Soil microbial community assembly analysis asks why the composition of microbial communities varies among sites. If you need to understand the beta nearest taxon index (βNTI), beta mean nearest taxon distance (βMNTD), Raup–Crick Bray–Curtis (RCbray) values, null models, heterogeneous selection, homogeneous selection or dispersal-related categories, begin by separating the observed community pattern from the model used to explain it.
A βNTI score above two is not a direct measurement of an environmental force, and an RCbray classification does not prove that individual microorganisms physically travelled or failed to travel between plots. These methods compare observed phylogenetic or compositional turnover with patterns generated under specified null-model assumptions. Conventional thresholds help summarise departures, but the biological interpretation depends on the marker data, phylogenetic tree, regional species pool, sampling design and environmental context.
This guide explains soil microbial assembly inference with first principles, step-by-step definitions, worked calculations, interpreting βNTI and RCbray together, common statistical traps, independent validation and clear scientific English. All numbers in examples are invented for teaching and are not field observations. The goal is to help a reader recognise what a null-model result supports and where a causal explanation requires further evidence.
Quick answer: what do these metrics try to distinguish?
Community assembly describes the processes shaping which microorganisms are present and how their observed communities differ across sites. Ecologists often discuss environmental selection, dispersal, ecological drift and diversification among relevant influences. A βNTI-style null-model analysis investigates whether between-community phylogenetic turnover is greater or less than expected under a selected randomisation. An additional RCbray-style calculation may examine composition-based turnover for pairs not showing a strong phylogenetic departure.
The output is an interpretation framework, not direct observation of each ecological process. A common classification labels large positive βNTI as compatible with greater-than-null phylogenetic turnover and large negative βNTI as less-than-null turnover. Values nearer zero may prompt a second comparison, but even those categories are conditional on the null model. The strongest reports explain exactly which question the randomisation asked.
Start with the ecological question
Suppose you are comparing microbial communities in wet and dry soil patches. Do differences mainly reflect selection by water availability, barriers to dispersal, chance fluctuations or changing habitat history? A community assembly framework can organise these hypotheses. However, a simple correlation between moisture and community composition does not identify one dominant mechanism. The study needs appropriate sampling, environmental variables and models with explicit assumptions.
Write the intended explanatory question before choosing an analysis. A comparison of adjacent plots in one field may warrant different expectations from a comparison across distant regions. A disturbance experiment can supply stronger causal evidence than a one-time geographical survey, if designed and replicated appropriately. Null-model analysis is most valuable when it guides these questions instead of functioning as an automatic label generator.
Deterministic and stochastic are broad categories
‘Deterministic’ is often used for influences such as environmental filtering and biological selection that make some taxa more likely than others in given conditions. ‘Stochastic’ describes chance-related processes including ecological drift and components of dispersal and diversification under some frameworks. The distinction is conceptual: real microbial communities can be influenced by both, and their relative importance can vary across space, time and biological groups.
Do not translate stochastic as ‘random and meaningless’ or deterministic as ‘perfectly predictable.’ A stochastic model can have strong probability structure and biologically constrained outcomes. Selection can be complicated and contingent. These categories help organise evidence but cannot by themselves explain specific interactions, metabolic rates or causal chains. The method determines what can be inferred from observed patterns.
Observed communities come from a sampling chain
A soil ASV table contains detected marker features across biological samples after collection, extraction, amplification, sequencing and bioinformatic processing. The method typically does not count every living cell or species. Soil depth, season and microbial habitat heterogeneity affect the input. A null-model analysis inherits this measurement scope rather than correcting every laboratory bias.
For the primary feature meaning, read the soil ASV table guide. Before interpreting assembly categories, identify the marker, feature filtering, sequencing-depth decisions and sample independence. A robust ecological result requires that the differences among communities are not primarily artefacts of inconsistent processing.
Why phylogenetic relationships enter the analysis
Ecologists sometimes reason that closely related microorganisms may share certain conserved ecological traits, making phylogenetic turnover informative about habitat selection. This is an assumption to test, not a guarantee. Many microbial functional traits vary substantially among close relatives because of gene gain, loss, horizontal transfer and environmental regulation. Deep or short-marker phylogenies can also have limited resolution.
