COS 274-5 - Airborne species mapping enhanced by AI-guided sampling for rare species allows enhanced landscape analysis of patterns and drivers of structural, functional and species diversity and composition

Abstract: Biodiversity is changing rapidly as species composition is shifting in response to local to global environmental change. A deep understanding of the processes driving these changes requires identifying drivers and patterns across multiple scales and for different components of biodiversity. Combining field data with remote sensing can provide large scale estimates of biodiversity related to species, structure, and functional traits and allow understanding their environmental and management drivers across scales. The goals of this study were to use algorithm (AI)-guided sampling to increase representation of rare species and reveal pockets of undetected diversity in airborne biodiversity surveys of canopy trees and analyze the resulting biodiversity and composition patterns at a landscape scale. We asked how structure, function and species composition and diversity varies with each other and with topography and prescribed fire frequency.



For the Ordway Swisher National Ecological Observatory Network (NEON) site in Florida, we used NEON vegetation structure field data to train a baseline species classification of NEON airborne image data to generate predictions for all detected trees. The baseline species predictions and uncertainties guided additional field data collection, which increased the number of rare species, the amount within-species trait variation, and improved the classification model. Using the airborne-based predictions of species, traits and structure for all trees, we examined how topography and fire regime impact relative species abundance, intraspecific trait variation and species, structural and functional biodiversity.



Improving upon a baseline model that could accurately predict 3 species, our improved model predicted 14 tree species with 70% accuracy. Functional and species diversity had weak positive correlations across the site. Structure metrics, such as mean canopy height and biomass, showed a negative relationship with functional diversity and no relationship with species diversity. The rarer hardwood species occur at low elevations with their abundance stratified by small elevation changes. In the dominant sandhill ecosystem type, more frequent prescribed fires is positively correlated with relative abundance of longleaf pine, but has more complex relationships with co-occurring oaks. Leaf mass per area of sandhill pines and oaks increased with fire frequency and elevation. In conclusion, biodiversity patterns at the landscape scale differ substantially depending on if structural, functional and species diversity is considered. Intraspecific variation in functional traits is related to both topography and management. Future work will apply these methods and analysis to sites across the NEON network to allow cross-site comparison of diversity and composition and their drivers.