Automated Phenostage Annotation with Community Science Plant Images workflow

summary: 

This study presents a machine learning workflow to automatically classify community-contributed plant images into several phenological stages: flowering bud, flower, unripe fruit, ripe fruit, and senescence. Using a small set of training images, we trained an SVM-based classifier that achieved 96% overall accuracy across nine common woody shrub and tree species. We validated our automatically annotated phenostage data using phenological observations from the German Meteorological Service (DWD), finding strong agreement in both spatial and temporal patterns.

The R code provided shows how we can use the extracted image features.

This study is published in the International Journal of Biometeorology.

DOI: https://doi.org/10.1007/s00484-025-02972-x