Source-linked AI summary
The Pan-STARRS Moving Object Processing System
Larry Denneau, Robert Jedicke, Tommy Grav, Mikael Granvik, Jeremy Kubica, Andrea Milani, Peter Veres, Richard Wainscoat, Daniel Chang, Francesco Pierfederici, N. Kaiser, K. C. Chambers, J. N. Heasley, Eugene. A. Magnier, P. A. Price, Jonathan Myers, Jan Kleyna, Henry Hsieh, Davide Farnocchia, Chris Waters, W. H. Sweeney, Denver Green, Bryce Bolin, W. S. Burgett, J. S. Morgan, John L. Tonry, K. W. Hodapp, Serge Chastel, Steve Chesley, Alan Fitzsimmons, Matthew Holman, Tim Spahr, David Tholen, Gareth V. Williams, Shinsuke Abe, J. D. Armstrong, Terry H. Bressi, Robert Holmes, Tim Lister, Robert S. McMillan, Marco Micheli, Eileen V. Ryan, William H. Ryan, James V. Scotti
TL;DR
The paper addresses the need for automated, end-to-end processing of transient detections from next-generation asteroid surveys. It presents MOPS and evaluates it on simulated Pan-STARRS4-class data and prototype Pan-STARRS1 data, finding high efficiency overall while identifying reduced multi-night orbit performance on Pan-STARRS1.
Problem
Next-generation surveys require automated systems that can process transient detections, produce discoveries and identifications, and reduce follow-up inefficiency.
Method
The paper describes MOPS, an integrated moving-object processing system, and evaluates it with simulated survey populations and adapted Pan-STARRS1 data.
Results
> 99% efficiency is reported for detecting moving objects and computing orbits for most solar system objects.
Takeaways & Limitations
MOPS is a capable tool for detecting moving objects, supporting NEO and comet searches, and characterizing the main belt.
Takeaways & Limitations
MOPS performs less effectively with real Pan-STARRS1 data, and reduced efficiency is associated with Pan-STARRS1 cadence and astrometry yielding less suitable orbits.
Abstract
from arXiv · showhide
We describe the Pan-STARRS Moving Object Processing System (MOPS), a modern software package that produces automatic asteroid discoveries and identifications from catalogs of transient detections from next-generation astronomical survey telescopes. MOPS achieves > 99.5% efficiency in producing orbits from a synthetic but realistic population of asteroids whose measurements were simulated for a Pan-STARRS4-class telescope. Additionally, using a non-physical grid population, we demonstrate that MOPS can detect populations of currently unknown objects such as interstellar asteroids. MOPS has been adapted successfully to the prototype Pan-STARRS1 telescope despite differences in expected false detection rates, fill-factor loss and relatively sparse observing cadence compared to a hypothetical Pan-STARRS4 telescope and survey. MOPS remains >99.5% efficient at detecting objects on a single night but drops to 80% efficiency at producing orbits for objects detected on multiple nights. This loss is primarily due to configurable MOPS processing limits that are not yet tuned for the Pan-STARRS1 mission. The core MOPS software package is the product of more than 15 person-years of software development and incorporates countless additional years of effort in third-party software to perform lower-level functions such as spatial searching or orbit determination. We describe the high-level design of MOPS and essential subcomponents, the suitability of MOPS for other survey programs, and suggest a road map for future MOPS development.
1. Introduction
The paper presents MOPS as an integrated, automated system for discovering, identifying, and determining orbits for moving objects from survey detections. It addresses inefficiencies in follow-up and existing moving-object processing while supporting high survey efficiency and accuracy.
- Contribution: MOPS is an integrated, end-to-end moving object processing system for Pan-STARRS.It combines moving-object detection, computation, database management, and orbit-related processing.
- Existing survey processing: Modern surveys link 3-5 detections into tracklets by testing transient detections for constant linear motion.This approach supports detection at per-detection signal-to-noise levels of approximately 1.5 to 3σ.
