Global effects of land-use intensity on local pollinator biodiversity

The study addresses the global question of how land-use type and land-use intensity affect local pollinator biodiversity across taxa and regions. Motivated by widespread reports of pollinator declines and the high economic and food security value of animal pollination, the authors highlight gaps in the evidence base due to geographic and taxonomic bias and methodological differences. They propose a global, multi-taxon synthesis to test: (1) overall effects of land-use intensity (LUI) on pollinator biodiversity; (2) whether responses within cropland differ between tropical and non-tropical zones; and (3) how responses vary among taxonomic groups and biodiversity metrics. The work aims to inform policy initiatives on pollinator conservation by providing globally representative, local-scale analyses.

Prior work links pollinator declines to land-cover and land-use intensification and climate change, with much evidence biased toward Europe and North America. Landscape composition and configuration (distance to natural/managed land, fragmentation, edge density) influence pollinators, and chemical inputs (pesticides, fungicides, herbicides, fertilizers) often typify intensity effects. Neonicotinoids are associated with bee declines and altered behaviour; fungicides and herbicides can increase pesticide toxicity and reduce floral diversity; nitrogen fertilizers reduce plant diversity and alter rotations, lowering floral resources. Responses vary by taxon and traits (dietary specialization, mobility, nesting), and some pollinators benefit in semi-natural or urban habitats due to floral availability. Sensitivity may differ between tropical and non-tropical regions due to longer land-use history in temperate zones (extinction filtering), higher tropical climate sensitivity, and greater functional specialization in tropical pollination systems. Comprehensive, multi-metric, multi-taxon, globally representative analyses have been lacking.

Design: Space-for-time global synthesis using local site-level data to model pollinator biodiversity responses to land-use type and intensity. Data sources: (1) PREDICTS database (site-level ecological surveys with land-use type and intensity categories). (2) Newly compiled global list of likely animal pollinators built via semi-automatic text mining of Scopus abstracts containing 'pollinat*' and animal binomials, manual verification, extrapolation to higher taxa where appropriate, and expert review by seven pollination ecologists. Pollinator dataset construction: Initial automated genus scraping using Taxonfinder and Neti Neti (taxize R package), confirmation through abstract/full-text evidence graded at confidence levels 1–4 (from experimental pollination to non-destructive visitation). Extrapolation to families/subfamilies/tribes where evidence suggested all members pollinate; exclusion of groups with predatory or destructive floral visitation (e.g., ants, crab spiders). Merge confirmed and extrapolated pollinator taxa with PREDICTS species records, retaining only sites with at least one pollinator record. Expert curation added/removed taxa. Final dataset: 4502 pollinating species (3862 invertebrates, 640 vertebrates) across 303 studies and 12,170 sites; 8639 sites retained where both land-use type and intensity were known. Site coverage spanned all continents, highest in Europe, North America, and Africa. Response variables: Site-level species richness (raw and Chao1-estimated), total abundance (effort-adjusted), Simpson diversity (reciprocal index). Abundances divided by rescaled sampling effort per study. Non-integer values allowed due to standardization and some density measures. Predictors: Land-use type (primary, mature/intermediate/young secondary vegetation, plantation forest, pasture, cropland, urban); land-use intensity (minimal, low, high). Combined as LUI factor following De Palma et al. Additional analyses focused on croplands: intensity by geographic zone (tropical vs non-tropical), and by taxonomic order. Continuous intensity proxies included total fertilizer application rate (EarthStat; N, P, K on 17 major crops; 5×5 min resolution) and, in supplementary analyses, pesticide application (PEST-CHEMGRIDS). Models: Generalized linear mixed-effects models (GLMM Poisson) for species richness (and Chao1), linear mixed-effects models (LMM) for log-transformed total abundance and Simpson diversity (log(x+1)). Random effects: study identity (methodological/geographic variation), spatial block within study (spatial structure); for richness, an observation-level random intercept for site to account for overdispersion. Random structures selected to minimize AIC. Models compared to intercept-only; models with AIC greater than null excluded. Effects expressed relative to primary vegetation with minimal use baseline. Fixed-effect uncertainty propagated by sampling from variance-covariance matrices to compute percentage changes and 95% intervals. Validation and sensitivity: Tested for spatial autocorrelation in residuals (Moran's I); compared LMM to zero-inflated negative binomial for abundance; added climatic covariates (WorldClim max temp of hottest month, precipitation of wettest month); jack-knifed by continent; compared raw vs Chao1 richness. For croplands, also analyzed baseline forest cover strata (≤40% vs ≥60% forest), balanced resampling across zones (1000 sites each, replicated), and restriction to main crop-pollinating groups. Fertilizer models included interactions with zone or taxonomic order; similar structure for pesticide models. All analyses in R v4.0.3.

