Your counts
Result
Every site
Coverage: the share of individuals belonging to species already found. Chao1 and ACE are lower-bound estimates of richness; the Shannon and Simpson columns are estimated for the whole assemblage, in effective numbers of species.
By sample size
By sample coverage
Solid: rarefaction (smaller samples drawn from yours). Dot: your sample. Dashed: extrapolation. Shaded: bootstrap interval. Richness should not be extrapolated far past twice the sample.
Compared at a common coverage
Compared at a common sample size
More sampling
The example: ten years of desert rodents at Portal, Arizona
The Portal Project has trapped rodents on 24 fenced plots in the Chihuahuan Desert near Portal, Arizona, every month since 1977, and publishes every capture under CC0. Since 2015 ten plots are open controls, eight have gates too small for kangaroo rats, and six try to keep all rodents out (some still get in). The bundled file counts the distinct tagged animals first caught on each plot from 2016 to 2025: 6,094 rodents of 21 species.
- Richness is saturated; evenness is not. Pooled by treatment, every group has 20 or 21 species at coverage above 99.9%, and Chao1 finds almost nothing left to find. The difference is in how the individuals are spread: at the common coverage of 99.97%, Simpson diversity is 3.51 effective species on the controls (95% interval 3.37 to 3.65) and 2.95 where kangaroo rats are kept out (2.75 to 3.15), where one pocket mouse, Chaetodipus penicillatus, is 1,044 of the 1,900 animals.
- The smallest sample is not the poorest plot. Plot 10 caught only 105 animals and 9 species, the fewest of all 24. Compared at the common coverage of 98.1%, it is ninth of 24, with 11.9 species and a wide interval (7.1 to 16.7).
- Chao1 can swing on one doubleton. Plot 4 found 15 species, six of them only once and one twice, so Chao1 says 32.9. Read the interval, and the ACE estimate beside it, before the point.
These are descriptive readings of the public census, pooled over plots and years; they are not a re-analysis of the experiment's published conclusions. Counting tagged individuals once over ten years is one choice of unit (eval/build-example.py); captures per survey would give different numbers.
Method and citation
From the counts Xi of the S species in a sample of n individuals the page takes the number of species seen once (f1) and twice (f2). Diversity is measured by Hill numbers qD = (Σ piq)1/(1−q) (Hill 1973) at q = 0, 1 and 2. For the whole assemblage: Chao1, S + ((n−1)/n) f1²/(2f2) (Chao 1984), with Chao's (1987) variance and log-normal interval; ACE (Chao and Lee 1992); the Chao-Wang-Jost entropy estimator (2013) for q = 1; the unbiased Σ Xi(Xi−1)/(n(n−1)) for q = 2. Rarefaction to a size m < n is the expected diversity of a random subsample, computed exactly from hypergeometric probabilities (Hurlbert 1971; Chao et al. 2014); extrapolation to m > n uses Chao et al. (2014) for q = 0 and 2, and for q = 1 the rule of the iNEXT package (Hsieh, Ma and Chao 2016): a geometric approach to the estimated asymptote at the rate the rarefaction curve ends with. Sample coverage and its curve follow Chao and Jost (2012), refining Good (1953) and Turing's 1 − f1/n. Sites are compared at the lowest coverage, and the lowest size, any of them reaches at twice its sample size, as iNEXT does. Intervals come from a seeded bootstrap of an estimated assemblage that includes the undetected species (Chao et al. 2014).
Random, independent individuals. Each individual is assumed drawn at random from the assemblage, as in a well-mixed sample; clumped species, or counts of the same animal twice, break this. Use one row per distinct individual.
Not here (yet): incidence data (presence in sampling units: Chao2, ICE), phylogenetic and functional diversity, beta diversity between sites, and orders of q other than 0, 1 and 2.
Checked against: an independent Python reference (scipy 1.17.1 hypergeometric probabilities, mpmath 1.4.1, scikit-bio 0.7.4) on 13 assemblages, simulated and real, to 1e-10 or better; 20,000 random subsamples; and a coverage simulation of 1,000 samples in each of four scenarios (README, "Verification").
