Fig. 3

Plant-plant networks based on pollen transfer patterns of eight sampled grassland communities. Each node represents one plant species and each link represents HP transfer between two species. Direction of arrows indicates HP transfer from donor to receptor species and width of arrows indicates the number of transferred pollen grains. Nodes with the same colour are in the same module within one community. Numbers in community names show sampling date: I = early summer, II = mid-summer, III = late summer. Eselsburger Tal I is not shown as no modules were detected. OOB: the estimate of error rate from random forest analysis shows the percentage of misclassifications, i.e. assignment to the wrong module. The three most important variables for the module classification are given for each community (Table 2). The full results of the random forest analysis are given in Additional file 1-5. For RB III, two solutions of the modularity algorithm (Blondel et al. 2008) are shown as both were equally likely: solution one with four modules / solution two with five modules because one module was split into two plant-plant networks (computed in Gephi version 0.9.2 [41] by using the Fruchterman Reingold layout)