Our research is exploring to improve the safety and resiliency of communities at the wildland urban interface (WUI). Billions of dollars are spent each year on protecting these communities. We are investigating how to reduce the ignitability of the built environment and perform rapid forecasting of wildland fire spread.
Firebrand Structure Ignition
Firebrands generated by wildfires ignite structures causing significant damage, but we don’t understand the conditions and mechanisms behind their ignition so we can design better construction.
We are conducting research to understand the fundamental heat transfer between firebrands and building materials to support designing improved, ignition resistant communities.
What is a firebrand? Firebrands are small pieces of partially burned vegetation or building material typically less than one inch long that can fly over a mile from the main part of a wildfire and ignite homes. It is difficult to determine where these “spot fires” occur, which usually results in a large amount of damage before firefighters arrive to extinguish them.
Wildland Fire Spread Forecasting
Knowing where a wildfire might be in the next 24 hours or a week is important to provide community awareness and assist emergency responders in fire planning. Forecasting wildland fires currently requires significant computational time making it not possible to predict where a wildland fire might be in the future before new satellite data is available.
Our group is developing a more detailed understanding of how wildfires spread and efficient ways to predict the behavior using historic fire data and deep learning. These efficient predictions can be used to support vegetation management to reduce wildfire severity and forecasting wildfires for use by communities and emergency responders.
The types and distribution of landcover (vegetation, roads, buildings, soil, etc.) have a significant impact on how wildfires spread; however, quantifying vegetation over the large regions is still a time consuming process that includes detailed surveys and years of processing.
We are exploring using multi-spectral and hyperspectral imaging from planes and satellites to train deep learning models to efficiently predict landcover. With these improved tools, we are attempting to reduce the process of quantifying landcover in days.