Monitoring plant-pollinator networks with AI

Floral Vision: Monitoring plant-pollinator networks by integrating citizen science, computer vision, and museum specimens

Changing biodiversity
Authors

James Boyko

Nathan Fox

Yutong Wang

Yiluan Song

Yu Zhou

(random order)

Published

November 10, 2023

Keywords

species interactions, AI, segmentation, phenological mismatch, citizen science, museum, computer vision

Code

Highlights

  • We validate and expand plant-pollinator interaction networks with citizen science data (iNaturalist images).
  • We process images with interacting plant and pollinators using text-promptable image segmentation and identify species with computer vision models like Pl@ntNet.
  • We compare phenology of interacting species inferred from museum specimens and identify pairs with high risks of phenological mismatch under climate change.

iNaturalist images with plants and pollinators

(Left) An iNaturalist image with bee and with bee removed using LangSAM, text-promptable image segmentation. [Source] (Right) A common Eastern bumble bee Bombus impatiens visiting an unlabeled flowering plant [Source]

A plant-pollinator network generated based on a set of iNaturalist images with both species labeled using Graphistry.

What’s the insect on?

An iNaturalist image with bee and with bee removed using LangSAM, text-promptable image segmentation.

Examples of bee removal leading to different plant species identification by Pl@ntNet.

Phenology of interacting species

Example of small overlap in the seasonal activity of pollinator species (orange) and the flowering of plant species (green).

Example of large overlap in the seasonal activity of pollinator species (orange) and the flowering of plant species (green).