Abstract
Japanese native bumblebees (Bombus ignitus) are widely used as pollinators in Japanese greenhouse crop production; however, their diurnal activity patterns and responses to greenhouse environmental conditions remain insufficiently quantified. We integrated artificial intelligence (AI)-based continuous monitoring with Poisson and negative binomial generalised linear models (GLMs) and generalised linear mixed models (GLMMs) to quantify environmental and temporal associations of inbound and outbound bumblebee movements from a nest box in a commercial strawberry (Fragaria × ananassa) greenhouse. The monitoring system detected bumblebee activity with high accuracy during the experimental period, with precision, recall, and F1-score values of 0.978, 0.991, and 0.984, respectively. Both inbound and outbound movements exhibited pronounced diurnal patterns, increasing rapidly in the morning, peaking around midday, and declining gradually toward evening. Based on model comparison, the negative binomial GLMM was selected as the final model. Light intensity and nest temperature were positively associated with bumblebee activity. After adjusting for hour-of-day effects, air temperature was negatively associated with activity, whereas relative humidity showed no clear association. These findings demonstrate that AI-based monitoring combined with environmental and temporal modelling can convert nest-box traffic data into a quantitative indicator of colony activity. This proposed framework may support smart greenhouse management by enabling data-driven assessments of pollinator activity and informing decisions related to pollinator management and environmental control.