This study was conducted in the southwestern region of Hainan Island, China (18°45′ N, 109°17′ E). The area has a tropical marine monsoon climate. The average annual temperature is 19–26°C. The mean annual precipitation is 1,400–1,800 mm. Land-use patches are dominated by smallholders, forming a highly heterogeneous mosaic landscape mainly comprised of forest, rubber (Hevea brasiliensis), mango, longan (Dimocarpus longan), papaya (Chaenomeles sinensis), areca (Areca catechu L.), and farmland (rice and vegetables) patches. Mango orchards, which are managed similarly and are the main income source for many smallholders in the region, have been affected by the invasive pest B. dorsalis for many years. B. dorsalis is thought to have invaded Hainan Island as early as 1934 (Liu et al., 2019). Smallholders would use insecticides to exterminate B. dorsalis 1–2 times a year during the near-ripening period of mangoes.

Taking the mango patch as the center, we set 15 landscapes as a gradient based on forest cover, with a radius of 1.5 km surrounding the mangoes (Figure 1A). This landscape radius was chosen based on previous studies that observed traces of multiple pest activity (Gardiner et al., 2009; González et al., 2020). Among the selected landscapes, forest cover percentage varied from 0 to 30%, with varying degrees of landscape fragmentation.

First, we obtained a high-resolution geographic image map for each landscape based on their coordinates using ArcGIS 10.2 software. We divided the landscape patches into 11 categories [forest, mango, rubber, longan, areca, banana, papaya, farmland, water area, residential land, and others (Figure 1B)] and then used ArcGIS 10.2 software to outline each patch. The patches were identified by combining high-resolution satellite images (0.83 m) and data from field surveys conducted in June–August 2021, after which the vector graphics were converted into raster maps to calculate compositional and configurational heterogeneity metrics. We used FRAGSTATS 4.2 software (McGarigal et al., 2012) to calculate the landscape indices.

Since longan, papaya, farmland and banana have a small percentage of area in the landscape, while forest, mango, rubber and areca are the main land types that make up the composition of the landscape, the compositional heterogeneity indicators included mango cover percentage (%), forest%, rubber%, areca%, and Shannon's diversity index. Configurational heterogeneity was indicated by edge density, which can reflect the degree of landscape fragmentation (Martin et al., 2019).

Two spatial scales were set: spatial distance scale and spatial combined scale. The spatial distance scale refers to the circle of buffer rings centered on the mango patch with radii of 0.5, 1, and 1.5 km, respectively (Figure 1B), which can clarify the effect of landscape variables with spatial distance.

The spatial combined scales were defined based on whether the buffer ring with a radius of 0.5 km was cut off from the larger 1 and 1.5 km rings. The nested landscape formed without being cut off was called the coupled (Figure 1B), and the central blank landscape formed by cutting off was called a decoupled. Thus, the spatial combined scale contains two categories of coupled and decoupled (Figure 1B). Combined scales were used to explore the effects of surrounding habitat composition and configuration at different distances.

In September 2021, the newly grown leaves of mango trees in the current year were in the mature stage. The invasive pest (B. dorsalis) and its associator (D. melanogaster) were identified in mango patches at the center of each landscape and other surrounding patches (such as forest and rubber) using yellow sticky traps (length × width: 20 × 25 cm) and yellow funnel traps (attractant: methyl eugenol). Three yellow sticky traps and three funnel traps were set at a height of ~1–2 m above the ground and were displayed at least 3 m apart within each patch. All traps were placed at least 6 m from the edge of each patch to reduce edge effects. Except for 15 mango patches at the center of the landscape, other surrounding patches were randomly selected in each landscape to determine whether the land cover types were habitats for B. dorsalis and D. melanogaster. In total, 75 patches were investigated in 15 landscapes, including 4 forest, 15 mango, 15 areca, 13 rubber, 9 farmland, 9 longan, 7 banana, and 3 papaya patches; thus, we placed 225 yellow sticky traps, and funnel traps, respectively. After 3 days, we collected all yellow sticky and funnel traps and counted the number of each species.

During the flowering (October 2021), young fruit (November 2021), and near-ripe stages (March 2022) of mangoes, we measured the invasive pest and its associator in central mango patches using two yellow traps methods by the same sampling method as in September 2021. Therefore, the numbers of each species in mango patches were determined four times. The species abundance of each patch measured each time was obtained by summing the two methods, following previous study (Perrot et al., 2022).

