Biodiversity sensing: A NEON-based approach toward a regional understanding of spatial vegetation diversity in northeastern U.S.

Link To Share This Poster: https://cdmcd.co/Rrm9vy
Live Discussion Link: https://cdmcd.co/3wqGX7

Di Yang, Wyoming Geographic Information Science Center, University of Wyoming, Laramie, WY, Jessica Mitchell and Ryan Rock, Spatial Analysis Lab, University of Montana, Missoula, MT, Martha Farella, O'Neill School of Public and Environmental Affairs, University of Indiana, Bloomington, IN, Tao Liu, College of Forest Resources and Environmental Science, Michigan Technological University, Houghton, MI, Nancy F. Glenn, Geosciences, Boise State University, Boise, ID

Background/Question/Methods
Understanding the spatial distribution of vegetation diversity is crucial for ecological, conservation, and sustainability-related applications; however, continental to global scale plant diversity mapping based on occurrence data is biased by uneven sampling. The increasing complex patterns of vegetation diversity across scales have recently been recognized, but understanding is limited. Patches of biodiversity can reflect areas where patterns of species diversity are mechanistically related to each other as represented by environmental variables. The National Ecological Observatory Network (NEON) is the first open-source continental-scale ecological observatory that combines in-situ field observations with advanced airborne remote sensing technologies. The availability of multi-scale ecosystem observations provides essential opportunities to pursue new types of biodiversity surveys that can help answer mechanistic and functional ecology questions. The availability of field data that can be combined with NEON site-wide coverages of Lidar and hyperspectral remote sensing provides a framework for developing and validating scalable vegetation diversity mapping across sites in the Eastern US that span regional gradients.
Here, we calculated biodiversity metrics (total number of species, Simpson’s Index, beta diversity) at the site-level using Woody Vegetation Plant Structure field datasets from NEON Terrestrial Observational Sampling (TOS). We also developed a robust workflow that considers Bidirectional Reflectance Distribution Function (BRDF) corrections, which greatly improved the model performances. Biodiversity calculations were treated as response variables and merged in Random Forest models with predictor variables derived from hyperspectral and lidar data acquired from the NEON Airborne Observation Platform (AOP) platform to map vegetation diversity patterns.
Results/Conclusions
Across NEON sites at the 1-m scale, lidar variables (e.g., elevation, roughness) were consistently the strongest predictor of biodiversity indices. The relative importance of optical variables varied across sites and provides evidence of different drivers of biodiversity patterns for each site. The overall accuracy of all NE NEON sites ranges from 78.21% in Mountain Lake Biological Station (MLBS) to 93.82% in the Great Smoky National Park (GRSM). For macroscale pattern investigations, we have evaluated Simpson Index patterns summarized at the site level and found a pattern of decreasing species diversity and resilience along a south-north gradient.
The incorporation of open-source data through the NEON portal, enables our effort to test predictive modeling for regional upscaling effort highly transferrable and repeatable. Moreover, hyperspectral and Lidar mosaics at the NEON site level is a scaling bridge that allows us to develop upscaling techniques by relating high-resolution airborne remote sensing variables to coarser satellite Earth observations.