The identification of non-trophic species interactions has a long history in ecology, and there has been renewed impetus to estimate these interactions from species co-occurrence data. Most such attempts have been applied to spatial co-occurrence data, using partial correlations, Markov random fields, and joint species distribution models. Using high-resolution phytoplankton records from lake sediments, the long-term dynamics of phytoplankton have revealed how these communities have responded to environmental forcing on scales from years to millennia. In response to eutrophication and climate change, our previous work has identified a switch from compensatory to synchronous dynamics, and identified the core groups of species associated with this change. Here, I used exponential random graph models (ERGMs) to identify the specific taxa associated with these changes, estimate the strength of these associations, and how these have varied in response to environmental change.
Results/Conclusions
Graphical models have been widely used to recover the underlying structure of networks. Here, species are the nodes in the network or graph whilst edges between nodes describe the strength of the association between species. Higher-order associations are defined by considering not just the neighbours of a particular node but also the neighbours’ neighbours. Typically, the conditional distribution of a node is assumed to be Gaussian; this is often a poor choice for species abundance data. ERGMs extend the choice of conditional distribution to the exponential family, in the same way that generalised linear models extend linear models. In a northern Swedish lake, the ERGM identified a strong set of positive inter-relationships between Navicula minima and Fragilaria brevistriata, two benthic-associated taxa, and the planktonic taxon Aulacoseira lirata, which were strongly negatively associated with A. distans, another planktonic taxon. These associations are only strongly identified in the 400-year period prior to an extended extremely cold climate anomaly ~ 75BCE. This period was also where compensatory dynamics were strongly observed in the diatom community, and in combination suggest quasi-periodic variation in the relative partitioning of resources in the lake and competition for nutrients sources, with A. distans being abundant during periods where supply of nutrients from the catchment was relatively high. These results reflect indirect qualitative inferences drawn from the data using standard ordination techniques.
ERGMs can be extended to derive a smoothly-varying set of networks. We discuss this extension with some early results and address recently raised questions as to the ability of approaches to accurately recover species interactions from co-occurrence data.