Indicators of plant species richness in AVIRIS spectra of a mesic grassland

Abstract Hyperspectral imagery of the Konza Prairie Biological Station in northeastern Kansas was used to evaluate upwelling spectral radiance, prairie spectral reflectance and band ratios of each as potential indicators of vascular plant species richness in a mesic grassland. The extent to which spatial variability in these parameters related to plant species richness also was investigated. A 224 channel hyperspectral data cube acquired in June 2000 by the Airborne Visible and Infrared Imaging Spectrometer (AVIRIS) provided complete coverage of the 400–2500 nm range at approximately 10 nm per channel. After band deletions accounted for detector overlap and strong atmospheric attenuation features, 176 bands were retained for analysis and spanned the 404–2400 nm range. Prairie reflectance was estimated via radiative transfer modeling and scaling to a library spectrum of highway construction material. Data were sampled from pixels having a 19 m ground sample distance (GSD) to represent each of 93 vegetation sampling transects. Reflectance and radiance at mid-infrared wavelengths (e.g., 1553 nm), and band ratios that were based on atmospheric windows in the red, near-infrared and mid-infrared spectra estimated species richness to within 6 to 7 species per transect. The 856 to 780 nm radiance or reflectance ratio yielded maximum adjusted coefficients of determination (r2) of approximately 0.4 in regressions with richness when data from bison-grazed and ungrazed areas were combined. These regressions remained significant (p ≤ 0.001) when only ungrazed areas were assessed although r2 reduced to approximately 0.2. Richness was related significantly also to the 433 to 674 nm reflectance ratio for grazed-plus-ungrazed and ungrazed-only areas. In contrast, the effectiveness of the 433 to 674 nm radiance ratio was reduced by atmospheric backscatter. Species richness did not correlate strongly or consistently with transect spatial variability (coefficient of variation or range) in radiance, reflectance or band ratio value, apparently as a consequence of the relatively small area sampled for each transect (approximately 0.5 ha). Relationships between richness and prairie spectral features were explained by the influence of soil exposure on both parameters. Richness and estimated soil exposure tended to increase from ungrazed lowlands, to ungrazed slopes, to ungrazed uplands to grazed areas. Remotely sensed estimates of soil exposure may be particularly useful in addressing plant species richness on grazed grasslands owing to an overall similarity in spectral reflectance among dominant plant species.