- Survey performance: Existing surveys had strong results, identifying approximately 79% of ≥1 km near-Earth objects before 10 June 2008.The 90% discovery goal for ≥1 km asteroids was later met.
- MOPS approach: MOPS determines orbits from observations over multiple nights instead of merely reporting tracklets to the Minor Planet Center.This is intended to improve discovery rates and support more effective candidate assessment.
- Motivation: Follow-up inefficiency can waste telescope time through uncoordinated and unnecessary observations.The paper motivates automated reporting and follow-up coordination as surveys discover more and fainter objects.
- Conclusion: MOPS resolves several survey-processing issues and is capable of very high efficiency and accuracy.The system is presented as the next step in the evolution of asteroid surveys.
2. Pan-STARRS and Pan-STARRS1
Pan-STARRS combines wide-field survey telescopes with specialized observing programs and a transient-detection pipeline. Pan-STARRS1 provides the prototype platform on which these components and MOPS were validated.
- Pan-STARRS development: The Pan-STARRS project developed Pan-STARRS1 as a prototype for a next-generation distributed-aperture survey system.Pan-STARRS2 was being constructed as a second telescope essentially identical to Pan-STARRS1.
- Pan-STARRS1: Pan-STARRS1 has a 1.8 m telescope, approximately 7 deg^2 field of view, and approximately 1.4 gigapixel camera.Its observing system uses 0.26′′ pixels and can image the visible night sky from Hawai‘i to approximately r ∼21.2 in about 5 or 6 nights.
- Survey programs: The Pan-STARRS survey plan includes 3π, Medium Deep, and solar system surveys suitable for moving-object discovery.All survey components obtain at least one image pair of each field per night, while these three are specifically suitable for moving-object discovery.
- Photometric system: Pan-STARRS1 uses six passbands: gP1, rP1, iP1, zP1, yP1, and wP1.The filters were designed for different photometric and asteroid-observation objectives.
- Processing pipeline: The Image Processing Pipeline produces catalogs of transient detections that MOPS searches for moving objects.Transient detections include new or changing sources not otherwise identified as false transients such as cosmic rays or image artifacts.
3. The Pan-STARRS Moving Object Processing System (MOPS)
MOPS is an integrated system that processes transient detections through orbit determination, precovery, and attribution while independently measuring end-to-end efficiency. Its modular, incrementally processed pipeline is designed for reliable operation, though out-of-order data require further development.
- Overview: MOPS processes per-exposure transient detection lists through orbit determination, precovery, and attribution as an integrated moving-object system.
- Overview: MOPS independently characterizes subsystem and end-to-end efficiency to support correction for observational selection effects.
- Pipeline processing: Detections are grouped into intra-night tracklets and inter-night tracks, which are evaluated for asteroid orbits before rejected tracklets return to the unassociated pool.
- Pipeline processing: A derived orbit drives searches for additional detections that can further refine the asteroid orbit.
- Pipeline design: The linear nightly pipeline handles observations incrementally, but inserting data into the temporal middle forces subsequent nights to be reprocessed.
- Reliability: MOPS remains operationally resilient: database restoration returned production online within hours, and hardware failures caused no significant processing interruption.
4. MOPS Verification
MOPS was verified through large-scale simulations representing Pan-STARRS4 and Pan-STARRS1 observing conditions. It achieved very high tracklet and derived-object efficiency, with lower multi-night performance attributable partly to untuned processing limits and survey-specific conditions.
- 2-year S3M simulation: 99.99997% tracklet efficiency and 99.26% derived object efficiency were achieved in the 2-year full S3M simulation.The simulation used the full S3M, 100,424 fields, repeated visits, and simulated poor weather.
- 2-year S3M simulation: Better than 99% efficiency converted three or more tracklets into differentially corrected orbits for most populations within 14 days.NEOs, Jupiter-family comets, and long-period comets were exceptions to this general result.
