Defusing the hidden tinderbox:  How CISL research advances wildfire prediction accuracy

"With more accurate information, our ability to save lives and property can be dramatically improved."
— Scott Pearse, CISL Software Engineer

When the East Troublesome Fire ignited in October 2020, it defied expectations. 

In just nine days, the blaze consumed nearly 200,000 acres – including 87,000 acres in a single, devastating 24-hour period. 

Wildfire simulations are vital tools for modeling fire behavior, but the 2020 predictive models struggled because they were built on a flawed assumption: that the forest was healthy and upright.

In reality, prolonged drought, severe winds, and beetle infestations had turned vast swaths of the forest floor into a tinderbox of dead, downed trees. 

Because these real-time conditions were not captured in the available fuel datasets, models lacked the precision needed to forecast the fire's true behavior.
 

"It is our mission at CISL to support scientists at NSF NCAR and throughout the greater research community. And for me, it is a fulfilling mission to have in my life." — Scott Pearse

To tackle this challenge, Scott Pearse, a Software Engineer at NSF NCAR’s Computational and Information Systems Laboratory (CISL), joined forces with researchers across institutions to transform how fire behavior is modeled and visualized.

Today, a breakthrough collaboration between CISL and the Research Applications Laboratory (RAL) is changing that equation. 

By pairing AI-driven fuel data with high-performance supercomputers and high-resolution 3D rendering, researchers can now generate post-event simulations that mirror reality with startling precision. (Visit RAL’s dedicated web page on wildfire research.)

While scientists at RAL developed the AI models to process satellite imagery of dead trees, Pearse used CISL’s Visualization & Analysis Platform (VAPOR) to turn those massive, complex datasets into dynamic 3D visuals. 

 

“Please don't hesitate to reach out for help with visualizing and explaining your research." — Scott Pearse
 

When Pearse loaded the updated fuel data into VAPOR alongside the original East Troublesome Fire simulation, the stark contrast was immediate. “The unmodified fuel simulation shockingly underestimated the actual fire extent that occurred in 2020,” Pearse recalled, noting that the most striking takeaway was “the difference in the burnt fire area between the unmodified fuel simulation and the one with updated fuels.”

While VAPOR is designed as a scientific research and data exploration tool rather than an operational tool for active fire response, it can serve to bridge the gap between raw data and emergency planning. 
 

See related 2022 feature article from UCAR News.
 

By allowing scientists to investigate the accuracy of new AI-driven models, VAPOR helps researchers refine predictive frameworks and supports societal studies on how decision-makers interpret simulation data. 

VAPOR has also been instrumental in creating “explainer videos” to help the public and decision-makers interpret simulation data. 

Explainer examples include this one, produced by RAL graphic designer Cindy Halley Gotway, and this one, by Pearse. 

“Physical and statistical analysis will always be core to geophysical research, but qualitative visualizations represent data in a way that allows people to see phenomena with the naked eye,” said Pearse. “This helps scientists conduct and communicate their research, both of which align directly with NSF NCAR’s core mission: Science for a resilient, sustainable society.”

This interdisciplinary approach has earned widespread recognition, including the HPCWire Readers' Choice Award (2023) for "Best Use of High Performance Data Analytics & Artificial Intelligence" and the PEARC 2023 Best Visualization Award. The research has also driven key scholarly contributions across the field, including publications in Ecology and Society (December 2024), Fire Ecology (May 2024), and a PEARC 2023 Conference paper exploring how AI and visualization can drive wildfire simulations with better predictive skill.

The project's success is also opening international doors. Researchers at the Korea Institute of Science and Technology Information (KISTI) are currently adapting VAPOR to run on a massive high-performance graphics cluster, rendering the East Troublesome Fire simulations inside their world-class Computing and Advanced Visualization Environment (CAVE).

“It's thrilling to see this scientific work being applied half-way across the planet.” — Scott Pearse
 

“CAVE systems are immersive visualization environments, and KISTI's is the biggest I've ever seen, ” said Pearse. 

He added: “Check out the desk on the left side of this picture to get a sense of KISTI’s visualization environment’s massive scale.  The desk stands at less than three feet tall.” (See photo at right.)

As researchers continue refining these tools, grounding simulation models in real-time environmental data brings the scientific community closer to predicting megafire behavior with the precision required to better protect vulnerable communities.

For Earth System Science (ESS) researchers looking to push the boundaries of their own data, Pearse’s team is ready to collaborate. “It is our mission at CISL to support scientists at NCAR and throughout the greater research community. We welcome you to contact the VAPOR team, contact CISL HPC User Support, or see the VAPOR web page for assistance with rendering and communicating your scientific findings."

The results of this cutting-edge research can change the way we understand – and survive – the most challenging megafires.