Deep learning image segmentation reveals patterns of UV reflectance evolution in passerine birds

Area of Science:

• Avian ecology and evolution
• Bioacoustics and visual ecology
• Computational biology and machine learning

Background:

• Ultraviolet (UV) coloration is a key signaling method in numerous bird species.
• Limited understanding exists regarding the prevalence and evolutionary drivers of UV plumage.
• Previous segmentation methods hindered large-scale analysis of avian coloration.

Purpose of the Study:

• To investigate the prevalence and evolutionary factors of UV plumage coloration in passerine birds.
• To develop and apply a deep learning-based image segmentation pipeline for avian plumage analysis.
• To provide broad insights into UV coloration as a component of avian signaling.

Main Methods:

• Developed a deep learning image segmentation pipeline to analyze whole-body plumage color.
• Applied the pipeline to over 24,000 photographs of museum specimens from >4500 passerine species.
• Quantified UV reflectance across diverse passerine bird species.

Main Results:

• UV reflectance is widespread in passerine birds, often as a component of other colors.
• UV plumage coloration is strongly phylogenetically conserved across species.
• Evidence supports the role of the light environment in driving UV plumage evolution.
• A weak trend suggests higher UV reflectance in species with UV-sensitive visual systems.

Conclusions:

• Deep learning offers a powerful tool for ecological and evolutionary research using museum specimens.
• UV plumage coloration is a significant, evolutionarily constrained trait in passerine birds.
• Environmental light conditions and visual systems are key factors shaping UV coloration in birds.