Thu, Aug 18, 2022: 5:00 PM-6:30 PM
ESA Exhibit Hall
Background/Question/Methods: Image spectroscopy is a promising tool in the endeavor to measure plant biodiversity across large spatial extents at frequent intervals. The probable success of remote sensing efforts for biodiversity assessment will depend on the degree to which species or communities can be separated from each other across spectral scales, ranging from leaves through canopies to aerial images. The degree of separability can then be used to inform classification decisions at the scale of remotely collected imagery, e.g., angle thresholds in spectral angle mapping. This study examines the potential for spectral separability to identify three dominant species in a relatively low-diversity system, a tidal mangrove swamp within Florida’s Long Key State Park. Our data leverages spectral samples measured on leaves and canopies (ASD FieldSpec 4 spectrometer 1 nm sampled; 350nm - 2500nm) with hyperspectral imagery (210 bands sampled; 452nm - 1045nm) collected with the Portable Remote Imaging Spectrometer (PRISM) in 2014. We introduce novel methods to compare the magnitude of interspecific and intraspecific variation with hyperspectral reflectance data.
Results/Conclusions: Using a spectral angle-based variation method, we found that intraspecific spectral variation increased as spectral scales increased–an anticipated result given known spectral degradation at increasing spectral scales. At the leaf and canopy scale, interspecific spectral angle variation was greater than intraspecific variation, indicating spectral separability between species; however, these differences largely disappeared with spectra extracted from PRISM imagery. We also examined commonly used spectral indices such as Normalized Difference Vegetation Index (NDVI) and Photochemical Reflectance Index (PRI). These demonstrated more consistent separation across scales and species. Based on our analyses, we describe a decision tree workflow with either spectral indices or angle thresholds that can be used to classify dominant species within PRISM imagery. While mangroves are an ecosystem with low species richness, our results demonstrate how one can use ground-based measurements to make inferences on species identification within hyperspectral imagery. This is an important proof-of-concept for biodiversity remote sensing methods that attempt to measure biodiversity through the direct detection of plant species.
Results/Conclusions: Using a spectral angle-based variation method, we found that intraspecific spectral variation increased as spectral scales increased–an anticipated result given known spectral degradation at increasing spectral scales. At the leaf and canopy scale, interspecific spectral angle variation was greater than intraspecific variation, indicating spectral separability between species; however, these differences largely disappeared with spectra extracted from PRISM imagery. We also examined commonly used spectral indices such as Normalized Difference Vegetation Index (NDVI) and Photochemical Reflectance Index (PRI). These demonstrated more consistent separation across scales and species. Based on our analyses, we describe a decision tree workflow with either spectral indices or angle thresholds that can be used to classify dominant species within PRISM imagery. While mangroves are an ecosystem with low species richness, our results demonstrate how one can use ground-based measurements to make inferences on species identification within hyperspectral imagery. This is an important proof-of-concept for biodiversity remote sensing methods that attempt to measure biodiversity through the direct detection of plant species.