Can 3-D X-Ray tomography imaging improve the estimation of saturated hydraulic conductivity of soils?

Saturated hydraulic conductivity (K_sat) is one of the most fundamental parameters in soil hydrology. It governs the rate of saturated flow through porous media and functions as a scaling factor for unsaturated flow. Knowledge of K_sat is key to understanding the movement of water in soils, transport and recharge of groundwater, suspended and dissolved transport in soils, and soil-air water exchange. In hydrology and climate modeling K_sat is often estimated through pedotransfer functions. A large effort has been committed to the development of these models, using an array of differing algorithms and methods. However, estimating K_sat has been somewhat troublesome, since the commonly measured soil properties, such as soil texture, bulk density and organic matter content, used as predictor variables in PTFs do not explain K_sat variation well. Instead, K_sat is largely controlled by pore-network characteristics especially in highly-structured soils. Using an extended, methodologically homogeneous dataset of commonly measured soil physical properties, 3-D X-ray computed tomography imaged pore-network parameters, and quasi-continuous particle-size measurements using the Integral Suspension Pressure method, we assess the benefits of using combined soil textural and structural information on the estimation of K_sat. Using this dataset, we have built models that estimate K_sat using a boosted random forest algorithm (XGboost) and used explanatory model analysis to tune and evaluate the models. Three input data scenarios included (i) basic soil inputs only (ii) imaged pore metrics only, and (iii) their combination. Using or adding imaged pore metrics as inputs greatly improved the K_sat estimations that were reflected, for example, by the respective coefficients of determination, evaluated using a cross-validation scheme (R² = -0.05/0.60/0.58 for the three input scenarios respectively). 3-D imaging of soil and the subsequent characterization of its pore-space may serve multiple research purposes, but such data are still not routinely collected due to cost of measurement and general lack of access to equipment. Our study confirms, however, that when collecting such metrics will become economically feasible through e.g. better automation of image processing using tools like SoilJ, having those metrics will show great potential towards improving the estimation of the soil’s water transport properties.