The Gold Standard of Ecology | Acoustic Indices, Soundscape Saturation and Community Inference

1. An index is a transformation of recorded sound

Back to contents

Acoustic complexity, entropy, diversity and band-ratio indices turn many time–frequency measurements into a smaller number. Acoustic Complexity Index summarises amplitude fluctuations within frequency bins; other indices represent evenness, occupied acoustic space or relative energy.

The number describes the file after choices about microphone, gain, sample rate, filtering, segment length and threshold. It does not directly count species, individuals or ecological interactions.

2. Define the ecological target before choosing the index

Back to contents

A study may target calling-species richness, occupancy of focal taxa, biological sound production, habitat condition or change in a whole soundscape. These are different estimands. One index can be useful for one and misleading for another.

Write the causal chain: community state produces sounds; propagation and weather alter them; the recorder samples them; processing creates features; an index summarises those features. Mark every source of variation that can change the index without changing the community.

3. Standardise devices, placement and processing

Back to contents

Use matched microphones or measured sensitivity corrections, consistent gain, height or depth, orientation, housing and distance from reflective surfaces. Record firmware, clock drift, sample rate, bit depth and deployment dates.

Apply one documented processing pipeline. Frequency limits, window size, noise reduction and clipping rules can change the result. Calibration tones or co-located recorders make device effects estimable instead of invisible.

4. Sampling must cover biological time

Back to contents

One minute at dawn cannot represent a 24-hour soundscape. Species partition calling by hour and season; rain, wind, insects, vessels and human activity also follow schedules. Luypaert and colleagues explicitly frame soundscape diversity across spectro-temporal space.

Stratify or model time of day, season and weather. Use the same recording duty cycle across sites, or weight observations to a common target window. A larger file count does not correct a systematically missed chorus.

5. Saturation and dominance can hide community change

Back to contents

When many signals occupy the same time–frequency cells, an acoustic-space measure may approach a ceiling. Adding a species can then change the community while barely changing the index. A loud chorus can also dominate energy and mask quieter taxa.

Constant sustained signals pose a different problem: published work notes that ACI can be insensitive to them because it rewards temporal fluctuation. A stable hum may be biologically important yet contribute less complexity than intermittent rain or snaps.

6. Worked example: the same ratio, different soundscapes

Back to contents

Use a simplified normalised difference soundscape ratio: (biological-band energy − human-band energy) divided by their sum. Site A has illustrative energies 60 and 20, so the ratio is 40/80 = 0.50. Site B has 120 and 40, so the ratio is also 80/160 = 0.50.

The ratio hides that Site B recorded twice the energy in both bands. It also cannot tell whether the biological band contains one loud chorus or many taxa. The calculation is a transparent illustration, not a biodiversity estimate; inspect spectra, annotated clips and independent community data before giving it ecological meaning.

7. Validate against the community claim

Back to contents

Select recordings across the full index range, sites, seasons and noise conditions for blinded annotation. Compare the index with focal-call occupancy, verified events, visual surveys, eDNA or other relevant observations. Cross-validation should hold out sites or time periods that represent future use.

Report discrimination, calibration and uncertainty, not only a correlation in the development sample. If a threshold is used, show false-positive and false-negative consequences. An association with habitat class is not automatically an association with biodiversity.

8. Changed-context transfer check

Back to contents

Move a terrestrial dawn-chorus index to a coral reef. Standardisation, temporal coverage and validation transfer. The frequency bands, propagation, sensor housing, dominant sound sources and biological interpretation do not.

Snapping shrimp can dominate reef recordings; vessels and water movement change background energy. A terrestrial index may separate sites for reasons unrelated to fish community structure. Revalidate the processing and target rather than carrying over the original label.

9. Limits and common misconceptions

Back to contents

A high index is not always good habitat, and a low index is not necessarily depauperate. Wind, rain, cicadas, shrimp, machinery and one abundant caller can move indices strongly. Silent organisms and species outside the recorded band remain unmeasured.

Do not choose an index because its name resembles the desired construct. Do not compare raw values from different gains or pipelines. Do not call an occupied acoustic cell a species, or infer causation from a site-level correlation without addressing habitat and detectability.

10. Audit the recording-to-inference chain

Back to contents

Listen to a stratified sample and inspect spectrograms before modelling. Plot indices by device, hour, weather and clipping rate. Recompute under plausible window, band and noise settings. Map missing files and deployment failures.

Archive raw-file identifiers, metadata, code, package versions and parameter settings. Keep the calibration model separate from the ecological model. Report whether conclusions persist after excluding rain, loud machinery or dominant choruses.

11. Frequently asked questions

Back to contents

Which acoustic index is best?

None is universally best. Choose by target, soundscape, validation evidence and robustness to the deployment conditions.

Can several indices be combined?

Yes, but prespecify or regularise the model and validate out of sample. Many correlated indices can overfit a small set of sites.

Does manual annotation solve everything?

No. Annotation supplies valuable reference evidence, but its sampling, label definitions and observer error also require audit.

12. The gold standard keeps sound and community distinct

Back to contents

Acoustic indices are useful because they compress vast recordings and reveal temporal or spatial patterns. They become credible ecological indicators only after the compression has been linked to a defined community quantity under the intended conditions.

Before interpreting an index, ask what signal changes it, what signal it ignores, where it saturates and which independent observations validate the claim. The goal is not the most complex number; it is an inference that survives listening.