Speaker
Description
Quantifying intraspecific trait variation on a large scale is essential to understanding how biodiversity responds to environmental changes. However, high-resolution phenotypic data has historically been scarce in complex, mountainous landscapes. This study presents a scalable, automated pipeline that extracts flower color traits (FCTs), such as CIELAB coordinates and hue, from unstructured biodiversity observations. The study focuses on the vibrant blue flowers of Gentiana acaulis and G. verna in the European Alps. Specifically, we investigated whether these species differ in their mean FCTs and examined whether the resulting intraspecific variation is distributed nonrandomly across space and time. We used a dataset of 23,000 flowers from 6,000 locations gathered over five years. Using a deep learning segmentation framework, we isolated the flowers in the raw images to extract the FCTs in the CIELAB color space. We modeled these traits against spatiotemporal and abiotic predictors using multivariate regressions. Our results demonstrate distinct color identities: G. acaulis exhibited a hue distribution that was significantly shifted toward violet-purple, and G. verna was characterized by a more intense blue. Both species exhibited significant geographical structuring and temperature-driven gradients. However, G. verna displayed additional color shifts associated with elevation and interannual variation. Furthermore, G. verna showed a correlation between color change and the day of the year, which indicates the color fading phenological effect as the season progresses. Our study shows how combining computer vision with macroecological modeling can transform crowdsourced data into robust evidence for trait-based research.
| Status Group | Postdoctoral Researcher |
|---|---|
| FOR TALKS: Poster Presentation Option | No, I prefer to present only as a talk. |