3D shape analysis of grass silica short cell phytoliths: a new method for fossil classification and analysis of shape evolution

Abstract

Summary Fossil grass silica short cell phytoliths (GSSCP) have been used to reconstruct the biogeography of Poaceae, untangle crop domestication history and detect past vegetation shifts. These inferences depend on accurately identifying the clade to which the fossils belong. Patterns of GSSCP shape and size variation across the family have not been established and current classification methods are subjective or based on a 2D view that ignores important 3D shape variation. Focusing on Poaceae subfamilies Anomochlooideae, Pharoideae, Pueliodieae, Bambusoideae and Oryzoideae, we observed in situ GSSCP to establish their orientation and imaged isolated GSSCP using confocal microscopy to produce 3D models. 3D geometric morphometrics was used to analyze GSSCP shape and size. Classification models were applied to GSSCP from Eocene sediments from Nebraska, USA, and Anatolia, Turkey. There were significant shape differences between nearly all recognized GSSCP morphotypes and between clades with shared morphotypes. Most of the Eocene GSSCP were classified as woody bamboos with some distinctive Nebraska GSSCP classified as herbaceous bamboos. 3D morphometrics hold great promise for GSSCP classification. It accounts for the complete GSSCP shape, automates size measurements and accommodates the complete range of morphotypes within a single analytical framework.

“You Are Not My Type”: An Evaluation of Classification Methods for Automatic Phytolith Identification

“…Approaches to archaeobotanical remains, such as the one presented here, are still scarce. Regarding Evett & Cuthrell (2016), Cai & Ge (2017), or Gallaher et al (2020) although they all develop similar methodologies to identify phytoliths, the objectives, as well as the selected criteria, are not comparable to the present study.…”

“…We have used general categories that all specialists recognize, but that can be further subdivided. The case of grass short cells is paradigmatic, and several researches carry out subclassification processes when those morphotypes can provide further taxonomic information (Barboni & Bremond, 2009; Gallaher et al, 2020). The existing internal variability within our samples is probably producing a loss of accuracy on the results.…”

“…There have been several studies that used quantitative phytolith morphometric size and shape parameters for the identification of morphometric characteristics (Ball et al, 2016;Out & Madella, 2016;Portillo et al, 2019). Recently, researchers have started to design computing methods for the automatic identification of phytoliths (Evett & Cuthrell, 2016;Cai & Ge, 2017;Gallaher et al, 2020). Evett & Cuthrell (2016) established the conceptual basis for the application of semi-automated classification methods to morphometric phytolith analysis, describing detailed procedures and strategies to be tested while acknowledging current technical limitations.…”

“…They used morphometric data from scanning electron microscopy (SEM) images to train a classifier to successfully distinguish different genera within the Oryzeae even though they all produce the same phytolith morphotypes. Gallaher et al (2020) used three-dimensional (3D) geometric morphometrics and supervised classification algorithms to: (1) analyze the shape and size of modern grass phytoliths from 70 species of the subfamilies Anomochlooideae, Bambusoideae, Oryzoideae, Pharoideae, and Puelioideae; (2) build a classification model based on the short cells extracted from these modern samples; and (3) classify fossil grass phytoliths from Eocene sediments through the application of the previous resulting model. The results showed high classification scores among clades at different taxonomic levels even when different clades shared the same morphotypes.…”

“…There have been several studies that used quantitative phytolith morphometric size and shape parameters for the identification of morphometric characteristics (Ball et al, 2016; Out & Madella, 2016; Portillo et al, 2019). Recently, researchers have started to design computing methods for the automatic identification of phytoliths (Evett & Cuthrell, 2016; Cai & Ge, 2017; Gallaher et al, 2020).…”