Metapopulations

A metapopulation is a regional collection of local populations ("subpopulations" or "demes") occupying discrete habitat patches, connected to one another by the migration of individuals between patches. The defining feature is not just spatial separation — many species are patchily distributed without being metapopulations in the technical sense — but that local populations are subject to appreciable local extinction (a patch's population goes to zero) and that empty patches can be refounded through colonization by dispersers from occupied patches. The regional population is thus a dynamic mosaic: patches blink on and off over time, occupied → extinct → recolonized → extinct again, while the species as a whole can remain present in the landscape indefinitely. This is a genuine conceptual departure from single-population thinking. In classical Population Regulation theory, the question is: what density-dependent processes (resource competition, predation, disease) hold a population near carrying capacity within one patch? In metapopulation theory the local population is often treated as doomed — subject to demographic stochasticity, environmental catastrophe, or simply resource collapse, such that extinction within any given patch is a matter of when, not if. The interesting dynamics move up a level, to the turnover of occupied and empty patches across the whole network. The currency of the model is no longer "number of individuals" but the fraction of patches currently occupied, conventionally written p. Levins's foundational insight, formalized in 1969, was that this patch-occupancy process can be described with an equation structurally identical to a single-species logistic-like model, but where "births" are colonizations of empty patches and "deaths" are extinctions of occupied ones. This produces a testable equilibrium and — crucially — a threshold: below a critical rate of colonization relative to extinction, the metapopulation cannot persist regionally at all, collapsing to permanent absence even though nothing is wrong with the patches themselves. That threshold, and the trade-offs organisms make between dispersal ability and local competitive success (tying back to Reproductive Trade-offs), are the two ideas that make metapopulation biology more than bookkeeping.

Step 1 — Define the patch network. Consider a landscape of many discrete, suitable habitat patches — ponds for a frog, host trees for an insect, forest fragments for a small mammal. Patches are assumed similar enough in quality that we don't need to track their individual identities, only whether each is occupied (population present) or empty. Patches may differ in area and in distance from other patches, which will matter for the probability of colonization, but the core Levins model simplifies this away for a first pass.

Step 2 — Local populations go extinct at some rate. Within any occupied patch, the local population is vulnerable to demographic stochasticity (small populations can go extinct from bad luck in birth/death sequences alone), local catastrophes (drought, fire, disease outbreak, a pond drying up), and boom-bust dynamics driven by resource depletion or predator pressure. Let e be the extinction rate — the fraction of occupied patches that go extinct per unit time. Smaller and more isolated patches typically have higher effective e.

Step 3 — Empty patches are recolonized by dispersers. Occupied patches produce emigrants; a fraction of these successfully reach and establish in empty patches. The colonization rate depends on the fraction of patches already occupied (more occupied patches → more sources of dispersers) and on a colonization parameter c describing the per-patch, per-occupied-patch rate of successful colonization. Successful colonization requires that dispersers survive transit and that a colonizing propagule be large enough (or lucky enough) to avoid immediate extinction — this is why colonization is a rate, not a certainty, and why isolated or distant empty patches are colonized more slowly.

Step 4 — Patch occupancy dynamics: the Levins equation. Let p be the proportion of patches occupied at time t (0 ≤ p ≤ 1). The rate of change is: dp/dt = c·p·(1 − p) − e·p

• The c·p·(1−p) term is colonization: occupied patches (p) supply colonists, and only empty patches (1−p) are available to receive them — structurally the same "encounter" logic as the logistic equation's density-dependent term.
• The e·p term is extinction: only occupied patches (p) can go extinct, at per-patch rate e.

Step 5 — Solve for equilibrium. Setting dp/dt = 0 and excluding the trivial p = 0 solution: c·p(1 − p) = e·p c(1 − p) = e p\ = 1 − e/c</strong> This is the long-run, dynamic equilibrium fraction of occupied patches. It is a statistical steady state — the identity of which patches are occupied keeps changing (turnover continues indefinitely) even while the overall fraction occupied stabilizes at p.

Step 7 — Refinements the basic model ignores (and why they matter). The classic Levins model treats all patches as identical in size and equally accessible from all others — biologically unrealistic. Real metapopulations show a rescue effect (frequent immigration into small, at-risk populations lowers their effective extinction rate by topping up numbers before a stochastic crash) and effects of patch geometry (larger patches support bigger, longer-persisting populations; more isolated patches receive fewer colonists). Incidence-function models (developed later, notably by Hanski) explicitly weight colonization probability by patch area, distance to neighboring occupied patches, and local population size, producing far better fits to real data — but the qualitative logic of the extinction–colonization balance, and the existence of a persistence threshold, survives intact.

