Concept

Metapopulation dynamics

T-087Home BU-305Threads systems · evolution

Statement

Colonisation and extinction across habitat patches.

Why it matters

island-biogeography already showed that a single isolated habitat's species richness is set by a balance of colonisation and extinction; metapopulation-dynamics generalises that same balance to a single species distributed across many patches, treating patch occupancy itself, not species number, as the quantity that rises and falls. Where island-biogeography asks "how many species live here," this result asks "is this species present here," repeated across a whole network of habitat patches connected by dispersal.

It matters directly for conservation: ecological-succession and keystone-species describe what happens within a community once established, but for a species confined to fragmented habitat — increasingly the normal condition, not the exception, in human-altered landscapes — whether it persists at all is a metapopulation question, not a single-population question.

Hypotheses

Suitable habitat exists as discrete, spatially separated patches rather than one continuous range.Without discreteness there is no meaningful notion of "occupied" versus "empty" patches to track; the whole framework specifically addresses species whose habitat is fragmented, not species with continuous, uniform range. Local populations within a patch can go extinct, and empty patches can be recolonised by dispersal from occupied ones.If local extinction never occurred, every patch would eventually be occupied and stay that way, collapsing the model to ordinary island-biogeography-style saturation; if recolonisation never occurred, every local extinction would be permanent and the whole metapopulation would inevitably decay to zero occupancy. The classical Levins model further assumes an infinite, homogeneous array of identical patches, all equally accessible to dispersal — a strong simplification. Real landscapes have patches differing in size, quality and isolation, and more detailed spatially explicit models relax this idealisation at the cost of analytical tractability.

Proof

\[ \frac{dp}{dt} = cp(1-p) - ep \]

Let \(p\) be the fraction of patches currently occupied. Colonisation of an empty patch requires both an occupied source patch (probability \(\propto p\)) and an empty target patch (probability \(\propto 1-p\)), giving a colonisation term \(cp(1-p)\); each occupied patch independently goes locally extinct at rate \(e\), giving an extinction term \(ep\) (the Levins model). B

\[ \hat p = 1-\frac{e}{c}\quad(\text{setting } dp/dt=0) \]

At equilibrium the colonisation and extinction terms balance exactly; solving \(cp(1-p)=ep\) for the nonzero root gives the equilibrium occupied fraction \(\hat p\), which is stable whenever \(c>e\) (Step 3). B

\[ \text{Persistence requires } c>e; \text{ if } c\le e,\ \hat p\le 0 \text{ and the only stable state is total extinction.} \]

Because \(\hat p\) must lie between 0 and 1 to be biologically meaningful, the colonisation rate must strictly exceed the extinction rate for any nonzero equilibrium occupancy to exist; a metapopulation whose patches are too isolated (low \(c\)) or too small/poor-quality (high \(e\)) is predicted to go globally extinct even though no single local population is doomed on its own. A

\[ \text{The rescue effect: high immigration into small, extinction-prone patches lowers their effective } e. \]

Frequent immigration from occupied neighbouring patches can prevent a small local population from actually reaching zero even when its own birth and death rates alone would predict extinction, effectively reducing \(e\) for well-connected patches relative to isolated ones. A

\[ \text{Source-sink structure: some patches have local birth}>\text{death (sources), others birth}<\text{death (sinks), sustained only by immigration.} \]

Not every occupied patch contributes equally: source patches export more colonists than they receive and would persist alone, while sink patches would go locally extinct without continual immigration from sources — occupancy alone (Steps 1–3) does not distinguish the two, but their conservation implications differ sharply. A

Result

\[ \frac{dp}{dt}=cp(1-p)-ep,\qquad \hat p = 1-\frac{e}{c}\ \ (c>e) \]

Reading. A species can persist regionally as a shifting mosaic of occupied and empty patches, with no single patch permanently occupied, provided colonisation outpaces local extinction on average across the whole network.

Scope. The Levins model is a mean-field idealisation (Hypotheses); it predicts whether persistence is possible at all, not which specific patches will be occupied at any given moment.

Corollaries & converses

r-k-selection connects here: species with high dispersal and rapid local population growth (more r-selected) tend to have higher \(c\), favouring metapopulation persistence in fragmented landscapes even at high local \(e\).
Habitat fragmentation typically raises \(e\) (smaller patches, more edge effects) and lowers \(c\) (greater inter-patch distance, reduced connectivity) simultaneously — both changes push \(\hat p\) in Step 2 toward zero, which is why fragmentation is disproportionately damaging compared to simple habitat-area loss alone.
Converse: observing a species persisting stably across a fragmented landscape, with individual patches regularly blinking in and out of occupancy, is itself indirect evidence that \(c>e\) holds for that network, even without measuring either rate directly.

