Just Accepted
This article has been peer reviewed and accepted for publication. It is in production and has not been edited, so may differ from the final published form.
Learning from Wildfire Decision Support: large language model analysis of barriers to fire spread on a census of large wildfires in the United States (2011-2023)
Abstract
Background: Barriers are the landscape features that firefighters leverage to stop wildfire spread. In the United States, decision-makers discuss barrier availability in a framework called the Wildland Fire Decision Support System (WFDSS). Aims: This study analyzes WFDSS text from 6630 large wildfires and examines the barriers identified. Methods: A large language model was trained and validated, then used to detect 13 different barriers. Burn scar and fuel treatment barriers were compared to their availability near each fire. Key Results: Decision-makers recognize barriers on most wildfires (75%) and explicitly state when they are not present. Roads (42% of incidents) burn scars (26%), and natural fuel variability (25%) are the most common barriers. There is a strong relationship between the amount of burned area available and the probability that it is identified as a barrier. The relationship between fuel treatment barriers and their availability is weak, but fuel treatments are recognized as barriers at lower landscape thresholds than burn scars. Conclusions: Prior wildfire yields more opportunities for stopping fire spread than fuel treatments. However, less area must be treated than burned naturally before fire managers consider it a barrier. Implications: This study helps direct policy towards expanding useful barriers.
WF25051 Accepted 04 August 2025
© CSIRO 2025
