) of active objects in the ACO + JFC population to be 4.6%.We reiterate our query did not restrict our population selection by any physical property (e.g., albedo, color; Licandro et al. 2016) other than observed activity, and we did not limit our ACOs or JFCs populations using the Tancredi (2014) method.
The other classes cannot be meaningfully evaluated at this time due to, for example, currently unresolved ambiguities in overlapping class definitions (e.g., JFC, QHC).We carried out analyses to better understand performance when classifying data.Here we discuss the total number of submitted thumbnail images, as well as metrics from select dynamical classes of relevance to this discussion.At the time of these analyses, there were 406,082 sample (i.e., non-training) images in the Active Asteroids project on Zooniverse.Of these, 4,171 (1.03%) of the images qualified as candidates by our analysis system (Section 4).Our team flagged 526 (12.6%) of these candidates as warranting additional investigation, which we define as reaching a threshold of ≥ 4 based on our activity likelihood score (Section 3.4).If we do not limit our assessment to candidates flagged by volunteers, we found an additional 138 thumbnail images that our team had previously flagged as candidates that the project had not; these images are members of dynamical classes that we examined extensively during project preparations.
The lowest fraction of objects classified as candidates by volunteers were the main-belt asteroids.Of the 300,526 images of Main-belt asteroids submitted for classification (Table 7.3), 2,707 (0.9%) qualified as candidates based on analysis of volunteer classifications.Of these, our team flagged 258 (9.5%) as warranting followup.Conversely, the highest fractions occurred with the comets.Of the 1,150 sample (non-training) images of known comets, 300 (26.1%) were flagged as candidates, and 267 (89.0%) of these our team also classified as warranting follow-up.

Thumbnail Classification Rate and Completeness
As of 2023 July 3 -the date of our last Zooniverse data export -volunteers averaged 12, 770 ± 10, 750 classifications per day, with a maximum rate of 129,338 classifications/day taking place on the project launch date, 2021 August 31.These figures include both training and sample images, and exclude dates with < 1, 000 classifications, which typically occur when no new sample data is available on the project for volunteers to classify.Of the 6,543,368 classifications in this data export, 353,058 (5.4%) were training images.Taking this training fraction into account, the mean retirement rate (nominally 15 classifications per image; Section 3) is 805 ± 678 sample images/day, and our nominal peak rate covers ∼8,000 sample images/day.Some of the variation we observe in the classification rate is due to external factors, such as media attention and publications of our findings.Internal factors are driven by when we email newsletters to participants, and lulls between subject sets are due to pauses in-stilled while we examine previous results and prepare a new batch.The cause of the remaining variation is unknown, though we have speculated that seasonal societal effects, such as vacation times, for example, may be partially responsible.Nevertheless, with nearly two years of data to draw upon, we will consider the average and peak rates mentioned above for the remainder of this discussion.
It is worth mentioning here that we typically consider three modes of optimizing results through our Citizen Science project.All three modes can either increase the number of images examined or reduce the time it takes to complete the examination of an entire subject set.(1) Additional participation by existing volunteers, or an increased number of participants.( 2) Optimized analysis of classification data that is capable of reducing the time to retirement for at least a portion of subjects in a subject set.(3) Reducing the amount of data needing classification, through either (a) more advanced automated vetting, or (b) measured decisions to exclude certain data.The discussion that follows focuses on decisions to reduce the volume of data, however we continue to work towards improving all three areas, especially techniques involving applications of artificial intelligence (AI).However, those areas are still under development and outside the scope of this manuscript.

