Green AIAI image-analysis of community-science photos (~87% expert agreement) — not a botanical record
In the garden
Herb layer — Sits in the herb of a layered food forest or polyculture.Open guide →
derived roles
AI = read by an AI vision model · DERIVED = a computed estimate, not a direct measurement. The “How we know this” section below details each.
The garden yearbloom → fruit → fall colour
J
F
M
A
M
J
J
A
S
O
N
D
Bloom
Bloom (the flower's colour)
Bloom · May – Aug — 197 obs · Herbarium specimens — Park et al. 2023 (CC BY 4.0)
Species characteristics from USDA PLANTS (public domain) + TRY (CC BY) — general guidance, not a
guarantee for your exact site. Deer "browsing" is documented palatability, not a deer-proof claim.
Named species (a documented Nearctic sample, not exhaustive) from NHM HOSTS (CC0).
Species thumbnails re-hosted from iNaturalist — Creative Commons, credited per image (hover for credit). Click any species to see it on iNaturalist. Not exhaustive; many taxa have no openly-licensed photo yet.
Same genus / family, ranked by ecological value — from the PlantKey catalog
How we know this (1)Methods & honest limits ▸
Flower colourDerived
McKenzie et al. assigned each species a flower colour with a GPT-4V vision model over iNaturalist photos. We use the confident tier plus a separately-labelled lower-confidence top-up (mostly inconspicuous green/brown flowers), and render it as a small tint — never as an asserted fact, and kept out of the written synopsis.
Honest limits: AI image inference (~87% expert agreement on the confident tier); a decorative, confidence-tiered indicator, not a measured trait.
McKenzie, P., Berardi, A.E., Hopkins, R. (2025). flower_color_phenology (MIT).
Derived values are computed from open data by a published method — labelled, cited, and never shown as a direct observation.
Full methodology →
Spot an error or have a better photo? Every field links to its source, and the whole dataset is open.
Cite this pageOpen data, please attribute ▸
PlantKey’s data is open under CC BY-SA 4.0 — free to reuse and adapt, with attribution and the same licence. Photos keep their own per-image licence + credit (see Sources above).
