Skip to main content
v2026.11,986 entries · CC-BY 4.0

Occupancy Models: Principles, Mathematical Formulations, and Ecological Applications

Occupancy models estimate true species occurrence by separating presence from imperfect detection using repeated surveys and hierarchical models.

Written and maintained by CASRAI Editorial Board

Published

On this page

Occupancy models are hierarchical statistical frameworks used in ecology and biosurveillance to estimate the true probability of species occurrence across spatial sites while explicitly correcting for imperfect detection. Developed by quantitative ecologists to overcome the negative bias of raw presence-absence surveys, an occupancy model uses repeated detection and non-detection observations across sample units to estimate two distinct parameters: the true probability of site occupancy (ψ) and the conditional probability of detection (p). By decoupling the underlying biological state from the observational sampling process, occupancy models allow field researchers, environmental agencies, and laboratory biosurveillance teams to determine true geographic distributions, evaluate habitat suitability, and track population dynamics without requiring costly individual marking or animal handling.

Definition and scope of ecological occupancy models

In quantitative ecology, determining species distributions across landscapes is fundamental. Traditional field surveys record whether an organism is detected at a site. However, species are rarely detected with certainty due to cryptic coloration, low abundance, and dense cover. Analyzing raw detection data without correcting for missed sightings confounds biological absence with detection failure. Occupancy models resolve this issue by simultaneously estimating occurrence and detection probabilities.

The scope of occupancy modeling spans conservation biology, wildlife management, and environmental monitoring across terrestrial, freshwater, and marine ecosystems. It accommodates diverse field methods, including point counts, camera traps, acoustic arrays, environmental DNA (eDNA), and botanical quadrats. By integrating sampling and environmental covariates, occupancy models isolate the true ecological drivers of species distribution.

What is occupancy?

Occupancy is the proportion of habitat patches, sample units, or landscape cells occupied by a target organism. In population ecology, site occupancy is a binary state variable: a site is either inhabited (z = 1) or uninhabited (z = 0) by the focal species during a defined period. Occupancy serves as an effective proxy for abundance when counting individuals is technically infeasible or excessively invasive.

What is ecological occupancy?

Ecological occupancy refers to the genuine biological occurrence of a species within an ecosystem, reflecting fundamental processes such as resource selection, niche partitioning, and metapopulation dynamics. Unlike opportunistic sightings, ecological occupancy represents a latent state that cannot be directly observed with certainty when detection probability is less than 100%. Formally, it quantifies the probability that a randomly chosen site contains at least one individual of the target species interacting with local habitat.

What does “naive occupancy” mean?

Naive occupancy is the unadjusted proportion of surveyed sites where a species was detected at least once, calculated as sites with positive detections divided by total sites sampled. Naive occupancy assumes perfect detection (p = 1.0). When detection is imperfect (p < 1.0), naive occupancy systematically underestimates true occurrence, distorting habitat associations and masking population trends.

How occupancy modeling works in practice

Occupancy modeling relies on repeated observations across discrete spatial sites during temporal closure, where no colonization or extinction occurs between visits. Repeated visits are conducted across consecutive days, by independent observers, via automated sensors, or through replicate environmental samples.

Sampling is structured as a two-level hierarchical model separating a latent biological state process from an observational detection process. For S sample sites and K repeated visits, the unobserved true occupancy state for site i is a binary variable zi: zi ~ Bernoulli(ψi).

During survey visit j (where j = 1, …, K), the observer records detection (yij = 1) or non-detection (yij = 0). The observation process is conditional on latent state zi. If unoccupied (zi = 0), the species cannot be detected under the assumption of no false positives: yij = 0 with probability 1.0. If occupied (zi = 1), detection occurs with conditional probability pij: yij | zi ~ Bernoulli(zi × pij).

Over K visits, each site generates an observational detection history of ones and zeros (Bailey and Adams, USGS Fact Sheet 2005-3096). For a site surveyed four times (K = 4) yielding history Hi = 1001 (detected on visits 1 and 4, missed on visits 2 and 3), the site is occupied (zi = 1). Assuming survey independence, the probability of this history is: Pr(Hi = 1001) = ψ × p1 × (1 – p2) × (1 – p3) × p4.

If the species was never detected (Hi = 0000), the site was either unoccupied (probability 1 – ψ) or occupied but undetected across all visits (probability ψ ∏ (1 – pj)). The mixture probability is: Pr(Hi = 0000) = (1 – ψ) + ψ × [(1 – p1) × (1 – p2) × (1 – p3) × (1 – p4)]. For K occasions, Pr(Hi = 00…0) = (1 – ψ) + ψ ∏j=1K (1 – pj). The full likelihood multiplies history probabilities across all S sites, solved via maximum likelihood estimation.