At the patch level, we confirmed the presence of both species in the selected habitat types, and then analyzed the differences in B. dorsalis and D. melanogaster abundances among habitats using ANOVA. Multiple comparisons were then made using the Tamhane's T2 due to the heterogeneity of variance in the species abundance data. We explored the correlation between B. dorsalis and D. melanogaster abundance using linear regression analysis across all patches. Species abundance was log-transformed before doing the regression analysis, as were the following regressions and correlation analysis.

At the landscape level, we first removed the covariance between landscape variables in two steps. (a) Pearson's correlation analysis was performed on the landscape variables at each spatial scale (Supplementary Figure 1), and only those variables with weak correlations (r < 0.65) were retained (Da Silva Carneiro et al., 2022); (b) The retained landscape variables were subjected to multiple regression analysis to log-transformed species abundance, and the landscape variables were further filtered by VIF<2 (Perrot et al., 2022). The landscape variables that were thus screened were edge density, forest cover percentage (%), mango%, and Shannon's diversity (Supplementary Table 1). The general linear model was then used to analyze the relationship between each landscape variable and species abundance, with landscape variables as fixed effects, species abundance as response variables, and sampling site and month as covariates in the spatial distance and combined scales.

To explore the effects of spatial distance and combined scales on the landscape variable-species abundance relationship, we explored differences in linearly fitted relationships between landscape variables (Supplementary Table 1) and species abundance using grouped regression. The spatial distance scale varies from 0.5 to 1 km and then to 1.5 km. Spatial combined scales for coupled and decoupled were compared at distance scales of 1 and 1.5 km, respectively. Differences in linearly fitted relationships were checked using the Chow test (Zeileis et al., 2002; Wen et al., 2022).

To reveal the optimal landscape variables and spatial scales, we subjected all landscape variables (Supplementary Table 1) and species (B. dorsalis and D. melanogaster) abundance to multiple linear regression analysis in a stepwise approach, with landscape variables as independent variables and species abundance as dependent variables. Standardized regression coefficients were used to compare the relative effects of landscape variables on species. All of the above data analyses were conducted using IBM SPSS software (version 21.0).

Configurational heterogeneity (edge density) was negatively correlated with B. dorsalis abundance (P < 0.05, Figure 3A, Supplementary Figure 2A), but the correlation with D. melanogaster was insignificant (P > 0.05, Figure 3D, Supplementary Figure 2D). In contrast, compositional heterogeneity was positively correlated with B. dorsalis and D. melanogaster abundance owing to an increase in the percentage of forest cover (P < 0.05, Figures 3B, E, Supplementary Figures 2B, E). The percentage of mango cover was not significantly associated with either species at 1 and 1.5 km scales (P > 0.05, Figures 3C, F).

Negative relationships between edge density with B. dorsalis abundance was observed across spatial distance scales of 0.5, 1, and 1.5 km (Figure 3A, Figure 2A). In addition, there were no significant changes in the slopes of these fitted lines (Figure 3A), implying that variation in the strength of these indicator effects is limited. Positive correlation between forest cover percentage with B. dorsalis and D. melanogaster abundances did not change with spatial distance scale from 0.5, 1 to 1.5 km (Figures 3B, E). The slope of the fitted lines between forest cover and species abundance did not change significantly at the scales of 0.5,1, and 1.5 km (Figures 3B, E), indicating that the effect of forest cover had limited variation in intensity. Overall, spatial distance scales did not alter the direction and intensity of compositional and configurational heterogeneity affecting the species (Figure 3).

At spatial coupled and decoupled, the direction of linear relationships between edge density with B. dorsalis abundance was consistent, regardless of 1 or 1.5 km space distances (Figure 3). In addition, there was no difference in the strength of the linear relationship, as the differences in the regression coefficients of the fitted straight lines were insignificant (P > 0.05, Table 1, Supplementary Table 2). Similarly, the direction and strength of the correlation between forest percentage with B. dorsalis and D. melanogaster abundance did not change from spatial coupled (Figures 3B, E) to decoupled, regardless of 1 km (Table 1) or 1.5 km space distances (Supplementary Table 2). Therefore, spatially combined scales had limited effects on compositional and configurational heterogeneity effects.

