Hyperspectral remote sensing offers rich opportunities for biodiversity assessment. Remote sensing-based biodiversity metrics provide the opportunity to map biodiversity at large spatial extents in a short amount of time. However, the metrics are sensitive to a number of confounding factors. Main goals of this study were to 1) investigate the effect of correcting for varying soil exposure and 2) tame “curse of dimensionality”. The “curse of dimensionality”, a rather counter-intuitive concept, is related to the problem of sparsity of useful data embedded in high dimensions. This issue can be addressed by dimension reduction, which can select optimal bands of the hyperspectral data. In this study, α-diversity (species richness) was used as a measure of plant biodiversity. We used two imaging spectrometry data sets from the Cedar Creek Ecosystem Science Reserve in Central Minnesota, USA, at two levels: proximal and airborne. The spatial resolution of the airborne and proximal datasets were 0.75 m and 1 cm, respectively. This combination of hyperspectral data sets included varying degrees of soil background sampled at two different spatial scales.
The results of five optical biodiversity metrics, including the coefficient of variation, convex hull volume, spectral angle mapper, spectral information divergence, and a newly proposed dimension reduction-based metric called “convex hull area” are presented in this study. After removing the soil background effect from the proximal data set, the performance of all biodiversity metrics improved significantly in terms of R2