The demand for agent-based models to explore the effects of environmental change on pollinator population dynamics is growing. However, models need a simple yet flexible interface to enable adoption by a wide range of stakeholders. We introduce BEE-STEWARD: a research and decision-support software tool, enabling researchers, policymakers, land management advisors and practitioners to predict and compare the effects of bee-friendly management interventions on bumblebee populations over several years. BEE-STEWARD integrates the BEESCOUT and Bumble-BEEHAVE agent-based models of bumblebee behaviour, colony growth and landscape exploration into a user-friendly interface, with reconstructed code, and expanded functionality. Bespoke automatic reports can be created to illustrate how different land management interventions can affect the densities of bumblebees and their colonies over time. BEE-STEWARD could be an important virtual test bed for scientists exploring the impacts of different stressors on bumblebees and used by those with little or no modelling experience, enabling a shared methodology between research, policy and practice.

World-wide declines in pollinators, including bumblebees, are attributed to a multitude of stressors such as habitat loss, resource availability, emerging viruses and parasites, exposure to pesticides, and climate change, operating at various spatial and temporal scales. Disentangling individual and interacting effects of these stressors, and understanding their impact at the individual, colony and population level are a challenge for systems ecology. Empirical testing of all combinations and contexts is not feasible. A mechanistic multilevel systems model (individual-colony-population-community) is required to explore resilience mechanisms of populations and communities under stress. We present a model which can simulate the growth, behaviour and survival of six UK bumblebee species living in any mapped landscape. Bumble-BEEHAVE simulates, in an agent-based approach, the colony development of bumblebees in a realistic landscape to study how multiple stressors affect bee numbers and population dynamics. We provide extensive documentation, including sensitivity analysis and validation, based on data from literature. The model is freely available, has flexible settings and includes a user manual to ensure it can be used by researchers, farmers, policy-makers, NGOs or other interested parties. Model outcomes compare well with empirical data for individual foraging behaviour, colony growth and reproduction, and estimated nest densities. Simulating the impact of reproductive depression caused by pesticide exposure shows that the complex feedback mechanisms captured in this model predict higher colony resilience to stress than suggested by a previous, simpler model. Synthesis and applications. The Bumble-BEEHAVE model represents a significant step towards predicting bumblebee population dynamics in a spatially explicit way. It enables researchers to understand the individual and interacting effects of the multiple stressors affecting bumblebee survival and the feedback mechanisms that may buffer a colony against environmental stress, or indeed lead to spiralling colony collapse. The model can be used to aid the design of field experiments, for risk assessments, to inform conservation and farming decisions and for assigning bespoke management recommendations at a landscape scale.

Social bees are central place foragers collecting floral resources from the surrounding landscape, but little is known about the probability of a scouting bee finding a particular flower patch. We therefore developed a software tool, BEESCOUT, to theoretically examine how bees might explore a landscape and distribute their scouting activities over time and space. An image file can be imported, which is interpreted by the model as a “forage map” with certain colours representing certain crops or habitat types as specified by the user. BEESCOUT calculates the size and location of these potential food sources in that landscape relative to a bee colony. An individual-based model then determines the detection probabilities of the food patches by bees, based on parameter values gathered from the flight patterns of radar-tracked honeybees and bumblebees. Various “search modes” describe hypothetical search strategies for the long-range exploration of scouting bees. The resulting detection probabilities of forage patches can be used as input for the recently developed honeybee model BEEHAVE, to explore realistic scenarios of colony growth and death in response to different stressors. In example simulations, we find that detection probabilities for food sources close to the colony fit empirical data reasonably well. However, for food sources further away no empirical data are available to validate model output. The simulated detection probabilities depend largely on the bees’ search mode, and whether they exchange information about food source locations. Nevertheless, we show that landscape structure and connectivity of food sources can have a strong impact on the results. We believe that BEESCOUT is a valuable tool to better understand how landscape configurations and searching behaviour of bees affect detection probabilities of food sources. It can also guide the collection of relevant data and the design of experiments to close knowledge gaps, and provides a useful extension to the BEEHAVE honeybee model, enabling future users to explore how landscape structure and food availability affect the foraging decisions and patch visitation rates of the bees and, in consequence, to predict colony development and survival.