Understanding biodiversity patterns, notably community composition turnover (or beta diversity), is critical to improving biodiversity conservation and human well-being due to their impacts on ecosystem function and services. In eastern forested landscapes, spatial differentiation in breeding bird communities can be determined by transitions in structural and compositional attributes of forest ecosystems, all factors targeted by management activities, typically through even or uneven aged practices, and threatened by intensifying land use and climate change. We used an approach of visualizing and predicting ordination position in an indirect framework as a RGB color composite to map bird communities over a 5,258-km2 area in Southeast Ohio. We used non-metric multidimensional scaling (nMDS) to summarize community variation of 36 species among 304 point count stations collected over two field seasons onto three axes. Axes coordinates were related to a suite of remotely-sensed descriptors, including LiDAR-derived forest structure and topography and Landsat 8 spectral reflectance data, with Random Forests models. Our goal was to test the feasibility of such an approach, and to provide a spatial tool box in the form of a beta diversity map to resource managers that describes how management activities might affect distribution patterns and community dynamics of wildlife.
Results/Conclusions
The nMDS ordination converged with a stress value of 18.8. In the Random Forest models, the first axis was related to local forest structural variables (~ 40% of explained variation), the second axis was related to topography (15% of variation), and the third axis was related to landscape-scale composition and configuration of woody cover (3% of variation). Score coordinates were projected across the landscape and compiled into unique colors, visualizing beta diversity across our study area. Finally, for each focal species the Euclidean distance in SD units of their centroid to a given site was used to map species distributions of focal species, and its predictive performance was compared to more traditional habitat modeling approaches, including generalized linear models, using receiver operating characteristic curves. This community first approach had similar predictive performance as systemically modeling individual distributions of our three focal species but with the advantage of summarizing all variation in species occurrence in a single ordination. Our maps produced spatially-explicit estimates of community composition across our landscape, identified potential drivers of community change, and served as an efficient approach to summarize community variation to help inform sound landscape-scale conservation efforts both locally and broadly through forest management practices.