Food webs

Abstract

Ecosystems, whether they are forests, lakes or coral reefs, consist of a diverse range of species that interact with each other in many ways. As ecologists we want to understand how these complex systems function as a whole and how they respond to human impacts such as harvesting and environmental change. A mechanistic understanding of the underlying biological processes will allow us to predict the cascading effects of species extinctions and species invasions, something that is particularly relevant at present in view of predicted species range shifts and expansions in response to climate change. The first step in this process is to reduce the seemingly endless complexity of an ecosystem down to a form that can be studied systematically. A food web is essentially a distilled version of this complexity. It is the network of feeding interactions in the ecosystem, with species as the nodes that are linked when one feeds on the other (Figure 1). Describing ecosystems in this way opens them up to network analysis, a field that has been developed especially in the social and information sciences and that has now become a truly multi-disciplinary science. As is explained in the first section that follows, this approach allows us to quantify structural characteristics of an ecosystem and compare these among systems. The second section then deals with modelling food webs. With dynamic models we can ask how the network properties affect the stability or resilience of a system, and with assembly models we can test our theories of how ecosystems assemble themselves by comparing the network structure of real webs with those predicted by theory. In order to describe the food web of an ecosystem, one needs to observe the feeding (trophic) interactions between the species. Ideally the food web will contain all the species and trophic links within a more or less self-contained system but, for simply practical reasons, this is rarely possible. Webs that do describe the whole network of interactions are referred to as community webs. In reality, they can rarely be complete; microbes and parasites in particular are often excluded. Groups of species are often combined into single nodes referred to as trophic species. The assumption is that the diversity within such a group is essentially redundant because all species share the same prey and natural enemies. Whether this is always justified is debatable. An alternative approach is to sacrifice inclusiveness for resolution by focusing on clearly defined subsets of the community as a whole. Examples of this are source webs, which trace the feeding links from a single basal species to all the species that consume it and from there to higher trophic levels, and sink webs, which trace the links from a predator back down the food chain. Another approach is to focus on distinct taxonomic groups. Webs of insect hosts and their parasitoid wasps are particularly popular subjects in this respect because they are clearly defined subsets of the community and the feeding interactions can be readily observed and quantified by the rearing of parasitoids from the hosts. The structure of a network can be characterized using a range of descriptors. These are all based on the number of species (nodes), the number of links and the distribution of the links. An important concept is that of complexity or connectance: the proportion of all possible links that is realised. A number of equations for calculating connectance can be encountered and it is important to understand the underlying assumptions (Figure 2). For a given connectance, the distribution of the links in the network can still vary. For example, links can be clustered, dividing the network up into distinct compartments of strongly interacting species that are only weakly linked to other such compartments, or links may show a nested distribution or be randomly distributed. A simple intuitive example of nested links is the situation that has been observed especially in aquatic food webs, where predation appears to be mostly size-based. Here the diet of smaller predators tends to be a subset of the diet of larger predators, and the diet of more specialised feeders therefore tends to be a subset of that of the generalist feeders (rather than being prey that the generalists do not feed on). Another non-random pattern in the distribution of links that is commonly observed in food webs is that food chains are generally short. This means that the vast majority of species are only a few steps removed from the abiotic environment on which the producers feed. There are good reasons for this. First, the inefficiency of energy transfer from prey to consumer (5–15%) means that, as you move higher up the food chain there is less and less biomass to go around. There is also a population dynamic argument: Longer food chains are less stable, so top predators are more prone to extinction. These two explanations are not mutually exclusive and evidence exists for both. As more food webs data sets are collected with quantitative data on the population sizes and the strengths of interactions, new food web statistics have been developed to take this extra information into account. These methods are based on Shannon information theory, which in ecology is best known for its use in quantifying diversity. Put simply, the diversity index is at its maximum value, equal to the number of species, when all species are equally abundant. The less even the abundances are, the lower the diversity index. The same principle can be applied to the links in a food web when their strengths have been quantified, and so food web descriptors can be derived that are not based on raw numbers of species and links but on their diversities as defined by the Shannon index. Comparisons of these quantitative methods with their binary counterparts have shown that the former are far less sensitive to the effects of rare species and rare links, and are thus less sensitive to sampling effort. The quantitative measures have also revealed, for example, that the intensity of land use can affect the evenness of link strengths in a food web, with important implications for ecosystem functioning, while the binary measures of the same webs revealed no effects. In the early 1970s, Robert May showed that theoretically the stability of food webs should decrease with increasing complexity (connectance). One can imagine why this would be the case: in a highly connected network, effects on one species can easily propagate through the wider food web because most species will only be a few links removed from the affected species. However, ecological