What are the odds?: Automated system calculates the likelihood that asteroids will hit Earth

by Abbey Nastan
Tuesday, January 12, 2016

By some estimates, Earth is pelted each year by tens of thousands of asteroids — from souvenir-sized nuggets 10 grams and up to the occasional meters-wide boulder. Spotting larger Earth-bound impactors that could cause serious destruction, while they’re still in space, has been a priority of the planetary science community for years. But many space objects are small enough that they can only be detected days or hours before impact — if at all. In a new study, scientists have tested and automated a new technique called systematic ranging that’s intended to rapidly calculate the probability — from only a handful of observations — that newly discovered asteroids will hit Earth.

The traditional method of calculating an asteroid’s orbit, known as the standard differential correction procedure, relies on many repeated observations of an asteroid to calculate its speed and direction relative to Earth. This works well for large objects that are far away, but small asteroids are naturally harder to spot, so astronomers can’t usually detect them before they approach Earth. In fact, only two asteroids — both about 3 to 4 meters in diameter — have been observed in space prior to impacting Earth: one in 2008, which was observed several times prior to impact in Sudan, and one in 2014 that was observed briefly but not identified as a possible impactor until it fell into the Atlantic.

In recent years, new methods to estimate asteroid orbits have made use of Bayesian statistics, which offers advantages over traditional “frequentist” statistics. Using Bayesian statistics, for example, allows researchers to account for the orbits of previously known asteroids in determining the behavior of newly discovered bodies, and offers information on the full range of possible behaviors. In 2001, researchers introduced a method called statistical ranging — not to be confused with systematic ranging — which uses the Bayesian approach to assess the likelihood that randomly chosen individual orbits may match that of a newly discovered asteroid.

Systematic ranging, first proposed in 2005, differs from statistical ranging in that it tests a “grid” of possible orbits, yielding more information while requiring less computation time. Systematic ranging is designed for circumstances when “we don’t actually have enough information to say where [an asteroid] is” with certainty, says Steve Chesley, senior research scientist at NASA’s Jet Propulsion Laboratory (JPL) in Pasadena, Calif., and an author of the study, published in the journal Icarus.

In the study, researchers led by Davide Farnocchia, a scientist at JPL, tested the performance of the systematic ranging method and developed an automated system using the method to compute the likelihood that an asteroid would hit Earth as soon as the asteroid was detected. Farnocchia and his colleagues assessed the method’s performance under three “priors” — sets of previously known information that could influence the outcome of the calculations. One was a generic prior used in Bayesian statistics called “Jeffreys prior”; another was based on a realistic model of the known distribution of asteroid orbits in the solar system; and the third was a “uniform” prior, which assumed that we know nothing about an asteroid’s possible orbit. The researchers used these priors to retroactively compute the impact likelihood for the 2008 and 2014 asteroids based on the first few observations that had been made of each. They did the same, again based on initial observations, for asteroids originally thought to be very close to Earth. (Following early observations, the orbits of these were subsequently determined with greater accuracy and found to be located relatively far away in the asteroid belt.)

The results showed that Jeffreys prior gave too many false positives of impact, while not giving the true orbits of known impactors a likely probability. The prior based on the asteroid population model produced the most accurate results, but it also took the most computational time due to its complexity. The uniform prior performed almost as well as the asteroid population model, while requiring substantially less computation time. Additionally, Farnocchia says, whereas the model-based prior may bias predictions by assuming newly discovered asteroids will behave like most previously known bodies, the uniform prior limits this bias. Therefore, he and his colleagues concluded that the uniform prior was the most effective of those they tested for use in systematic ranging.

The systematic ranging method could be a powerful tool to predict smaller asteroid impacts with little warning, but only if the computations are done in time. Therefore, Farnocchia and his colleagues designed an automated system that performs systematic ranging calculations on new discoveries posted to the Near Earth Objects (NEO) confirmation page of the International Astronomical Union’s Minor Planets Center. If the system calculates an impact probability above 1 percent, it notifies the researchers, who can then contact observatories to request additional data that can be used to determine whether an impact is indeed going to occur. Currently, the system gives notifications a few times a month.

The new work is “exciting,” says Dagmara Oszkiewicz, a postdoctoral researcher at Lowell Observatory and Northern Arizona University, who has worked on other methods for detecting asteroids on short notice. “The big advantage,” she says, is that the method “has been implemented as an automated system to continuously analyze objects reported on the Minor Planet Center NEO Confirmation page.”

Peter Jenniskens, a senior research scientist at the SETI Institute and NASA Ames Research Center, who studied the 2008 asteroid impact and coordinated the effort to recover many of the resulting meteorites in Sudan, agrees: “Previously, much of this was done by hand and results would come out only when it was likely that the impact would actually happen.” With the advance warning that the systematic ranging system could provide, it would be possible to capture approaching asteroids on radar, Farnocchia’s team noted in their study, which would greatly aid in efforts to both characterize the asteroid and find meteorites after impact. It might also be possible to observe an asteroid impact from an airplane, Jenniskens says, something he says he’d like to do.

Meanwhile, Farnocchia and his colleagues are hoping to refine their systematic ranging system while it’s in action, and in the near future, they would like to re-evaluate their choice of prior. “We’ve found a solution that works fairly well, but it is probably not the ideal one,” Farnocchia says. Chesley agrees: After comparing actual results to the program’s predictions for a while, “we’ll know our batting average.”