Saturated hydraulic conductivity prediction from microscopic pore geometry measurements and neural network analysis

Traditional models to describe hydraulic properties in soils are constrained by the assumption of cylindrical capillarity to account for the geometry of the pore space. This study was conducted to develop a new methodology to directly measure the porosity and its microscopic characteristics. The methodology is based on the analysis of binary images collected with a backscattered electron detector from thin sections of soils. Pore surface area, perimeter, roughness, circularity, and maximum and average diameter were quantified in 36 thin sections prepared from undisturbed soils. Saturated hydraulic conductivity K_sat, particle size distribution, particle density, bulk density, and chemical properties were determined on the same cores. We used the Kozeny-Carman equation and neural network and bootstrap analysis to predict a formation factor from microscopic, macroscopic, and chemical data. The predicted K_sat was in excellent agreement with the measured K_sat (R² = 0.91) when a hydraulic radius r_H defined as pore area divided by pore perimeter and the formation factor were included in the Kozeny-Carman equation.