Using agent-based models to predict pollen deposition in a dioecious crop

Abstract

AbstractPollination involves complex interactions between plants and pollinators, and variation in plant or pollinator biology can lead to variability in pollination services that are difficult to predict. Models that effectively predict pollination services could enhance the ability to conserve plant-pollinator mutualisms in natural systems and increase crop yields in managed systems. However, while most pollination models have focused either on effects of plant or pollination biology, few models have integrated plant-pollinator interactions. Moreover, crop management causes variation in plant-pollinator interactions and pollination services, but management is rarely considered in pollination models. Here we used extensive datasets for kiwifruit (Actinidia chinensis var. deliciosa) to develop an agent-based model to track insect-provided pollination services with variation in crop cultivars, pollinator traits, and orchard layouts. This allowed us to predict pollination outcomes in a dioecious crop under a range of management scenarios. Our sensitivity analysis indicated that flower density and the proportion of female flowers are the most important factors in successful pollination, both of which growers control via cultivar selection and cultural management practices. Our analysis also indicated that economically viable pollination services and crop yields are attained with ∼60% female flowers and a peak foraging activity of 6 to 8 bees per 1,000 open flowers with diminishing returns for additional pollinators. The quality of pollination service varied across simulated orchard layouts, highlighting the potential use of this model as a framework to screen novel orchard configurations. More broadly, linking complex plant and pollinator interactions in pollination models can help identify factors that may improve crop yields and provide a framework for identifying factors important to pollination in natural ecosystems.HIGHLIGHTS- We develop a model using extensive empirical datasets to predict pollen deposition based on the interactions between flowers and pollinators in a dioecious crop system- We conducted a thorough sensitivity analysis, and analysis of the effect of stochastic variance between model runs, which can be used to inform future design of stochastic agent-based models- Our model effectively predicted the outcomes of varying management regimes of orchard layouts and pollinator introductions on pollination in a dioecious crop- Our model can be extended for other functionally dioecious crops or plant communities where managers want to understand how their decisions impact pollination