Bee Tracker – an open-source machine-learning based video analysis software for the assessment of nesting and foraging performance of cavity-nesting solitary bees
Running headline: Video analysis software for solitary bees
Anina C. Knauer1*, Johannes Gallmann2 and Matthias Albrecht1
1 Agroscope, Agroecology and Environment, Zürich, Switzerland;
2 Ubique Innovations AG, Zürich, Switzerland
*Corresponding author: Anina Knauer, Reckenholzstrasse 191, 8046 Zürich, Switzerland, anina.knauer@agroscope.admin.ch
Abstract
  1. The foraging and nesting performance of bees can provide important information on bee health and is of interest for risk and impact assessment of environmental stressors. While radio-frequency identification (RFID) technology is an efficient tool increasingly used for the collection of behavioral data in social bee species such as honey bees, behavioral studies on solitary bees still largely depend on direct observations, which is very time-consuming.
  2. Here, we present a novel automated methodological approach of individually and simultaneously tracking and analyzing foraging and nesting behavior of numerous cavity-nesting solitary bees. The approach consists of monitoring nesting units by video recording and automated analysis of videos by a machine learning based software. This Bee Tracker software consists of four trained deep learning networks to detect bees that enter or leave their nest and to recognize individual IDs on the bees’ thorax as well as the IDs of their nests according to their positions in the nesting unit.
  3. The software is able to identify each nest of each individual nesting bee, which permits to measure individual-based measures of reproductive success. Moreover, the software quantifies the number of cavities a female enters until it finds its nest as a proxy of nest recognition, and it provides information on the number and duration of foraging trips. By training the software on 8 videos recording 24 nesting females per video, the software achieved a precision of 96% correct measurements of these parameters.
  4. The software could be adapted to various experimental setups by training it to an according set of videos. The presented method allows to efficiently collect large amounts of data on cavity-nesting solitary bee species and represents a promising new tool for the monitoring and assessment of behavior and reproductive success under laboratory, semi-field and field conditions.
Key words: behavior, fitness, Osmia bicornis , risk assessment, sublethal
Introduction
Bees provide pollination services to wild plants and crops and are essential for biodiversity and human food supply (Klein et al., 2007; Ollerton et al., 2011). They hold important flagship and indicator species and are used for the monitoring and impact assessment of environmental stressors such as habitat degradation, pesticide exposure or pathogens (Potts et al., 2010; Potts et al., 2016; Schönfelder & Bogner, 2017; Woodard et al., 2020). An important component in the evaluation of bee health is the assessment of reproductive success and foraging behavior, as key drivers of population development and provisioning of pollination services (Artz & Pitts-Singer, 2015; Ganser et al., 2020; Henry et al., 2012; Siviter et al., 2021). Such assessments requires, however, accurate and efficient tools to collect the often large amount of data required to assess bee health, especially if data on individual bees shall be collected (Crall et al., 2018; Nunes-Silva et al., 2019). Recent research and environmental risk assessments have mainly focused on the honey bee, Apis mellifera , and a few other social bee species (e.g. Bombus terrestris) as indicator species (Goulson et al., 2015; Potts et al., 2016). Only relatively recently research and risk assessments increasingly consider also other bee species for the monitoring of impacts of stressors on bee pollinators, prominently including cavity-nesting solitary bee species (Boff et al., 2020; Rundlöf et al., 2015; Stuligross & Williams, 2020; Zurbuchen et al., 2010). In Europe for example, the European Food Safety Authority (EFSA) has proposed to integrate two cavity-nesting solitary bee species, Osmia bicornis and O. cornuta for risk assessment of plant protection products on bees, including higher-tier assessments of sub-lethal effects on reproductive success (EFSA, 2013; Franke et al., 2021). This development has been fueled by the increased recognition of the fact that the effect of different environmental drivers can substantially vary between bee species and depend on their functional and life-history traits such as sociality, body size, foraging or nesting traits (Brittain & Potts, 2011; Sgolastra et al., 2019).
Bees can respond through changes in their nesting and foraging behavior to various environmental stressors as pesticides, habitat degradation or pathogens (Leonhardt et al., 2016; Li et al., 2013; Siviter et al., 2021). However, while foraging behavior of individuals of social bee species such as A. mellifera can automatically be recorded with RFID technology (Nunes-Silva et al., 2019), no such tool is, to our knowledge, currently available for the collection of such data for solitary bees. As studies with cavity-nesting solitary bees typically require nesting units with numerous scattered nesting cavities (Fig. 1), RFID, which has a short reach of the signal (Nunes-Silva et al., 2019), is difficult to implement. Furthermore, tracking foraging behavior and reproductive success of multiple individual females requires correct identification and assignment of the cavities used for nesting by individual females, which can only be achieved with a large number of readers at high costs. So far, studies on solitary bee species have therefore largely depended on direct visual observation to monitor foraging behavior or the nesting progress of individual females (Artz & Pitts-Singer, 2015; Franke et al., 2021), which is very time consuming, hampering research and environmental risk assessment with solitary model bee species.
Software can be used to automatically detect animals in images or analyze animal behavior recorded with videos (Eikelboom et al., 2019; Pennington et al., 2019). Here, we present a new machine learning based software, which can automatically extract and analyze data on the foraging and nesting behavior of individually marked, cavity-nesting solitary bees from videos. The Bee Tracker analyses videos of nesting units and records the entering and leaving of cavities by individually marked bees. The software is provided free and open-source including the underlying Python code, as well as a user manual, which makes the software also accessible to users who have no programming background. The above-mentioned measurements of bee behavior are provided as csv files and can easily be further processed (e.g. for statistical analysis). Additionally, the software creates visualization videos of the machine learning based analysis, which allows users to evaluate software performance including the precision of the provided measurements. The machine learning networks that permit to train the software and parameters of the input file can be adapted to specific requirements, which allows to use the software in a wide range of experimental setups.
Methods