Covariates enter through logit link functions. Site-level covariates influence occupancy ψi, while survey-level covariates (such as temperature and observer bias) influence detection pij: logit(ψi) = β0 + β1Xi1 + … + βmXim and logit(pij) = α0 + α1Vij1 + … + αnVijn. This prevents omitted variable bias by isolating habitat relationships while controlling for detection artifacts.

What are the different types of ecological models?

Ecological models are formal representations of biological systems designed to understand patterns and predict environmental responses. In quantitative ecology, models are categorized by mathematical structure:

  • Phenomenological models: Regressions describing empirical correlations without specifying underlying biological mechanisms.
  • Correlative species distribution models: Algorithms associating occurrence records with GIS layers to predict geographic suitability, including MaxEnt and models in species distribution modeling software like dismo.
  • Mechanistic models: Simulations modeling biological processes such as thermal tolerances and energy budgets to determine suitability.
  • Metapopulation models: Formulations tracking local population turnover across habitat patches through extinction and colonization dynamics.
  • Individual-based models: Simulations treating organisms as distinct agents within landscapes, used in movement ecology frameworks.
  • Hierarchical state-space models: Statistical models separating latent ecological processes from observation errors, including occupancy models and N-mixture abundance models.

Within this hierarchical framework, quantitative ecologists recognize five primary occupancy model classes:

  • Single-season occupancy models: The foundational framework introduced by MacKenzie et al. (2002) for a closed season, estimating static occupancy (ψ) and detection (p) via occu() in the R package unmarked.
  • Dynamic multi-season occupancy models: Formulated by MacKenzie et al. (2003) for multi-year monitoring, estimating initial occupancy (ψ1), colonization (γt), extinction (εt), and survey detection (pt,j) via colext() in unmarked.
  • Multi-scale occupancy models: Formulations for hierarchical sampling where secondary units are nested within primary sites, estimating large-scale occupancy (ψ), small-scale habitat use (θ), and survey detection (p).
  • Community and multi-species models: Hierarchical Bayesian frameworks evaluating entire assemblages simultaneously, utilizing statistical shrinkage to improve parameter estimates for rare species.
  • False-positive occupancy models: Extended models relaxing the zero false-positive assumption to accommodate misidentifications in citizen science or eDNA surveys via occuFP() in unmarked.

Comparative analysis of occupancy modeling frameworks

Selecting an appropriate occupancy model requires matching the ecological question and field protocol with the correct mathematical structure. The table below compares the five principal occupancy modeling frameworks utilized in contemporary research and biosurveillance.

Model Framework Core State Parameters Observational Parameters Key Biological Assumptions Primary R Function Foundational Reference
Single-Season Occupancy ψ (Occupancy probability) p (Detection probability) Demographic and geographic closure during survey window; no false detections; independent sites occu() in unmarked MacKenzie et al. (2002) / Bailey and Adams (2005)
Dynamic Multi-Season ψ1 (Initial occupancy); γt (Colonization); εt (Extinction) pt,j (Survey-specific detection) Closure within primary seasons; open between seasons; Markovian state transitions colext() in unmarked MacKenzie et al. (2003) / Fiske and Chandler (2011)
Multi-Scale Occupancy ψ (Large-scale occupancy); θ (Small-scale conditional use) p (Sample-level detection) Closure at primary and secondary sampling scales; nested spatial sampling design occuMulti() in unmarked Nichols et al. (2008) / Mordecai et al. (2011)
Community / Multi-Species ψs (Species occupancy); community hyper-parameters ps (Species detection); community hyper-parameters Closure during surveys; species responses drawn from shared community distributions communityOccu() in unmarked Dorazio and Royle (2005) / Fiske and Chandler (2011)
False-Positive Occupancy ψ (Occupancy); γ (Colonization); ε (Extinction) p11 (True detection); p10 (False-positive detection); b (Certainty probability) Subset of unambiguous detections or multi-method validation; closure within sampling occasions occuFP() in unmarked Miller et al. (2013) PLOS ONE

Empirical benchmarks: naive occupancy versus model-estimated occupancy

To evaluate the impact of imperfect detection, ecologists examine field datasets where naive observations directly contrast with detection-corrected occupancy estimates. Two empirical studies demonstrate how failure to model detectability distorts scientific conclusions.

USGS Great Smoky Mountains salamander monitoring

In a wildlife study documented in USGS Fact Sheet 2005-3096, Larissa Bailey and Michael Adams analyzed field surveys of terrestrial salamanders (Desmognathus imitator complex) in Great Smoky Mountains National Park across 39 trail transects (spaced 250 meters apart) surveyed across four sampling occasions.

Salamanders were observed at least once at 10 of 39 sites, yielding a naive occupancy of 0.26 (26%). Researchers hypothesized that occupancy depended on forest disturbance history, while detection varied with rainfall and moisture. In PRESENCE, candidate models evaluated constant occupancy ψ(.) versus disturbance-dependent occupancy ψ(dist), alongside constant detection p(.), survey-specific detection p(t), and disturbance-specific detection p(dist).