Levins's original motivation (1969). Richard Levins developed the model while studying insect population genetics on agricultural land, where crop management created a shifting mosaic of habitable and inhabitable fields. His mathematical formalization, "Some demographic and genetic consequences of environmental heterogeneity for biological control," established the framework but was not itself an empirical field test — it was theory awaiting data. For over two decades the model remained mostly a conceptual tool; the decisive empirical test came from long-term fieldwork on a single butterfly.

Ilkka Hanski and the Glanville fritillary (Melitaea cinxia) on the Åland Islands, Finland — the landmark long-term study. Beginning in 1991, Hanski's research group mapped essentially the entire network of suitable habitat patches (dry meadows containing the butterfly's host plants) across roughly 50 km² of the Åland archipelago — about 4,000 discrete patches — and censused every patch for larval nests annually for decades. Key findings that validated the metapopulation framework:

• Extensive turnover. Individual local populations went extinct frequently — annual local extinction rates on the order of 10–20% of occupied patches in a typical year — yet the regional metapopulation persisted continuously across the whole study period. This directly confirmed the core prediction: local instability with regional stability.
• Patch-area and isolation effects on occupancy. Occupancy probability rose with patch area (bigger patches held bigger, more resilient populations) and fell with distance to the nearest occupied patch (isolated empty patches were less likely to be recolonized) — exactly the modifications predicted by spatially explicit extensions of the Levins model (the incidence function model Hanski developed from these data).
• Rescue effect confirmed. Small populations that received more immigrants had measurably lower extinction probabilities than equally small, more isolated populations, demonstrating that dispersal doesn't just recolonize empty patches — it stabilizes small occupied ones too.
• Extinction–colonization balance measured directly. By tracking which patches were occupied, which went extinct, and which were newly colonized each year, Hanski's group empirically estimated c and e for a real system and showed that the regional metapopulation sat close to the dynamic equilibrium the model predicted, with observed occupancy fluctuating around 40–70% of suitable patches depending on the year and local weather.
• Evidence for extinction debt. Comparative and modeling work using this dataset (and parallel studies on other fragmented systems) showed that after habitat patches are lost or fragmented further, the surviving metapopulation does not collapse instantly — it can take many generations of accumulating local extinctions exceeding colonizations before regional occupancy fully re-equilibrates downward, a lag now called extinction debt.

A second classic system: the American pika and mainland "island" extinctions. Long-term surveys of pika (Ochotona princeps) populations across the Great Basin's mountain-top talus patches — isolated from each other by intervening lowland desert too hot for the species — have documented repeated local extinctions from formerly occupied sites with no recolonization at some sites over multi-decade timescales, consistent with a metapopulation in which warming climate has pushed local e above achievable c across parts of the range, illustrating the persistence threshold operating in real time under environmental change.

Laboratory microcosm confirmation. Because field metapopulation dynamics unfold over years to decades, researchers have also tested Levins-type predictions in controlled laboratory microcosms — protozoan or mite populations distributed across small connected patches (e.g., wells or vials linked by tubing that permits controlled dispersal). These experiments manipulate patch connectivity directly and show, as predicted, that reducing dispersal rate below a threshold causes metapopulation-wide collapse even though each patch, treated alone, could sustain a population indefinitely — a controlled demonstration of the c > e threshold that field data alone (subject to countless confounds) cannot cleanly isolate. Together, the Åland fritillary work stands as the discipline's flagship long-term empirical validation — comparable in stature, within population ecology, to how Kettlewell's peppered moth work anchors natural selection or the Grants' finch work anchors quantitative evolutionary ecology — while the microcosm experiments supply the controlled, manipulative proof that correlational field data alone cannot.

Example 1 — Solving the Levins equilibrium and diagnosing collapse risk. Suppose ecologists monitoring a metapopulation of a wetland amphibian across 60 ponds estimate, from several years of occupancy surveys:

• colonization parameter c = 0.35 per year
• extinction rate e = 0.22 per year

Step 1: Check persistence condition: c > e → 0.35 > 0.22 ✓. The metapopulation is predicted to persist. Step 2: Compute equilibrium occupancy: p = 1 − e/c = 1 − (0.22/0.35) = 1 − 0.629 = <strong>0.371</strong> So at equilibrium, roughly 37% of the 60 ponds (≈ 22 ponds) are occupied at any given time — even though which 22 ponds are occupied keeps changing year to year. Step 3: A wetland-drainage project removes 25 of the 60 ponds, and — because remaining ponds are now farther apart on average — effective c drops to 0.18/year while e (driven mainly by local pond drying, unaffected by neighboring ponds) stays at 0.22/year. New check: c > e? 0.18 > 0.22 is <strong>false</strong>. The persistence threshold has been crossed. Step 4: Predicted outcome: p would be negative under the formula, which is biologically read as p\* = 0 — regional extinction is now the only stable state. Because turnover is stochastic and generational, this collapse will not be instantaneous; the amphibian may linger in a shrinking number of ponds for several more years (extinction debt) even though its regional fate is already sealed by the drainage project. This is exactly the "sleeper effect" documented in the Åland fritillary system and is the central argument for assessing connectivity, not just current patch count, before permitting habitat removal.

