The data collection protocol is described in Sánchez et al.19 and followed Reporting Standards for Systematic Evidence Syntheses (ROSES) guidelines20. Philibert et al.21, also proposed eight criteria for conducting high quality meta-analysis, which overlap to some extent with ROSES guidelines. Our methods fulfil the Philibert et al. requirement to use a repeatable procedure for paper selection, provide a list of references, and ensure availability of the dataset, while other quality criteria are only relevant at the meta-analysis stage.
Search process
The literature search was conducted on 29 November 2019 and updated on 5 January 2021, and aimed to identify relevant English language articles published in peer-reviewed literature. We searched in titles, abstracts and keyword lists of literature in the Scopus and Web of Science databases, using the following search string (formatted for Scopus; see Dataset 1 for the equivalent string formatted for Web of Science): TITLE-ABS-KEY (“agricultur*” AND “biodiversity”) AND TITLE-ABS-KEY (“agro?ecology” OR “agro?biodivers*” OR “agroforestry” OR “border plant*” OR “riparian buffer” OR “woodlot” OR “hedgerow” OR “cover crop*” OR “crop rotation” OR “crop divers*” OR “inter?crop*” OR “mixed crop*” OR “cultivar mixture” OR “plant divers*” OR “polyculture” OR “tree divers*” OR “variet* diversity” OR “fallow” OR “field margin*” OR “grass strip*” OR “*flower strip*” OR “insect* strip” OR “conservation strip” OR “vegetation strip” OR “catch crop” OR “inter?crop*” OR “crop variety” OR “crop sequenc*” OR “mixed farming” OR “land sparing” OR “landscape heterogeneity” OR “heterogeneous landscape” OR “landscape diversi*” OR “divers* landscape” OR “homogeneous landscape” OR “landscape homogeneity” OR “landscape complexity” OR “simplif* landscape” OR “complex landscape” OR “multi?function* landscape” OR “integrated crop-livestock” OR “integrated crop-forest” OR “land sharing”) AND TITLE-ABS-KEY (“ richness” OR “ abundance” OR “species diversity” OR “functional diversity” OR “index”) AND TITLE-ABS-KEY (“crop yield” OR “crop production”) AND (LIMIT-TO (LANGUAGE, “English”)). We extracted the primary studies included in all relevant meta-analyses identified from the database search. In addition, we included a small number of peer-reviewed articles known to scientists consulted through the Sustainable Foods project and which were not retrieved by the search string or from previous meta-analyses. In total, 1590 articles with the potential to be included in the meta-analysis were identified (Fig. 1).
Article screening
All identified articles were screened at full-text level. We used the PICOC (Population, Intervention, Comparator, Outcomes, Context) framework to define the inclusion-exclusion criteria as described in Sánchez et al.19. These criteria required that, to be included: (i) the article presents a quantitative comparison of a diversified farming system (Intervention) compared to either a relatively simplified farming system (first Comparator) or to natural habitat (second Comparator), ii) the article reports quantitative outcomes for any terrestrial organism that is non-domesticated (Population), iii) the article provides the mean or median, variance and sample size for biodiversity outcomes, and outcome measures in comparator and intervention sites were collected using comparable sampling approaches (Outcome), iv) results are from primary field studies and not from experiments conducted in greenhouses or laboratories (Context).
Diversified farming systems were defined as agricultural plots where: i) more than one plant species or variety is cultivated at multiple temporal and/or spatial scales, such as crop rotations, intercropping or agroforestry, or ii) semi-natural habitat such as hedgerows and flower strips is embedded into the system, or iii) crop production is integrated with livestock or fish production, such as aquaculture or integrated crop-livestock systems. Simplified farming systems were agricultural plots with less diversity than in eligible interventions, i.e., plots with relatively fewer plant species or varieties (usually monocultures), less semi-natural habitat embedded, or no mixed crop-animal production. Where natural habitat was used as a comparator, this was defined as habitat that is not actively used for human activities, such as primary and secondary forests, wetlands, unmanaged grasslands and shrublands.
Suitable outcome metrics for biodiversity included any comparable quantified measure, such as richness, abundance, or Shannon’s diversity index. While studies only needed to report biodiversity outcomes to be considered for inclusion, we recorded harvested yield in all cases where this was reported and met our inclusion criteria. For yield outcomes to be included, the article must have provided means or medians, variance and sample sizes, and outcome measures at intervention and comparator sites must have been collected using comparable sampling approaches. Suitable outcome metrics for yields included the land equivalent ratio, weight of harvested produce per unit land area, or counts of harvested produce per standardized unit (e.g. grape bunch per plant, apples per branch). For comparisons comparing intercropped or agroforestry systems against simplified farming systems, the land equivalent ratio was prioritized as the outcome metric while other metrics were used only when the land equivalent ratio could not be calculated.
In total, 237 (14.9%) of retrieved articles met our inclusion criteria (Fig. 1).
Data extraction
From each article that met our inclusion criteria, we extracted qualitative data on: the literature source (e.g. authors, publication year, title); crop type (common name, scientific name); agricultural system (e.g. intercropping, monoculture, agroforestry, integrated crop-livestock system, crop rotation, set aside); non-domesticated taxa sampled (common and scientific names); functional group of the non-domesticated taxa, if specified (e.g. pest, decomposer, predator); biodiversity outcome metric (e.g. species richness, abundance, Shannon’s diversity index); yield metric (e.g. kilogram per hectare, grams per plant, land equivalent ratio); sampling method used (e.g. transect, trap); pesticide use (yes or no, and kg/ha); fertilizer use (yes or no, and chemical fertiliser use yes or no); soil management (e.g. tillage, no tillage, slash and burn); landscape characteristics (e.g. % agricultural land use, climate); and study location (local name, country and geographic coordinates). Following initial data-entry, we classified several variables into categories to facilitate data exploration and analysis. This included categorizing crops by the Food and Agriculture Organisation of the United Nations commodity group, woodiness (e.g. tree, shrub, herb), and growth cycle (perennial, annual), and documenting the phylum, class, order and functional group of each non-domesticated taxon.
We extracted quantitative data on: biodiversity outcome means or medians, variance and sample size; yield means or median, variance and sample size; farm size, if specified; length of time that the land has been in its current state, and; sampling duration (in days, from start to finish). Data on biodiversity outcomes and yield were extracted from figures using GetData Graph Digitizer 2.26 or WebPlotDigitizer v4.2. Where outcome values or units in an article were unclear or not provided, the corresponding author was contacted by email to request this information. If the author did not respond, the data entry was removed. We provide a dictionary of how the extracted data were recorded and coded in Dataset 2.
Data were organized using R-4.0.0 (R-Core Team, 2013) such that each row contained a pair of biodiversity outcomes and, where provided, a pair of yield outcomes, for a single comparator-intervention pair. In total, 237 studies containing 4076 comparisons of biodiversity outcomes and 1214 comparisons of yield outcomes were retained for analysis (Fig. 1).
Source: Ecology - nature.com
