Food Webs and Community Structure

A food chain is a linear sequence of "who eats whom" — producer → herbivore → carnivore → top carnivore. It is pedagogically useful but almost never describes reality, because most species feed at multiple trophic levels and are fed upon by multiple predators. A food web replaces this line with a directed network: nodes represent species, functional groups, or trophic species (sets of taxa with identical predators and prey), and directed edges represent energy-yielding feeding links, drawn from prey to predator in the direction of energy flow. Several quantities let ecologists compare webs of wildly different size and habitat. Species richness (S) is the node count. Linkage density (L/S) is the average number of links per species. Connectance (C) is the fraction of all possible directed links that actually occur, C = L / S², where S² is the maximum possible number of feeding links (including cannibalism and omnivory) among S species. A trophic level is a position in the chain of energy transfer — primary producers occupy level 1, herbivores level 2, primary carnivores level 3, and so on — though in real webs many species (omnivores) occupy fractional or multiple levels simultaneously, so trophic level is often calculated as a weighted average of a consumer's prey's levels plus one.

Food chain length (FCL) is usually measured as the number of links (or levels) along the longest — or, more robustly, the mean — path from a basal resource to a top consumer. Two forces are said to regulate community structure from opposite directions: bottom-up control, where the abundance of primary producers (set by light, nutrients, water) limits productivity and biomass at every level above it, and top-down control, where predators suppress prey populations enough to release the resources those prey would have consumed, producing alternating high-low-high biomass patterns down the chain — the trophic cascade. Neither force acts alone; virtually every community is shaped by both simultaneously, and much of community ecology since the 1960s has been devoted to measuring their relative strength.

Step 1 — Defining the nodes. Constructing a food web begins with deciding the resolution: individual species, functional guilds (e.g., "filter-feeding zooplankton"), or life stages (a tadpole and a frog may occupy different trophic positions). Coarser resolution loses biological detail but makes cross-ecosystem comparison tractable — this is why most published connectance values use "trophic species" rather than raw taxonomic species counts.

Step 2 — Mapping the links. Each feeding link is established empirically: gut-content analysis, stable-isotope ratios (δ¹⁵N enriches ~3–4‰ per trophic step, allowing quantitative trophic-level assignment), stomach-flushing, DNA metabarcoding of prey remains, or direct observation. The result is an adjacency matrix — rows as predators, columns as prey (or vice versa) — with a 1 for a realized link and 0 otherwise.

Step 3 — Energy flows and shrinks at each transfer. Only a fraction of the energy consumed at one trophic level is converted into biomass available to the next. Gross ecological efficiency (biomass production at level n+1 ÷ biomass production at level n) averages around 10% (Lindeman's "ten percent law," itself an approximation with a real range of roughly 5–20% depending on taxon), because of respiration losses, egestion of undigested material, and energy spent on movement, reproduction, and heat (especially in endotherms). This energetic bottleneck is the principal reason food chains cannot lengthen indefinitely: a chain long enough to need a level-6 predator would require an implausibly large base of primary production to support even one individual at the top.

Step 4 — Connectance responds to richness and to a "link-scaling" constraint. As species richness S increases, empirical surveys show total link number L increases roughly linearly with S (roughly L ≈ 2S, the "link-species scaling law" proposed by Cohen and colleagues), which means connectance C = L/S² declines with richness — bigger webs are proportionally sparser, not denser. This is thought to reflect a biological ceiling: each consumer can physiologically process, recognize, and specialize on only a limited number of prey types regardless of how many species exist in the community, a constraint related to the foraging-efficiency trade-offs behind Resource Partitioning.

Step 5 — Top-down control propagates as an alternating cascade. When a predator is removed or added, the immediately-below trophic level's biomass changes in the opposite direction, the level below that changes in the same direction as the top predator, and so on in alternating sign down the chain — this is the trophic cascade mechanism (detailed under Trophic Cascades). The cascade's strength depends on how tightly coupled the levels are: a generalist predator with many prey items or a prey community with many competing consumers dampens the cascade, because compensatory species can substitute for the suppressed one.

Step 6 — Bottom-up control sets the ceiling the cascade operates within. Even a maximally efficient predator cannot sustain more biomass than primary production allows. Oksanen, Fretwell, and colleagues formalized this with the Exploitation Ecosystems Hypothesis (EEH): as primary productivity increases along a gradient, ecosystems can support successively more trophic levels, and which level is top-down-limited versus resource-limited flips in a predictable stair-step pattern — very low productivity supports plants only (self-limited), moderate productivity adds herbivores that suppress plants, higher productivity adds carnivores that suppress herbivores and release plants, and so on.