Considering all landscape variables in spatial distance and combined scales, edge density at coupled 1 km scale and forest cover at coupled 0.5 km scale were important factors for B. dorsalis, and the importance of the former (absolute value of the standardized coefficient) was higher than the latter (Table 2). For D. melanogaster, the proportion of forest cover at coupled 1 km scale was more important than edge density at coupled 1 km scale, although the latter had a significant effect (Table 2).

Compositional heterogeneity was positively correlated with the invasive pest and associator dependent on forest cover rather than other habitats, which differs from the results of previous studies showing that forest cover is often thought to reduce pest abundance (Medeiros et al., 2019; González et al., 2020). For example, forest cover is negatively associated with a coffee leaf pest (Leucoptera coffeella) (Medeiros et al., 2019) and fruit pest (coffee berry borer) (Aristizábal and Metzger, 2018). This conflicting result may be attributed to the differences in the ability of forests to control native and invasive pests. For native pests, various factors exist in the forests that limit their populations, such as competitors and natural predators (Henri et al., 2015; Aristizábal and Metzger, 2018). For invasive pests, however, there is not only a lack of competitors and natural enemies, but also sufficient alternative food sources, overwintering sites, and refuges provided by forests with high plant diversity (Tscharntke et al., 2016; Gurr et al., 2017; Tamburini et al., 2020). This is potentially conducive to the survival and reproduction of invasive pests, thereby increasing the number of invasive pests in agricultural land adjacent to forests. Similarly, in addition to obtaining resources from surrounding forests, the associator can better maintain populations by following the invasive pest, as shown by the significant positive correlation between the abundance of the invasive pest and the associator.

In contrast, configurational heterogeneity was negatively correlated with invasive pest abundance. Both negative and positive effects of landscape configuration on insects have been reported in previous studies (Fahrig, 2017; Martin et al., 2019). For pests, configurational heterogeneity with high edge density may reduce crop cover available as a food resource, echoing the resource concentration hypothesis (Tscharntke et al., 2016), while high edge density resulting from fragmentation promoted pest transfer from crop patches to surrounding habitat due to dispersal effects (Martin et al., 2019; Haan et al., 2020). Fragmentation increases new habitats that are more suitable for pests to survive, as in the present study, and rubber plantations are more suitable for B. dorsalis than mango orchards, accelerating the outward spread of B. dorsalis, thereby making configurational heterogeneity negatively related to this invasive pest. However, the associator was insignificantly associated with configurational heterogeneity, which may be attributed to the insignificant differences in associator abundance across habitat types (Figure 2B), that is, the number per unit area remained unchanged regardless of configurational heterogeneity. This indicates that species respond differently to landscape changes (Martin et al., 2019), even though the two species are closely related.

Notably, compositional heterogeneity contributed less to invasive pest abundance than configurational heterogeneity; however, the opposite was true for the associator. Although landscape composition with high forest cover promoted the invasive pest, invasive pest overflow from forests would be buffered by configurational heterogeneity with high edge density through dilution effects (Haan et al., 2020), similar to that observed in other insects (Martin et al., 2019; Souza et al., 2020). Other habitats (e.g., areca and banana) did not differ significantly from mango orchards in attracting invasive pests (Figure 2A), but some habitats (e.g., rubber) were more suitable for the survival of invasive pests than mango orchards. This also facilitates the escape of invasive pests in response to strong disturbances in mango orchards due to habitat fragmentation with high edge density, which provides a longer common boundary (Souza et al., 2020; Moore et al., 2022). These factors make configurational heterogeneity more important for invasive pests, supporting the habitat diversity hypothesis (MacDonald et al., 2018). In contrast, compositional heterogeneity had a greater effect on the associator, mainly because it was positively affected by forest cover, whereas edge density showed limited effects. However, at the landscape level, the relative effects of landscape variables on specific species remain elusive, as species respond to landscape heterogeneity, which varies widely across diverse landscape contexts (Martin et al., 2019; Moore et al., 2022).

