RootShape: Automated Analysis of In Situ Fine Root Images

The quantification of in situ fine root traits is an important task in the comprehensive study of natural and agricultural environments and critical to under- standing how plants respond to various changes in those environments, especially as those changes relate to energy and environmental challenges. Minirhizotron experiments enable nondestructive imaging of in situ root systems, but their analysis is complicated by a labor and expertise intensive manual image segmentation task. Images of roots must be manually traced before data can be extracted, a process that can take several months of a technician’s time and is a severe bottleneck for actionable results. This combined Phase I/II project is building a software tool to automate the analysis of minirhizotron images. Automation is achieved by using tools from computa- tional differential geometry to find representations of roots in images that can be applied in general settings. During Phase I, a prototype software pipeline was developed to validate the automated root identification procedure. Also, a prototype application was implemented to allow a user to inspect results and engage with data in a limited active learning activity, which was used to further filter out artifacts. Trait analysis data was produced on image data collected during the course of the project, and the project approach was validated by demonstrating that the resulting data was sensitive to experimental effects. During Phase II, prototype software will be refactored and packaged into a commercial software product. Several improvements to the algorithmic approach that were identified during the Phase I research will be implemented. Minirhizotron image data will be generated at higher resolutions and with plant breeds that will enable validation of the software’s application to more fine root features. Root phenotyping is of growing interest in the commercial agriculture sector. Lowering the cost to entry (in terms of capital, time, and expertise) to high-throughput minirhizotron analysis will be of immense impact in a sec- tor where the value of big data is being increasingly leveraged. Also, adaptable automated processing will increase data output of minirhizotron experiments, advancing scientific understanding of the nature and dynamics of a critical part of ecosystems.