A βNTI analysis uses phylogenetic distances as its core input, so its biological meaning depends on the quality of the tree and how strongly phylogenetic relatedness tracks traits relevant to the environmental gradient. If traits are not phylogenetically conserved at the chosen scale, simple selection interpretations weaken. Good papers disclose the tree method and evaluate whether their ecological assumptions are plausible.
A phylogenetic tree is not an ecological interaction map
A phylogenetic tree describes inferred evolutionary relationships among marker features. It does not show which organisms cooperate, compete, feed on one another or move together. Short branches can indicate closer inferred ancestry on the chosen molecular scale without guaranteeing identical nutrient use. Long branches describe more sequence-based evolutionary separation, not automatically greater functional difference.
This matters because a null-model result may be described using the word ‘selection’. It does not necessarily identify the trait under selection or any direct interaction. Additional environmental measurements, cultivation, metagenomics or process assays may be needed to connect phylogenetic pattern with mechanism. The tree is one evidence layer among several.
What beta mean nearest taxon distance measures
Beta mean nearest taxon distance, abbreviated βMNTD, summarises phylogenetic nearest-neighbour distances between taxa represented in two communities under a specified weighting convention. For each taxon in one community, the method finds a nearest relative in the other community on the phylogenetic tree; a corresponding mean can incorporate abundances depending on implementation. The measure captures a particular aspect of between-community phylogenetic separation.
It is not identical to the full phylogenetic distance between every pair of taxa, nor to Bray–Curtis dissimilarity of counts. Details such as whether weights are abundance-based and how directions are combined matter. A caption must identify the software function and options rather than relying solely on an acronym that hides method choices.
An intuitive nearest-neighbour illustration
Imagine one soil sample with three marker features and another with three. Each feature in sample A has a nearest relative in sample B at distances 0.1, 0.2 and 0.3 arbitrary phylogenetic branch-length units. A simple unweighted mean is 0.2. A different sample pair might have nearest distances of 0.1, 0.5 and 0.6, giving mean 0.4. Under that illustrative calculation, the second pair is more phylogenetically separated.
Real βMNTD implementations can use weights, bidirectional relationships and additional rules. The toy example teaches the nearest-neighbour idea, not a universal reproduction of package output. The units depend on the constructed phylogeny. The number alone does not tell us whether an observed separation is unusually large; that requires comparison with the null distribution.
Why a null model is necessary
A null model constructs a distribution of expected values under a defined randomisation or sampling process. Instead of asking merely whether two communities have a large observed βMNTD, the researcher asks whether their turnover differs from what the chosen null process would generate. Randomisation rules define which features can be assigned or shuffled, what abundances are preserved and what regional species pool is considered.
There is no single universally correct null model. Changing the allowed randomisation can change the expectation and thus the standardised score even when the observed community data remain unchanged. A complete result must specify which constraints were preserved and why that null scenario represents the intended scientific question.
What βNTI standardisation does
A commonly used βNTI standardises the observed βMNTD against a null distribution: βNTI equals observed βMNTD minus the mean null βMNTD, divided by the standard deviation of the null distribution. The numerator is the departure from model expectation; the denominator scales that departure by null variation. The resulting score is unitless under this formulation.
This transformation does not make the output a probability or physical transport measurement. It expresses how unusual the observed turnover is in standard-deviation units relative to the chosen null model. Small changes in the null mean or variance can change the score substantially. The method should report the number of randomisations and the randomisation rules so the reader can understand how expectations were constructed.
Worked βNTI calculation: a positive departure
Suppose observed βMNTD is 0.36 branch-length units, the null mean is 0.24 and the null standard deviation is 0.04. Subtracting gives 0.12; dividing 0.12 by 0.04 yields βNTI equal to 3.0. This is a positive departure three null standard deviations above the selected model expectation.
Under a frequently used heuristic, a βNTI above +2 is classified as greater-than-null phylogenetic turnover, often discussed in relation to heterogeneous selection. The arithmetic is clear; the ecological causal label remains conditional. It would be incorrect to claim the value directly measures three units of environmental selection or identifies the particular environmental variable responsible.
