Development of a Predictive Model of Honey Bee Foraging Activity Under Different Climate Conditions

Pollination services provided by managed honey bees are essential to a variety of agricultural crops. In early spring, 70% of the nation’s beehives are rented by almond growers for pollination. The bee hour, which is defined as an hour with temperature above $55^{\circ}\mathrm{F}$, wind speed below 15 mph, and no rainfall, has emerged as a tool for almond growers to predict pollination. This allows them to plan treatment to extend the bloom period if forecasted bee hours are low and time pesticide sprays to reduce the risk of honey bee exposure to pesticides, one of the leading causes of colony loss. However, as a binary classification, the bee hour calculation does not offer high precision. In this research, a machine learning model is developed to predict foraging activity under different climate conditions. In total, 57,333 honey bee foraging trips were recorded and analyzed through radio frequency identification (RFID) tracking of 1,508 foragers in summer, fall, and early spring, and 9 climate factors were recorded using a weather station. To take advantage of the temporal nature of the dataset, two recurrent neural network architectures, Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), were selected to analyze the data. GRU performed best, achieving R² scores of 0.84-0.91 across seasons after optimization. Compared to the bee hour method, the model provided accurate and precise predictions of foraging activity. Additionally, two potential novel discoveries were made. Firstly, a correlation analysis between several foraging activity metrics and weather revealed that metrics based on the time spent outside by individual foragers have a significantly stronger correlation with weather than metrics based on ingress and egress rate. Furthermore, the model identified multiple days of significant foraging activity under weather conditions previously considered non-permissive that all followed days of low or no foraging activity, suggesting that honey bees may adjust foraging behavior in response to a previous lack of foraging.