Abstract
Connecting the spectral variability in aerial remote sensing imagery to plant communities across time and space has great potential for conservation efforts. Variability among species at different points in the growing season, across years, across platforms, and across scales can elucidate the best times and approaches to detect invasive plant species for management efforts. Spectral variability within species can be used to better understand functional trait variation and ecosystem functioning through remote sensing. Variability in images can also be used to understand plant community dynamics across time and space.
This dissertation explores the temporal and spatial variability in species-specific spectral signatures and vegetation communities in northwestern Virginia at the biological field station Blandy Experimental Farm, which contains 80 ha of fields in various stages of succession with abundant invasive plant species. The first two chapters explore the remote detection of three invasive plant species that outcompete and displace native plants and that are of interest to land managers in Virginia and much of the U.S., Ailanthus altissima (tree of heaven), Elaeagnus umbellata (autumn olive), and Rhamnus davurica (Dahurian buckthorn). First, within a single growing season using fine resolution drone-based imagery, then across multiple growing seasons using aerial hyperspectral imagery collected by fixed-wing aircraft by the National Ecological Observatory Network (NEON), a different platform, sensor, and spatial resolution.
The results demonstrate that both UAV and NEON (fixed-wing aircraft) hyperspectral imagery can be used to detect the three species of interest, however, accuracies varied over time and were greatest when algorithms were produced using in situ data (e.g. from the same platform, on the same date). Drone-based algorithms were most consistent across the growing season for E. umbellata, while NEON-based detection was least consistent. NEON-based detection of R. davurica was most consistent across growing seasons and platforms. A. altissima algorithms were also relatively consistent across years but used different spectral features in the drone-based and NEON-based algorithms. These results demonstrate the usefulness of flexible sampling times within and across growing seasons.
The last two chapters explore the partitioning of spectral variability at different scales and their ecological implications. First, at multiple organizational scales (at the leaf, canopy, species, and community levels) within a single growing season in drone-based images, then at multiple spatial scales (within and among plots) by pairing field surveys of species composition and NEON-based images. Within a growing season, spectral variability in biochemical-associated spectral regions within individual canopies and among canopies of the same species exceeded among-species variability, suggesting a lack of agreement with the SVH as biochemical traits become increasingly variable at finer organizational scales as leaves mature over a growing season. Spectral variability within plots was greater in biochemical traits than in structural traits, but among-plot spectral variability was greater in structural traits than biochemical traits, suggesting vegetation communities are stable in different traits at different scales. These violations of the SVH were driven by both spatial and temporal factors.
This dissertation demonstrates that species-based assumptions about traits and spectra are not necessarily accurate across space and time and the importance of considering the wide range of spectral and trait variability within a species. Understanding trait variation at different scales and times can facilitate answering major questions in community ecology to further the understanding of plant communities and ecosystems. Spectroscopy can be used to this end and will benefit from increasingly available hyperspectral airborne data and new satellite missions.