Worked βNTI calculation: a negative departure
Now suppose observed βMNTD is 0.10, the null mean is 0.22 and the null standard deviation is 0.04. The difference is minus 0.12, and division gives βNTI equal to −3.0. Observed nearest-neighbour phylogenetic turnover is lower than the null mean by three null standard deviations.
A common interpretation is that pairwise communities are phylogenetically more similar than expected, potentially consistent with homogeneous environmental selection. But shared dispersal or other constraints may also influence patterns, depending on the model and dataset. The correct description begins with lower-than-null turnover and then evaluates possible ecological explanations rather than treating the heuristic as a directly observed mechanism.
What the plus-two and minus-two thresholds mean
Thresholds near βNTI greater than +2 and less than −2 are widely used heuristics for identifying stronger departures from a null distribution under particular implementations. Values between −2 and +2 are often treated as not showing strong evidence of deterministic phylogenetic turnover at that threshold. These boundaries are conventions, not natural laws written into soil ecosystems.
The chosen cutoff interacts with null-distribution shape, randomisation count and statistical calibration. A score of 2.01 should not be treated as a fundamentally different ecological universe from a score of 1.99. Readers should look at distributions, effect directions, method descriptions and sensitivity checks instead of relying solely on a hard classification boundary.
‘Homogeneous selection’ is an inference
When multiple sites experience similar environmental filters, ecologists may hypothesise that those filters favour related microbial lineages, yielding lower-than-null phylogenetic turnover. This is commonly described as homogeneous selection. Examples might involve similarly acidic soils or consistently oxygen-limited habitats, but the environmental trait must actually be measured to evaluate a specific explanation.
A negative βNTI alone does not show that pH, oxygen or any named factor caused the pattern. The statistic concerns phylogenetic similarity relative to a null expectation. Linking it to an environmental mechanism requires evidence of the relevant gradient, trait conservation and suitable comparisons. A careful sentence says ‘consistent with stronger homogeneous-selection influence under the model,’ not ‘the index proved pH selected these taxa.’
‘Heterogeneous selection’ also needs a tested gradient
A positive βNTI can reflect more phylogenetic turnover than expected under a randomisation, often interpreted as compatible with different environmental filters across sites. This is frequently called heterogeneous selection. Two soil patches with different salinity, redox state or plant communities could plausibly favour distinct microbial lineages.
But the statistic does not select among competing environmental explanations. The samples may differ in many correlated factors and have different dispersal histories. A study must measure those variables and, ideally, use controlled experiments or stronger designs to identify the driver. The phrase ‘heterogeneous selection’ is a framework category supported to varying degrees, not an independently measured chemical force.
What happens when absolute βNTI is below two?
If a pair has βNTI between −2 and +2 under the common convention, its phylogenetic turnover is not strongly displaced from the null expectation according to that cutoff. This does not prove all assembly was random, and it certainly does not show absence of environmental selection. Some effects may be weak, balanced by other processes or invisible to the marker and phylogenetic resolution.
Researchers sometimes use an additional composition-based null metric for these pairs. The goal is to distinguish patterns potentially compatible with different dispersal-related regimes or an ‘undominated’ fraction. This second stage should be read as an operational partition of pairwise comparisons, not a direct measurement of how cells moved.
What RCbray is intended to compare
RCbray commonly refers to a Raup–Crick-type null-model metric built from Bray–Curtis community dissimilarity, rescaled under a particular implementation. It compares observed composition-based turnover with what the null model would generate. Depending on the formulation, strongly positive values indicate higher-than-null compositional dissimilarity and strongly negative values lower-than-null dissimilarity.
The metric is not the same as ordinary Bray–Curtis dissimilarity. A raw Bray–Curtis value is a function of two observed feature vectors; RCbray is a model-standardised comparison involving a null distribution. The transformation and sign convention must be specified, because similar names can conceal implementation differences.
A two-stage classification is conditional by design
Many studies use βNTI first to identify pairs with relatively large phylogenetic departures. For pairs within the middle band, they then apply an RCbray threshold. This produces a table of pairwise comparisons assigned to operational categories such as heterogeneous selection, homogeneous selection, dispersal limitation, homogenising dispersal or undominated processes.
