The conversion of forests into agricultural areas is considered the main threat to global biodiversity. This process leads to the loss of native species and, consequently, threatens the provision of ecosystem services and the functioning of ecological processes (Ramankutty et al., 2018; Watling et al., 2020). Therefore, preventing forest loss and increasing forest cover in degraded landscapes are considered essential principles to enhance biodiversity conservation (Riva et al., 2024). Species conservation can also be achieved by improving matrix quality (Perfecto and Vandermeer, 2010; Arroyo-Rodríguez et al., 2020). For instance, agroecosystems that maintain native tree species, such as shaded agroforests, can exhibit a vegetation structure similar to that of native habitats, therefore prone to support greater biodiversity and consequently preserving a wide range of wildlife species (Costa et al., 2021; Niether et al., 2020; Ferreira et al., 2020). However, each agroforestry system can have different local environmental conditions (e.g., vegetation structure and management intensity) and be embedded in different landscape scenarios (e.g., remaining forest cover), which can determine its conservation value (de la Mora et al., 2013; Ferreira et al., 2024a). Therefore, understanding the relative importance of local and landscape predictors for preserving species diversity in agroforestry systems becomes crucial to promote biodiversity-friendly agricultural practices in human-modified landscapes (Arroyo-Rodríguez et al., 2020).

In tropical countries, cocoa plantations are important agroforestry systems in terms of socio-economic-environmental benefits. This is the case of traditional cocoa (Theobroma cacao) plantations (hereafter, shaded cocoa agroforest) in southern Bahia, Brazil, where cocoa is grown under the shade of emergent trees, including species native to the threatened Atlantic Forest (Myers et al., 2000). Such a combination of native and cocoa trees increases the heterogeneity of vegetation structure, and can provide supplementary habitat for native forest-dweller species (Cassano et al., 2009). In fact, these agroforests are recognized for their high conservation value, as they are used as temporary or permanent habitats by different terrestrial and flying mammals (Faria and Baumgarten, 2007; Ferreira et al., 2020, 2025), reptiles, and amphibians (Cervantes-López et al., 2025).

Shaded cocoa agroforests can also be highly valuable for insect conservation, particularly for bees and wasps (Bos et al., 2007; Ferreira et al., 2024a; Nascimento et al., 2025). Although bees are not considered the main pollinators of this crop (Toledo-Hernández et al., 2017), both bees and social wasps can use these agroforests as supplementary habitats (Ferreira et al., 2024a). However, as cocoa plantations exhibit different vegetation structure mostly driven by contrasted management practices, the local condition can differ among agroecosystems, with subsequent influence on species persistence. For example, the reduction of shade tree richness in Indonesian cocoa agroforests (i.e., greater management intensification) resulted in local changes in terms of temperature, humidity, canopy openness and herbaceous extract, which subsequently lead to the reduction in bees and wasp’s diversity (Bos et al., 2007). As the availability of nesting substrate is important for determining the occurrence of bees and wasps (Araújo et al., 2021; Morato and Martins, 2006), especially for species that nest above ground, the structural vegetation complexity can also predict the diversity of these insects in agroforests. Finally, landscape forest loss can also limit the availability of resources, as food and nesting substrate for bees and social wasps (Ferreira et al., 2024a), and therefore, the effect of management intensity may depend on the remaining forest cover (i.e., an interacting effect). However, to our knowledge, no study to date has assessed how local characteristics associated with landscape composition determine the diversity of bees and wasps in these agroecosystems.

Here, we investigated the effect of local (i.e., management intensity, tree basal area, and tree diversity) and landscape (i.e., forest cover) variables on the number and composition of bee and social wasp species across 29 shaded cocoa agroforests from the northeastern Brazilian Atlantic Forest. We predicted that farms under greater management intensity would retain lower species number, since such practices will imperil the persistence of more sensitive species. In addition, shaded cocoa agroforests with greater tree diversity, higher basal area of shade trees and inserted in landscapes with higher forest cover possibly offer a greater diversity of floral resources, availability of nesting sites, and suitable habitat, respectively. As a result, we expect these characteristics to be positively associated with the number of bee and wasp species in these agroforests (Basset et al., 2012; Flores et al., 2018; Medeiros et al., 2019). We also predicted that agroforests with high management intensity, lower tree diversity, lower basal area of shading trees, and inserted in landscapes with lower forest cover would show high species composition dissimilarity, due to the loss of bee and wasp forest-dweller species and the dominance of generalist species commonly found in degraded environments (Ferreira et al., 2015; Souza et al., 2010). Since landscape context can modulate the effects of local management (Aycart-Lazo et al., 2025), we expect that the response of bee and social wasp assemblages to local characteristics will be weaker in landscapes with greater forest cover.