Current DECam Dataset
Time to Complete Existing Thumbnails -HARVEST has produced roughly 18 million vetted thumbnail images (Section 2).At the peak rate (8,000 images/day) this works out to 2,250 days, or about six years until our program has classified all of the DECam-derived images, though this assumes the dataset is static (it is not; see below).At the mean rate, however, the completion time would be 22,360 days, or roughly 60 years.
Staying Current -So far Active Asteroids has exclusively shown volunteers DECam images, from instrument first light (2012 September) to present.The archive continues to grow, and HARVEST runs daily.We estimate the occurrence of minor planet images to be that of our average number of vetted thumbnail images produced per day, ∼5,000.At our current average daily retirement rate of 805 images/day, there is a significant deficit (i.e., we will not catch up at this rate).The peak rate (8,000 images/day) would be sufficient to stay current but would result in significant delays in processing the remainder of the existing thumbnails (see below).
Time to Completion while Staying Current -As mentioned above, we are not presently able to examine 100% of the thumbnails produced daily by the HARVEST pipeline.The peak rate would leave just nine hours (0.375 days) daily for classifying the remaining 18 million images.In this case, it would take roughly 16 years to get caught up while also staying current with newly available vetted images supplied by the HARVEST pipeline.Aside from increasing project participation, we can overcome this classification shortfall by implementing some of the procedures discussed in the subsequent section.Then it becomes possible to examine all of the DECam dataset with a reasonable degree of activity completeness before the commencement of LSST (mid-2025).

Considering LSST
The Legacy Survey of Space and Time (LSST) will be an all-sky survey conducted in the southern hemisphere with an 8.4 m diameter telescope at the Vera C. Rubin Observatory atop Cerro Pachón in Chile (Ivezić et al. 2019).The survey strategy to acquire 1,000 images per night with its 3.2 gigapixel camera is expected to produce on the order of 100 Tb of data per night.The challenges of working with this scale of data are extraordinary (e.g., Kelley et al. 2021;Vera C. Rubin Observatory LSST Solar System Science Collaboration et al. 2021;Breivik et al. 2022;Schwamb et al. 2023), and Citizen Science endeavors are no exception.Our program, initially selected for funding by the NSF Graduate Research Fellowship Program (GRFP), was designed with LSST in mind, and as survey commencement approaches (nominally mid-2025), we revisit the implications of such a data deluge for the Active Asteroids program.
Unfiltered Nightly Output -Ivezić et al. ( 2019) estimated roughly 5.5 million minor planet detections of the 11 million objects they simulated -a roughly 50% detection rate -so the estimated 8,000 minor planets within each LSST field translates to 4,000 detections per image.Depending on the final cadence selection, LSST plans to image ∼1,000 fields per night.Thus, we estimate 4 million minor planet detections per night.
Default HARVEST Vetting -Our automated vetting, such as the delta magnitude limits (Section 2.5), filters out ∼70% of thumbnails.Applied to nightly LSST data, this would leave ∼1.2 million minor planet detections per night.For perspective: LSST data processed via our HARVEST pipeline will produce the same quantity of vetted thumbnail images that make up our entire project (∼12 years worth of DECam data) every two weeks.A single night of LSST minor planet detections would require 150 days of citizen science efforts at our peak rate (8,000 images/day) to classify, clearly an impossibility.
