The Gold Standard of Ecology | Dynamic Occupancy Models, Colonisation, Extinction and Detection

1. Define the site and the state before the model

Back to contents

A site may be a pond, grid cell, forest patch or sampling station, and occupancy can mean residence, breeding, seasonal use or at least one visit. Fix the ecological interpretation and spatial grain before estimating transitions. If animals routinely move among small sites, the state may be use rather than population presence. Colonisation and extinction inherit this definition; changing the site boundary changes the process being measured.

2. Separate the ecological process from observation

Back to contents

The latent state changes between primary periods, while repeated surveys within each period generate detections or non-detections. A zero can mean true absence or an occupied site missed by observers. Detection may vary with weather, effort, observer, season and abundance. Model these predictors where justified. Without repeat information or external detection data, declining observations can be confounded with declining occupancy.

3. Understand initial occupancy, colonisation and extinction

Back to contents

Initial occupancy describes the probability sites are occupied in the first period. Colonisation is the probability an unoccupied site becomes occupied by the next period; local extinction is the probability an occupied site becomes unoccupied. These are site-state transitions, not automatically births, deaths, immigration or species-wide extinction. The expected occupancy in the next period combines persistence of occupied sites with colonisation of empty sites.

4. Keep closure within periods and change between periods

Back to contents

Repeated surveys within a primary period usually assume the state is closed, even though detection can vary. Transitions occur between periods. If individuals enter and leave during the survey window, detection may combine availability with observation, and apparent colonisation or extinction can reflect temporary movement. Choose short, biologically coherent secondary windows or use models that represent availability explicitly.

5. Design replication to identify both transition and detection

Back to contents

More years provide transition information; repeated surveys within years identify detection. Sites should span the covariates used to explain colonisation and extinction, with consistent effort and enough transitions in both directions. Sparse detections, near-perfect persistence or few primary periods can make parameters weakly identified. Simulate the planned design under plausible occupancy and detection before collection, not after a failed fit.

6. Worked example: frogs across seasonal ponds

Back to contents

Thirty ponds are surveyed three evenings in each wet season for four years. Naive presence counts fall after a dry year. The model includes rainfall in detection and hydroperiod in colonisation and extinction. Detection is lower on windy evenings, while local extinction rises in short-hydroperiod ponds and recolonisation is strongest near permanent wetlands. One pond visited only once in year three contributes little detection information. The conclusion concerns changing pond use under the defined season, not deaths of individual frogs.

7. Check false positives, heterogeneity and dependence

Back to contents

A recorded detection can be wrong, and unmodelled observer expertise may create both false positives and missed detections. Nearby sites can share dispersal, habitat and survey conditions, violating simple independence. Use verification data, expert confirmation or false-positive models when misidentification is plausible. Consider spatial or random effects when supported, but remember that a flexible model cannot recover transitions the sampling design never observed.

8. Changed-context transfer check: urban pollinator use

Back to contents

Move from frogs at ponds to pollinators visiting rooftop gardens. Occupancy may now mean use during a flowering period rather than residence. Gardens change plantings between years, and detection depends on temperature and survey duration. The transferable structure separates latent use, transitions and detection. The changed context requires explicit resource timing and perhaps multi-scale states because absence on one visit may reflect flower availability rather than local extinction.

9. Know the limits and common misconceptions

Back to contents

Dynamic occupancy does not identify abundance, demographic birth or individual survival. Local extinction may be temporary absence at the chosen grain. Covariate effects are conditional associations unless design supports causation. Unmodelled changes in detection can appear as distribution dynamics. Strong smoothing or priors can hide weak information, while annual site replacement destroys the transition meaning. State which sites and periods are comparable.

10. Audit histories, definitions and transition evidence

Back to contents

Publish site coordinates or defensible masks, primary and secondary occasion rules, effort, weather, observer identity, raw detection histories, false-positive checks, covariates, missing-survey codes, model equations, priors and code. Plot observed detections and predicted states by site and year. Test alternative closure windows, detection models and spatial structures, then simulate from the fitted model to see whether transition patterns are reproduced.

11. Frequently asked questions

Back to contents

Is colonisation the same as immigration? No; it is a transition to occupied at the site scale. Is local extinction species extinction? No. Why repeat surveys? To separate non-detection from absence. Can a site change state within a season? It can biologically, but the basic model assumes closure; redefine the season or model availability. Does more covariates solve low detection? No; design information remains essential.

12. The gold standard keeps state, transition and sighting distinct

Back to contents

A strong dynamic occupancy study defines what an occupied site means, builds repeat observation into each period and interprets colonisation and extinction at the correct spatial and temporal scale. It exposes false positives, closure failures and missing effort instead of letting zeros carry every meaning. The reward is not merely a smoother distribution map, but a transparent account of where use persisted, appeared or disappeared and how confidently the surveys could see it.