Experimental Studying of Extracting of Pedo Transfer Function by Regression and GMDH Method to Estimate Infiltration in Surface Irrigation

Infiltration is a major soil hydraulic property that has effective role on water and soil resources researches. Field methods of infiltration measuring are time consuming and costly. Therefore an indirect estimation method such as transfer functions is concerned. In this study were derived pedotransfer functions of infiltration of surface irrigation in both GMDH neural network and regression methods by the chemical and physical properties of soil and water. Due to this purpose, 17 soil samples were gathered from Foumanat plain of Guilan province. Physical and chemical properties of soil including soil texture, soil size distribution, particle density, bulk density, soil water retention curve, electrical conductivity, sodium absorption ratio, organic matter content and pH were measured and infiltration experiment were conducted in physical model to evaluate the effects of three treatments of water height on soil surface including 3, 5 and 7 cm and three treatments of sodium adsorption ratio including 1.7, 2.5 and 9. The results showed that GMDH neural network method estimate infiltration more accurate (R2 =0.82) than regression method (R2 =0.75). Regression equations showed that hydraulic parameters of irrigation water including water standing on soil surface and contact surface of water with soil had more effect on infiltration than chemical properties of water and soil. In GMDH method, SAR of water resource, sand percentage and soil moisture in 100 cm matric potential were recognized more effective parameter on infiltration estimation.