Perihelion Proximity Filtering -To reduce the number of images volunteers are asked to examine we can impose additional requirements on sample data, but we acknowledge that these will result in the loss of discoveries.For example, we can require that thumbnails show objects within 20% of their perihelion distance (as described by our percentage to perihelion metric; Section 3.3).This would reduce the number of images to classify by roughly 2/3 but at the cost of missing discoveries of objects that are active at times other than near perihelion, such as the notable case of ( 6478) Gault (Section 7.1.2).Importantly, this approach has a significant advantage in that image data need not be accessed to achieve this reduction.
Delta Magnitude Limits -We can reduce the number of images needing classification by adopting a stricter delta magnitude limit (how many magnitudes brighter an object appears above depth; Section 2.5) than our default threshold of ∆ mag ≤ −1.A ∆ mag ≤ −2 limit would provide a 30% reduction of viable thumbnails (reducing 1.2 million/day to 840,000/night), or a ∆ mag ≤ −3 a ∼60% reduction (down to 480,000/night).However, this approach favors objects that are closer to perihelion because they are brighter at that point in their orbit, and disfavors faint objects, such as Centaurs, which may always be faint.For example, with a typical limiting magnitude of DECam in our data of V ≈ 23 (Chandler et al. 2018), imposing ∆ mag < −3 eliminates all thumbnail images in which an object is fainter than an apparent magnitude of 20.
As with Perihelion Proximity Filtering, a major advantage of applying ∆ mag limits is that image data need not be downloaded or accessed to accomplish the filtering, unlike, for example, the automated source analysis vetting we carry out (Section 2.8).Moreover, the HARVEST pipeline applies only a rough estimate, but Rubin will provide precise photometry for each minor planet detection in LSST, allowing us to apply our filters based upon real measurements, not computed expected apparent magnitudes provided by ephemeris services.
Practical Considerations for LSST Citizen Science -We extract 126 ′′ × 126 ′′ thumbnail images to allow for extended tails which, in practice, may be much longer (e.g., Figure 1).The most comprehensive approach would be to transfer all 4 million nightly detections, requiring 4 Tb of bandwidth and temporary storage at 1 Mb per FITS cutout.The transfer rate would need to be ∼100 Mb/s to move the data in <12 hours.If we apply our stricter ∆ mag < −3 threshold we can reduce the number of thumbnails to roughly 1 million images (1 Tb storage/bandwidth, 25 Mb/s throughput).Even considering our peak classification rate, these data still need to be reduced by an additional two orders of magnitude.Our team is actively pursuing machine learning techniques with the intent to filter out images with an inactive minor planet, however this is a work in progress, and it is yet unclear whether or not AI applications will perform well enough to accomplish the requisite filtering for LSST-scale data.The remaining 10,000 thumbnails per day would require 10000 × (1Mb + 0.5Mb) = 15Gb of long-term storage daily, of which 5 Gb would nominally be transferred to Zooniverse as subjects for classification.Unaccounted for are tabular results and other output stemming from the analysis that will need to be saved.
The computing requirements for this work are significant.Consider a scenario in which all data must be processed within 12 hours, following the 12 hours of data transfer described above.We acknowledge that these could be accomplished in tandem if enough resources are available, so we will consider the lack of overlap a safety buffer to accomplish all compute and transfer tasks.For the full set of data (4 million observations), automated vetting examinations would need to take place at a rate of ∼100/s for a single Central Processing Unit (CPU)/Graphics Processing Unit (GPU) requirement.Similarly, if 1 s is required per examination, 100 CPU/GPU pairs would be needed.Dividing all of these requirements by four, should we first apply the stricter ∆ mag < −3 limit, would result in requirements of either ∼25 examinations/s, or 25× greater CPU/GPU resources employed in parallel.Computing facilities with these capabilities already exist, thus these tasks can be accomplished once LSST commences operations, assuming the advanced (likely AIdriven) vetting proves viable.

