Beta diversity measures how much species composition changes from one site, sample, or time point to another. Calculating it typically involves building a matrix of pairwise dissimilarity values between all your samples using an index suited to your data type, then interpreting those values through ordination plots, statistical tests, and partitioning methods. The choice of index matters more than many researchers expect, and the same dataset can tell different stories depending on whether you focus on species replacements between sites, losses of species subsets, or shifts in relative abundances.
What Beta Diversity Actually Measures
Robert Whittaker originally defined beta diversity as the variation in species composition among sites in a geographic area.1Ecological Monographs. ANALYZING BETA DIVERSITY: PARTITIONING THE SPATIAL VARIATION OF COMMUNITY COMPOSITION DATA That definition sounds simple, but it hides a lot of complexity. Two communities can differ because they share no species at all, because one is a depleted subset of the other, or because abundant species in one community are rare in the other. Each of those patterns has different ecological meaning, so the index you choose determines which kind of difference you detect.
Every beta diversity calculation starts with a community data table: rows are your sites or samples, columns are your species or operational taxonomic units, and cells contain either presence-absence records (1s and 0s) or counts and relative abundances. From that table you compute a dissimilarity value for every pair of samples, producing a square distance matrix. Values usually range from 0 (the two samples are identical) to 1 (they share nothing in common). That matrix is the raw material for everything that follows: ordination, hypothesis testing, and decomposition into ecological components.
Presence-Absence Indices
If your data records only whether a species is present or absent, the two most common indices are Sørensen dissimilarity and Jaccard dissimilarity. Both compare the number of shared species between two samples against the total number of species found across them, but they weight shared species differently. Sørensen gives double weight to species found in both sites, making it more sensitive to compositional overlap. Jaccard treats shared and unique species equally, producing slightly higher dissimilarity values for the same data.
A practical consideration that often gets overlooked is how robust these indices are to errors in your dataset, whether from misidentified organisms, incomplete sampling, or patchy detection. A study testing multiple presence-absence indices found that Jaccard dissimilarity was the most robust to such errors overall, while some commonly used indices like Sørensen and Simpson were relatively unreliable when taxonomy or sampling introduced mistakes.2Ecosphere. How robust are popular beta diversity indices to sampling error? If your sampling is uneven or your taxonomic identifications are uncertain, that robustness advantage is worth knowing about before you commit to an index.
A third presence-absence metric, Simpson dissimilarity (sometimes labeled βsim), focuses specifically on species replacement and is mathematically independent of differences in species richness between two sites. That independence can be a strength when you want to isolate true turnover from the confounding effect of one site simply having fewer species. Work comparing decomposition frameworks confirmed that the replacement component related to Simpson dissimilarity is genuinely independent of richness difference, while analogous components in some alternative frameworks are not.3Methods in Ecology and Evolution. Comparing methods to separate components of beta diversity
Abundance-Based Indices
Presence-absence data throws away a lot of information. A site where a species is represented by a single individual looks identical to a site where that species dominates the community. Abundance-based indices fix this by incorporating counts or proportional abundances. The most widely used is Bray-Curtis dissimilarity, which sums the absolute differences in abundance for each species and divides by the total abundance across both samples. A Bray-Curtis value near 0 means two communities have nearly identical species in nearly identical proportions; a value near 1 means they are completely different.
Bray-Curtis works for pairwise comparisons, but extending abundance-based dissimilarity to situations where you want a single number summarizing how different many sites are simultaneously required new formulations. Multiple-site extensions of both Bray-Curtis and the related Ruzicka index have been developed, and these can be decomposed into two components: balanced variation in abundance (where individuals of one species are replaced by individuals of another across sites) and abundance gradients (where some sites have systematically more or fewer individuals overall).4Methods in Ecology and Evolution. Partitioning abundance‐based multiple‐site dissimilarity into components: balanced variation in abundance and abundance gradients That decomposition parallels the turnover-versus-nestedness split for presence-absence data, which is covered next.
Splitting Beta Diversity into Turnover and Nestedness
A raw dissimilarity number between two sites tells you they differ, but not why. The difference could arise from species replacement, where species found at one site are swapped for entirely different species at the other, or from nestedness, where the poorer site’s species list is a subset of the richer site’s. These two processes have very different ecological implications. Turnover often points to environmental filtering or barriers to dispersal. Nestedness often suggests selective extinction or colonization with species being lost from some sites in a predictable order.