Those labels are not independent measurements of five physical processes. They are interpretations attached to combinations of null-model outcomes under predefined rules. If the assumptions change, so can the category proportions. A good article describes the workflow and names the computational criteria before presenting the resulting proportions of sample pairs.
Interpreting high positive RCbray
Under a common convention, an RCbray value greater than +0.95 among pairs not already identified as phylogenetically selected is treated as unusually high compositional dissimilarity and discussed in relation to dispersal limitation. The idea is that restricted movement or establishment might produce communities more different than expected under the null reference.
That interpretation needs care. Large compositional separation may also reflect unmeasured habitat differences, historical contingencies or model misspecification. The value does not count dispersing cells or estimate a physical dispersal rate. When reporting it, use ‘operationally classified as consistent with dispersal limitation’ unless independent spatial, experimental or transport evidence justifies a stronger conclusion.
Interpreting low negative RCbray
Under the same common convention, RCbray less than −0.95 may be labelled homogenising dispersal, a pattern in which observed composition is more similar than the chosen null expectation. High movement among sites can be one possible mechanism maintaining similarity, but shared environmental conditions and other processes may also contribute.
A negative score does not demonstrate measured microbial migration. Real dispersal would require appropriate spatial, movement or transplant data to establish rates and pathways. The label is useful as a hypothesis within a comparative framework, yet it must not be presented to land managers as a directly observed volume of microbial exchange.
What the undominated category means
When both βNTI and RCbray remain within their respective conventional central intervals, analyses sometimes classify a comparison as ‘undominated’. This means the selected tests did not identify a strong departure of the types used to assign the other categories. It does not mean no ecological processes operated or that the community was assembled without any constraints.
Potential explanations may include drift, weak selection, mixed processes, diversification or insufficient sensitivity. The exact mix cannot generally be reconstructed from the category label alone. Readers should treat undominated as ‘not assigned to one of the more extreme operational bins under this analysis,’ not a measured fraction of purely random ecology.
Worked example: applying the two-stage rules
Imagine three illustrative sample pairs under the common threshold convention. Pair A has βNTI +2.8 and is placed in the higher-than-null phylogenetic turnover category; there is no need to use RCbray to give it that primary label. Pair B has βNTI +0.7 and RCbray +0.97, placing it in a positive RCbray category. Pair C has βNTI −0.4 and RCbray +0.20, leaving it in the central undominated category.
These are classifications, not proofs of actual causal transport or selection. The numerical example teaches the decision logic. A useful published table should show both metrics and the criteria, enabling readers to reproduce the classification instead of accepting a single-colour summary as self-explanatory.
Negative and positive signs are not ‘good’ or ‘bad’
A positive βNTI does not mean a healthy soil, and a negative value does not mean a damaged soil. The sign describes direction of departure from a null phylogenetic expectation. Likewise, positive RCbray does not mean biological improvement, and negative RCbray does not mean that microbes lost functions. Neither statistic measures crop yield, pathogen pressure, fertility or climate impact directly.
Ecological value is a separate research question requiring independently chosen outcomes and contextual judgement. When explaining a figure to students, avoid colour legends that imply red is harmful and green is beneficial unless the evidence and legend explicitly justify that interpretation. Technical metrics should retain their mathematical meanings before being linked to environmental consequences.
The regional species pool defines the randomisation universe
A null model may draw from a regional pool of observed taxa or apply another rule about which features could potentially occur in each community. The choice of pool influences expected phylogenetic and compositional turnover. A global pool spanning distant environments might produce different null distributions from a pool restricted to one field and season, even with unchanged observed pair data.
This is why two studies can report different βNTI values for superficially similar soils. Their null hypotheses may differ. The question ‘What could have occurred by chance under this model?’ must be answered before the model’s deviation is interpreted. A transparent methods section defines the taxon pool, occurrence constraints and abundance-handling rules.
Abundance weighting changes the outcome
Nearest-taxon and composition-based metrics can be calculated with or without abundance weighting, depending on software and implementation. A community dominated by one ASV may look different from the same community treated as a simple list of presences. In amplicon data, read abundance itself is influenced by marker copy number and laboratory biases, creating another layer of interpretation.
