Comparison of different Artificial Intelligence methods in modeling water retention curve (Case study: North and Northeast of Iran)

Soil moisture in different potentials is one of the most important input parameters in majority of models, which related to agriculture, water and soil sciences. Pedotransfer functions (PTFs) predict the less readily properties using easily collected soil parameters; so they have these advantages to be inexpensive and easy deriving. Two important targets were designed in this paper. First target is performance evaluation of Fuzzy Neural Networks (Fuzzy-NN) and Genetic Algorithm Neural Network (GA-NN) in comparison routine neural networks such as Multilayer Perceptrons (MLPs) in predicting moisture in predefined potential points. The second target is introducing and evaluating a new PTF, pseudo parametric and comparing its performance in modeling water retention curve with point and parametric PTFs. For achieving these targets, 122 soil samples from north and northeast of Iran in variety of soil textures, such as loam, clay, clay loam and sandy loam were selected and modeling results from different networks were compared. Results showed that in general the performance of all structures of neural networks was acceptable, so that the average of R2 and RMSE statistics were 0.0316 and 0.842 respectively. The best and worse results belonged to pseudo parametric (with R2=0.92 and RMSE=0.022) and Parametric PTFs (with R2=0.72 and RMSE=0.044), respectively. In addition, according to results we can say using Fuzzy-NN could not improve the performance of MLPs but GA-NN could.