Project Outlook
With the help of thousands of volunteers Active Asteroids has produced over 20 discoveries thus far, resulting in numerous publications, and dozens more candidate objects are actively under investigation by our team.Clearly Active Asteroids is successful at accomplishing its primary goal of making active body discoveries while engaging the public in the scientific endeavor.Our engagement with volunteers continues to grow, with several Active Asteroids participants now included as authors on publications, including this manuscript.As we continue to innovate optimizations for the entire process, from HARVEST pipeline to Citizen Science to follow-up investigation, we optimistically anticipate discoveries to only increase in frequency as the project moves forward.Moreover, these optimizations include improved image vetting designed for both the current project and the upcoming LSST.Anyone with an internet connection who can visually examine images can participate by visiting http://activeasteroids.net.

SUMMARY
We set out to discover active asteroids and other active minor planets in order to further our understanding of astrophysical processes at play in the solar system, and to help map the solar system's volatile distribution so that we may better understand, for example, the origins of water on Earth (Section 1).We selected DECam archival images as our primary data source because of its wide aperture that enables the detection of faint activity.We demonstrated the suitability of these data for activity detection in our proof-of-concept Searching Asteroids For Activity Revealing Indicators (SAFARI; Chandler et al. 2018), summarized in Section 7.1.1.
Because of the overwhelming volume of images (> 16 million) we sought to scrutinize, we decided to seek help from the public by constructing a Citizen Science program, Active Asteroids (Section 3).There we ask volunteers to examine the images of known minor planets we produced with our pipeline, Hunting for Activity in Repositories with Vetting-Enhanced Search Techniques (HARVEST), that we built for this purpose (Section 2).The project, now a NASA Partner, launched on 31 August 2021 on the Zooniverse platform.
As of UT 2023 July 8, some 8,300 participants have carried out 6,700,000 classifications of ∼430,000 images of minor planets we provided.These data occupy 636 Gb of storage: 424 Gb of FITS thumbnail images for scientific analyses, and 212 Gb of PNG thumbnails for the Active Asteroids project hosted on Zooniverse.We derived these cutouts from ∼141,000 archival DECam images, about 40 Tb of data.Given the 2.2 • DECam FOV, the total searchable image area within this dataset is roughly 682,440 deg 2 .
The novel classification analysis approach we introduced in this work (Section 4) has been crucial for yielding the numerous promising activity candidates we actively investigate through archival image searches and follow-up observations (Section 5).Our team has examined over two million thumbnail images by eye, including as part of our follow-up investigation into candidates identified through the Active Asteroids program.As of 2023 September 18, our Citizen Science program has yielded ∼230 unique minor planet activity candidates that our team has subsequently vetted, including 145 known cometary objects.
We emphasize that our experience with Citizen Science as a paradigm for addressing image-based science questions has made it clear that volunteers alone cannot possibly examine all of the image data output by LSST-scale programs.However, justified analytic filtering (Section 2) can substantially reduce the amount of data needing examination, and further applications of AI-informed filtering (Section 8.3) will enable fruitful Citizen Science with LSST-scale datasets.
Our follow-up observing campaign is designed to efficiently leverage telescopes with apertures appropriate to the faintness of the objects we are investigating.The telescopes range in diameter from the 1.8 m Vatican Advanced Technology Telescope (VATT), to the Apache Point Observatory (APO) 3.5 m and Lowell Discovery Telescope (LDT) 4.3 m, to the 6.5 m Baade, 8.1 m Gemini telescopes, and the twin 8.5 m Large Binocular Telescope (LBT).Our observations with these facilities, both during project preparations and after the launch of the Active Asteroids program, spanning well over 100 nights of observations (including partial nights and nights with poor observing conditions).We have observed hundreds of activity candidates, many of which we are still actively pursuing.
We reiterate that the metrics that follow are preliminary as they have yet to be debiased, an investigation that will be performed as part of a follow-up work.Active Asteroids volunteers carried out 6,700,000 classifications (Section 3), with about 1.3% of all (430,000) images they examined classified as showing activity (as indicated by our enhanced classification analysis; Section 4).Of the asteroids, participants found 25 (0.010%) were activity candidates as defined by our enhanced classification analysis (Section 4).This value is consistent with the frequently-cited 1 in 10,000 estimates (Hsieh et al. 2015;Jewitt et al. 2015;Chandler et al. 2018).
Volunteers examined roughly half of the known Centaur population and identified 5.7% as active.Our examination of the whole population indicates a Centaur activity occurrence rate of around 9%, a value significantly lower than previously reported (13%; Jewitt 2009), though our sample size is roughly four times larger than what was available then.We examined the activity of the ACO + JFC population (i.e., 2 < T J < 3), including the activity discoveries from our project, and computed the occurrence rate (i.e., observed fraction) of activity in the ACO + JFC population to be 4.6%.
The Active Asteroids project is ongoing and can be accessed through the project website http:// activeasteroids.net.Participation is easy, and intuitive, and can take as little as a few minutes to contribute.These characteristics also make Active Asteroids an excellent tool for teaching solar system astronomy to a wide range of audiences.Pence (1999).World Coordinate System (WCS) corrections facilitated by the Astrometry.netsoftware suite (Lang et al. 2010).
This work was supported in part by NSF award 1950901 (NAU REU program in astronomy and planetary science).This research has made use of data and/or services provided by the International Astronomical Union's Minor Planet Center.This research has made use of NASA's Astrophysics Data System.This research has made use of the Institut de Mécanique Céleste et de Calcul des Éphémérides (IMCCE) SkyBoT Virtual Observatory tool (Berthier et al. 2006).This work made use of the FTOOLS software package hosted by the NASA Goddard Flight Center High Energy Astrophysics Science Archive Research Center.This research has made use of SAOImageDS9, developed by Smithsonian Astrophysical Observatory (Joye 2006).This work made use of the Lowell Observatory Asteroid Orbit Database astorbDB (Bowell et al. 1994;Moskovitz et al. 2021).This work made use of the astropy software package (Robitaille et al. 2013).
These results made use of the Lowell Discovery Telescope (LDT) at Lowell Observatory.Lowell is a private, non-profit institution dedicated to astrophysical research and public appreciation of astronomy and operates the LDT in partnership with Boston University, the University of Maryland, the University of Toledo, Northern Arizona University and Yale University.The Large Monolithic Imager was built by Lowell Observatory using funds provided by the National Science Foundation (AST-1005313).NIHTS was funded by NASA award #NNX09AB54G through its Planetary Astronomy and Planetary Major Equipment programs.
) Test project viability via a Zooniverse "beta release."(3) Formally launch Active Asteroids for public use.(4) In a cyclic fashion, (a) interact with volunteers, (b) download and analyze results, (c) prepare and upload a new batch

