COS 242-5 - Separating grass lineages through leaf spectroscopy

Background/Question/Methods

Grass functional types are typically classified based on photosynthetic pathway (C3 or C4). However, the C4 photosynthetic pathway has evolved independently multiple times under different ecological settings, resulting in different traits among lineages with C4 taxa. Thus, evolutionary lineages may better capture functional differences among C4 taxa. Hyperspectral spectroscopy using the visible, near-infrared, and shortwave infrared ranges (VIS-NIR-SWIR) may be capable of separating lineages, enabling larger-scale mapping of lineages and their respective functional traits. In this study, leaf spectra from 43 grass species, representing the Chloridoideae, Pooideae, and Panicoideae lineages, from 3 grassland sites were collected using hyperspectral imaging spectroscopy (350 to 2500 nm). We processed the reflectance spectra by trimming wavelengths shorter than 400 nm or longer than 2400 nm, splicing the sensor overlap regions of each spectrum, and subsampling the spectra to 10 nm resolution using spline interpolation. Linear discriminant analysis (LDA) was used to determine whether grass lineages can be spectrally separated.

Results/Conclusions

The results indicate Chloridoideae, Pooideae, and Panicoideae have distinct spectral signatures. The LDA model using the optimal wavelengths for discrimination was able to classify the leaf spectra resulting in an overall accuracy of 88% in separating lineages with a Kappa coefficient of 0.81. The best-classified group was Panicoideae, with a median accuracy of 95%, whereas Chloridoideae had the lowest median accuracy with 88%. Because Panicoideae contains both C3 and C4 species, when lineages were further broken down by photosynthetic pathway, the overall accuracy across all lineages improved to 90% with a Kappa coefficient of 0.85. C4 Panicoideae had a median accuracy of 97%, whereas C3 Panicoideae had a median accuracy of 85%. These results suggest that both evolutionary lineage and photosynthetic pathway capture differences in leaf traits that result in unique reflectance spectra. The results also show potential for using hyperspectral imaging spectroscopy as a tool to map grass lineages and their respective leaf functional traits. Developing a spectral library for grass species and lineages would benefit plant phenotyping for further research in leaf traits.