Metapopulation

A metapopulation—coined by the mathematical ecologist Richard Levins in 1969 as a "population of populations"—is a spatially structured assembly of local breeding subpopulations occupying discrete, suitable habitat patches embedded within an inhospitable or semi-permeable landscape matrix and connected by limited inter-patch dispersal. Unlike a single panmictic population where all individuals mix freely, or a set of completely isolated island populations with zero gene flow, each local subpopulation in a metapopulation is small enough to face a substantial risk of stochastic local extinction due to demographic fluctuations, environmental disturbances, inbreeding depression, or localized predation, while empty habitat patches are recurrently recolonized by migrants dispersing from extant occupied patches.

Whereas MacArthur and Wilson's 1967 theory of island biogeography modeled species richness across oceanic islands colonized from a permanent, inexhaustible mainland source pool, Levins's metapopulation framework demonstrated that a single species can persist indefinitely at the regional landscape scale even when every individual local population is transient and unstable. Together with Ilkka Hanski's spatially realistic Incidence Function Model (IFM) and Ronald Pulliam's source–sink demographic theory, metapopulation dynamics provide the quantitative foundation for reserve network design, habitat fragmentation assessment, and wildlife corridor engineering.

Derivation of the Levins Ordinary Differential Equation

To analyze regional persistence analytically, Levins (1969, 1970) formulated a spatially implicit patch-occupancy model assuming an infinite network of identical, equally connected habitat patches, each existing in one of two binary states: occupied (ignoring internal local population size N_i against local carrying capacity K_i) or empty. Let p(t) ∈ [0, 1] denote the fraction of patches occupied at time t, so that 1 − p(t) is the fraction of empty suitable patches. Assuming that occupied patches go extinct independently at a constant per-patch rate e (time⁻¹), and that each occupied patch sends out propagules at rate c (time⁻¹) distributed uniformly across all patches (mass-action "propagule rain"), the rate of new colonizations is proportional to the product of occupied patches p and empty patches (1 − p). The continuous-time rate of change in patch occupancy is governed by the Levins differential equation:

dp / dt = c · p(1 − p) − e · p

Factoring the right-hand side as dp/dt = c · *p*[(1 − e/c) − p] reveals a mathematical isomorphism with the Verhulst logistic growth equation (dN/dt = *rN*[1 − N/K]), where the intrinsic metapopulation growth rate at low occupancy is r_m = c − e and the effective regional carrying capacity is the non-trivial equilibrium occupancy p\*:

*p*\* = 1 − (e / c)

Two fundamental ecological insights follow from this equilibrium:

1. Persistence Criterion (c > e): A metapopulation persists regionally (*p*\* > 0) if and only if the colonization rate parameter c exceeds the local extinction rate e. Even when c > e, a fraction e/c of intact, high-quality habitat patches remains unoccupied at any given survey snapshot simply because recolonization lags behind local turnover.
2. Habitat Destruction and the Extinction Threshold: Suppose human land-use conversion permanently destroys a fraction D of the original habitat patches, leaving a fraction h = 1 − D of patches habitable. Because dispersing propagules landing in destroyed patches perish, the colonization term becomes c · p(1 − D − p), modifying the differential equation to:

dp / dt = c · p(1 − D − p) − e · p

Setting dp/dt = 0 yields the habitat-depleted equilibrium occupancy:

*p*\*(D) = 1 − D − (e / c) = h − (e / c)

Consequently, the metapopulation collapses to regional extinction (*p*\* = 0) when the fraction of remaining suitable habitat h falls to the critical extinction threshold h_c = e/c (or equivalently when destroyed habitat reaches D_c = 1 − e/c), even though a fraction e/c of pristine, unoccupied habitat patches still remains on the landscape. Moreover, when habitat loss abruptly drops h below e/c, regional decline follows an exponential decay trajectory toward zero over dozens of generations, creating a protracted extinction debt (Tilman et al., 1994).

The Rescue Effect and Hanski's Incidence Function Model

Levins's classical formulation assumes that inter-patch migration influences only the colonization of empty patches. In 1977, James H. Brown and Astrid Kodric-Brown introduced the rescue effect, recognizing that immigrants arriving at an already-occupied patch bolster local census size N_i and restore allelic diversity via genetic rescue, thereby reducing the local extinction rate E_i as regional occupancy p or connectivity S_i increases:

E_i = e · (1 − p) or E_i = e / (1 + C_i)