Example 2 — Reserve design: patch number vs. patch connectivity trade-off. A conservation agency has a fixed budget sufficient to protect either: (A) 10 large, well-separated reserves (average distance between reserves: 40 km), or (B) 25 smaller reserves at higher density (average distance between neighboring reserves: 8 km) for an old-growth-dependent salamander with limited dispersal capacity (typical successful dispersal distance ≈ 5 km). Reasoning through the Levins logic:

• Colonization parameter c scales with the probability a disperser reaches another patch, which falls off steeply with distance once distance exceeds typical dispersal range. At 40 km spacing (option A), successful inter-patch colonization events are essentially zero — c ≈ 0. Each of the 10 reserves functions as an isolated population with no rescue effect and no recolonization after local extinction: effectively 10 independent single populations, each individually doomed by demographic and environmental stochasticity over the long run (extinction is a matter of time, and there is no c term to counteract it).
• At 8 km spacing (option B), distances are close to but somewhat above the ~5 km typical dispersal range, so c is small but nonzero — occasional successful dispersal is possible, especially between near-neighbor patches, giving some realistic chance that c > e.
• Even though option A protects more total hectares per reserve, option B is predicted to have a much higher regional persistence probability, because it is patch density and connectivity, not patch size alone, that determines whether the c > e threshold is met.

This is precisely the reasoning behind modern corridor and "stepping-stone" reserve design: many conservation programs deliberately choose configurations of moderate-sized, closely spaced reserves (or reserves linked by corridors) over a few large, isolated ones, when the target species has limited dispersal — a design principle derived directly from the Levins threshold condition.

Richard Levins (1930–2016) was an American population biologist and mathematical ecologist whose career spanned tropical agriculture, evolutionary genetics, and — later — social epidemiology and Marxist philosophy of science. Before formalizing the metapopulation model, he worked on pest population genetics in Puerto Rico and the Caribbean, where he confronted the practical reality that agricultural insect pests occupied fields as a shifting patchwork of colonized and abandoned sites rather than one uniform population — this concrete agronomic problem, not abstract theorizing, is what generated the 1969 patch-occupancy model. Levins was known for insisting that models should be judged by the robustness of their qualitative conclusions across a family of related assumptions rather than by precise quantitative fit to any one dataset — a philosophy visible in how the Levins model's core prediction (a persistence threshold) has survived even as its details (identical patches, no spatial structure) have been extensively revised by later spatially explicit models such as those developed by Hanski.

Conservation reserve design. Agencies designing protected-area networks now explicitly model colonization and extinction rates across candidate reserve configurations rather than maximizing total protected area alone, favoring connected networks (via corridors, stepping-stone patches, or proximity) for dispersal-limited species — a direct application of the c > e threshold.

Predicting extinction debt from historical fragmentation. Landscape ecologists use metapopulation models to identify species that are likely to decline in the coming decades purely as a delayed consequence of past habitat fragmentation, even without further habitat loss — informing which currently "stable-looking" populations most urgently need connectivity restoration.

Invasive species and biological control. The same colonization-extinction logic that Levins originally used for agricultural pests is applied in reverse for invasive species management: reducing effective connectivity between patches of suitable habitat (e.g., isolated water bodies for an invasive fish, or host plants for an invasive insect) can push an unwanted species below its own persistence threshold.

Disease ecology and epidemiology of fragmented host populations. Wildlife disease dynamics (e.g., chytrid fungus in amphibian metapopulations, or canine distemper in fragmented carnivore populations) are increasingly modeled with metapopulation-style patch occupancy, treating "infected patch" versus "disease-free patch" analogously to occupied versus empty habitat patches.

Climate-driven range fragmentation. As warming shifts suitable climate envelopes upslope or poleward (as documented for pika populations in mountain-top habitat islands), metapopulation models are used to forecast whether remaining patches retain enough connectivity for the species to track its shifting habitat, or whether isolation will outpace the species' colonization ability.

Setting up source-sink dynamics. The Levins model's simplifying assumption — that all patches are equivalent — is the natural next thing to relax: real patches differ systematically in quality, some producing a surplus of dispersers (sources) and others sustained only by immigration (sinks), which is the subject taken up directly in Source-Sink Dynamics.

Ask the tutor: How does relaxing the Levins model's "all patches are identical" assumption lead directly into source-sink dynamics?