In the Bee Lab, the forage mapping project is using camera-equipped drones and machine learning to create resource density maps of different landscapes. The main purpose of these density maps is to allow researchers see how the foraging behavior of honeybees depends on the resource distribution around their hive. What if we could use these density maps to simulate bee foraging behavior in software and compare our results to real-life observations?
This is the premise for a new project in the HMC Bee Lab. The completed program would allow researchers to input parameters such as colony size and a resource density map, and the simulation would display the interactions between bees, flowers, and their central hive. Individual-based simulations such as this have been built before to draw conclusions on bee foraging; this model will be particularly insightful for the bee lab because we can input our own resource density maps. The figure below outlines the major components that will help create the simulation, which will be explained in more detail below.
Process for generating honeybee simulations.
The software I have found to be both powerful and intuitive enough for the task at hand is called Repast Simphony, which is an agent-based modeling system that runs Java. Repast uses object-oriented programming to simulate interactions between instances of different classes. For my simulations, I plan to implement objects of three different classes: Bees, Flowers, and Hives.
The Bee class will be the backbone of the simulation. There will be five different states for every bee in the simulation: resting, wandering, targeting, harvesting, and returning. “Resting” is the state for every bee in the hive, and it is where bees consume nectar and regain energy. “Wandering” is the state in which a bee is flying aimlessly without a particular flower patch in mind, and bees in this state lose energy while looking for neighboring flowers. “Targeting” indicates a bee has found a flower and is flying toward it. “Harvesting” refers to bees sitting on flowers, and the food storage of those bees increases. Finally, “returning” occurs if a bee is losing energy or has collected enough nectar; these bees then return to the hive. The figure below displays the finite state machine diagram for the bee simulations.
Diagram of the Bee’s finite state machine.
These states form the foundation of the simulation program. After this is implemented, I may decide to add additional states such as “waggle-dancing” or perhaps even “going-to-the-bathroom”. In all seriousness though, complex honeybee simulations could be a valuable tool in accompanying and confirming field observations.
For example, let’s imagine the “waggle-dance” implemented in software. Of all the bees returning to the hive, a fraction of the bees that collected lots of food will perform the waggle dance. Some of the bees leaving the hive to forage will use that dance to skip the “wandering” state and instead enter the “targeting” state, which will allow them to find valuable resources faster. In the simulation environment, changing a single variable will turn the waggle dance behavior on or off, and by changing other variables such as resource density or colony size we can draw conclusions about the effectiveness of the waggle dance in different contexts.
This sounds ideal, but it is not possible to quickly build a waggle-dance feature in software; it must be defined and then analyzed in the field extensively. How often is the dance performed? How many bees (on average) observe the dance? How accurately are bees able to describe the distance and direction of resources through the dance? There are so many parameters that need to be calculated, and the only way to find these numbers is through real-life observations. Fortunately, past research has answered many of these questions and will be invaluable in building these simulations.
There are many characteristics of honeybees (such as the waggle dance) that must be analyzed in the field before they can be realistically simulated in software. The purpose of the simulation is to effectively combine all of these characteristics once they have been studied. Once combined, we can use the density maps generated by the forage mapping project to analyze the foraging behavior of honeybees. As Tessa and Clayton collect field observations of native pollinators, we can compare these observations to our simulation results and refine the model as needed. These comparisons ideally will teach us how colonies distribute foraging effort across different resources and the efficiency of this effort.
Further Reading
Agent-Based Models in Biology
Bee Foraging Behavior
Existing Simulation Models of Honeybee Foraging
Schmickl, Thomas, and Karl Crailsheim. 2004. “Costs of Environmental Fluctuations and Benefits of Dynamic Decentralized Foraging Decisions in Honey Bees.” Adaptive Behavior 12 (3-4): 263–77. doi:10.1177/105971230401200311.
Dornhaus, Anna, F Klugl, C Oechslein, F Puppe, and Lars Chittka. 2006. “Benefits of Recruitment in Honey Bees: Effects of Ecology and Colony Size in an Individual-Based Model.” Behavioral Ecology 17 (3): 336–44. doi:10.1093/beheco/arj036.
Beekman, Madeleine, and JB Lew. 2008. “Foraging in Honeybees—when Does It Pay to Dance?” Behavioral Ecology 19 (2): 255–61. doi:10.1093/beheco/arm117.
No comments:
Post a Comment