Simulating Honey Bee Large‐Scale Colony Feeding Studies Using the BEEHAVE Model—Part I: Model Validation

Abstract

In pesticide risk assessments, semifield studies, such as large-scale colony feeding studies (LSCFSs), are conducted to assess potential risks at the honey bee colony level. However, such studies are very cost and time intensive, and high overwintering losses of untreated control hives have been observed in some studies. Honey bee colony models such as BEEHAVE may provide tools to systematically assess multiple factors influencing colony outcomes, to inform study design, and to estimate pesticide impacts under varying environmental conditions. Before they can be used reliably, models should be validated to demonstrate they can appropriately reproduce patterns observed in the field. Despite the recognized need for validation, methodologies to be used in the context of applied ecological models are not agreed on. For the parameterization, calibration, and validation of BEEHAVE, we used control data from multiple LSCFSs. We conducted detailed visual and quantitative performance analyses as a demonstration of validation methodologies. The BEEHAVE outputs showed good agreement with apiary-specific validation data sets representing the first year of the studies. However, the simulations of colony dynamics in the spring periods following overwintering were identified as less reliable. The comprehensive validation effort applied provides important insights that can inform the usability of BEEHAVE in applications related to higher tier risk assessments. In addition, the validation methodology applied could be used in a wider context of ecological models.

Simulating Honey Bee Large‐Scale Colony Feeding Studies Using the BEEHAVE Model—Part II: Analysis of Overwintering Outcomes

“…This was repeated for honey stores (in kg, where 1 kg ≈ 2000 cells), adult bees, pupae, and the ratio of adult bees to honey stores. Pollen, larvae, and egg measures were not used due to high measurement error (see Schmolke et al 2020, this issue). Data on Varroa destructor mite and Nosema spp.…”

“…The ranges of 4 of these aspects of colony variability were tested one by one systematically with BEEHAVE as follows: 1) colony conditions observed at the time of study initiation, including initial number of adult worker bees, brood, and honey and pollen stores per hive; 2) feeding timing and amount, including both treatment feedings and supplemental feedings with sugar syrup; 3) landscape composition around apiaries, reflecting spatial and temporal bee resource availabilities; and 4) weather impacting daily foraging hours available to simulated foraging bees. Each is represented by a BEEHAVE input file unique to each scenario (discussed in Schmolke et al 2020, this issue). In this initial set of BEEHAVE simulations, all 4 aspects were examined for potential impact on fall colony conditions using the calibrated BEEHAVE model (Schmolke et al 2020, this issue).…”

“…Supplemental sugar was provided to the colonies between August and April as standard practice to help achieve sufficient honey stores for winter survival. A more detailed description of the LSCFSs used for the present study can be found in Schmolke et al (2020; this issue).…”

“…The 5th, 50th, and 95th percentiles of initial honey stores and initial numbers of adult bees were used as inputs in these simulations. The initial measures of pollen, pupae, and uncapped brood (larvae + eggs) were omitted due to their considerably smaller effects, as found in an earlier analysis (see Supplemental Data, Section 4.0 and Figure S24); pollen, larvae, and eggs also had high measurement error (see Schmolke et al 2020, this issue). Instead, these metrics were assigned their median values in all scenarios.…”

“…The BEEHAVE model has explicit representations of foraging in the landscape and in‐hive processes such as brood raising (i.e., development of eggs, larvae, and pupae; Becher et al 2014). Environmental conditions and beekeeping activities can be explicitly represented in the model, which was previously calibrated and validated with data from the control colonies of the 7 LSCFSs (see details in Schmolke et al 2020, this issue). Data from these colonies were grouped and analyzed over time, showing that the last assessment in the fall prior to overwintering was the most predictive of overwintering outcome.…”