How Science Works | How Habitat Patches Stay Connected Through Local Populations

A habitat patch is an area treated as a local unit for the question being asked. Its boundary should reflect the focal organism’s resources and movement, rather than merely a convenient line on a map. A small meadow may be a useful unit for one insect but only a minor part of a bird’s daily range.

A local population consists of the focal species in that patch. Occupancy records whether the population is present under a stated definition. Abundance records how many organisms are present. Ten occupied patches containing two individuals each differ greatly from ten containing two hundred each, even though their occupancy values match.

Classical patch models deliberately set much abundance detail aside. This can reveal how the availability of occupied sources and empty habitat affects regional dynamics. It also means the model cannot answer every question about population size, breeding success or genetic diversity.

Not every spatially structured population is well described by frequent local extinction and recolonization. Some have large, persistent populations linked by occasional exchange. Others have sources continually supporting populations that would otherwise decline. The metapopulation perspective is useful when its representation matches the system, not simply whenever a map contains several habitat shapes.

Before counting patches, therefore, state what qualifies as a resident population. A passing individual may show movement without establishment. Evidence of reproduction can be important, especially when the scientific claim concerns a continuing local population rather than temporary use of a site.

Contents · Next chapter

Local extinction means that a population disappears from a specified patch. Regional extinction means that all relevant local populations in the defined region have disappeared. The distinction depends on spatial scale: a regional loss in one study could still leave populations outside that study’s boundary.

Recolonization occurs when a previously occupied patch is occupied again after a local extinction. Colonization more generally concerns establishment in an empty patch, whether or not a previous population is known. Immigration into a population that is still present is different again.

Immigrants can reduce the chance that a small existing population disappears. This demographic reinforcement is often called a rescue effect. It should be distinguished from re-establishment after a population has already vanished, even though coarse surveys may blur the difference.

Diagram to sketch: draw three suitable patches, A, B and C. At the first time, A and B are occupied and C is empty. At the second, A is empty, B remains occupied and C has been colonized. Occupancy stays at two out of three, but the landscape has experienced both a local loss and a gain. Add arrows only where dispersal is part of the stated example.

A stable total therefore need not mean nothing happened. Tracking patch identities and transitions reveals turnover that a regional count hides. Conversely, seeing the same patch occupied on two dates does not prove continuous persistence between them: extinction and recolonization could have occurred during the unobserved interval.

Contents · Previous chapter · Next chapter

Consider an original simplified model with a fixed number of equally suitable patches. Let p be the fraction occupied at the start of an interval. A fraction 1 − p is initially empty. We model occupancy rather than the number of individual organisms.

Let e be the probability that an initially occupied patch becomes empty during the interval. Let the probability that an initially empty patch becomes occupied be c × p, where c is a specified colonization parameter. More occupied patches mean more potential sources in this deliberately well-mixed description.

The update is p next = p(1 − e) + (1 − p)cp. The first term represents occupied patches that remain occupied. The second represents initially empty patches that become occupied. Equivalently, change in occupancy is cp(1 − p) − ep: gains minus losses.

The timing rule matters. In this model, a patch becoming empty during the interval is not recolonized within that same step. Colonization applies only to patches empty at the start. We also omit outside immigration, differences among patches, rescue effects and correlated disturbances. Those are choices, not claims that such processes are absent in nature.

Take c and e between zero and one so the stated probabilities remain valid for p between zero and one. Both refer to the chosen interval. They are not continuous-time rates that can be transferred unchanged into a differential equation or multiplied freely when the interval changes.

The recurrence is a deterministic mean-field teaching model. One update can summarize conditional expected transitions under its rules, but repeatedly applying it is not generally the exact expected trajectory of a small stochastic network with fluctuating occupancy.

Contents · Previous chapter · Next chapter

Set c = 0.4, e = 0.1 and p = 0.5. Imagine one hundred patches, initially fifty occupied and fifty empty. The probability of colonizing an initially empty patch is cp = 0.4 × 0.5 = 0.2.

Among the fifty occupied patches, expected losses are 50 × 0.1 = 5. Among the fifty empty patches, expected gains are 50 × 0.2 = 10. The conditional expected number occupied next is 50 − 5 + 10 = 55, corresponding to p next = 0.55.

The denominators differ. Extinction probability applies to occupied patches; colonization probability applies to empty patches. Ten gains among fifty empty patches means twenty per cent of the available empty patches became occupied, not ten per cent. Ten per cent describes the gains as a fraction of all one hundred patches.

