Investigation Artificial Neural Networks For Point Simulation Of Soil Water Retention Curve

摘要

In this research 35 soil samples with Loamy texture were gathered form region near Karaj-Alborz province-Iran. Different points of soil water retention curve (SWRC) in tensions points : 0, 10, 33, 100, and 1500 Kpa asdependent variables have been meseaured using pressure plate and presure membrane. Soil properties organiccarbon, bulk density, soil particles size distribution, Caco3, mean and Geometric standard deviation of particlediameter were measured as independent soil peroperties. Independents variables divided into 3 groups ofvariables. Statistical investigation of pedotransfer functions showed that regression pedotransfer functionestablished with soil particle size distribution, bulk density and organic carbon as dependent variable s resultedin best primary prediction also this pattern of input variables used as input variables of artificial neural networksmodels with Marquardt-levenburg training algorithm 3 layer procepteron structure with 6 neuron in hiddenlayer. R2 and RMSR ranged between 0.79- 0.82 and 2.53-1.21 for regression pedotransfer functions. In theother hand R2 and RMSr ranged between 0.85-0.94 and 0.88-1.30 for artificial neural networks. FinalInvestigation of result showed that artificial neural networks had more precious prediction.