Professor-Student Team Helps Improve Disaster Preparedness of Coastal Areas
In October 2016, North Carolina’s coastal residents were swept into the hurried and anxious routine of closing shops, boarding up windows and moving to higher ground. The governor declared a State of Emergency as Hurricane Matthew’s Category 4 winds spiraled closer.
Several hundred miles to the south, at Benedict College in Columbia, South Carolina, Anton Bezulgov, Ph.D., and Reinaldo Santiago tracked the storm’s trajectory with their own sense of urgency: if their recent summer research is vetted by peers and adopted by emergency management teams, their storm surge computer model could help hundreds of thousands of residents prepare for similar natural disasters in the future.
“The goal of our summer research was to make coastal areas safer,” said Bezuglov, a computer science professor at Benedict College. “When a storm surge is estimated quickly and accurately, emergency managers and other officials can spend more time carrying out emergency preparedness activities.”
Bezuglov and Santiago, a senior in computer engineering at Benedict College, conducted their research on storm surge predictions through their participation in the U.S. Department of Homeland Security (DHS) Summer Research Team (SRT) Program for Minority Serving Institutions. The purpose of the DHS SRT Program is to increase and enhance the scientific leadership at Minority-Serving Institutions (MSIs) in research areas that support the mission and goals of DHS.
This program provides faculty and student research teams the opportunity to conduct real-world, meaningful research at the university-based DHS Centers of Excellence (DHS Centers). The SRT Program and DHS Centers are sponsored by the DHS Science and Technology Directorate Office of University Programs.
The research was conducted at the Coastal Resilience Center of Excellence at the University of North Carolina at Chapel Hill, which is colocated with the Renaissance Computing Institute (RENCI). Bezuglov and Santiago developed a software program to help predict storm surges. They worked under the mentorship of coastal oceanographer Brian Blanton, Ph.D., who serves as RENCI’s Director of Environmental Initiatives.
“Existing methods for storm surge predictions can provide great accuracy, but their computational complexity is generally very high, and they may take many hours to execute,” said Bezuglov, the team leader. “Our project, which uses artificial neural networks (ANNs), is a significantly quicker, cheaper method.”
Bezuglov explained that ANNs somewhat resemble neural networks in animals and comprise a significant number of interconnected processing elements called inputs. The inputs in this case are parameters like a hurricane’s location, central pressure and radius to maximum winds. ANNs function similarly to mammalian brains in that they can be “trained” over time to recognize and respond to various inputs.
Bezuglov and Santiago spent their time developing the neural network model, experimenting with different inputs, and finding and fixing subtle errors in the model. Every day, they generated and tested new ideas and approaches, compared the networks and evaluated their performances.
“Outside observers might think our days were absolutely boring, as there were no special equipment or devices—only computers,” said Bezuglov. “But that could not have been further from the truth. There was nothing more rewarding and enjoyable than seeing our software work as designed, especially when it meant tracking down and fixing an error in code that had been bothering us for a few days.”
As the team leader, Bezuglov developed the general research direction, led weekly team meetings, and guided Santiago through complex software development problems. He also challenged Santiago to reengineer top-performing models to be even better.
“I suggested a form of competition, where I gave my best performing model to Santiago and asked him to ‘beat’ it, i.e., train a better, faster model,” said Bezuglov. “At the end of the ‘competition,’ Santiago developed a model five times as accurate as the initial network.”
By the end of the DHS SRT MSI Program, the team developed a storm surge model that estimates the storm surge at 10 coastal locations in North Carolina with an average accuracy of three inches and an operational time of less than one second.
Along the way, Santiago improved his Python™ programming skills and gained experience with NumPy and SciPy libraries, TensorFlow™, Git and GitHub, Linux and LaTeX text preparation. These are all specialized and in-demand technical skills he can use along his path to a master’s degree in computer science and beyond.
“This research experience cannot compare to anything else,” said Santiago, who enjoyed the upbeat, collaborative and creative atmosphere of the research experience. “It has had a great impact on my career planning and allowed me to learn so many new things. I would definitely recommend the program to others.”
Bezuglov couldn’t agree more.
“One of the most beneficial aspects of the DHS SRT MSI Program is the 100 percent immersion in a specific research project for 10 weeks, five days a week, eight hours a day. This is rarely possible under other circumstances,” he said. “Overall, the program is an absolutely positive experience that I will highly recommend to others.”
The DHS SRT Program is funded by DHS and administered through the U.S. Department of Energy’s (DOE) Oak Ridge Institute for Science and Education (ORISE) through an interagency agreement between DOE and DHS. ORISE is managed by ORAU for DOE.