Fails without

Drop recolonisation (Hypotheses): if dispersal between patches is severed — a new road or barrier, for instance — each local population becomes an isolated population subject only to its own extinction risk, with no rescue effect (Step 4) possible; local extinctions become permanent and the metapopulation as a whole decays toward \(p=0\) regardless of how many patches remain suitable in principle.
Let colonisation rate fall below extinction rate (\(c\le e\), Step 3): \(\hat p\) becomes zero or negative, meaning no stable positive occupancy level exists at all; the metapopulation is predicted to go globally extinct even though the habitat itself has not disappeared, purely because patches are being colonised more slowly than they are being emptied.

Common errors

Treating a metapopulation as if it were one large, well-mixed population rather than a set of discrete local populations connected only by limited dispersal (Hypotheses).
Assuming every occupied patch contributes equally to persistence, ignoring the source-sink distinction (Step 5) — protecting a sink patch alone, without its source, does not guarantee persistence.
Confusing local extinction (a patch losing its population, still recolonisable) with true metapopulation extinction (\(p\to0\) globally, Step 3), which requires the entire network's colonisation-extinction balance to fail, not just one patch.
Overlooking the rescue effect (Step 4) and treating each patch's extinction risk as independent of its connectivity to occupied neighbours.

Discussion

Richard Levins introduced the occupancy model of Step 1 in 1969, originally in the context of agricultural pest management rather than conservation; its central insight — that a species can persist indefinitely at the regional scale while going locally extinct constantly at the patch scale — became one of the founding ideas of landscape ecology once habitat fragmentation emerged as a dominant global conservation concern.

Because the classical Levins model assumes identical, equally connected patches, applying it quantitatively to a real fragmented landscape usually requires a spatially explicit extension that weights colonisation probability by inter-patch distance and patch size, and weights extinction probability by patch area (smaller patches generally support smaller, more extinction-prone local populations) — the qualitative persistence logic of Step 3 survives this added realism, but \(\hat p\) itself is no longer a single clean number.

Common misconception: that fragmenting a fixed total area of habitat into more, smaller patches is harmless as long as the total area is unchanged. Step 3 and its Corollary show this is generally false: smaller, more isolated patches typically raise \(e\) and lower \(c\) simultaneously, so the same total habitat area can support a persistent metapopulation when consolidated but fail to do so when fragmented.

Worked examples

\[ c=0.4\ \text{yr}^{-1},\quad e=0.1\ \text{yr}^{-1} \]

A patch network has an estimated colonisation rate \(c=0.4\) and local extinction rate \(e=0.1\) (arbitrary consistent units). Since \(c>e\) (Step 3), a stable positive equilibrium occupancy exists. A

\[ \hat p = 1-\frac{e}{c} = 1-\frac{0.1}{0.4}=0.75 \]

Applying Step 2's equilibrium formula directly: at equilibrium, roughly three-quarters of all patches are occupied at any given time, even though which specific patches are occupied changes continually as local extinctions and recolonisations occur. A

\[ \hat p = 0.75 \]

Reading. A metapopulation can be robustly persistent at the regional scale (75% mean occupancy) while every individual patch experiences repeated local extinction and recolonisation.

Scope. Doubling \(e\) alone (to 0.2) would drop \(\hat p\) to 0.5; raising \(e\) to 0.4 or above would eliminate the positive equilibrium entirely (Step 3).

Further example. A metapopulation of a rare butterfly inhabits a network of 500 meadow patches. Field estimates indicate a colonisation rate of \(c = 0.60\ \text{yr}^{-1}\) and a local extinction rate of \(e = 0.20\ \text{yr}^{-1}\). A proposed development project will fragment the habitat, reducing the effective colonisation rate to \(c' = 0.15\ \text{yr}^{-1}\) while the extinction rate remains unchanged. Calculate the equilibrium fraction of occupied patches (\(\hat{p}\)) before and after the fragmentation, and determine the percentage loss in the number of occupied patches.

\[ \hat{p}_{\text{initial}} = 1 - \frac{e}{c} = 1 - \frac{0.20}{0.60} \]

The Levins model equilibrium is derived from setting \(\frac{dp}{dt} = 0\), which yields \(c(1-p) = e\). Solving for \(p\) gives the standard formula. A

\[ \hat{p}_{\text{initial}} = 1 - 0.3333 = 0.6667 \]

This represents the stable state where the rate of new colonisations exactly balances the rate of local extinctions. A

\[ \hat{p}_{\text{final}} = 1 - \frac{e}{c'} = 1 - \frac{0.20}{0.15} \]

