Soil microbial alpha diversity is a way of describing the diversity of detected microbial features within a sample. If you need to interpret a Shannon index, observed ASV richness, Simpson diversity, Pielou evenness, Faith’s phylogenetic diversity or an alpha-rarefaction curve, first identify the feature table and the metric’s mathematical definition. Different diversity measures answer different questions, even when they appear together in one laboratory report.
A higher alpha-diversity score does not automatically mean that soil is healthier, more fertile or better managed. The calculation depends on which sequences were recovered, the sequencing and filtering pipeline, the relative distribution of features and the measurement’s sensitivity to rare taxa. The scores describe a particular sampled feature population, not an exhaustive inventory of living soil organisms or a direct process rate.
This guide explains how to read soil microbial alpha diversity results from first principles, including richness versus evenness, Shannon and Simpson formulas, effective numbers of features, sampling-depth effects, rarefaction, uncertainty and practical report-writing. Worked numbers are teaching examples, not claimed results from a particular soil experiment. The aim is to help students and practitioners make defensible comparisons without exaggerating what one index can reveal.
The quickest useful answer
Alpha diversity describes variation among detected features within one sample, or a clearly defined biological unit. Richness asks how many distinct features are observed. Evenness asks how uniformly counts are distributed among those features. Shannon and some Simpson measures combine aspects of both, with different sensitivity to the common and rare portions of the feature table. The terms are related but not interchangeable.
For a responsible summary, name the precise metric and the unit being sampled: ‘The treatment samples had a higher observed ASV richness at the reported sequencing depth’ is better than ‘The treatment had more species.’ The observation concerns detected marker-sequence variants, not guaranteed species counts. Where a laboratory reports more than one index, explain what each adds rather than collapsing the values into a single vague claim of biodiversity.
Start with the biological unit, not the equation
A biological unit might be one soil core, a plot composite, a root-adhering sample or a defined depth horizon. This identity determines what the index means. Three separately sequenced aliquots from one homogenised core can provide useful technical information, but they do not automatically represent three independent field replicates. Comparing a shallow surface sample with a deeper mineral sample may be ecologically meaningful only when their contexts are explicitly reported.
Read the sample metadata before comparing scores. Soil moisture, season, plant cover, organic matter, management, extraction method and collection time can all shape the recovered community. An index without a sampling unit is a number without a convincing ecological referent. The question ‘Which field is more diverse?’ is ambiguous until we specify biological group, marker, depth, sampling design and diversity definition.
What counts as a feature?
In many current microbiome workflows, a feature is an amplicon sequence variant, or ASV, inferred from marker-gene reads after sequence quality control and denoising. Other studies use operational taxonomic units, species calls, contigs or functional gene categories. Richness is calculated from whichever feature definition is in the table. Changing that definition can change the index even if the original soil has not changed.
An ASV is not necessarily a unique microbial species. Several strains or genomes may share the same marker, and one organism may contribute more than one marker variant. Read the soil amplicon sequencing guide before treating observed ASVs as organisms. The calculation may be correct for the table while a species-level interpretation would be too strong.
Richness is the count of observed categories
Observed richness is simply the number of different features represented in the table for a given sample under the analysis rules. If sample A contains nonzero counts for eight ASVs and sample B contains nonzero counts for twelve, observed richness is eight and twelve respectively. This is an exact count of categories in those processed tables, not an exact count of species inhabiting the original physical samples.
The number of observed features usually depends on sequencing depth, read-quality filtering and rarity of the community. A sample with more retained reads has more opportunities to encounter rare sequences. Therefore, comparing raw observed richness across libraries with radically different depths can be misleading. Ask whether depth has been harmonised or explicitly modelled, and whether sensitivity analysis shows the result is stable to reasonable choices.
Evenness asks a different question
Two samples can contain the same number of features but differ greatly in their distributions. Suppose each contains four ASVs. In one, reads are divided equally among all four. In another, one ASV has 70% of reads while each of the remaining three has 10%. Both have observed richness four, but the first has higher evenness because its feature distribution is less dominated by a single category.
