Imaging spectroscopy of grass species and evolutionary lineages

Keywords: Spectroscopy, Hyperspectral, grass species, NEON,PLS-DA, LDA, MDA
Session Type: Virtual Poster Abstract
Day: Sunday
Session Start / End Time: 2/27/2022 05:20 PM (Eastern Time (US & Canada)) - 2/27/2022 06:40 PM (Eastern Time (US & Canada))
Room: Virtual 32

Abstract

Grass functional types are typically classified based on C3/C4 photosynthetic pathway. However, the C4 photosynthetic pathway has evolved independently multiple times under different ecological settings, resulting in different traits among clades. Hyperspectral spectroscopy using the visible, near-infrared, and shortwave infrared ranges (VIS-NIR-SWIR) may be capable in separating clades enabling larger-scale mapping of grass clades and their respective functional traits. In this study, spectral reflectance of grass species and clades from 3 NEON sites during peak greenness in the summers of 2020 and 2021 using spectroscopy (350 to 2500 nm) was performed through collecting spectral leaf samples on Chloridoideae, Pooideae, and Panicoideae clades. Partial least squares-discriminant analysis (PLS-DA), Linear discriminant analysis (LDA), and Mixture discriminant analysis (MDA) were used to determine whether grass clades can be spectrally separated using discriminant analysis. Results in all methods indicated that grass clades expressed distinct spectral signatures. MDA outperformed with an overall accuracy of 88 % in separating clades with a Kappa coefficient of .81, which displays a well-performed model. The best-classified class was Panicoideae, with a median accuracy of 95 %, Chloridoideae had the lowest median accuracy with 88 %. The results show great potential for using hyperspectral data in the VIS-NIR-SWIR reflectance spectroscopy as a tool to predict grass clades in the Continental United States. Developing a grass species and clades spectral library would benefit plant phenotyping for further research in leaf traits.