Can NextEarthquake.com Really Forecast Earthquakes?

Natalie Wolchover, Life's Little Mysteries Staff Writer

Date: 24 October 2011 Time: 12:34 PM ET

Every two weeks, Reginald Roberts posts a new set of earthquake forecasts on his website, NextEarthquake.com. Between 2,000 and 10,000 people visit the site each day, and Googling "earthquake prediction" returns NextEarthquake.com near the top of the search results.

"This is a good indication of how many people worldwide are using the site and finding it helpful," Roberts said. "My work has been attributed with saving a number of lives."

Roberts says his earthquake forecasts, which included a prediction of the earthquake that shook Hawaii on Thursday (Oct. 20), are accurate over 90 percent of the time. Though he refuses to divulge his methodologies for fear of intellectual property theft, he claims his background as an entomologist studying insect population dynamics has made him an expert on pattern recognition, and says he has worked out the rules governing the movements of Earth's tectonic plates.

"Earthquake forecasting is not an exact science, but my work is showing that we can, nevertheless, get considerable insights into earthquake dynamics and therefore anticipate [them]," Roberts told Life's Little Mysteries.

If Roberts is right, then all the world's seismologists and earthquake geologists are wrong when they say earthquakes are too chaotic to be predicted. If tectonic plates really do interact according to some underlying pattern and if Roberts has discovered that pattern then from now on, people could be forewarned of impending quakes and safely evacuated. Thousands of lives could be saved. [What Should You Do When an Earthquake Hits? ]

But not everyone believes his method really works.

How the predictions work

As seen in the "long-term forecast" tables on his site, Roberts makes about 40 earthquake predictions for each upcoming "moon period" the two weeks following each new and full moon . (This implies that he thinks the moon's phases somehow influence seismic activity.) Later, when one of his forecasted quakes happens, he makes note of it in the table.

Not counting all the earthquakes that he fails to predict more than 1.4 million measurable earthquakes actually happen worldwide each year, or 60,000 during each of Roberts' moon periods Roberts says his forecasts have an accuracy of 90 percent. "To be a successful forecast, the forecast quake must occur in the two-week lunar window (plus or minus one day to allow for events on the cusp of the moons), and within the nominated forecast zone," he said, noting that the 41 zones in which he makes his predictions, some of the most seismically active regions in the world, range from 225 to 1,000 kilometers (140 to 621 miles) in radius.

Furthermore, the observed quake must have a magnitude within 1.1 of the predicted magnitude in order for a forecast to count as successful, Roberts said. (For example, if he forecasts a magnitude of 4.0, the actual quake must be in the range from 2.9 to 5.1).

But, does Roberts' ability to hit those marks 90 percent of the time mean that his system actually works, or would any system random guessing, for instance hit with as much accuracy?

Safe bet, not forecast

According to John Vidale, a seismologist at the University of Washington in Seattle and director of the Pacific Northwest Seismic Network, it is definitely the latter. "Eighty or 90 percent is meaningless," Vidale told Life's Little Mysteries. "Anyone can construct a scale that scores whatever number they like the question is whether the earthquakes tend to arrive when they were anticipated or not." Roberts' A-grade stems from the fact that he gives his forecasts 16 days of wiggle room, Vidale said. [Psychology of Fear: Why People Listen to Earthquake Prophecies ]

Looking at a recent example, Roberts' appears to have successfully forecast the magnitude-4.5 earthquake that happened in Hawaii on Oct. 20. His original prediction held that a magnitude-3.7 quake would strike between Oct. 11 and Oct. 26 in a 96,526-square-mile (250,000 sq. km) region around Hawaii. Because he had pinpointed the correct moon period, the correct seismic region, and a quake magnitude that was within 1.1 of the actual magnitude, he judged his forecast as successful.

Earthquakes in Hawaii on October 19-20, 2011.
CREDIT: USGS/Hawaii Volcano Observatory

View full size image

However, U.S. Geological Survey records show that Hawaii typically experiences several minor earthquakes each day. This means Roberts was almost guaranteed to guess correctly that a quake somewhat close to magnitude 3.7 would happen within his 16-day grace period.

"Any legitimate predictor will compare the results of their forecasts with the same list of forecasts with scrambled times, i.e., are they doing better than random guessing? I've never seen a phony predictor try ... to surpass such a hurdle, and most wouldn't pass this test were they to try," Vidale said.

Lastly, Roberts is far too lenient when it comes to judging his guesses of impending earthquakes' magnitudes. The magnitude scale is logarithmic , not linear, meaning that the 4.5 quake that happened in Hawaii was actually about 15 times more powerful than the 3.7 quake he predicted. Thus, his guess came within only 7 percent of the Hawaii earthquake's actual energy. Had there been no powerful earthquakes in Hawaii in the entire 16-day moon period, but merely a tiny magnitude-2.6 tremor, Roberts still would have judged his prediction a success, since 2.6 is within 1.1 of 3.7. But a 2.6-magnitude quake is hundreds of times less powerful than the one that actually struck.

The vast majority of Roberts' "forecasts" are more like "safe bets." Occasionally he goes out on a limb by forecasting stronger earthquakes. These typically don't come to pass, accounting for Roberts' 10 percent failure rate.

Follow Natalie Wolchover on Twitter @nattyover. Follow Life's Little Mysteries on Twitter @llmysteries, then join us on Facebook.