Neither state is inherently good or bad in every ecological setting. A specialised environment may support a community strongly dominated by organisms adapted to that niche. An equal distribution could arise under other conditions without establishing superior nutrient cycling. Evenness is a quantitative description of read allocation across detected categories. It should be interpreted alongside habitat, depth, seasonal pattern and research question.
What the Shannon index calculates
Shannon entropy is commonly calculated as negative the sum, across features, of each proportion multiplied by its natural logarithm: H equals minus the sum of p times ln(p). Here p is the feature’s proportion within the defined sample table. The calculation increases when more features are represented and when their proportions are more evenly distributed, although exactly how it changes depends on the full distribution.
When a category has a proportion of zero, its contribution is conventionally treated as zero by the limiting value of p ln p. The logarithm base should be specified because it changes the numerical scale. Most microbiome implementations use natural logarithms. Shannon H is an index on a mathematical scale, not a percentage of organisms or a measurement per gram of soil. A value of two is not literally twice the species count of a value of one.
Worked calculation: balanced and unbalanced communities
Imagine two illustrative samples with two detected features each. Sample A is divided 50% and 50%; Shannon H using natural logs is about 0.693. Sample B is divided 90% and 10%; H is about 0.325. Richness is two in both samples, yet Shannon is higher for the more even sample. The calculation reflects a change in distribution without requiring additional features.
Another way to explain the result is to ask how unpredictable the identity of a randomly selected read is within the table. Balanced categories are less predictable, so the entropy is higher. The word ‘read’ matters. The calculation does not prove that a randomly selected living cell follows the same proportions, because recovery, primer amplification and sequencing can differ among organisms. Preserve the evidence level in the accompanying sentence.
Effective numbers make some comparisons clearer
The exponential of Shannon entropy, exp(H), is often called a Hill diversity of order one or the effective number of equally common features. For the 50–50 example, exp(0.693) is approximately two. For the 90–10 example, exp(0.325) is approximately 1.38. This translation converts an abstract entropy value into a more interpretable comparison of effective diversity under a precise mathematical definition.
Effective diversity is still calculated from the observed feature proportions; it does not restore undetected organisms or correct all laboratory biases. Use it to communicate the meaning of an index, not as a claim that a fractional number of biological species lives in the soil. If a report uses an effective-number interpretation, identify the underlying order and metric. Different orders weight rare and dominant features differently, so ‘effective species’ without the order can hide essential context.
Simpson is a family of related measures
The term ‘Simpson index’ is ambiguous unless its formula is given. Simpson concentration can be written as the sum of squared feature proportions. A high concentration means reads are concentrated in relatively few features. The Gini–Simpson form, one minus that sum, moves in the opposite direction and increases as diversity increases. Inverse Simpson, one divided by the sum, is another common expression.
For two equally represented features, concentration is 0.5, Gini–Simpson is 0.5 and inverse Simpson is two. For a 90–10 split, concentration is 0.82, Gini–Simpson is 0.18 and inverse Simpson is approximately 1.22. A paper stating ‘Simpson increased’ without naming its convention cannot be reliably interpreted. Always retrieve the formula or software implementation before translating the direction of a result.
Why metric direction can reverse a headline
Suppose one paper reports a Simpson value of 0.82 and another reports 0.18 for the same underlying distribution. The numbers may not conflict at all: one could be concentration, the other Gini–Simpson. A phrase such as ‘higher Simpson diversity’ is especially hazardous when the author has not indicated which version they use. The mathematical convention must come before the ecological interpretation.
Write the metric name as displayed by the software, the formula and whether larger values indicate greater diversity or greater dominance. This is an excellent exercise in accurate technical language because the same familiar label can hide two opposite meanings. It also demonstrates why copying a sentence from another paper without checking its definition is an unreliable way to explain a graph.
Pielou evenness and its denominator
Pielou’s evenness is often calculated as Shannon H divided by the natural logarithm of observed richness, provided the richness is greater than one. It rescales a form of Shannon diversity relative to the maximum possible H for that number of observed features. Values near one indicate relatively even distributions. At richness one, the direct formula involves a zero denominator, so software may treat the case specially.
