T67 Future Leaders in Paleontology

Sunday, 22 September 2019
1:30 PM - 5:30 PM
Room 227ABC, North Building

OLD PLANTS, NEW TRICKS: APPLYING MACHINE LEARNING TECHNIQUES TO THE PALAEOBOTANICAL RECORD

Paper number: T67-4
Program: Technical Sessions
Time: 2:15 PM – 2:30 PM

Authors

Abstract

Plant fossils can be used to reconstruct various aspects of the palaeoenvironment, including past vegetation, atmospheric and climatic conditions. One approach to palaeoclimatic estimation uses the bioclimatic envelopes of the extant representatives of fossil taxa (the nearest living relative approach). However, the environmental ranges inhabited by the modern relatives of some co-occurring fossil species do not overlap – these instances are referred to as 'anomalies' in the fossil record. It is not clear whether these anomalies represent extinction, evolution of the climatic niche or non-analogous environments. By using machine learning to detect ecological overlap, we can quantify taxonomic, environmental and temporal patterns in these anomalies and gain insight into the evolution of conifer lineages, as well as strengthen our confidence in paleoclimatic estimates.

Another way to reconstruct past environments is via the use of proxies. The relationship between stomatal frequency and pCO2 is well documented, but this uses only one of many characters that are preserved by the cuticle. In fact, the cuticle preserves the shape, size and arrangement of all the cells it overlays. Currently, comprehensive extraction of these data is difficult and time-consuming, requiring manual tracing of cell outlines. This has prohibited broad-scale studies into the links between epidermal architecture and environment. We combine a deep convolutional neural network with image processing techniques to automate the process of cell measurement from cuticle images. These measurements can then be used to explore the complex relationship between anatomy and environment in both extant and fossil plants.

We demonstrate that the use of sophisticated machine learning algorithms facilitates novel analyses of palaeobotanical datasets and present our findings to date.

Geological Society of America Abstracts with Programs. Vol. 51, No. 5

doi: 10.1130/abs/2019AM-336326

© Copyright 2019 The Geological Society of America (GSA), all rights reserved. Permission is hereby granted to the author(s) of this abstract to reproduce and distribute it freely, for noncommercial purposes. Permission is hereby granted to any individual scientist to download a single copy of this electronic file and reproduce up to 20 paper copies for noncommercial purposes advancing science and education, including classroom use, providing all reproductions include the complete content shown here, including the author information. All other forms of reproduction and/or transmittal are prohibited without written permission from GSA Copyright Permissions.