• Overall LUI effects: Increasing land-use intensity significantly altered pollinator biodiversity (species richness F=9.4384; total abundance F=4.8075, p<0.01; Simpson diversity F=11.6691, p<0.01). Land-use type was also significant (species richness F=8.9440; total abundance F=8.0346, p<0.01; Simpson diversity F=4.4150, p<0.01), with stronger declines within a land-use type as intensity increased.
• Relative to primary minimal-use baseline, low-intensity levels often had higher biodiversity across many land-use types; cropland and young secondary vegetation were exceptions. With the exception of cropland and young secondary vegetation, at least one of low or intermediate intensity showed significantly higher richness and/or abundance than the baseline.
• Urban areas showed the strongest intensity-related declines: from minimal to intense use, species richness decreased by 43% and total abundance by 62%. Plantation forests had a 38% decline in species richness. Pasture showed a 75% decline in total abundance with increasing intensity, while species richness did not significantly decline. Cropland did not show significant richness or abundance declines across intensity in the global LUI model. Young secondary vegetation richness declined 16% from minimal to high intensity; mature and intermediate secondary vegetation showed no significant differences across intensity.
• Croplands by geographic zone: In non-tropical zones, cropland species richness and abundance did not differ significantly among intensities and were significantly higher in minimal-intensity cropland than in primary vegetation. In tropical zones, species richness and total abundance decreased by 44% and 49%, respectively, from primary vegetation to high-intensity cropland. These patterns were robust to baseline forest cover differences and not attributable to sample size disparities.
• Taxonomic responses in croplands: Invertebrate orders showed strong abundance declines at high intensity relative to primary vegetation (>70% for all orders; up to ~80% for Lepidoptera and Diptera). Lepidoptera exhibited consistent negative responses across richness, abundance, and Simpson diversity along the intensity gradient. Diptera showed higher richness and abundance at intermediate vs minimal cropland intensity but overall negative response to the combined LUI measure. Among vertebrates, Apodiformes declined by at least 20% from medium-intensity cropland to primary vegetation across all metrics (no high-intensity sampling), and Passeriformes declined 30% (abundance), 36% (richness), and 26% (Simpson diversity) from baseline to high-intensity cropland.
• Fertilizer application (croplands): Responses differed by taxon. Hymenoptera and Lepidoptera declined strongly with increasing total fertilizer application rate (e.g., +1000 kg/ha associated with −44% Hymenoptera abundance and −50% Lepidoptera abundance), while Diptera increased markedly (+760% richness; +374% abundance). Coleoptera responses were not significant across metrics. Among vertebrates, Apodiformes richness increased by 163%, whereas Passeriformes showed no marked change. Zone-by-fertilizer interactions were not significant for richness, abundance, or Simpson diversity.
• Additional robustness: Zero-inflated models, inclusion of climatic covariates, continent jack-knifing, and abundance-controlled richness (Chao1) yielded qualitatively similar conclusions. Some Simpson diversity models were excluded due to AIC > null.

The findings demonstrate that land-use intensity drives significant, context-dependent changes in local pollinator biodiversity. Low to intermediate intensity can support higher biodiversity than minimally used primary vegetation in several land-use types, highlighting potential benefits of moderate disturbance and habitat heterogeneity, notably in low-intensity urban settings. However, intensification within anthropogenic land uses generally reduces pollinator richness and abundance, with the strongest effects observed in urban, plantation forest, and pasture systems. In croplands, the pronounced sensitivity of tropical pollinators to intensification contrasts with the relative stability or increases seen in non-tropical regions, likely reflecting the interplay of historical land-use filtering and higher climate sensitivity in the tropics. Taxonomic differences underscore trait-mediated responses: specialist and less mobile groups (e.g., Lepidoptera, many Hymenoptera) are more negatively affected, whereas Diptera often show resilience or increases with fertilizer inputs, possibly linked to larval habitat preferences and lower specialization. Vertebrate pollinators appear less sensitive, potentially due to greater body size and mobility, but their increases do not offset losses in key invertebrate pollinators. These patterns imply risks to pollination services and crop yields, particularly in tropical agriculture, where gains in resilient taxa may not compensate for declines in primary crop pollinators. The study advances a globally representative, multi-taxon perspective that refines understanding of how intensity, land-use type, geography, and taxa interact to shape pollinator biodiversity.

This study provides the largest global, local-scale assessment to date of how land-use type and intensity affect pollinator biodiversity across taxa and regions. It reveals that while low-intensity use can enhance biodiversity relative to primary baselines in several land uses, intensification generally reduces pollinator richness and abundance, with especially strong negative effects in tropical croplands and among invertebrate taxa such as Lepidoptera and Hymenoptera. Differential taxonomic responses to specific intensity components (e.g., fertilizer) further highlight the need for targeted management. Given the central role of animal pollination in crop production, these changes threaten pollination services, particularly in the tropics. Future research should disentangle the relative contributions of historical land use and climate change, incorporate finer-scale and multi-faceted intensity metrics (including pesticides) globally, expand temporal datasets to capture extinction debt, and improve trait-based and life-stage-specific understanding across underrepresented taxa and regions to guide conservation and sustainable land-use strategies.

• Space-for-time inference may miss extinction debts; long-term time series are limited globally.
• Primary vegetation baselines vary (e.g., forest cover differences between tropics and non-tropics) affecting comparisons; although baseline forest cover stratification did not change overall conclusions, magnitude of differences can vary.
• Low explanatory power with substantial variance explained by random effects; analyses focus on general patterns rather than site-specific prediction.
• Species richness may be confounded by abundance; however, Chao1 estimates aligned with raw richness.
• Spatial bias toward non-tropical regions (Europe, North America); overall impacts may be underestimated given stronger tropical declines. Jack-knife and resampling suggest robustness of main patterns.
• Fertilizer data are coarse (5×5 min), potentially mismatched to pollinator response scales; local-scale intensity measures could alter inferences.
• Extrapolation from genera/family-level evidence may not represent all species or life stages; some included taxa may have variable pollination roles.
• Evidence of pollination on specific plants may not generalize to all flowering plants or crops; functional contribution to crop pollination varies by taxon.