A widely used framework additively partitions total Sørensen or Jaccard dissimilarity into a turnover component and a nestedness-resultant component, so the two components sum exactly to the total.5Global Ecology and Biogeography. Partitioning the turnover and nestedness components of beta diversity This works for both pairwise comparisons and multi-site situations. In practice, you compute the total dissimilarity, compute the turnover-only component, and then obtain nestedness as the remainder. If turnover dominates, your sites are exchanging species along some gradient. If nestedness dominates, your less diverse sites are impoverished versions of your most species-rich site.
The R package betapart makes this partitioning straightforward, computing total Sørensen or Jaccard dissimilarity along with their respective turnover and nestedness components.6Methods in Ecology and Evolution. betapart: an R package for the study of beta diversity When reporting results, presenting all three values together (total, turnover, and nestedness) gives a much richer picture than reporting total dissimilarity alone.
Phylogenetic and Functional Beta Diversity
Standard beta diversity treats all species as equally different from one another. A site with five unique beetle species looks just as dissimilar from a reference site as one with five unique species spanning beetles, birds, and fungi. Phylogenetic beta diversity addresses this by incorporating evolutionary relationships. If two communities differ only in closely related species, their phylogenetic beta diversity is low; if they differ in distantly related lineages, it is high.
In microbial ecology, UniFrac distances are the standard phylogenetic beta diversity metric. Unweighted UniFrac considers only presence-absence on the phylogenetic tree, while weighted UniFrac incorporates abundance, giving more weight to lineages that are common rather than rare.7PubMed Central. Quantitative and qualitative beta diversity measures lead to different insights into factors that structure microbial communities The two versions frequently lead to different conclusions from the same dataset: unweighted UniFrac is more sensitive to rare lineages and can pick up the presence of unusual taxa, while weighted UniFrac reflects shifts in dominant community members. Running both and comparing results is common practice in microbiome studies.
Functional beta diversity takes a different angle entirely, comparing communities based on their organisms’ traits rather than their taxonomy or evolutionary history. Instead of asking “do these sites share species?” it asks “do these sites share ecological strategies?” A variety of indices exist, differing in how they account for between-species similarities in trait space when calculating community dissimilarity. A review of available trait-based dissimilarity indices identified important conceptual and technical differences among them that researchers need to consider before choosing one.8Ecography. A guide to between‐community functional dissimilarity measures Using taxonomic, phylogenetic, and functional beta diversity together can reveal whether community differences are driven by wholesale lineage replacement, by gains and losses of closely related species with similar niches, or by shifts in functional roles regardless of which species fill them.
Testing Whether Observed Differences Are Statistically Real
A dissimilarity matrix shows you how different your communities are, but it does not tell you whether those differences are larger than you would expect by chance. For that, you need a statistical test. The most widely used is PERMANOVA (permutational multivariate analysis of variance), which partitions variation in the distance matrix according to grouping factors, such as habitat type or treatment, and then assesses significance by randomly shuffling sample labels thousands of times.9Wiley StatsRef: Statistics Reference Online. Permutational Multivariate Analysis of Variance The test produces a pseudo-F statistic analogous to the F-ratio in classical ANOVA: larger values indicate that between-group differences are large relative to within-group variation.10Bioinformatics. Power and sample-size estimation for microbiome studies using pairwise distances and PERMANOVA
PERMANOVA has a well-known sensitivity issue: it can flag a significant difference when the groups actually have similar centroids but different amounts of spread (dispersion) around those centroids. In other words, it might tell you two habitats differ when really one habitat just has more variable communities than the other. To check for this, researchers typically pair PERMANOVA with a test for homogeneity of multivariate dispersions, often called BETADISPER, which measures the average distance from individual samples to their group centroid and compares those average distances across groups.11PubMed Central. Multivariate dispersion as a measure of beta diversity If PERMANOVA is significant but BETADISPER also shows significantly different dispersions, you need to be cautious: the PERMANOVA result may reflect dispersion differences rather than genuine location shifts in community composition.
