Why we need more mechanisms in pollination models

Plant-pollinator systems are undergoing fast and significant modifications due to anthropogenic pressures and climate change. To remain insightful, predictive models of ecological interactions must thus be robust to these changing environmental conditions. Current pollination models dominantly infer statistical correlations from empirical data and propagate these correlations to novel environments (phenomenological modelling). Here, we argue that integrating more realistic biological mechanisms into pollination models (mechanistic modelling) is needed to achieve high levels of model robustness under changing conditions. We support this claim by highlighting recent case studies showcasing an advance in mechanistic modelling for three different aspects of animal-mediated pollination: plant-pollinator species interaction networks, spatial distribution and abundance of pollinator species, and pollination quality. We then discuss how to choose mechanisms and how to integrate them into models, based on the scientific question and the ecological scale under consideration.