Taxonomic and Spatial Geometry in Phytoplankton Monitoring Data

This early-stage study, investigates how geometry-aware metrics and models can support the analysis of phytoplankton monitoring data. We study subsets of 169,015 records from the German Marine Environmental Database. The data naturally exhibit two structures guiding modelling choice: a hierarchical taxonomy over plankton observations and a spatial graph induced by the monitoring stations. We show that Tree-Wasserstein is more robust to variation in taxonomic resolution than a well-established metric for plankton community comparison, Bray-Curtis.We further identify spatial distance-decay in the data and show that graph-based forecasting improves over a local baselines at most stations.