Linking Seasonal Foliar Chemistry to VSWIR-TIR Spectroscopy Across California Ecosystems

Abstract:

Synergies between the Visible Near Infrared/ Short Wave Infrared (VSWIR) and Thermal Infrared (TIR) spectra for identifying plant species’ foliar chemistry have been largely unexplored. Here we evaluate: 1) the capability of VSWIR and/or TIR spectra to predict levels of lignin, cellulose, nitrogen, leaf mass area, and water content; 2) whether these relationships between spectra and foliar chemistry can be extended to the reduced spectral resolution available in airborne and proposed spaceborne sensors, including the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS), the Hyperspectral Thermal Emission Spectrometer (HyTES), and the Hyperspectral Thermal Imager (HyspIRI); and 3) how these predictive relationships might change seasonally and among plant functional types.

In the 2013 spring, summer, and fall seasons, fresh leaves from sixteen common shrub and tree species were sampled from the Sierra Nevada Mountains, the Central Valley, and coastal Santa Barbara. Partial Least Squares (PLS) regression analysis was used to relate spectral response at wavelengths from 0.3 µm to 15.4 µm to laboratory-measured biochemical properties. For each component, three PLS regression models were fit using different portions of the spectrum: VSWIR (0.3 - 2.5 µm), TIR (2.5 – 15.4 µm), and the entire spectrum (0.3 – 15.4 µm). Three additional models were fitted using spectra resampled to AVIRIS (0.4 - 2.5 µm), HyTES (7.5 – 12 µm), and HyspIRI (0.38 - 12 µm).

The majority of the highest performing models used either the TIR spectrum or entire spectrum. When using simulated sensor spectra, HyspIRI produced the highest performing models, followed by HyTES. From model results the combination of VSWIR and TIR increased the R² of regression models compared to VSWIR alone, signifying that the inclusion of TIR data would improve predictions of foliar chemistry. Also, we found that model accuracy varied by seasons and across plant functional types. Models developed for all seasons resulted in a decreased R² value or required twice the number of factors of a single season to explain the variance. These analyses provide a quantitative estimate of the full spectrum’s potential to predict plant chemistry as it fluctuates seasonally.