All top-ranking AIC models retained forest disturbance as a predictor. Model-averaged occupancy was 0.70 (SE = 0.15) for undisturbed forest sites and 0.19 (SE = 0.15) for disturbed sites. Naive occupancy (0.26) severely obscured this disparity by blending inhabited undisturbed sites with depauperate disturbed areas, demonstrating how raw presence-absence data mask habitat degradation when detectability is low.

Plant biosurveillance in granite rock outcrop pools

In an open-access botanical study in Ecology (PMC13598730), Anna W. Wyngaarden and Megan L. DeMarche evaluated imperfect detection in Gratiola amphiantha (snorkelwort), a rare annual wetland plant endemic to ephemeral granite outcrop pools in the southeastern United States.

Botanical surveys frequently assume perfect detection (p = 1.0). Wyngaarden and DeMarche evaluated pool occupancy across three seasons (2023-2025) at Arabia Mountain and Heggie’s Rock. In the 2025 dataset, single-survey sampling designs (naive occupancy) estimated high occupancy probability (ψ ≥ 0.50) in only 56% of pools at Arabia Mountain and 8% at Heggie’s Rock. In contrast, a four-survey protocol analyzed with single-season occupancy models in RPresence estimated occupied pool proportions of 79% at Arabia Mountain and 51% at Heggie’s Rock.

Single surveys underestimated pool occupancy by 23% at Arabia Mountain and 43% at Heggie’s Rock due to vegetative overgrowth and ephemeral phenology. Mean estimated detection probability was p = 0.87 (range 0.798 to 0.977). Using cumulative formula p* = 1 – (1 – p)K, achieving 95% certainty (p* ≥ 0.95) required Nmin = 1.46 surveys. Exactly two repeat visits within a season were sufficient to achieve 95% detection confidence in occupied pools, proving that modest repeat sampling eliminates substantial undercounting bias.

Standardized protocol for designing and analyzing occupancy studies

To implement an occupancy modeling workflow in field biosurveillance and laboratory monitoring, researchers follow a structured six-step sequence:

  1. Delineate spatial sampling units: Define discrete sample sites based on habitat patches, grid cells, or transects, separated by buffer distances to ensure spatial independence.
  2. Establish the temporal closure window: Define a survey timeframe during which occupancy remains constant (such as a breeding season or flowering window).
  3. Conduct standardized repeated surveys: Execute K repeated surveys (typically 2 to 5 visits) at each site within the closure window, standardizing search effort and observer protocols.
  4. Construct detection history matrices: Assemble observations into an S × K matrix of ones and zeros, compiling environmental covariates and survey-specific sampling covariates.
  5. Fit hierarchical models and rank candidate sets: Import detection histories into modeling software (such as R package unmarked or PRESENCE). Fit candidate models for ψ and p, ranking models using Akaike’s Information Criterion (AIC).
  6. Assess goodness-of-fit and derive predictions: Evaluate model fit using MacKenzie-Bailey bootstrap tests, calculate model-averaged occupancy predictions, and map spatial distribution patterns across the study region.

Occupancy modeling in laboratory operations and field biosurveillance

Occupancy modeling principles extend beyond wildlife ecology into modern laboratory operations protocols, environmental diagnostics, and disease biosurveillance. The adoption of molecular tools—specifically environmental DNA (eDNA), quantitative PCR (qPCR), and digital droplet PCR (ddPCR)—makes hierarchical occupancy modeling an essential analytical quality control tool.

In eDNA laboratory workflows, a physical sample represents a primary site, while technical PCR replicates represent repeated survey occasions. Because target DNA fragments in water bodies are often present at low concentrations near assay detection limits, PCR may fail to amplify sequences even when DNA is present. Fitting multi-scale occupancy models allows laboratories to separate three nested probabilities: species presence in the water body (ψ), DNA capture in a sample bottle (θ), and PCR amplification success in a 96-well plate (p). This prevents false-negative reporting in testing for endangered species, invasive mussels, and pathogens like Batrachochytrium dendrobatidis.

In vector-borne disease surveillance—such as testing mosquito pools for viral pathogens—occupancy models account for imperfect detection arising from sample degradation, extraction inefficiencies, and primer binding variations. Incorporating probabilistic graphical models enables laboratory directors to calibrate minimum replicate thresholds, ensuring diagnostic screening protocols achieve certified statistical confidence before declaring an area disease-free.

