Skip to content

Metapopulation Model

Dynamical model — instantiates Source–Sink Viability Management

Runs a network of coupled patches forward from their per-patch birth–death and dispersal rates to forecast whether the whole persists — and which patches are true sources versus occupied-but-doomed sinks.

A Metapopulation Model is the predictive engine of the archetype: it treats the system as a set of coupled patches, each with its own local net rate of production and loss, linked by dispersal, and it simulates the whole forward in time to ask whether it persists. Its one defining move — the thing no sibling does — is deriving aggregate viability from per-patch rates plus movement between patches, which is what exposes the archetype's core trap: a patch can be reliably occupied and still be a demographic sink whose local rate is below replacement, kept full only by immigration from a source. Occupancy is not viability, and only a model that runs the coupling forward can tell the two apart. Where the Dispersal or Transfer Tracer measures the rates and the Source–Sink Patch Map labels the patches, this model computes the consequences of those rates for survival of the network.

Example

A conservation team studies a frog whose ponds form a patchwork across a valley. Field counts say all eight ponds are occupied, so the habitat "looks fine." The team builds a metapopulation model: each pond gets a local net reproductive rate (its own births minus deaths, ignoring arrivals) and a dispersal estimate for how many froglets move pond-to-pond. Run forward, the model tells a different story — three ponds have a net rate below one (each generation would shrink on its own) and stay full only because two large, high-output ponds spill dispersers into them season after season.

The output is not a snapshot but a forecast with a threshold: the network persists as long as the two source ponds hold, but if either is lost, occupancy collapses across the sinks within a few generations. That reframes the whole problem — the "healthy" eight-pond system is really a two-pond system wearing a disguise, and protection should concentrate on the sources rather than being spread evenly across ponds that cannot pay their own way.

How it works

  • Parameterize each patch by its unaided rate. Every patch carries a local net rate of production versus loss, estimated without imports, so sources (surplus) and sinks (deficit) separate on their own demographics rather than on how full they currently are.[1]
  • Couple the patches by dispersal. A movement kernel says how much surplus reaches which patches, turning a list of independent sites into one interdependent system.
  • Simulate occupancy and extinction forward. Run the coupled dynamics over time to get persistence probability and the extinction threshold — the point where too few sources remain to keep the sinks occupied.
  • Attribute persistence to sources. Trace which patches the network's survival actually depends on, so the model outputs not just "will it persist?" but "on whose surplus?"

Tuning parameters

  • Spatial grain — whether a "patch" is a single site or an aggregated cluster. Finer grain reveals local sink-dependence but multiplies parameters you cannot estimate well.
  • Dispersal-kernel shape — how far and how much surplus travels. Optimistic kernels manufacture rescue that isn't there; the model is most sensitive to this dial.
  • Stochasticity — deterministic averages versus demographic and environmental noise. Adding noise reveals fragility that mean-value runs hide, at the cost of needing distributions, not point estimates.
  • Time horizon — how many generations forward you project; short horizons can miss a slow collapse already baked in.

When it helps, and when it misleads

Its strength is making hidden sink-dependence and cascade risk visible before the shock: it separates real sources from rescued sinks, sizes how much slack the network has, and shows which single loss would unravel it. It is the input that keeps stewardship from being spread evenly across patches that will never be self-viable.

It misleads mainly through its inputs. Local rate and dispersal estimates are hard to get, so a tidy persistence curve can carry false precision over parameters that are really guesses. The classic error is calibrating the model to occupancy data — which is exactly the signal a rescued sink corrupts — so the model confidently reproduces a healthy-looking network while missing that it is propped up. And like any model it can be run backwards, tuned until it endorses a protection plan someone already chose. The discipline that guards against this is to estimate patch rates independently of occupancy, carry parameter uncertainty into the persistence forecast, and treat a source-loss scenario as the headline result rather than a footnote.

How it implements the components

Metapopulation Model realizes the predictive slice of the archetype's machinery — the components that require running the coupled system forward, not measuring or governing it:

  • sink_viability_threshold — the model computes the import level (and the number of surviving sources) below which a sink, and then the network, drops out of viability; that threshold is its central output.
  • scenario_stress_test — losing a source, a bad breeding year, a severed dispersal route are run forward as scenarios to see which ones cross the threshold.

It does not measure the raw flows and net balances it runs on — those come from the Dispersal or Transfer Tracer; it does not classify or label the patches — that is the Source–Sink Patch Map; and it does not decide what to do about a doomed sink — that is the Restoration Priority Matrix.

Notes

The model's most useful and most dangerous property is the same one: it can keep a network looking occupied right up to the threshold, because rescued sinks stay full until the source that fills them fails. A persistence forecast should therefore always be read alongside which patches it depends on — a high occupancy concentrated on two sources is far more fragile than the same occupancy spread across self-viable patches.

References

[1] The source–sink distinction rests on the local net reproductive rate: a source's own births exceed its deaths (surplus for export), a sink's do not, so a sink persists only by immigration. This is the standard demographic definition of a source versus a sink, and it is why classification must be built on unaided rates rather than on how occupied a patch looks. ↩