In an innovative leap for disaster management, a professor and his team at LSU have introduced “DeepFire,” an advanced tool harnessing artificial intelligence to forecast the onset of wildfires before they ignite. This cutting-edge system aims to enhance preparedness and response to such emergencies.
DeepFire operates by predicting the likelihood of wildfires in specific areas, enabling emergency managers to efficiently distribute resources in anticipation of potential disasters. This project took shape in 2019, when a group of undergraduate researchers approached Supratik Mukhopadhyay, a professor in the Department of Environmental Science, with the idea.
Mukhopadhyay emphasizes that DeepFire is groundbreaking, as it serves as the first tool capable of detecting wildfires prior to their emergence. “Currently, wildfires erupt unexpectedly—these destructive events don’t provide advance warning,” he explained. “Fire departments are often caught off guard due to their limited resources.”
The predictions offered by DeepFire help disaster managers prioritize fires over various timelines, including four, seven, 15, 28, and 35 days in advance. Mukhopadhyay notes that the accuracy of DeepFire’s forecasts hovers around 90%.
“If I predict there will be a fire in 15 days in a specific location, there’s a 90% likelihood it will occur,” he stated, highlighting the advantage of readiness in mobilizing response teams and resources effectively. “Preventing fires can be incredibly challenging, and often, it’s hard to implement measures that can stop them.”
Emerging as pioneers in wildfire technology, Mukhopadhyay’s team recently secured a finalist spot in the XPRIZE Wildfire competition in Australia, vying for a grand prize of $11 million, and they have already won $85,000 in earlier rounds.
The research group consists of Mukhopadhyay, Rubayet bin Mostafitz (assistant research professor with the LSU AgCenter), Saiful Sajol (doctoral student and graduate research assistant in the Department of Environmental Science), and Thomas Douthat (assistant professor of environmental science).
Mukhopadhyay considers their communication framework and rapid prediction capabilities as distinctive advantages over competing teams, although they face challenges with satellite data refresh rates, which can vary based on factors like cloud cover.
Moreover, the team must adapt their approaches to the diverse climatic conditions in various regions. For instance, Australia’s dry, nutrient-poor landscape contrasts sharply with the more fertile soils found in places like Alberta, Canada, or California—regions where DeepFire has been put to the test.
Looking ahead, the team aspires to broaden their focus to address other natural hazards such as hurricanes, floods, tornadoes, and droughts—disasters that similarly require urgent attention from emergency managers. Bin Mostafiz pointed to Hurricane Katrina and the significant Baton Rouge floods in 2016, suggesting that a predictive tool could have made a notable difference in those situations.
“We are aiming to adapt this technology for various hazards,” Bin Mostafiz shared. “We plan to submit a proposal to the National Science Foundation to investigate not only the detection of these hazards but also how individuals react during evacuation scenarios.”
