Spectral wavelength range influences the performance of chemometric models estimating various foliar functional traits

1. Hyperspectral reflectance can potentially be used to non-destructively estimate a diverse suite of plant physiochemical functional traits by applying chemometric approaches to leverage absorption features related to chemical compounds and physiological processes associated with these traits. This approach has considerable implications in advancing plant physiological and chemical ecology. For complex functional traits, however, there is a lack of well-defined absorption features and features may be unevenly distributed across the reflectance spectrum, suggesting that the influence of wavelength ranges on the performance of chemometric models is potentially important for accurately estimating foliar functional traits. 2. Here, we investigate the influence of spectral ranges on the performance of models estimating six tree functional traits: CO2 assimilation rate, specific leaf area, leaf water content and concentrations of foliar nitrogen, sugars and gallic acid. Using data collected from multiple different experiments, we quantified plant functional trait responses using standard reference measurements and paired them with proximal leaf-level hyperspectral reflectance measurements spanning the wavelength range of 400–2400 nm. A total of 100 different wavelength range combinations were evaluated using partial least squares regression to determine the influence of wavelength range on model performance. 3. We found that the influence of starting or ending wavelength on model performance was trait specific and better model outcomes were achieved when the starting and ending wavelengths encompassed absorption features associated with the specific leaf trait modelled. Interestingly, we found that including shortwave-infrared wavelength ranges (1300–2500 nm) improved performance for all trait models. 4. Collectively, our findings underscore the importance of optimal spectral range selection in enhancing the accuracy of chemometric models for specific foliar trait estimates. An emergent outcome of this work is that the approach can be used to (1) identify the important spectral features of traits that currently lack known absorption features or have multiple or weak absorption features, (2) expand the current suite of plant functional traits that can be estimated using spectroscopy and (3) ultimately advance the integration of a spectral biology approach in ecological research.