Case Study: Application of an Ecological Model (BEEHAVE) to Large-Scale Honey Bee Colony Feeding Studies
In our June newsletter, we mentioned the recent release of a 2-part publication series in Environmental Toxicology and Chemistry presenting 1) a model validation of BEEHAVE and 2) an application of BEEHAVE. This month, we’re taking a deeper dive into these two publications and, through a case study, will show both how the BEEHAVE model can potentially be used for large-scale colony feeding studies (LSCFSs) and how it could inform future study designs in order to improve overwintering success in control hives as well as guide consistency across studies.
Part 1 of this series presents the validation of the BEEHAVE model with control data from large-scale colony feeding studies (LSCFSs). LSCFSs are studies used to assess the potential risks of pesticide exposure at the colony level and are typically conducted in higher-tier honey bee risk assessments. These studies are very costly and time intensive and carry significant resource concerns should unexpected issues arise. They also carry challenges such as potential high overwintering losses in untreated controls (observed in some studies). In addition, study designs can vary with respect to bee keeping activities and study schedules.
BEEHAVE is a mechanistic model of a honey bee colony and its interactions with the landscape. It provides a tool to systematically assess multiple factors (e.g., weather, landscape composition, and beekeeping activities) that influence colony outcomes and can inform study design. To use it properly, the model should be validated to demonstrate that it can appropriately reproduce patterns observed in the field prior to use.
In our study, we used untreated control data from seven LSCFSs that were conducted in North Carolina between 2014 and 2017 in BEEHAVE’s validation. Initial conditions of the colonies, the spatial and temporal bee resource composition of the landscape surrounding the study apiaries, the weather, and feeding of the untreated control colonies were used to parameterize the model. Two of the seven studies were used for the calibration of the model and the remaining five were used for validation of the model. The validation results suggested that the calibrated BEEHAVE model provided good agreement with apiary-specific data for the first study year. However, conditions in the springtime following overwintering presented a challenge for the model outputs since they did not match the observed colony conditions.
Based on the above observations, the calibrated BEEHAVE model is recommended to be useful in predicting a colony’s dynamics in LSCFSs prior to overwintering. Additionally, the results from the model are typically less variable than observations from colonies in the field. In other words, predictions from BEEHAVE should not be used as precise predictions of individual colony properties but rather applied to compare impacts of different environmental and beekeeping scenarios applied to colonies kept similar conditions. Our in-depth analysis informs the usability of BEEHAVE in applications related to higher tiered risk assessments and a path to model validation using multiple study data sets.
In Part 2, we present the application of BEEHAVE to simulate untreated control colonies in LSCFSs under a range of beekeeping and feeding scenarios. The goal was to identify the most important factors that impact control colony conditions in the fall, and inform study design aspects that increase the likelihood of overwintering success in control colonies. The composition of the landscape, feedings patterns, initial control colony conditions, and weather conditions were derived from the seven LSCFS. The following four aspects were examined for potential impact on fall colony conditions using the calibrated BEEHAVE: 1) colony conditions reported at the time of study initiation, 2) feeding timing and amount, 3) landscape composition around apiaries reflecting spatial and temporal bee resource availability, and 4) weather impacting daily foraging hours available to simulated foraging bees. Different inputs were used for each of the four aspects to represent the ranges of variability that across the available LSCFSs. Feeding schedules and initial conditions were most impactful to colony conditions in the fall and were targeted in more detailed simulation scenarios. Further analysis suggests the importance of colony conditions at study initiation and sugar feeding amounts and timing for colony fall conditions and subsequent overwintering success. The results can be used to inform a more standardized study design with increased likelihood of overwintering survival of controls.
This case study for applying a colony-level validated model to simulate a large-scale feeding study is a demonstration of how ecological modeling tools can be used to inform study designs for field-based, resource intensive studies.