We apply the same equilibrium formula to the post-fragmentation parameters. Note that \(c' < e\) is not the case here, but \(c'\) is close to \(e\), indicating high sensitivity. A

\[ \hat{p}_{\text{final}} = 1 - 1.3333 = -0.3333 \]

A negative equilibrium occupancy is biologically impossible. In the Levins model, if \(c \le e\), the only valid equilibrium is \(p = 0\), indicating global extinction. A

\[ \text{Percentage loss} = \frac{0.6667 - 0}{0.6667} \times 100\% = 100\% \]

Since the metapopulation goes extinct, the loss of occupied patches is total. A

\[ \hat{p}_{\text{initial}} \approx 0.67, \quad \hat{p}_{\text{final}} = 0 \]

Reading. The initial metapopulation is stable with about two-thirds of patches occupied. However, the reduction in colonisation rate to 0.15 yr\(^{-1}\) drops below the extinction rate of 0.20 yr\(^{-1}\). This violates the persistence condition \(c > e\), leading to the complete collapse of the metapopulation.

Scope. This result holds for the standard Levins model with constant rates. In reality, stochasticity and spatial structure might delay extinction, but the deterministic prediction of collapse remains valid when \(c \le e\).

Problems

A conservation programme increases habitat connectivity, raising \(c\) from 0.3 to 0.5, while \(e\) remains at 0.2. Compute \(\hat p\) before and after the intervention.
Solution
Before: \(\hat p=1-0.2/0.3=0.33\). After: \(\hat p=1-0.2/0.5=0.60\). Increasing connectivity alone, with extinction rate unchanged, nearly doubles the equilibrium occupied fraction.
Two metapopulations have identical \(c=0.3\), but network A has \(e=0.1\) and network B has \(e=0.35\). Determine which, if either, is predicted to persist, using Step 3.
Solution
Network A: \(c=0.3>e=0.1\), so a stable positive \(\hat p=1-0.1/0.3\approx0.67\) exists; predicted to persist. Network B: \(c=0.3\lt e=0.35\), so \(c\le e\) and no stable positive equilibrium exists (Step 3); network B is predicted to go globally extinct despite sharing the same colonisation rate as network A.
Explain, using the source-sink concept (Step 5), why removing a small, seemingly low-value patch from a reserve network could cause unexpected declines in several other, larger patches.
Solution
If the removed patch happens to be a net source (birth exceeds death locally, exporting more colonists than it receives), the other patches may be sinks that only persist because of continual immigration from it. Removing the source patch would not appear costly by patch-occupancy or patch-size criteria alone, but would eliminate the immigration keeping the sink patches populated, potentially collapsing occupancy well beyond the removed patch itself (Common errors).
A wetland metapopulation of amphibians has a colonisation rate \(c = 0.25\ \text{yr}^{-1}\) and extinction rate \(e = 0.15\ \text{yr}^{-1}\). A climate change projection suggests that \(e\) will increase by 50% while \(c\) remains constant. Calculate the percentage decrease in the equilibrium occupancy \(\hat{p}\) resulting from this change.
Solution
Initial equilibrium: \(\hat{p}_1 = 1 - \frac{0.15}{0.25} = 1 - 0.60 = 0.40\). New extinction rate: \(e_{new} = 0.15 \times 1.5 = 0.225\ \text{yr}^{-1}\). New equilibrium: \(\hat{p}_2 = 1 - \frac{0.225}{0.25} = 1 - 0.90 = 0.10\). Percentage decrease: \(\frac{0.40 - 0.10}{0.40} \times 100\% = \frac{0.30}{0.40} \times 100\% = 75\%\). The equilibrium occupancy drops by 75%.
In the Levins model \(\dfrac{dp}{dt} = c\,p\,(1-p) - e\,p\), \(p\) is the fraction of habitat patches occupied, \(c\) the colonisation rate per occupied patch and \(e\) the local extinction rate. A network has 100 patches, \(c = 0.05\ \text{yr}^{-1}\) and \(e = 0.02\ \text{yr}^{-1}\). Find the equilibrium number of occupied patches, and the condition for the metapopulation to persist at all.
Solution
At equilibrium \(c(1-\hat p) = e\), so \(\hat p = 1 - e/c = 1 - 0.02/0.05 = 0.60\), i.e. 60 of the 100 patches. The metapopulation persists only if \(\hat p \gt 0\), i.e. \(c \gt e\): colonisation must outpace extinction. Destroying patches lowers the effective colonisation rate, so a network can collapse even while suitable habitat remains, a key insight for conservation.