Development and evaluation of a low-cost, automated camera trap for surveying bumble bee communities

Abstract

Widespread declines in insect diversity and abundance underscore an urgent need for standardized, nonlethal monitoring methods for important pollinators such as bumble bees (Bombus spp.). Camera traps are widely used for non-intrusive, continuous surveys of large animals but have not yet been extensively adopted for monitoring insects. An automated camera trap system could improve current insect survey methods, which often rely on lethal traps or in-person observation. We developed an open-source, low-cost camera trap for monitoring wild insects and evaluated its performance relative to established sampling approaches. The system used a low-power microcomputer to collect time-lapse images on colored platforms. We trained whole-image and tiled deep-learning object detection models to detect insects in images captured by the traps. We found that tiled inference models significantly improved detection accuracy and outperformed human review. Bumble bee visitation was highest on platforms featuring a fluorescent bullseye pattern; adding a fluorescent coat to blue platforms increased visitation modestly. With continuous monitoring, the cameras recorded bumble bee visits during all daylight hours. Over 18 days, camera traps recorded six Bombus species, yielding community composition and diversity estimates comparable to those obtained by hand netting and blue vane traps. Using our observed data, we simulated the effect of deploying variable numbers of cameras at sites with distinct levels of diversity. Adding cameras substantially increased sampling completeness, particularly in species-rich communities. Our findings demonstrate that low-cost, automated camera traps paired with deep-learning image analysis can enable scalable, nonlethal studies of bumble bee diversity and behavior. Our work establishes a foundation for monitoring other diurnal insect communities.