Compartmentalization increases food-web persistence

Abstract

It has recently been noted that empirical food webs are significantly compartmentalized; that is, subsets of species exist that interact more frequently among themselves than with other species in the community. Although the dynamic implications of compartmentalization have been debated for at least four decades, a general answer has remained elusive. Here, we unambiguously demonstrate that compartmentalization acts to increase the persistence of multitrophic food webs. We then identify the mechanisms behind this result. Compartments in food webs act directly to buffer the propagation of extinctions throughout the community and augment the long-term persistence of its constituent species. This contribution to persistence is greater the more complex the food web, which helps to reconcile the simultaneous complexity and stability of natural communities.

Fig. 1.

Potential effects of compartmentalization. (A) A hypothetical compartmentalized food web made up of three compartments. We explore here the possible consequences of a single species extinction. Suppose that the species highlighted in red and indicated by the arrow goes extinct. (B) Upon the extinction of this species, it is hypothesized that compartmentalization will predominantly restrict the effects of perturbations to within the same compartment as opposed to other compartments. We would then expect that secondary extinctions (the enlarged species highlighted in the color of their respective compartment) are more probable within the same compartment. (C) Everything else being equal, the effects of extinction of a species will be felt throughout the community, independent of where the initial extinction occurs, if compartmentalization had no influence.

Fig. 2.

Effect of compartmentalization on food-web persistence. (A) Mean contribution of compartmentalization—quantified by modularity—to the long-term persistence of species in the community. The greater the compartmentalization of a food web is, the greater the persistence of its constituent species. The SEs of the reported averages are shown as error bars but are small. (B) The range of compartmentalization observed in 15 empirical food webs (see SI Materials and Methods, Section S1 for a list of the empirical webs). The middle line marks the median, the box covers the 25th–75th percentiles, and the maximum length of each whisker is 1.5 times the interquartile range. Points outside this range show up as outliers.

Fig. 3.

Community response to manipulated species extinctions. (A) Mean relative number of extinctions that occur in the same compartment as an eliminated species, as a function of the web's modularity. Values greater than zero imply that the subsequent species that go extinct as a consequence of the original extinction have a higher probability of belonging to the same compartment. (B) Mean relative time to extinctions that occur in the same compartment as the eliminated species, as a function of the web's modularity. Values less than zero imply that these species tend to go extinct earlier, as a consequence of the original extinction. The SEs of the reported averages are shown as error bars.

Fig. 4.

Propagation of extinctions within food webs. (A) We compare the ability of different factors to predict the next species to go extinct after the earlier extinction of a species in a food web. As the connectance of the food web increases, the tendency to observe consecutive extinctions of directly connected species decreases (white triangles). For species within the same compartment, the same tendency increases with increasing connectance (red circles). Values close to zero imply that this tendency is close to the random expectation. (B) We separate within-compartment extinctions into those that occur between (i) directly connected species (gray squares) and (ii) nondirectly connected species (blue diamonds). We find that the probability of consecutive extinctions between two nondirectly connected species shows a strong increase with increasing connectance. The SEs of the reported averages are shown as error bars but are small. (C) The range of connectance observed in 15 empirical food webs, as in Fig. 2.