Multi-habitat landscapes are more diverse and stable with improved function

Conservation policy and landscape management have shifted from protecting individual species and habitats to ecosystem- and landscape-level approaches. Habitat heterogeneity and the number of habitats within a landscape are thought to contribute to species richness and ecosystem functioning, yet a mechanistic understanding of how the number of habitats shapes community structure and function remains limited. This understanding is critical for managing landscape-scale ecosystem services that depend on species interactions, such as pollination and pest control. While ecological networks provide a framework to link biodiversity and function, datasets that span multiple interaction guilds across multiple habitats are scarce, potentially obscuring important cross-habitat cascades and functional effects. Previous efforts have linked networks across habitats or interaction types, but the lack of independent functional measures has hindered mechanistic links to functional outcomes. Moreover, dispersal among habitats can drive redundancy or complementarity in community roles and buffer function against disturbance, potentially differing across trophic groups. It remains unclear whether landscapes are simply the sum of their habitat parts or exhibit emergent properties—such as increased stability or function—that cannot be explained by component habitats alone. To address this, the study evaluates how increasing the number of habitats within a fixed area influences biodiversity, network structure, stability (robustness to species loss) and pollination function across multiple interaction types in replicated landscapes.

The study situates itself within a body of work advocating landscape-level conservation and the role of habitat heterogeneity in promoting biodiversity and function. Prior research has highlighted: (1) benefits of habitat diversity and crop heterogeneity for multi-trophic diversity and ecosystem stability, particularly in agricultural landscapes; (2) impacts of landscape simplification on trait filtering and biotic homogenization; (3) the use of ecological network approaches to bridge biodiversity changes and ecosystem function; and (4) emerging attempts to link networks across habitats and interaction types. Theoretical and empirical work on spatial insurance posits that dispersal can stabilize ecosystem function via redundancy and complementarity among species, with differential responses across trophic levels. However, previous network studies often lacked independent functional measures, limiting mechanistic inference about how network structure (e.g., evenness, complementarity, robustness) translates into ecosystem services such as pollination. This study addresses those gaps by combining multi-guild, cross-habitat network data with a manipulative field assay of pollination and modeling to test for emergent properties beyond additive effects of component habitats.

Study design: Thirty independent 9-ha field sites in southwest United Kingdom were selected and standardized in area but varied in the number of constituent habitats: ten monads (single 9-ha habitat), ten dyads (two 4.5-ha habitats), and ten triads (three 3-ha habitats). Habitats were chosen from six types (grassland, heathland, woodland, salt marsh, sand dune, scrub) to balance replication across multi-habitat configurations and avoid confounding habitat identity effects. Data collection spanned two years.

Community and interaction sampling: Across sites, the authors recorded 11,482 interactions involving 154 plant species and 954 insect species, comprising 5,729 flower-visitor interactions, 2,345 plant–leaf miner interactions, 697 plant–caterpillar interactions, 1,240 plant–seed-feeding interactions, and 1,471 herbivore–parasitoid interactions. Species were represented as nodes and interactions as links within multilayer networks, with layers corresponding to habitats. Metrics quantified included floral abundance, plant species richness, insect abundance, insect species richness, insect species evenness, and interaction evenness.

Statistical analyses of diversity and structure: A MANOVA tested for multivariate differences among monads, dyads, and triads, followed by pairwise MANOVAs and generalized linear models (GLMs) to identify drivers of observed differences in individual metrics.

Robustness modeling: Community stability was measured as network robustness to species loss via simulations that removed plant species from least to most abundant (mimicking bottom-up habitat degradation, reflecting higher extinction risk of rare species). The model allowed rewiring, enabling species to reallocate interactions after resource loss, and accounted for shared species across interaction types (e.g., ontogenetic shifts from herbivorous caterpillar to pollinating butterfly). Robustness was evaluated across landscape types and under varying diet flexibility and extinction thresholds (25%, 50%, 75%). Variability in robustness was compared using Brown–Forsythe tests; random species removal was also analyzed as a comparator.

Manipulative field experiment (pollination function): To quantify pollination function, 20 potted, wild-type Fragaria vesca plants grown under standardized conditions were placed at the center of each monad and triad at the onset of flowering. Plants remained in situ for 14 days for natural pollination, then were transferred to a pollinator-free greenhouse for 28 days to ripen. Fruits were weighed and graded as Class I (perfectly symmetrical) or Class II (otherwise). Differences in fruit weight and Class I proportion between monads and triads were assessed, as was variance homogeneity (Brown–Forsythe).

Interaction complementarity: For pollinator communities, interaction complementarity was quantified as dietary dissimilarity among all recorded flower visitors. Principal coordinate analysis (PCoA) was used to derive community dispersion as a measure of the breadth of dietary dissimilarity at each site, with higher dispersion indicating greater complementarity. Robustness to rare or undersampled species was assessed via sensitivity analyses.

Null models for emergent properties: To test whether triad properties were additive versus emergent, for each empirical triad the authors generated 1,000 null triads by aggregating independent monad observations while preserving the total number of sampled interactions, and then additionally controlling for the number of plant species. They compared interaction evenness and pollinator interaction complementarity (functional dispersion) between empirical and null triads. Relationships with plant phylogenetic diversity and sampling completeness were examined to interpret observed differences.

