Fine roots are often defined as the roots less than about 2 mm in diameter, and they produce and die repeatedly in a short period. It is estimated that fine root biomass accounts 33% of net primary production in a forest ecosystems. Therefore, studies of fine-root dynamics are important for revealing carbon cycling in a forest systems. However, studies of forest belowground part are very laborious.
Scanner method is a technique that takes pictures of soil profile including roots and rhizosphere in a field. This method can be used for studying fine-root dynamics by comparing the extracted fine root areas between two time series images, which are taken a fixed location. However, since visually distinguish fine roots in a soil image is difficult because of their similarity in color, extracting fine roots from a soil image requires huge amount of time. Recent years, deep learning-based software for extracting automatically fine root has been developed. They are expected to reduce the time of fine-root analysis in soil images drastically. However, it is not revealed whether they assess actual fine root dynamics by root growth and death correctly or not.
This study aimed to search the best software for analyzing fine-root dynamics automatically. In this study, four kind of automated tracing methods, ARATA, RootPainter, SegRoot and TrenchRoot-SEG were examined. First, we manually produced fine roots extracted images. Then, each software was trained using the manual fine roots extraction images and their original images. Fine root images were thereafter automatically produced using each trained software. Additionally, images of growing and dying roots were generated from the difference of time-series fine roots images for both of manual and automatic extraction. Finally, we compared the images of growing and dying roots between the manual and automatic extraction methods and evaluated the accuracy of the automatic method. In the result, only ARATA successfully assessed fine-root dynamics as same as manual tracing method. Additionally, we revealed ARATA software extracted approximately 70% of growing and dead roots.