Abstract Background and Aims Phase segmentation is a crucial step in X-ray computed tomography (CT) for image-based analysis (CT-IBA) to derive soil and root information. How segmentation uncertainty (SU) affects CT-IBA of vegetated soil has never been explored. Methods We proposed a new framework enabled by machine learning to measure SU and its propagation from the first to the second-order parameters derived from CT-IBA. Vegetated glass beads of varying moisture contents and plant species were CT scanned. Segmented images were used to determine volumetric fractions and morphological properties of each phase for determining the absolute permeability (K). Results Although the root phase is susceptible to SU, its influence on CT-IBA is minimal when the root content is low. However, its SU was magnified when the water phase is present. The grain phase has a lower SU susceptibility, but due to its large volumetric content, the IBA can be affected significantly. Fine roots were found to exhibit higher SU than coarse roots, indicating that root architecture has an effect on the segmentation of the root phase, and thus higher-order properties like K. Conclusion Segmentation of the grain phase is sensitive to SU. A small SU will lead to a remarkably erroneous estimation of pore morphological properties and K. To reduce SU, we suggest reducing the water content to a discontinuous state of a cohesionless vegetated porous media specimen before sending it for CT scanning and IBA. However, caution should be taken when fine roots were dried and experienced excessive shrinkage.
Read full abstract