Speaker
Description
Pollen allergies affect around 20% of the population in Central Europe. Anthropogenic changes, such as pollution and warming, are intensifying symptom severity, making pollen allergies a significant public health concern. Many allergy sufferers rely on pollen forecasts to avoid allergen exposure, yet these forecasts often lack sufficient local-scale resolution. Since the flowering phenology of allergenic plants is one of the best predictors of local pollen concentrations, fine-scaled phenological data hold considerable potential for improving local pollen load estimations. Here, we leverage this potential by developing automated recognizers of fine-scaled flowering stages in the key allergenic species of Central Europe using citizen science images. We launched the citizen science project "PollenNet" within the plant identification app Flora Incognita during the pollen season 2024/25, where users are invited to contribute images of flowering key allergenic species, including hazel, alder, birch, grasses, mugwort, and ragweed. To date, citizen scientists have contributed over 61,000 images of hazel, 12,000 of alder, 25,900 of birch, and 10,000 of grass species to the project. We use these images to develop automated classifiers capable of recognizing flowering stages that correspond with the release of pollen. For each species, we train a classifier using feature representations extracted from a pre-trained convolutional neural network (CNN). For hazel, the classifier achieved an overall accuracy of >96% in differentiating four detailed flowering stages. The resulting classifiers will provide spatio-temporal phenological data that will ultimately help improve local pollen load estimations and enable near real-time phenological monitoring with Flora Incognita data.
| Status Group | Doctoral Researcher |
|---|