Introduction
Ecologists recognize that species interactions are a cornerstone in determining biodiversity, community assembly, and ecosystem functioning (Bascompte & Jordano 2007; Goudard & Loreau 2008; Tonkin et al.2017). Thus, species interactions have far reaching biological implications (Tylianakis et al. 2008). For instance, species interactions are used in the evaluation of community stability (Ives & Carpenter 2007) and hence are essential for determining community resilience and resistance to perturbations. In addition, positive species interactions have been shown to produce species-rich model communities among competitors supported by a single limiting resource (Gross 2008), and facilitate increased species richness in harsh environments (Cavieres & Badano 2009). As well, interactions have been shown to mediate the negative effects of environmental change (Brooker 2006; Suttle et al. 2007), and hence are likely essential for continued ecosystem persistence in the face of global change.
While species interactions vary across both space and time (Rasmussenet al. 2013), fewer studies have focused on the changes in temporal interactions (Alarcón et al. 2008; Thompson et al. 2012) despite evidence linking them to the coexistence of competing species (McMeans et al. 2020). More recently, temporal heterogeneity, in particular seasonality, has been invoked to explain biodiversity and community structure (Tonkin et al. 2017). For example, seasonality has been shown to minimize competitive interactions and help stabilize total species abundances (Shimadzu et al.2013), as well as play a role in maintaining structure and diversity in communities (e.g., Fitzgerald et al. 2017). Consequently, as many environments experience seasonal oscillations (Tonkin et al.2017) with different community compositions being favoured across different seasons, the influence of seasonality on species interactions is widespread.
Typically, species interactions are analyzed using network theory (Rasmussen et al. 2013) and seasonality can be incorporated by constructing and comparing separate networks representing different seasons. Often these types of ecological interaction networks are constructed using species co-occurrence methods (Freilich et al.2018). Traditionally, these approaches assume that species with significant co-occurrence have beneficial interactions while species with significant co-exclusion have negative interactions (Cazelleset al. 2016). However, co-occurrence data and their corresponding methods have been criticized for elucidating false interactions and for failing to detect true pairwise species interactions (Blanchet et al. 2020). A false interaction may arise due to species responding similarly to the same environmental factors (Peres-Neto et al.2001) while true interactions may not be detected due to the coarseness of presence/absence data (Sander et al. 2017). More recently, joint species distribution models (e.g., Pollock et al. 2014; Ovaskainen et al. 2016) have been touted as a more robust method to infer community structure over other co-occurrence methods as they incorporate abiotic factors into their analysis (D’Amen et al.2018). However, while this approach controls for environmental factors, it is still likely limited by its reliance on co-occurrence data to infer species interactions (Blanchet et al. 2020). A promising approach proposed by Momal et al. (2020) addresses these issues by utilizing species abundances, instead of presence/absence data, with a joint species distribution model. Including abundance measures provides richer information for capturing interactions (Blanchetet al. 2020) while the inclusion of environmental factors help prevent spurious interactions in the network.
One of the strong utilities of using species interaction networks are the many developed tools available for comparing networks (e.g., Delmaset al. 2019). Of particular interest are metrics that evaluate topological differences due to species turnover and interaction rewiring (i.e. the changes in the interactions between the same species across space or time despite both species remaining present) (Poisot et al. 2012). While one of these two processes may be more dominant in a system than the other, these processes are not mutually exclusive. For example, the temporal changes in plant-pollinator networks are the result of both species turnover and rewiring (Alarcón et al.2008; Petanidou et al. 2008; Olesen et al. 2011) and not a singular process. Identifying the relative contribution(s) of each process to network topology is important as systems dominated by rewiring, rather than species turnover, may be more robust to perturbations (CaraDonna et al. 2017). Additionally, as traits have been shown to be an important driver of ecological network structure (Eklöf et al. 2013), determining how traits relate to rewiring is a critical component for understanding ecosystem dynamics.
Due to the sampling effort required, few systems have the appropriate biological data needed to produce temporal interaction networks (Alarcónet al. 2008). The difficulty of collecting large spatial or temporal-scale data primarily limits seasonal network analysis to physically small organisms such as those that make-up bipartite plant pollinator networks (e.g., Alarcón et al. 2008; Petanidouet al. 2008; Olesen et al. 2011; Rasmussen et al.2013; Burkle et al. 2016; CaraDonna et al. 2017). However, notable exceptions include seasonal networks of tropical fish (Winemiller 1990), frugivorous birds (Carnicer et al. 2009), a forest predator-prey community (Saavedra et al. 2016), and a consumer-resource intertidal community (Lopez et al. 2017). Given the release of a recent time series dataset of stream fish abundances for sites across the United States (NEON 2020), there is a unique opportunity to investigate seasonal networks in a freshwater stream ecosystem. Since stream communities experience regular seasonal variations from differences in shading, temperature, disturbance, and productivity (Thompson & Townsend 1999), they are likely a valuable study system to understand the effects of seasonality on multiplex networks.
Here, we investigate seasonal changes in a freshwater stream fish community using NEON data (NEON 2020) and quantify both species turnover and interaction rewiring therein. Specifically, our objectives are to: (i) determine if there are measurable differences between Fall and Spring networks, using modularity and interaction turnover (i.e. beta-diversity); (ii) quantify the influence of interaction rewiring and species turnover in these communities; and, (iii) evaluate whether seasonal changes in species interactions are related to species-specific traits. To do so, we apply the recently proposed method of Momalet al. (2020) to construct two freshwater species multiplex networks for Fall and Spring. We find that: (i) Spring interaction networks have higher modularity than Fall networks, and between seasons, there is a large amount of interaction turnover; (ii) most topological change across seasons are the result of species rewiring (c.87%) as compared to species turnover (c.13%); and (iii) species’ maximum length and its piscivore/non-piscivore status help explain a species total number of rewiring connections (adjusted R2=0.34).