The behavior of massive wildland fires is incredibly complex, but as the western U.S. experiences these events with increasing frequency, the need to better understand those systems has become crucial.
Civil & Environmental Engineering Assistant Professor Hamed Ebrahimian is pioneering new scientific computation techniques to enhance prediction, decision-making and safety in critical areas such as earthquake engineering and wildfire prediction.
With support from the National Science Foundation (NSF) Faculty Early Career Development Program (CAREER), Ebrahimian is working to transform how machine-learning and physics-based computational models can be integrated to create “smart digital twins”: advanced virtual replicas of dynamic systems that can learn and adapt using measurement data.
The NSF CAREER program supports early-career faculty who have the potential to serve as academic role models in research and education.
“Our idea to develop new computational frameworks that can address critical shortcomings of learning methods in complex real-world applications, including limited model interpretability and generalizability, and limited training data,” Ebrahimian said. “The idea has been inspired by years of experience in developing computational models for critical engineering applications, including predictive modeling of wildland fires and post-earthquake damage assessment of civil structures.”
The project also supports the next generation of engineers.
“The project has a distinct objective,” Ebrahimian said, “to improve engineering education and help train the new generation of civil engineers and researchers at the cross-roads of computer science, applied mathematics and computational mechanics to solve use-inspired problems affecting our society.”
How does this new method work?
When trying to understand and predict the behavior of complex systems, engineers and researchers generally use physics-based models, which describe how a system should behave based on the laws of physics, Ebrahimian says. He asserts that those models can fall short because of the inherent idealization, simplifications and assumptions.
Ebrahimian’s proposed solution is to incorporate machine-learning components within the core of physics-based models to help them learn fast from data collected from real-world systems to improve their predictive capabilities. This technique promises to result in models that are more accurate and informative than a physics-based or data-driven model alone, Ebrahimian says.
These novel digital twin models could be applied to tackle wildland fires — an area of concern in recent years.
“Active-fire response is perhaps the most challenging and time-sensitive stage of wildfire management,” Ebrahimian said. “The high stakes and large uncertainties in predicting fire behavior places emergency responders and communities alike in a reactive position.
“We need to use computationally efficient and accurate models to get ahead of the fire, to predict the fire dynamics and to be able to compare and decide about the best response strategies in advance. Nevertheless, our (current) modeling techniques are often inaccurate, resulting in incorrect predictions.”
Ebrahimian and his team will work to develop machine-infused physics-based models for wildfire predictions. He argues that these models can learn quickly by observing the initial fire behavior and ideally would provide accurate predictions of fire dynamics hours in advance.
Broad use
Using computational models and modeling techniques to simulate complex systems has been an interest for Ebrahimian. He has been using models and data for different applications, including assessing post-earthquake damage in civil infrastructures and predicting the remaining useful life of offshore wind turbines in addition to wildland fire simulation. But wildfires remain top-of-mind in the public interest, as they have increased in size and frequency in recent years due to climate change and the expansion of communities within the wildland.
Across the U.S., an average of more than 70,000 wildfires have burned an average of 7 million acres – equal to the size of Hawaii – every year since 2000, according to the National Interagency Fire Center (NIFC). That is more than double the annual area scorched by wildfires between 1980 to 1999.
Last September, the Davis Fire in Washoe Valley resulted in widespread evacuations in communities south Reno and destroyed at least 12 structures.
The total economic impacts of wildfires across the nation is estimated to be in the range of hundreds of billions of dollars each year, according to the NIFC, and is expected to increase with exacerbating climate conditions. This level of impact is especially troubling, Ebrahimian adds, because various vulnerable communities, such as low-income, migrant, Indigenous and older populations, disproportionally bear the burden.