Readers should check whether weights use read counts, proportions or transformed values. If filtering removes low-abundance features, phylogenetic distances and null expectations may change. A conclusion stable across reasonable feature and weighting choices is stronger than one dependent on an undocumented selection. The specific data transformation belongs beside the reported categories.
Tree uncertainty must be acknowledged
A phylogenetic tree inferred from a short marker region has limited resolution, especially among closely related microbes. Alignment quality, branch-length estimation and rooting conventions can affect nearest-taxon distances. Some features may be placed ambiguously, and reference databases may have uneven coverage across lineages. These uncertainties can propagate into βMNTD and βNTI.
When an ecological conclusion hinges on small metric differences near a threshold, sensitivity to phylogenetic reconstruction deserves attention. Alternative reasonable trees or branch-length models may help evaluate robustness. A null model applied to a questionable tree does not gain certainty from thousands of randomisations. Computational repetition cannot correct a biological representation that is inappropriate for the question.
Trait conservation is an important interpretive assumption
For phylogenetic turnover to imply particular kinds of selection, traits associated with environmental performance must show relevant phylogenetic structure at the analysed scale. Microbial traits can be conserved or labile, and horizontal gene transfer can weaken correspondence between marker relatedness and function. Environmental filtering may act on traits that a 16S tree does not capture well.
If a study has trait data, evaluate whether the traits of interest are phylogenetically patterned. If it does not, state the mechanism as a hypothesis. A βNTI framework can still describe departures from expected phylogenetic structure, but its ecological labels require more than the numerical score. The inferential chain is strongest when marker phylogeny, trait evidence and measured environmental conditions align.
Pairwise comparisons are not independent observations
For n samples there are n multiplied by n minus one divided by two unique unordered sample pairs. With ten samples, that gives forty-five pairs. Yet those forty-five distances share the same original ten samples. Treating every pair as an independent field replicate can underestimate uncertainty and make category percentages appear more statistically precise than justified.
A report might state that 60% of pairs fell into one operational category. That is a description of the pairwise comparison set, not evidence from 60 independently sampled ecosystems. Use models and summaries appropriate to the dependence among pairwise distances. When aggregating by treatment or spatial group, show how the pair sets were defined and how uncertainty was assessed.
Worked pair-count calculation
Suppose a study has twelve independent soil samples. The number of unordered sample pairs is twelve times eleven divided by two, equal to sixty-six. If fifteen of those pairs exceed a chosen positive βNTI threshold, the proportion of pairs in that category is fifteen divided by sixty-six, approximately 22.7%. That percentage describes pair classifications among the twelve samples.
It does not mean 22.7% of microbes were selected by the environment or that 22.7% of assembly events were observed directly. The pairs share observations, so the percentage also should not be given an ordinary independent-binomial confidence interval without attention to dependence. The denominator and statistical unit matter as much as the result.
Spatial distance is a separate measurement
Geographic distance may be measured in metres or kilometres between soil sampling sites. βNTI and RCbray are not spatial distances; they concern biological profiles and null expectations. To test whether geographic separation relates to community turnover, the researcher must record coordinates or distances and use appropriate spatial analysis.
A pattern where distant sites differ more may be consistent with dispersal limitation, environmental distance, historical separation or multiple mechanisms. Spatial autocorrelation complicates simple significance tests. Readers should not infer direct microbial travel distances from an RCbray label, nor assume every beta-diversity difference demonstrates a barrier to dispersal.
Environmental gradients need independent measurement
If communities along a pH gradient show positive βNTI departures, the result may be discussed as compatible with heterogeneous selection. But to specifically implicate pH, the study should measure it and consider correlated factors such as organic matter, moisture, vegetation and mineral composition. A model built only from phylogenetic data does not isolate chemical drivers.
A stronger design may cross pH or moisture treatments experimentally while controlling other conditions, or examine repeated measurements across independent gradients. Null-model results can then contribute to an interpretation supported by other evidence. In plain language, say which variable was measured and which process is inferred, rather than treating a statistical category as a laboratory assay of selection.
