New Automated Pipeline Digs Years Into Archival Sky Surveys to Find Asteroids That Were Hiding in Plain Sight
Featured image: Archival Zwicky Transient Facility image showing asteroid trails; credit: ZTF/Caltech
A team of researchers from Georgia Tech and Lawrence Livermore National Laboratory has developed an automated probabilistic pipeline that can find near-Earth asteroids hiding in years-old archival survey images, extending observational arcs by years and dramatically improving impact risk assessments.
Published in the August 2026 issue of The Astronomical Journal (DOI: 10.3847/1538-3881/ae7c73), the method developed by Sage Li, Alex Geringer-Sameth, and Nathan Golovich addresses a fundamental problem in asteroid tracking: when a new asteroid is discovered, historical images of the same patch of sky may already contain it, but no one was looking for it at the time. Finding these archival appearances (a process called prediscovery or “precovery”) extends the known observational arc, often by years, which is the single most important factor in determining whether an asteroid poses a threat to Earth.
How It Works
The pipeline operates in four stages. First, it refits the orbit of a newly discovered near-Earth asteroid using data from the Minor Planet Center, computing a full six-parameter covariance (position and velocity in three dimensions). It then propagates this uncertainty backward and forward in time to the epochs of archival survey images.
Instead of calculating a single ephemeris point, the method constructs probabilistic sky maps using Monte Carlo orbit samples. For asteroids with huge uncertainty regions (sometimes hundreds of square degrees), this captures the full banana-shaped probability distribution of where the object could be in old images.
The pipeline then builds source catalogs from archival Zwicky Transient Facility (ZTF) images at low signal-to-noise thresholds, detecting very faint objects that standard processing pipelines would miss. Finally, a probabilistic linking algorithm uses a likelihood ratio to connect detections across multiple images, statistically ruling out false positives without hard cuts.
The method is survey-agnostic: it works with any archival image or source catalog.
What It Found
Applied to approximately 3,000 recently discovered near-Earth asteroids in ZTF data, the pipeline found that roughly 500 objects had their observational arcs doubled. For the potentially hazardous asteroid 2021 DG1, the arc was extended by 2.5 years, reducing its sky-plane uncertainty from many degrees to arcseconds for future apparitions.
The most dramatic case was 2025 FU24, a recently discovered near-Earth asteroid that the pipeline recovered in archival images taken nearly seven years before its first known observation. The search region for this object covered hundreds of square degrees across thousands of ZTF images, a scale that would be infeasible with manual techniques.
“Reducing orbital uncertainty immediately after discovery” is the stated goal of the pipeline, which is designed to produce better orbits in days rather than months.
Why It Matters for Planetary Defense
The timing is significant. The paper’s authors explicitly cite the 2024 YR4 near-miss event, which briefly had a small chance of Earth impact in 2032, as a motivating factor. While the team was unsuccessful in finding 2024 YR4 in archival data, the methodology applies directly to future such cases.
Every year of archival data that can be recovered for a potentially hazardous asteroid translates directly into improved impact probability calculations. A two-year arc extension for 500 objects represents a meaningful improvement in the global asteroid risk assessment.
The approach will become even more powerful as new survey telescopes come online. The Vera C. Rubin Observatory, set to begin full operations in the coming years, will generate petabytes of imaging data and discover thousands of new near-Earth objects annually. This pipeline provides the computational framework to handle that flood of discoveries, finding prediscovery detections in Rubin’s own archival data as well as in existing ZTF and Pan-STARRS archives.
“The pipeline is designed to handle the step function in NEO discovery rates that Rubin will deliver,” the authors note.