Do not use Pielou as an independent census of species or as an ecosystem performance score. It inherits limitations from the feature table, richness estimate and Shannon calculation. Comparisons across very different feature-detection conditions may remain biased even after rescaling. A readable caption says, for example, ‘Pielou evenness of detected ASVs after specified filtering,’ rather than making the value sound like a universal property of the soil.
Shannon and Simpson care differently about rare features
Both indices use proportions, but Simpson-type concentration places comparatively greater emphasis on dominant categories because proportions are squared. Shannon responds to the whole distribution with a different weighting of rare and abundant features. This can produce different rankings across communities. Such disagreement does not mean the statistical system has failed; the measures are answering distinct mathematical questions.
A useful report compares metrics according to the research need. If the ecological concern is the balance among dominant lineages, a Simpson-family measure may be informative. If the aim is a broader distribution-sensitive description, Shannon may add context. Observed richness is useful for the count of detected variants. None by itself answers whether beneficial functions are active or whether the sampled system is resilient to disturbance.
Rarefaction addresses uneven sampling effort
Rarefaction is a procedure for estimating or comparing observed diversity after subsampling feature counts to a common number of reads. In a typical read-subsampling rarefaction, libraries with adequate reads are randomly reduced to a selected depth. This can make certain comparisons more directly matched for sequencing effort but discards information, and samples below the threshold may be excluded entirely.
A chosen rarefaction depth is therefore an analytical decision with trade-offs. A high threshold can discard many lower-depth samples; a low threshold retains more samples but sacrifices information from deeply sequenced libraries. Read how the threshold was selected, whether rarefaction was repeated and how conclusions respond to alternative reasonable depths. No single universal depth is appropriate for every soil dataset.
How to read an alpha-rarefaction curve
An alpha-rarefaction plot typically places sampling or read depth on the horizontal axis and estimated or observed within-sample diversity on the vertical axis. The curve often rises quickly as common features are encountered and rises more slowly as additional sequencing reveals rarer features. Different curves may represent samples or groups, with replication and uncertainty displayed depending on the software.
A curve approaching a plateau suggests that additional reads yield relatively fewer new observed features under that sampling and pipeline. It does not prove that every microbial organism or gene in the original soil has been discovered. A curve still climbing steeply at the maximum depth signals limited sampling of the observed feature pool. Interpret the shape alongside exclusions and confidence estimates rather than treating a smooth graph as a certificate of completeness.
Rarefaction is not the same as extrapolation
Rarefaction studies subsampling within the observed data, while extrapolation predicts what might be observed with greater sampling effort using a model and assumptions. Mixing the terms can make a chart sound more empirically grounded than it is. If a curve extends beyond actual sequencing depth, verify whether that extension is a model prediction and what uncertainty accompanies it.
The distinction matters when explaining biodiversity to a wider audience. A model can help estimate unseen richness, but it does not directly count organisms never sequenced. Its reliability depends on the observed frequency distribution, sampling design and estimator assumptions. Use explicit verbs: ‘observed under subsampling’ for rarefaction and ‘estimated under extrapolation’ for modelled additions. That vocabulary preserves the strength of evidence.
Richness estimators try to infer unseen categories
Some analyses report Chao1 or related richness estimators alongside observed ASVs. These methods use information about low-frequency features to estimate how many categories may have been missed. They are sensitive to the structure of the rare portion of the table and to errors, filtering and sequencing artefacts. Such values are estimates under assumptions, not direct measurements of hidden soil species.
Read the exact estimator version and its suitability for the data, including whether the input is count-based and how singletons or doubletons have been treated. A pipeline that removes low-count features can alter the evidence these estimators depend upon. Do not compare a Chao1 value with an observed richness number as though they share the same measurement status. An honest report names which is observed and which is inferred.
A zero count does not prove a species absent
Low-abundance features may be missed because a finite library samples only part of a much larger DNA pool. Sampling, extraction, marker specificity, read filtering and denoising all affect detectability. A zero in a feature table therefore usually means that no reads of that feature were retained for the sample under the analysis, not that the organism could not have been present in the original soil.
