Species diversity and food web structure jointly shape natural biological control in agricultural landscapes

The study investigates how landscape composition influences parasitoid-driven biological control and hyperparasitism in agricultural systems by explicitly considering species diversity and food web structure. Biodiversity loss due to land-use change and intensification threatens ecosystem services such as pollination and pest control. While diverse landscapes can buffer species loss, responses of predators and parasitoids to landscape composition are inconsistent, leading to variable outcomes for ecosystem services and disservices. Food web metrics, notably generality (mean number of hosts per consumer) and vulnerability (mean number of consumers per host), can mediate these outcomes. The research focuses on cotton agro-landscapes in northern China, where Aphis gossypii is a key pest and hymenopteran parasitoids provide biological control. The authors aim to disentangle how landscape composition affects (1) biodiversity of ES and EDS providers, (2) food web structure, and (3) resulting ecosystem functionality (parasitism and hyperparasitism), and to determine cascading effects via structural equation modeling.

The paper synthesizes prior work showing: (i) landscape simplification often reduces insect biodiversity but effects on ES providers (predators/parasitoids) are inconsistent; (ii) agricultural intensification and on-farm practices (plant diversity, tillage, agrochemicals) influence biological control; (iii) landscape composition simultaneously shapes ecosystem disservices such as pest colonization, hyperparasitism, and intraguild predation, which can dampen control; (iv) hyperparasitoids may thrive in complex landscapes and destabilize parasitoid communities; (v) adopting a multitrophic food web perspective can clarify variable landscape–functionality relationships. Food web descriptors such as generality and vulnerability link structure to function: higher generality can buffer species loss by offering multiple hosts to consumers, whereas higher vulnerability indicates shared resources and potential competition leading to fragility. Host–parasitoid networks are well-suited for quantifying interaction structure and have been used to show landscape effects on parasitism and hyperparasitism. However, the mediating role of specific food web features in translating landscape composition to ES/EDS delivery remains underexplored, particularly in dynamic agroecosystems.

Study system and sampling: Over three years (2014–2016), the authors surveyed cotton fields at 25 sites across four regions in northern China. They collected mummified (parasitized) Aphis gossypii from fields to assemble tri-trophic aphid–primary parasitoid–hyperparasitoid networks. In total, 2153 mummified aphids were collected. Species identification and network assembly: DNA-based molecular detection was used to identify primary parasitoids and hyperparasitoids within mummies and to reconstruct quantitative trophic interactions. The assembled network included one aphid pest, three primary parasitoid species, and seven hyperparasitoid species, comprising 2503 parasitoid and hyperparasitoid individuals involved in 2386 interaction events. Landscape variables: Landscape composition metrics around fields included non-crop habitat (NCH) cover, secondary crop (SC) cover, cotton area, maize cover, etc. Landscape predictors were preselected based on correlation screening and principal component analysis (PCA). Food web and diversity metrics: Quantitative food web indices were calculated: generality (Gq), vulnerability (Vq), and connectance (Cq). Biodiversity metrics included species richness and Shannon diversity for primary parasitoids and hyperparasitoids. Statistical analyses: The authors first tested direct effects of food web metrics on ES (parasitism rate of A. gossypii) and EDS (hyperparasitism rate) using GLMs with model selection via AICc. They then evaluated direct landscape effects on ES/EDS, food web metrics, and diversity. To consider multiple predictors simultaneously, linear mixed-effects models (LMMs) were fitted with groups of predictors (landscape composition, richness/diversity, food web features), with model averaging over candidate models (ΔAICc < 4) and assessment of variable importance. Variance inflation factors were used to diagnose collinearity. Finally, they performed path analysis using piecewise structural equation modeling (SEM) to quantify combined and cascading effects on ES and EDS separately, after confirming no direct linear relationship between parasitism and hyperparasitism. SEMs were fit in two steps, beginning with full models including landscape variables and then removing nonsignificant paths; model fit was evaluated with Fisher’s C and d-separation tests. Indirect and total effects were computed as products and sums of path coefficients. Analyses were conducted in R (packages: car, stats, lme4, MuMIn, piecewiseSEM, among others). Data comprised 25 independent field measurements.

