More than twenty-five years ago, Tom Seeley argued in his influential book The Wisdom of the Hive that the honey bee colony is a system complex enough to demand computer simulation if we ever hope to predict its responses to stimuli. That vision has recently gained institutional weight: the European Food Safety Authority has endorsed the development of a mechanistic colony model as the foundation for environmental risk assessment of multiple stressors to bees at the European scale. The model now under construction, called ApisRAM, uses an agent-based modelling approach in which individual bees are represented as digital agents with their own behavioural rules, interacting with each other, with the hive’s physical and chemical properties, and with the external environment. The colony-level outcomes that emerge from these interactions are not programmed in directly; they arise from the bottom up, as consequences of thousands of individual decisions.

Agent-based models of bee foraging are not new. Researchers have previously explored the metabolic costs of foraging, the exploitation of rewarding resources, the behavioural rules governing recruitment, and the feedback mechanisms that control foraging through nectar receivers. But these earlier efforts shared serious limitations. The landscapes in which model bees foraged were simplistic, often containing only a handful of food sources. Weather variables were either ignored or disconnected from the landscape representation. Crucially, none of the previous foraging models fully integrated the spatial and temporal dynamics of landscape resources, and the role of the colony itself in shaping forager numbers and behaviour was rarely considered at the individual level. Studies evaluating colony strength and production, meanwhile, seldom accounted for environmental conditions at all, typically defining landscape structure with broad land-use categories like urban, grassland, or forest, without any quantitative assessment of the resources actually available to bees.

The new foraging sub-model, developed as part of ApisRAM and described in the Food and Ecological Systems Modelling Journal, overcomes these constraints by embedding foraging bees within the Animal, Landscape and Man Simulation System, known as ALMaSS. This landscape representation typically operates on a ten by ten kilometre window at a spatial resolution of one square metre, combining detailed land-use and land-cover mapping with information on farming systems, farm practices, weather, and plant growth. Fields belong to farm units classified by type, each with associated crop rotation plans that allow realistic modelling of crop patterns changing in space and time. Country-specific management plans describe soil cultivation, fertiliser, and pesticide applications within time windows and probabilities of occurrence. Vegetation growth models supply daily biomass figures for all modelled vegetation types and crops, driven by weather conditions such as mean daily temperature.

Layered onto this dynamic landscape are floral resource models that describe pollen, nectar, and sugar production levels and their changes throughout the year for every vegetated polygon. Flowering time and resource production are tied to accumulated growing degree-days, calculated from daily average temperatures. For annual crops, degree-days accumulate from the sowing date using crop-specific thermal requirements; for other vegetated elements, the models superimpose the production of all individual plant species composing a given habitat. The result is daily information on nectar and pollen quantity in milligrams per square metre, with nectar quality defined as sugar content. When a model bee forages at a location, the foraged amount is depleted from the corresponding resource map, so the landscape genuinely responds to bee activity. Hourly weather data, including wind speed, rainfall, temperature, and solar radiation, complete the environmental picture delivered to each foraging agent.

The behavioural rules for individual model bees were defined with care. Low temperatures below ten degrees Celsius, darkness, rain, and strong winds exceeding twenty-five metres per second prevent foraging or scouting. Under favourable conditions, the possibility of flight is determined by hourly solar radiation and temperature, following an exponential equation relating the radiation threshold to ambient temperature. Each model bee has a maximum flying distance of twenty kilometres per day and can carry fifty milligrams of nectar or eight milligrams of pollen per flight, defined as a full crop. If a bee cannot fill its crop within a single square metre of a resource polygon, it performs searching flights in the surrounding square metres, up to a maximum local area of twenty-five square metres, before the flight is counted as unsuccessful. Successful bees memorise the relative position, quantity, and quality of the resources they found, and bring this information back to the colony.

The researchers then tested four distinct strategies that model bees could use when deciding where to forage. The distance strategy prioritised the closest polygons to the colony, reflecting evidence that honey bees have poor stereo vision and that smaller colonies tend to forage closer to the hive early in the season. The quality strategy, applied to nectar, prioritised polygons with the highest sugar content, drawing on the bee’s gustatory sensilla that allow rapid assessment of sugar levels. For pollen, where no single quality factor has been clearly identified, the equivalent strategy prioritised quantity. The random strategy had bees choose randomly among polygons visited during scouting trips, acknowledging that the mechanisms driving real scouting choices remain poorly understood. Finally, an energetic efficiency strategy, tested only for nectar, prioritised polygons offering the best balance between energy gained from the resource and energy spent travelling to it, calculated from sugar quantity and distance from the colony.

The simulations, run across one hundred different colony locations within a real Danish landscape near Ringkøbing and repeated to produce twelve hundred one-year simulations, yielded striking results. When scout bees applied the distance strategy, the colony collected the highest average amounts of sugar, between ninety-one and one hundred twenty-five kilograms depending on the foraging strategy, but at the cost of many daily flights with a lower success rate, and with enormous variability in outcomes. This variability revealed that distance-prioritising bees could find the most profitable foraging sites only locally, not at the landscape level, exploring habitats further afield only after the closest resources were depleted. Contrary to expectations, the energetic efficiency strategy did not deliver the highest sugar haul; bees using it performed fewer but more successful flights, most likely because a twenty-kilometre daily flight budget limited their trips. For pollen, the highest collection, between twenty-six and thirty-four kilograms on average, came when the distance strategy was used by scouts or foragers, with the unexpected best combination being distance scouting coupled with random foraging, which kept social information flowing and resources depleting slowly.

Perhaps the most important lesson concerns the scout bees, whose supposedly minor role proved decisive: the choice of scouting strategy influenced the annual resource collection of the whole colony for both nectar and pollen. The researchers conclude that foraging models must include a reliable implementation of scouting strategies, and that nectar foraging should not be determined by distance alone, since such a strategy ignores resource profitability. Future development will integrate pollen quality and diversity, refine how bees weigh private versus social information when deciding to switch foraging locations, and couple the foraging sub-model to other colony processes like food processing and brood rearing. Field-collected data from EFSA projects will be used to validate the tested strategies against real colony populations, weather, resource supply, and honey gains.

The significance of this work extends well beyond model calibration. By predicting the distribution of foraging bees and their use of resources within a dynamic landscape, the ApisRAM foraging sub-model aims to correctly predict exposure pathways to pesticides and other stressors, providing regulators with a tool that captures the feedback loops between environment and colony. The study also highlights how much fundamental behavioural knowledge is still missing, from the drivers of scouting decisions to the thresholds that cause bees to abandon known foraging sites, and makes a strong case for modern tracking technologies to fill these gaps. What emerges is a portrait of the colony as a genuinely emergent system: its success in gathering food arises not from any central plan, but from the accumulated decisions of individuals navigating a landscape that changes beneath their wings.

Subject of Research: Modelling foraging strategies of honey bees as agents in a dynamic landscape representation

Article Title: Modelling foraging strategies of honey bees as agents in a dynamic landscape representation

Article References: Modelling foraging strategies of honey bees as agents in a dynamic landscape representation. (n.d.). https://doi.org/10.3897/fmj.5.99103

Image Credits: AI Generated

DOI: 10.3897/fmj.5.99103

Keywords: Modelling, foraging, strategies, honey, bees, agents, dynamic, landscape, representation, scientific research