Tornadoes are commonly understood to be independent of each other. Most forecasters assume that, given the right conditions, the number of tornadoes that occurs on a given day is just as likely as any other number. To check this assumption James Elsner from Florida State University, US, and colleagues examined daily tornado counts in the US over the period 1994–2012. When the researchers plotted the frequency distribution of tornado days by how many tornadoes took place on that day on a log-log graph they found that the size of a tornado day follows a power law relationship (Environmental Research Letters (ERL)). “Our study found that there is more order in tornado statistics than commonly assumed,” said Elsner.

It seems that, like earthquakes, solar flares and the firing of neurons in our brains, tornadoes follow a power law and are not entirely independent of each other. In the case of tornadoes, Elsner and his colleagues discovered that the log-log plot had an exponent of 1.64.

“Because the exponent is less than two it tells us that average statistics (for example, there are an average of 4.5 tornadoes per tornado day) are practically worthless for estimating the probability of rare events, like the tornado outbreak of 27 April 2011,” he explained. In fact, the log-log plot shows that the chances of having 145 tornadoes in one day are 0.014% (just over one in 10,000), not the far smaller one in a million that random chance would predict. Given that probability, Elsner and his team calculate that an event like this is likely to occur approximately every 71 years in the US.

When studying tornado size, the scientists found that the total number of tornadoes of a particular damage category follows an exponential rule, with the number of tornadoes doubling for progressively weaker damage categories. For example, on a day where one EF5 tornado occurs there are also likely to be two EF4 tornadoes, four EF3 tornadoes, eight EF2 tornadoes and 16 EF1 tornadoes.

Although these statistics are for North American tornadoes, other tornado-prone regions, including Argentina, Bangladesh and Europe, could well exhibit power law distributions too, although probably with a different exponent.

How can this new information help us be better prepared? Elsner suggests that the results should be used in tornado hazard models. “More accurate risk assessments might lead to practical measures such as safe rooms and building codes in areas like the US South or Midwest,” he said. In the short term the results can improve tornado forecasts, giving greater prior warning for violent and extreme tornado outbreaks, and some hope of mitigating the damage.

Related links

Related stories