The results showed that the strength and direction of the relationships between compositional and configurational heterogeneity with the invasive pest and its associator did not vary with spatial distance. Some studies suggest that the effects of landscape variables on species vary with spatial distance scales (Aristizábal and Metzger, 2018; Redlich et al., 2018), which contradicts our results. This may be due to differences between species with different functional traits and dispersal abilities. Winged insects (such as B. dorsalis and D. melanogaster) can travel faster and farther than other insects because of their ability to fly (Miguet et al., 2016; Zhang et al., 2021), and generalist pests (such as B. dorsalis) spread farther than specialists because of extensive habitat adaptation (Miguet et al., 2016; Pan et al., 2022). In particular, invasive pests can survive in a variety of habitats and spread easily among patches owing to the formation of patch connectivity. Furthermore, the spatially coupled scales formed by small distances (e.g., 1 km) nesting within a large distance (e.g., 1.5 km) make it difficult to separate the effects from each other (Martin et al., 2016), thus making the direction and intensity of landscape variable effects constant with spatial distance.

Another novel finding of the present study is that spatially combined (coupled and decoupled) scales did not change the intensity and direction of landscape effects, which provides new insights into the effects of landscapes on pests and associators. Previous results have only shown that landscape structures control pests at spatially coupled (Aristizábal and Metzger, 2018; Kheirodin et al., 2020; Zhang et al., 2021), making it difficult to distinguish the role of surrounding landscape heterogeneity at different spatial distances from the sampled crop patches (Da Silva Carneiro et al., 2022). The comparison of results in the present study revealed that spatial coupling and decoupling did not change the landscape effects, whether the spatial distance was 1 or 1.5 km, indicating that farther and nearer distance habitat heterogeneity has similar effects on species, at least for B. dorsalis and D. melanogaster. This may be because species with a flexible diet and flight ability have an efficient dispersal capability and adaptability regardless of habitat type (Miguet et al., 2016; Zhang et al., 2021), especially for invasive pests, there may also be a lack of natural enemies in geographic space. Furthermore, habitat type and landscape variable changes were limited after spatial decoupling compared to coupling, regardless of distance.

The present findings have implications for habitat conservation and biological control, especially in the context of landscape heterogeneity due to global land-use change (Martin et al., 2019; Zheng et al., 2019). Although the results showed that forest cover was positively associated with an invasive pest, especially at a distance scale of 0.5 km radius, reducing forests to control invasive pests is not recommended. Most studies have found that forests conserve biodiversity, including plants, arthropods, birds, and mammals (Zhang et al., 2017; Aristizábal and Metzger, 2018; González et al., 2020). The benefits of controlling invasive pests by destroying forests are likely to be far less than those of biodiversity conservation by protecting forests, not to mention the fact that forests potentially provide many natural enemies to control agricultural pests (Aristizábal and Metzger, 2018; Varela et al., 2018; Martin et al., 2019). Habitat fragmentation had a negative effect on the invasive pest. Maintaining the diversity of habitat types in the surrounding landscape composition is beneficial for the control of invasive pests and can be attempted in landscape management at the 1 km landscape radius. In addition, the effects of spatial scales were not evident in the present study, it is suggested that spatial distance and combined scale should be combined to elucidate landscape effects in broader regions (Medeiros et al., 2019; Da Silva Carneiro et al., 2022), especially at spatially decoupled, as scale effects may depend on landscape contexts and species functional groups (Martin et al., 2019; Haan et al., 2020).

Pests are not only regulated by landscapes but also affected by natural enemies (Kheirodin et al., 2020; Zhang et al., 2021), which are thought to be closely related to compositional and configurational heterogeneity (Tscharntke et al., 2016; Martin et al., 2019; Kheirodin et al., 2020). Although we used a pest sampling method known to trap parasitic natural enemies (Böckmann et al., 2015), we did not capture parasitoids associated with the invasive pest that have been reported (Liu et al., 2019). It is still necessary to explore the presence or absence of other natural enemies associated with invasive pests and their responses to landscape heterogeneity at multiple spatial scales, which is important for unraveling the mechanisms of landscapes controlling invasive pests. Moreover, the effects of invasive pests on crops need to be assessed in detail, such as fruit damage rates and economic losses to smallholders, which are directly related to policymaking. Taken together, we suggest that a cascading framework of landscape structures, natural predators, invasive pests, and crop loss should be constructed to address the possible impacts of land-use change on sustainable agriculture in the future.