These space rocks can raze entire cities to the ground – why didn't NASA notice them?
A team of researchers from Leiden University in the Netherlands has developed a neural network called Hazardous Object ID, which they say can predict if an asteroid is on its way to impact Earth.
The new AI has identified 11 asteroids previously unclassified by NASA as dangerous, larger than 100 meters in diameter – large enough to explode with the force of hundreds of nuclear warheads when they hit Earth.
They also focused on space rocks that can fly within 4.7 million kilometers of Earth, as detailed in an article published in Astronomy & Astrophysics earlier this month.
Using a supercomputer, the researchers were able to simulate 10,000 years of orbital motions of the planets in the solar system. The team then tweaked the simulation by simulating future asteroids impacting the Earth, throwing them off Earth, and tracking their exact location and orbits.
“If you rewind time, you'll see collisions with known asteroids again,” said co-author and astronomer Simon Portegis Zwart of Leiden University. “So you can create a library of the orbits of asteroids that have hit Earth.”
This simulation served as a testing ground for the neural network, which then looked for patterns in the data common to simulated asteroids that could eventually collide with Earth.
To find out if their AI was really good at detecting asteroids, the team tested it on known data from 2,000 asteroids listed by NASA. The neural network determined with an accuracy of 90.99% which of them were dangerous and which were not.
The team is now working to make its neural network even more accurate.
“We now know that our method works, but we certainly would like to delve deeper into research with a neural network and a large amount of input data,” said Zwart. “The tricky part is that small glitches in orbit calculations can lead to major changes in conclusions.”
