Introduction
Biodiversity is heterogeneously distributed, and the geographical, ecological, and evolutionary processes that cause its variation across space and time has attracted much attention (Soininen et al., 2018). The biodiversity component that expresses this heterogeneity, beta diversity, captures the dynamic nature of species’ distribution patterns (Whittaker, 1960). Beta diversity patterns across gradients are used to understand the underlying factors that causes variation in diversity, i.e., the processes causing a pattern (Baselga, 2010). It is thus crucial to understand sampling effects on beta diversity assessments, especially in plant-animal interactions that can be sampled by many different methods (Vizentin-Bugoni et al., 2018). Moreover, not only species but their interactions are known to vary across ecological gradients (Poisot et al., 2012; Trøjelsgaard et al., 2015; Tylianakis & Morris, 2017), including plants and their pollinators (Burkle & Alarcón, 2011; Carstensen et al., 2014; CaraDonna et al., 2017). Trøjelsgaard et al., 2015; CaraDonna et al., 2017). A better comprehension of interaction beta diversity would help better predict pairwise interactions, and ultimately contribute to conservation of interactions and associated ecosystem services (Trøjelsgaard et al., 2015; Burkle et al., 2016).
The plant-centred approach is by far the most common one in the literature on pollination networks (Vizentin-Bugoni et al., 2018), possibly because data collection is easier and does not require the tedious laboratory work in identifying pollen grains carried by pollinators afterwards. This overrepresentation of one method in relation to another may be especially relevant, as indices derived from the two methods sometimes lead to distinct estimates of network structure (Bosch et al., 2009; Ramírez-Burbano et al., 2017; Zhao et al., 2019). For instance, the addition of pollen load to visitation data recovered higher species degree and connectance in a Mediterranean pollination network (Bosch et al., 2009).
While the plant-centred method concentrates on static organisms, the animal-centred method focuses on highly mobile organisms, usually with flight abilities. This fundamental difference in mobility between the two focal groups also means that while there is a delimited spatial sampling unit in the plant-centred method, this is not the case in the animal-centred approach which captures pollinator species with distinct mobility (Jordano, 2016; Ramírez-Burbano et al., 2017) and potentially comprises a larger spatial context. Since flying organisms tend to show lower beta diversity across communities than nonmobile organisms such as plants ( et al., 2014), it is likely that sampling method affects the estimates of beta diversity of interactions.
Here we analysed plant-pollinator interaction networks from the largest continuous floodplain in the world, the Pantanal. We ask: (1) how sampling methods (animal and plant-centred approaches) influences the patterns of beta diversity of interactions, network metrics and species level indices; (2) if there are differences in species-level indices assessment related to plant traits (e.g., pollinator system, resource offered, size and type of flowers). Due to differences in vegetation structure and composition in the Pantanal landscape, we expect considerable turnover of species and interactions across habitats ( & Morris, 2017). We also expect to find lower beta diversity of interactions in animal-centred networks, owing to higher mobility of pollinators which could blur the spatial limits of a plant-centred approach and yield lower variability on interacting partners across space. We expect differences on the assessment of species level metrics related to plant traits because it is known that pollen grains may adhere more or less to the bodies of pollinators, or they may have different behaviours during the visit to the flowers, affecting estimates of interaction specialization (Ballantyne et al., 2015; Zhao et al., 2019). In addition, plants presenting flowers with more open generalized morphologies and pollination systems will possibly be associated with higher differences in species level metrics between the two sampling methods, as they receive a greater number of opportunistic visits from less specialized pollinators that can vary considerably in pollen loading. Therefore, we designed this study to fill in a key gap in the understanding of how the sampling approach affects the assessment of interaction networks’ beta diversity, and hence the inferences on spatial variation of ecological networks.