Effect of resolution, reconstruction settings, and segmentation methods on the numerical calculation of saturated soil hydraulic conductivity from 3D computed tomography images

Over the last decade, computed tomography (CT) images have been used increasingly to gain a more complete understanding of the microscale parameters that control the soil processes. Images that are used for that purpose typically are subjected to a number of successive treatments. This article evaluates for a sandy-loam soil, the impact of two CT scan resolutions (26 and 54 mu m), three noise reduction levels, four settings of conversion from a 32-bit to an 8-bit image and four segmentation methods (Hapca et al., 2013; Houston et al., 2013a; Schliiter et al., 2010; Otsu, 1979) in an incomplete factorial design, on the morphological metrics and on the numerical calculation of the saturated hydraulic conductivity, K-sat. The calculations reveal variations of up to two orders of magnitude of K-sat. Houston's and Schluter's segmentations appear the most sensitive to the identification of gray voxels in the key throats controlling the water flow. When combined with high noise reduction levels they produce in some extreme cases disconnections of the percolating pores. Hapca's segmentation produces more robust results. The Kozeny-Carman relation successfully predicted the saturated hydraulic conductivity when using the critical path diameter as the characteristic length, instead of the macroscopic hydraulic radius which appears too integrative to identify the extent of the variation of key throats.