In early August, the moment that Bill Borucki had been dreading finally arrived. As the principal investigator of NASA’s Kepler space telescope, Borucki had been working with his colleagues to restore the spacecraft’s ability to precisely point itself. The planet-hunting telescope has four reaction wheels—essentially, electrically driven flywheels—and at least three must be functional to maintain positioning. But in the past year, two of those wheels had been on the fritz. One went off line in July 2012 after showing elevated levels of friction, and a second followed suit in May 2013, effectively ending science operations. After a few months of recovery efforts, the telescope team was finally forced to call it quits, six months after the mission was originally scheduled to finish but years before they hoped it would.
The failures mark the end of an era for Kepler. With only two reaction wheels, the telescope can’t steady itself well enough to ensure that light from each star hits the same fraction of a pixel on its charge-coupled devices for months on end without deviation. That’s what Kepler needs in order to detect, with high precision, the transit of a planet: the slight dip in the brightness of a star that occurs when an orbiting planet crosses in front of it.
But the Kepler spacecraft might still have its uses, and the data it has already gathered almost certainly will. The telescope’s managers are currently evaluating proposals for what might be done with a two-wheeled spacecraft. And the telescope’s analysis team is gearing up for the rest of the science mission: a two- to three-year effort to systematically crawl through the four years of data that Kepler has collected since its launch in 2009.
That analysis effort, which will incorporate new machine-learning techniques and a bit of human experimentation, could yield a bounty of new potential planets on top of the 3500 that Kepler has found so far. “We expect somewhere between several hundred more planets to maybe as many as a thousand,” Borucki says. If all goes well, the revised hunt might even uncover the first handful of terrestrial twins—or at the very least, near cousins: roughly Earth-size planets on nearly yearlong orbits around sunlike stars.
Uncovering those Earth analogues won’t be easy. The orbits are slow and the planets themselves are small. “You’re looking for a percent of a percent” dip in the brightness of a star, says Jon Jenkins, the telescope’s analysis lead. “That’s a very demanding and challenging measurement to make.”
The task will be made even more difficult by an unexpected complication: Stars vary in brightness due to sunspots and flares, and Kepler’s observations reveal that these variations are greater than scientists had previously estimated. Those fluctuations can hide the presence of a planet, reducing the telescope’s sensitivity to terrestrial transits by 50 percent.
In April 2012, NASA granted Kepler a four-year extension that would have compensated for the extra noise. But with the failure of the reaction wheels, Jenkins and his colleagues now must find a different way to uncover planetary signals.
Earlier this year, they moved the data processing from a set of computer clusters containing 700 microprocessors to the Pleiades supercomputer at the NASA Ames Research Center, in Moffett Field, Calif., where they have the use of up to 15 000 of the machine’s more than 160 000 cores. The team is also working on implementing a machine-learning process using an algorithm called the random forest, which will be trained with data already categorized by Kepler scientists. Once it’s up and running, the software should be able to speedily differentiate false positives and data artifacts from promising candidates. Eventually, Jenkins says, the analysis team will insert fake data into the pipeline to test the performance of both the humans that ordinarily do the processing and the automated algorithms. “We need to know for every planet we detect how many we missed,” Jenkins says.
No one can predict exactly how many planets Kepler will find. The telescope’s main goal was to determine how common planets are in and around the habitable zones of stars—the areas around stars with the right temperature range for liquid water to be present. Such statistics could help astrophysicists decide how practical it would be to build a space telescope capable of directly detecting light from Earth-like planets, which is necessary to determine whether they have atmospheres that could support life.
For Earth-size planets in settings similar to our own, developing a good statistical estimate will be difficult. With small numbers, the uncertainty in the size of the overall population will be large. “The best-case scenario is that Kepler could still have, with very large error bars, a number for us at the end of the day,” says Sara Seager, a professor of planetary science and physics at MIT and a participating scientist on the team.
Even if Kepler finds no Earth analogues, Seager says, the mission is a success. “Kepler revolutionized exoplanet science and, arguably, big-data astronomy,” she says. “We’ll see the data being mined for years to come.”
This article originally appeared in print as "Keplers Continuing Mission."