Global changes in biodiversity caused by climate change and environmental degradation have spurred the need to cost-effectively monitor shifting large-scale dynamics in plant species distributions and community composition. New remote sensing technology, in particular improved hyperspectral spectrometers, are fast overcoming technical limitations that have previously hindered these efforts. Thus, questions of what biologically relevant information can be extracted from leaf reflectance spectra need to be answered from the individual to the community/landscape levels. To address this, we sampled over 1,600 individuals in 8 community plots for their leaf reflectance spectra in the Greater Cape Floristic Region (GCFR) of South Africa. The GCFR is recognized as a global biodiversity hotspot with over 11,500 plant species, providing a model system for testing scalability using remotely sensed data in the world’s biodiverse regions threatened by changing conditions in the Anthropocene.
Conventional methods of analyzing hyperspectral information rely on techniques that sample hundreds of points along the wavelengths of a reflected light curve. This would be computationally prohibitive on a large scale and is limited mathematically in its inferences. In our analysis, we tested the use of a new and rapidly growing method in statistics, functional data analysis (FDA), in order to analyze family scale and regional differences in leaf reflectance spectra. Functional data objects were created using B-spline basis systems for plot averages and for three plant families that define vegetation types in the GCFR: Proteaceae, Ericaceae, and Restoniaceae. Variation was then compared using the functional first and second derivatives, phase-plane plots and contour plots of the bivariate variance-covariance surfaces.
Results/Conclusions
Analyses to date among the predominant families of the GCFR revealed significant differences in the first derivative plots that were not obvious in the spectra plots. Contour plots and phase-plane plots further confirmed this variation among families. Within families, the contour plots denoted high amounts of variation in the photosynthetically active region (PAR) and near-infrared region (NIR) of the Proteaceae while the Ericaceae and Restioniaceae had variation primarily in the NIR. Among the 8 community plots sampled, significant differences appeared in the NIR, but signals were less well differentiated in derivative plots.
Based on the results, hyperspectral leaf reflectance spectra can distinguish between plant family types and different communities. This is promising for the future interpretation of remotely sensed data and provides a case study for uses of FDA in other ecological applications.