Skip to main content

Sky survey classifies more than 10,000 supernovae

Northwestern scientists help the Zwicky Transcient Facility's record-breaking effort
December 4, 2024 | By Amanda Morris
zwicky supernovae
Over the past five years, Zwicky Transient Facility's Bright Transient Survey has become the world’s primary discovery machine for astronomical transients. Image by Caltech Optical Observatories

With contributions from Northwestern University astrophysicists, the Zwicky Transient Facility (ZTF) astronomical survey officially has classified more than 10,000 new supernovae. This major milestone marks a new record for discovering and confirming supernovae, the brilliant flashes of light emitted from dying stars.

By compiling a growing dataset of supernovae, astronomers are better equipped to answer questions about stellar evolution and death, how dark energy drives the expansion of the universe and how, exactly, stars die. 

“We need large samples of supernovae to understand their relative rates,” said Northwestern’s Adam Miller, a ZTF board member and co-principal investigator on the project. “For example, suppose there is an unusual type of white dwarf explosion that only manifests itself in 1 out of every 100 such explosions. On average we would need at least 100 discoveries just to find one of these, and we’d need more than 1,000 or so discoveries to make a robust statistical measurement of their prevalence. Finding lots of supernovae allows us to determine these rates.”

Miller is an assistant professor of physics and astronomy at Northwestern’s Weinberg College of Arts and Sciences. Miller also is a member of the Center for Interdisciplinary Exploration and Research in Astrophysics (CIERA) and of the NSF-Simons AI Institute for the Sky (SkAI, pronounced “sky”).

Northwestern AI bot adds automation

Located at Caltech’s Palomar Observatory just east of San Diego, ZTF uses a wide-field camera to scan the entire visible sky every other night. Over the past five years, ZTF’s Bright Transient Survey has become the world’s primary discovery machine for astronomical transients — fleeting phenomena that flare up suddenly and then quickly fade.

Launched in 2017, ZTF received a boost last year when Northwestern scientists led the development of Bright Transient Survey Bot (BTSbot), a new tool that automates the entire search for supernovae across the night sky. BTSbot discovers, reports and requests additional observations of new supernovae — effectively removing humans from the process.

supernova galaxy
Artificial Intelligence

First supernova detected, confirmed, classified and shared by AI

New artificial intelligence tool removes humans from entire search, discovery process

Not only does the new tool rapidly accelerate the process of analyzing and classifying new supernova candidates, it also bypasses human error.

“Since BTSbot began operation, it has found about half of the brightest ZTF supernovae before a human found them,” said Nabeel Rehemtulla, a Northwestern graduate student in astronomy who co-led the technology development. “For specific types of supernovae, we have automated the entire process, and BTSbot has so far performed excellently in more than a hundred cases. This is the future of supernova surveys especially when the Vera Rubin Observatory begins operations.”

High ‘purity of predictions’

To develop BTSbot, Rehemtulla trained an AI algorithm with more than 1.4 million historical images from nearly 16,000 sources, including confirmed supernovae, temporarily flaring stars, periodically visible stars and flaring galaxies. In October 2023, the new tool successfully detected, confirmed, classified and reported its first supernova.

“After we deployed it, I meticulously monitored every one of BTSbot’s actions for about eight months,” said Rehemtulla, who also is a member of CIERA and SkAI. “It runs on new data during the night while observations are occurring, so I check on it every morning. I kept detailed daily logs and calculated that approximately 96% of targets which BTSbot sends to our second telescope for follow-up observations are genuine supernovae. In other words, the purity of predictions is very high.”

By topping 10,000 confirmed supernovae, ZTF has more data to feed into BTSbot. So far, BTSbot has trained on more than 5,200 supernovae. And its performance will continue to improve as its trained on thousands more.

“We need lots of data to train machine-learning models in the first place,” Miller said. “If we only had 100 or even 500 supernovae, that wouldn’t be enough to train a model that works as well as BTSbot.”

Preparing for the future

When the Vera Rubin Observatory opens in 2025, it will be much more sensitive than ZTF. Astronomers predict the new observatory, located in the Chilean mountains, will detect millions of new supernovae. And AI will help with that search.

“BTSbot in its current form will not be deployed at the Rubin Observatory,” Rehemtulla said. “But we will deploy some evolved version of BTSbot, which will be based on similar principles.”

“The machine learning and AI tools we have developed for ZTF will become essential when the Vera Rubin Observatory begins operations,” said Daniel Perley, an astronomer at Liverpool John Moores University in the U.K. who developed the search and discovery procedures for the BTS. “We have already planned to work closely with Rubin to transfer machine-learning knowledge and technology.”

Funded by the National Science Foundation, ZTF is an international partnership among Northwestern; Caltech; University of Stockholm in Sweden; Ruhr University Bochum in Germany; University of Warwick in the U.K.; Drexel University; Cornell University; University of Maryland, College Park; University of Wisconsin, Milwaukee; and University of California, Berkeley.

Editor’s Picks

Deinococcus radiodurans

Related Stories

exoplanet