Food webs have long been the archetypal road map for interaction networks in ecosystems. However, the relative importance of trophic interactions as compared to other interaction types (e.g. competition and mutualism) in ecological networks is not yet well understood. Co-occurrence analyses are commonly used for quantifying ecosystem interactions generally; however, these methods do not capture the non-linear nature of ecosystem interactions. Further, both food webs and correlation-based networks tend to depict a static ecosystem; they do not show that interactions may change over time, or only occur under specific conditions (e.g. competition over a shared nutrient only when the nutrient is scarce).
Here we utilize convergent cross mapping (CCM), a method that measures dynamic causation among ecosystem variables from time series data. Unlike correlation-based measures of causation, CCM accommodates the fact that interactions in real ecosystems are ever-changing: they can appear and disappear, be episodic, or change sign in ways typical of non-linear dynamical systems.
We construct causal interaction networks for five distinct ecosystems (aquatic and terrestrial) that were notable as long-term monitoring studies with extensive time series data spanning multiple decades. From these networks, we analyze how interactions and network structure change over time.
Results/Conclusions
Food web links are indeed central to the interaction network in that every trophic link is measured to be a significant interaction; however, across all ecosystems, food web links only account for about half of the significant interactions. Further, correlation in species abundances does not imply two species interact: less than half of all significant causal interactions occur between cross-correlated species.
We also find that interactions are dynamic, changing from significant to non-significant in strength. We find that interactions are significant (p < 0.05) for only ~10% of the time on average; a percentage surprisingly consistent across all systems studied. Because only a fraction of interactions identified over the course of the full study may be occurring in any given year, ecosystem connectance (# of interactions / # possible interactions) changes over time. The variability of connectance shows interesting dynamics that are likely driven by multiple factors such as seasonal cycles, population cycles and changes in individual interactions (e.g. diet-switching). We also find that connectance may also be indicative of system-wide disturbances (e.g. flood) or regime shifts (e.g. introduction of a keystone species). This work provides evidence that ecosystems are much more connected and dynamic than food webs would suggest.