Maximizing human effort for analyzing scientific images: A case study using digitized herbarium sheets.

Area of Science:

• Botany
• Digital Imaging
• Citizen Science

Background:

• Herbarium specimens are vital for historical plant data.
• High-quality human annotations are necessary for utilizing digitized herbarium images.
• This study provides guidance for botanical image annotation projects.

Purpose of the Study:

• To evaluate the accuracy of human phenological annotations on herbarium specimen images.
• To compare annotation accuracy between in-person and online community science settings.
• To identify factors influencing annotation accuracy.

Main Methods:

• Utilized a gold-standard dataset to test annotation accuracy.
• Compared in-person, structured sessions with an online community science platform.
• Analyzed the impact of botanical expertise, career level, time, traits, taxa, setting, and scorer.

Main Results:

• Botanical expertise and time had minimal impact on annotation accuracy.
• Annotation accuracy was influenced by traits, taxa, setting, and individual scorer.
• In-person annotations were more accurate than online, but both yielded high-quality data.
• Multiple independent annotations per image enhanced overall accuracy.

Conclusions:

• Community science platforms can generate high-quality botanical data.
• Best practices for human annotation of plant images are provided.
• Careful selection of taxa, phenophases, and informative training are essential for success.