Community composition and stability: Across three years and 25 sites, the network comprised 3 primary parasitoids (n = 1569 individuals) and 7 hyperparasitoids (n = 934 individuals). Binodoxys communis dominated primary parasitoids (91% ± 2%), and Syrphophagus spp. comprised 40% ± 4% of hyperparasitoids. The tri-trophic network remained stable over years. Direct effects of food web metrics: GLMs showed food web generality (Gq) was negatively related to ES (parasitism rate) (Coeff = -0.13, P = 0.020). Food web vulnerability (Vq) was positively related to EDS (hyperparasitism rate) (Coeff = 0.12, P = 0.007). Connectance was not highlighted as significant. Landscape direct effects: No landscape factor had strong direct effects on ES/EDS. NCH showed marginal positive association with ES (GLM Coeff = 0.25, P = 0.066; LMM P ≈ 0.055–0.061) and marginal positive association with EDS (GLM Coeff = 0.88, P = 0.077; LMM P = 0.076). Secondary crop cover (SC) was negatively related to food web vulnerability (GLM Coeff = -3.21, P = 0.034). No direct effects of landscape variables on Gq or on parasitoid/hyperparasitoid richness or Shannon diversity were detected. Combined predictors (LMM): Parasitism rate was negatively related to primary parasitoid species richness (Coeff ≈ -0.06, P = 0.001–0.002) but not to Shannon diversity. When controlling for other predictors, Gq was not significant for ES (P = 0.685). For hyperparasitism, Vq remained positively related to EDS (P = 0.010), whereas NCH had a marginal positive effect (P = 0.087); hyperparasitoid richness and diversity effects were not significant directly in LMMs. Path analysis (SEM): ES (parasitism): Direct negative effect of primary parasitoid richness on ES (β = -0.490 to -0.524, P ≈ 0.043). No direct effect of Gq or NCH on ES. Positive paths: parasitoid diversity → Gq (β = 0.899, P < 0.001) and Gq → parasitoid richness (β = 0.693, P < 0.001). Total effects on ES: parasitoid richness -0.524 (direct), Gq -0.363 (indirect), parasitoid diversity -0.326 (indirect). EDS (hyperparasitism): Direct positive effect of Vq on EDS (β = 0.415; total effect 0.514, P ≈ 0.009–0.044). Hyperparasitoid diversity positively affected Vq (β ≈ 0.817–0.824, P < 0.001) and indirectly increased EDS (total indirect effect ≈ 0.424). Hyperparasitoid richness increased diversity (β = 0.778–0.829, P < 0.001), indirectly elevating EDS (total ≈ 0.329). SC cover directly reduced Vq (β ≈ -0.424 to -0.449, P < 0.001) and indirectly lowered EDS (total ≈ -0.231). NCH did not have significant paths within SEMs. Overall: Biological control (parasitism of A. gossypii) was directly attenuated by higher primary parasitoid richness and indirectly weakened by greater food web generality and parasitoid diversity. Hyperparasitism increased with food web vulnerability and was modulated by landscape SC cover and hyperparasitoid diversity.

The findings demonstrate that on-farm biodiversity and trophic network structure jointly mediate how landscape composition translates into ecosystem services and disservices. Primary parasitoid richness, shaped indirectly by food web generality and parasitoid diversity, directly reduced parasitism rates, contradicting a common expectation that higher richness enhances ES. Complex food webs (high generality) were associated with more inter-trophic links and, in the presence of diverse parasitoid communities, coincided with lower parasitism—potentially through enhanced support for hyperparasitoids and altered interspecific interactions. In contrast, hyperparasitism was strongly governed by food web vulnerability, reflecting higher consumer sharing of hosts and potential for competition and cascading top-down effects. Landscape composition had nuanced roles: non-crop habitat showed only marginal positive associations with both ES and EDS, while greater secondary crop cover reduced food web vulnerability and thereby indirectly suppressed hyperparasitism. These results help reconcile inconsistent landscape–biocontrol relationships by revealing the mediating roles of food web properties and diversity, emphasizing that predictions of ES delivery require integrating multitrophic interactions with biodiversity metrics.

This study shows that species diversity and food web structure jointly determine biological control outcomes in agricultural landscapes. Parasitism of Aphis gossypii is directly reduced by higher primary parasitoid richness and indirectly by greater food web generality and parasitoid diversity, while hyperparasitism is directly elevated by food web vulnerability and indirectly modulated by hyperparasitoid diversity and secondary crop cover. Landscape management that increases secondary crop heterogeneity can reduce food web vulnerability and hyperparasitism, potentially enhancing net biological control. The work underscores the need to incorporate biodiversity and network structure into landscape-scale pest management and ecological intensification strategies. Future research should disentangle the relative contributions of non-crop habitats versus crop heterogeneity, expand to systems with richer herbivore-parasitoid communities (e.g., cereals), and refine measures of parasitoid-mediated pest suppression beyond parasitism rates.

The assessment is constrained by the species-poor herbivore community in temperate cotton agro-ecosystems in northern China and a simplified quantification of parasitoid-mediated pest suppression (parasitism rate as a proxy). Potential confounding from aphid outbreak dynamics (affecting parasitism rate denominators) and the lack of direct landscape effects on certain food web metrics also limit generalizability. The observational design across 25 sites limits causal inference beyond the SEM framework.