Automated processing of webcam images for phenological classification.

Area of Science:

• Ecology
• Remote Sensing
• Climate Change Science

Background:

• Global climate change drives interest in phenological shifts, such as growing season timing.
• Scientific webcams capture daily images for phenological pattern analysis.
• Manual definition of regions of interest (ROIs) in webcam images is labor-intensive, limiting large-scale studies.

Purpose of the Study:

• To develop and evaluate automated methods for defining ROIs in webcam images.
• To enable scalable phenological analysis across large networks of webcams.
• To assess the effectiveness of automated methods compared to expert-defined ROIs.

Main Methods:

• A semi-supervised approach using pixel greenness time series correlation with prototypes.
• An unsupervised approach clustering pixels based on singular value decomposition scores.
• Implementation in the R package 'phenofun' for public accessibility.

Main Results:

• Automated ROI methods yield informative results comparable to expert selection.
• Methods successfully applied to publicly available internet webcams.
• Analysis of 13,988 AMOS database webcams demonstrates suitability for big data applications.

Conclusions:

• Automated ROI definition significantly enhances the scalability of webcam-based phenological studies.
• These methods facilitate large-scale monitoring of plant phenology and its response to climate change.
• The 'phenofun' R package provides accessible tools for researchers to implement these techniques.