networks are typically highly complex, which begs the question how they can persist. One possible explanation is in the distribution of links: May assumed that, for a given connectance, the links are distributed randomly among the nodes, and all the available evidence shows that the distribution of links within a food web is far from random. For one thing, the food chains in random webs are longer than in real webs and we have already seen that this can affect stability. These observations led to the development of a succession of new food web assembly models in which the distribution of links is in some way constrained. The first example of this was the Cascade model, in which species are ordered along some biological gradient or niche dimension (usually thought of as body size), and species can only feed on other species that have a lower value along this dimension, with the probability of a link being realised dependent on connectance. This model captures a number of aspects of real food webs reasonably well, such as species richness at different trophic levels but it fails to reproduce food chain length, for example. One reason why the Cascade model does not reproduce realistic food webs is the constraint that species can only feed on others that have a lower ‘niche value’, and features of real food webs such as cannibalism and feeding loops are therefore not possible. This was addressed by the Niche model in which a predator is linked to all species within a ‘niche range’ along the gradient that may include the predator itself and species with a higher value. This approach produces model webs that better match the structure of real food webs. If we assume the niche dimension along which the species are ordered to be body size, then the Niche model assumes that a consumer will eat all prey types within a given size range, regardless of their taxonomic type. This is an assumption that is difficult to defend because the diet of most species shows some degree of taxonomic specialisation. This is why we classify animals into herbivores and carnivores, but also into more specialised groups such as insectivores and molluscivores. The Nested Hierarchy Model captures this level of organisation by assuming that, if one consumer shares one prey species with another, then the rest of its prey species are more likely to include the prey species of that other consumer rather than any randomly chosen species. This succession of models has shown that, for a given number of species and a given number of links, the application of some simple rules can reproduce many aspects of food web structure. They do have their limitations though. First, like many of the early empirical food webs, they produce binary networks that depict the presence and absence of species and links but contain no information on the abundances of species and the strength of the interactions. As already mentioned above, including quantitative information can significantly affect the interpretation of food web structure, and weak links have a potentially stabilising effect on food webs so information on link strength is important. Second, these models produce static webs, while real food webs are highly dynamic with feeding links appearing and disappearing or varying in strength as species adjust their diet over the course of a season to suit their nutritional needs, for example, or in response to the population dynamics of potential prey. Third, all these models require as their input the number of species and the number of links, so they can never answer the fundamental questions of the mechanisms that determine the diversity and complexity of an ecological community. Recent developments in the field are moving away from this largely phenomenological approach towards a more mechanistic understanding of the processes that structure food webs and in so doing address some of the limitations of the previous models. Optimal foraging theory is a branch of behavioural ecology that has been very successful in providing an evolutionary explanation of the diet choices of individual animals. The Diet Breadth Model applies this theory to predict the connectance of real webs. The Allometric Diet Breadth Model then assumes that foraging decisions of individuals are dependent on their own body size and that of available potential prey, and by doing this it can go one step further: Not only can it be used to predict the general pattern of a food web, for example its connectance, it can also predict the individual links that make up this pattern. This simple, size-based foraging model can in some cases successfully predict up to 65% of the actual links in a food web. A great advantage of this is also that these food web assembly models could be incorporated into dynamic food web models in which the realisation of links and their strength can vary in response to population dynamics. Focusing on size-based foraging decisions in this model represents a return to the single niche dimension of the Cascade and Niche models and it is likely that once it too is extended to include taxonomic constraints on diet, as in the Nested Hierarchy Model, its explanatory power will further improve. Many models aimed at describing the dynamics and stability of food webs essentially assume that these networks are self-explanatory: All the necessary information for predicting their behaviour is contained with the network of trophic interactions. It is important to remember, however, that, despite their complexity, food webs are still simplified representations of the ecosystem as a whole. Organisms interact in many ways that do not always involve eating each other. The presence of members of one species, for example, may affect the behaviour of another species which in turn may affect its trophic interactions. This happens when plants release volatile chemicals in response to herbivores feeding on them which then attract the natural enemies of the herbivore, or when the alarm calls of one species alert another to the presence of a predator. There are many empirical examples of these and related phenomena, usually referred to as trait-mediated effects or interaction modifications. Microcosm experiments and modelling approaches have demonstrated that such effects can stabilise food webs that from the trophic interactions alone are predicted to be unstable. Advances in our understanding of the behaviour of the complex systems that food webs are will likely come from dynamic models that incorporate these kinds of behavioural effects along with the individual foraging decisions of the Allometric Diet Breath Model with appropriate phylogenetic constraints.