Three-dimensional imaging of pore water diffusion and motion in porous media by nuclear magnetic resonance imaging

We report on the use of a pulsed gradient spin-echo imaging sequence for the three-dimensional (3D) imaging of water transport properties in two porous media: 2 mm glass-beads and 0.15 mm quartz-sand mixed with 2 mm glass-beads. In contrast to tracer methods, which monitor the tracer motion by its effect on the signal relaxation of 1H, this sequence measures the echo signal intensity I0 without and I with applied diffusion gradient, respectively. For the wide-pore glass-bead system, the intensity loss is controlled by nearly free self-diffusion in the pores. A mean apparent diffusion coefficient is calculated from the ratio ln(I0/I) as 〈Dloc〉=1.9×10−9 m2 s−1, which is slightly lower than that of free water (D=2.3×10−9 m2 s−1). Increasing the mean pore flow velocity from 0 to 0.14 mm s−1 results in a linear increase of 〈Dloc〉 to 2.3×10−9 m2 s−1, caused by mechanical dispersion. The spatial distribution is of the log-normal type, where the width increases with increasing pore velocity. Correlation lengths are also calculated. For the fine porous medium, frequent contacts of the water molecules with the pore boundaries lead to a significant decrease of I0 by increased T2 relaxation. The resulting ratio of the signal intensities ln(I0/I) is then smaller than expected for pure diffusion, which is caused by the restricted diffusion in the fine pore system. The spatial distribution (normal) is broader than for the glass-bead system and the mean local apparent diffusion coefficient is calculated as 1×10−9 m2 s−1, a dependence on the pore flow velocity could not be detected. For the glass-bead system, the 3D image clearly shows regions of increased dispersivity (50% greater than the 〈Dloc〉), caused by packing errors, leading to preferential flow. This macroscopic effect on the column scale is quantified by a numerical simulation of tracer transport, based on the 3D diffusion coefficient field, assuming a linear relation to local velocities. From this simulation, the effective dispersion coefficient is obtained for the column scale (Deff=130×10−9 m2 s−1), which is comparable to that obtained from classical break-through curves with tracer substances.