PhenoVision: Machine Annotations of Phenology for iNaturalist Plant Photos

Challenge

PhenoVision turns 27 million iNaturalist plant photos into machine-detected flowering and fruiting events. Using day_of_year, latitude and trait, can you map the phenological calendar of the world's plants: when do flowers and fruits peak, how does that timing flip between hemispheres, and how does the flowering season march across latitude bands? Then dig into a family or genus you care about and describe its own seasonal signature.

Where to start

The data is a single ~12 GB CSV at /srv/data/phenovision/annotations_all_w_headers_9cf8ad8.csv (do not load it whole). Open it with DuckDB (duckdb.connect(); read_csv_auto('.../annotations_all_w_headers_9cf8ad8.csv')) and aggregate on disk. Group count(*) by trait ('flower'/'fruit'), day_of_year (1-366), and a hemisphere flag from latitude (>=0 Northern, else Southern); convert day_of_year to a month for calendar plots. For a taxon deep-dive add a WHERE family = '...' or genus = '...' filter (e.g. family='Asteraceae', the most common). Useful extra columns: scientific_name, year, prediction_probability, coordinate_uncertainty_meters.

What to share

Share a short notebook or post with 2-3 charts: a flowering-vs-fruiting seasonal curve per hemisphere and a latitude x month flowering heatmap, plus a paragraph interpreting the hemisphere offset, the flower-to-fruit lag, and whatever your chosen family or genus reveals. State any caveats (all rows are High-certainty positive detections, some verbatim years look erroneous, and sampling follows where photographers are).

💬 Discuss this dataset, ask questions and share your results in its discussion thread. Challenges are open-ended — there's no single right answer.