Whitney Palmer

But they will be better prepared in the future to answer these questions – not just about one hurricane, but about all hurricanes during a given year. Thanks to a new predictive model from researchers at N.C. State, storm watchers and scientific investigators alike will be able to forecast a season’s hurricane activity with 15 percent more accuracy over existing models.

The goal of the research, published in Data Mining and Knowledge Discovery,

Nagiza Samatova, an N.C. State associate professor of computer science, is co-author of the hurricane modeling report. Courtesy: NCSU

investigators said, is twofold.

“Hopefully, this approach will give policymakers, insurance providers, and people responsible for preparedness a more reliable method for anticipating hurricane activity,” said Nagiza Samatova, N.C. State associate professor of computer science and study co-author. “We also hope it will impact climate change modeling and work that can predict what climate changes we can anticipate.”

New system vs. old model

Despite the number of storms that happen annually, historical data surrounding hurricanes is surprisingly small. Meteorologists currently use information dating back to 1950 to estimate how many storms will occur during a single hurricane season. Data about temperature, humidity and precipitation from a small number of hot spots are run through a statistical algorithm to predict how many storms the East Coast will experience. Each hot spot is a location around the globe known to be the origination points for hurricane activity.

“The problem with this model is it has a relatively small amount of data to work from,” Samatova said. “Sixty or so years of information from a few areas of activity worldwide isn’t enough to really predict how a hurricane season will play out.”

Consequently, determining which information correlates to low activity and what indicates high activity has been challenging. The new N.C. State model, however, aims to alleviate that problem by pulling data from nearly 10,000 points across the globe. Samatova likened this model to a social network where each node is a different person. Just as more is known about an individual and how they connect with others when they provide more details within a social network, a growing bed of meteorological knowledge will help scientists forecast from where the most intense storms will spring.

By entering historical data into the new algorithm, Samatova and her N.C. State colleague, Fred Semazzi, a professor of marine, earth, and atmospheric science, have confirmed that their model predicts hurricane activity with 80 percent accuracy over the 65 percent accuracy rate of traditional methods.

Benefits of the model

In addition to providing more precise estimations for storm activity, the N.C. State model has also helped identify previously unknown hotspots, Samatova said. For example, there are several points around southern Africa that scientists never realized directly affected hurricane activity in the North Atlantic region – the area covering the East Coast.

Professor Fred Semazzi of N.C. State, in front of an image of Hurricane Fran: Adding historical data and unusual characteristics into the new algorithm makes the N.C. State model better at predicting hurricanes. Courtesy: NCSU

“Identifying new hot spots is an advantage, because now we’re able to discover new, interesting features of what makes a location a hot spot that we didn’t already know,” she said. “This way, we’ll be able to build out even more models, and we’ll be better able to identify high-activity or low-activity seasons.”

Activity levels are a sliding scale based on geographic region. In the North Atlantic, eight or more hurricanes equates to high activity; five to seven storms is average; and anything less is considered low activity.

Through the new model, researchers also identified additional characteristics that can help predict hurricane activity. Incorporating unusual behaviors, such as strange patterns or unusual storm edges, can also enhance the model’s predictive ability.

“By using additional factors, we hope to learn more about what influences a hurricane’s behavior,” Samatova said. “This could allow for an impact on climate change work if we’re able to forecast storm behavior well into the future.”

New models needed

Previous research from the University of Notre Dame supported Samatova and Semazzi’s work that new predictive modeling is needing to accurately forecast not only the occurrence, but also the impact of hurricanes. Although that research focused on the Gulf Coast region and anticipating storm surge, researchers emphasized the need for more data gathered from an extensive number of points across the globe.

“Accurate model forcing is necessary to simulate the correct storm surge propagation,” wrote Jesse Feyen, a former Notre Dame doctoral student and current researcher in the National Ocean Office in the National Oceanic and Atmospheric Administration.

In the wake of Hurricane Katrina, researchers at the Massachusetts Institute of Technology also began asking questions similar to the ones N.C. State’s model aspires to answer. MIT’s team seconded Feyen’s assertions and noted one practical concern for which N.C. State’s model could eventually be greatly beneficial.

“There is a time component. Scientists simply cannot predict hurricanes early enough for cities to be completely prepared,” they said. “There is no certainty in the position of a hurricane until it is too late to respond. Hurricane predictions in the future need to be more accurate earlier on in the forecasting process.”