How AI tools are transforming disaster response, preparedness

Dr. Ali Mostafavi sees Texas as ground zero for natural disasters. The state regularly sees hurricanes, flash floods, wildfires and tornadoes — hazards during which a matter of minutes can mean the difference between safety and catastrophe.

During extreme rainfall events, for example, expediting the prediction of a flooded road or neighborhood 30 minutes sooner could help save hundreds of lives, said Mostafavi, a professor of civil and environmental engineering in the Texas A&M University College of Engineering. Researchers at his UrbanResilience.AI Lab — who recently partnered with Meta — are developing artificial intelligence systems that can be translated into technologies for augmenting situational awareness and resilience capabilities across all stages of a weather hazard to reduce the impact of disasters.

Take flash flooding: If a local emergency manager learns of an upcoming storm with potential for heavy rainfall, AI could be used to quickly analyze massive, disparate datasets — everything from which parts of the community flooded during the past 20 years of storm events to the predicted amount of rainfall in the next six hours based on readings from rainfall sensors and stream gauges — to predict the neighborhoods and ZIP codes most likely to flood.

It’s a significant amount of information for humans to digest, but it’s a quick task for AI. Mostafavi said AI systems have significant computing power and speed to reliably provide the insights humans need to make decisions during times of crisis when every minute counts.

As part of the partnership with Meta, for example, Mostafavi and his colleagues will work to build AI models that leverage large language models (such as Llama) in creating Disaster Management Companion AI systems.

“If we are able to integrate these applications into disaster management processes, we can significantly improve the anticipation, situational awareness and response to these events,” he said. “These events will be more frequent, and we cannot let them have the impacts they currently have. We see these AI systems as next-defense barriers for cities and communities.”

AI advantages

If Texas is ground zero for disasters, Texas A&M is “ground zero for solutions,” Mostafavi said.

While large language models and generative AI systems that create text and images — like ChatGPT or Gemini — are what come to mind for most people when thinking about artificial intelligence, the UrbanResilience.AI lab is at the forefront of using predictive, analytic and generative AI to build, test and refine new methods and models to gather and analyze big data for decision-making.

Mostafavi explains that predictive AI uses data from historical and current events to predict future outcomes, while analytic AI focuses on analyzing data to extract insights from trends and patterns. Each type of system has different use cases across the phases of a disaster, and when used in tandem can help emergency responders, local and state governments, emergency managers and residents better predict and cope with crises.

In addition to speed and computing power, another benefit of AI systems is their capability to handle multiple modalities of data, like readings from rainfall sensors and stream gauges, historical information on damages and losses, satellite images, location-based cellphone activities, news articles and even photos and videos posted by residents to social media. 

“It takes the process of understanding the situation, the risks and the impacts more complete, and it removes blind spots,” Mostafavi said. “That’s one of the main capabilities we have been working on to improve situational awareness during disasters. The ability to digest and provide near real-time insights at critical times is essential, and that’s where AI systems can play a significant role.”

AI solutions across the disaster cycle

Mostafavi’s lab has developed several AI systems for use across each phase of a hazard event: mitigation and preparedness, response and recovery.

In the mitigation phase, predictive models can evaluate which neighborhoods are most likely to flood before a hurricane makes landfall based on historical flood risk and topographic and hydrologic characteristics and even anticipate how people will evacuate from high-impact areas. During the active response phase, AI can monitor the status of neighborhood-level evacuations, power outages and property damages, providing insights to decision-makers on where to deploy search and rescue personnel and other resources in near real-time. And in the recovery stage, it can make rapid impact assessment using high-resolution satellite and street-level imagery.

This is just to name a few of the tools Mostafavi’s lab has developed. He said it isn’t about replacing people, but rather augmenting human decision-making in times of crisis when every minute counts. Another project under development: an emergency operating center “companion” functions like ChatGPT for emergency managers during disaster events, helping them expedite decision making by providing quick access to historical disaster reports and other data sources.

Real-world testing

Mostafavi has field tested several of these tools during some of the country’s most impactful weather events — including hurricanes Beryl, Milton and Helene, the Los Angeles wildfires and the deadly July 4 flooding in the Texas Hill Country.

“It’s an essential part of the process because during each field test we identify additional features and capabilities and improvements that our AI applications need to have,” he said. “This requires very strong partnerships between universities, tech companies working in this space and public agencies.”

He anticipates multiple AI applications will become standard tools for emergency management within the next three to five years. But for that to happen, Mostafavi said, fundamental research and development needs to be supported in this space, and agencies need to be incentivized to field test the technologies so they can be refined and scaled into products to be used across the nation.       

Disasters can leave people feeling helpless, Mostafavi said, but AI systems can help predict and mitigate their impacts.

“These technologies have a significant ability to save lives, protect communities, reduce the impacts and help us deal with this increasing frequency and magnitude of hazard events,” he said.