Spectral discrimination using infinite leaf reflectance and simulated canopy reflectance

Abstract

ABSTRACT An important application of hyperspectral remote sensing is monitoring the diversity of plant species and functional types, which correlate with differences of leaf chemical constituents. Diffuse reflectance of an infinitely thick leaf ( R ∞) is directly related to chemical absorbance and may be calculated from leaf spectral reflectance and transmittance based on Kubelka-Munk theory or other radiative transfer models. The PROSPECT-D leaf optics model was used to create leaf optical properties of nine pseudo-species, and R ∞ were calculated using Stokes’, Lillesaeter’s, Goudriaan’s, and Hapke’s equations. Conceptually, R ∞ is assumed to equal canopy reflectance (canopy ρ ) at maximum leaf area index. For each pseudo-species, canopy ρ at a leaf area index of 10 was obtained using the PROSAIL model, but the simulated canopy ρ did not match R ∞. Similarity between R ∞ and canopy ρ for each pseudo-species was calculated using five spectral information metrics: Euclidean distance (ED), an adjusted spectral correlation measure (SCM), spectral angle mapper (SAM), spectral information divergence (SID), and the product of SID × sin(SAM). Differences using ED between canopy ρ and R ∞ were large, especially using Stokes’ equation. However, the SCM, SAM, SID, and SID × sin(SAM) metrics showed R ∞ from the Stokes’ equation was the most similar to canopy ρ , because these metrics are insensitive to differences in magnitude. The combined SID × sin(SAM) metric had the greatest similarity between R ∞ and canopy ρ compared to the other spectral information metrics. While R ∞ was not a good substitute for canopy ρ, R ∞ may be important for assessing differences of leaf chemical composition among species, thereby avoiding effects of taxon-independent factors on canopy ρ .