This matters for richness because every undetected feature affects the apparent category count. Adding more sequencing may reveal new ASVs, while a different primer set may reveal a different segment of the community. Define the biological scope of the index instead of claiming ‘total microbial biodiversity.’ In environmental communication, the phrase ‘detected marker features’ is often more scientifically faithful than ‘all species present.’
Good’s coverage has a specific technical interpretation
Good’s coverage is a sample-coverage estimator often connected to the proportion of sequence observations belonging to already represented categories. Its exact calculation and treatment of low-frequency features depend on the chosen implementation. It can help assess sampling completeness for the observed feature distribution, but a high coverage estimate does not mean every rare species in the original soil has been detected.
If an author uses a coverage percentage, ask for the denominator and estimator. An apparently excellent value can coexist with many unobserved rare organisms. Coverage is a statistical property of sampling a feature distribution, not a guarantee that the lab extraction recovered every cell. Readers should preserve this distinction, particularly where an ecological conclusion rests on small differences in richness between treatments.
Faith’s phylogenetic diversity adds evolutionary structure
Faith’s phylogenetic diversity sums branch lengths in a phylogenetic tree connecting the observed features under a specified construction. It therefore asks a different question from the raw number of ASVs. Two samples with equal richness may span different amounts of evolutionary divergence if one contains closely related sequences and the other contains lineages separated by longer branches.
Tree construction, alignment, rooting, marker suitability and branch-length accuracy affect the score. A fungal ITS dataset may not automatically support the same phylogenetic treatment or assumptions as a bacterial 16S analysis. If a report includes Faith PD, inspect which feature tree was used. Do not translate a higher PD into a proven increase in ecological services, because the score expresses tree-based sequence diversity, not measured ecosystem function.
Depth and library filtering can change the ranking
Suppose one sample retains 60,000 quality-controlled reads and another retains only 8,000. More features may be detected in the deeper library even if the underlying communities have comparable biological richness. A pipeline’s minimum-count and prevalence filters can also remove rare ASVs differently across samples. The resulting observed-richness ranking may be sensitive to processing choices.
Before concluding that a treatment reduced richness, ask whether the same marker, software settings and read-sampling method were applied. Check whether one group lost disproportionately more samples at the rarefaction depth. A robust finding should ideally survive sensible analytical sensitivity checks and be visible across independent biological replicates. If it does not, the uncertainty is worth stating prominently rather than burying in a supplement.
What is the actual statistical unit?
If a study collects five cores from one plot and pools them into one DNA extract, the plot composite is the biological observation for some analyses. Running the same library through a sequencer several times may improve technical reliability but does not generate additional independent plots. Statistical tests that treat technical repeats as independent can underestimate uncertainty and make small differences appear more certain.
Read the design: nested plots, repeated dates, paired sampling sites and regional clusters require thought. A change in alpha diversity across samples should be analysed at the appropriate experimental unit, with pairing or random effects where warranted. This is especially important when comparing soil management systems because fields, depths and seasons can introduce structured variation far greater than the nominal treatment effect.
Statistical significance and ecological significance differ
A difference in mean Shannon diversity may achieve a low p value in a well-powered dataset while having a small practical effect. Conversely, an ecologically interesting pattern may be imprecisely estimated with few biological replicates. Statistical significance is a description about a test under assumptions, not a general statement that the result is important for soil productivity or resilience.
Ask for effect sizes, confidence intervals, replicate distributions and the intended ecological interpretation. If many metrics and comparisons are tested, multiple-testing considerations apply. A good figure showing group distributions is more informative than a naked star above two bars. Interpret both the estimated difference and the uncertainty, then connect the finding to independently measured ecological outcomes if the application requires that bridge.
A worked example with four features
Consider two illustrative samples of four ASVs each. In soil A the proportions are 25%, 25%, 25% and 25%. In soil B they are 70%, 10%, 10% and 10%. Both have observed richness four. Shannon entropy is approximately 1.386 for A and 0.940 for B using natural logarithms. The difference arises from evenness, not from a difference in the number of observed features.
