As early as last Wednesday, meteorologists like Slate's Eric Holthaus were predicting severe tornado outbreaks over large areas on the Southeast and Central Plains this weekend. On Sunday, those forecasts proved sadly accurate as large thunderstorms and tornadoes wreaked havoc in Arkansas, killing as many as 17 people. 

The storms caused devastating destruction, but those tornado predictions were vital forewarnings for people who were more prepared than they might have been just a few years ago. Thanks to Big Data and more powerful computers, the accuracy of tornado forecasts and warnings is rapidly improving. In fact, it just may be pushing up on the theoretical limit of tornado predictions.

The fact that meteorologists can predict volatile events like tornadoes about four days away is a direct consequence of "the incredible increase in computing power," Holthaus said in an interview with The Wire. The key improvement has less to do with an increase in data volume than with the ability to manipulate that data with modeling. 

The National Weather Service relies on computing power
for its work (Reuters/Gene Blevins)

Indeed, Holthaus said he had been eyeing this particular storm for 36 hours before even writing a story last Wednesday, as he held off for a bit to fully confirm the breadth of the weather warning. But he's not the only one seeing the trends. The Storm Prediction Center's [SPC] forecast stretches out to eight days ahead of time, and according to CNN, the Severe Storms Laboratory forecasts seven-to-eight days in advances. The SPC's FAQ page writes that "the most important hardware for forecasting at the Storm Prediction Center is the human hand." But it's clear the increased sophistication and speed of modern machines is leading the future of predictions. 

Holthaus in particular singled out an analog model that compares the current weather conditions to a 30-year database of similar conditions. This allows meteorologists to examine a significant sample size of historical data to build their models on. "It's basically a Big Data project that they're looking for patterns that match the current forecast," he explained, as it uses history as a guide. "That's not a traditional forecast," he said. "It's more of a number-crunching exercise." With that model, Holthaus noticed early last week that the best comparisons for the current weather conditions in the middle of the country matched conditions that brought some of the biggest storms in U.S. history. That model helped solidify the scope of his warnings.

Multiple radar monitors track the storms (Reuters).

Of course, these predictions apply on a regional scale, with wider time frames, and still can't provide a concrete warning that a tornado will touched ground in a specific neighborhood. But even these more traditional warnings have improve drastically in recent years.

The average warning time for tornado threats has increased from five minutes to about 13 minutes in the last two decades, according to The VergeThanks to a recent upgrade in 3D-radar technologies, scanners can pinpoint spiraling storms that are sending large debris flying through the air. That can help discover where powerful tornadoes could be landing before the funnel cloud even fully forms.

As these forecasts improve, though, they push up on the theoretical limit to tornado prediction. Can meteorologists really issue an immediate tornado emergency warning for an area before that tornado (or the storm itself) technically exists? "Issuing the warning based on a forecast of the tornado forming – that's not something that's ever been possible before," Holthaus said. But as it proves possible, it could save untold lives as people receive ample time to seek shelter.

There's always a concern, of course, that too many warnings will lower people's guard for the truly important emergencies. But as these forecasts get more accurate and the number of false warnings decreases, they could eventually lead people to take every single warning more seriously. That would mean those who see or hear a warning will find shelter first, rather than take time to look out the window or call neighbors to confirm.

"Forecasters are getting more confident in models because models are getting better," Holthaus explains. Similarly, as forecasters get better, people believe them more. The better the models, the better the forecasts, the more lives can be saved.