Our results show strong spatio-temporal variability in plant-pollinator interactions driven by differences in pollinator specialization levels as a consequence of changes in plant and pollinator species richness. This variability at small spatial scales has already been observed within synthesis works (e.g., (Trøjelsgaard & Olesen 2016)), but to the best of our knowledge this is the first attempt at simultaneously evaluating temporal and spatial variabilities to infer their different effects. Interestingly, we find that although temporal variability is important, and in particular drives changes in the composition of pollinator species communities, spatial differences in interaction composition seem to override temporal differences. In this case, these differences are not a consequence of changes in the overlap of species, since we focus on the turnover of shared species across space or time, but rather are a consequence of changing interacting patterns between shared species due to changes in the overall composition of plants and pollinators. For example, we find that species level specialization changes through space, increasing with pollinator species richness and decreasing with plant species richness. It therefore seems that in the presence of direct competition pollinator species become more specialized, while they become more generalized when the availability of different plant resources increases.
We find that sampling completeness varied as a function of how data were pooled. Specifically, when spatial and temporal dynamics were not considered, i.e., when data were pooled for each site and period, sampling completeness greatly increased, giving a false sense of complete sampling coverage, when in reality such a reduced sampling size is omitting differences in species and link composition across space and time. These results are important because in many cases, the validity of network structure, summarised by a number of descriptive metrics, depends on how representative a sample is from the overall population. Neglecting to take into account spatial or temporal dynamics could thus be affecting our understanding of many natural ecosystems.
Within our dataset, we find that while most plant species were consistently recorded across both study years, a significant proportion of the interactions and pollinator species were uniquely recorded during one of the study years. In the case of pollinator species, we find values of pollinator persistence across years ranging from 54 to 73%. These values are intermediate compared to those found in other studies. For example, (Olesen et al. 2008b) found that 80% of the pollinator species present in an arctic heathland were recorded during two consecutive study years, (Dupont et al. 2009) found that this value decreased to <25% of the total pollinator fauna, while in the case of (Petanidou et al. 2008) only 20.5% of pollinator species were detected in all of four study years. Interestingly, these results arise despite our large sampling efforts, spanning multiple locations and times, and including several surveys throughout the day to account for daily dynamics. This could mean that our sampling was still too low to uncover all existing species and interactions, or it could mean that some of these interactions follow neutral processes, and are a consequence of random encounters between plants and pollinators based on species abundances (Vázquez et al. 2009), which do not necessarily replicate every year as species abundances shift through time (e.g., bet hedging Danforth 1999).
A re-estimation of sampling completeness removing these interactions that were uniquely recorded during one study year greatly improves coverage values in the case of plant-pollinator links, but decreases the sampling completeness of pollinator species in the case of Gorbea. This shows that a significant portion of the pollinator species in this area were involved in unique interactions not recorded both years, which could be a consequence of pollinator species in this area having more of an “opportunistic” behavior, changing their plant preference as a consequence of changing plant species abundances. Indeed, Gorbea, located in a mountain area features much larger environmental differences throughout the day as well as between weeks, which probably means pollinators are more adaptive to changing conditions (Ploquin et al. 2013). Pollinator communities in this area are more plastic in their use of resources, and will adapt better to future perturbations, such as climate change, or maybe our findings show that they are already adapting to it, as mountain areas are one of the most impacted by changing climates (Inouye 2019).
Further, our results show that while the community of plant species is relatively constant across space and time, pollinator species turnover is greatest through time, i.e., the composition of pollinator species changes more for a particular site through time than across sites for a particular time period. In the case of plant-pollinator links occurring between shared plant and pollinator species, we find the contrary. In this case, turnover is largest across space, i.e., given the same species of plants and pollinators, these interact more similarly through time for a given location than they do across space for a given period. Reducing our dataset to only interactions common in both years, significantly reduces plant and pollinator species turnover, but retains similar values of interaction turnover, particularly spatial turnover, which means that variability in interaction composition amongst shared species through space is a prevalent trend across both study areas. Therefore, not only do we have a significant portion of interactions occurring only in one study year, but amongst those that occur across multiple years, there are strong spatial differences in the frequency in which they occur.
A closer inspection at community structure metrics, shows that it is also space that acts as the main driver of differences in both community and species-level specialization values. In the case of species-level specialization, part of the variability can be explained through spatial differences in plant and pollinator species richness, such that pollinator specialization increases in the face of increasing competition by other pollinator species and decreases with the increase in plant resources. This therefore answers, at least partially, the question posed by (Trøjelsgaard & Olesen 2016), “why do some species interact one year but not the subsequent although they co-occur in both years?” Our results show that together with trait-matching and neutral processes driven by differences in species abundances (Vázquez et al. 2009), the composition of both interacting communities plays an important role in determining species interactions. This probably explains why predictive exercises have been somewhat successful in predicting macroscopic features of plant and pollinator communities, but no so much in predicting pairwise interactions (Vázquez et al. 2009)(Olito & Fox 2014), which vary across space and time (Trøjelsgaard & Olesen 2016).
In summary, our results show the importance of simultaneously considering spatial and temporal dynamics within natural communities. A step further should try to understand the mechanisms that drive these spatial and temporal differences and the biotic and abiotic factors that are driving them, and that will ultimately help us improve our predictions. These results are also important for conservation or restoration efforts, that should also consider the necessary micro-scale dynamics needed to maintain the stability and functioning of natural ecosystems.