In This Story
When an emergency occurs—be it a natural disaster or public health crisis—effectively directing people to safe shelter is critical, but providing accurate information and ensuring shelter resilience in a rapidly changing situation is fraught with challenges.
George Mason University's Hemant Purohit is working with the Virginia Beach Department of Information Technology and Department of Emergency Management to develop PCExplorer (Physical & Citizen Sensing Exploration tool), which will use AI to help officials guide a distressed public to safety through effective decision support to plan shelters and communicate to the public.
The National Science Foundation awarded Purohit and his research partners a $1.25 million grant to fund the three-year project. He is collaborating with a multidisciplinary team of experts, including Qian Hu from George Mason's Schar School of Policy and Government; computer scientist Ayan Mukhopadhyay and infrastructure engineering expert Hiba Baroud of Vanderbilt University; and social scientists Joshua Behr, Rafael Diaz, and Wie Yusuf of Old Dominion University.
"Consider an impending hurricane. The big challenge for emergency managers is three-fold—how to make people voluntarily move out of the harm's way, how to estimate shelter demand, and how to optimize the allocation of limited resources to manage shelters, because operating a shelter is a resource-intensive, costly responsibility," said Purohit, an associate professor in the Department of Information Sciences and Technology. "Frequently, the tools that public officials have on hand are not integrated and rely on leveraging limited data sources such as weather forecasts, simulations of evacuation times, and historical data of people's past behavior in using shelters."
PCExplorer will collect and integrate data from a variety of sources to give authorities real-time information on risks, using AI-powered analyses to continuously understand threats to infrastructure crucial to manage shelters. For instance, social media posts and traffic sensors may indicate a critical road leading to a shelter is blocked, or the power is out at the shelter, making it inaccessible and unsustainable. Decision-makers can rapidly update information given to the public, redirecting them appropriately.
"You may have traffic sensor data, weather forecast data, water level data, public transportation systems and social vulnerability indices," said Purohit. "Then you have citizen sensing, such as people posting about situations and reporting on apps like Waze, and when you combine that with other infrastructure data sources, you have a powerful mechanism for dynamic modeling of risks to infrastructure and their inter-dependencies." The integrated data is analyzed and provided to authorities via a simple, natural-language interface coupled with a map for explaining complex AI-based analyses.
Hu, an expert in emergency management and public administration with a focus on understanding interorganizational communication and coordination, added, "Evacuation and sheltering operations span emergency communications, transportation, infrastructure, health, and community engagement. Their success depends on coordinated action across organizational boundaries. Thus, our multidisciplinary approach will help integrate technical, social, and administrative expertise to understand the needs of these diverse stakeholders and find actionable information to aid decisions under extreme uncertainty."
Purohit has long been interested in working in disaster relief and assistance, which was the focus of his PhD work. In addition, via the United Nation's Young Innovator program, he had a fellowship in the UN's International Telecommunication Union. He earned the fellowship for designing open-source technologies for disaster management.
He aims to follow this passion of AI for social good with PCExplorer, which represents a step toward a future where data, insight, and collaboration help communities weather the storms ahead.