Predicting Soil Hydraulic Properties of Sandy Soils Using Neural Networks.

Knowledge of hydraulic proprieties are essential in the field of hydrology and soil physics. As reported in the literature, pedotransfer functions have proven to be quite useful to predict soil hydraulic properties. On the other hand, several authors such as Schaap et al., 1998 have used neural network analysis for predicting the properties of mineral soils, from clay to sand. However to date, there are few methods capable of accurately predicting the hydraulic properties of sands. Indeed, following the distribution of the particles, sands have a wide range of hydraulic properties that are not always well represented in existing functions and models. The objective of this study was to apply a neural network method for sandy soils. A database of 343 values of K_s (saturated hydraulic conductivity under drying conditions), 69 values of K_sw (saturated hydraulic conductivity under wetting conditions), 379 values of α (van Genuchten parameter of the drying curve), 379 values of α_w (van Genuchten parameter of the wetting curve), 379 values of n (van Genuchten parameter of the drying curve), 72 values of τ (Mualem tortuosity-connectivity parameter), 112 values of θ_r (residual water content) were used to train a neural network using the neuralnet package (Günther and Fritsch 2010) of the R software.  To reduce the processing time, we implemented the neural network analysis within the parallel computing environment of Université Laval supercomputing facility. More than 1715 configurations of neurons and synapses were tested to determine the most accurate. The results demonstrated that using neural network models can lead to very accurate predictions with determination coefficients varying between 0.93 and 0.99. The results of this study will provide the basic information in our future numerical simulation studies with HYDRUS 2D/3D (Šimùnek et al. 2016).

Schaap, M. G., Leij, F. J., & van Genuchten, M. T. (1998). Neural network analysis for hierarchical prediction of soil hydraulic properties. Soil Science Society of America Journal, 62(4), 847-855.

Šimùnek, J., M. Th. van Genuchten, and M. Šejna, Recent developments and applications of the HYDRUS computer software packages, Vadose Zone Journal, doi: 10.2136/vzj2016.04.0033, 2016 (in press).

van Genuchten, M.T., 1980. A Closed-form Equation for Predicting the Hydraulic Conductivity of Unsaturated Soils1. Soil Sci. Soc. Am. J. 44(5), 892-898.

Mualem, Y., 1976. A new model for predicting the hydraulic conductivity of unsaturated porous media. Water Resources Research 12(3), 513-522.

Günther, F., & Fritsch, S. 2010. neuralnet: Training of neural networks. The R Journal, 2(1), 30-38.