Pattern Recognition in Transient Astronomical Light Curves with the Assistance of Artificial Intelligence

Rafael Santos Oliveira & Maurício Pinheiro

The Large Synoptic Survey Telescope (LSST), scheduled to begin operations in 2023, represents a significant advancement in the observation of transient objects in the sky, i.e., astronomical objects or events that last for a short period of time on astronomical timescales, such as: supernovas, which are bright and ephemeral stellar explosions; neutron star mergers, an event resulting in the generation of gravitational waves; and meteors, which are objects that cross the Earth’s sky. Therefore, it is of great importance to detect and catalog these astronomical events and objects.

The Vera C. Rubin Observatory, formerly known as the Large Synoptic Survey Telescope (LSST), is an astronomical observatory located in the Province of Elqui, Coquimbo Region, Chile. Situated at an altitude of 2,663 meters, the observatory houses the Simonyi Survey Telescope, a wide-field reflecting telescope with an 8.4-meter primary mirror. This telescope operates in the wavelength range of 320–1060 nm and has a collecting area of 35 square meters. It features a three-mirror anastigmatic design and is mounted on an altazimuth support. The LSST will conduct a ten-year imaging survey covering 18,000 square degrees of the sky, capturing images with a 3.2-gigapixel CCD camera, which is the largest digital camera ever constructed. (Photo courtesy of LSST Corp./NOAO).

The distinctive feature of LSST lies in its ability to generate massive volumes of astronomical data. Each night of observation, it will produce approximately 15 terabytes of information. This substantial volume of astronomical data is a result of its capability to capture the sky in a panoramic and repetitive manner, mapping the universe in an unprecedented way. However, this abundance of astronomical data brings a significant challenge: how to efficiently analyze and classify these transient objects before they fade away?

In 2018, recognizing this challenge, LSST launched a competition on the Kaggle platform, offering a reward of $25,000. The goal was to develop a neural network capable of automatically classifying the type of observed object based on its light curves. The winner was Kyle Boone, and his solution is available here.

In the challenge presented by the Large Synoptic Survey Telescope (LSST), a substantial dataset was provided, totaling 40 gigabytes in CSV format tables. These data tables serve as the foundation for training the AI, which must be able to classify the type of astronomical object observed based on their light curves. Each object in the dataset has an ID and a classification, totaling 14 objects/events, including Single micro-lens, TDE, Eclipsing Binary, SNII, SNIax, Mira, SNIbc, Kilonova, M-dwarf, SNIa-91bg, AGN, SNIa, RR lyrae, and SLSN-I.

Light curve of asteroid 201 Penelope based on images taken on October 6, 2006, at the University of Mount John Observatory. The bottom scale is in days, and one complete rotation occurs from approximately -0.14 to 0.02 (or about 3.7 hours). The light curve has two peaks and two troughs because asteroid 201 Penelope has (in first order) an ellipsoidal shape. It rotates around its short axis, so the troughs occur when we see it head-on, while the peaks occur when we see it from the side. Created using Matlab. Created on October 19, 2006, by Evil Monkey. CC BY-SA 2.5 Source: Wikimedia Commons.

When observing a star, it’s possible to notice variations in its brightness over a certain period of time, which can be represented through a graph called the light curve of a celestial object. Each type of object, such as supernovas, neutron stars, asteroids, etc., has a characteristic light curve. Therefore, the AI is trained to recognize these characteristics and, based on the observed light curves, automatically classify the object into one of the predefined categories.

Automated classification is of vital importance, considering the significant volume of data that LSST is about to generate. It would be humanly impossible to manually analyze and classify all these light curves in a timely manner. Each curve would require meticulous analysis by an expert, and thus, the development of an artificial intelligence capable of performing this task becomes a fundamental necessity for the use of this telescope.

Such an initiative exemplifies the increasingly important role that artificial intelligence plays in space exploration and the analysis of large astronomical datasets. Through challenges like this, humanity advances toward a deeper understanding of the cosmos, and LSST is one of the most exciting pieces of this cosmic puzzle. As we prepare for the operational era of LSST in 2023, we can expect a deluge of astronomical discoveries aided by artificial intelligence.

#Astronomy #AutomatedClassification #UnderstandingTheCosmos #LightCurves #SpaceExploration #AI #ArtificialIntelligence #Kaggle #LargeSynopticSurveyTelescope #LSST #TransientObjects #AstronomicalDataVolume

Glossary:

Kaggle is a platform dedicated to data science competitions. On this platform, both individuals and companies can freely provide datasets for others to download and utilize for study and research purposes. Some of these datasets are considered classics for beginners, such as the “Iris Species” dataset. In this example, participants are challenged to correctly classify the species of a flower based on a set of provided features in the data. Kaggle also provides a platform for users to share their solutions and approaches to the challenges, promoting collaborative learning. In addition to collaborative spaces, companies can offer financial rewards through competitions for participants who successfully solve their proposed challenges. This platform has become a hub for learning and innovation in the field of data science and machine learning, attracting a diverse community of data enthusiasts and professionals.