This project pioneers the use of very-high-resolution remote sensing and advanced deep learning to map individual dead trees across the Western United States from 2010 to 2025. This innovation addresses a critical gap in wildfire science by capturing tree mortality at unprecedented resolution and scale, reducing one of the largest uncertainties in fire modeling.
By integrating cutting-edge computer vision models with ecological knowledge, this team will create a generalizable framework that improves accuracy while minimizing the need for intensive manual labeling, setting a new standard for ecological monitoring.
The impact of this work includes incorporating spatially explicit dead tree datasets into fire models - enabling more precise predictions of wildfire behavior, directly supporting forest managers, the timber industry, and carbon cycle researchers.
The co-development of a web-based data portal with USGS and national park stakeholders ensures practical application, allowing managers to identify high-risk areas, prioritize interventions, and anticipate fire severity. By bridging AI innovation with ecological modeling, the project strengthens wildfire resilience strategies and lays the foundation for more advanced ecological studies.