- Pan-STARRS1-like simulation: MOPS achieved essentially 100% tracklet efficiency in the Pan-STARRS1-like simulation, losing 10 of 105,439 tracklets.The simulation used four years of simulated opposition and sweetspot observations and the NEO subset of the S3M.
- Pan-STARRS1-like simulation: 81.7% orbit-determination performance in the Pan-STARRS1-like simulation was attributed to larger astrometric uncertainty.The simulation used a 0.1′′ baseline astrometric uncertainty and nearly four years of observations.
- Grid simulations: Grid simulations showed consistently high derived-object efficiency across broad orbital elements, including semi-major axes beyond 100 AU and retrograde orbits.Efficiency degraded for e ∼1, and polar regions exposed conservative acceptance limits for some fast-moving objects.
5. MOPS Validation and Real Data
MOPS was adapted to real Pan-STARRS1 data, where reduced fill factor, systematic false detections, and survey cadence differ from the idealized Pan-STARRS4 design assumptions. The adaptation added operational tools and procedures while exposing important efficiency limits.
- Pan-STARRS1 adaptation: MOPS was adapted to maximize asteroid detection and NEO discovery from the Pan-STARRS1 data stream.The adaptation included tools for rejecting false detections and tracklets and enabling human review before MPC submission.
- Pan-STARRS1 limitations: ∼75% effective camera fill factor limited maximum end-to-end derived-object efficiency to 26–75% for objects in the field of view.The bound applied even to objects that would otherwise be imaged at high efficiency.
- Pan-STARRS1 adaptation: Four-exposure quads reduced false-tracklet rates as the number of detections increased, but false tracklets remained even with four detections.The exposures were typically separated by about 15 minutes, producing tracklets with roughly 45-minute arcs.
- False detections: Systematic artifacts dominated Pan-STARRS1 false detections, reaching ∼8200/deg2 at ≥5σ and forming spatially clumped false tracklets.The rate near the Galactic Plane was 10–50× higher than elsewhere, creating substantial downstream contamination.
- Data quality: Pan-STARRS1 astrometry for slow-moving objects remained strong, with average RMS uncertainty of about 0.13′′ despite the false-detection problems.The paper describes this performance as excellent by contemporary standards.
5.3. Tracklets
MOPS forms intra-night tracklets from transient detections despite substantial real-data contamination and complex observing sequences. Synthetic tests show near-complete recovery, while real Pan-STARRS1 data reveal false-tracklet and duplicate-tracklet burdens.
- Robustness: MOPS could still maintain high tracklet-creation efficiency without detection orientations and lengths, but at the cost of a much higher false-tracklet rate.The system also maintained integrity at more than 10× higher random false-detection rates in earlier tests.
- Tracklet contamination: Systematic false detections can align into asteroid-like tracklets, substantially slowing derived-object processing through false-track proliferation.Pan-STARRS1 false detections were reported as 10–50× higher than expected at ≥5σ.
- Tracklet creation: 99.98% tracklet creation efficiency was measured after injecting synthetic detections into real Pan-STARRS1 data.The algorithm correctly identified almost every possible detection set despite real detections and noise.
- Tracklet classification: About 10–15% of generated tracklets were MIXED or BAD, although these represented a small fraction of the CLEAN rate.Non-synthetic detections were treated as noise when evaluating recovery of synthetic solar-system objects.
- Deep-drilling sequences: ∼25% duplicate-tracklet ratio occurred in Pan-STARRS1 eight-exposure Medium Deep sequences, with essentially zero lost tracklets.The collapseTracklets procedure had not been optimized for these deep-drilling cadences.
5.4. Tracks
MOPS links tracklets across nights into candidate tracks using spatial searches and quadratic sky-plane motion fits, then filters them through orbit determination. Pan-STARRS1 linking efficiency was about 80% because of contamination, duplicate tracklets, and untuned operational parameters.
