A species can be rare without being especially vulnerable. This study argues that extinction risk depends not just on how much abundance a species has, but on how its dispersal strategy fits into the entire community. Predicting species persistence within ecological communities is a fundamental challenge for both empirical and theoretical ecology.

Existing methods range from mechanistic models, whose parameters are difficult to estimate from data, to statistical tools whose context-specific parameters are less interpretable. The framework integrates the key processes governing species survival into a single measurable quantity: the competitive balance.

This metric measures vulnerability to competitive exclusion beyond abundance alone by incorporating dispersal-strategy diversity and the structure of interspecific interactions. Greater heterogeneity in dispersal strategies reduces vulnerability to competitive exclusion for a given abundance, and the framework is validated with tropical and temperate forest data.

Human impacts on biodiversity have prompted concerns about a potential mass extinction. Despite signals such as population declines and range contractions, the magnitude and trajectory of ongoing extinctions remain debated. Resolving these uncertainties requires theoretical frameworks that link local population dynamics to spatial community structure across scales and can ultimately be validated with empirical data.

Spatial patterns can be quantified with spatial point pattern analysis and joint species distribution models. Recent work has begun integrating spatial patterns with interaction-based descriptions of communities, but integrating spatial structure into mechanistic models of interacting communities remains challenging.

Analytical results typically become intractable, while numerical simulations depend sensitively on model assumptions and parameterization and offer limited scalability. Data-driven methods combine statistical inference with some mechanistic elements, but often focus on single-species dynamics and typically require detailed demographic and environmental data.

The need is for frameworks that combine spatial structure with multispecies interactions while remaining analytically tractable and compatible with the coarse-grained data available for most species. The framework quantifies each species’ proximity to its competitive-exclusion boundary in metacommunities using a parsimonious ecological model in a heterogeneous spatial environment.

Analytical methods from the physics of disordered systems produce ecologically interpretable metrics for how species exploit locally available space and are affected by community-mediated interactions. Species vulnerability is defined as proximity to the competitive-exclusion threshold while explicitly accounting for ecological context.

The approach does not require previous knowledge of species-specific traits, demographic parameters or interaction coefficients, and can be applied directly to empirical data. This is especially valuable in species-rich communities, where detailed species-level information is often unavailable and might be strongly context-dependent.

Comparison with forest datasets shows that species vulnerability cannot be fully predicted by abundance alone. The framework quantifies how the entanglement among dispersal strategies, spatial distribution patterns and total population size determines proximity to exclusion from a metacommunity.

These metrics remain interpretable without relying on model specifics, suggesting broader ecological relevance. The study uses a general metacommunity model describing temporal dynamics of species abundances in a habitat composed of colonizable sites, each providing finite available space or a limiting resource.

The sites may correspond to distinct patches, although the framework does not strictly require explicit physical boundaries. The model incorporates death, reproduction, dispersal, colonization and competition for space. Their combined effect is captured by an interaction-dispersal kernel matrix, whose elements provide a coarse-grained description of how species interact and spread across patches.

The model does not explicitly separate local competitive and facilitative interactions; instead, it captures their net effect on dispersal efficacy. The link between model and data is the vacancy-adjusted abundance, or VAA, a metric that emerges from the analytical solution using generalized dynamical mean-field theory.

For a species at a given site, VAA is the fraction of space it occupies divided by the locally available, unoccupied space. This normalization accounts for local space saturation, so VAA depends on how empty a local site is relative to its capacity. A subplot with very low free space amplifies VAA for a species occupying that spot, while larger free space reduces its VAA contribution.

Ecologically, VAA upweights space-limited patches and rescales local relative space occupation to reflect effective colonization effort. For each species, the VAA distribution parameters quantify dispersal heterogeneity and an effective dispersal strength that depends on the surrounding community.

Fitting empirical VAA distributions enables inference of species-specific dispersal strategies without direct measurements of dispersal or interaction parameters. Fitting the theory to data also enables inference of the interaction-dispersal kernel, which encodes species-specific spatial strategies.

The VAA distribution links observable space-use patterns with the underlying interaction-colonization structure and can be estimated from spatial abundance data. In panel c, the colored data bars closely follow the corresponding model curves across the VAA axis, for both the orange and green examples.

Panel d then shows several distinct probability curves for dispersal strength, with their peaks and tails separated across the horizontal scale, indicating visibly different inferred strategies among the tree species. The theory is validated on three forest datasets: Barro Colorado Island in Panama, Pasoh Forest Reserve in Malaysia and Michigan Big Woods in the United States.

Each dataset provides a spatially complete census of all trees, making it a testbed for a framework that models spatially explicit abundances of species competing for shared space. Each plot is partitioned into square subplots with a side length of fifty metres, balancing sufficient sampling within subplots with the need to retain spatial resolution.

The analysis estimates the space used by each species from tree size using the established allometric scaling relation between trunk and crown size. The per-subplot effective total space capacity is inferred with a global fitting procedure, and the empirical VAA histogram is then constructed from its definition.

