Why Individual-Based Predator-Prey Models Behave Differently from the Lotka-Volterra Equations

Two ways to write down a predator-prey system

For most of the twentieth century, "modelling a food web" meant writing down a pair of differential equations. The Lotka-Volterra model, proposed independently by Alfred Lotka and Vito Volterra in the 1920s, describes a prey population and a predator population as two continuous numbers that change smoothly over time according to fixed rules: prey grow exponentially in the absence of predators, predators decline exponentially in the absence of prey, and the encounter rate between the two populations sets how fast one is converted into the other. Solve the equations and you get a closed loop in phase space — predator and prey numbers chasing each other in a perpetual, perfectly repeating cycle.

The simulation on this page takes a different approach, one that has become common in ecology, epidemiology, and conservation planning since computers made it cheap: instead of two abstract population numbers, it tracks dozens or hundreds of individual organisms, each with its own position, its own energy reserve, and its own private die-roll for reproduction and death. This is called an individual-based model or agent-based model (IBM/ABM). Nobody writes a differential equation for the whole population; instead, population-level behaviour — including the same boom-and-bust cycles the Lotka-Volterra equations predict — emerges from thousands of local, individual decisions. The interesting scientific question is not just "do both approaches show cycles?" but "in what ways do the two approaches disagree, and why does that matter for real conservation decisions?"

What an energy budget actually buys you

In the individual-based version, every organism carries a scalar quantity — call it energy — that behaves like a bank balance. It costs a small amount just to stay alive each tick (basal metabolism), it goes up when the organism successfully finds and consumes food within some short encounter radius, and it must cross a threshold before the organism is allowed to reproduce, at which point it pays a lump-sum energetic cost to produce an offspring. If the balance ever drops to zero, the organism dies of starvation, full stop — no differential equation "reduces" the population smoothly; an actual individual is removed from the list.

This bookkeeping does more than add cosmetic realism. It makes reproduction and death depend on an organism's actual, individual foraging history rather than on the instantaneous population-wide average. Two prey animals with identical starting conditions can end up with wildly different fates depending on whether they happened to wander near a cluster of plants early on. The Lotka-Volterra equations have no way to represent that — they assume every prey individual is, in effect, interchangeable and instantaneously well-mixed with every predator. An energy-budget model reintroduces exactly the kind of individual variation that field ecologists actually observe: some animals are well-fed and fecund, others are on the edge of starvation, and which is which depends on their concrete history of encounters, not a population average.

Discreteness, demographic noise, and the risk of extinction

A second, subtler difference concerns what happens near small population sizes. In the Lotka-Volterra equations, population size is a continuous real number, so a predator population can shrink to 0.0001 individuals and then recover — mathematically that's a perfectly valid trajectory. Real populations, and individual-based simulations, don't have that luxury: population size is a non-negative integer, and once it hits exactly zero, there is nothing left to recover. This is called demographic stochasticity, and it is one of the main reasons wildlife managers care about individual-based models specifically: they capture extinction risk in a way that smooth equations structurally cannot.

In the simulation on this page, this shows up directly in the status readout. A run where the prey population dips low enough by chance can crash the predator population before either has a chance to rebound, an outcome the mean-field Lotka-Volterra curves would simply never produce for the same starting parameters, because the ODEs describe only the average behaviour of infinitely many replicate ecosystems, not the particular noisy trajectory of any single one. Run the same individual-based model twice with the same starting population sizes and you can get a stable oscillation in one run and a predator die-off in the other, purely from the randomness in who finds food, who reproduces, and who gets found.

Space matters: local encounters instead of a well-mixed soup

The Lotka-Volterra equations assume "mass-action" mixing: every predator is equally likely to encounter every prey individual, as if the whole ecosystem were a single well-stirred pot. An agent-based simulation instead gives every organism an explicit x-y position and a finite interaction radius — a predator only has a chance to catch a prey animal if the straight-line distance between them falls under some threshold, typically a few percent of the world's width.

That single change has large consequences. Predators can locally deplete prey in one region of the map while prey elsewhere are untouched and free to reproduce, producing spatial refuges that a well-mixed model cannot represent at all. It also means the encounter rate is not simply proportional to the product of the two population sizes (as Lotka-Volterra assumes) but depends on how clumped or dispersed each population happens to be at that moment — which is itself an emergent, constantly shifting property of the simulation, not a fixed parameter. Ecologists studying real trophic cascades, from wolves and elk in Yellowstone to kelp forests and sea otters, know that spatial structure is often the single biggest factor separating a textbook cycle from what actually happens in the field, which is exactly why individual-based, spatially explicit models have become the standard tool for testing management interventions before they're tried in the real world.

Trophic cascades: the third rung matters too

The simulation includes a third trophic level below the classic predator-prey pair: plants, which regrow at a fixed low probability each tick up to some carrying capacity, and which prey animals must find and consume to keep their own energy balance positive. This turns the system into a three-level food chain, and three-level chains behave in ways two-level ones cannot, a phenomenon ecologists call a trophic cascade. When predators are abundant, they suppress prey numbers, which in turn releases plants from grazing pressure and lets plant biomass recover — the classic "landscape of fear" effect documented in real systems such as the reintroduction of wolves to Yellowstone National Park, where elk browsing pressure on willow and aspen dropped measurably once wolves returned. Remove the predator level entirely and prey numbers can rise until they overgraze the plant base faster than it can regrow, at which point the prey population itself collapses from resource scarcity rather than from predation — a different failure mode with a different signature in the population data.

Because energy budgets are explicit numbers in this kind of model, you can actually quantify that cascade: how much plant biomass is present, on average, when predators are near their carrying capacity versus when they've been wiped out, is a straightforward measurement in the simulation and a very expensive, multi-year field study in reality.

Why ecologists use both approaches, not just one

None of this means the Lotka-Volterra equations are obsolete — they remain the fastest way to build intuition about why predator-prey systems oscillate at all, and closed-form results about their cycle period and stability are still taught in every introductory ecology course for good reason. Individual-based, energy-budget models are computationally heavier and harder to analyse mathematically; you generally can't write down a clean formula for their long-run behaviour, you have to run them and see. What they buy in exchange is the ability to represent exactly the features real conservation biologists most need: discrete extinction events, spatial refuges, individual variation in condition, and multi-level cascades through a food web with more than two species. Modern wildlife management software — used for setting fishing quotas, planning predator reintroductions, and modelling disease spread through wild populations — is built almost entirely on the individual-based approach for precisely these reasons, with the Lotka-Volterra equations kept around mainly as a sanity check and teaching tool.