To incorporate real-world patch geometry—where patch areas A_i and pairwise inter-patch distances d_ij vary across several orders of magnitude—Ilkka Hanski (1994, 1998) developed the spatially realistic Incidence Function Model (IFM). In a discrete-time Markov chain, the stationary probability J_i (incidence) that patch i is occupied balances patch-specific colonization probability C_i and extinction probability E_i (including the rescue effect):

J_i = C_i / (C_i + E_i − C_iE_i)

Here, local extinction probability scales inversely with patch area as E_i = min(1, e / A_i^x)—because larger patches support larger populations with lower demographic stochasticity—while colonization probability C_i = S_i² / (S_i² + y²) depends on the patch connectivity index S_i, where S_i is summed over all other occupied patches j ≠ i weighted by a negative exponential decay dispersal kernel exp(−α·*d_ij*):

S_i = ∑_*j*≠*i* p_j · e^{−α · d_ij} · A_j^b

where 1/α is the mean species dispersal distance. Hanski and colleagues validated the IFM across a network of roughly 4,000 dry-meadow patches inhabited by the Glanville fritillary butterfly (Melitaea cinxia) across the Åland Islands of southwestern Finland. Annual censuses showed that although ~200 local larval webs go extinct and ~150 empty meadows are colonized each summer (with average dispersal distances of 1/α ≈ 1.0 km), overall regional occupancy remains stable across the archipelago and matches IFM predictions.

The Four Metapopulation Paradigms

Susan Harrison (1991) and H. Ronald Pulliam (1988) classified spatially structured populations along two orthogonal axes—the rate of inter-patch dispersal and the heterogeneity of patch size or quality—into four distinct spatial paradigms:

In Pulliam's (1988) source–sink dynamics, habitat patches differ intrinsically in birth rates (b) and mortality rates (d). Source habitats are high-quality patches where local reproduction exceeds mortality (b > d, finite rate of increase λ > 1), generating a net demographic surplus of emigrants. Sink habitats are marginal or degraded patches where local mortality exceeds reproduction (b < d, λ < 1); without continuous immigration from adjacent sources, sink subpopulations inevitably decay to local extinction. When anthropogenic disturbance converts an attractive habitat into a high-mortality sink that organisms preferentially select using misleading environmental cues, the sink becomes an ecological trap.

At the opposite end of the dispersal spectrum, when inter-patch dispersal is negligible while local extinction persists, a regional network degenerates into a non-equilibrium (relict) metapopulation—such as montane mammal populations isolated on sky islands since the Pleistocene—where occupied patches blink out one by one without recolonization until regional extinction occurs, unless conservation managers re-establish habitat corridors or conduct assisted translocations.

Why are some pristine habitat patches always empty in a healthy metapopulation?

In a Levins metapopulation at equilibrium (*p*\* = 1 − e/c), local subpopulations experience stochastic extinctions at rate e while empty patches require finite time to be discovered and recolonized by dispersers at rate c. Because local extinctions continuously create newly vacant patches faster than dispersers can instantaneously fill every site, an equilibrium fraction e/c of completely suitable habitat remains unoccupied at any point in time.

What is an "extinction debt" in fragmented landscapes?

When habitat destruction reduces the fraction of remaining patches h near or below the extinction threshold (e/c), the metapopulation does not vanish overnight. Instead, regional occupancy p(t) relaxes gradually toward its new lower or zero equilibrium over many decades—especially in long-lived perennial plants, woodland insects, and forest vertebrates. This transient surplus of extant species or occupied patches awaiting inevitable stochastic loss is termed an extinction debt.

How can conservation planners distinguish a demographic source from a sink?

Measuring local population density alone is insufficient because territorial exclusion by dominant individuals in a source patch can force huge numbers of subordinate juveniles to crowd into an adjacent low-quality sink, making the sink appear denser than the source. Distinguishing sources from sinks requires measuring local per-capita fecundity (b) and survival (s) independent of immigration to verify whether the intrinsic finite growth rate λ = s + b exceeds 1.0 (source) or falls below 1.0 (sink).

1. Levins, R. (1969). "Some demographic and genetic consequences of environmental heterogeneity for biological control." Bulletin of the Entomological Society of America, 15(3), 237–240. https://doi.org/10.1093/besa/15.3.237
2. Pulliam, H. R. (1988). "Sources, Sinks, and Population Regulation." The American Naturalist, 132(5), 652–661. https://doi.org/10.1086/284880
3. Hanski, I. (1998). "Metapopulation dynamics." Nature, 396(6706), 41–49. https://doi.org/10.1038/23876
4. Tilman, D., May, R. M., Lehman, C. L., & Nowak, M. A. (1994). "Habitat destruction and the extinction debt." Nature, 371(6492), 65–66. https://doi.org/10.1038/371065a0