Environmental DNA surveys can fail to detect a species even when it occupies a site. DNA may be patchy, degraded, diluted or missed during filtration, extraction or PCR. An eDNA occupancy model separates the probability that a species is present from the probability that the survey detects it. Repeated visits, water samples and laboratory replicates provide the information needed to estimate these hidden stages. A common three-level model distinguishes site occupancy, availability of DNA in a field sample, and laboratory detection in a PCR replicate. The method improves inference, but it cannot rescue a weak design automatically. Replication must occur at meaningful levels, assumptions must be stated, contamination must be controlled, and probabilities should not be misread as proof that an organism is physically present at the exact sampling point.
Key takeaways
- Non-detection does not necessarily mean absence.
- Replication allows occupancy and detection probabilities to be estimated separately.
- eDNA studies may model site, field-sample and laboratory stages.
- False positives and contamination require explicit controls or extensions.
- Detection probability changes with season, flow, abundance and assay performance.
- The model’s spatial and temporal definition of a site must match the question.
What occupancy means
Occupancy is the probability that a defined site is occupied during a defined period. The ecological state is hidden because observations are imperfect.
A site might be a pond, stream reach or grid cell. Its definition determines interpretation; DNA transported downstream can complicate the connection between detection and local residence.
Why eDNA is missed
Organisms shed different amounts of DNA, and particles are unevenly distributed. Sunlight, microbes, temperature and chemistry degrade molecules. Filters capture only sampled water.
Extraction can lose DNA and inhibitors can prevent amplification. The DNA guide explains the molecule, while occupancy modelling focuses on the observation process.
The detection-history idea
Repeated surveys produce histories such as 101, meaning detection, non-detection, detection. Across many sites, the pattern reveals how often an occupied site is missed.
A single visit cannot usually distinguish absence from failed detection without external information. Replication creates the statistical leverage.
Three levels of replication
Site visits repeat environmental conditions; multiple field samples test whether DNA enters a bottle or filter; PCR replicates test laboratory detection. A hierarchical model can assign probability to each stage.
Replication at one level does not replace another. Ten PCR wells from one bottle say little about whether the bottle represented the whole stream.
Covariates
Occupancy may vary with habitat, while detection varies with discharge, temperature, season, sample volume or inhibition. Including justified covariates explains systematic differences.
Too many predictors with few detections create unstable estimates. Study design should anticipate the main sources of variation rather than search indiscriminately after results are known.
False positives
Contamination, assay cross-reaction or laboratory error can produce detection when the target species is absent. Field blanks, extraction blanks, no-template controls and positive controls reveal different failures.
Some models estimate false-positive probability, but they need validation data or repeated confirmation. A rare, surprising detection warrants particularly careful review.
Sampling design
Sites should represent the target landscape, not merely easy access points. The sampling-bias guide explains why convenience sampling limits generalisation.
Pilot data can help allocate effort among more sites, more visits, more bottles and more PCR replicates. The best balance depends on which stage contributes most non-detection.
Interpreting abundance
Detection probability may rise with abundance, but eDNA concentration also depends on shedding, transport, decay and water volume. Occupancy models generally estimate occurrence, not numbers of organisms.
Abundance relationships require separate calibration and strong assumptions. A higher number of positive replicates is not automatically a population count.
Freshwater applications
Surveys can screen for invasive species, rare fish or amphibians and community change. The freshwater-conservation guide places detection within wider habitat monitoring.
Management decisions may use thresholds or repeat confirmation. A model provides probabilities; agencies still need decision rules that weigh the costs of missed and false detections.
Model checking and transparency
Researchers examine goodness of fit, sensitivity to priors, parameter identifiability and predictive performance. Sparse detections can make occupancy and detection difficult to separate.
The peer-review guide describes one layer of scrutiny, but sharing protocols, detection histories and code makes the analysis more reproducible.
What the result can say
An estimated occupancy probability describes a modelled ecological state under stated definitions and assumptions. It should be reported with uncertainty intervals and detection estimates.
It does not prove breeding, population viability or precise location. Complementary capture, visual, acoustic or habitat evidence may be needed for those claims.
Assumptions behind the basic model
A basic occupancy model assumes the site’s state does not change during the survey season, detections are independent conditional on occupancy, and species identification is correct. Violations can bias estimates. Rapid movement or DNA transport may make closure especially difficult.
Researchers can shorten the survey window, redefine occupancy as use rather than permanent residence, or fit dynamic models across seasons. The interpretation must follow the assumption chosen rather than the everyday meaning of occupied.
Dynamic occupancy
Repeated seasons allow estimation of colonisation and extinction probabilities in addition to detection. This can track range expansion, restoration or invasion while allowing some surveys to miss the target.
For eDNA, laboratory methods and assay sensitivity must remain comparable through time. A change in filtration, primer or threshold can look like ecological change unless modelled or calibrated with overlapping samples.
Decision thresholds and management
Managers may act when posterior occupancy exceeds a chosen probability or when any confirmed detection occurs. The threshold depends on consequences: missing an invasive species can be costly, while a false closure can also harm livelihoods.
Decision analysis combines ecological probabilities with costs, benefits and values. It makes clear that the model estimates evidence, while policy determines how much evidence is sufficient for an action. Follow-up sampling can be targeted where it has the greatest expected value.
A replicated pond survey example
Researchers survey fifty ponds for a rare amphibian. At each pond they make three visits, collect two water samples per visit and run four PCR replicates per extract. A three-level model estimates pond use, the chance DNA enters a field sample when the species uses the pond, and the chance PCR detects target DNA when present in the extract. Water temperature and vegetation are occupancy covariates; turbidity and inhibition are detection covariates.
A pond with one positive PCR among twenty-four wells is examined against blanks and assay controls rather than automatically labelled occupied. Another pond with repeated negatives may still retain moderate occupancy probability if its samples were highly inhibited. The output preserves these differences instead of converting every pond to a confident yes or no.
Design efficiency and simulation
Before fieldwork, simulation can test how well proposed replication recovers realistic occupancy and detection probabilities. When field detection is poor, visiting more sites once may add less information than collecting repeated samples. When occupancy is rare, broader coverage may be essential.
Cost also differs among travel, filtration, extraction and PCR. Optimising design means balancing statistical precision with logistics and contamination risk. Pilot studies can estimate stage-specific failure, after which effort is moved to the level where additional replication most improves inference.
Conclusion
eDNA occupancy models turn repeated detections and non-detections into a more realistic estimate of species occurrence by acknowledging that field and laboratory methods miss DNA. Their strength comes from design: clearly defined sites, replication at the correct levels, controls for contamination and measured covariates. Used carefully, they improve surveillance and prevent a single negative sample from being treated as absence. Used carelessly, they can attach sophisticated probabilities to ambiguous sampling.
Frequently asked questions
Why are repeated PCR tests needed?
They help estimate laboratory detection from DNA that reached the extract, but do not measure field sampling by themselves.
Does a positive eDNA result prove a living organism is nearby?
Not always. DNA can be transported or persist, so ecology and hydrology affect interpretation.
What is imperfect detection?
It is the chance of missing a species or its DNA even though the target state is present.
Can occupancy models estimate abundance?
Not directly in their basic form; abundance requires additional information and assumptions.
How are false positives controlled?
With validated assays, clean workflows, multiple blanks, confirmation rules and sometimes explicit statistical models.
