We used nonmetric multidimensional scaling (NMDS) to visualize
compositional diversity patterns of phytoplankton communities
from different regions. First, we constructed a community
matrix d samples in rows and species in columns. The joint community
matrix, incorporating all regions, had 44871 samples and
2146 taxa. To scale down the task to computationally feasible size
we collapsed the data to geographically distinct sampling locations
(stations) and averaged the species-specific biomasses. We further
restricted the data to the summer season (July to Sept in the
northern hemisphere, and Dec to March in the southern hemisphere).
For the Baltic Sea, which had heterogeneous spatial sample
coverage, we collapsed the data of the seventeen sub-basins
(Table 1). The Baltic Sea data were temporally constrained to
1990e2005 in order to reduce the effect of known long-term trend
in phytoplankton compositional diversity (Olli et al., 2011), and we
also excluded stations with very close proximity to the shoreline
(