Network structure in dry forest versus rainforest in relation to null models
Contrary to initial predictions that mutualistic networks in the rainforests would become more modular and less nested during El Niño than in a normal year, while networks in the dry forests would become less modular and more nested, we observed similar changes to network structure in response to El Niño for both forests. In both forests, aspects of the observed changes in network structure are likely to have contrasting consequences for network resilience. For example, nested mutualistic networks are thought to contribute to an increase in the maximum amount of biodiversity supported in the environment (Bastolla et al. 2009). A decrease in nestedness, as observed in both forest types, may thus be related to an increase in effective competition (Bastolla et al. 2009) driving niche separation. This is important as lower nestedness was found across most of the networks in the present study and nestedness helps to buffer against secondary extinctions and temporal fluctuations (Tylianakis, Laliberté, Nielsen, & Bascompte, 2010). Similarly, the decrease in connectance is worrying as this network metric is thought to contribute to ecosystem functional stability during fluctuating environmental conditions (Tylianakis et al. 2010).
Most observations of robustness to species extinctions also suggest a decrease in the stability of the communities and resilience of biological interactions, likely as a result of decreases in connectance and nestedness in the networks (Thébault & Faontaine, 2010). These effects are particularly important as connectance and nestedness are thought to show little temporal variation within and between years (Dupont, Padrón, Olesen, & Petanidou, 2009; Vázquez, Blüthgen, Cagnolo, & Chacoff, 2009). In habitats such as forest and savannah, recovery to the conditions before disturbances such as floods and droughts is slow (Maron, McAlpine, Watson, Maxwell, & Barnard, 2015). Thus a significant deviation in network structure in normal years following these extreme climatic events would be expected. Overall, this effect might reduce the biodiversity supported in these ecosystems, especially when taking into consideration the expected increase in the frequency of strong El Niño events and the worldwide trend for wet areas to become wetter and dry areas to become drier (Chou et al. 2013; Cai et al. 2014).
We observed higher values of modularity than those expected under null models for both forests, also suggesting that the current interacting species are showing higher differentiation in their niche use. Modularity was not only significantly higher than expected by chance, but values for both forests were also higher than the calculated ones using a similar algorithm for previously observed mutualistic networks of phyllostomid bats in other regions of South America during normal conditions (Mello et al. 2011). Following a similar trend, the increase in compartmentalization of both habitats might reduce the number of coexisting species as fully connected networks promote a reduction in the effective interspecific competition (Bastolla et al. 2009). On the other hand, compartmentalization has been linked to greater stability, slower spread of disturbance, and smaller likelihood of trophic cascades in networks (Tylianakis et al. 2010).
It is interesting to note that the similar increases in compartmentalization and modularity alongside a decrease in nestedness might have arisen due to the same causes in each forest. Changes in rainfall have an impact in different groups of herbivorous mammal populations through alterations in the amount and quality of food resources (Mandujano, 2006; White, 2008). Severe droughts in some Pacific areas provoked by El Niño were responsible for increased production of flowers and fruit of the entire plant community (Wright & Calderon, 2006), meanwhile in rainforests flowering was triggered by heavy rain (Wright, 1991). In Central American tropical forests, the fall of leaves after droughts that occurred during El Niño events tended to be associated with subsequent increases in seed production (Detto, Wright, Calderón, & Muller-Landau, 2018). These events, when both droughts and floods were associated with increased productivity of fruits and flowers could likely be the explanation to pattern that we have witnessed where the dry forests and rainforests showed similar changes in network structure. On the other hand, the drought that occurred in the dry forests of ACG promoted by the strong El Niño of 2015 caused a reduction in seed production that remained even after the return to normal levels of the rainfall (O’Brien et al. 2018). Thus, this effect was probably the main responsible for the changes in our observed networks for ACG, with the reduction in fruit availability leading to a higher resource specialization, which promoted an increase in modularity but a decrease in nestedness. Despite the contrasting causes, similar responses to opposite water stress in two very dissimilar species communities suggests a generalized response to stress that may become more prevalent as extreme weather cycles increase in frequency (also see Butt et al. 2015).
One of the limitations of our comparisons is that we do not have data collected for a normal year during both seasons from any of the forest types. Thus, it is hard to fully understand how our results are limited to the data and null models that we have used for the comparisons, or if they also reflect a real comparison with values gathered from a normal year for both forests. Another limitation is that for all almost all networks, except for two, we have lower values of sampling completeness than the rule of thumb proposed by Macgregor et al. (2017) (90%), though not by much. In addition, most of the rarefaction curves built to estimate the number plant species present in the diet of each bat species did not reach an asymptote. However, we focused our study mostly on network metrics that do not show a strong bias by network size which should minimize the impact of these issues. Finally, it is hard to assess the influence of markers choice that we used for plant identification on the values of network metrics. Multiple genetic markers have been proposed in various combinations to identify different plant species (matK , trnH-psbA , rbcL , ITS2), but still not sufficient to discriminate closely related species in some taxonomic groups, especially those with recent and intense species radiation (Hollingsworth et al. 2011). For example, the fig tree (Ficus ), one of the common genera consumed by bats, is extremly specious and demonstrated poor resolution on species level usingrbcL and ITS (Ronsted, Weiblen, Clement, Zerega, & Savolainen, 2008). As such, some of our identificaitons should be treated provisionally. However, our analysis of data limited only to genera (see supplement) suggests our observations are robust to these effects. One major benefit of our molecular approach is the inclusion of plant species which might otherwise be missed when their seeds are not consumed. Our ability to identify plants from consumed pollen or fruit pulp provides a more complete perspective than many previous analytical approaches.