Hyperspectral sensing of aboveground biomass and species diversity in a long-running grassland experiment

Abstract

Vegetation properties can be assessed through analysis of canopy reflectance spectra. Early techniques relied on simple two-band vegetation indices (VIs) that exploit leaf reflectance properties at key wavelengths. As the technology matures it is now possible to gather and test hyperspectral data. Little evidence exists on how different management regimes, such as nutrient addition, might affect hyperspectral reflectance and thus influence derived estimates of plant diversity and productivity. At a grassland experiment in southern England, we used a portable spectroradiometer to sample 96 plots exposed to multifactorial treatments combining herbivory, plant competition, soil pH and fertility. Our objective was to compare the predictive performance of popular two-band VIs with a multivariate partial least square regression (PLSR) model that uses all available wavelengths. We found that the PLSR models showed higher predictive power than the best performing VIs – that was especially true for our measure of species diversity ($R_{\mathrm{cv}}^2$ = 0.36 compared with a Pearson correlation of 0.21). The predictive power for our PLSR model of biomass ($R_{\mathrm{cv}}^2$ = 0.54) compares favourably with values reported in earlier grassland studies. These results confirm that hyperspectral measurement combined with multivariate regression techniques is a promising approach for monitoring grassland properties. There is evidence of particular benefit in capturing narrow bands associated with the red edge region of the spectrum (700–750 nm). Remotely sensed hyperspectral images at a fine spatial scale offer the prospect for matching with sampling units as small as the 2 × 2 m nutrient subplots measured here.