Quantifying rangeland fractional cover in the Northern Great Basin sagebrush steppe communities using high-resolution unoccupied aerial systems (UAS) imagery

Satellite products of fractional vegetation cover are often used to manage rangelands. However, they frequently miss the details of heterogeneous landscapes. The use of unoccupied aerial systems (UAS) to produce high spatial resolution rangeland fractional cover maps could fill that gap at local scales. We evaluated the capabilities of UAS imagery for mapping rangeland fractional vegetation cover in sagebrush steppe communities of the Northern Great Basin, USA. We applied segmentation and machine learning models for image classification, and established regression functions with field-measured herbaceous cover and multiple spectral indices to quantify herbaceous fraction in bare/herbaceous mixed polygons. Finally, we conducted a correlation analysis to compare UAS-derived rangeland fractional cover with satellite-derived products. Overall classification accuracies for the UAS-derived rangeland fractional cover maps were high (89–98%). Modified Soil Adjusted Vegetation Index was the most important spectral index for predicting photosynthetic classes and including Brightness Index in a multiple index approach improved classification of shadows and bare ground. Regression models effectively estimated herbaceous fractions within bare/herbaceous mixed polygons with high accuracy (R 2 = 0.71–0.88). UAS-derived rangeland fractional cover estimates captured within-site variability, while satellite-derived products did not, specifically for herbaceous and litter. This study demonstrated a workflow using UAS and intensive ground sampling for estimating rangeland fractional cover in sagebrush communities. We found a disagreement between UAS-derived and satellite-derived fractional cover products at two sagebrush communities in the Northern Great Basin. We recommend the application of UAS when estimating rangeland fractional cover at local scales.

Перевод пока недоступен