Preferential flow revealed by hydrologic modeling based on predicted hydraulic properties

Pedotransfer functions have shown a reasonable reliability and accuracy for predicting soil hydraulic properties. However, the contribution of the range of macroscopic features leading to preferential water flow is not readily taken into account. We modified and used the hydrological component of the reactive transport model MIN3P and the neural network-based code ROSETTA in an attempt to simulate 4 yr of daily measurements of the soil water content in a forest soil covered by Douglas-fir (Pseudotsuga menziessii Franco). A good fit of the mean measured water contents was obtained during periods of low soil moisture, while the model tended to overpredict water contents during periods of high soil moisture. This behavior is typical for the presence of significant preferential flow. Slightly better results were obtained by using predicted values of the saturated hydraulic conductivity, while the assumption of a water table located at shallow depth increased discrepancies. A good match was obtained by calibration of a simple preferential flow scheme, which was based on the assumption that the retention properties of the porous network control preferential flow. Accordingly, preferential flow seemed to initiate within the capillary pore domain. This causes a much greater sensitivity of the results to the position of the water table than with other schemes that consider pure gravity-driven flow in large macropores. Knowledge of the functional pore size is needed to ascertain the type of preferential flow scheme to be used.