• Community structure: Landscapes with more habitats exhibited higher insect species evenness and trends toward higher plant species richness and interaction evenness. MANOVA indicated significant overall community differences among monads, dyads, and triads (MANOVA F ≈ 285.366; P=0.001). Pairwise MANOVAs showed significant increases from monads to triads (F1,18=5.552; P=0.005). GLMs indicated:

  • Plant species richness (mean ± s.d.): monads 41.4 ± 20.77, dyads 43.3 ± 22.5, triads 59.1 ± 22.99; F1,28=3.244; P=0.082 (non-significant increase).
  • Interaction evenness: monads 0.48 ± 0.07, dyads 0.52 ± 0.06, triads 0.52 ± 0.03; F1,28=3.767; P=0.062 (non-significant increase).
  • Insect species evenness: monads 0.71 ± 0.11, dyads 0.81 ± 0.06, triads 0.83 ± 0.04; F1,28=14.92; P<0.001 (significant increase).
  • Floral abundance, insect species richness, and insect abundance: no significant differences (all F1,28<0.714; P>0.405).

• Robustness to species loss: Mean robustness did not differ among monads, dyads, and triads (F2,27=0.183; P=0.83), but variability in robustness decreased significantly with more habitats: IQRs were 0.105 (monads), 0.064 (dyads), 0.047 (triads) (Brown–Forsythe F2,1499=1,272.9; P<0.001). The trend strengthened under more flexible rewiring rules and was stronger for rare-to-common species removal than for random removal, suggesting a role of the distribution of rare species across habitats rather than species loss per se.

• Pollination function: Strawberries from triads were 30.3% more frequently Class I (symmetrical) than those from monads (t113=3.263; P=0.007), with more consistent outcomes at triads (Brown–Forsythe F1,16=10.65; P=0.007). Fruit weights did not differ (F1,16=0.091; P=0.122). Although pollinator abundance and richness did not increase at triads, pollinator interaction complementarity (dietary dissimilarity) was higher at triads (t=8.42; P<0.001). Interaction complementarity predicted the proportion of Class I fruits (P=0.045) but not fruit weight (P=0.342).

• Emergent properties: Empirical triads typically exhibited higher interaction evenness (7/10 sites) and lower interaction complementarity (7/10 sites) than null triads constructed additively from monads while preserving interaction counts; these differences persisted, though weakened, when also controlling for plant species richness. Plant phylogenetic diversity was positively correlated with interaction complementarity (repeated-measures correlation r≈0.158; P<0.001) and negatively with interaction evenness (r≈−0.238 to −0.201; P<0.001). Constraining models for equal sampling completeness produced qualitatively similar trends.

The study demonstrates that increasing the number of habitats within a landscape yields communities with higher evenness and greater interaction complementarity, leading to improved and more consistent ecosystem function (pollination quality) and decreased variability in robustness to species loss. Crucially, these outcomes are not fully explained by additive effects of component habitats; rather, empirical triads show emergent properties in interaction evenness and complementarity relative to null expectations. The reduced variability in robustness in multi-habitat landscapes suggests a buffering effect at the landscape scale, where robustness in some habitats compensates for lower robustness in others, thereby lowering the risk of cascade failures in any single site. The modeling indicates that this stabilizing effect strengthens when species can rewire interactions (diet flexibility), highlighting the potential role of behavioral and trophic plasticity in community responses to environmental stressors. The link from landscape composition to pollination function is mediated by community interaction structure: triads harbor pollinator assemblages with more complementary diets, which in turn increase the proportion of high-quality fruits, independent of pollinator abundance or richness. Plant phylogenetic diversity appears to tune these relationships, correlating positively with complementarity and negatively with interaction evenness. These findings underscore the importance of maintaining multiple, contiguous natural habitats to support stable, functional landscape-scale ecosystems and associated services.

Multi-habitat landscapes support communities with higher insect species evenness, greater interaction complementarity, more consistent robustness to species loss, and improved pollination quality. These benefits are not simply additive combinations of single-habitat communities but reflect emergent properties likely driven by landscape-scale environmental heterogeneity and facilitated by species’ capacity to rewire interactions. The work provides a mechanistic link from landscape composition, through community network structure, to ecosystem function. Management should prioritize maintaining diverse, connected habitat mosaics to bolster ecosystem services and resilience, particularly under global change. Future research should: (1) replicate specific habitat combinations to enable formal path analyses of causal mechanisms; (2) incorporate temporal dynamics to assess stability components beyond spatial variability; (3) explore how species’ trait distributions and phylogenetic diversity modulate complementarity and robustness; and (4) test generality across regions, habitat types, and other ecosystem functions beyond pollination.

The robustness analysis focused on spatial variability and did not include temporal dynamics, limiting inference about temporal stability. The design did not include large numbers of replicates for each specific habitat combination, precluding detailed path analyses to disentangle mechanistic pathways. One author affiliation (superscript 11) was not specified in the provided text. Although the authors addressed potential confounders such as surrounding patch size and sampling completeness, residual differences in sampling completeness and plant phylogenetic diversity could influence interaction metrics. Null model assumptions and the operationalization of rewiring and extinction thresholds may affect robustness estimates, and generality beyond the study region and habitat pool remains to be tested.