Appendix S1 The response of three phytoplankton size fractions (pico-, nano- and microplankton) to changes in copepod abundance. The bars represent mean values and the error bars indicate the standard deviation. At higher predation pressure (i.e. higher copepod abundance) larger phytoplankton are favored while at lower predation pressure (lower copepod abundance) smaller phytoplankton are favored.


Appendix S2 Explained variance as function of number of clusters for two different configurations of our dataset: (a) with irradiance and (b) without irradiance. For the four-clustering, the explained variance increases from 82% for the dataset that includes irradiance (a) to 87% for the dataset that does not include irradiance (b). The explained variance is calculated as the ratio of the between-group variance to total variance and is also known as F-test statistic. The red circle in panel b marks the ‘elbow’ beyond which the gain in the explained variance is very marginal.


Appendix S3 Principal component analysis of environmental parameters (including irradiance) and normalized phytoplankton size fractions. From all environmental variables, irradiance has the lowest loading with respect to the first PC indicating that irradiance has the least explanatory power.


Appendix S4 Phytoplankton community size structure and environmental conditions obtained with a four-cluster partitioning on the dataset that includes irradiance. (a) Geographical distribution of the clusters. (b) Distributions of size-classes clustered according to water temperature, irradiance and nutrient concentrations with contours corresponding to the convex hull of the size-fraction distribution of each cluster. Symbols denoting the mean values of the size fraction distribution or each cluster called: Low Temperature – High Nutrients – Low Irradiance (LTHNLI), Mid Temperature – High Nutrients – High Irradiance (MTHNHI), Mid Temperature – Low Nutrients – Low Irradiance (MTLNLI), High Temperature – Low Nutrients – High Irradiance (HTLNHI). The color-coding reflects the cluster classification.


Appendix S5 Summary statistics of a model comparison for the data excluding cruises AMT13 and AMT14 with temperature (Temp), nitrite+nitrate concentration (NOx), phosphate concentration (PO4), silicate concentration (SiO4), degrees of freedom (Df), Bayesian information criteria (BIC), and Akaike weights (wm). M3 = Temp + PO4 is the preferred model because of its lowest BIC score and its highest Akaike weight.


Appendix S6 Global distribution of pico-, nano- and microplankton normalized size fractions in four selected months. The figure shows the normalized size fraction (ranging from 0 to 1) at given nutrient concentration and temperature, calculated with equations (4) and (5) adapted to the model M3 = Temp + PO4, and for the dataset excluding cruises AMT13 and AMT14 (see Appendix S5). Compare that the temporal and the spatial variability of both model results (i.e. M2 = Temp + NOx in Fig. 5 and this model result) are qualitatively the same.


Appendix S7 Loadings of first principal component (PC) for the dataset incuding (black bars) and excluding (grey bars) cruises AMT13 and AMT14. The left set of columns is the PC analysis including irradiance and the right set of columns is the PC analysis excluding irradiance. The letters on the bottom axis stand for the variables nitrite+nitrate (N), phosphate (P), silicate (S), temperature (T) and irradiance (I). The results suggest that irradiance is the least explanatory variable and that nitrite+nitrate (N) and phospathe (P) are highly correlated on both datasets (i.e. the datasets including and excluding AMT13 and AMT14).

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.