Blown Away: 3 Questions about the science of hurricane prediction

Dr. Montserrat Fuentes is part of NC State’s hurricane research team led by Dr. Lian Xie, professor of marine, earth and atmospheric sciences. Ph.D. student in statistics Danny Modlin rounds out the team.

Abstract: How accurate can a hurricane model be, really?  What are the issues that hinder accuracy in hurricane prediction, and how can we improve our forecasts going forward?

Fuentes: Weather is a chaotic system and hurricanes, especially landfalls, are very difficult

to forecast. Advances of the last half-century have brought tremendous improvements in hurricane forecasting and, despite a growing coastal population, have resulted in a  dramatic decline in hurricane-related fatalities. The real challenge is that our forecast is for an entire season, and it relies on weather patterns and climate indices (functions of sea-surface temperatures [SST]), both of which are very difficult to predict.

We do update the hurricane forecast later in the season, June 1, when we have more current information about those important predictors in our regression model.  As more statisticians get involved we should be able to provide better measures of uncertainties associated with the forecast, and as we get access to higher quality information to characterize weather patterns and SST  (the most relevant predictor), such as satellite data, networks of ground- and ocean-based sensors,  and Hurricane Hunter aircraft we should be able  to improve the forecast even more.

Abstract: How did you go about creating a model that could predict hurricanes?

Fuentes: I collaborate with oceanographer Dr. Lian Xie to create a seasonal hurricane prediction—that’s the one that the meteorologists on the news start talking about every year around April 1– that estimates how many named storms are predicted and how many hurricanes will make landfall. Our model evaluates data from the last 100 years on Atlantic Ocean hurricane positions and intensity, as well as other variables — including weather patterns and SSTs. We use a regression model, which takes the relationship between many variables and their effect on one another into account, to summarize the data. And we estimate the relationship between the number of hurricanes and the observed historical data (on weather patterns and SST). We then plug in our current information about SSTs and weather patterns, and make our prediction.

Named storms are typically predicted based on past occurrences and current measures of factors in the climate.  We provide probabilities; we can’t say that the fifth named storm of the season will hit the Outer Banks of N.C. on September 2nd, 2010. We can only say that there is some probability (this year a 70 percent chance) of  a major hurricane hitting the coast of the southeastern United States between June 1 and Nov. 30.

Abstract: How does a statistician get involved with the science of predicting hurricanes?

Fuentes: All the models used for the seasonal prediction of hurricanes are statistical models. One of the important challenges in hurricane prediction is to quantify and characterize  uncertainty in the forecast, and since statistics is the science of uncertainty I think it’s completely natural for statisticians to get involved in predicting hurricanes. However, we certainly could not provide reasonable and useful models without a good understanding of climate and weather.