Format: ORAL

Ross D. Stewart1, N. Bard3, M. van der Bank2, T. Jonathan Davies2,3

1 Sol Plaatje University, Kimberly, South Africa 2 African Centre for DNA Barcoding (ACDB, University of Johannesburg, Auckland Park, South Africa 3 University of British Columbia, Vancouver, British Columbia, Canada

Phenological patterns, representing the timing of recurrent biological events, such as flowering, are essential for understanding plant life cycles, ecological interactions, and how species respond to climate change. However, given the diversity of South African flora (ca. 24,000 species), manually recording species phenology presents an immense challenge. In this study, we explore machine learning (ML) application to data sourced from the iNaturalist database and describe the flowering phenology of plants in the South African National Botanical Gardens. We generated a training dataset from 10,000 photographic images, encompassing a diverse range of species from various locations. Next, we applied a Convolutional Neural Network (CNN) to classify images as flowering versus non-flowering. Using metadata associated with each image, including the date the photograph was taken, we were able to derive the timing of peak flower production and length of the flowering season for each species in the database. Our analysis illustrates how ML can leverage the vast wealth of citizen science biodiversity data in South Africa and describe large-scale phenological dynamics across the region. Applying ML and other advanced data tools to big data provides an opportunity for more informed decision-making and sustainable practices in the context of biodiversity conservation and management.