Improving Wildfire Prediction with Cutting-Edge Satellite Imagery and Artificial Intelligence (AI)


Fire suppression personnel cannot predict the spread and intensity of an emerging wildland fire without knowing how dry the fuels are that lie in its path. We use models to help predict this behavior, but for the model to accurately simulate a wildfire, it requires up-to-date information about the conditions. This includes the local weather and terrain, as well as the characteristics of the plant matter that provides fuel for the flames. Is it dead or alive? Is it moist or dry? Standing or fallen? Until recently, the available data were very outdated and misleading, which hampered efforts to predict fire behavior.


Modeling wildfires with precision is dependent upon the quality and truth of the inputs. However, the available moisture fuel data used during the East Troublesome Fire was out of date, causing predictions to underestimate growth and spread. This is because collecting this information is painfully time consuming and inefficient, requiring manual inputs from human observers on the ground. To remedy this, our researchers turned to the Sentinel satellites, deployed by the European Space Agency’s Copernicus program. Sentinel-1 provides information about surface texture, which can  identify vegetation type. Sentinel-2 provides information about plant health as derived from its "greenness." Our scientists fed the satellite data into a machine learning (AI) model known as a “random forest” survey that estimates tree mortality. In fact, the AI model was able to accurately update the LANDFIRE fuel data, defining the majority of the fuels previously categorized as “timber litter” or “timber understory” to “slash blowdown,” the term used for heavy tree mortality. When WRF-Fire first simulated the East Troublesome Fire using the unadjusted LANDFIRE fuel dataset, it substantially underpredicted the amount of area the fire would burn. When the model was run again using the adjusted dataset, it predicted the burned area with far greater accuracy, indicating that the dead and downed timber accelerated the fire’s spread much faster than if the trees had still been alive.


The AI model is currently designed to update existing fuel maps, and it can do so in a matter of minutes. But the success of the project also holds promise of using an AI system to produce and update fuel maps from scratch for large regions at risk from wildfires.

This research is part of a larger trend of investigating AI applications for wildfire, including using AI to more quickly estimate fire perimeters. NCAR researchers are also hopeful that AI may be able to help solve other persistent challenges for wildfire behavior modeling. For example, AI may be able to improve our prediction of the ember properties produced by a fire, as well as the likelihood that those embers could cause spot fires. The better we are at predicting fire behavior, the more support we can deliver to fire suppression management to plan their attack strategies and protect their personnel from harm.

Advancing Wildland Fire Forecasting - NCAR | RAL


Please direct questions/comments about this page to:

Amy DeCastro

Assoc Scientist II