To evaluate how sampling methods (plant- or animal- centred) capture the variation of species and interactions across habitats, we calculated the β-diversity between all pairwise combinations (a total of 1296 pairwise combinations) and the β-diversity of plants and pollinator species separately (β-plants and β-pollinators). First, β-plants and β-pollinators, based on quantitative data, were calculated using the vegdist function (with Ruzicka index) in the vegan package (Oksanen et al., 2018):
βRuzicka = (B + C)/(A + B + C)
where A is the sum of intersections (or minima) of species abundances at two sites, B is the sum at site 1 minus A, and C is the sum at site 2 minus A (Legendre, 2014; Noreika et al., 2019). Second, interaction beta diversity was estimated with pairwise comparison of all networks using quantitative data based on frequencies of interactions. We used the framework available in the betalink package (Poisot et al., 2012) with modifications to include a quantitative estimate of beta diversity previously used in Magrach et al., (2017) and Noreika et al., (2019) following Bartomeus, (2019) and incorporated in the function betalinkr_multi in the bipartite package version 2.15 (Dormann et al., 2008). We calculated network beta diversity of interactions (βWN), its component generated by dissimilarity of species (βST - turnover) and the dissimilarity in interactions between co-occurring species in the local networks, i.e., beta diversity of interactions between shared species (βOS - rewiring). We used the recommended method for additive partitioning with the argument commondenom, which implements a method originally proposed by Novotny (2009). This method incorporates a secondary partitioning of dissimilarity (βWN and βOS) and derive "true" components (βWN.repl, βOS.repl) that remove the dissimilarity generated by the difference in richness between the networks, where:
βWN = βWN.repl (link turnover) + βWN.rich (differences in the number of links)
and
βOS = βOS.repl (link turnover among shared species) + βOS.rich (differences in the number of links among shared species).
The procedure used here follows Legendre (2014 and references therein) and was first applied to networks by Noreika et al., (2019) modified from Carvalho et al., (2012). The beta diversity metrics take values of 0 when the interactions or species composition are the same, whereas values of 1 indicate a completely different composition and/or interactions. Additionally, we tested whether the spatial location of plots would be related to the beta diversity of interactions and species. Thus, we verified whether the distance between the plots can be a determining factor in the dissimilarity of interactions (results can be seen in Fig. S1: βOS: χ2 = 0.81, P >0.05, βST: χ2 = 0.64, P >0.05, βWN: χ2 = 0.99, P >0.05, pollinators β diversity: χ2 = 0.91, P >0.05, plants β diversity: χ2 = 0.99, P >0.05). We also tested our models for spatial autocorrelation (see below). In addition, we analysed the β-diversity measures within the same habitat (Canjiqueiral, Riparian forest and Paratudal) and between habitats in order to verify whether the beta diversity would be greater between different habitats than within the same habitats.