Beyond PERMANOVA, visualization through ordination is essential for interpretation. Principal coordinates analysis (PCoA) plots your samples in low-dimensional space based on the dissimilarity matrix, letting you see clustering and gradients. Non-metric multidimensional scaling (NMDS) does something similar but prioritizes preserving the rank order of dissimilarities rather than exact distances, making it especially useful when the relationship between ecological distance and compositional difference is nonlinear. Both are implemented in standard software packages.
Distance-Decay Relationships
One of the most consistent patterns in ecology is that communities become less similar the farther apart they are geographically. This distance-decay of similarity is essentially beta diversity viewed along a spatial axis. Plotting pairwise dissimilarity against the geographic distance between sites typically shows a positive trend: nearby sites share more species than distant ones.12PubMed Central. A general framework for the distance-decay of similarity in ecological communities
The slope and shape of that curve carry ecological meaning. A steep decay suggests strong dispersal limitation or sharp environmental gradients. A shallow decay suggests that species can move freely across the landscape or that environmental conditions are fairly uniform. Regression of similarity against geographic distance unites dispersal processes and environmental filtering into a single analytical framework, providing an effective approach for gauging spatial turnover.13Ecography. The distance decay of similarity in ecological communities You can also swap geographic distance for environmental distance (difference in temperature, pH, soil nutrients, or other variables) to test whether environmental filtering drives community differences independently of spatial separation.
When computing distance-decay curves, the choice of data transformation matters. Applying chord or Hellinger transformations to community composition data before calculating Euclidean distances produces dissimilarity matrices with desirable mathematical properties for downstream analysis.14Ecography. Box–Cox‐chord transformations for community composition data prior to beta diversity analysis Hellinger transformation, which takes the square root of relative abundances before applying the chord transformation, is especially popular because it down-weights dominant species and gives rare species more influence, producing ecologically interpretable distances.
Identifying Which Species Drive the Differences
Once you know that two groups of sites differ in composition, the natural next question is which species are responsible. SIMPER (Similarity Percentage analysis) addresses this by performing pairwise comparisons between groups of samples and ranking all species according to how much each one contributes to the overall average dissimilarity. Species that are consistently present in one group but absent or rare in the other rise to the top of the list.15PLoS ONE. Integrating Taxonomic, Functional and Phylogenetic Beta Diversities: Interactive Effects with the Biome and Land Use across Taxa
SIMPER is intuitive and widely used, but it has a well-known bias toward species with high variance in abundance, sometimes flagging species as important contributors simply because they are patchy. Complement SIMPER results with indicator species analysis or differential abundance testing to confirm that the species flagged as drivers genuinely discriminate between your groups rather than just being noisy.
Dealing with Uneven Sampling Effort
In sequencing-based studies, uneven library sizes are the norm. One sample might yield 50,000 reads while another yields 500. That variation inflates dissimilarity estimates because more deeply sequenced samples detect more rare taxa, making them look compositionally different from shallowly sequenced samples even when the underlying communities are identical. Studies have found as much as 100-fold variation in sequencing depth across samples within a single project.16PubMed Central. Rarefaction is currently the best approach to control for uneven sequencing effort in amplicon sequence analyses
Rarefaction, the practice of randomly subsampling all libraries down to the size of the smallest one, remains the most reliable way to control for this problem when measuring beta diversity. It discards data, which feels wasteful, and alternatives like variance-stabilizing transformations or proportion-based normalization have been proposed. But empirical comparisons keep finding that rarefaction is the only method that consistently controls the effect of uneven sequencing effort on both alpha and beta diversity metrics. If you skip rarefaction and your samples vary widely in depth, your beta diversity results may be driven partly by a technical artifact rather than genuine biological differences.
Software and the Practical Workflow
Most beta diversity analyses happen in R. The vegan package is the workhorse: its vegdist() function computes dissimilarity matrices using Euclidean, Bray-Curtis, Jaccard, Canberra, and other distances, while metaMDS() handles NMDS ordination and adonis2() runs PERMANOVA.17Oxford Academic. The best practice for microbiome analysis using R For PCoA, the ape package’s pcoa() function is standard. betapart handles turnover-nestedness partitioning as described earlier. In microbiome work, QIIME 2 and phyloseq integrate many of these steps into pipelines that handle rarefaction, distance computation, ordination, and PERMANOVA in sequence.
