An open-source workflow for scaling burn severity metrics from drone to satellite

Abstract

Wildfires are increasing in size and severity across much of the western United States, exposing vulnerable wildland-urban interfaces to post-fire hazards. The Mediterranean chaparral region of Northern California contains many high sloping watersheds prone to hazardous post-fire flood events and identifying watersheds at high risk of soil loss and debris flows is a priority for postfire response and management. Uncrewed Aerial Systems (UAS; aka drones) offer post-fire management teams the ability to quickly mobilize and survey burned areas with very highresolution imagery (~1 cm), facilitating emergency management and post-fire hazard assessment. However, adoption of this technology by hazard response teams may be hindered by complicated workflows for UAS data acquisition, image processing and analysis, which often involve the use of costly proprietary software. We present an open-source workflow using mature Geographic Information Systems (GIS) software and Python packages in a Jupyter Notebook environment that guides users through classification of true-color UAS imagery to generate high resolution burn severity maps which can then be scaled across larger watersheds using Sentinel-2 normalized burn ratio (NBR) images. Soil burn severity classifications using a weighted brightness (WB) image and Char Index (CI) generated from UAS imagery were validated with in-situ data and random stratified points, resulting in the CI having the highest overall accuracy of 87.5%. CI also displayed a marginally stronger relationship over the WB with the post-fire Sentinel-2 NBR, R2 = 0.79 and R2 = 0.78 respectively. Open-source methods offer the unique opportunity to standardize GIS workflows, promoting replication through transparency, while improving the user's understanding of scientific GIS functionality.