Percent Shrub Cover for the Desatoya Mountains, Nevada, Derived from Worldview-2 and UAS Imagery (2018)

Accurate maps of habitat availability for Greater Sage-grouse (Centrocercus urophasianus) across broad extents are of paramount importance to conservation efforts in sagebrush ecosystems across the Great Basin, particularly for habitat assessments and mitigation efforts. However, the ability to model sage-grouse habitat at fine spatial scales necessary for microhabitat assessment is constrained by the spatial and spectral resolution of most measurements remotely sensed of vegetation composition. Fractional approaches that scale plot-based measures of vegetation composition to satellite imagery that yield estimates of percent cover at relatively coarse resolution (e.g., 30 x 30 meter pixels) are well suited for regional and local estimates at relatively broad spatial scales (e.g., 3rd order macrohabitat selection and availability), but precision at individual pixels that could represent finer selection patterns by sage-grouse (e.g., 4th order microhabitat selection) is often poor. Here, we present results from a proof-of-concept study at a sagebrush site in central Nevada, where estimates of shrub cover in training plots derived from ultra-high-resolution (near 3 centimeters) imagery collected with uncrewed aircraft systems (UAS) are applied to multi-spectral WorldView-2 scenes using regression tree (RT) models. The approach yields fractional estimates of shrub cover at a 2 meter resolution that more closely correlate with actual field measurements that reflect finer selection patterns of sage-grouse compared to other widely available fractional products most suited for macrohabitat applications. An advantage of the UAS-RT approach is that large volumes of training data can be collected rapidly from UAS compared to traditional ground-based vegetation surveys, which could ultimately yield previously lacking estimates of microhabitat availability across macrohabitat or landscape level extents. Our map was highly accurate (mean absolute error of 7.28) compared to the measured accuracy of other available products, captured land cover and shrub variability in high-resolution, and can be used to develop predictive models of sagebrush and sage-grouse response to management actions.