A typical workflow looks like this:
- Normalize: Rarefy or transform your community table to control for uneven sampling.
- Compute distances: Choose an index appropriate to your data type and question, then build the pairwise dissimilarity matrix.
- Visualize: Run PCoA or NMDS to see how samples cluster in reduced-dimensional space. Color points by your grouping variable to look for visual separation.
- Test: Use PERMANOVA to test whether groups differ and BETADISPER to check whether dispersion differences might be confounding the result.
- Decompose: Partition total dissimilarity into turnover and nestedness to understand what is driving the differences.
- Identify drivers: Run SIMPER or indicator species analysis to find which taxa contribute most to the observed dissimilarity.
Not every project needs all of these steps. A straightforward comparison of two treatment groups might stop after PERMANOVA and an ordination plot. A biogeographic study might skip SIMPER entirely and focus on distance-decay modeling. Tailor the workflow to the question you are asking.
Applications in Microbiome Research
Beta diversity has become one of the default metrics in human microbiome studies, where comparing microbial community composition across health states, body sites, or time points is central to the field. Quantitative comparison among microbiomes can link microbial beta diversity to environmental features, enabling prediction of ecosystem properties or dissection of host-microbiome interactions.18PubMed Central. Elucidating the Beta-Diversity of the Microbiome: from Global Alignment to Local Alignment
A meta-analysis across multiple gut microbiome datasets found that beta diversity and beta dispersion were among the characteristics that consistently associate with disease states.19PubMed Central. A Metagenomic Meta-analysis Reveals Functional Signatures of Health and Disease in the Human Gut Microbiome Patients with various conditions often show not only shifted community composition (higher beta diversity from healthy controls) but also greater variability among themselves (higher beta dispersion within the disease group). That dispersion finding is itself informative: it suggests that there may not be a single “disease microbiome” but rather many different ways a gut community can deviate from a healthy configuration.
Applications in Conservation Planning
Conservation has traditionally focused on protecting areas with the most species, an alpha diversity approach. But two highly diverse areas might contain largely the same species, making one of them redundant for regional biodiversity goals. Beta diversity identifies where communities change in composition, and those transition zones are often where unique species assemblages live. Recent work argues that both areas of high species richness and high species turnover are inadequately protected, and that integrating beta diversity into conservation mapping provides a more ecologically robust foundation for long-term planning.20PubMed Central. Integrating α and β diversity in conservation planning
This idea has been put into practice. In designing a protected-areas network for disturbance-sensitive mammals in Canada’s Yukon Territory, researchers found that both regional heterogeneity and compositional turnover between non-adjacent sites were significant predictors of how many protected areas were needed to represent all target species within each ecoregion. In other words, beta diversity told planners not just where to put reserves but how many and how far apart they needed to be.21Conservation Biology. Beta Diversity and Nature Reserve System Design in the Yukon, Canada
Temporal Beta Diversity
Most beta diversity methods were developed to compare communities across space, but the same logic applies to comparing a single community with itself at different points in time. Temporal beta diversity asks: how much has this community changed between year one and year ten, or between pre-disturbance and post-disturbance conditions? You can apply the same indices, the same turnover-nestedness partitioning, and the same statistical tests, but several additional challenges arise. Establishing a meaningful baseline is harder in time than in space, because “the original community” may never have been sampled. Seasonal cycles can masquerade as directional change if sampling intervals are not carefully chosen. And appropriate null models for temporal change are less developed than their spatial counterparts.22Global Ecology and Biogeography. Temporal β diversity—A macroecological perspective
When temporal beta diversity is driven primarily by turnover, it suggests active species replacement, perhaps because environmental conditions are shifting and new species are colonizing while others disappear. When it is driven by nestedness, it suggests progressive impoverishment, with species dropping out over time without new arrivals taking their place. That distinction matters enormously for management: a community undergoing turnover may be adapting to change, while one undergoing nested loss may be collapsing. The same partitioning framework that works in space gives you that diagnostic in time as well, but the interpretive bar is higher because confounding factors like observer turnover and methodological drift accumulate across years in ways they do not across sites sampled simultaneously.