Automating Phenological Stage Detection from Citizen Science Images for Plant Phenology Monitoring

Plant phenology, the study of seasonal events in plants' life cycles such as budburst, flowering onset, leaf-out, fruit ripening, and senescence, is intrinsically linked to climatic conditions and plays a crucial role in ecosystem processes like carbon and nutrient cycling. Due to its ecological importance, many countries have established phenological monitoring networks based on systematic protocols. However, declining volunteer participation in recent decades has raised concerns about the continuity of these invaluable datasets.

Advancements in technology, machine learning, and smartphone accessibility have spurred the development of plant identification apps. These apps enable users to identify plant species without prior botanical knowledge, generating vast datasets of plant occurrences.

This study investigates the potential of applying machine learning to citizen science-derived plant image data for phenological monitoring. By utilizing a pre-trained deep learning model, we extracted relevant image features and classified 39 species-specific phenostages for nine common plant species in Germany using a Support Vector Machine (SVM) classifier. Our model achieved an impressive overall accuracy of 96%, enabling the automated annotation of over 600,000 plant occurrence images from the Flora Incognita app into corresponding phenological stages.

With this approach, not only did we capture additional fine-granular phenostages, such as flower bud and unripe fruit stages, which are less commonly resolved in traditional phenological network datasets, but we also observed the interannual variability of each phenostage across different years. This demonstrates the feasibility of integrating opportunistic citizen science data into phenological monitoring schemes. By addressing the challenges posed by declining volunteer participation, this method significantly enhances the temporal and spatial resolution of phenological datasets, offering innovative opportunities for phenology monitoring and ecological research.