By Paul Gabrielsen, senior science writer, University Marketing and Communications
At about 3 p.m. on Jan. 26, 2011, as a normal, albeit cold, Washington, D.C., workday drew to a close, heavy snow started to fall. By the time the surprise snowstorm ended six hours later, low visibility, slick roads and 7 inches of snow left hundreds of cars smashed and many others abandoned in roadside snowbanks.
The storm was a disaster, and area transportation officials were unprepared to respond quickly in order to mitigate the traffic hazards. Richard Medina, a University of Utah geography professor, along with Guido Cervone of Pennsylvania State University and Nigel Waters of the University of Calgary, used geospatial tools to analyze patterns of traffic patterns in Fairfax County, Virginia, during the storm. Their results, published in June in The Professional Geographer, track the progress of the desperate homeward commute that day, and identify patterns and trends in the accident data.
“The characteristics of the urban system environment give us patterns where we can predict where accidents are going to happen,” Medina says.
Snow on the Beltway
The Washington, D.C., area is no stranger to large snowstorms, most recently January’s storm that some dubbed “Snowzilla.” But weather forecasters usually can see the storms coming, allowing schools and workplaces to close and residents to stay home and off the roads.
“When any snow falls, everything closes down,” says Medina, a former resident of Fairfax County.
But the 2011 snowstorm was unexpected. School and work had gone on as normal. Workers, already used to legendary gridlock even on a good day, faced the prospect of inching their way home through an evening commute marked by blinding snow.
According to data that Medina’s team obtained from Fairfax County via a Freedom of Information Act request, local emergency personnel received more than 1,500 calls to 911 during the storm, and reported around 1,000 accidents, with 1,300 police dispatches and110 dispatches of emergency medical services.
The team combined this data, which included collision location, with geographical information system (GIS) maps of Fairfax County roads and speed limits. They further incorporated the spatial pattern and intensity of precipitation over the course of the storm to see which areas were most impacted by snow.
Medina says that patterns of accident incidence changed over the course of the commute. “We could track the drive home,” he says. An initial pattern of accidents in high-speed-limit areas gave way to more low-speed collisions as drivers moved from freeways to surface streets. Most accidents occurred in densely populated residential areas. Several, Medina writes, occurred near mass-transit rail stations.
The researchers developed a model to predict where accidents might occur, given the overlain GIS and weather data. As the model progressed through the accident data, its predictive power increased. After about 50 accidents, the model could reasonably predict the locations of half of the future accidents, and predict nearly 100 percent of future accidents after only 200 collisions. Since the snowfall affected all Fairfax County areas roughly equally, the main factors in predicting accident location were road type, speed limit and zoning (commercial vs. residential, for example).
Eyes on Salt Lake
Medina’s work aimed to examine accident patterns during severe weather events, which may increase in frequency and intensity in a warming climate. Next, he will analyze similar extreme weather accident data in the Salt Lake City area. The Salt Lake urban system is different than D.C., with fewer people and a population more used to driving in winter conditions. Closures due to weather are much rarer in Utah.
“Just because people in Salt Lake are used to the snow doesn’t mean they get in less accidents, or are less prone to accidents,” Medina says. “We could potentially see more here because things don’t close down.”
Identifying accident hotspots may help local officials design roads and cites to minimize hazards ahead of time, Medina says. Further, if emergency responders can track data in real time, predictive models like Medina’s could help reduce emergency response time and get more people home safely in bad weather.
The full study can be found here.