Studying aquatic population dynamics using spatio-temporal monitoring data is associated with a number of challenges and choices. One can let several samples represent the same population over larger areas, or alternatively model the dynamics of each sampling location in continuous space. We analysed the spatio-temporal population dynamics of six phytoplankton taxa in the Baltic Sea applying multivariate state-space models with first-order density dependence. We compared three spatial scales and three models for spatial correlation between predefined subpopulations using information theoretic model selection. We hypothesised that populations close to each other display similar dynamic properties and spatial synchrony decreasing with the distance. We further hypothesize that intermediate-scale grouping of data into subpopulations may parsimoniously represent such dynamics. All taxa showed constant density dependence across space and strong spatial synchrony, consistently requiring a parameter for spatial correlation whenever models included several population states. The most parsimonious spatial structure varied between taxa, most often being one panmictic population or ten intercorrelated population states. Evidently, longer time-series, containing more information, provide more options for modelling detailed spatio-temporal patterns. With a few decade-long plankton time-series data, we encourage determining the appropriate spatial scale on biological grounds rather than model fit.