Urban fragmentation leads to lower floral diversity, with knock-on impacts on bee biodiversity

Urbanisation drives habitat fragmentation, degradation and loss, with pervasive effects on biodiversity, including bees and flowering plants. Pollinators rely on floral resources, and many plants depend on pollinators, creating coupled communities whose diversity often correlates. However, the direction and mechanisms of bee–plant diversity relationships within urban landscapes remain unresolved. This study investigates whether urban fragmentation directly and indirectly shapes bee and flowering plant species richness, functional diversity, and phylogenetic diversity in urban green spaces. The authors test for co-variation in diversity, evaluate trait-based matching (e.g., nectar holder depth vs bee size), and use structural equation modelling to infer causality—predicting negative effects of fragmentation on both partners and a bottom-up pathway (plants influencing bees more than bees influencing plants).

Prior work shows strong positive correlations between flowering plant and pollinator species richness, and parallel declines in insect pollinators and insect-pollinated plants, suggesting coupled dynamics though causality is unclear. Increased floral richness can stabilize pollinator communities and reduce extinction risk; conversely, pollinator diversity can influence plant diversity, density, reproduction, and seedling dynamics. Plant–pollinator networks are structured by trait matching, implying potential links between the functional and phylogenetic diversity of partners. Urbanisation has mixed effects on bee richness (negative, neutral, or positive) and can favor particular traits (parasitic, social, cavity-nesting, generalist, large-bodied, and exotic bees). Urban effects on plants are also mixed; German cities can be rich in native plant species, yet urbanization can alter plant functional traits and increase phylogenetic diversity of certain urban-adapted groups, with shifts towards biennial, self- or wind-pollinated, scleromorphic/succulent leaves, and exotics. Despite these findings, how bee and floral diversity co-vary within urban settings, and the causal direction of their relationship, remains insufficiently understood. Simulation studies suggest mutualistic networks are more robust to animal than plant loss, implying stronger bottom-up control (plants → pollinators).

Study area: Halle (Saale), Germany (~240,000 inhabitants; ~13,000 ha), with heterogeneous land uses (paved 37.7%, agriculture 25.6%, forest 16.9%, parks 14.1%). Eight urban semi-natural, low-management sites were selected across a gradient of surrounding urban cover, ≥1.5 km apart. Sampling design: Bees and flowering plants were sampled via variable transect walks (30 min) at each site, twice per month (morning and afternoon) from June 10 to August 18, 2017, totaling six visits per site and 1440 min. Bees visiting flowers and the visited flowers were collected. Sampling occurred under standardized favorable weather conditions and by the same collector. Specimen processing: Bees were preserved in ethanol and identified to species using keys and COI DNA barcoding; flowering plants were identified to species using local floras. Bee specimens are housed at Martin-Luther University; COI sequences were deposited in GenBank (MW065567–MW065776). Diversity metrics: For both bees and plants, computed abundance, species richness, functional diversity, and phylogenetic diversity. Functional traits: Bees—intertegular distance (ITD; body size), sociality, tongue length, nesting behavior, voltinism, lecty; Plants—nectar holder depth (mm), flower restrictiveness/shape, color, breeding system, flower sex timing, plant longevity. Community weighted means (CWM) for bee ITD and plant nectar holder depth were calculated. Functional diversity was quantified using standardized Rao’s Q via a null model (1000 randomizations controlling for richness). Phylogenetic diversity (PD) used Faith’s PD based on gene trees (bees: COI, EF1-α; plants: rbcL, matK), standardized against null expectations. Environmental variables: Local (patch) variables included percent exposed (bare) soil from ten 1 m² random quadrats per site per visit (proxy for ground-nesting resources) and patch size. Landscape heterogeneity was quantified at multiple radii (250–1500 m) using QGIS and Geofabrik land cover; Shannon diversity correlations identified 1000 m as the most explanatory scale. Metrics included proportions of parks, forest, allotments, semi-natural areas, agriculture, residential cover; edge density; and fragmentation (number of disconnected green patches divided by total green area). Statistical analysis: Linear mixed-effects models (LMMs) for functional and phylogenetic diversity and Poisson GLMMs for abundance and richness, with month and site as random effects. Bee abundance was included as a covariate in richness models to control for sampling effort. Automated model selection (all subsets) with AICc and model averaging (ΔAICc ≤ 2, up to three predictors) identified parsimonious predictors. Fourth-corner analysis (mvabund, negative binomial, LASSO penalty, 1000 resamples) assessed trait–environment associations for bee lecty, nesting, and body size. Piecewise structural equation modelling (piecewiseSEM) tested causal pathways among fragmentation, bare soil, plant richness, and bee richness/abundance; model fit via Fisher’s C and AIC, with Sobel tests for indirect effects. Predictors were standardized, VIF < 3, and residual spatial autocorrelation (Moran’s I) was absent.

