Metapopulation networks unlock the effects of landscape fragmentation on agricultural pests and natural predators

Abstract

In agricultural landscapes, patches containing diverse species can serve as nodes, and species diffusion among these patches can serve as edges, forming metapopulation networks. However, the impact of landscape fragmentation on these networks and consequences for keystone crop species remain unclear. Here, we investigated thrips (pests) and lacewings (natural predators) in mango orchards and surrounding habitats across 15 landscapes. Using a machine learning neural network, we built metapopulation networks for thrips and lacewings based on the relationship between patch properties and species abundance. We evaluated these networks with metrics like clustering coefficient, strength, and path length. Furthermore, we analyzed the effects of landscape fragmentation, as indicated by edge density and landscape Shannon diversity, on network structures and the subsequent effect on species abundance using piecewise structural equation modeling. The results show that the neural network effectively constructed the metapopulation network, accurately predicting lacewing and thrips abundance in various patches with an accuracy rate exceeding 80%. Positive correlations were observed between edge density and network strength for both species. However, a positive relationship between landscape Shannon diversity and network path length was identified only for lacewings. Edge density had a greater impact on lacewings and thrips compared to landscape Shannon diversity, with a negative effect on lacewings and a positive effect on thrips. Edge density affected lacewing abundance both directly and indirectly through network strength. However, no network structure mediated the effect of landscape fragmentation on thrips abundance. Our conclusion highlights that the effects of landscape fragmentation on species are primarily driven by edge effects. These effects were clarified by the metapopulation network structure, which varied depending on the species. The findings provide new perspectives on metapopulation networks to unlock the landscape fragmentation effects.

Keywords

• Fragmented landscape
• Habitat loss
• Species abundance
• Metapopulation structures
• Machine learning