Purdue's new Hansen cluster supercomputer models to aid hurricane-evacuation planning

Researchers are developing detailed models to predict how populations behave during hurricane evacuations to better plan for the disasters.

The models will be used by public policymakers to improve how evacuations are carried out, said Satish Ukkusuri, an associate professor of civil engineering at Purdue University.Satish Ukkusuri

"For example, during Hurricane Rita many people evacuated at the same time, not following the evacuation instructions," he said. "Then everyone was stuck on the freeway for a very long time and ran out of gas, further worsening the traffic congestion. The gas stations were depleted of fuel; people were out of food and water. But if you could stage the evacuation properly and we could understand how households react to evacuation warnings and make evacuation decisions, we could build better evacuation strategies."

Rita hit Texas and Louisiana in 2005, causing extensive damage, killing seven people directly and 120 overall from other factors, including evacuation mishaps.

"The models will lead to a practical means of predicting specifically how people will behave during a hurricane," Ukkusuri said. "Who will evacuate and who will stay? When will they evacuate, where will they go and which routes will they use? Understanding these issues is important to make evacuations safer and more effective."

The interdisciplinary work involves both engineering and the social sciences. Purdue is working with researchers at Florida International University and Virginia Tech.

"There are two major pieces to this work: behavioral research and computational modeling," Ukkusuri said. "The behavioral part is to learn how people in different demographic groups respond to hurricanes."

Survey data have been collected from people in regions impacted by hurricanes Katrina, Ivan and Rita. The engineering piece focuses on building transportation models for safe evacuation using the survey data. Different scenarios are analyzed with the computational models to improve the evacuation efficiency for future hurricanes.

"We are building a large-scale simulation model using computational techniques," Ukkusuri said. "There is a great need now to bridge varied disciplines and build tools from a holistic point of view, and that is what we are doing. These kinds of models are going to be more important in the future because of increasing storm severity from climate change and rising population density in coastal areas."

The work has been funded with several grants from the National Science Foundation totaling about $1.25 million. A current portion of the research, funded with a new $500,000 NSF grant, focuses on understanding the impact of evacuation warnings on household behavior using complex network science tools. These models will be tested in the Miami-Dade and Houston areas.

Researchers are using Purdue's new Hansen cluster supercomputer at the university's Rosen Center for Advanced Computing, the research arm of ITaP, or Information Technology at Purdue. The models will include vital information about how people use various tools such as mobile phones, social media and community social ties to make hurricane evacuation decisions.

Similar modeling approaches also are useful for other types of disasters.

"People get information from TV, the local emergency preparedness office, text messages and emails," Ukkusuri said. "We want to understand how that information spreads within the social network. People are connected differently. Some have tight-knit, family oriented communities, where conceivably news spreads fast, but others are more isolated and rely more on TV, Internet and social media tools as opposed to their pastor or friends."

{hwdvs-related}3{/hwdvs-related}