• Sampling overview: 845 wild bees (63 species, 23 genera, 6 families) from 1440 min transects; additional 291 honey bees recorded (not influential in models). Bee traits: 71% ground-nesting, 67% small-bodied, 65% solitary (by species) though social individuals were 52%. Plants: 58 flowering species; most open flower shape (62%), yellow color most common (38%), majority perennials (66%) and allogamous (66%).
• Bee diversity drivers:
  • Species richness increased with flowering plant richness (β=0.148, P=0.021).
  • Abundance increased with flowering plant richness (β=0.266, P=0.001) and percent bare soil (β=0.186, P=0.017).
  • Functional diversity decreased with higher plant CWM nectar holder depth (β=−0.462, P=0.023).
  • Community mean bee body size (ITD) increased with plant CWM nectar holder depth (P<0.001) and with local flower richness (significant in model set); bee phylogenetic diversity had no significant predictors (intercept-only best model).
• Plant community drivers:
  • Species richness increased with bee abundance (β=1.611, P<0.001) and decreased with fragmentation (β=−0.866, P=0.019).
  • Functional diversity decreased with fragmentation (β=−1.076, P<0.001).
  • Phylogenetic diversity decreased with fragmentation (β=−0.698, P=0.002).
  • Plant CWM nectar holder depth increased with bee CWM body size (β=2.136, P<0.001); fragmentation not significant here.
• Trait–environment associations (fourth-corner):
  • Oligolectic bee abundance: positive with residential cover; negative with plant CWM nectar depth.
  • Polylectic abundance: positive with local plant richness, allotment gardens, and parks.
  • Above-ground nesters: positive with residential cover.
  • Ground nesters: positive with local plant richness and percent bare soil.
  • Bee body size: positive with local plant richness and plant CWM nectar depth.
• Structural equation modelling (best model fit: AIC=33.387; Fisher’s C=9.387, df=8, P=0.311):
  • Urban fragmentation negatively affected flowering plant richness (standardized path −0.408, P=0.02; R² plants=0.38).
  • Flowering plant richness positively affected bee richness (0.148, P=0.02; R² bee richness=0.17) and bee abundance (0.151, P<0.01), alongside a positive effect of bare soil on bee abundance (0.229, P<0.001; R² bee abundance=0.88).
  • Indirect effects: fragmentation had negative indirect effects on bees via plants (Sobel: bee richness −0.06, P=0.04; bee abundance −0.09, P=0.02).
  • Alternative top-down SEM (bees → plants) had higher AIC by 4.941, indicating less support for that direction of causality.

Findings demonstrate strong coupling between floral resources and bee communities in urban semi-natural patches, with clear bottom-up control: plant richness increases bee richness and abundance, while urban fragmentation reduces plant richness, functional diversity, and phylogenetic diversity. This plant-mediated pathway leads to indirect negative effects of fragmentation on bees. Local nesting resources (bare soil) further enhance bee abundance, underscoring the joint importance of food and nesting resources. Trait-based results indicate that increasing nectar holder depth filters bee communities toward larger, long-tongued, often social species, reducing overall bee functional diversity; size-related trait matching thus plays a dominant role in structuring bee–flower community assembly. Bee responses also depended on landscape context and traits: residential areas, parks, and allotments supported various functional groups, while ground-nesting bees benefited from bare soil and local floral richness. The absence of direct fragmentation effects on bees likely reflects bees’ mobility and ability to utilize resource mosaics, yet declines in floral diversity due to fragmentation can still propagate to bees. Plants appear less directly dependent on pollinators in this urban context—potentially due to alternative reproductive strategies—reinforcing the asymmetry in dependency within these mutualisms.

This study shows that urban fragmentation reduces flowering plant richness and both functional and phylogenetic diversity, which in turn indirectly diminishes bee richness and abundance, evidencing bottom-up control in urban plant–pollinator communities. Trait matching centered on nectar holder depth and bee body size further shapes community composition, favoring larger, long-tongued bees as corolla depth increases. From a management perspective, enhancing local native floral diversity and nesting resources (e.g., maintaining bare soil patches) and supporting diverse green spaces (parks, allotments, residential gardens) can bolster urban bee communities. Reducing fragmentation and improving connectivity of green patches should help sustain floral diversity and, consequently, pollinators. Future research should extend across multiple cities and seasons, integrate manipulative experiments to test causality, and assess long-term dynamics and network-level functional outcomes under varying urban design and management scenarios.

• Spatial and temporal scope: eight sites within a single city (Halle, Germany) sampled over one summer (June–August 2017), which may limit generalizability and seasonal coverage of bee–plant dynamics.
• Observational design: causal inference via SEM, while informative, is based on observational data and cannot fully exclude unmeasured confounders.
• Trait set and metrics: bee functional and phylogenetic diversity predictors were limited; bee phylogenetic diversity lacked significant predictors in this dataset.
• Scale selection: landscape effects were analyzed at 1000 m radius chosen by correlation peaks; responses at other scales or in different urban contexts may differ.
• Honey bee abundance was included as a covariate but not influential; potential competitive effects may vary in other settings.