When Wildfires Erase History: Using AI to Map What America Stands to Lose

 

Somewhere in the American West, there is a building that has stood for more than a century. It might be a one-room schoolhouse, a Spanish mission, an inn, or a ranch house where a beloved entertainer and philosopher once lived.

Many of these places are still standing. For now.

Although wildfire is a natural part of living in the Western United States, it has unintentionally destroyed these kinds of cultural structures. The July 2025 Dragon Bravo Fire took the Grand Canyon Lodge in its wake, which was built in 1927. Six months earlier, the Will Rogers Ranch House was lost in the January 2025 Palisades Fire. 

These weren't just structural losses. They were the erasure of places that held collective memory, identity, and belonging for generations of people.

According to a new study published in Scientific Reports by researchers at NCEAS, these losses are at risk of becoming far more common and uncovering just how widespread the threats are requires researchers to work at a scale made possible with AI.

 

 

A Gap in the Map

For lead author and previous NCEAS analyst, Mona Farnisa, the research grew out of a question about what wildfire planning leaves out.

"In wildfire planning there is a lot of emphasis on how to protect individuals, homes, infrastructure, and biodiversity," Farnisa says. "Cultural and historical heritage is not often given center stage."

Countries including Italy, Spain, and Greece had already conducted national assessments of wildfire risk to cultural heritage. The United States had not.

The project initially took shape as one component of the broader Wildfire Resilience Index, a research effort to understand and measure how communities and ecosystems weather the growing threat of fire. But as Farnisa dug deeper, she saw an opportunity to fill an important gap: identify which of America's historic and cultural resources are most exposed to wildfire.

Doing this analysis on a national scale meant bringing together two enormous and complex datasets.

Farnisa combined information from the National Register of Historic Places, maintained by the National Park Service, with a collection of data from the U.S. Forest Service that can help estimate the likelihood of a given landscape area burning in a wildfire (commonly called the FSim).

The National Register contains more than 100,000 historic sites, while the FSim provides a nationally consistent picture of the probability of wildfire burning across the landscape.

But combining the datasets was not as simple as putting two maps on top of each other.

The data from the FSim were far too large to simply download and the National Register dataset itself was messy, with numerous duplicate entries for individual sites. Not to mention, the team needed to determine which of more than 90,000 historic sites fell within areas with modeled wildfire exposure.

This is where artificial intelligence (AI) became part of Farnisa's research workflow.

AI-assisted coding helped her develop ways to download and work with the large FSim datasets, identify National Register sites located within burn areas, and clean and organize poorly drawn spatial boundaries within the National Register database.

Rather than replacing the analysis, AI helped Farnisa develop and refine the code needed to handle a massive amount of information and turn it into something the researchers could analyze.

"AI was particularly useful for the coding and data-cleaning parts of the project," Farnisa explains.

The technology also helped at the end of the research process by making the findings accessible.

 

 

From Data to Understanding 

Once the analysis was complete, the researchers wanted people to be able to explore the results themselves by building a map visualizing the information, with the help of AI. Instead of presenting the findings only as a static paper or dataset, the interactive experience gives users a way to engage directly with the results and see where cultural resources are most exposed to wildfire.

That matters because the goal of the research isn't simply to document risk. It's to help people act on it, and knowing what is at risk is the first step.

Farnisa hopes the research will help state historic preservation officers and local officials identify vulnerable heritage within their regions and incorporate those places into wildfire planning before a fire starts.

"Resources during an active wildfire are incredibly thin," she explains. "Deciding what to protect and save, whether it be lives, infrastructure, homes, biodiversity, or heritage is not simple."

The study's open-source dataset gives agencies at every level a consistent, nationally comparable foundation for making those decisions. The interactive webpage offers another way to explore that information and understand where the greatest risks are.

AI was one piece of the process that helped get the researchers there, but the larger goal remains decidedly human: helping communities decide what they value, what they want to protect, and how they can prepare before the next fire.

"By identifying where and what is most at risk," says Farnisa, "this study provides a foundation for proactive planning to safeguard the places that anchor community identity and national heritage before they are permanently lost."

The schoolhouse, the mission, the inn, the ranch house. Most of them are still standing.

Whether they still are a generation from now may depend on what we do next.

 

 

The research was funded by the Gordon and Betty Moore Foundation.