AI Helps Researchers Mine Citizen Science Data to Track Ecosystem Health
Artificial intelligence tools can scan millions of comments on participatory science platforms to extract rare data on species interactions, according to a study published in the Proceedings of the National Academy of Sciences. Researchers led by the University of Michigan found that large language models process unstructured text from birdwatchers and nature enthusiasts to build ecological datasets that reveal how biological communities respond to global change.
The Tech TL;DR:
- Large language models like Claude and ChatGPT automate the extraction of species interaction data from public observation platforms like eBird and iNaturalist.
- Researchers successfully deployed the technique to detect non-native plant species creeping into West Coast ecosystems.
Extracting Ecological Data from Citizen Observations
Millions of people worldwide capture rare and valuable data regarding ecosystem health on free online platforms without realizing the scientific value of their notes. While technological advances give researchers access to richer metrics on individual species, interactions between different species remain difficult to monitor.
To access this hidden layer of data, the research team focused on two specific case studies. One examined plant-pollinator interactions involving the West Coast lady butterfly using observations recorded on iNaturalist. The other tracked bird interactions found in the comments of observations recorded on eBird. While users upload crucial metadata like time and location, vital behavioral notes remain buried within an ocean of unstructured text comments.
Accelerating Classification Workflows with Large Language Models
Reviewing millions of comments manually to identify behaviors such as predation, cooperative flocking, or competition at bird feeders is unfeasible for small teams of trained ecologists. To solve this bottleneck, the team tested whether modern large language models could accurately isolate and categorize these textual descriptions.
Working alongside undergraduate researchers, the team designed prompts that enabled Claude and ChatGPT to extract interactions with high accuracy and precision. To measure the efficiency gains, undergraduate research assistant Olivia Stein manually classified a random sample of nearly 500 eBird comments. The team then had Claude perform the same task.
Funding Sources and Collaborative Scope
The research study titled Large language models unlock the ecology of species interactions received support from federal funding provided by the U.S. National Science Foundation and the U.S. Department of Agriculture. Additional financial backing came from the David and Lucile Packard Foundation, the Alfred P. Sloan Foundation, the Institute for Global Change Biology at SEAS, the U-M Undergraduate Research Opportunity Program, and Michigan State University’s Ecology, Evolution, and Behavior Program. Computational resources and services were supplied by Advanced Research Computing at U-M.