Estimation of Soil Tensile Strength Using Different Modeling Methods in Some Pistachio Orchards of Rafsanjan, Iran

Tensile strength is one of the most important indicators for soil physical quality, which is equivalent to the maximum stress applied to an aggregate without any disruption. The purpose of this study was to investigate the performance of different modeling methods for estimating soil tensile strength of some Rafsanjan pistachio orchards. For this purpose, soil samples (80 samples from 0- 30 cm depth mostly sandy loam) were taken and some soil physical and chemical properties were determined. Aggregate tensile strength was also measured in different sizes. The average EC, pH, and SAR indicated that soils of the study area were saline and sodic. Multiple regression between tensile strength and other soil properties were investigated. Tensile strength modeling was also performed using multilayer perceptron neural network and decision tree. The mean squares of error and coefficient of determination were used to evaluate different modeling models. The results of model evaluation showed that the use of regression decision tree for predicting tensile strength was better than the other modeling methods because of the lowest error (R2=0.88 and RRMSE = 14%) compared to the two methods of multiple regression and the multilayer perceptron neural network. Also, the results of different tensile strength modeling showed that the percentage of clay, percentage of dispersible clay, adsorption ratio of sodium, percentage of calcium carbonate equivalent, and the percentage of organic matter are the most influential variables on tensile strength. According to the results, it seems that the most effective way to increase soil tensile strength and reduce soil bulk density in pistachio orchards is to increase the percentage of soil organic matter.