Why disturbance history matters
Soil communities reflect past fires, cultivation, flooding, drought, amendments and dispersal events. Two fields with similar present pH and moisture can still contain different lineages because their historical trajectories differ. A null-model category summarises present comparative patterns but does not reconstruct the entire sequence of historical events.
Record land-use history, sampling date and disturbance timing when interpreting assembly mechanisms. A one-time observation cannot easily distinguish persistent selection from transient post-disturbance recovery. Longitudinal sampling and targeted experiments can strengthen the link between history and current turnover. Avoid narrating a complete ecological past from one βNTI number.
Diversity and assembly are related but different questions
Alpha diversity summarises within-sample features and their relative distributions. Beta diversity summarises between-sample community differences under selected metrics. Assembly analysis adds an explanatory null-model layer to phylogenetic or compositional turnover. A soil can have high alpha diversity while pairwise turnover is unusually large or small under a null model; those statements are not contradictory.
The soil alpha diversity guide covers Shannon and richness, while the soil beta diversity guide covers Bray–Curtis, PCoA, NMDS and PERMANOVA. This article asks the next question: what departures from a defined null model are compatible with candidate assembly processes?
Network graphs do not identify assembly forces
A microbial co-occurrence network displays inferred statistical relationships between features, often based on correlations or conditional associations. βNTI describes phylogenetic turnover relative to a null distribution. RCbray addresses composition-based departure under another null comparison. These analyses answer different questions and cannot validate one another merely by producing visually compatible patterns.
The soil network guide explains why network edges are not direct ecological interactions. A highly connected node does not prove community assembly is deterministic. Similarly, a βNTI category does not establish direct relationships between individual taxa. Use each method for its defined purpose.
Common mistake: converting model categories into causal percentages
A paper might state ‘70% deterministic assembly’ after finding that 70% of analysed pairwise comparisons fell outside a βNTI central threshold. That phrase can suggest researchers directly observed and quantified physical selection events. The more precise description is that 70% of the selected pair comparisons were assigned to the operational higher- or lower-than-null phylogenetic turnover bins under the chosen framework.
If the authors discuss deterministic influence, they should make clear that it is a model-based interpretation. The result is conditioned on the null pool, randomisation method, tree, thresholds and sampling design. Reporting these dependencies does not weaken the analysis; it tells other scientists how to reproduce and evaluate the conclusion.
Common mistake: calling central βNTI pure randomness
Values within a conventional βNTI central band do not establish that the community lacks selection or ecological structure. The method may not detect weak selection, opposing processes or effects acting on traits not conserved in the phylogeny. Likewise, central RCbray values can reflect an undominated mixture rather than pure drift.
Replace ‘entirely random assembly’ with ‘not strongly distinguishable from the chosen null under these metrics’ when that is the evidence. A carefully qualified statement can then list plausible alternative processes and direct measurements that would distinguish them. This restores the important gap between statistical nonsignificance and absence of an ecological mechanism.
Common mistake: ignoring null-model sensitivity
Changing the regional species pool, abundance weights, rare-feature filter or phylogenetic tree can alter null expectations. When a study presents precise percentages of assembly categories without examining sensitivity, it may suggest a certainty that the modelling assumptions do not deserve. A robust comparison should check whether the main qualitative conclusion survives plausible alternatives.
The objective is not to run every imaginable null model until one agrees with a preferred story. It is to justify a scientifically relevant reference process and evaluate reasonable alternatives. The chosen null should reflect the biological question. Where inference depends heavily on one arbitrary modelling decision, say so clearly and avoid definitive causal claims.
A careful interpretation of a sample-pair table
Begin with the two sample identifiers and their biological grouping. Confirm the input feature table and phylogenetic tree. Find the observed βMNTD, null mean and standard deviation, and calculate or verify βNTI. If the pair lies in the central βNTI range and the workflow uses RCbray, inspect that value and classification rule. Finally, ask which empirical environmental and spatial observations support the proposed mechanism.
Do not interpret the colour label alone. An explicit sample-pair example makes the analysis auditable and reveals when categories depend on borderline scores. If a table omits the underlying null-model settings, the causal explanation should remain provisional. A reader cannot independently assess an inference whose randomisation rules are unknown.
