FTIR spectra from grass pollen: a quest for species-level resolution of Poaceae and Cerealia-type pollen grains

Palynological analysis based on spore and pollen morphology is well established in the field of palaeo-environmental reconstruction but is currently not fully exploited for understanding the history and development of cereal cultivation due to difficulties in visually differentiating between grass species (Poaceae). Here we employ a chemotaxonomic approach, by examining the chemical differences among Poaceae taxa, based on Fourier-transform infrared (FTIR) microspectroscopy data to overcome problems associated with morphological similarities across the Poaceae family. FTIR spectra of untreated and acetolysed pollen from 19 Poaceae taxa were used in our study. We used both populations and individual pollen grains to explore how we can minimise the effect of Mie scattering (spectral distortions caused by scattering of the incident IR beam) on spectra from individuals. Random forest classification algorithms were applied to explore our ability to differentiate taxa at the species level. We found that pollen grains treated with acetolysis yield better classification results (86% for individuals and 97% for populations) compared to untreated samples (65.7% for individuals and 83% for populations), since they are less affected by Mie scattering. The high classification success at species level on acetolysed individual pollen grains suggests that our chemotaxonomic method holds substantial promise in numerous areas of grass and in particular cereal pollen research, including elucidating the history of agriculture.