Common misunderstandings in occupancy modeling

Despite the widespread adoption of occupancy estimation across ecology, several persistent misconceptions recur in study design and data interpretation:

  • Conflating non-detection with biological absence: Treating raw survey zeros as true absences systematically underestimates range limits and overestimates habitat specificity whenever detection probability is less than 1.0.
  • Assuming temporal closure is unconstrained: Occupancy models require that sites do not change state during the survey period. If individuals move in or out between visits, parameter ψ reflects “habitat use” rather than true occupancy.
  • Overlooking false-positive errors: Standard occupancy models assume zero false positives (p10 = 0). In citizen science or cross-contaminated PCR plates, false detections violate likelihood assumptions unless explicit false-positive models are applied.
  • Equating site occupancy directly with numerical abundance: While occupancy often correlates with density, it represents spatial distribution rather than total individual counts. A species can maintain stable occupancy while undergoing numerical declines.
  • Failing to account for unmodeled heterogeneity: If individuals or sites possess unmeasured traits that alter detectability, detection probability exhibits unmodeled heterogeneity, causing positive bias in occupancy estimates.

The data table below compiles empirical detection probabilities, survey effort thresholds, and parameter biases across diverse taxonomic groups, contrasting raw survey counts against hierarchical model estimates.

Frequently asked questions

What is ecological occupancy?

Ecological occupancy is the true biological probability that a defined habitat patch is inhabited by a target species during a specific time period. It reflects underlying ecological processes, separated from observational detection as a latent state variable.

What are the different types of ecological models?

Ecological models encompass phenomenological regressions, correlative species distribution models (such as MaxEnt), mechanistic physiological models, metapopulation models, individual-based simulations, and hierarchical state-space models. Occupancy models represent a major hierarchical class analyzing detection and non-detection data.

What is occupancy?

Occupancy is the proportion of discrete spatial sampling units, habitat patches, or landscape grid cells where a target organism is present. It serves as a binary state indicator (present or absent) that functions as a robust metric for monitoring population distributions.

What does “naive occupancy” mean?

Naive occupancy is the simple ratio of surveyed sites with at least one confirmed species detection divided by total sites sampled. It assumes perfect detection (p = 1.0) and underestimates true occurrence when detectability is imperfect.

How many repeat surveys are required to confirm species absence?

The number of repeat surveys (K) required to confirm absence depends on per-survey detection probability (p). Using formula K ≥ ln(1 – C) / ln(1 – p) for confidence C (such as 0.95), if single-visit detection is 0.60, a minimum of 4 surveys reaches 97.4% confidence that an undetected site is genuinely unoccupied.

What software packages fit occupancy models?

The primary open-source packages for fitting occupancy models are R package unmarked (using maximum likelihood estimation via C++ backends) and USGS PRESENCE. Bayesian formulations are implemented in R using packages such as jagsUI, nimble, or brms connected to JAGS or Stan.

Sources

Data: Occupancy Modeling Parameters and Imperfect Detection Benchmarks Across Taxa

Comparative compilation of empirical single-visit detection probabilities, estimated true occupancy rates, naive occupancy rates, and required repeat survey effort to reach 95% detection certainty across published field studies.

Period covered: 2005-2026. Coverage: United States.

Occupancy Modeling Parameters and Imperfect Detection Benchmarks Across Taxa
Study Taxon and SystemSingle-Visit Detection Probability (p)Naive Occupancy Proportion (proportion)Model-Estimated True Occupancy (proportion)Occupancy Underestimation Gap (percentage points)Surveys for 95% Certainty (surveys)
Desmognathus imitator salamanders (Undisturbed forest, Great Smoky Mountains)0.430.260.7446
Desmognathus imitator salamanders (Disturbed forest, Great Smoky Mountains)0.380.260.19-77
Gratiola amphiantha aquatic snorkelwort (Arabia Mountain granite pools)0.870.560.79232
Gratiola amphiantha aquatic snorkelwort (Heggie's Rock granite pools)0.820.080.51433
Canis lupus gray wolves (Montana public survey and telemetry)0.650.380.52144

Sources

Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.

Follow CASRAI

Research-administration guidance, standards updates and independent tool reviews.

Ask CASRAI · full access

Ask about Occupancy Models: Principles, Mathematical Formulations, and Ecological Applications

Full access: large monthly usage, up to 10 source passages behind each answer, and MCP and API access for Claude Code, Codex and other CLIs. Cancel any time.

An AI assistant for research: funder rules, compliance, methods and publishing. It links the sources behind its answers and labels answers that come from a web search.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Works on this site and inside Claude, Cursor and the AI tools you already use.

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays open to read, with no account and no card.

Referenced across the research world

University of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logoUniversity of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logo
  • University of Cambridge logo
  • Columbia University logo
  • Crossref logo
  • University of Edinburgh logo
  • Harvard University logo
  • University of Oxford logo
  • Princeton University logo
  • Stanford School of Medicine logo
  • University College London logo
  • ORCID logo

View CASRAI adoption →