Estimating Least Limiting Water Range (LLWR) of Soil Using Artificial Neural Networks and Genetic Algorithms

Quick and accurate estimation of soil available water as one of the most critical soil quality indices plays an essential role in agricultural water resources management. The present study estimated the least limiting water range (LLWR) for 250 soil samples taken from Khanmirza plain in Chaharmahal and Bakhtiari province. Artificial intelligence method (combining genetic algorithm (GA) with artificial neural network (ANN)) and readily available soil properties were used for this purpose. The LLWR was considered as output variable, and sand, silt and clay percentages, organic carbon content, bulk density (BD), particle density (PD), pH, electrical conductivity (EC) and equivalent calcium carbonate (CCE) were considered as input variables. From 250 data, 200 were allocated to model training and 50 to model testing. The statistical analyses showed that the artificial neural network had a reasonable estimate of LLWR with a coefficient of determination of 0.93. Finally, the combined model of artificial neural network-genetic algorithm (ANN-GA) with the highest coefficient of determination (R2 = 0.96) was identified as the most appropriate model for predicting LLWR. The two models of artificial neural network and genetic algorithm generally showed better performance than the regression equations.