Detection of honey bees (Apis mellifera) in hypertemporal LiDAR point cloud time series to extract bee activity zones and times

Abstract. Given the vital role bees play in our ecosystems and their increasing endangerment, it is highly important to develop new methods that assist in gaining a deeper understanding of the spatial dimension of insect behavior. Conventional methods for monitoring bees are subject to accuracy limitations, experimental setup complexity, and lack the explicit spatial dimension. This study presents a novel approach for detecting and identifying honey bees (Apis mellifera) and Asian hornets (Vespa velutina) using hypertemporal LiDAR point clouds. We employed an experimental setup of a permanent terrestrial laser scanner (Riegl VZ-600i) to capture point clouds in a region of interest of 3 × 2 × 5 m at regular intervals (30 s) over ca. 1.8 h. By training a random forest classifier based on local neighbourhood features, the classified points can then be clustered in single and distinct objects of bees/hornets. Ultimately, a simple logical operator is employed to ascertain whether an object is a bee or hornet, according to definable knowledge-driven thresholds (e.g., size of bees). Our proposed method demonstrates high accuracy and precision in bee (n = 7,084, acc. = 97.44%, prec. = 99.07%) and hornet detection (n = 296, acc. = 87.71%, prec. = 67.65%), offering a fully automatic and 3D spatial monitoring alternative to traditional techniques. Furthermore, it allows for the identification of insect activity zones and times, as well as their relative change over time. We could identify zones of bee activity in front of the hive with observable flying slowdowns before entering and defensive behaviors in response to predators. This approach provides new insights into the spatial and temporal dynamics of insect populations, especially in the context of environmental and climate change.