Pollination is critical to many ecosystem services, such as the maintenance of biodiversity and human nutrition. Addressing how plant-pollinator interactions change across environmental gradients requires quantifying pollinator visitation. We need new tools that allow for rapid data collection to increase the spatial and temporal sampling that is possible.
Ecologists typically observe pollinators in the field, record the species identity of the easily distinguished individuals, and collect with a net the remaining individuals so that they can be later identified to species in the lab. This is labor and time intensive, and thus prevents the massive data collection that is necessary to achieve a general, predictive understanding of how pollination responds to global change.
We use machine learning methods and technology to automatize ecological data collection on plant-pollinator interactions.
A first step is to develop and test a camera system capable of capturing clear images of visiting insects on flowers. Our first results use mobile phones for time-lapse photography, capturing images of flowers every 1-2 seconds with the goal of capturing images of insects visiting flowers (Fig .1). The pollinators in the collected images are identified by entomologists in our group down to the lowest possible taxonomic level (order > family > genus > species).
The second part of the project involves training state of the art Convolutional Neural Networks (CNN) with the goal of insect identification down to the lowest possible taxonomic level (Fig. 2). For this, we make use of our field images as well as images from online databases (e.g., iNaturalist and Observation.org). We focus on four orders of insect pollinators: Coleoptera, Diptera, Hymenoptera and Lepidoptera.
Our main goal is to develop an automated pipeline that can quantify the abundance and diversity of insects that visit flowers and the structure of plant-pollinator interactions across various environmental gradients in Europe.
How can you contribute?
We also welcome your thoughts, ideas and collaboration.
DLR: Prof. Hannes Taubenböck, Thomas Stark, Dorothee Stiller, Dr. Michael Wurm & Robin Spanier
Boussioux, L., Giro-Larraz, T., Guille-Escuret, C., Cherti, M., & Kégl, B. (2019). InsectUp: Crowdsourcing Insect Observations to Assess Demographic Shifts and Improve Classification. arXiv preprint arXiv:1906.11898.
Steen, R. (2017). Diel activity, frequency and visit duration of pollinators in focal plants: in situ automatic camera monitoring and data processing. Methods in Ecology and Evolution, 8(2), 203-213.
Robin Spanier presents the results of his master thesis at the Helmholtz AI conference.
Valentin Ștefan started a 3 months research stay at the DLR in the team of Prof. Hannes Taubenböck, City and Society. The research stay is funded by the Helmholtz Information & Data Science Academy - HIDA Trainee Network.
Find out more about our project in the Helmholtz AI News - "Helmholtz AI project call showcase: Pollination Artificial Intelligence. Transferring existing large-scale AI methods to the field of pollination ecology for the automatic classification of insect pollinators."
We kicked off our UFZ-DLR collaboration for Pollination Artificial Intelligence (PAI).
Many thanks to the participants: Thomas Stark, Dorothee Stiller, Robin Spanier, Michael Wurm, and Prof. Hannes Taubenböck from DLR, and Valentin Ștefan and Tiffany Knight from UFZ/iDiv.
Robin Spanier started working on his MSc thesis " Pollination AI, Deep Learning approach to identify pollinators and their taxa using the YOLO architecture"
Our project PAI - Pollination artificial intelligence is one of the 17 funded projects selected by the Helmholtz AI.
This sets our collaboration with the team of Prof. Hannes Taubenböck, City and Society.