Estimating Penetration Resistance in Agricultural Soils of Ardabil Plain Using Artificial Neural Network and Regression Methods

Introduction: Penetration resistance is one of the criteria for evaluating soil compaction. It correlates with several soil properties such as vehicle trafficability, resistance to root penetration, seedling emergence, and soil compaction by farm machinery. Direct measurement of penetration resistance is time consuming and difficult because of high temporal and spatial variability. Therefore, many different regressions and artificial neural network pedotransfer functions have been proposed to estimate penetration resistance from readily available soil variables such as particle size distribution, bulk density (Db) and gravimetric water content (θm). The lands of Ardabil Province are one of the main production regions of potato in Iran, thus, obtaining the soil penetration resistance in these regions help with the management of potato production. The objective of this research was to derive pedotransfer functions by using regression and artificial neural network to predict penetration resistance from some soil variations in the agricultural soils of Ardabil plain and to compare the performance of artificial neural network with regression models. Materials and methods: Disturbed and undisturbed soil samples (n= 105) were systematically taken from 0-10 cm soil depth with nearly 3000 m distance in the agricultural lands of the Ardabil plain ((lat 38°15' to 38°40' N, long 48°16' to 48°61' E). The contents of sand, silt and clay (hydrometer method), CaCO3 (titration method), bulk density (cylinder method), particle density (Dp) (pychnometer method), organic carbon (wet oxidation method), total porosity(calculating from Db and Dp), saturated (θs) and field soil water (θf) using the gravimetric method were measured in the laboratory. Mean geometric diameter (dg) and standard deviation (σg) of soil particles were computed using the percentages of sand, silt and clay. Penetration resistance was measured in situ using cone penetrometer (analog model) at 10 replicates. The data were divided into two series as 78 data for training and 27 data for testing. The SPSS 18 with stepwise method and MATLAB software were used to derive the regression and artificial neural network, respectively. A feed forward three-layer (8, 11 and 15 neurons in the hidden layer) perceptron network and the tangent sigmoid transfer function were used for the artificial neural network modeling. In estimating penetration resistance, The accuracy of artificial neural network and regression pedotransfer functions were evaluated by coefficient of determination (R2), root mean square error (RMSE), mean error (ME) and Akaike information criterion (AIC) statistics. Results and discussion: The textural classes of study soils were loamy sand (n= 8), sandy loam (n= 70), loam (n= 6) and silt loam (n= 21). The values of sand (26.26 to 87.43 %), clay (3.99 to 17.34 %), organic carbon (0.3 to 2.41 %), field moisture (4.56 to 33.18 mass percent), Db (1.02 to 1.63 g cm-3) and penetration resistance (1.1 to 6.6 MPa) showed a large variations of study soils. There were found significant correlations between penetration resistance and sand (r= - 0.505**), silt (r= 0.447**), clay (r= 0.330**), organic carbon (r= - 0.465**), Db (r= 0.655**), θf (r= -0.63**), CaCO3 (r= 0.290**), total porosity (r= - 0.589**) and Dp (r= 0.266*). Generally, 15 regression and artificial neural network pedotransfer functions were constructed to predict penetration resistance from measured readily available soil variables. The results of regression and artificial neural network pedotransfer functions showed that the most suitable variables to estimate penetration resistance were θf, Db and particles size distribution. The input variables were n and θf for the best regression pedotransfer function and also Db, silt, θf and σg for the best artificial neural network pedotransfer function. The values of R2, RMSE, ME and AIC were obtained equal to 0.55, 0.89 MPa, 0.05 MPa and -14.67 and 0.91, 0.37 MPa, - 0.0026 MPa and -146.64 for the best regression and artificial neural network pedotransfer functions, respectively. The former researchers also reported that there is a positive correlation between penetration resistance with Db and a negative correlation between penetration resistance with θf and organic carbon. Conclusion: The results showed that silt, standard deviation of soil particles (σg), bulk density (Db), total porosity and field water content (θf) are the most suitable readily available soil variables to predict penetration resistance in the studied area. According to the RMSE and AIC criteria, the accuracy of artificial neural network in estimating soil penetration resistance was more than regression pedotransfer functions in this research.