Conceptual framework illustrating hypothesized mechanisms linking urban affinity to interspecific body-size shifts.

These include dispersal and mobility constraints under habitat fragmentation44,45, thermophily and the temperature–size rule driven by the urban heat island effect15,30, size-biased competition and survival94,95, and size-biased human preferences64. Urban fragmentation of habitat resources can select for increased mobility (e.g., larger butterflies) or reduced mobility (e.g., larger seeds) depending on isolation severity. Elevated urban temperatures favor thermophily, which often negatively correlates with size as it affects the heat balance via thermal inertia. Similarly, these higher temperatures generally favor smaller-bodied adult ectotherms because they accelerate development and reduce time available for growth (i.e., temperature-size rule). In plants, the increased CO₂ and nutrient availability associated with anthropogenic environments—due to heating- and traffic-related CO2 emissions and eutrophication—provides a competitive advantage to larger plant species, and human preferences too may favor larger species (e.g., tree-lined streets), whereas smaller species may be advantaged in colonizing built infrastructure.

Species urbanness distributions (SUDs) exemplified for eight subrealms.

Plotted are all species per subrealm (A), with the images highlighting an example ‘hyper-exploiter’ species from each of these subrealms (i.e., with a high urban affinity score). The x-axis shows the urban affinity measure whereas the y-axis is the number of species within that bin. There were consistent patterns for kingdoms, classes, and orders (B) as shown by similar central tendencies despite variation in distributional shape. The vertical dashed line represents where species are neutral towards urbanization. Photos acquired from iNaturalist CC BY-NC and background was removed by authors: Brown rat (© Ouwesok), Monk parakeet (© Juan Emilio), Cape dwarf chameleon (© Berkeley Lumb), American cockroach (© Len Worthington), Flaming kay (© Lyubo Gadzhev), Juno silverspot (© Rigoberto Ramírez Cortés), Hibiscus harlequin bug (© Sam Fraser-Smith), and Mascarene island leaf-flower (© Douglas Goldman).

Effect sizes between body size and urban affinity across the tree of life and individual effect sizes for animals and plants families.

(A) Effect sizes of the relationship between body size and urban affinity for the 371 families included in our analysis, plotted along a phylogenetic tree of life (see Methods); Plantae are highlighted and shaded in green. Colors indicate the direction of the effect: orange indicates negative, petrol indicates positive, grey indicates neutral (i.e. any effect sizes between -0.05 and 0.05). (B) and (C) histograms of individual effect sizes for each family, for animals (B) and plants (C). Orders are shown along the outside edge of the phylogenetic tree, each with a bar and icon, for any order with more than 3 families. An interactive version for full exploration of our results at both family and order level is available here.

We used subrealms as our geographical aggregation.

Subrealms were quantified from aggregating bioregions as identified by One Earth (see more here: https://www.oneearth.org/bioregions/). The subrealms level was chosen after exploring the tradeoff between accounting for geographic differences in urban affinity and the number of species that could be included.

Violin plots showing the distribution of urban affinity values by kingdom across the 12 subrealms with the greatest sample sizes.

Violin plots showing the distribution of urban affinity values by class across the 12 subrealms with the greatest sample size.

Only classes represented by ≥ 50 species within a subrealm are included.

Violin plots showing the distribution of urban affinity values by order across the 9 subrealms with the greatest sample size.

Only orders represented by ≥ 50 species within a subrealm are included.

An illustration of the species-specific distribution of observations and the VIIRS values, in average radiance, of those observations for six different species within the Great European Forests subrealm.

Of note is that the urban affinity is not shown, as some are negative values, but these values are shown in the text.

The distribution of Magnolia warbler Setophaga magnolia observations and the VIIRS values, in average radiance, of those observations in four different subrealms.

Shown using dashed lines are the species-specific mean (green), the subrealm-specific mean of VIIRS (violet), and the resulting urban affinity measure (yellow).

The total number of species for which we calculated urban affinity scores, stratified by subrealm and shown separately for Animalia (top) and Plantae (bottom).

The subrealms with their names are shown in Figure S1. The dataset of species’ urban affinity scores is provided in Table S1.

The total number of subrealms for which a species had an urban affinity score.

The majority of species (64%) only had an urban affinity score from one subrealm, but the range was from 1 to 34.

An illustrative example, showing the relative urban affinity scores for Hymenoptera in the Northeast American Forests subrealm.

The plot is for illustrative purposes and the values for each species can be found in Data S1.

An illustrative example, showing the relative urban affinity scores for Lepidoptera in the Southeast Asian Forests subrealm.

The plot is for illustrative purposes and the values for each species can be found in Data S1.

An illustrative example, showing the relative urban affinity scores for Asterales in the Scandinavia & West Boreal Forests subrealm.

The plot is for illustrative purposes and the values for each species can be found in Data S1.

A summary of the number of species, shown per class, gained through taxonomic harmonization, and therefore included in the analysis dataset.

The number of taxa, on a log10-transformed scale.

These can be found in Table S3.

We chose to model each taxonomic group (i.e., family, order, class, phylum, and kingdom) independently of one another to avoid the influence of where individual families could have no effect but result in a positive effect if modeled jointly.

The top panel shows simulated raw data for two families; the middle panel shows the posterior distribution of a bayesian model fit for each of these families separately; the bottom panel shows the posterior distribution of a bayesian model fit for a model including family as a random effect and for a model not including family as a random effect.

An illustrative example showing the influence of using a random slope for body size, by showing the (A) Interaction between body size (scaled and log10 transformed) and body size measurement type (metadata; see Methods for details) for the family Accipitridae.

Facets represent different types of body size metrics used in the dataset, labeled with a letter identifier and corresponding sample size (n). Lines represent predicted urban affinity based on body size, with 95% credible intervals shown as shaded ribbons. These are extracted from a model fit with an interaction between metadata and body size to purposefully investigate the influence of metadata and the relationship with body size. (B) Posterior distributions of the slope for body size (scaled and log10 transformed) under two model structures: V1 includes a random slope for body size by subrealm and a random intercept for metadata (as presented in main results), while V2 adds a random slope for body size by metadata. The inclusion of this additional random slope in V2 increases uncertainty and pulls slope estimates toward zero, particularly when data are sparse within body size measurement types as illustrated here. Also see Fig. S15.

An illustrative example showing the influence of using a random slope for body size, by showing the (A) Interaction between body size (scaled and log10 transformed) and body size measurement type (metadata; see Methods for details) for the family Apidae.

Facets represent different types of body size metrics used in the dataset, labeled with a letter identifier and corresponding sample size (n). Lines represent predicted urban affinity based on body size, with 95% credible intervals shown as shaded ribbons. These are extracted from a model fit with an interaction between metadata and body size to purposefully investigate the influence of metadata and the relationship with body size. (B) Posterior distributions of the slope for body size (scaled and log10 transformed) under two model structures: V1 includes a random slope for body size by subrealm and a random intercept for metadata (as presented in main results), while V2 adds a random slope for body size by metadata. The inclusion of this additional random slope in V2 increases uncertainty and pulls slope estimates toward zero, particularly when data are sparse within body size measurement types as illustrated here. Also see Fig. S14.

A list of 41 potential ‘types’ of body size that were used for potential inclusion in our body size dataset.

We aimed to incorporate as many types of body size measures as possible and were not restrictive in our searching for body size measures.