@article{hill1973diversity,
author = {Hill, M. O.},
title = {Diversity and evenness: a unifying notation and its consequences},
journal = {Ecology}, volume = {54}, number = {2}, pages = {427--432}, year = {1973},
doi = {10.2307/1934352}}
@article{chao2014rarefaction,
author = {Chao, Anne and Gotelli, Nicholas J. and Hsieh, T. C. and Sander, Elizabeth L. and Ma, K. H. and Colwell, Robert K. and Ellison, Aaron M.},
title = {Rarefaction and extrapolation with {Hill} numbers: a framework for sampling and estimation in species diversity studies},
journal = {Ecological Monographs}, volume = {84}, number = {1}, pages = {45--67}, year = {2014},
doi = {10.1890/13-0133.1}}
@article{chao2012coverage,
author = {Chao, Anne and Jost, Lou},
title = {Coverage-based rarefaction and extrapolation: standardizing samples by completeness rather than size},
journal = {Ecology}, volume = {93}, number = {12}, pages = {2533--2547}, year = {2012},
doi = {10.1890/11-1952.1}}
@article{chao1984nonparametric,
author = {Chao, Anne},
title = {Nonparametric estimation of the number of classes in a population},
journal = {Scandinavian Journal of Statistics}, volume = {11}, number = {4}, pages = {265--270}, year = {1984}}
@article{chao1987unequal,
author = {Chao, Anne},
title = {Estimating the population size for capture-recapture data with unequal catchability},
journal = {Biometrics}, volume = {43}, number = {4}, pages = {783--791}, year = {1987},
doi = {10.2307/2531532}}
@article{chao1992ace,
author = {Chao, Anne and Lee, Shen-Ming},
title = {Estimating the number of classes via sample coverage},
journal = {Journal of the American Statistical Association}, volume = {87}, number = {417}, pages = {210--217}, year = {1992},
doi = {10.1080/01621459.1992.10475194}}
@article{chao2013entropy,
author = {Chao, Anne and Wang, Y. T. and Jost, Lou},
title = {Entropy and the species accumulation curve: a novel entropy estimator via discovery rates of new species},
journal = {Methods in Ecology and Evolution}, volume = {4}, number = {11}, pages = {1091--1100}, year = {2013},
doi = {10.1111/2041-210X.12108}}
@article{good1953population,
author = {Good, I. J.},
title = {The population frequencies of species and the estimation of population parameters},
journal = {Biometrika}, volume = {40}, number = {3-4}, pages = {237--264}, year = {1953},
doi = {10.1093/biomet/40.3-4.237}}
@article{hurlbert1971nonconcept,
author = {Hurlbert, Stuart H.},
title = {The nonconcept of species diversity: a critique and alternative parameters},
journal = {Ecology}, volume = {52}, number = {4}, pages = {577--586}, year = {1971},
doi = {10.2307/1934145}}
@article{hsieh2016inext,
author = {Hsieh, T. C. and Ma, K. H. and Chao, Anne},
title = {{iNEXT}: an {R} package for rarefaction and extrapolation of species diversity ({Hill} numbers)},
journal = {Methods in Ecology and Evolution}, volume = {7}, number = {12}, pages = {1451--1456}, year = {2016},
doi = {10.1111/2041-210X.12613}}
@misc{portaldata,
author = {{Weecology}},
title = {Portal Project data: long-term monitoring of a {Chihuahuan Desert} ecosystem},
howpublished = {\url{https://github.com/weecology/PortalData}}, note = {CC0, retrieved 2026-11-18}}
Questions
Why not just compare the number of species each site found?
Because the count rises with effort. A site where you caught 100 animals will show fewer species than one where you caught 300, even if the two are identical. Rarefaction cuts the bigger samples down to a common size; Chao and Jost (2012) showed it is fairer still to compare at a common sample coverage, the share of individuals belonging to species already found, because a rich site needs more individuals to be sampled equally well.
What are Hill numbers?
Effective numbers of species (Hill 1973). Order q = 0 is species richness and counts every species equally; q = 1 is the exponential of Shannon entropy and counts species in proportion to their abundance; q = 2 is inverse Simpson and counts mainly the dominant ones. All three are in units of species, so a community of 10 equally common species scores 10 at every order, and the gap between orders says how uneven the community is.
How far can I extrapolate?
Chao et al. (2014) recommend extrapolating richness to no more than about twice the sample size: beyond that the prediction leans on Chao1, which is a lower bound and can fall well short in uneven communities. The q = 1 and q = 2 curves can be pushed further because they depend less on rare species. The page draws extrapolation dashed and compares sites at most at twice the smallest sample.
Is Chao1 the number of species there really are?
No: it is a lower bound. It uses only singletons and doubletons, and it is accurate when the rare species are all about equally rare. In uneven communities sampled lightly it falls short; in the page's own simulation with 120 lognormal species and 200 individuals it averaged 83, and its 95% interval held the true 120 only half the time. Read it as at least this many.
Can I use this for things that are not animals or plants?
Yes, for any count of how many items fall into each class where unseen classes matter: OTUs or amplicon variants in a microbiome sample (where singletons may be sequencing errors, so read Chao1 with care), words in a text, distinct bugs found, customers by product, coins by die. The assumption is that items were drawn at random and independently.
What is sample coverage?
The share of the assemblage's individuals that belong to species already in your sample, estimated from the singletons and doubletons (Good 1953; Chao and Jost 2012). At 99% coverage, the next individual you catch has about a 1% chance of being a species you have not seen. The page also says how many more individuals would take coverage to 95% and 99%.
How well do the intervals work?
In 1,000 simulated samples per scenario, the 95% intervals for rarefied richness held the truth 92% to 98% of the time, for richness extrapolated to twice the sample 87% to 99%, and for Shannon and Simpson diversity 85% to 96%, lowest for the smallest sample of the most uneven community. Chao1's interval held the true richness only 50% to 93% of the time: 50% when a 120-species community was sampled with 200 individuals, because Chao1 is a lower bound, and 55% when every species had already been found, because its interval sits above the species seen. The table is in eval/coverage-output.txt.
Is my data uploaded anywhere?
No. Every number is computed by JavaScript in this browser tab. There are no accounts, cookies or analytics vendors.