For A, the effective number exp(H) is four; for B it is about 2.56. These values illustrate how dominant sequences reduce the effective diversity of the observed distribution. They do not establish that B is damaged soil. A dominant lineage could be expected in a particular environment or experimental condition. A scientifically meaningful next step would ask how the sampling and habitat explain the distribution and whether process measurements differ.
Do not confuse alpha and beta diversity
Alpha diversity concerns what is detected within samples. Beta diversity describes differences in entire feature compositions among samples. Two soils can have the same Shannon score but completely different ASVs. Conversely, the same set of features can occur in both soils at changed proportions, producing different alpha scores but only some kinds of compositional difference.
The mathematical questions are distinct. If a paper claims ‘no change in microbial diversity’ based only on one alpha index, ask whether community composition may still have shifted. Diversity is not one number that answers every biological question. A careful report names alpha metrics when discussing within-sample distribution and uses appropriate distance-based analysis when discussing between-sample differences.
Soil function cannot be read directly from Shannon
Soil organisms interact with substrate supply, water, temperature, oxygen and each other. High alpha diversity may coexist with limited current metabolic activity, while a less diverse community may perform a specialised function efficiently. Richness can sometimes support hypotheses about redundancy or resilience, but these relationships depend on which functions are measured and the environmental context.
Pair diversity analysis with complementary measures when making claims about processes. Soil respiration reports gas flux under its measurement conditions; soil enzyme assays report defined catalytic signals. These are not substitutes for diversity indices, nor do the indices substitute for them. Their combination can clarify which hypotheses warrant further testing.
Rarefaction can exclude the most informative difficult samples
A stringent rarefaction threshold sometimes removes samples with low read depth. If those samples cluster in one ecological group, the retained dataset may no longer represent the full experimental comparison. The final diversity figure can look cleaner because difficult cases disappeared. An honest methods section reports exclusion counts, reasons and the sample attributes associated with excluded libraries.
When readers are shown only the retained samples, ask whether the initial design was balanced and whether the remaining observations support the original claim. Sensitivity analyses using alternative depths or methods can help show whether a conclusion depends on a particular threshold. Avoid assuming that more data processing automatically increases ecological truth. Careful filtering can improve measurement quality while also narrowing the question that remains answerable.
Fungal and bacterial alpha scores are not automatically comparable
A bacterial 16S ASV richness of 100 and a fungal ITS richness of 80 do not establish that the bacterial community is biologically more diverse. The markers sample different organism groups, have different copy-number patterns, variable sequence behaviour, primer coverage and reference resolution. The ASV feature definitions are not a universal unit shared between those assays.
Compare diversity within a well-defined assay and analysis when possible. If broader cross-domain comparison is required, explain why the chosen metric and sampling design are appropriate, or avoid direct ranking. Do not convert numbers of detected marker variants into a common species currency merely by placing them on one chart. The apparent precision of integers should not hide the biological differences between measurement systems.
Time-series diversity has its own questions
Seasonal soil moisture, plant growth, freezing, drying and disturbance can alter detected communities. A diversity index from one date is a snapshot, not a permanent property of a field. When the same sites are sampled repeatedly, observations are correlated. A plot that rises in spring and falls in summer may reflect time, sampling changes, treatment effects or all three.
Use a time-aware design and identify whether the graph shows individual trajectories, group means or pooled sample estimates. Match the sample unit across dates and report any changes in laboratory processing. Stronger analyses distinguish within-site temporal change from differences among sites. A time-series caption should avoid turning one seasonal cycle into a universal trend for all soils.
Translate index names without losing their meaning
Words like richness, evenness, diversity and coverage sound familiar, but they are technical terms with specific definitions. ‘Richness’ in an ASV report is not a claim about agricultural wealth. ‘Coverage’ is not necessarily geographic coverage. ‘Significant’ can refer to a statistical criterion rather than ecological importance. Confusing these senses can distort an otherwise correct translation into ordinary language.
A useful glossary should explain each index in one sentence, name the unit and state what it cannot tell the reader. Cross-linguistic transfer helps explain why similar-looking terms can mislead across languages. The guide to translation failure through lost context and meaning reinforces the need to preserve distinctions, not merely substitute words.