A second deterministic update uses the new model value p = 0.55. Surviving occupancy is 0.55 × 0.9 = 0.495. New occupancy is 0.45 × 0.4 × 0.55 = 0.099. The next value is therefore 0.594.

This does not predict that an actual hundred-patch landscape must contain 59.4 occupied patches. The fraction is a smooth model summary. Real patch counts are whole numbers, and random outcomes can differ. A stochastic model would describe a distribution of possible counts and could include dependence between patches.

The arithmetic’s purpose is to expose a balance: sources help create gains, but gains also require empty suitable places. When every patch is occupied, this model has no initial empty patches available for colonization during that step, even though local losses can still occur.

Contents · Previous chapter · Next chapter

At a positive fixed point, gains and losses balance: cp(1 − p) = ep. Dividing by p, which is valid only when p is positive, gives c(1 − p) = e. The nonzero equilibrium is therefore p* = 1 − e/c when c is positive and greater than e.

For c = 0.4 and e = 0.1, p* = 1 − 0.25 = 0.75. At this fraction, the loss term is 0.1 × 0.75 = 0.075. The gain term is 0.4 × 0.75 × 0.25 = 0.075. The occupied fraction is steady because equal flows oppose one another, not because every patch is permanent.

For a hundred-patch interpretation, those terms correspond to 7.5 expected losses and 7.5 expected gains per interval. Again, fractional expected counts summarize possibilities; they do not describe half an actual event.

The zero-occupancy state is also a fixed point. With no occupied sources and no outside arrivals, the model cannot regenerate a population from nothing. When e is positive and c is less than or equal to e, there is no positive fixed point, and positive occupancy declines under these rules. At the special corner c = e = 0, no transitions occur and every initial occupancy remains unchanged.

Most importantly, a positive deterministic equilibrium is not a guarantee of indefinite persistence in a finite real network. Chance can eliminate the remaining populations, and a shared disturbance can affect many patches together. The model identifies a mechanism and a conditional balance; estimating extinction risk needs a representation of stochasticity, network structure and the relevant observation period.

Contents · Previous chapter · Next chapter

Two landscapes may each contain ten occupied and ten empty patches. In one, occupied patches lie close to most empty suitable patches. In the other, all occupied patches form an isolated cluster. The single number p = 0.5 hides that difference.

A spatial model can assign each empty patch its own colonization probability based on the locations, sizes and states of potential sources. Distance alone may still be insufficient: the intervening landscape can alter movement and establishment. Connectivity is functional and species-specific, rather than simply the straight-line distance between patch centres.

Patch area and quality can affect local extinction as well as disperser production. A large poor-quality patch is not automatically equivalent to a smaller productive one. A model must state how the measured habitat features connect to demographic processes rather than assigning importance solely by appearance.

Hanski and Ovaskainen’s 2000 theoretical study developed metapopulation capacity, a landscape measure incorporating patch areas and connectivities. In their framework it is the leading eigenvalue of an appropriate landscape matrix, and persistence depends on its comparison with a species-specific threshold. This is a model-based way to evaluate configuration, not a universal score independent of organismal biology. Original research abstract.

The beginner’s lesson does not require calculating an eigenvalue. It is that two landscapes with the same total habitat or occupied fraction can differ in their capacity to sustain exchange. A more realistic model earns its additional complexity when it improves predictions using measured differences that the simpler occupancy model omits.

Contents · Previous chapter · Next chapter

A strong dataset records identified patches repeatedly, including those where the focal species was not detected. Surveying only occupied locations would discard the very empty habitat needed to estimate colonization opportunities. Changes in the patch inventory also matter: destroyed habitat is not simply another suitable but currently empty patch.

For each survey interval, investigators can compare which occupied patches remained occupied and which empty patches gained populations. They can then examine whether transitions depend on patch area, host resources, isolation, weather and neighbouring populations, rather than assuming one common c and e.

A 2017 Glanville fritillary study used twenty-two years of records across 4,415 small meadows in the Åland Islands. It compared a spatial metapopulation model with observed occupancy and examined ecological and genetic contributions to persistence. Many of the defined patch networks fell below its estimated long-term persistence threshold. That threshold was a model inference tied to the studied networks and species, not a claim that every occupied meadow was about to lose its butterflies. Original research.

