Food Webs: How a Disturbance Travels Through an Ecosystem

Removing a single link from a chain can be enough to break it. In an ecosystem, removing one species can trigger a cascade of changes that travels through several levels and transforms entire landscapes. This propagation is not random: it follows the structure of the food web — the complete set of feeding relationships connecting organisms in a habitat.

Food web ecology is a discipline that combines population biology, graph theory and thermodynamics. It seeks to understand how energy and matter flow through ecosystems, how disturbances propagate, and which structural properties confer stability or fragility.

2. Structure of a Food Web

A food web is a directed graph. Nodes represent species or functional groups (for example, "medium-sized herbivores"). Directed edges indicate predation relationships: an arrow from A to B means A is consumed by B. Network density — the ratio of observed links to possible links — is a measure of connectivity.

Real food webs are rarely simple linear chains. Most species have multiple prey and multiple predators. Omnivory — consuming at multiple trophic levels — is common. Feedback loops exist: a predator may consume a prey that itself consumes a prey of the predator. These complex structures make analysis difficult but also more realistic.

Functional redundancy refers to the presence of multiple species filling a similar ecological role. A network with high redundancy is more robust to the loss of one species: if one disappears, others can compensate. A network without redundancy is vulnerable to cascading secondary extinction.

Energy enters an ecosystem primarily through photosynthesis. Primary producers — plants, algae, cyanobacteria — convert solar energy into biomass. Primary consumers (herbivores) feed on producers. Secondary consumers feed on herbivores, and so on. Decomposers — bacteria and fungi — recycle dead organic matter.

At each trophic transfer, a large fraction of energy is dissipated as heat through respiration. Trophic transfer efficiency is typically around 10%: to produce 1 kg of predator requires about 10 kg of prey, which required 100 kg of plants. This 10% rule is a rough approximation — real values range from 5 to 20% depending on the system — but it explains why food chains rarely exceed four or five levels.

The trophic pyramid represents biomass or energy at each level. It is generally wider at the base (producers) and narrower at the top (apex predators). Inversions are possible in some aquatic systems where phytoplankton turns over very rapidly despite low instantaneous biomass.

The keystone species concept was introduced by Robert Paine in 1969 from experiments on Pacific rocky shores. By removing the sea star Pisaster ochraceus, Paine observed that mussels — its main prey — invaded the space and eliminated most other species. The sea star, though not abundant, maintained community diversity. Its influence was disproportionate to its biomass.

A trophic cascade occurs when modification of one trophic level propagates downward (top-down cascade) or upward (bottom-up cascade). In a top-down cascade, reduction of a predator releases its prey, which can then overexploit their own resources. In a bottom-up cascade, reduction of primary producers propagates to higher levels.

Trophic cascades have been documented in many systems: lakes, oceans, tropical forests, savannas. Their intensity varies considerably. In some systems, effects are strong and well documented; in others, they are weak or masked by other factors. Generalization is risky.

The reintroduction of gray wolves to Yellowstone National Park in 1995–1996 is one of the most cited examples of a trophic cascade. The simplified version of the story is: wolves reduced elk populations, which had been overgrazing willows and cottonwoods in river valleys; vegetation recovered, stabilizing riverbanks, altering stream channels and benefiting beavers, birds and other species. Some authors even spoke of "rivers of fear" to describe how the mere presence of wolves altered elk spatial behavior.

The reality is more complex. Subsequent studies showed that wolf effects on vegetation vary by zone within the park, climatic conditions, and the presence of other factors such as drought, fire and human pressure. Ripple and Beschta (2012) documented positive effects on riparian vegetation, but other researchers noted that vegetation recovery had begun before wolf reintroduction in some areas, and that the behavioral effects of wolves on elk are difficult to isolate from numerical effects.

Yellowstone illustrates both the reality of trophic cascades and the difficulties of interpretation in complex natural systems. A single factor — wolf reintroduction — cannot explain all the observed changes. The mechanisms are multiple, interactive and context-dependent. This is precisely what food web theory predicts: effects propagate, but their magnitude depends on network structure and local conditions.

The stability of a food web can be defined in several ways. Local stability refers to return to equilibrium after a small perturbation. Resilience refers to the speed of this return. Robustness refers to the capacity of the network to maintain its structure and functions after the loss of one or more species. These three properties are not equivalent and can even be in tension.

Robert May showed in 1972, in a foundational paper, that large random networks with high connectivity tend to be unstable. This result, counterintuitive relative to the popular idea that diversity promotes stability, triggered a debate that continues today. Subsequent work showed that the non-random structure of real networks — with few strong links and many weak links — can reconcile complexity and stability.

Secondary extinction occurs when the disappearance of one species causes the disappearance of other species that depended on it. In a highly connected network, an extinction can propagate in cascade. A network's vulnerability to this phenomenon depends on its topology: networks with highly connected nodes (hubs) are vulnerable to the loss of these nodes but robust to random extinctions.

