Submitted to: Ag Data Commons
Publication Type: Database / Dataset
Publication Acceptance Date: 5/3/2024
Publication Date: 5/29/2024
Citation: Woodruff, C.D., Clark, P. 2024. Johnston Draw high resolution pre-fire vegetation map 2023. Ag Data Commons. https://doi.org/10.15482/USDA.ADC/25684290.v1.
DOI: https://doi.org/10.15482/USDA.ADC/25684290.v1

Interpretive Summary: Mapping vegetation is the first step to quantifying pre-fire conditions, but many maps often suffer from moderate spatial resolutions and generalization of the complex vegetation types. We trained a random forest to model a 10 classes vegetation map specific to the Johnston Draw area at a high resolution (0.5-meter spatial resolution) using two pansharpened 8-band WorldView 2 images: one image collected at peak greenness and the other collected two weeks prior to a prescribed fire. We demonstrated map accuracy of 83.3% when validated using 54, 30-meter diameter, plots selected to represent the following dominant vegetation types: deciduous/riparian (classes were collapsed into a single class for validation), living juniper, dead juniper, sagebrush, mixed low sage and bunchgrass, bitterbrush, and annual grasses. Megafire (>100,000 acres) frequency is increasing in the Great Basin and high resolution specific vegetation maps are essential to mitigate fire potential by classifying fuel types for fuel reduction management, analyzing post fire damage, and identifying post fire rehabilitation site selection which was a $1.5 billion expense in 2023 including suppression costs (Mitigation, 2023).

Technical Abstract: The variability of vegetation in rangelands can be over generalized in spatial representation and vegetation types mapped by moderate resolution vegetation maps. A high resolution (<1 meter), site specific, vegetation map may better represent the diversity and spatial complexity of rangelands – a necessity for analyzing pre-fire conditions. We pansharpened two 8-band VNIR Worldview 2 scenes to map pre-fire vegetation in Johnston Draw in the Reynolds Creek Experimental Watershed in Southwest Idaho. The two Worldview 2 scenes represent peak greenness (June 14, 2023) and pre-fire (September 23, 2023) conditions with spatial resolutions of 50 centimeters and 42 centimeters, respectively. A prescribed fire burned the area on October 6, 2023. We trained a pixel based random forest classifier to map 10 site specific vegetation classes at a 50-centimeter spatial resolution. We applied a majority filter to remove speckling. Map accuracy was 83.3% when validated using a test set of 54, 30-meter diameter, plots selected to represent the following dominant vegetation types: deciduous/riparian (classes were collapsed into a single class for validation), living juniper, dead juniper, sagebrush, mixed low sage and bunchgrass, bitterbrush, and annual grasses. Barren and water classes were not validated. Training data was developed through a combination of site visit based knowledge and a training set of 30-meter diameter dominant vegetation class plots. The 1.5-billion-dollar cost of fire prevention, suppression, and restoration is stretched thin over the vast area of wildfire occurrence, where site-specific high-resolution vegetation maps are essential to mitigate fire potential and address post fire recovery.