Title: A trait-based taxonomy for phytoplankton biomass modeling and prediction
Authors: Crispin Mutshinda - Dalhousie University (Canada) [presenting]
Andrew Irwin - Dalhousie University (Canada)
Zoe Finkel - Dalhousie University (Canada)
Abstract: A Bayesian model is developed for phytoplankton biomass dynamics and estimate trait values of phytoplankton taxa in the Western English Channel using 7 years of weekly data recorded at Station L4. In addition to the usual classification of species into diatom and dinoflagellate functional types, we create and evaluate an alternative trait-based grouping. While our ultimate interest is in biomass modeling and projection, species-level biomass data are fraught with missing values due either to actual absences or abundances below detection thresholds. Consequently, we initially rely on occurrence data, which are exempt from these issues. We analyze the factors that determine the presence-absence of individual phytoplankton species and subsequently apply a clustering algorithm to these traits to define a new taxonomy. We contrast the performance of the biomass dynamics model under three different groupings namely, (i) our trait-based taxonomy, (ii) the null clustering of identical structure where cluster memberships are randomly assigned to species, and (ii) the functional type taxonomy, with regard to the quality of cluster-specific trait value estimates and the model skill for total biomass prediction.