In both strictly theoretical and more applied contexts it has been historically assumed that metapopulations exist within a featureless, uninhabitable matrix and that dynamics within the matrix are unimportant. In this article, we explore the range of theoretical consequences that result from relaxing this assumption. We show, with a variety of modeling techniques, that matrix quality can be extremely important in determining metapopulation dynamics. A higher-quality matrix generally buffers against extinction. However, in some situations, an increase in matrix quality can generate chaotic subpopulation dynamics, where stability had been the rule in a lower-quality matrix. Furthermore, subpopulations acting as source populations in a low-quality matrix may develop metapopulation dynamics as the quality of the matrix increases. By forcing metapopulation dynamics on a formerly heterogeneous (but stable within subpopulations) population, the probability of simultaneous extinction of all subpopulations actually increases. Thus, it cannot be automatically assumed that increasing matrix quality will lower the probability of global extinction of a population.

The capacity of highly diverse systems to prevail has proven difficult to explain. In addition to methodological issues, the inherent complexity of ecosystems and issues like multicausality, non-linearity and context-specificity make it hard to establish general and unidirectional explanations. Nevertheless, in recent years, high order interactions have been increasingly discussed as a mechanism that benefits the functioning of highly diverse ecosystems and may add to the mechanisms that explain their persistence. Until now, this idea has been explored by means of hypothetical simulated networks. Here, we test this idea using an updated and empirically documented network for a coffee agroecosystem. We identify potentially key nodes and measure network robustness in the face of node removal with and without incorporation of high order interactions. We find that the system's robustness is either increased or unaffected by the addition of high order interactions, in contrast with randomized counterparts with similar structural characteristics. We also propose a method for representing networks with high order interactions as ordinary graphs and a method for measuring their robustness.

No abstract