Estimating Wet Aggregates Stability from Easily Available Soil Properties in North West of Lake Urmia

Direct measurement of mean weight diameter (MWD) of wet aggregates in the laboratory is time consuming, laborious and expensive. The objective of this study was to derive regression, artificial neural networks (ANNs) and neuro-fuzzy pedotransfer functions (PTFs) to estimate the wet MWD in the northwest of Lake Urmia. Total of 100 disturbed and undisturbed soil samples were taken from 0-10 cm soil depth for determining some readily available soil variables in bare and agricultural lands of Shabestar region. The MWD of wet aggregates was measured by wet sieving in the laboratory. The data were divided into two series, so that 80 data points were applied for training and remaining 20 data points as testing set. There were found positive and significant correlations between sand and organic carbon (OC) (0.43**) and also between clay and sodium adsorption ratio (SAR) (0.60**). There were found positive and significant correlations between the MWD with sand (0.60**) and OC (0.58**) and negative and significant correlations between the MWD with clay (-0.48**) and SAR (-0.57**). The results of PTFs showed that OC, sand and SAR were the most important readily available soil variables to estimate the MWD. The values of R2, root mean square error (RMSE) and mean error (ME) were obtained to be 0.84, 0.192 mm, -0.122 mm and 0.84, 0.154 mm, 0.030 mm and 0.87, 0.215 mm, -0.161 mmfor the best regression, ANNs and neuro-fuzzy PTFs, respectively, in estimating the MWD according to testing data set. Therefore, the performance of the ANNs in estimating the MWD was more than regression and neuro-fuzzy PTFs in the soils of studied region, since they had lower RMSE and ME values.