Discerning individual plant species in satellite imagery using field-measured hyper-spectral reflectance data at the Blandy Experimental Farm, north-central Virginia

Background/Question/Methods

Invasive plant species can threaten native species composition, community biodiversity, and ecosystem function. Knowing the location and dispersal of non-native invasive species that were accidentally or intentionally introduced would allow us to avoid such introduction and dispersal in the future. Locating them remotely would save resources otherwise needed for extensive field surveys. The ability to map individual species using remote sensing depends on differences in their spectral signatures, and detectability of those differences from remote sensing platforms. Ground-level spectral measurements were obtained for five non-native invasive plant species at the Blandy Experimental Farm (BEF) in north-central Virginia, using an Analytical Spectral Devices FieldSpec 3 spectro-radiometer during the summer of 2011. Spectra were analyzed to determine ground-level species differences among Ailanthus altissima (tree of heaven), Rhamnus davurica (Dahurian buckthorn), Galium verum (yellow bedstraw), Cirsium arvense (Canada thistle), and Carduus acanthoides (spiny plumeless thistle).

Results/Conclusions

A discriminant analysis using reflectance, first derivatives, and second derivatives of certain wavelengths selected by principal components analysis (PCA) revealed species differentiability, except for confusion among the two thistles. The four reflectance values selected from the PCA were subsequently used to locate these species in a pan-sharpened QuickBird satellite image. Spectral angle mapping was conducted using average field spectra of target species and non-target surfaces like gravel, pavement, plant litter, and soil. A multiple endmember spectral mixture analysis was also conducted, using multiple spectra to represent each species, and allowing a pixel to be represented by more than one species. These techniques allowed mapping of these species apart from the surrounding environment, but similar potential locations for multiple species from the analyses suggest the need for more bands for species discrimination from a remote sensing platform. Frequencies of reflectance values at each band for the BEF image and each species reveal similar reflectance values in the visible bands, but variation in the near-infrared region, which may be exploited for species differentiation via satellite.