Speaker
Description
Plant-pollinator interactions are a candidate Essential Biodiversity Variable (EBV) but remain absent from the eLTER Standard Observations and from the operational core of EuropaBON, largely because traditional sampling is labour-intensive, costly, and requires taxonomic expertise that is in increasingly short supply. The Nature Restoration Regulation and the proposed EU Pollinator Monitoring Scheme recently created a legally backed demand for scalable, standardised data. In response, the Species Interaction Ecology (SIE) group at iDiv (UFZ and MLU) has developed integrated computer-vision pipelines designed to augment (not replace) traditional methods, enabling dense spatiotemporal observations and lower per-site costs.
This poster synthesises several interconnected components. For field acquisition, we developed an open-source smartphone time-lapse protocol alongside Insect Detect DIY camera traps (with on-device real-time detection and tracking) and a night-vision variant for nocturnal pollinators. For detection and classification, we trained pollinator detectors for automated annotation and in-field deployment, and CNN image classifiers with uncertainty quantification for 15 European fly-pollinator families. For honest assessment, we performed an explicit out-of-distribution evaluation on smartphone images captured in field conditions, and within the SEPPI consortium (Biodiversa+), we compare automated and traditional sampling across eight European countries along various ecological gradients (e.g. altitude, restoration, fragmentation, land-use intensity). For fine-grained taxonomy, ongoing work targets Syrphidae genus and species classification.
We also contribute to community best-practice guidelines for insect sensing systems and are planning to extend the pipelines to PollObs within the PhenObs botanical-garden network (Jena, Halle, Leipzig).
Code, trained models, technical documentation, and protocols are openly available. We welcome ecological use cases, image datasets, taxonomic expertise, and dialogue with monitoring schemes, conservation practitioners, land managers and other stakeholders on which biodiversity indicators they need to deliver, so AI tools can be co-designed with end users.
| Status Group | Postdoctoral Researcher |
|---|