Estimation of index points of soil moisture retention curve of some calcareous soils with readily available soil properties by artificial neural network

Since the measurement of soil moisture retention curve and its index points like field capacity and permanent wilting point are costly and time-consuming using direct methods, one of the quick and low costly estimation methods of these properties is estimating of those properties with respect to soil basic properties including particle size distribution and organic matter content as well as bulk density. In this study, index points of soil moisture retention curve such as field capacity, permanent wilting point and available water were estimated using MLP type neural network with three activate functions and regressions methodology with respect to soil basic properties. In order to conduct the aforementioned issue, 60 soil samples were collected from the case study, the region of Robatkarim of Karaj city, with an area of 24 ha. Then, the moisture content of the index points of soil moisture retention curve was measured by using the pressure plate. Further soil basic properties including the percentages of clay, sand, bulk density and organic matter as well as lime and the moisture content of the index points were measured using the aforementioned methodology. The results of the investigations showed that the artificial neural networks by sigmoid activate function a network with 6 neurons in the input layer and 10 neurons in the hidden layer as well as 3 neurons in the output layer with Performance indicators R2 and RMSE variations of 0.6197–0.8902 and 1.6846–2.0225, respectively was introduced as the efficient network which had the best performance in estimation of wanted outputs in comparison with another activate functions (method of artificial neural network) and regression method. Percentage of lime according to results of sensitivity analysis is considered as an effective factor in the range of soils texture tested.