Explaining patterns of biodiversity across spatial scales with improved detection and attribution of disturbances

Abstract:

Disturbance regimes can strongly influence geographic patterns of biodiversity. The types of disturbances and their frequencies can have varying impacts on different dimensions of biodiversity and taxonomic groups, and their influence can also vary with spatial scale. Yet disturbance layers are lacking at sufficiently high spatial resolution and extent to uncover these relationships with biodiversity. We present disturbr - a flexible R package to detect and attribute disturbances from any sensor and user-defined time series. disturbr leverages spatial information from surrounding pixels and relies on random forest classifiers to improve detection and attribution of multiple disturbances across different ecoregions. We show how disturbr can be applied to Landsat imagery across National Ecological Observatory Network (NEON) sites in different ecoregions, and we evaluate spatial relationships between disturbances and dimensions of biodiversity (taxonomic, functional and phylogenetic) for several taxonomic groups including woody plants, small mammals, fish, and birds. In addition to providing an open and reproducible workflow with which to apply disturbr anywhere on Earth, we are using disturbr to generate continuous 30-m geospatial disturbance layer data products across the United States.