Designing a stronger field test
To test a dispersal-limitation hypothesis, sample across distance while measuring environmental variation and, where feasible, manipulate connectivity or exchange. To examine selection, use controlled environmental gradients or reciprocal-transplant approaches with adequate replication. Record trait, process and community changes through time rather than relying on a single cross-sectional analysis.
These designs have practical complications, but they identify what the null-model categories cannot settle alone. A strong scientific project uses βNTI and RCbray to prioritise hypotheses, then measures the mechanisms more directly. For environmental management decisions, a supporting process measurement is often more actionable than a category percentage in isolation.
Terminology for ordinary English and translation
‘Assembly’ here means the ecological formation and maintenance of observed communities, not a manufactured structure. ‘Selection’ is an inference about differential ecological filtering, not necessarily conscious choice. ‘Null’ names a defined comparison expectation, not absence of all biology. ‘Dispersion’ refers to movement or spread only in the appropriate ecological model, while ‘dispersal limitation’ is an inferred process category.
Words such as deterministic, stochastic, turnover, standardised, homogeneous and heterogeneous can be misunderstood without definitions. Use the cross-linguistic transfer guide to see how technical meaning can drift across languages. The translation-and-context guide encourages preserving the mathematical relationship between observations and inferences.
Frequently asked questions
Is βNTI a direct measurement of environmental selection? No; it is a standardised deviation of observed phylogenetic turnover from a defined null expectation. Does positive βNTI always mean healthy soil? No. What does βNTI below −2 commonly suggest? Lower-than-null phylogenetic turnover, often discussed as compatible with homogeneous selection under the model. Does a central value prove all assembly was random? No.
Is RCbray the same as ordinary Bray–Curtis? No; it uses a null-model comparison. Does RCbray above 0.95 prove cells could not disperse? No, it is conventionally associated with a dispersal-limitation interpretation under a particular workflow. Can thresholds and null pools change the classification? Yes. Are sample pairs independent biological replicates? No; many pairs share the same original samples.
Independent practice with explained answers
An observed βMNTD of 0.31, null mean 0.23 and null standard deviation 0.04 produce βNTI equal to (0.31 minus 0.23) divided by 0.04, or +2.0. Under a strict rule requiring greater than +2, that boundary value is not in the positive extreme category. Why does this matter? Classification depends on the stated inequality and the precision of the underlying calculation.
Next, a pair has βNTI +0.3 and RCbray −0.97 under a framework using central βNTI and RCbray less than −0.95 for a homogenising-dispersal-compatible category. How is it classified? Under those declared rules, it enters the negative RCbray category. Does that prove microbes physically migrated between the fields? No. Finally, fifteen samples generate how many unique unordered pairs? Fifteen times fourteen divided by two equals 105, but only fifteen original biological samples exist.
Authoritative sources and next reading
The mSystems analysis of stochastic and deterministic microbial assembly explains a frequently used βNTI and RCbray classification, including conventional thresholds and process labels. Read such primary research alongside the specific study’s implementation, because model conventions and ecological assumptions need not be identical across applications. Technical categories are most useful when their exact construction remains reproducible.
Continue with eduKateSG’s soil beta-diversity guide, soil alpha-diversity guide, soil microbial network analysis and soil source tracking. Their results can contribute different parts of an ecological story, but none should be made to answer a question outside its measurement design.
Closing lesson: a model is a disciplined comparison
Community assembly null models are valuable because they replace unstructured speculation with defined comparisons between observed biological patterns and explicit expectations. βNTI can identify unusually high or low phylogenetic turnover under a selected reference process; RCbray can help examine compositional turnover in a second stage. Their conventional labels organise hypotheses about selection, dispersal and other influences.
The reader’s task is to preserve the difference between a classification rule and a measured causal event. Name the phylogenetic tree, null pool, weighting, thresholds, biological sampling units and independent environmental evidence before making claims about assembly processes. That discipline makes a specialised scientific figure understandable, useful and honest without flattening uncertainty into a claim that the model has observed forces it only infers.