The Lotka-Volterra model describes interactions between two species — a predator and a prey — through two coupled differential equations. The prey grows exponentially in the absence of predator and is consumed at a rate proportional to the product of both populations. The predator grows through prey consumption and declines through natural mortality. This model produces cyclic oscillations in the phase plane.

The generalized model extends this approach to n species. For each species i, the dynamics are $\frac{dN_i}{dt}=N_i\left(r_i+\sum_j\alpha_{ij}N_j\right)$, where $r_{i}$ is the intrinsic growth rate and αᵢⱼ is the interaction coefficient between species i and j. The matrix A = (αᵢⱼ) encodes the entire network structure. The stability of the equilibrium depends on the eigenvalues of the Jacobian matrix evaluated at that equilibrium.

Energy-flux-based approaches offer an alternative. Rather than modeling abundances, they track carbon or nitrogen fluxes between compartments. These models are easier to parameterize from biogeochemical data, but they lose information about individual population dynamics.

Overfishing is one of the most documented trophic disturbances. Removal of large marine predators — sharks, tuna, cod — releases their prey and can trigger cascades that alter the structure of marine communities. In several regions, overfishing of apex predators has led to jellyfish and sea urchin proliferation and the collapse of seagrass beds.

Invasive species introduce new links into the network. They can be predators without natural enemies, superior competitors or disease vectors. Rainbow trout introduced into fishless mountain lakes eliminated the amphibians and aquatic invertebrates they fed on, profoundly altering the structure of lake communities.

Climate change alters food webs in several ways: range shifts, phenological mismatch (prey and predators no longer in the same place at the same time), changes in growth and metabolic rates. These changes can break coevolved interactions and create new vulnerabilities.

Building a complete food web is a difficult undertaking. Interactions are numerous, often rare, seasonal or size-dependent. Field data are costly to collect and often incomplete. Published networks are approximations that aggregate species into functional groups and ignore many weak interactions.

The debate between complexity and stability remains open. May's (1972) work suggests complexity destabilizes; empirical studies and more realistic models suggest that the non-random structure of real networks can compensate for this effect. The answer likely depends on the type of complexity (diversity, connectivity, link strength) and the type of perturbation.

The propagation of a disturbance through a food web is not a spiral in the geometric sense. It is a diffusion through a graph, with direct and indirect effects, amplifications and attenuations depending on network structure. The spiral metaphor can illustrate the idea that effects amplify or propagate in concentric circles, but it must not be confused with a precise mechanism.

An arrow between species can mean “eats,” “is eaten by,” or “changes the abundance of.” These conventions are not interchangeable. Our first sketch follows biomass from resource to consumer: plants → herbivores → predators. It indicates possible transfers of matter and energy, not their quantities or efficiencies. The three boxes are functional groups in an abstract example, not measurements from a particular site.

Real food webs contain additional levels, detrital pathways and omnivores. Links also depend on life stage: an organism may change diet during development. A map is therefore a documented hypothesis about interactions at a particular scale and time. Paine’s classic experiment illustrates why relationships need to be connected to observations and perturbations to understand diversity beyond simply counting species. ^[1]

In the second sketch, arrows now describe a local effect on another compartment’s growth. More predators tend, other things equal, to increase herbivore mortality: a negative sign. More herbivores can provide more resources to predators: a positive sign. One feeding relationship thus yields two effects of opposite sign. A positive arrow does not mean a benefit to the whole ecosystem; it describes the direction of a local response.

The matrix J collects these derivatives near a particular state. Signs determine neither strength nor stability: networks with identical signs can respond differently when interaction strengths, timescales or self-regulation differ. A temporary population increase also differs from permanently removing a species. Interpreting a result requires specifying the perturbation and response variable, then distinguishing immediate response from a new equilibrium.

Along the single path predators → herbivores → plants, two negative signs multiply: more predators may reduce herbivory and favor plants. This is an indirect positive effect through an intermediate compartment. “May” matters: multiplying signs predicts a direction for this isolated path, not the total response of the whole web. The curved green arrow summarizes the reasoning; it does not depict a species directly eating a plant.

Paine’s study concerns an intertidal system and predator manipulation with changes in community composition. A universal three-level drawing cannot substitute for that study. Testing a cascade elsewhere requires controls, an appropriate observation period, measurements of herbivory or other relevant interactions, and attention to differences among sites. Sketches help formulate paths to test; data establish their importance. ^[1]

Now add direct plant consumption by the upper compartment. Two paths connect it to plants: the previous indirect path is positive, while the direct path is negative. Signs alone do not tell us which dominates. We would need interaction strengths and how they vary with abundance. Network complexity guarantees neither stability nor instability; it increases the number of mechanisms a model must make explicit.

When a spiral appears in an ecological phase portrait, it describes joint changes in abundances, such as prey and predators. It is not inherent in a food-web drawing. Different equations and parameters can yield nonoscillatory convergence, damped oscillations, a limit cycle or other behavior. Talking about a spiral therefore requires measurable axes, states tracked over time and a checked dynamic model: a loop drawn in a network is not an observed rotation in state space.