Basal area loss from fire using field-calibrated remote sensing refines western US fire severity measurements

Fire is a critical ecological process and the spatial patterns of fire severity have profound effects on forest structure and resilience. Cost-effective fire severity measurements are critical for informing post-fire restoration strategies and for improving our understanding of the patterns, drivers, and ecological consequences of wildfires. Remote sensing technology has advanced burn severity mapping through space and time. However, different severity mapping methods vary in interpretability, scalability, generalizability, and how well they align with field measurements. One common meaningful metric of fire effects on forests is percent basal area loss, but existing methods present challenges to researchers and managers due to the lack of region-specific field reference data and a scalable framework for automatic mapping. To address these issues, we compiled field reference data from 123 fires in forests across the western US to calculate the proportion of fire-induced basal area loss. We then tested spatially cross-validated machine learning models concurrently with hyperparameter tuning to select a skillful, parsimonious model from an extensive candidate set comprising various combinations of climatic, topographic, and surface reflectance-based covariates. Model performance varied geographically, highlighting that this new method currently provides a critical resource for managers in some regions such as forests in Oregon and Washington, while providing a roadmap for researchers in other regions to continue adding data to the project to improve fire severity mapping efforts. To support fire severity assessments, we created a tool in Google Earth Engine to produce estimates of percent basal area loss for any fire perimeter in the western US.