Johnston Draw (Idaho) high resolution burn severity map 2023

Authors: BYRNE, AKIRA - Boise State University; Woodruff, Craig; ENTERKINE, JOSHUA - Boise State University; Clark, Patrick; HUBER, DAVID - University Of Texas - El Paso

Submitted to: Ag Data Commons
Publication Type: Database / Dataset
Publication Acceptance Date: 2/4/2025
Publication Date: 2/7/2025
Citation: Byrne, A., Woodruff, C.D., Enterkine, J., Clark, P., Huber, D. 2025. Johnston Draw (Idaho) high resolution burn severity map 2023. Ag Data Commons. https://doi.org/10.15482/USDA.ADC/28330952.v1.
DOI: https://doi.org/10.15482/USDA.ADC/28330952.v1

Interpretive Summary: Accurate mapping of rangeland burn severity is the first step towards managing and mitigating post-fire consequences, but the spatial resolution of freely-available, remotely-sensed products are often too coarse to effectively represent the complexity of burned rangelands. We trained a support vector machine to classify burn severity over Johnston Draw (a 1.8 square kilometer prescribed fire study area in southwestern Idaho) at very high spatial resolution (0.5-meter) using post-fire pan-sharpened 8-band Worldview 3 imagery, co-registered with a complimentary pre-fire vegetation map. Burn severity map accuracy was quite high at 88.3% when validated at 1000 randomly distributed points within the study area. In 2023 alone the Bureau of Land Management treated fuels in over 1.2 million acres of rangeland, a quarter of which was treated with prescribed fire, and managing postfire consequences like soil erosion and vegetation recovery requires high resolution burn severity mapping to inform mitigation on these vast acreages.

Technical Abstract: Remote sensing of burn severity in complex rangelands is often too coarse to represent burn complexity. The goal of this work is to develop a high spatial resolution (0.5 meter) burn severity map for a prescribed fire that took place in Johnston Draw, Idaho on October 6, 2023, to compliment an existing high resolution pre-fire vegetation map. We used a pansharpened 8-band WorldView 3 scene collected on October 8, 2023, to classify burn severity. Spatial offset discrepancies between the pre-fire vegetation map and the post fire WorldView 3 scene required us to co-register the images. We used a second order polynomial with 20 ground control points to co-register the image. We selected distinctive terrain features as ground control points in the pre-fire WorldView 2 imagery (June 14, 2023) used in the pre-fire vegetation map and matched the features in the post fire WorldView 3 scene (October 8, 2023). Co-registration Root Mean Squared Error (RMSE) was 0.99 and below the acceptable threshold of 1.0 (Jensen, 2016). A first order support vector machine was trained to model burn severities as ash color – black, gray, and white ash representing low, moderate, and high severity burns, respectively. Unburned areas were first modeled into vegetation classes, then aggregated into a single unburned class. Training data was developed through heads up digitizing. To validate the resulting map, 1000 points were randomly generated and burn severity was attributed using the WorldView 3 scene as reference. Mapping overall accuracy was 88.3%. Timely, high resolution, and spatially explicit burn severity maps are the critical decision support tools land managers require to manage and monitor the growing threat of megafires (>100,000 acres) that are increasing in frequency in the Great Basin.