We collected 254 bees belonging to 38 species, and 346 wasps from 24 species (Supplementary Material 2). The average number of species of bees and social wasps per site was 4.3 ± 2.3 (range = 0–9 species) and 3.4 ± 2.1 (range = 0–10 species), respectively. Our model selection approach revealed that the number of bee species was best predicted by landscape forest cover alone (marginal pseudo-R² = 0.37; β = 0.25; p = 0.01), with a positive influence (Fig. 2A), whereas for the composition of bee species, the best model included only the shade tree diversity (marginal pseudo-R² = 0.20; β = −0.10; p = 0.03; Table 1 and Fig. 2B). Regarding the social wasp species number, the best model included the interactive effect of management intensity and landscape forest cover (marginal pseudo-R² = 0.29; β = −0.47; p = 0.01; Table 1). We observed that increased management intensity positively affected the number of social wasp species, but only in landscapes with low forest cover (Fig. 2C). Finally, for the composition of social wasp species, our model selection included the null model among the parsimonious models, and therefore we considered that none of the variables was able to explain the change in species composition (Table 1). The values for all models are presented in Appendix E in Supplementary Material.

Fig. 2.

Effect of predictor variables on the number of species of bees and social wasps in shaded cocoa agroforests in southern Bahia, in the Brazilian Atlantic Forest. We only show relationships from the selected models, shown in Table 1. In panel C, low (19%), medium (37%), and high (55%) values correspond to the first, second, and third quartiles, respectively, of forest cover in the landscape.

In this study, we assessed the combined effects of local and landscape variables on bee and social wasp taxonomic diversity in shaded cocoa agroforests. As expected, we unveiled that bees and social wasps respond differently to habitat modifications at local or landscape scales. In particular, we demonstrated that only landscape forest cover explained the bee species number, but that the species composition was influenced by shade tree diversity. On other hand, only management intensity explained the wasp species number, but this effect was dependent on forest cover, while none of the variables explained the change in species composition. Considering that we are facing a global diversity and pollinator crisis (Betts et al., 2017; Levy, 2011), which calls for urgent management measures in human-modified landscapes (Arroyo-Rodríguez et al., 2020), our results demonstrate that both local management practices and landscape context should be considered to maximize bee and wasp species in cocoa agroforests.

Contrary to our expectations, none of the local variables explained the bee species number in cocoa agroforests. We attribute this low variation in the species number to the high heterogeneity characteristic of these agroforests, especially due to the presence of emergent and native Atlantic Forest trees (Cassano et al., 2009). In fact, in structurally more complex agricultural systems, bee species number may be similar to, or even higher than those found in native remnants within agricultural landscapes (Hoehn et al., 2010; Serralta-Batun et al., 2024), as also observed for these agroforests (Ferreira et al., 2024a). Furthermore, although the ages of the sampled agroforests unknown, this is a traditional system has expanded since the early 20th century and currently constitutes the main type of land use in this region (Faria et al., 2021). Therefore, considering the rapid life cycle of insects, it is possible that the bee assemblage present in these agroecosystems is composed of tolerant species to disturbance intermediate levels. Nonetheless, we observed that species composition was influenced by the shade trees diversity. This is to be expected, given that bees are highly dependent on floral resources (Michener, 2007) and that the occurrence of certain species may be determined by the availability of specific resources. Thus, greater tree diversity at the local scale may favor the occurrence of bees with more specialized relationships with host plants (González‐Chaves et al., 2024).

