Introduction
Movement capacity and foraging performance are key traits in ecology affecting survival and dispersal and have been explored in a large range of animals, like marine predators (Humphries et al., 2010), mammals (Clarin et al., 2013; Sinclair, 1992), birds (Naef Daenzer & Keller, 1999), and insects (Capaldi et al., 2000; Holway & Case, 1999; Sumpter & Pratt, 2003). The miniaturisation of technology has allowed the development of automated tracking devices (e.g. GPS, Argos; Kissling et al., 2014) surpassing the performance of traditional Movement Capacity Record (MCR) methods. However, automated tracking devices are often restrictive in terms of sample size (1 to 20 individuals in general) and size of animals targeted (related to the weight of the tracking device). Different tracking technologies can be selected depending on the locomotory mode of the targeted animal (e.g. flight, walk, swim), their environment (e.g. water, air, ground surface), the weight of the tracking device and its optimal attachment to the animal. But such tracking tools require individual manipulation and frequently the addition of extra weight can impact individual behavioural performance (Batsleer et al., 2020).
Image-based tracking is a good alternative that is increasing in use in animal ecology (Dell et al., 2014). These devices are not invasive as they do not rely on catching the individual nor the attachment of e.g. a microchip/GPS tag, and allow the tracking of several targets simultaneously, enabling the observation of complex behaviours and interactions between multiple individuals (Bozek et al., 2021; Gernat et al., 2018). Two dimensional (2D) image-based tracking to study foraging behaviour, learning, and/or vigilance of animals is quite common (Noldus et al., 2002; Peters et al., 2016; Wajnberg & Colazza, 1998). However, most animals move in three dimensions (3D). Birds, bats, and flying insects move in 3D in the air, as do fish and sea mammals in the water, which limits the accuracy and precision of 2D data recording. Given the ability to assess the distance from the object in all three dimensions, 3D-image based tracking can describe adjusted flight/swim behaviours (e.g. speed, curvature, orientation), even if individuals are close to one another (Campbell et al., 2008; Chiron et al., 2015). An additional advantage of 3D-imaging devices such as stereovision cameras is their ability to recover target positions directly in metric coordinates, as these systems are pre-calibrated in advance for a specific need (e.g. close focal length, wide field of view). In comparison, traditional 2D-image based tracking devices (e.g. common cameras) would need the use of a test chart to convert those 2D pixel expressed coordinates to metric coordinates, and would be less accurate with a varying 3rd dimension. 3D-image based tracking devices have been used to study Malaria mosquito flight for example (Spitzen et al., 2013), and bat flight patterns in different landscapes (Falk et al., 2014). Stereovision cameras are a tool often used in medicine (Skvara et al., 2013) and engineering (Gao et al., 2011; Huynh et al., 2016), but less in ecology (but see Theriault et al. (2010) and Matzner et al. (2020) for bats and birds, and Rachinas-Lopes et al. (2019) for water mammals, Chiron (2014) for insects, and more generally Straw et al. (2011)) although benefits are numerous when studying the behaviour of 3D-moving animals.
Here we applied 3D-image based tracking to a multi predator-prey relationship, focussing on two model species: the invasive Asian hornetVespa velutina nigrithorax (also known as the Yellow-legged Asian hornet; called “Hornet” in this study) and its prey, the Western honey bee Apis mellifera (called “Honey bee” in this study). We analysed time series in flight speed, flight curvature, and hovering of this hornet and honey bees in front of one beehive looking for potential effects of biotic and/or abiotic factors. We also question whether predation success could be linked with a mis-match in flight performances of both prey and predators.
Vespa velutina nigrithorax is an invasive alien hornet in Europe that now needs to be considered amongst the multiple stressors affecting honey bee survival (Monceau et al., 2014; Requier et al., 2019). On top of being a generalist predator of insects, it is capable of predating honey bees in high numbers in front of their hives, demonstrating a specific predation behaviour described as “hawking”, when the hornet hovers in front of the hive waiting for its prey (Monceau et al., 2014; Tan et al., 2007). This increasing predation pressure through summer and autumn is leading to heavy honey bee colony losses (between 5-80% of colony losses in France, 30% of colony loss in average; Kennedy et al., 2018; Requier et al., 2019), via two phenomena. The first one is the direct impact of predation which decreases the number of available foragers in the hive (Tan et al., 2007). The second is “foraging paralysis”, when the honey bee colony stops sending foragers out and there is a consequent decline in incoming food resources (Requier et al., 2020).
Using a stereovision camera, we carried out automated processing of 3D-image based tracks in the field to record 3D-adjusted behavioural parameters and to focus on specific inter-individual interactions. This study aims at understanding the effects of predator density (the Asian hornet) on both the flight behaviour of predators and prey (honey bees), in a biogeographical context (western Europe) where the prey-predator interaction did not evolve or co-adapt. First, we developed an automated process to select “scenes of interest”, when both prey and predator are present on the screen at the same time. Such an automated process helps ecologists in video analyses and reduces potential observation bias. Second, we explored flight performances in terms of speed, curvature, and hovering by honey bees and hornets looking for potential drivers of predation success. We assumed that the flight performances of hornets and honey bees differ due to the morphological differences between these species. Moreover, we hypothesized that the predation success could be influenced by predator density due to a disturbance in flight performance of both predator and prey.

Material and methods