Datensatz

Zusammenfassung

Aim Develop a biogeographical classification of phytoplankton size distributions for the Atlantic Ocean and predict the global phytoplankton size composition based on prevailing environmental conditions. Location Atlantic Ocean and Global Ocean Methods Using phytoplankton size composition data, nutrient concentrations (nitrite+nitrate, phosphate, and silicate), irradiance, temperature and zooplankton abundances of the Atlantic Meridional Transect programme, we derived and tested an environmental classification method of phytoplankton size distribution with a k-means clustering technique. We then used principal component and Dirichlet multivariate regression analyses to disentangle the relative influence of different environmental conditions on the phytoplankton size composition. Subsequently, we evaluated different probabilisitic models and selected the most parsimonious one to estimate the global phytoplankton size distributions in the world oceans based on global climatology data of the World Ocean Atlas 2009. Results Based only on prevailing environmental conditions and without a priori knowledge concerning, for example, the position of oceanic fronts, the primary productivity, the distribution of organisms or any geographical information, our classification method captures the size structures of phytoplankton communities across the Atlantic. We find a strong influence of temperature and nitrite+nitrate concentration on the prevalence of the different size classes, and we present evidence that both factors may act independently on structuring phytoplankton communities. While at low nitrite+nitrate concentrations temperature has a major structuring impact, at high nitrite+nitrate concentrations its influence is reduced. Finally, we show that the global distribution of phytoplankton community size structure can be predicted by a probabilistic model based only on temperature and nitrite+nitrate. Main conclusion The global distribution of phytoplankton community size structure can be predicted with good approximation using a parsimonious probabilistic model forced by only temperature and nitrite+nitrate data.