MANUAL CLASSIFICATION OF POLLEN GRAINS IN THE DIGITAL ERA

Computational classification of pollen grains is increasing in popularity. For some, computation is a way of increasing the number of samples that can be analysed. For others, it is a way of classifying taxa that humans struggle with, usually because the morphology of the taxa in question is very similar. In this presentation, I outline a recent investigation of the ability of human analysts to classify grass pollen using scanning electron microscopy images of surface texture. This provides baseline data for computational classifications of the same material, and gives a sense of the difficulty of this classification problem. I also present some preliminary data on the accuracy and consistency of human analysts undertaking a palynological classification problem that most would regard as easy. Using examples from both cases I attempt to highlight strengths and weaknesses in the way humans record and represent pollen morphology during the process of classification.