Bio
I am a wildfire and geospatial researcher with a particular passion for artificial intelligence. I use machine learning, deep learning, computer vision, and large geospatial datasets to explore how AI can help us understand wildfire growth, evaluate suppression operations, interpret remote-sensing imagery, and support better fire-management decisions. My work brings together satellite and airborne imagery, aircraft-tracking data, fire-progression records, and other complex spatial information to develop practical AI-driven approaches to wildfire research.
My perspective is shaped by both technical research and fire-service experience. As a volunteer firefighter, I am especially interested in AI that is scientifically rigorous, transparent, and useful to people making decisions on the ground. Outside of research, I create glass art and handmade marbles through Woodlot Pyro Crafters, which gives me another way to explore complex patterns, color, and light.
Selected publications
High-level overviews of published and current work.
Open aircraft-tracking data were used to compare CAL FIRE and U.S. Forest Service aviation response across 7,214 California wildfires. The work shows how standardized ADS-B records can reveal cross-agency response patterns and support better resource coordination during initial attack.
View full paperThis study tests whether multimodal AI can scan high-resolution aerial imagery for ponds that appear accessible to fire apparatus. The workflow identified candidate water sources for local review while showing how AI screening can support field verification and firefighter judgment.
View full paperNational Infrared Operations perimeters were organized into a consistent, reusable dataset that captures daily wildfire growth across the western United States. The resulting resource supports fire-behavior modeling, risk analysis, ecological assessment, and operational research.
View full paperA neural network was trained on a decade of management records from 11 Colorado national forests to study how treatment types have been selected. Wildfire risk, expected fire behavior, and proximity to structures emerged as important influences across six treatment categories.
View full paperAircraft trajectories from ADS-B were paired with verified suppressant-drop records to train sequential deep-learning models. The best model identified drop events with an F1 score of 0.922, demonstrating a path toward more complete, near-real-time awareness of aerial suppression activity.
View full paperEmployment histories for Interagency Hotshot Crew members were analyzed to understand why experienced firefighters stay or leave. Greater workload and cumulative experience were associated with stronger retention, while wages in alternative occupations were not a significant predictor.
View full paperUAV hyperspectral imagery was evaluated as a way to find bear-damaged redwoods before the damage becomes obvious from the air. Healthy trees could be separated from trees with older damage, while very recent damage remained difficult to distinguish, helping define the next steps for earlier detection.
View full paperFine-scale UAV imagery and photogrammetry were used to measure dune movement and map vegetation at California’s Manila Dunes. The study connected dune stability with trails, topography, and plant cover, providing a baseline for managing recreation and invasive species in a dynamic coastal system.
View full paperGlass work · color studies · small worlds
Add marble photos to the MARBLES folder to show them here.