Application of artificial neural networks in modeling soil solution electrical conductivity.

In various applications in soil science and agriculture, there is a need for accurate measurements of soil solution electrical conductivity (EC). Time-domain reflectometry (TDR) has become an important means for measurement of soil water content and bulk EC. The TDR-measured EC (σ a ) and dielectric constant (K a ) can be used to calculate the soil solution EC (σ p ). Hilhorst (2000) found that using this linear relationship, measurements of σ p can be made in a wide range of soil types without soil-specific calibrations. In the present study, the linear model was evaluated using detailed TDR. We also attempted to model the σ p -σ a -K a relationship using artificial neural networks (ANN). To develop ANN models, we used TDR to measure K a and σ a along with five soil physical parameters (sand, silt, clay, organic matter content, and bulk density) in 10 different soil types. In total, 265 K a and σ a measurements were obtained. The ANN estimation of σ p was found to have mean square error values between 0.071 and 0.41 dS m -1 for the 10 different soil types, whereas the mean square error of the linear model was 0.315 dS m -1 . A sensitivity analysis showed that the ANN model was more sensitive to σ a and two soil physical parameters (organic matter and clay content) more than other inputs as they affected the σ p -σ a -K a relationship.