- Track linking: MOPS searches 7–14 days of prior observations with a kd-tree and fits candidate linkages to quadratic sky-plane motion.Candidate combinations are retained when their fitted motion satisfies configured error limits.
- Track linking: ∼80% tracklet-linking efficiency was obtained for most solar-system object classes, from the inner solar system to beyond Neptune.The result reflects Pan-STARRS1 processing rather than the higher efficiencies found in some simulations.
- Performance limits: Linking efficiency is reduced by false-track contamination, duplicate tracklets from Medium Deep sequences, and parameters not tuned to Pan-STARRS1.The paper reports that suitable survey strategy, false-detection rates, and configuration parameters can raise derived-object efficiency to nearly 100%.
- Candidate filtering: Although fewer than 1% of candidate tracks are real, subsequent motion, residual, and orbit cuts eliminate essentially all bad tracks.The reported track-linking accuracy is similar to the ≪1% accuracy reported by Kubica et al. (2007).
- Orbit determination: Orbit determination calculates a six-parameter orbit, accepts tracks meeting an RMS-residual requirement, and returns rejected tracklets for other linkages.The process begins with initial orbit determination and continues with least-squares differential correction.
- Discordance handling: Discordance identification selects the linkage with significantly smaller RMS residuals or rejects all competing linkages when no correct choice is evident.This handles cases where tracklets are assigned to multiple distinct linkages or share detections.
5.6. Initial Orbit Determination (IOD)
MOPS uses initial orbit determination to convert linked detections into six-parameter orbits, with OrbFit selected and evaluated for efficiency, accuracy, speed, and support. The IOD stage is essentially 100% efficient for correctly linked observations, while downstream Pan-STARRS1-derived-object efficiency is typically 70–90%.
- IOD produces six-parameter orbits from sets of detections, and MOPS implements the OrbFit package for this task.
- OrbFit was selected based on orbit-production efficiency, orbit accuracy, computational speed, and availability and support.
- Millions of synthetic tracks were used to measure orbit-determination performance, with efficiency and accuracy treated as explicit objectives.
- Essentially 100% IOD efficiency was achieved for all classes of solar-system objects when detections were correctly linked.
- 70–90% derived-object efficiency was realized for most solar-system-object classes in the Pan-STARRS1 survey.
- MOPS remains agnostic to prior object knowledge, creating tracklets, tracks, and derived objects without using information about known objects.
5.10. Precovery & Attribution (‘PANDA’) of derived objects
PANDA extends MOPS derived objects through attribution and precovery by matching predicted positions and velocities to tracklets and refining orbital solutions. Its efficiency is about 93% for attribution, but contamination, sparse cadence, and long time gaps constrain reliability.
- PANDA predicts derived-object positions and velocities, searches nearby tracklets for matches, and accepts associations after differential-orbit residual checks.
- Attribution efficiency is ∼93%, although the cumulative statistic is dominated by main-belt asteroids and does not account for ephemeris accuracy.
- Attribution generally becomes easier with increasing semi-major axis because distant objects move slowly and have lower sky-plane density.
- Contaminating derived objects with unrelated detections degrades orbital quality and makes future attributions less likely to be real.
- The contamination-mitigation techniques were never implemented because the attribution algorithm assumed at least three tracklets for most asteroids per lunation.
- Orbit identification efficiency was close to 100% for simulated Pan-STARRS4 data but only ∼26% for Pan-STARRS1.
5.12. Data Rates & MOPS Timing
MOPS timing is stable for per-night processing but scales with the growing derived-object catalog. Precovery is the principal computational concern, while Pan-STARRS1 detection efficiency is constrained largely by camera fill factor and operational filtering.
- Per-night generation and tracklet stages remain generally constant, while derived-object stages grow linearly or worse with simulation time.
