Fig. 1 (a) Nesting Osmia bicornis female bee marked with an ID tag (unique color-digit combination) attached to its thorax, (b) nesting unit composed of layers (wooden boards) with 10 cavities each, (c) layer with cavities covered with plastic foil for which nesting progress and offspring production can be tracked, (d) specifications of layers used for nesting units: black :170 mm; red: 18 mm, green: 11.5 mm; blue: 9 mm; yellow: 8 mm
Results
The Bee Tracker software could successfully assign 67% of the alive females to a nest. The proportion of assigned females did not depend on video recording time (t = 1.25, P = 0.23) which was between 2 and 4 h. Of each video, the software generated nest_recognition files with a sample size of 80 h-1 on average while the flight_list files contained a mean sample size of 61 h-1. 96.1% of the checked events were detected correctly (including the color and digit of the bee ID, the cavity ID and the direction of the movement). This precision is valid for all extracted measurements: assignment of females to their nest cavity, flight duration and number of probed cavities. The 7 errors that were found all related to the bee ID. Three errors were caused by a wrong color detection: green was classified falsely as yellow in all these cases. The remaining 4 errors were caused by an ID swap between two bees that had crossing movement paths, which led to a commutation of the IDs between bees. Precision did not depend on the video that was used for the analysis (χ2 = 13.949, P = 0.45; Fig. 2).