Accuracy Assessment of National Land Cover Database Shrubland Products on the Sagebrush Steppe Fringe

Remotely sensed data products depicting physical and ecological attributes of a landscape are becoming invaluable tools in wildlife and rangeland management. However, if such geospatial tools and data layers are to be used in management, their accuracy and appropriateness for such use needs to be vetted and validated. We assessed accuracy of two National Land Cover Database (NLCD) shrubland products for use in western South Dakota—percent sagebrush and sagebrush height—by comparing them to ground-truthed data. Western South Dakota sagebrush communities are an ecotone between sagebrush (Artemisia spp.) and grassland. This ecotone is typified by shorter- and lower-density sagebrush than interior sagebrush steppe ecosystem. This distinction could make it difficult to remotely detect and map sagebrush in this area. We determined NLCD correlations with ground estimates of sagebrush canopy cover (r = 0.17) and sagebrush height (r = 0.40). The NLCD percent sagebrush accurately predicted sagebrush presence ∼73‒76% of the time once resampled to 100-m pixels and 50-m mean values, respectively. Cohen’s kappa values were estimated to determine if the ground-truthed and remoted-sensed data were in agreement when determining sagebrush presence. Kappa values were 0.26 ± 0.06 and 0.28 ± 0.06 for mean values within 50-m and resampled 100-m pixels, respectively, indicating a “fair” level of agreement between the ground-truthed and remote-sensed data types when determining presence of sagebrush. The NLCD data sufficiently described the presence of sagebrush in South Dakota, which is useful for estimating geographic distributions of sagebrush obligate species, species distribution models in which presence or absence of sagebrush is of interest, or mapping the occurrence of sagebrush in South Dakota. Inaccuracies of the NLCD shrubland products in predicting sagebrush height and sagebrush canopy cover may limit their utility as continuous variables in species distribution models, habitat selection, and suitability models or when assessing rangeland quality in South Dakota.