The Gold Standard of Ecology | Acoustic Spatial Capture–Recapture, Call Density and Cue-Rate Uncertainty

1. Define whether the target is calls or animals

Back to contents

Write the estimand before placing recorders. Call density counts vocalisations per area and time; animal density counts individuals per area. They are linked only through a cue rate and assumptions about which animals were available to call. A survey can estimate call density well while animal density remains uncertain. Specify species, call type, time window, demographic group and spatial region. If only territorial males call, the animal-density target may be calling males rather than the whole population. Clear language prevents a numerical conversion from silently changing what was measured.

2. Synchronise the array and map every detector

Back to contents

Acoustic spatial capture–recapture depends on matching detections from the same cue across microphones and often on time-of-arrival differences. Clock drift, unknown sensor latency and coordinate errors distort localisation. Synchronise recorders, measure positions, test timing before and after deployment and record gain, orientation, height and habitat around each sensor. Calibration sounds at known locations reveal systematic delay and level differences. A dense array with poor timing can be less informative than a smaller well-calibrated one. Keep raw channels aligned so that uncertain matches can be revisited rather than frozen into a capture history.

3. Construct cue capture histories cautiously

Back to contents

For each call, record which detectors registered it. Matching is difficult when many animals call at once, echoes create duplicates or distant signals arrive out of order. Use timing, waveform shape, frequency and bearing where available, and retain match probabilities or ambiguous alternatives. Requiring detections at two sensors reduces some false positives but changes the detection process and discards genuine singletons. Manual matching is not infallible; automated classifiers introduce false positives and false negatives. The model should reflect the rule used to build histories rather than treating them as perfect identities.

4. Model detection in space and sound

Back to contents

Detection probability usually declines with source–sensor distance but also depends on source level, direction, vegetation, wind, rain, topography and recorder settings. A simple radial detection function assumes these effects are adequately represented or averaged. Variable source levels can mimic distance, and directional calls can make one side of an array seem empty. Include measured covariates when justified, examine residual patterns and use calibration or propagation experiments. The effective sampled area is estimated by the model; it is not merely the convex hull around microphones.

5. Keep localisation uncertainty inside the analysis

Back to contents

Arrival-time differences or bearings constrain a call’s location, but error can be large when sensors are nearly collinear, sound speed varies or the call is weak. A single triangulated point hides that uncertainty. Spatial capture–recapture can integrate over possible source locations while using detection histories and auxiliary data. Plot uncertainty ellipses or posterior surfaces for representative calls and check whether edge calls are pulled toward the array. Mislocalisation changes both distance and effective area, so it can bias density even when the map looks plausible.

6. Worked example: frog calls are not frog counts

Back to contents

A four-recorder wetland array estimates 240 target calls during a two-hour session and a model-based effective area of 0.30 square kilometres, giving 400 calls per square kilometre per hour. An independent focal study estimates that calling males produce 20 eligible calls per hour, suggesting 20 calling males per square kilometre. But focal rates range from 8 on cool nights to 32 on warm nights. Using 20 as exact hides a much wider density range, and females and silent males remain outside the target. The defensible report presents call density directly, propagates cue-rate uncertainty and names the population subset represented.

7. Estimate cue rate in a compatible context

Back to contents

Cue rate should match the same call definition, season, time of day, weather, disturbance and animal state as the array survey. Rates measured from captive animals or peak choruses may not transfer. Individuals differ and may call in bursts, so sample size and observation duration matter. Link focal animals, tags or close-range recordings to known individuals where possible. Model rate variation and propagate it into animal density rather than dividing by a single mean. If cue rate and density are estimated jointly, inspect identifiability: extra model complexity does not create information absent from the data.

8. Changed-context transfer check: gibbons in forest

Back to contents

Move from frogs to gibbon groups. Calls travel farther, groups may duet and terrain affects propagation. A detector array can estimate call density, but group density requires the probability and rate of groups producing the selected call during sampling. Group size is a further conversion if individual density is desired. The transferable method defines the cue, builds multi-sensor histories, models detection and propagates cue-rate uncertainty. The new context adds group coordination, long calls and strong habitat effects, so frog-specific rate assumptions cannot be carried over.

9. Know the limits and common misconceptions

Back to contents

More recordings do not automatically mean more independent animals. Repeated calls from one individual, false classifier positives and unmatched duplicates can inflate cue counts. Silent animals are not detected by an acoustic method. A high localisation precision near the centre does not guarantee accurate effective area at the boundary. Detection functions, source-level distributions and spatial density may trade off. Multiplying confidence intervals after a cue-rate division is not valid uncertainty propagation. Avoid calling an estimate population density when it is call density or the density of calling groups.

10. Audit from waveform to density estimate

Back to contents

Archive raw multichannel audio, synchronization tests, sensor coordinates, calibration files, classifier versions, thresholds and manually reviewed samples. Preserve matching rules, ambiguous histories, localisation inputs, excluded periods and weather covariates. Publish the spatial mask, detection function, source-level model, cue-rate data and code. Refit models under plausible false-positive rates, matching rules, propagation covariates and cue-rate distributions. Report how much uncertainty arises from call detection, localisation, effective area and conversion to animals.

11. Frequently asked questions

Back to contents

Is acoustic SCR the same as ordinary acoustic occupancy? No; it uses spatially linked detections to estimate cue or animal density. Must every call be localised? Not always, but auxiliary timing or bearing information can improve inference. Can a classifier be used? Yes, if false positives and missed calls are evaluated and incorporated. Why are singletons difficult? They may be real distant calls or false detections and cannot be matched across sensors. Does dividing call density by mean call rate solve everything? No; compatibility, variation and target-population coverage remain essential.

12. The gold standard preserves every conversion

Back to contents

A strong acoustic spatial capture–recapture study keeps the chain visible: waveform to species decision, detections to cue history, history to source location and effective area, cue density to animal density. Each link has uncertainty and biological assumptions. The most useful result may be a precise call-density estimate with an explicitly wider animal-density range. Honest separation makes the evidence portable and shows which next measurement—timing calibration, cue-rate study or classifier validation—would most improve the ecological claim.