Identification and photometric classification of extragalactic transients in the Vera C. Rubin Observatory's Data Preview 1

The Big Picture

Imagine trying to spot a single flickering candle from across a crowded city while thousands of other lights turn on and off every second. That’s roughly what astronomers face when hunting for supernovae. The night sky is noisy. Finding a star detonating in a galaxy millions of light-years away takes not just a powerful telescope, but a pipeline that can sort signal from noise at massive scale.

The Vera C. Rubin Observatory, perched in the Chilean Andes, is about to become the most powerful survey telescope ever built for tracking changes in the night sky. Its full survey, the Legacy Survey of Space and Time (LSST), will generate up to 10 million alerts per night, each flagging a difference compared to a stored reference image. That’s more than ten times the output of the current frontrunner, the Zwicky Transient Facility.

But before that firehose opens, astronomers needed to test the plumbing. A team led by James Freeburn at UNC Chapel Hill, including IAIFI affiliate V. Ashley Villar at Harvard, did exactly that. Using an early test dataset called Data Preview 1 (DP1), collected with a prototype camera before the full telescope goes live, they showed that Rubin’s detection and classification pipeline can handle the flood.

Searching just a fraction of Rubin’s commissioning data, the team identified eight new likely supernovae from nearly 370,000 candidate events, confirming that the telescope’s detection pipeline and machine learning classifier work ahead of the full survey launch.

How It Works

The commissioning campaign used LSSTComCam, a smaller prototype camera, to observe seven sky fields between November and December 2024. Each field covers roughly 0.4 square degrees. The team focused on three of these fields for their supernova search.

First comes difference image analysis (DIA): subtract a stored “template” image from a new observation, and anything that has changed shows up as a residual. This produced 369,644 candidate sources. The vast majority were not supernovae. Cosmic rays, instrument artifacts, poorly subtracted galaxy cores, and other false positives (collectively called “bogus” detections) dominated the catalog.

To cut through the noise, the team required each candidate to have at least five detections, show an association with a galaxy beyond the Milky Way, and pass quality checks on its light curve (the graph of how brightness changes over time). Cross-matching against existing astronomical databases recovered three previously catalogued supernovae. Eight additional candidates passed all cuts as new, unreported events.

Classification relied on Superphot+, a machine learning tool trained on simulated supernova light curves that assigns probabilistic sub-type labels. The two main categories are Type Ia (thermonuclear explosions of white dwarf stars, used as cosmological distance markers) and Type II (core-collapse explosions of massive stars).

Superphot+ needs multi-band photometry and enough data points to pin down the light curve shape. Of the eleven supernovae in the sample, only two got confident classifications above 95% probability: one Type Ia and one Type II. The rest didn’t have enough observations, a gap that LSST’s continuous cadence should close.

The team also ran simsurvey simulations to predict how many supernovae the campaign should have detected given its depth, cadence, and sky coverage. The answer: 15 ± 4. Finding eleven falls within that range. The shortfall probably comes from a known DP1 quirk: template images were sometimes taken during the same campaign, causing some supernovae to be subtracted away before analysis even began.

Why It Matters

The stakes extend well beyond eleven faint smudges of light. Type Ia supernovae are the primary “standard candles” astronomers use to measure cosmic distances and map the expansion history of the universe. LSST is expected to detect hundreds of thousands over its ten-year run, enabling precision measurements of dark energy that will either confirm the standard cosmological model or crack it open. None of that science arrives if the pipeline can’t work reliably at scale and in near-real time.

Rubin’s difference imaging produces clean, scientifically useful subtractions. Classifiers like Superphot+ can meaningfully categorize events even with sparse photometric coverage. The bottlenecks are now visible too: template quality, observation cadence, spectroscopic follow-up capacity. All fixable before the full survey begins.

The next phase brings LSSTCam, the 3.2-gigapixel instrument, covering 9.6 square degrees per exposure and generating data at many times the commissioning rate. The pipeline tested here will face that torrent head-on. This paper is the fire drill.

Eight new supernovae found in a sliver of commissioning data, results matching simulation predictions, and a classification pipeline that works under real-world constraints. Rubin’s system has passed its first real test, and the full LSST survey is cleared for launch.