The Gold Standard of Ecology | Sample Coverage, Rarefaction and Richness Comparisons

1. Separate observed richness from community richness

Back to contents

Observed richness is the number of species recorded in a sample. It is a lower bound on the species present in the assemblage because unseen species leave no direct count. The gap is largest when sampling is incomplete and many species are rare.

State the assemblage, taxonomic rules, sampling unit and time window. “The richer site” is not meaningful until the observation process and diversity order are defined.

2. Understand what sample coverage measures

Back to contents

Sample coverage estimates the fraction of the assemblage’s total incidence or abundance represented by the species already detected. High coverage means a new observation is relatively unlikely to belong to an unseen species; low coverage signals a long unseen tail.

Coverage is not the proportion of all species found. It weights by occurrence probability, so many extremely rare unseen species can coexist with high coverage.

3. Build comparable sampling data

Back to contents

Use abundance data when individuals are counted under a common protocol, or incidence-frequency data when species presence is recorded across replicated sampling units. Do not treat repeated detections of one individual as independent abundance.

Harmonise taxonomy, effort definitions, spatial grain and survey windows before rarefaction. A statistical curve cannot correct different habitat boundaries or incompatible field methods.

4. Read a rarefaction curve correctly

Back to contents

Interpolation, or rarefaction, asks how much diversity would be expected in a smaller subsample drawn from the observed data. Extrapolation projects beyond the reference sample using information about rare species. Both should show uncertainty.

A flattening curve suggests additional effort yields fewer new species, but it does not prove complete sampling. The curve’s domain and confidence interval matter as much as its shape.

5. Choose size-based or coverage-based standardisation

Back to contents

Size-based rarefaction compares communities at the same number of individuals or sampling units. This controls effort but may compare different completeness. Coverage-based methods compare at a common estimated completeness, asking what diversity is supported by equally complete samples.

Report why the standard was chosen. Coverage comparison is often fairer for richness when evenness differs, while common-size curves remain useful for showing discovery under equal effort.

6. Worked example: the even and uneven forests

Back to contents

Forest A and Forest B each yield 200 individual records. A contains 24 observed species; B contains 20. A size-only comparison seems to rank A higher.

In A, most records are concentrated in two dominant species and twelve species occur once. Estimated coverage is 0.86. B is more even and has coverage 0.96. Comparing both at 0.86 coverage shows that B would support more richness at the same completeness, despite its lower observed count at the full sample endpoint.

The report shows observed richness, common-size rarefaction and common-coverage comparison. It does not declare either forest universally more diverse; the ranking depends on whether the question concerns the collected sample, discovery under equal effort or estimated richness at equal completeness.

7. Use Hill numbers to keep diversity orders coherent

Back to contents

Hill numbers place richness, exponential Shannon diversity and inverse Simpson diversity on a common effective-number scale. Order q=0 weights all species equally and is most sensitive to unseen rare species; higher orders increasingly emphasise common species.

Plot several orders when the ecological claim involves both richness and evenness. A rank reversal across q is information about community structure, not a statistical inconvenience.

8. Changed-context transfer check: two eDNA libraries

Back to contents

Move the method to environmental-DNA libraries with different read depths. Rarefying reads to the smaller library controls sequence count, but PCR bias, marker resolution and detection thresholds also shape occurrence. Reads are not always direct organism abundance.

The transferable rule is to match the statistical unit to the observation process. Use replicated samples or incidence where appropriate, examine coverage and retain laboratory controls; do not let a deeper sequencing run masquerade as richer ecology.

9. Know the limits of extrapolation

Back to contents

Extrapolation becomes increasingly assumption-dependent beyond the observed sample, especially for q=0 when unseen rare species dominate. Confidence intervals widen and unobserved heterogeneity can invalidate the model. A numerical endpoint far beyond the data is not rescued by extra decimal places.

Do not compare confidence intervals mechanically as a universal test, pool sites with different species pools or use rarefaction to erase meaningful effort differences. If one assemblage never reaches the chosen coverage, lower the comparison target or report that the intended comparison is unsupported.

10. Audit the completeness claim

Back to contents

Archive raw counts or incidences, sampling units, taxonomy, exclusions, estimator version, diversity order, interpolation and extrapolation limits, coverage estimates and uncertainty method. Mark where each curve changes from observed interpolation to projection.

Repeat the comparison at several plausible coverage targets and after taxonomic harmonisation. Inspect singletons, doubletons and rare incidence classes because they drive unseen-species estimates.

11. Frequently asked questions

Back to contents

Does rarefaction throw away data?

Interpolation describes expected diversity at smaller effort without requiring one random subsample. The full data still inform the curve and coverage estimate.

Is higher coverage always better?

It means greater sample completeness under the model, not better field design or unbiased detection.

Can curves prove two communities are identical?

No. They quantify diversity under defined orders and sampling assumptions; composition and ecological process require additional analysis.

12. The gold standard compares information, not just counts

Back to contents

Fair richness comparison keeps observed effort, estimated completeness and diversity order distinct. Rarefaction and extrapolation are tools for a stated target, not automatic corrections.

Before ranking communities, ask whether their sampling units are compatible, whether the chosen standard gives equal information and whether the conclusion survives another reasonable coverage target. That is how a species count becomes an ecological comparison.