Variations in accuracy of leaf functional trait prediction due to spectral mixing. (March 2022)

• Record Type: Journal Article
• Authors: Hacker, Paul W.; Coops, Nicholas C.; Laliberté, Etienne; Michaletz, Sean T.
• Is Part Of: Ecological indicators. Volume 136(2022)
• Publication Date: 2022-03
• Subjects: Biodiversity -- Functional traits -- Spectroscopy -- Spectral mixing -- Partial least squares regression; Environmental monitoring -- Periodicals; Environmental management -- Periodicals; Environmental impact analysis -- Periodicals; Environmental risk assessment -- Periodicals; Sustainable development -- Periodicals; 333.71405
• DOI: 10.1016/j.ecolind.2022.108687
• Languages: English
• ISSNs: 1470-160X

Abstract

Highlights: Leaf trait prediction error changes when multiple species contribute to reflectance. %N and EWT were more affected by spectral mixing than the 6 other traits evaluated. Spectral mixes containing multiple plant lifeforms enhance prediction errors. %N and LMA can identify key species in mixed grassland-woodland ecosystems. Understanding spectral mixing can improve studies quantifying leaf functional traits. Abstract: Associations between leaf chemicals and reflectance values using spectroscopy enables the modelling, and prediction of, individual functional traits. Prediction accuracy is generally high when reflectance values are acquired from a single individual or species, however when the spectroscopy field of view is extended to include multiple species, the effects on predictive accuracy are less clear. Leaf spectroscopy of 23 plant species and the subsequent chemical derivation of 14 leaf traits enabled the creation of a species-inclusive partial least squares regression model for each trait (Validation R-2 = 0.28 – 0.85). Model effects were applied to mixed spectra containing various combinations of a target plant species and one, or multiple, other species to determine if, and at what amount of target species presence, spectral mixing significantly altered the trait prediction of the target species. Trait prediction accuracy does change due to spectral mixing and certain traits were more affected by mixing than others. For example, percent nitrogen (%N) and equivalent water thickness (EWT) were more affected by spectral mixing than chlorophyll a (chl a ) concentration and leaf mass per area (LMA). These results suggest that species-specific relationships between spectra and traits like %N and EWT are more important, whereas a general site model that holds across species is more achievable for chl a or LMA. Mixes containing various percentages of forbs, shrubs, graminoids and trees generate different prediction errors in some traits, including %N. LMA was relatively unaffected, except when Q. garryana, a tree, contributed to the mixed spectra. In mixed grassland-woodland ecosystems the prediction of LMA was relatively unaffected by spectral mixing, while %N highlights graminoid presence and isolates a key invader. Spectral mixing can complicate the accurate estimation of remotely sensed leaf functional traits, compromising the effectiveness of this technique.