Prediction of Mean Weight Diameter of Aggregates using Artificial Neural Network and Regression Models

Direct measurement of soil physical properties is time consuming, costly and sometimes unreliable because of soil heterogeneity and experimental errors. Stability of aggregates could be estimated from surrogate data such as soil texture, bulk density, organic carbon and CaCO3 using pedotransfer function (PTF).The objective of this research was to present regression PTFs and artificial neural network models to predict mean weight diameter (MWD) of aggregate from limited sets of soil properties and to assess the efficiency of the presented models to predict the MWD with the statistical criteria including the coefficient of determination (R2) and root mean square deviation (RMSE). In total, 100 soil sample were collected from Ardabil plain and analyzed for their physicals and chemicals properties. Soil samples were divided into two groups, so that, 80 samples were used for the development and remaining 20 samples for the validation of PTFs. The values of R2 and RMSE for regression PTFs and artificial neural networks were, respectively, 0.88, 0.42 for neural networks and 0.81, 0.054 for regression PTF. Results showed that two methods could be applied to predict the MWD in Ardabil plain. However, artificial neural networks performed better than regression model in this study.