A reader-first workflow for checking a diversity figure
Begin by identifying marker and feature type. Check biological sampling unit and number of independent replicates. Read the sequencing depth and exclusion summary. Find the exact metric definition and its direction. If the chart includes rarefaction, determine whether it displays observed subsampling or extrapolation. Inspect the variation among replicates, then look for independent evidence supporting any ecological claim.
Write your conclusion in two sentences: one for the measured pattern and another for the limitation. For example, ‘Observed ASV richness was higher under the reported depth-matching procedure in the sampled treatment libraries. That result does not establish higher absolute cell counts or improved ecosystem function.’ This style is concise and durable because it separates what the figure shows from what a wider claim would require.
Common mistakes and their repairs
Mistake one is calling observed ASVs species. Repair it by naming the feature definition. Mistake two is calling the highest index value the healthiest soil. Repair it by identifying the ecological outcome of interest. Mistake three is comparing unbalanced sequencing depths without checking processing. Repair it by reviewing read retention, rarefaction and sensitivity analyses. Mistake four is describing a Simpson index without its formula. Repair it by stating direction and convention.
Mistake five is treating technical reads as independent biological replicates. Repair it by recognising plots and samples as the proper experimental units. Mistake six is treating an apparent rarefaction plateau as a complete organism census. Repair it by defining the sampled feature pool. Mistake seven is calling a significant alpha result a causal proof. Repair it by checking design, effect size, potential confounding and functional validation.
Frequently asked questions
Does Shannon equal the number of species? No; it is an entropy-based index of the observed feature distribution. Can two samples have equal richness but different Shannon? Yes, because their evenness can differ. Does a high Simpson value always mean high diversity? No, because there are several Simpson conventions with different directions. Can rarefaction recover undetected species? It subsamples or estimates aspects of observed feature distributions; it does not directly reveal every organism.
Why does sequencing depth matter? Deeper sampling can reveal more rare features. Is Faith PD the same as feature richness? No, it also uses phylogenetic branch lengths. Does a significant alpha difference mean management caused a change? Not unless design and evidence support causal interpretation. Is the most diverse sample always most resilient? No; that wider ecological inference requires separate evidence and definitions.
Independent practice with explained answers
Sample X has four ASVs at 25% each; sample Y has four ASVs dominated by one at 85%, with the remaining 15% shared among three. Which is richer? Neither: both have observed richness four. Which is more even? X. Which is likely to have higher Shannon under these proportions? X, because its uncertainty across randomly sampled feature reads is greater. The answer follows the metric definition, not a judgement about soil quality.
A second question: suppose sample X has twice as many retained reads as sample Y and more observed ASVs. Can you conclude that X’s soil contains more biological species? Not from those figures alone. Examine depth harmonisation, extraction and independent replication. A third question: Shannon stays stable while a PCoA plot shows changed community composition. Is that impossible? No. Within-sample diversity and between-sample identity can vary independently.
Sources and next reading
The QIIME 2 diversity documentation defines available alpha metrics and alpha-rarefaction workflows. The QIIME 2 feature-table documentation explains the underlying feature data types and transformations. The National Cancer Institute microbiome diversity tutorial offers a practical comparison of richness, evenness and diversity indices. Read original method definitions when results depend on a particular index convention.
Within eduKateSG, read soil ASV tables for how features were inferred, soil shotgun metagenomics for functional-potential questions, and soil microbial biomass for a differently defined biological quantity. Links serve distinct reader needs; none should be treated as an interchangeable measurement of diversity.
The conclusion that holds across datasets
Alpha diversity becomes understandable when its two basic ingredients are visible: how many features were detected, and how reads are distributed among them. Shannon, Simpson, evenness, richness and phylogenetic metrics emphasise different aspects of that information. Rarefaction and sampling checks remind us that the observed table is a limited view of a complex soil habitat.
A sound translation from an index to plain language names the feature pool, metric definition, depth and uncertainty before adding ecological interpretation. That discipline helps readers appreciate genuine community patterns while avoiding unsupported claims about living cell numbers, universal soil health or process rates. In scientific work, a precise modest claim is more useful than an impressive conclusion that the data cannot defend.