Evaluating Principal Components Analysis for Identifying Optimal Bands Using Wetland Hyperspectral Measurements From the Great Lakes, USA

Mapping species composition is a focus of the wetland science community as this information will substantially enhance assessment and monitoring abilities. Hyperspectral remote sensing has been utilized as a cost-efficient approach. While hyperspectral instruments can record hundreds of contiguous narrow bands, much of the data are redundant and/or provide no increase in utility for distinguishing objects. Knowledge of the optimal bands allows users to efficiently focus on bands that provide the most information and several data reduction tools are available. The objective of this Communication was to evaluate Principal Components Analysis (PCA) for identifying optimal bands to discriminate wetland plant species. In-situ hyperspectral reflectance measurements were obtained for thirty-five species in two diverse Great Lakes wetlands. PCA was executed on a suite of categories based on botanical plant/substrate characteristics and spectral configuration schemes. Results showed that the data dependency of PCA makes it a poor, stand alone tool for selecting optimal wavelengths. PCA does not allow diagnostic comparison across sites and wavelengths identified by PCA do not necessarily represent wavelengths that indicate biophysical attributes of interest. Further, narrow bands captured by hyperspectral sensors need to be substantially re-sampled and/or smoothed in order for PCA to identify useful information.