The Gold Standard of Ecology | Occupancy Closure Violations, Temporary Emigration and Seasonal Availability

1. Occupancy is a latent state, not a raw detection rate

Back to contents

For site i, occupancy probability is commonly written ψ and detection probability p. Repeated detections such as 101 or 000 help estimate p while accounting for sites that are occupied but never observed.

The interpretation depends on what “occupied” means. It might mean an individual is physically present at an instant, the site lies within a home range, or the species uses the site during a season. Define the spatial unit and time window before fitting the model.

2. Closure is a statement about the chosen state

Back to contents

Single-season models assume that the occupancy state does not change across repeat surveys, or that movement does not bias the targeted parameter under an accepted site-use interpretation. This is a biological and design assumption, not a software option.

A breeding pond surveyed across two evenings may be nearly closed for a resident amphibian but open for a migratory bird. Shortening the interval can help, yet it can also create dependent observations. The right interval follows the process being estimated.

3. Availability sits between occupancy and detection

Back to contents

An occupied site may contain no animal available to the survey at that moment. A bird can be outside the point-count radius, a frog underground or a whale below the acoustic or visual window. Detection given occupancy can be understood as availability multiplied by detection given availability.

If availability changes among visits and is not modelled, the estimated detection parameter absorbs both processes. That may be acceptable for estimating seasonal site use, but not when the question requires instantaneous presence or temporary emigration itself.

4. Survey architecture decides what can be separated

Back to contents

A robust design nests secondary surveys within primary periods. Secondary visits estimate detection under closure or a site-use interpretation; primary periods allow colonisation and extinction or other distributional changes.

Record the exact spacing of surveys, effort, weather and phenological stage. Replicates taken minutes apart may share the same animal state and observer conditions. Replicates separated by days may gain independence while admitting availability changes.

5. Temporary emigration can imitate local extinction

Back to contents

Temporary emigration is a reversible change in availability. If all individuals using a site are unavailable during a primary period, a dynamic occupancy model can interpret that absence as extinction and a later return as colonisation.

Valente and colleagues showed that such confounding can bias dynamic-rate interpretation. Ask whether the timescale of estimated colonisation matches plausible demographic or range change, or merely movement in and out of the sampling frame.

6. Worked example: closure changes the estimand

Back to contents

At 100 wetlands, three weekly surveys produce detections at 40, 35 and 45 sites. Sixty sites are detected at least once. A closed model estimates ψ = 0.75 and p = 0.45. The probability of at least one detection at an occupied site is 1 − (1 − 0.45)^3 = 0.834.

If frogs are available at a used wetland on 0.75 of visits and detected with probability 0.60 when available, the product is also 0.45. The same detection histories can support seasonal site use of 0.75, but they do not identify availability and conditional detection separately without additional design or data. Calling ψ instantaneous occupancy would be too strong.

7. Choose the model by process and timescale

Back to contents

Closed occupancy may fit a brief resident-species survey. Multi-scale occupancy can separate site use from availability at a smaller scale. Staggered-entry models can estimate arrival and departure. Dynamic occupancy can estimate changes between primary periods when temporary availability is addressed.

Model choice is not a menu of increasing sophistication. Each model needs replication and identifiable parameters. Simulate the planned design under plausible ψ, availability and p to see which questions the data can answer.

8. Changed-context transfer check

Back to contents

Move the wetland example to camera traps for a wide-ranging cat. False absence and changing availability still matter. The site may now represent use of a camera viewshed rather than occupancy of an ecological territory.

Short visits by the same animal can create dependence, and movement among cameras can link sites. Redefine the state as site use over a stated interval, consider spatial encounter models or multiple scales, and avoid interpreting camera non-detection as territory abandonment.

9. Limits and common misconceptions

Back to contents

More repeat surveys do not rescue a mismatched state definition. Very close repeats may violate conditional independence; long repeats may violate closure. A model that converges can still estimate a parameter with the wrong biological meaning.

Do not assume a goodness-of-fit test has high power to detect every closure violation. Do not label reversible availability change as extinction. Do not compare ψ across seasons when survey windows cover different phenological stages without modelling the difference.

10. Audit the state–availability–detection chain

Back to contents

Plot detections against date, time, weather and effort. Review species movement, breeding and calling phenology. Check whether site size matches territory or movement range and whether the survey method samples the same space each visit.

Fit documented sensitivity models with alternative primary periods or site-use interpretations. If conclusions change when visits are regrouped, report that dependence. Archive survey calendars and code so another reader can reconstruct the closure claim.

11. Frequently asked questions

Back to contents

Does closure require every individual to stay put?

Not always. It requires the chosen site state to remain interpretable under movement. A site-use parameter can tolerate some movement that would violate instantaneous presence.

Can availability be estimated from detection histories alone?

Only with sufficient structure, scales or auxiliary data. In a simple model, availability and detection given availability often appear only as a product.

Should surveys always be closer together?

No. Closer spacing can reduce state change but increase dependence and may confound temporary emigration with dynamics at another scale.

12. The gold standard defines the ecological state first

Back to contents

Occupancy modelling is powerful because it makes imperfect detection explicit. It becomes misleading when a seasonal-use parameter is described as instantaneous presence or temporary unavailability is described as extinction.

Before interpreting ψ, γ or ε, ask what state the site can enter, how quickly it can change, and when an animal becomes available to the method. Those answers belong in the design and in the final words used for the result.