مقایسه توابع انتقالی رگرسیونی و شبکه عصبی مصنوعی در برآورد گنجایش

Measuring the cation exchange capacity (CEC) as one of the most important chemical soil properties is very time consuming and costly. Pedotransfer functions (PTFs) provide an alternative to direct measurement by estimating CEC. The objective of this study was to develop PTFs for predicting CEC of Guilan province soils using artificial neural network (ANN) and multiple-linear regression method and also determine whether grouping based on soil textural class and organic carbon content improved estimating CEC by two methods. For this study, 1662 soil samples of Guilan province were used from soil chemistry laboratory database of Rice Research Institute. 1109 data were used for training (the development of PTFs) and 553 data for testing (the validation of PTFs) of the models. The results showed that organic carbon was the most important variable in the estimation of cation exchange capacity for total data and all classes in textural and organic C groups in both methods. ANN performed better than the regression method in predicting CEC in all data, and grouping of data only improved the prediction of PTFs in Sand and Sandy clay loam classes by ANN method.