Spatiotemporal Topic Modeling Reveals Storm-Driven Advection and
Stirring Control Plankton Community Variability in an Open Ocean Eddy
Abstract
Phytoplankton communities in the open ocean are high-dimensional,
sparse, and spatiotemporally heterogeneous. The advent of automated
imaging systems has enabled high-resolution observation of these
communities, but the amounts of data and their statistical properties
make analysis with traditional approaches challenging. Spatiotemporal
topic models offer an unsupervised and interpretable approach to
dimensionality reduction of sparse, high-dimensional categorical data.
Here we use topic modeling to analyze neural-network-classified
phytoplankton imagery taken in and around a retentive eddy during the
2021 North Atlantic EXport Processes in the Ocean from Remote Sensing
(EXPORTS) field campaign. We investigate the role physical-biological
interactions play in altering plankton community composition within the
eddy. Analysis of a water mass mixing framework suggests that
storm-driven surface advection and stirring were major drivers of the
progression of the eddy plankton community away from a diatom bloom over
the course of the cruise.