We used the t-distribution with to calculate the 95% CI for the overall mean effects . We used the Restricted Maximum-Likelihood method for estimating all model parameters because it is more efficient and presents less bias when calculating heterogeneity . For abundance and species richness data, we examined the suitability of conducting the three-level meta-analysis compared to the two level model using one-sided log-likelihood-ratio tests for the variance at the second and third levels separately.In addition to the heterogeneity parameters tested by the model, we quantified the proportion of the observed variance for each of the three levels , I2 , I2 using the formulas provided in Cheung . The application of moderator analyses is justified when the I2is low compared to the total variance . We applied univariate meta-regression procedures to examine whether the functional groups , and landscape composition , heterogeneity , and configuration metrics moderated the effect of farming systems on biodiversity. Landscape configuration , was categorised for capturing differences at smaller and larger ranges beyond the 1 km threshold. An omnibus test based on the F-distribution was used to evaluate whether there were significant differences in effect sizes across the values of each moderator . We also examined if the influence of each landscape metric on the estimated effect size were the same across the six functional groups, by including the interaction between functional groups and each landscape metric as a moderator in the meta-regression models. To facilitate the interpretation of the results, we reported the mean effect sizes and their CI as percentages changes by back-transforming the log response ratios and converting them to percentages − 1.
A mean effect size > 0 indicates biodiversity was positively impacted by diversified farming, whereas a mean effect size < 0 indicates biodiversity was negatively impacted by diversified farming. A mean effect size equal to zero indicates little or no difference between diversified and simplified systems. We applied multivariate models to check which variables continued significantly moderating the overall effect when controlling for other variables hydroponic grow table. The inclusion of highly collinear moderators in the same model, however, may lead to problems of overfitting or spurious results . We tested for multicollinearity between moderators by computing the Cramer’s V correlation between categorical variables, Spearman correlation between continuous variables, and Kruskal-Wallis H test between categorical and continuous variables. We considered moderators as presenting severe collinearity between each other when Cramer’s V or Spearman correlation were ≥ 0.7 or Kruskal-Wallis H tests were significant . We constructed separated multivariate models for abundance and species richness by adding all non-collinear moderators found to be significant predictors during the univariate models. We ignored collinearity between functional groups and the landscape metrics, since they are conceptually very different . We used omnibus tests to compare the significance of each moderator that had been found to be significant in the univariate models. Finally, we used multi-model inference to check the model fit statistics of each applied model based on the corrected Akaike’s Information Criterion . We used the AICc as our selection method because it is considered to be less biased and selects models that are much closer to the truth . We found significant interaction effects between functional groups abundance and the landscape metrics . The positive impact of diversified farming systems on decomposers abundance increased with the increment of the percentage of natural habitats, land cover diversity, and when plots were near and very far away from natural habitats . The impact of diversified farming systems on natural enemies’ abundance were significantly positive in farming plots located near and moderately near to natural habitats.
Contrastingly, the abundance of pests was significantly reduced in diversified farming systems located very close or moderately near to natural habitats, and in landscape with high land cover diversity. The abundance of pollinators was significantly higher in diversified farming systems located moderately far and very far away from natural habitats, and in agricultural landscape with high land cover diversity. The omnibus tests from the multivariate models applied for abundance showed that functional groups and the interaction between Functional groups and Euclidean min-distance to natural or semi-natural habitats remained significant moderators . However, the interaction effects between Functional groups and Percentage of natural habitat, and Functional groups and Land cover Shannon’s diversity index were no longer significant when including Functional groups in the models.This meta-analysis showed that, on average, diversified farming systems improves the richness and abundance of non-domesticated taxa with potential benefits for local and global biodiversity and food production goals. The magnitude of the overall effect varies with functional group and landscape complexity, heterogeneity and composition. Similar to Lichtenberg et al. , our study showed that agricultural diversification substantially increased species richness, and had a non-significant positive effect on abundance. This may reflect that farming diversification helps support a wider range of species, probably as more diverse habitat and resources open up . However, diversification may limit simultaneously individual species dominance reducing the population numbers for some species . For example, our results show that while the abundance of beneficial functional groups tended to have positive effect sizes, pest populations significantly decreased in diversified systems and therefore reduced the overall mean effect size for abundance. The variable effects of diversified farming systems across functional groups supports the notion that different species respond to biotic interactions and abiotic conditions according to their functional traits . The negative effect of diversified farming systems on pest abundance may be related to higher predation and parasitism rates in diversified farming systems , due to synergistic effects between species . Indeed we found a higher richness of natural enemies in diversified systems, consistent with several previous meta-analyses . Our study extends on previous work by including pest plants and our findings suggest the pest control benefits of diversified farming systems extend to weed suppression.
