Representing Zooplankters: An Example from the Foraminifera

Abstract

Because of their excellent preservation record, testate zooplankters provide valuable proxy ocean climate data through the Quaternary–Recent. Commonly, specimen abundances are sought, which are time-consuming to collect manually and require taxonomic expertise. While machine learning models obviate these problems, it is questioned whether the current use of specimens selected by experts to train the models impartially captures the variation within the source populations. To illustrate the potential value of the latter and their relevance to the selection of representative specimens, the 2D outline shape of the planktonic foraminifer Truncorotalia crassaformis from four globally distributed, late-Quaternary–modern collections is examined. Large intra-sample variation is attributed to changes in the size and shape of the last-formed chamber, which often departs radically from its predecessors. Similar outlines occur in each collection, and no single axial shape is dominant when the aggregated data, aligned on their centroids and adjusted for size and position, are projected onto their principal components. Several partitions based on distance from the centroid of the standardized data are considered as sources of representative specimens, with that at ±1.645σ (standard deviations, nominally 90%) suggested as suitable. This procedure obviates the need for expert-based consensus sampling; for greater environmental resolution, it can be applied to individual water mass samples. It assists, but does not fully resolve, the following basic diagnostic question: which characters separate Truncorotalia crassaformis from its relatives?