Aim Interacting communities of species are organized into complex networks, and network analysis is reckoned to be a strong tool for describing their architecture. Many species assemblies show strong macroecological patterns, e.g. increasing species richness with decreasing latitude, but whether this latitudinal diversity gradient scales up to entities as complex as networks is unknown.We investigated this using a dataset of 54 community-wide pollination networks and hypothesized that pollination networks would display a latitudinal and altitudinal species richness gradient, increasing specialization towards the tropics, and that network topology would be affected by current climate. Location Global. Methods Each network was organized as a presence/absence matrix, consisting of P plant species, A pollinator species and their links. From these matrices, network parameters were estimated. Additionally, data about geography (latitude, elevation), climate at the network site (temperature, precipitation) and sampling effort (observation days) and extent (study-plot size) were gathered. Analyses were done using simultaneous autoregressive modelling (SAR). Results Species richness did not vary strongly with either latitude or elevation. However, network modularity decreased significantly with latitude whereas mean number of links per plant species (Lp) and A/P ratio peaked at mid-latitude. Above 500 m a.s.l., A/P ratio decreased and mean number of links per pollinator species (La) increased with elevation. Lp displayed mid-ambient peaks with temperature and nestedness and modularity displayed linear relationships with precipitation. Main conclusion Pollination networks showed macroecological patterns. No strong latitudinal or altitudinal gradient in species richness was observed. Lp and the A/P ratio peaked at mid-latitude whereas modularity decreased linearly. Both patterns are suggestive of a more specialized interaction structure towards the tropics. In particular, mean annual precipitation appeared influential on network topology as both nestedness and modularity varied significantly. Importantly, corrected regressions suggest that neither sampling effort nor extent affected the observed patterns.
Global Ecology and Biogeography, 2013, Vol 22, Issue 2, p. 149-162