Methods for quantifying the spectral variation within spectral libraries and imagery
Henry Frye, University of Connecticut, henry.frye@uconn.edu
Heidi M Dierssen, University of Connecticut, heidi.dierssen@uconn.edu (Presenter)
A central task for ecological remote sensing is to make inferences about organisms and the communities they comprise based on their spectral reflectance. Most spectral analyses in the ecological context rely on the assumption that the spectral signatures of groups like species or community types are spectrally separable, i.e., spectrally distinct despite variation arising inherently or from external sources. Assessing the spectral separability of taxonomic groups and communities can be an important exploratory step in spectral mixing and classification analyses and can provide inference into the underlying biological and ecological properties of the groups under study. For instance, organisms with similar spectral signatures are likely to have similar physical and chemical traits.
Despite the practical importance of spectral separability in ecological contexts, there are few explicit approaches for exploring spectral separability. Further, it is difficult to assess how comparable other separability results are given limited sources aggregating such values. Here, we propose two possible approaches, specifically for hyperspectral data, using modifications of existing metrics. The first approach examines a pairwise distance matrix within and between groups based on spectral angle values. The second approach is based on apparent wavelength, a weighted harmonic mean of reflectance values. We apply these metrics over five spectral libraries containing a total of 1,225 spectral measurements that span a wide breath of organisms in both terrestrial and marine domains. We document the amount of spectral variation within species, genera, and communities across these libraries to serve as a reference for spectral separability within organisms.
Associated Project(s):
Poster Location ID: 2-13
Presentation Type: Poster
Session: Poster Session 2
Session Date: Wed (May 10) 5:15-7:15 PM
CCE Program: BDEC