phenofun

phenofun automates definition of regions of interest in webcam images and extraction of percentage greenness time series for phenological analysis of vegetation.

Key Features:

• Automated ROI definition: Defines regions of interest in webcam images without manual delineation to enable large-scale phenological analysis.
• Semi-supervised pixel selection: Selects pixels by correlating pixel percentage greenness time series with prototype pixels' time series.
• Unsupervised SVD clustering: Uses singular value decomposition (SVD) to cluster pixels based on SVD scores for ROI determination.
• Percentage greenness extraction: Computes percentage greenness time series from selected ROIs to quantify vegetation phenology.
• Scalability: Demonstrated at scale with analysis of 13,988 webcams from the AMOS database.
• Webcam applicability: Applicable to scientific digital webcams and publicly accessible webcams for broader spatial coverage.
• Implementation: Implemented in the statistical software R (R package phenofun).

Scientific Applications:

• Phenological monitoring: Extraction of time series for detection and classification of seasonal vegetation stages.
• Large-scale network analysis: Automated analysis across extensive webcam networks to characterize regional and continental phenology.
• Climate-change impact studies: Quantification of vegetation response patterns to global climate change.
• Spatial phenology mapping: Capturing spatial variation in phenological signals across multiple camera sites.
• High-throughput time-series generation: Producing standardized percentage greenness series for ecological and remote-sensing comparisons.

Methodology:

Computational methods explicitly include computing percentage greenness time series per pixel, semi-supervised pixel selection via correlation with prototype pixels' time series, unsupervised pixel clustering using singular value decomposition (SVD) scores, evaluation against expert-defined ROIs, and application to 13,988 AMOS webcams; methods are implemented in R (R package phenofun).