Step 7 — Interaction strength, not link count, drives stability. Empirical webs (and the theoretical work of Robert May and later Kevin McCann) show that most feeding links are weak (small effect on prey population dynamics) with only a few strong links; this skewed distribution, rather than uniform link strength, is what allows large complex webs to persist without runaway oscillations, resolving a long paradox (May's 1972 result that random networks with many strong random interactions should be mathematically unstable, contradicting the observed complexity of nature).

Hairston, Smith & Slobodkin (1960) — the HSS hypothesis. In a short, highly influential theoretical paper, Nelson Hairston Sr., Frederick Smith, and Lawrence Slobodkin argued that terrestrial food chains have exactly three broad, alternating regimes of control: decomposers are food-limited (bottom-up), herbivores are rarely food-limited because they are kept below the carrying capacity of vegetation by predators (top-down), and predators are ultimately limited by their own food supply (again bottom-up, but of herbivores). Their key evidence was indirect but striking: the observation that terrestrial plant biomass is not, in most ecosystems, consumed to the point of exhaustion — implying herbivores are held below the level plants could otherwise support, which is only possible if predators (or disease/parasitism) are suppressing herbivore numbers. The paper triggered decades of debate and direct experimental tests, because HSS itself offered no manipulative proof — it was the first article to state top-down control as a testable, general community-level hypothesis rather than a single-system anecdote.

Robert Paine's sea star removal experiments (1966, 1969, Washington coast). Paine physically removed the predatory sea star Pisaster ochraceus from sections of rocky intertidal shoreline near Mukkaw Bay and monitored the resulting community for years while leaving adjacent control plots untouched. Within about a year, mussel (Mytilus californianus) populations — normally kept in check by Pisaster predation — expanded explosively and outcompeted other sessile species for space, and the number of species present in cleared plots dropped from 15 to roughly 8. This was the founding experiment for the very concept of a Keystone Species (Paine coined the term explicitly from this system) and remains one of the cleanest field demonstrations that a single strongly-linked predator can determine the diversity and composition of an entire local web through top-down effects that outweigh its modest numerical abundance.

Carpenter and Kitchell's whole-lake manipulations (1980s, Wisconsin). Stephen Carpenter and James Kitchell manipulated entire small lakes (not just cages or plots) by adding or removing piscivorous fish, testing whether cascading effects would propagate down through zooplankton to phytoplankton and measurable water chemistry. When piscivores (largemouth bass) were added or increased, planktivorous fish declined, zooplankton grazers increased, and phytoplankton biomass and primary productivity fell — a three-level cascade visible as a genuine change in lake water clarity and chlorophyll concentration, not just a laboratory artifact. This work was crucial because it extended Paine-style manipulation from a small, structurally simple intertidal community to a large, open, species-rich lake ecosystem, showing top-down cascades were not a peculiarity of simple systems.

Oksanen's productivity-gradient studies (1981, Fennoscandian tundra). Lauri Oksanen and colleagues tested the Exploitation Ecosystems Hypothesis by comparing plant and herbivore (microtine rodent, reindeer) biomass across a natural gradient of primary productivity in arctic and subarctic habitats. They found that at the lowest productivity sites, plant biomass tracked productivity directly (bottom-up, no effective herbivore suppression), while at higher productivity sites herbivore biomass increased sharply and plant biomass leveled off or dropped — consistent with herbivores becoming abundant enough for predators/food limitation to shift the controlling level, exactly as EEH predicted. This provided some of the first large-scale, non-manipulative field support that the position of top-down versus bottom-up control is not fixed but shifts predictably with ecosystem productivity.

Post et al. and stable isotope chain-length surveys (2000s). Using nitrogen-15 enrichment to assign quantitative trophic positions across dozens of lake food webs, David Post and colleagues showed that food chain length correlates most strongly not with productivity (as many had assumed) but with ecosystem size and, in aquatic systems, basin/lake area — larger systems support longer chains, likely because they support more species and more resource heterogeneity, giving room for additional specialist trophic levels to evolve or persist. This overturned a simple "more energy, longer chains" intuition and remains a benchmark large-scale correlational dataset in food web ecology.