For three representative species per dataset, theoretical VAA predictions fitted to empirical distributions show excellent agreement and capture the diversity of distributional shapes across species. These differences are naturally accounted for by species-specific dispersal strategies.

Goodness-of-fit analyses cover all analysed species, with explicit visualizations for two hundred and twelve species reported in the supplementary information. The model was tested for robustness by asking whether explicitly incorporating facilitation beyond the competition buffering already encoded in the inferred kernel could improve performance.

Adding facilitation produced only marginal changes in model fit across the three forest datasets. At the spatial and temporal scales considered, the dominant effects of facilitation may already be captured by the inferred kernel parameters.

To test generalization, parameters inferred on half of each forest plot were used to predict the mean fraction of space occupied by each species in the remaining half. This out-of-sample validation produced remarkable agreement for all three datasets, showing that the analytical predictions align with empirical observations.

In panel a, the dark theoretical curves closely trace the species-specific histograms across all three forest datasets, including both sharply concentrated and long-tailed VAA shapes. Panels c and d show the corresponding error distributions: the DMT results cluster near or below the dashed reference, while the negative-binomial comparisons extend to visibly larger errors.

Across all three datasets, the model’s absolute goodness-of-fit is consistent with the reference level expected for a correctly specified model. The species-specific model provides an adequate fit only for Michigan, while the neutral model has substantially larger errors across all datasets, suggesting model misspecification.

The framework consistently outperforms the neutral model across all datasets and improves on the species-specific model in Pasoh and Michigan, with comparable and slightly better performance in Barro Colorado Island. Additional analyses show that reliable parameter estimation is achievable even for smaller datasets and provide lower-bound estimates on data requirements.

Beyond fitting VAA distributions, the model yields two species-specific parameters with clear ecological information: dispersal heterogeneity and effective dispersal strength. In the Barro Colorado Island dataset, parameter values were compared across shade-tolerant, gap-exploiting and intermediate functional guilds.

The guilds showed no statistically significant difference in effective dispersal strength. Gap-exploiting and intermediate species exhibited significantly larger dispersal heterogeneity than shade-tolerant species, and this difference remained significant when all three guilds were treated separately.

An explicit integral equation connects each species’ competitive balance to the VAA distribution parameters of all species in the community. Competitive balances are coupled nonlinearly through a shared term, so community-level coexistence emerges as a collective property rather than a sum of individual contributions.

The same structure enables inference of all competitive balances from data. A species can persist when its competitive balance is greater than one and not too far below an equal-balance condition in which all species share identical balance values.

The theory predicts that competing species can coexist as long as their competitive balances are not too dissimilar, and this prediction aligns closely with empirical observations. Effective dispersal strength and dispersal heterogeneity show a large spread among species, reflecting high variability in effective dispersal strategies.

Competitive balances are different: their values are tightly clustered around the mean despite the broad range of individual species parameters. The closely aligned competitive balances inferred for all species are identified as enabling widespread coexistence in the forest communities.

In panel a, the shaded coexistence hypervolume narrows toward the dashed balanced-competition line, showing that allowable combinations shrink as species’ balances diverge. In panel b, the triangular all-species region sits near the diagonal, while moving away from it leads to regions labelled with fewer surviving species.

Numerical simulations show that species with positive vulnerabilities have stationary abundances approaching zero, consistent with the theory. Away from the persistence threshold, species abundances show a substantial spread, suggesting that low abundance does not necessarily imply high vulnerability.

In all three forest datasets, every species has a vulnerability value below the persistence threshold, reinforcing the conclusion that the observed communities lie within the predicted coexistence region and are stable. Abundance and vulnerability are strongly correlated only for species whose vulnerability is close to zero, near the extinction threshold.

Far from that threshold, average abundance spans a large range, so species with low average abundance are not necessarily at risk of competitive suppression. These findings are consistent across all three datasets and are supported by numerical simulations of the model.

Look first at panel a: the star sits inside the shaded coexistence region, while the dashed boundaries mark where either species would lose persistence. Panel b then shows the practical approximation, projecting the community’s displacement onto each vulnerability axis and comparing it with the corresponding threshold distance.

The theoretical framework was validated against three forest communities. Its metrics can be efficiently inferred from large-scale ecological data using an underlying dynamical metacommunity model. Although derived in a specific mathematical framework, the metrics admit model-independent interpretations.

One limitation is that the vulnerability metric does not provide information about the timescales over which species approach their steady state; it predicts only the final outcome. Numerical simulations showed that these timescales depend strongly on species-specific parameters, with some species stabilizing faster than others.

Transient dynamics are not captured by the vulnerability index and may be crucial in shaping short- and medium-term ecological interactions. The central message is that dispersal heterogeneity and community-wide interactions can buffer species against competitive exclusion, so abundance alone is an incomplete measure of vulnerability.

That matters for assessing persistence in real, species-rich ecosystems.