Figure 1 .
Figure 1.This UT 2014 March 28 Dark Energy Camera (DECam) thumbnail image of active asteroid (62412) 2000 SY178 (at center) received a score of 0.35 via our analysis system (Section 4), below the 0.473 threshold needed to qualify as an activity candidate.The faint tail seen oriented towards PA ∼ 150 • North through East (roughly 7 o'clock), extends beyond the edge of the image.The image FOV is 126 ′′ ×126 ′′ , with North up and East left, and an overlaid green reticle as shown to Active Asteroids volunteers.DECam image from Prop.ID 2014A-0479, PI Sheppard, observer S. Sheppard.

Metric 1 :
Training Image Accuracy -This metric considers users who perform well with training data as having more expertise than those who perform poorly, thus the more expert users should be given more weight.To quantify training image performance we measure the ratio of a user's successful training image classification, Y training , to their total number of training image classifications, T training , as

Figure 4 .
Figure 4. Minor planets with activity discoveries resulting from the Active Asteroids project.a -f are active asteroids and main-belt comet (MBC) candidates; g is an active Centaur; h -l are active quasi-Hilda asteroids; m -t are Jupiter Family Comets.In all panels the object is at center, North is up and east is left, and the FOV is 126 ′′ ×126 ′′ .The anti-solar (yellow filled arrow) and anti-motion (black arrow with red border) directions as projected on sky are shown in the top-left corner of each image.

Figure 5 .
Figure 5. Jupiter corotating frame plots for the objects presented in this work.In all panels, the Sun (star marker) is at the center, with Jupiter and the minor planet indicated by orange and blue markers, respectively.All axes are in units of au.Acronyms: active asteroid (AA), Jupiter Family Comet (JFC), Main-belt Comet (MBC), and Quasi-Hilda Comet (QHC).

2009 DQ118 -
We found > 20 images of activity of 2009 DQ 118 (Figure 4j; Prop.ID 2016A-0189; PI Rest; observers A. Rest, DJJ) with activity from this epoch, spanning two consecutive days, from UT 2016 March 8 to UT 2016 March 9, when 2009 DQ 118 was at a r H = 2.55 au and f = 322 •(Oldroyd et al. 2023a).Our follow-up observations with the Astrophysical Research Consortium (ARC) instrument on the APO 3.5 m telescope (Sunspot, USA) and the IMACS instrument on the 6.5 m Baade Telescope (Las Campanas Observatory, Chile) revealed 2009 DQ 118 was active again, indicating that sublimation is the most likely mechanism responsible for the observed activity(Oldroyd et al. 2023b).Our dynamical modeling (Section 6) indicated 2009 DQ 118 (T J = 3.004, a = 3.58 au, e = 0.32, i = 9.4 • , q = 2.43 au, Q = 4.72 au) is an active QHO.

