Descriptors of network structure
To determine network structure and resilience from each habitat during a whole El Niño year, and for each habitat during each season during the El Niño and normal year, we assessed network structure by measuring six key metrics. First, we quantified nestedness, which measures the extent to which the interactions of one species are a sub-set of the interactions of another species when the matrix of interaction is organized by decreasing number of links (Dormann, Fründ, Blüthgen, & Gruber, 2009). We calculated nestedness using the weighted NODF approach, which is a measure of nestedness that uses overlap and decreasing fill in the weighted matrix, that has been shown to outperform other methods for estimating nestedness in binary networks (Almeida-Neto & Ulrich, 2011). Second, we quantified modularity, characterized as more interactions within a module than between modules (Dormann & Strauss, 2014), using the QuanBiMo algorithm that is based on simulated annealing and is more specifically designed for weighted bipartite networks (Dormann & Strauss, 2014). Third, we calculated weighted connectance by dividing linkage density by the number of species in the network (Tylianakis, Tscharntke, & Lewis, 2007), which reveals the number of links in the network in relation to the total number of links (Altena, Hemerik, & Ruiter, 2016). Fourth, we measured/calculated number of compartments, which are defined as isolated sub-sets of nodes interacting with each other that do not have any connections with another compartment in the network (Dormann, Fründ, Blüthgen, & Gruber, 2009). Fifth, robustness was calculated as the area below the curve of secondary extinction of bats when primary extinction of plant species was simulated according to three methods: random extinction of plant species (random), extinction of most connected to least connected plant species (degree) and extinction from the least connected to most connected plant species (abundance). And finally, niche overlap among bat species was calculated using the Morisita-Horn index (Horn, 1966). Apart from robustness, all metrics chosen have little or no biases to sampling completeness and network size (Fründ, McCann, & Williams, 2015). We used the function network level from the Bipartite package (Dormann, Gruber, & Fründ, 2008) to determine network structure and resilience from each habitat during the whole year and for each habitat during each season and calculated the following network metrics: weighted NODF, modularity (QuanBiMo), number of compartments, niche overlap, and robustness.