The Wind Hazard Identification, Surveillance, and Prediction via Radar (WHISPR) system: Leveraging machine learning to enhance situational awareness of wind shifts in active wildland fire response

Convective weather cells and associated outflow winds, such as gust fronts, can rapidly alter a fire’s propagation speed and direction, leading to firefighter entrapment. One well-documented disaster occurred during the Yarnell Hill Fire on June 30, 2013, when a gust front from a nearby thunderstorm reversed a fire’s direction and dramatically increased its propagation speed, causing 19 firefighters to become entrapped and perish. While the Yarnell Hill incident was an extreme case, sudden wind shifts routinely lead to firefighter entrapments, injuries and fatalities. In this presentation we'll describe how Next Generation Weather Radar (NEXRAD) data can provide the high spatial resolution (~250 m) and rapid updates (every 4-6 minutes) needed to identify such wind hazards and actively alert stakeholders on the ground.

We curated a NEXRAD-based dataset of >300k labeled gust front events from 142 NEXRAD sites spanning CONUS and subsequently developed a deep learning algorithm to identify and track wind hazard events. We investigate the range limitations of NEXRAD-based detection of low-altitude gust fronts and show how geostationary weather satellite imagery may fill gaps in radar coverage in mountainous western CONUS. Finally, we will demonstrate a real-time operational prototype which can aid in wind hazard awareness on active fires.