Crop and farm diversification may be an effective weed management strategy helping to reduce the need for herbicide inputs . Our synthesis demonstrated pollinator richness and abundance benefited from diversified farming practices, consistent with previous quantitative syntheses . The positive effect of diversified farming systems on the concentration of pollinator species might lead to a greater provision of pollination services with benefits to crop yields . Ensuring pollinator abundance may be more important than richness for provision of pollination services since pollination in agroecosystems might be mostly delivered by a few abundant and widespread species of insects rather than rare pollinator species . Nevertheless, a high redundancy of pollinators can help ensure community stability and function provision against unexpected changes . Agroecosystem resilience is associated with high diversity of organisms responsible for maintaining soil structure and nutrients cycling . Conversely to previous studies that indicated local management had little effect on decomposers , our results – including a wider range of taxa – demonstrated that diversified farming systems substantially increased decomposer abundance. This may be because diversified farming systems have a higher concentration of soil organic matter, benefiting soil biota . Effects on decomposers may also be associated with changes in autotrophs identified in this study, since a higher richness of autotrophs can supply more diverse resources which may enhance the decomposer community . Simpler landscape configurations enhanced the positive effect of diversified farming systems on the overall biodiversity richness in our study, specifically when natural habitats were > 1000 m away or constituted ≈ < 60% of the surrounding 1 km radii area. Our results agree with the “intermediate landscape complexity hypothesis”, which stablish the positive effect of local-conservation practices might be higher in agricultural plots located in simple landscapes than in complex ones . This response may be because in landscapes with a lower proportion of natural habitats, diversified farming systems may offer more varied habitats and resources attracting a higher variety of organisms than monocultures . Hence, diversification strategies at the farm level should complement restoration approaches at the landscape scale for promoting biodiversity, and softening productivity-ecosystem services trade-offs . Similar to prior observations , we found some interaction effects between functional groups diversity and the surrounding landscape characteristics. Our synthesis showed the importance of farming diversification at increasing pollinators abundance and richness in highly simplified and heterogeneous landscapes. Farm-scale diversification strategies might counter the direct negative impact of the absence of surrounding natural or semi-natural habitats on pollinators number and diversity, and therefore,flood tray enhance pollination service provision .
Natural and semi-natural habitats close to farming plots might be important to sustain populations of predators which benefit from the resources offered by natural habitats as well as by adjacent crops . This may also explain the reduction in pest abundance in diversified farms close to natural and semi-natural habitats, where predators were more abundant helping limit infestations . Moreover, our results showed the positive effects of diversified farming systems on decomposer abundance tend to increase with landscape complexity, primarily the proportion of surrounding natural habitats and landscape heterogeneity, but also with the distance from natural and semi-natural habitats. These results suggest the colonisation of soil biota may depend on the surrounding landscape and organisms’ dispersal abilities . However, decomposer richness seems to be independent of landscape patterns , and more influenced by local characteristics such as soil type, climate, plant diversity, and temperature . On the other hand, autotrophs responded positively to farming system diversification as landscape complexity increased, which may reflect that higher landscape complexity facilitates seed pool dispersal and establishment . Besides, most of the autotrophs’ data came from the comparison effect of simplified and agroforestry systems, suggesting the establishment and development of non-crop plants also could be promoted by the lower herbicide input or soil disturbance practices that characterised these diversified systems . Results from the multivariate models revealed the robustness of the unique moderating effect of functional groups, and the interaction between landscape configuration and functional groups, on abundance. However, the interaction between landscape composition and heterogeneity with functional groups were no longer significant when accounting for other variables in the same model. These results may be because landscape composition and heterogeneity interact with functional groups in determining effect sizes, but functional group has a stronger effect on its own. For species richness, none of the non-collinear predictors with significant moderating effects within the univariate models remained significant when included together in the same model. The reduction in the significance of these variables may be related to the lower number of effect sizes in the species richness database , which leads to small class sizes when multiple variables are combined in a single model. When the sample size is small, multivariate models may be statistically less powerful than univariate models . While our results show that functional group, and distance to and proportion of natural and semi-natural habitat, moderate the effect of diversified farming on species richness, future research to expand the number of effect sizes in the richness database as new studies emerge would be valuable to confirm which moderators have the strongest influence. The sensitivity and publication bias analysis, in addition to the prepublication of the review protocol and the meta-data , guarantee the transparency and reproducibility of our meta-analysis . While sensitivity analyses confirmed the robustness of our overall findings for abundance, we identified some changes in the significance of the species richness results when excluding the effect size outliers and with a high risk of bias from the analyses. The exclusion of one highly influential effect size reduced the significance of the overall farming systems impact on species richness by widening its confidence interval. However, the significance of the overall species richness results was not affected by the exclusion of 175 effect sizes with a high risk of bias. Hence, this last finding may ensure the robustness of our overall results for species richness. On the other hand, the reduction in significance of positive effect of diversified systems on pollinators species richness may be related to the exclusion of almost 40% of the effect sizes. Moreover, we identified possible sources of publication bias that may reduce the generalisation of our conclusions.