Food codes as an AI-based framework to investigate bird foraging ecology and interspecific interactions

Food codes activated by provisioning food through a garden bird feeder have proven to be an effective tool for investigating bird behavior and bird-human interactions. The experimental setup included a garden feeder with carefully controlled variables, such as its location and structure, the type of food provided, and its temporal distribution. A total of 2.8 million image frames of birds at feeder were captured between November 2023 and May 2024 using a time-lapse camera system. Of these, 1,232,456 frames were classified through supervised image processing using the Squeezebrains SDK (https://fabervision.com), an AI-based image analysis tool specifically developed for wildlife monitoring. This approach enabled detailed insights into feeding preferences and interspecific interactions. Nine bird species were identified: Great tit (Parus major) (65.59 % of all visits), blue tit (Cyanistes caeruleus) (13.62 %), house sparrow (Passer domesticus) (8.28 %), red-billed leiothrix (Leiothrix lutea) (5.90 %), siskin (Carduelis spinus) (2.84 %), chaffinch (Fringilla coelebs) (1.69 %), dunnock (Prunella modularis) (1.00 %), European robin (Erithacus rubecula) (0.59 %), and common blackbird (Turdus merula) (0.45 %). These species, belonging to different genera, visit the feeder at different times. The cluster analysis has categorized the bird species into three distinct groups based on their temporal preferences, leaving the chaffinch as an unclustered species: Group 1 (Blue tit, great tit, house sparrow, red-billed leiothrix), group 2 (Dunnock, siskin), group 3 (Common blackbird, European robin). The potential interspecific competition, estimated by the number of frames shared concurrently by two species, was highest for the blue tit, great tit, and red-billed leiothrix, respectively. The chaffinch, siskin, common blackbird, European robin, and dunnock exhibited the highest scores for intraspecific, non-shared frames. Severe weather conditions appear to increase the number of visits to feeders. Overall, food codes represent a robust approach, providing valuable insights into bird community dynamics while offering perspectives relevant to ecological management.