Example 1 — Calculating connectance and comparing two webs. Suppose a small alpine meadow web has been resolved into 12 trophic species with 28 observed directed feeding links (including two cases of omnivory and one of cannibalism among carrion-feeding beetles). A nearby, more disturbed roadside web has only 7 trophic species but 24 links, because the surviving species are mostly generalists. Connectance is C = L / S². Meadow: C = 28 / 12² = 28 / 144 = 0.194 Roadside: C = 24 / 7² = 24 / 49 = 0.490 Although the roadside web has fewer species, it is far more densely connected (C ≈ 0.49 vs 0.19). This matches the general empirical pattern that species-poor, disturbed webs are often proportionally denser because the surviving species are generalists forced into broader diets by the scarcity of specialist prey/competitors — while the species-rich meadow web, despite having more total links, spreads them across many more possible pairings, giving lower connectance. This illustrates why raw link count (L) and connectance (C) tell different stories, and why comparing ecosystems by species richness alone can be misleading about network robustness — a sparse, species-rich web can actually be more vulnerable to losing specific links (specialist extinction) even while looking "more complex" by node count.

Example 2 — Applying the 10% rule and testing feasibility of a fifth trophic level. A wetland's primary producers fix and make available 8,000 kcal m⁻² yr⁻¹ of net primary production. Using an average ecological efficiency of 10% per transfer:

A large predatory bird at trophic level 5 with an estimated individual energy requirement of roughly 400 kcal m⁻² yr⁻¹ equivalent territory support (a simplified figure for illustration) would require far more than the 0.8 kcal m⁻² yr⁻�1 available at level 5 — meaning the wetland, at this productivity, cannot support even a single such specialist purely on a five-step chain; the real top predator in such systems is almost always an omnivore/generalist feeding across levels 2–4 simultaneously (shortening its effective trophic position and its energetic demand), which is precisely why very long strict food chains are rare in nature and why apex consumers in most real ecosystems have measured mean trophic levels (via isotope data) closer to 3.5–4.5 rather than a clean integer 5 or 6. This calculation demonstrates concretely why FCL saturates at low integers: it is an energetic ceiling, not an arbitrary convention.

Robert T. Paine (1933–2016) transformed food web ecology from a largely descriptive, correlational discipline into an experimental science built on deliberate, replicated manipulation of whole natural communities. Trained as an invertebrate zoologist, Paine worked for most of his career at the University of Washington, spending decades doing fieldwork on the rocky intertidal shorelines of the Pacific Northwest — a system he chose partly because its low diversity and clear zonation made cause and effect experimentally tractable in a way rainforest or coral-reef webs were not. His removal experiments were deceptively simple (pry the sea stars off the rocks, keep them off, watch what happens) but the interpretive leap — that a single predator species, not resource competition or physical stress, could determine the entire diversity of a shoreline community — was genuinely new, and it directly produced the vocabulary ("keystone species," used first in his 1969 paper) that ecologists still use to describe disproportionately important species in food webs today.

Fisheries management relies directly on food web structure: overfishing a mid-level predator can trigger a cascade that collapses commercially valuable species several links away, which is why modern stock assessments increasingly use whole-ecosystem models (e.g., Ecopath with Ecosim) rather than single-species quotas. Biological control in agriculture exploits top-down suppression deliberately — introducing or conserving a specialist predator or parasitoid to control a crop pest depends on understanding connectance and alternative prey availability, since a generalist predator with many other food options may fail to suppress the target pest strongly enough. Invasive species risk assessment uses food web position to predict impact: an invader that inserts itself as a new, strongly-linked node (a novel apex predator, or a novel highly connected omnivore) is far more likely to trigger cascading extinctions than one that slots into an already crowded, weakly-linked niche, directly applying connectance and interaction-strength theory to conservation triage. Rewilding and keystone reintroduction programs, most famously the reintroduction of gray wolves to Yellowstone in 1995, are explicitly designed around restoring a lost top-down link and observing (with mixed but real evidence) whether browsing pressure on willow and aspen relaxes and downstream effects on beaver, songbirds, and stream morphology follow — a large-scale, ongoing test of trophic cascade theory in a terrestrial system. Climate change impact modeling increasingly uses food web and trophic-level frameworks to predict which species are most vulnerable to warming-driven mismatches, since a species' vulnerability depends heavily on how many links connect it to the rest of the web and whether its trophic partners can shift phenology in step. Finally, food web architecture is foundational to how ecologists now measure and interpret Species Diversity and Its Measurement and predict community trajectories during Ecological Succession, since both diversity metrics and successional turnover are shaped by the same underlying connectance and trophic-length constraints described here.

Ask the tutor: How would you redesign Paine's removal experiment today using isotopes or DNA metabarcoding to quantify connectance changes, not just species counts?