Inferring plankton community structure from marine and freshwater long-term data using multivariate autoregressive models

Multivariate autoregressive (MAR) models have been useful in elucidating food web dynamics and stability from freshwater plankton monitoring data, but their applicability to marine datasets has not been as well explored. Characteristics of marine systems, such as the movement of water masses by tides and currents, may present unique challenges to MAR modeling of data gathered in marine environments. To explore the behavior of MAR models with marine plankton data, in the context of what we know about applying MAR to freshwater data, we applied MARs to each of three freshwater and four marine long-term datasets and compared results among them. We generated sets of replicate MAR models for each dataset and used the consistency of models within each set of replicates as a measure of MAR performance. Overall, replicate MAR models generated from the marine datasets were less consistent than those generated from the freshwater datasets, suggesting that MAR methods need fundamental reconfigurations to be applied to standard marine plankton data. Higher variability observed within the marine MAR results may be attributable to weaker biotic interactions as represented by the data, and to overparameterization when the criteria for lumping freshwater plankton taxa into model variables are directly applied to marine plankton taxa. Adjustments to dataset preparation for MAR application and to the modeling framework itself may address these issues associated with analyzing data from highly dynamic systems.