4
http://www.aerith.net/comet/catalog/2019OE31/index.html2005 XR132 -Active Asteroids volunteers classified a DECam image of 2005 XR 132 (Figure 4m; UT 2021 March 26, Prop.ID 2021A-0149, PI Zenteno, observer A. Zenteno) as showing activity, and our archival investigation revealed additional activity images from ZTF 2008 QZ44 -We identified activity in 2008 QZ 44 (Figure 4o; UT 2008 November 20 Canada France Hawaii Telescope (CFHT) MegaPrime, PI Hoekstra, observers "QSO Team") via two independent means (Chandler et al. 2023g).A member of our team discovered images of 2008 QZ 44 as part of a separate investigation, and volunteers from the Active Asteroids project flagged two images of 2008 QZ 44 as showing activity.The nine MegaPrime images, all from UT 2008 November 20 (r H = 2.43 au and f = 29 • ), clearly show a tail in the anti-solar direction.The second activity epoch (UT 2017 November 12 -13, r H = 2.90 au, f = 68 • ; Prop.ID 2014B-0404, PI Schlegel, observers C. Stillman, J. Moustakas, M. Poemba) is visible in DECam) images as a tail oriented between the anti-solar and anti-motion angles.We classify 2008 QZ 44 (T J = 2.821, a = 4.19 au, e = 0.44, i = 11.4 • , q = 2.35 au, Q = 6.04 au) as a JFC.2012 UQ192 -Volunteers flagged (551023) 2012 UQ 192 (Figure 4p; UT 2014 April 30, Prop.ID 2014A-0283, PI Trilling, observers D. Trilling, L. Allen, J. Rajagopal, T. Axelrod), alternate designation 2019 SN 40 , as showing activity (DeSpain et al. 2023).Our follow-up archival investigation revealed a total of four images from the same orbit that showed an unambiguous tail oriented towards the anti-motion direction, PA ∼300 • East of North (roughly the 2 o'clock position).At the time, (551023) 2012 UQ 192 was outbound from perihelion.Activity is evident in DECam images from UT 2014 April 30 (r H = 2.99 au, f = 96.5 • ), UT 2014 May 5 (r H = 3.02 au, f = 97.5 • ), and in > 20 ZTF images between UT 2020 November 12 (r H = 2.08 au, f = 40 • ) and UT 2021 May 5 (r H = 2.84 au, f = 90 • ).With recurrent activity near perihelion, the activity is most likely caused by sublimation.We classify (551023) 2012 2018 VL10 -The DECam images of 2018 VL 10 (Figure 4t; UT 2018 December 31, Prop.ID 2018B-0122, PI Rest, observers A. Zenteno, A. Rest) we identified having activity (?) range from UT 2018 December 31 (r H = 1.42 au, f = 0.0 • ) to UT 2019 February 01 (JFC), Main-belt Comet (MBC), Quasi-Hilda Comet (QHC).AA* denotes an MBC candidate.The first and last activity were identified or observed as part of this work, with the exceptions of Gault epoch #4, 282P epoch #1, and 433P epoch #1, as described in the text.f← and f→ are the true anomaly angles at the start and end of activity observations.rH,← and rH,→ are the corresponding heliocentric distances of an object for the observed ranges of activity.This research received support through the generosity of Eric and Wendy Schmidt by recommendation of the Schmidt Futures program.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 work was made possible in part through the State of Arizona Technology and Research Initiative Program."GNU's Not Unix!" (GNU) Astro astfits (Akhlaghi & Ichikawa 2015) provided command-line FITS file header access.C Flexible Image Transport System Input Output (CFITSIO) enabled FITS compression and more a l'Energes Atomique (CEA)/Département d'Astrophysique, de physique des Particules, de physique Nucléaire et de l'Instrumentation Associée (DAPNIA), at the Canada France Hawaii Telescope (CFHT) which is operated by the National Research Council (NRC) of Canada, the Institut National des Science de l'Univers of the Centre National de la Recherche Scientifique (CNRS) of France, and the University of Hawaii.The observations at the Canada-France-Hawaii Telescope were performed with care and respect from the summit of Maunakea which is a significant cultural and historic site.

Table 3 .
Activity Circumstances