COS 279-5 - An automated method for morphological analysis of leaf cuticle with implications for paleoecology and paleoclimatology

Abstract: Fossilized leaf cuticle - the waxy surface of leaves common to geologic sediments from the last 400 million years - remains an underutilized data source for reconstructing paleoclimate and paleovegetation assemblages. The morphology of a leaf’s upper epidermal pavement cells reflects the environmental conditions of its growth. Traditionally, cell morphology was determined by hand-tracing cell walls using digital tools. This is a slow and tedious process, with cell-outline quality highly dependent upon the setup used, with a stylus producing different results than a mouse or trackpad. Additionally, the resulting shape parameter data can differ by >5% between workers due to fatigue, hand steadiness, and differing setups. To avoid the issues associated with hand-tracing cell outlines, we developed a user-friendly automated image analysis workflow using the freeware programs ‘ImageJ’ and ‘R’. Our easily tunable code allows for the generation of large cell-shape datasets from images of leaf epidermis, with options for different measurement parameters and levels of automation. Use of this method results in a higher degree of cell-tracing consistency, while allowing for increased sample sizes by drastically decreasing the time required to generate these data.

We apply this method to an image set of 130 vouchered herbarium specimens extracted from The Cuticle Database (cuticledb.eesi.psu.edu), representing 127 species in 87 genera from 36 families. Images were chosen to maximize taxonomic breadth and span the range of cell shapes and sizes, utilizing the specimens with the highest quality image of the upper cuticle. For each image, our workflow produced a set of vectorized cell outlines, with associated shape parameters. Our initial results suggest that cell measurement ranges overlap between the hand-drawn and automated datasets for the same images, but they follow slightly different distributions, which might be expected given our increased sample size and different sampling strategy. This workflow can now be used to reconstruct parameters such as Leaf Area Index, and thus the canopy structure of forests in deep time, more effectively utilizing the vast repository of leaf cuticle in the fossil record.