Tool-Explore: exploring new tools for monitoring of alpine plant-pollinator interactions
2022 - 2023
Cooperation FinancingAlpine plant and insect communities are undergoing rapid changes due to climate warming, with species shifting their elevational ranges in response. These shifts can disrupt tightly linked interactions, for example, if insects emerge before plants bloom, it can lead to mismatches that affect both plants and pollinators. However, predicting how these changes will impact the structure and function of alpine interaction networks remains a challenge.
Automating Plant–Pollinator Observations ¶
Field observations are traditionally used to study plant–pollinator interactions, but these methods are time-consuming and provide limited temporal coverage. Recent advances in affordable video technology and computer vision have opened new possibilities for automated monitoring. Combined with machine learning, these systems can potentially identify insect visitors directly from images—but their application in remote, high-elevation environments is still limited.
In this project, we are testing and refining methods for observing pollinators in the field to track changes in community structure and ecosystem function over time. Building on the work of Droissart et al. (2021), we developed a rugged, low-power camera system based on Raspberry Pi, designed for long-term deployment under alpine field conditions. The system was field-tested at the Furkapass (UR) in summer 2023, with software improvements focused on reliability and stability.
Below you will find some sample clips of the videos recorded by the cameras:
Citizen science and AI for detecting pollinators ¶
Using this camera system, we recorded over 3,000 hours of footage. Insect visits are relatively rare in high alpine environments, meaning that most of the recorded images contain no insects. Reviewing this footage manually is highly time-intensive, and current AI models struggle with the specific challenges of our data: small, fast-moving insects, dynamic vegetation, and changing lighting conditions.
To address this, we launched a citizen science project on the Zooniverse platform: Alpine Bug Shot.
Volunteers from around the world contributed over 230'000 annotations, marking the presence, identity and position of insects in images. These valuable contributions are now being used to train new AI models capable of detecting insect visitors in complex visual environments. This will greatly reduce the time needed to turn raw footage into ecological insights.
We are developing these models in collaboration with the Swiss Data Science Center (SDSC) as part of the InterDetect project: https://www.datascience.ch/projects/inter-detect.