Estimating saturated soil hydraulic conductivity using water retention data and neural networks

The modified Kozeny-Carman equation K-sat = B phi(e)n is used to relate soil saturated hydraulic conductivity K-sat to effective porosity phi(e). However, different values of the coefficient B and the exponent n are found in different data sets. Our objective was to find out whether and how B and n are related to Brooks-Corey's air entry pressure h(b) and pore distribution index lambda. The Southern Region soil hydrologic database of about 500 samples was explored. All soils had both silt and clay content <70% Neural networks were used to relate B and n to h(b) and lambda, and a genetic algorithm was: applied to find weights in neural networks. Dependencies of B and n on h(b) and lambda had similar shapes. Values of B and n were almost constant for values of lambda greater than I and were close to 2.6 x 10(-3) m s(-1) and 2.5, respectively. As the values of lambda decreased from 1 to 0, values of B and n decreased. The larger the air-entry pressure, the steeper was the decrease in B and n.