%0 Journal Article
%T Estimating the parameters of Philip infiltration equation using artificial neural network
%J Iran Agricultural Research
%I Shiraz University
%Z 1013-9885
%A Abrishami-Shirazi, N.
%A Sepaskhah, A.R.
%D 2019
%\ 12/01/2019
%V 38
%N 2
%P 25-36
%! Estimating the parameters of Philip infiltration equation using artificial neural network
%K Artificial neural networks
%K Water infiltration
%K Philip equation
%K Sorption coefficient
%K Hydraulic conductivity
%R 10.22099/iar.2019.5472
%X Infiltration rate is one of the most important parameters used in irrigation water management. Direct measurement of infiltration process is laborious, time consuming and expensive. Therefore, in this study application of some indirect methods such as artificial neural networks (ANNs) for prediction of this phenomenon was investigated. Different ANNs structures including two training algorithms (TrainLM and TrainBR), two transfer functions (Tansig and Logsig), and different combinations of the input variables such as sand, silt, and clay fractions, bulk density (BD), soil organic matter (SOM), cumulative infiltration (CI) and elapsed time were used to predict sorption coefficient (S) and hydraulic conductivity (A) in Philip equation (I=S*t0.5+A*t), which corresponded to 30 soil samples from study areas located in the Agricultural College, Shiraz University, (Bajgah). A two-hidden layer ANNs with two and three neurons in the hidden layers, respectively and TrainLM algorithm performed the best in predicting S when Logsig and Tansig were used. Silt+ clay+ sand+ time+ CI combination was the most basic influential variables for the S prediction. Furthermore, a two-hidden layer ANNs with two and three neurons in the hidden layers, respectively and TrainBR algorithm performed the best in predicting A when Tansig and Tansig were used. Silt +clay +sand +BD + OM+ time+ CI combination was the most basic influential variables for A prediction. Results showed that increasing the hidden layers and input variables significantly improved the ANNs performance. The coefficient of determination (R2) confirmed that the ANNs predictions for A (84.6 %) fit data better than S (77.5 %).
%U https://iar.shirazu.ac.ir/article_5472_a812598ca75daced5b811347719f62aa.pdf