RoofPoll – Automated Pollinator Monitoring on Green Roofs

RoofPoll is an interdisciplinary research project at UFZ that combines sensor technology, camera systems, and artificial intelligence to automatically monitor pollinators on green roofs. The project aims to improve our understanding of urban biodiversity and develop new approaches for continuous ecological monitoring.

Example camera recording showing automated tracking of a hoverfly (Episyrphus balteatus)

Example recording from the RoofPoll monitoring system showing automated tracking of a hoverfly (Episyrphus balteatus).

Weekly Pollinator Abundance

Daily updated chart showing weekly pollinator abundance on the UFZ green roof in Leipzig

Daily updated visualization of weekly pollinator abundance based on camera observations collected on the UFZ research green roof in Leipzig.

As urban areas expand, green roofs provide important habitats for insects. RoofPoll investigates which pollinator species use these environments, how they behave, and how their activity changes over time.

The project integrates ecological research with technological and data-driven approaches to enable efficient, long-term biodiversity monitoring. A key aspect is the development of non-invasive monitoring methods: insects are observed in their natural behavior without being captured or disturbed.

At the core of RoofPoll is a compact “all-in-one” monitoring unit that integrates:

  • A camera to capture insect activity
  • Weather sensors (temperature, humidity, wind, rainfall)
  • An on-site computing unit for automated image analysis

The system operates autonomously and uses AI-based methods to detect and classify pollinators directly on site.

  • Species and taxonomic groups
  • Activity patterns (daily and seasonal)
  • Timing of occurrences (first and last observations)
  • Frequency and behavior
  • Environmental conditions influencing activity

These data provide new insights into the dynamics of urban pollinator communities.

In the coming phases, the system will be further refined, detection accuracy improved, and additional data streams (e.g. acoustic data) integrated.