Long records help because a patch can temporarily remain occupied even when conditions no longer favour long-term regional persistence. They also allow researchers to distinguish an unusual year from a repeated pattern. Nevertheless, many years of observations do not eliminate unmeasured habitat change or turn every association into a causal effect.

The most informative test asks whether the model predicts transitions or occupancy in data not used to tune it. A model that reproduces its training records but misses new years may be summarizing the past without capturing the processes needed for prediction.

Contents · Previous chapter · Next chapter

Suppose a fictional patch is occupied in September, empty in March and occupied again the following September. Annual September surveys alone would record presence followed by presence. They would miss the intervening local extinction and recolonization. More frequent observation changes which transitions can be identified.

This is different from imperfect detection. If a population is present but missed, an apparent extinction can occur in the observation record without a biological extinction. Repeated visits within an appropriate period help distinguish nondetection from absence, provided the assumptions about population closure and detection are reasonable.

In a simple original example, assume a population remains present and each visit independently detects it with probability 0.6. The chance of missing it in three visits is (1 − 0.6)³ = 0.064, or 6.4%. One missed visit would have a forty-per-cent probability under the same model.

Three misses still do not establish a 6.4% probability that the population is present. That reverses the conditional question. Inferring presence after nondetections also requires prior occupancy information and a valid observation model. Independence and constant detection are substantive assumptions, not automatic benefits of repeating a survey.

A 2024 Glanville fritillary study explicitly compared overwinter and annual extinction measures alongside genetic, demographic and environmental information. It noted that populations lost over winter could be recolonized before the next autumn survey. Its findings underline how census timing and demographic context affect apparent extinction relationships. Original PNAS research.

Contents · Previous chapter · Next chapter

Increasing exchange can replenish small populations and connect empty suitable habitat with sources. It can also change how closely local populations fluctuate together. If every patch experiences its lowest abundance at the same time, fewer strong sources may remain when recolonization is most needed.

Environmental synchrony can arise even without movement: the same drought or storm can affect many patches. Conversely, local environmental differences can maintain asynchronous outcomes. Dispersal and shared environmental forcing must therefore be distinguished when explaining why populations rise and fall together.

A 2001 experiment with bean plants, herbivorous mites and predatory mites found longer persistence with habitat structure. Crucially, the authors’ mechanistic modelling attributed the result to reduced predator success in locating prey outbreaks and asynchronous local cycles following random colonization, rather than to spatial population structure itself. A visually patchy system did not automatically validate the simplest spatial explanation. Original research abstract.

Other complications include source–sink dynamics. A sink may remain occupied because immigrants offset local demographic losses. Occupancy alone cannot establish that it is a self-sustaining source. Movement can also alter disease transmission and the exchange of locally adapted genetic variants, with consequences depending on the organisms and circumstances.

These possibilities do not justify blanket opposition to connectivity. They show why a useful claim specifies the process improved and measures the resulting demographic outcome. More movement, more mapped connections and lower regional extinction risk are related hypotheses, not three interchangeable observations.

Contents · Previous chapter · Next chapter

A useful worksheet starts with the habitat definition, patch boundary, occupancy criterion and survey interval. Then mark whether each apparent transition is a confirmed biological change, a detection-dependent inference or an unobserved possibility. This makes uncertainty visible before adding an equation.

For the original model, check five points. With p = 0.5 and c = 0.4, an initially empty patch has a colonization probability of 0.2. Expected gains among fifty empty patches are ten. Expected losses among fifty occupied patches at e = 0.1 are five. The next occupied fraction is 0.55. The positive fixed point at these parameters is 0.75, with continuing balanced turnover.

Next challenge the model. What if one large occupied patch supplies most dispersers? What if a drought affects every patch at once? What if empty patches are no longer suitable? Each question identifies a missing mechanism that could change the prediction. The appropriate extension depends on evidence, not on making the model as complicated as possible.

A defensible conclusion might say that repeated local losses and establishment, together with measured dispersal and habitat suitability, support a metapopulation account of regional persistence. It should also state whether the evidence distinguishes rescue from recolonization and whether abundance changes are hidden behind stable occupancy.

The central idea is hopeful but conditional: a local loss need not be the end of a regional population when suitable habitat, surviving sources and effective dispersal remain. Continue through Conservation Biology for broader persistence questions, or return to the How Science Works master guide for another scientific mechanism.

Contents · Previous chapter · Continue to Conservation Biology