Computer vision reveals flower visitor interactions are mediated by agroforestry management, driving cocoa yield

Cocoa ( Theobroma cacao L.) is a multi-billion-dollar crop affected by climate change. Improving poorly understood pollination services in agroforests offers a pathway to increase yields sustainably. Here, we combine embedded computer vision and structural equation models to assess interactions among flower visitors, shade-tree diversity, canopy cover, and fruit set in cocoa systems of China and Brazil. We show that flowers are visited by arthropods functioning as nectar and pollen foragers (23.1%), herbivores (6.1%), predators (2.3%), and multifunctional visitors (63%), showing that pollination extends beyond specialized pollinators. Forager activity increases with shade-tree diversity and canopy cover. In China, foraging midges enhance fruit set, while in Brazil multifunctional ants play a direct role via plant-arthropod interactions. In China, higher canopy cover reduces aphids in flowers and weakens aphid-ant pressure on midges, whereas in Brazil it is associated with reduced fruit set. Overall, agroforestry management can enhance pollination and support sustainable cocoa production under climate change.