An evaluation of genetic algorithm method compared to geostatistical and neural network methods to estimate saturated soil hydraulic conductivity using soil texture

Abstract

ABSTRACT-Determining hydraulic conductivity of soil is difficult, expensive, and time-consuming. In this study, Algorithm Genetic and geostatistical analysis and Neural Networks method are used to estimate soil saturated hydraulic conductivity using the properties of particle size distribution. The data were gathered from 134soil profiles from soil and lander form studies of the Ardabil Agricultural Organization. Results showed that Or denary cokriging has the best fit for the geostatistical methods. The best-fitted vario gram was the exponential model with anugget effect of 0 cm day-1 and sill of 156 cm day-1 which is the strength of the spatial structure and full effect of the structural components on the vario gram model for the region; also, in the or denary cokriging method, an accurate estimate was obtained using R2 = 0.93 and RMSE = 3.21.Multilayer perceptron (MLP) network used the Levenberg- Marquardt (trainlm) algorithm with are gression coefficient (R2) of 0.997 and Root Mean Square Error (RMSE) of 1.22 to estimate the hydraulic conductivity of saturated soil. For GA model, parameters of root mean square error (RMSE) cm day-1 and the coefficient of determination (R2) were determined as 1.35 and 0.926, respectively. Performance evaluation of the models showed that the Neural Networks model compared with geostatistical analysis and genetic algorithm was able to predict soil hydraulic conductivity with high and more accuracy and results of this method was closer to the measurement results.