Estimating near-saturated soil hydraulic conductivity based on its scale-dependent relationships with soil properties

Soil hydraulic conductivity near saturation (K_ns) is affected by various soil properties operating at different spatial scales. Using noise-assisted multivariate empirical mode decomposition (NA-MEMD), our objective was to inspect the scale-dependent interactions between K_ns and various soil properties and to estimate K_ns based on such relationships. In a rectangular field evenly across cropland and grassland, a total of 44 sampling points separated by 5 m were selected and measured for K_ns at soil water pressure heads of −1, −5 and −10 cm. At each point, the saturated conductivity K_s was estimated using Gardner’s exponential function, and six soil structural and textural properties were investigated. Decomposed into four intrinsic mode functions (IMFs) and a residue by NA-MEMD, each K was found to significantly correlate with all six properties at one spatial scale at least. The variations in K were primarily regulated by soil structure, especially at the relatively small scales. Multiple linear regression (MLR) failed to regress either IMF1 or IMF2 of each K from the soil properties of the equivalent scales and only accounted for 13.7 to 43.6% of the total variance in calibration for the remaining half of the IMF1s and IMF2s. An artificial neural network was then adopted to estimate IMF1 and IMF2, and the corresponding results were added to the MLR estimates at other scales for which each K was estimated at the measurement scale. This prediction greatly outperformed the MLR modeling before NA-MEMD and, on average, accounted for additional 74.4 and 73.4% of the total variance in calibration and validation, respectively. These findings suggest nonlinear correlations between K and the soil properties investigated at the small scales and hold important implications for future estimations of K_ns and K_s as well as other hydraulic properties.