Ph.D dissertation defense: Brian Blaylock
Brian was nearly as happy after his thesis defense as when he was at a BYU football game in Provo. The Utes scored and later won!
High-Resolution Rapid Refresh Model Data Analytics for Wildland Fire Weather Assessment
Tuesday, August 13, 2019
Threats associated with wildland fires are exacerbated by weather conditions conducive to rapid fire spread, such as strong winds, high temperatures, and low humidity. Incident Meteorologists rely on a variety of in situ and remote observations as well as numerical weather prediction model output to assess the potential influence atmospheric conditions will have on the fires. With the accelerating accumulation of available meteorological data, efficient computational solutions are needed to process, archive, and analyze the massive datasets in ways useful to Incident Meteorologists. This work demonstrates how object-based storage technology is used to efficiently archive multiple years of the High-Resolution Rapid Refresh (HRRR) model output—a convective-allowing operational forecast system that produces 0–18 h forecasts. The archive developed for this work now supports air quality and wildland fire research activities at the University of Utah and hundreds of other researchers. The historical HRRR model output was used to provide information on model behavior and skill that may be of great value to Incident Meteorologists.
An extensive set of empirical cumulative distributions for near-surface variables based on three years of model analyses was efficiently computed on the Open Science Grid—a high-throughput computing resources. The cumulative distributions are used to evaluate techniques that may be appropriate to discriminate between typical and atypical atmospheric conditions in a historical context for situational awareness of hazardous weather conditions like strong winds. Also, the skill of HRRR model lightning forecasts—which are important to firefighting operations because of the potential for convective outflows—was evaluated by comparing forecasted lightning threat with lightning observations from the Geostationary Lightning Mapper. Based on the fractions skill score, HRRR lightning forecasts skill decreases rapidly after the first two hours of model integration with better skill for longer lead times in the afternoon and evening hours in the western and central United States. With case studies of recent wildland fires, this dissertation illustrates how historical model data and objective evaluation of lightning forecast performance can help Incident Meteorologists identify hazardous weather conditions and interpret deterministic lightning forecasts from the HRRR model.
Supervisory Committee: Dr. John Horel, Dr. Jim Steenburg, Dr. John Lin, Dr. Erik Crosman, Dr. Philip Dennison