| Abstract Detail
Paleobotany Gallaher, Timothy [1], Perry, Jessica [2], Zhang, Jifan [3], Meng, Xiangyun [4], Jamieson, Kevin [3], Strömberg, Caroline [5]. 2D computer vision and 3D geometric morphometric shape analysis of grass silica short cell phytoliths (GSSCP): strengths and weaknesses of these approaches and implications for grass biogeography. The evolution of grasses and assembly of the grassland biome during the Cretaceous and Cenozoic vitally transformed Earth’s terrestrial ecosystems, but the timing and geographic contexts of these changes have been difficult to reconstruct because of the scarcity of informative grass macrofossils. Recent research using phytoliths has helped shed light on this key evolutionary and ecological series of events. However, gaining further insight into the evolutionary and biogeographical history of Poaceae depends critically on refined methods to identify specific grass lineages in the fossil record. Current methods for taxonomically placing grass phytoliths, so-called grass silica short cell phytoliths (GSSCP) are imprecise and subjective, employing qualitative shape traits and “expert” opinion, and attempts at solving this problem using quantitative measurements of GSSCP shape have failed to come up with accurate, precise, and practical solutions. Our research aims to enhance the use of GSSCP for robust taxonomic and ecological interpretations by developing two independent, quantitative, and objective ways of assessing taxonomic or phylogenetic affinity of fossil GSSCP: (I) Machine Learning and Computer Vision (ML&CV) analysis of 2D images; and (II) Landmark-Based Geometric Morphometrics (LBGM) on 3D images. We are currently assembling large databases of (1) 2-D brightfield images of GSSCP; and (2) 3-D models based on confocal images of stained GSSCP from 200 taxa from across the Poaceae phylogeny. We couple this work with phylogenetic mapping of functionally relevant traits (e.g., photosynthetic pathway, growth form) and environmental preferences. Our methods will vitally expand the number of taxonomically (and ecologically) well-placed fossils used for phylogenetic dating, biogeographic analysis, and ecological characterization of ancient grass communities.
1 - Bishop Museum, Natural Sciences- Botany - Herbarium Pacificum, 1525 Bernice St., Honolulu, HI, 96817, United States
2 - University of Washington, Department of Statistics, 4060 E Stevens Way NE, Seattle, WA 98195, Seattle, WA, 98195, USA
3 - University of Washington, Department of Computer Science, 3800 E Stevens Way NE, Seattle, WA, 98195, USA
4 - University of Washington, Department of Computer Science, 3800 E Stevens Way NE,, Seattle, WA, 98195, USA
5 - University of Washington and the Burke Museum o, Department of Biology, Life Sciences Building Box 351800, Seattle, WA, 98195, USA
Keywords:
Poaceae
Phytolith
fossil
Morphometrics
Machine Learning
Biogeography.
Presentation Type: Oral Paper
Session: PAL5, Paleobotany III: Patterns and trends
Location: Virtual/Virtual
Date: Tuesday, July 28th, 2020
Time: 3:45 PM
Number: PAL5004
Abstract ID:834
Candidate for Awards:None |