As expected, landscape composition proved to be important in determining the number of bee species in these agroforests. According to our predictions, landscape forest cover positively explained the increase in bee species number in surveyed agroforests, which can be explained by the habitat amount hypothesis (Fahrig, 2013). Landscapes with high forest cover may offer a greater availability and diversity of floral resources and nesting substrate, which can be accessed mainly by organisms that have a high vagility (Zurbuchen et al., 2010). Indeed, habitat loss has been one of the main causes of the decline of pollinators (Potts et al., 2010), including bees in agricultural landscapes (Saturni et al., 2016; Medeiros et al., 2019). Therefore, our results support the idea that maintaining or restoring forest remnants, especially in deforested landscapes, is an essential and priority measure to ensure the conservation of bee diversity in agricultural landscapes (Arroyo-Rodríguez et al., 2020; Riva et al., 2024), even in biodiversity-friendly systems.

For social wasps, we observed that the management intensity had a positive effect on species number, but only in landscapes with low forest cover. We attribute this response to factors such as the management practices adopted by cocoa producers and the resources provided by these agroforests, such as prey and nesting substrate. Our management intensity index includes practices such as pruning and thinning cocoa trees, which are less invasive than those employed in mechanized or intensive agricultural systems. These practices can also contribute to increase solar incidence, which may favor the activity of these wasps (da Silva et al., 2022) and increase prey density, typical of environments with more sunlight, such as forest edges and clearings (Barbosa et al., 2005; Pereira et al., 2017). Thus, as they are structurally complex agricultural systems, these agroforests may also provide an abundant supply of insects and nesting substrates (Ferreira et al., 2024b), on which social wasps are highly dependent (Barbosa et al., 2021). However, the positive effect of management intensity on the species number only in landscapes with low forest cover is unexpected, given that higher forest cover generally favors social wasp species richness in agricultural areas (Medeiros et al., 2019). In addition, it is important to note that the forest cover is negatively correlated with the amount of landscape agroforest (r = −0.77, p < 0.001). Therefore, less forest in the landscape implies an increase in the amount of supplementary habitat for these wasps, which may explain the positive response of these insects to intensified management in less forested landscapes. In this way, the environmental plasticity and generalist diet of social wasps, associated to the structural complexity of these agroforests, could favor these insects even under more intensive management.

The sustainability of agricultural practices depends on the adoption of strategies that minimize the trade-off between biodiversity conservation and agricultural yields. Despite the high conservation value of the shaded cocoa agroforests in the southern Bahia, this system has low productivity, considering the average production in other regions (Gama-Rodrigues et al., 2021). This lower economic gain can drive producers to adopt more intensive production methods with potential damage to biodiversity, like full-sun cocoa plantations or coffee monocultures, as previously observed in our region. However, studies suggest that the maintenance of shade trees associated with less invasive management practices (i.e., changes in the structural characteristics of agroforests such as the density of cocoa trees and associated trees), which is possibly the main bottleneck to greater productivity (Jagoret et al., 2017), can provide win-win scenarios for productivity and biodiversity conservation by enhancing the economic gains ensure the maintenance of high species number in these systems. In this context, our results indicate that agroforests inserted in landscapes with high forest cover can favor insect conservation, especially of bees, even under more intensive management. Additionally, these agroforests could experience greater productivity due to ecosystem services and ecological intensification associated, while maintaining local biodiversity (Araújo et al., 2025). Finally, although bees are not considered effective pollinators of cocoa, the pollination deficit is one of the factors associated with lower cocoa productivity (Toledo-Hernández et al., 2017). Thus, we can expect that maintaining a suitable habitat for bees (such as shaded cocoa agroforests), could also reflect in a favorable habitat for the maintenance of more effective cocoa pollinators (Toledo-Hernández et al., 2017). In addition, these benefits could also be extended to include the biological control offered by wasps and insectivorous vertebrates (Aycart-Lazo et al., 2025). Therefore, we suggest that restoration (in landscapes with low forest cover) or the maintenance of forest remnants (in landscapes with moderate or high forest cover), which could reduce the possible negative effects of intensified management on bee species richness, should be priority efforts to safeguard the taxonomic diversity of bees and social wasps in shaded cocoa agroforests.

The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.