Astronomers from the California Institute of Technology (Caltech) have completely automated the classification of 1,000 supernovae using a machine-learning (ML) algorithm. The Zwicky Transient Facility, or ZTF, a sky survey instrument located at Caltech’s Palomar Observatory, collected data that the algorithm was then used to analyse.
“We needed a helping hand, and we knew that once we trained our computers to do the job, they would take a big load off our backs,” says Christoffer Fremling, a staff astronomer at Caltech and the mastermind behind the new algorithm tagged as SNIascore.
A year and a half after SNIascore classified its first supernova in April 2021, they are approaching the pleasant milestone of 1,000 supernovae. Every night, ZTF scans the night sky for alterations known as transient events. This covers everything, from asteroids in motion to recently devoured stars by black holes to exploding stars known as supernovae.
ZTF notifies astronomers worldwide of these transient events by sending out hundreds of thousands of alerts each night. Other telescopes are then used by astronomers to monitor and learn more about the nature of the shifting objects. Thousands of supernovae have so far been found thanks to ZTF data.
Members of the ZTF team cannot organise all the data on their own due to the constant flow of data that comes in every night. According to Matthew Graham, project scientist for ZTF and research professor of astronomy at Caltech, “the traditional notion of an astronomer sitting at the observatory and sieving through telescope images carries a lot of romanticism but is drifting away from reality.”
Instead, to help with the searches, the team has created ML algorithms. SNIascore was created to categorise potential supernovae. There are two main categories of supernovae: Type I and Type II. In contrast to Type II supernovae, Type I supernovae are devoid of hydrogen.
When material from a companion star flows onto a white dwarf star, causing a thermonuclear explosion, a Type I supernova is produced. When a massive star collapses due to its own gravity, a Type II supernova happens. Type Ia supernovae, or the “standard candles” in the sky, can be classified by SNIascore. These are dying stars that explode with a steady-state thermonuclear blast.
Astronomers can gauge the universe’s expansion rate thanks to Type Ia supernovae. Fremling and colleagues are currently expanding the algorithm’s capabilities to classify additional types of supernovae soon.
Every night, after ZTF has recorded sky flashes that may be supernovae, it sends the data to the SEDM spectrograph at Palomar, which is in a dome a short distance away (Spectral Energy Distribution Machine).
To determine which supernovae are likely Type Ias, SNIascore collaborates with SEDM. As a result, the ZTF team is working quickly to compile a more trustworthy data set of supernovae that will allow astronomers to conduct additional research and, ultimately, learn more about the physics of the potent stellar explosions.
“SNIascore is incredibly precise. We have observed the performance of the algorithm in the real world after 1,000 supernovae” says Fremling. Since the initial launch in April 2021, they have found no clearly misclassified events, and they are now planning to implement the same algorithm with other observing facilities.
According to Ashish Mahabal, who oversees ZTF’s machine learning initiatives and is the centre’s lead computational and data scientist at Caltech, their work demonstrates how ML applications are maturing in near real-time astronomy.
The SNIascore was created as part of the ZTF’s Bright Transient Survey (BTS), which is currently the most comprehensive supernova survey available to the astronomical community. The entire BTS dataset contains nearly 7000 supernovae, 90 per cent of which were discovered and classified by ZTF while the remaining 10 per cent were contributed by other groups and facilities.