- Naive precovery has quadratic growth, but search-window optimizations reduce the scaling below quadratic to approximately O(n · log n).
- The maximum tracklet-detection efficiency across all four filters is ∼78%, driven largely by the effective Pan-STARRS1 camera fill factor of ∼80%.
- Objects imaged on live, unmasked camera pixels are detected with ∼97% efficiency.
- Pan-STARRS1+MOPS performance was roughly constant during the first year, apart from a dip caused by overly aggressive false-detection filtering.
- For bright main-belt asteroids with wP1 < 19, Pan-STARRS1 reported about one-third of the predicted number of tracklets.
6. Current Pan-STARRS1 Performance
Pan-STARRS1 performance improved through better image processing, detector-noise modeling, detection-morphology characterization, and survey scheduling. The system became a capable comet finder while increasing sensitivity and reducing false detections entering MOPS.
- Improved measurement and modeling of detector readout noise increased detection sensitivity in image processing.
- Finer detection-morphology characterization allows MOPS to reject many false-detection classes before they contaminate NEO-tracklet review.
- The Modified Design Reference Mission increased the fraction of 3π survey time spent observing in quads.
- Completion of 3π static-sky processing was expected to improve system limiting magnitude by at least 0.4 mag.
- The Pan-STARRS1 NEO discovery rate had been increasing since late 2010, despite monthly variation from weather losses.
- Pan-STARRS1 became a capable comet finder, discovering 30 comets to date and 8 during October.
7. Availability & Ongoing Development
MOPS is distributed with substantial third-party dependencies and has accumulated extensive development effort. It has also been used across multiple survey and research contexts, though some components are not freely available or fully integrated.
- MOPS is available under the GNU General Public License, with documentation and development notes accessible through the PS1 Science Consortium website.
- The package is large and unwieldy, uses multiple programming languages, and requires many Perl and Python module dependencies.
- Some MOPS subcomponents are unavailable in the distribution, including the JPL Solar System Dynamics package and other third-party software.
- Approximately 15–20 full-time-equivalent years have been invested in developing MOPS and the S3M.
- MOPS has processed Spacewatch and TALCS data and supported simulations for Pan-STARRS, LSST, and ATLAS.
- MOPS has also served as a research tool for graduate students and postdocs.
8. Future improvements
Future MOPS development targets greater simulation fidelity, faster and more flexible processing, easier deployment beyond Pan-STARRS, and improved handling of photometric and database complexities.
- Future work includes improving execution speed for Pan-STARRS4-like data volumes through reduced database I/O and greater in-memory processing.
- MOPS plans improved installation and configuration for surveys other than Pan-STARRS.
- The current linear processing model can be inadequate for evaluating production parameters and makes merging separately configured databases difficult.
- Future versions should support more agile processing, including nonlinear operations and dataset merging.
- Future simulations should model varying sky sensitivity, bright objects, detector fill-factor losses, and photometric artifacts more faithfully.
- Simplified two-body ephemerides are proposed for large-scale simulations instead of a perturbed dynamical model.
9. Summary
MOPS has been deployed effectively on Pan-STARRS1 while supporting high-efficiency moving-object detection and orbit computation under next-generation survey conditions. Its measured efficiencies also provide a basis for population studies, although real-telescope performance remains challenging.
- Pan-STARRS1 and Pan-STARRS4 differ in targeted capability, but MOPS has been applied effectively to the current Pan-STARRS1 transient stream.
- MOPS has been deployed effectively on Pan-STARRS1 for searching for near-Earth objects and comets and characterizing the main belt.
- MOPS estimates object-detection efficiency, providing a foundation for large-scale population studies.
- >99% efficiency is the design goal for detecting objects and computing their orbits across most solar-system populations with next-generation survey data quality.
- The desired performance is challenging to achieve with real telescope data.
- Further optimization and tuning of the Pan-STARRS1 pipeline is expected before applying MOPS to future data and combined telescope operations.