ALi Barikloo; Moslem Servati Khajeh; parisa alamdari
Abstract
Introduction: A variety of precise farming practices in arid and semi-arid regions such as Iran require periodic information on soil salinity and sodium content. Sodic soils have unfavorable physical and chemical properties due to the high percentage of exchangeable sodium (ESP). This problem reduces ...
Read More
Introduction: A variety of precise farming practices in arid and semi-arid regions such as Iran require periodic information on soil salinity and sodium content. Sodic soils have unfavorable physical and chemical properties due to the high percentage of exchangeable sodium (ESP). This problem reduces the capacity of available water and growth of plants. To measure the percentage of exchangeable sodium, it is necessary to measure the amount of cation exchange capacity (CEC). Because determining CEC are time consuming, it is appropriate and economical to develop a method that determines ESP indirectly from easy-measured properties. One of the methods to study the relationships and correlations between different soil properties and their quantitative expression is the use of some statistical models. These models, called transfer functions, include data mining, regression models, artificial neural networks, and the coherent neural-fuzzy integrated system (CANFIS).Materials and Methods: The aim of this study was to develop an intelligent model (CANFIS) for predicting soil ESP from soil easy-measured properties in approximately 1450 ha of salt affected soils, South East of Urmia Lake, Bonab region, East Azarbaijan Province. For this purpose, 209 soil samples were taken by grid survey method from surface (0-25 cm) and then carried out laboratory for measure necessary soil properties. Soil acidity and electrical conductivity of samples were measured in a ratio of 1: 5 soils to water, soil tissue by hydrometric method, sodium cation by flame photometer and calcium and magnesium by returned titration method.Results and Discussion: Pearson correlation method showed that the accuracy of estimating intelligent models depends on the correct choice of first layer input information. Therefore, using the correlation matrix, the relationship between soil parameters (independent variable) and the percentage of exchangeable sodium (dependent variable) was determined. Sodium absorption ratio (0.961), electrical conductivity (0.808), pH (0.638), clay content (0.524), sand (0.482) and silt (0.389) have the greatest effect on estimation Percentage of exchangeable sodium in soil. Also, the positive relationship between soil reaction and the percentage of exchangeable sodium on the one hand and the high correlation between the percentage of exchangeable sodium and electrical conductivity indicate the importance of the fine soil. In this study, the relationship of linear regression between the percentage of exchangeable sodium and the ratio of sodium uptake with an explanation coefficient of 0.91 was calculated, which is significant at the level of 5% probability. Two important targets were designed in this paper. First target is determining performance of Fuzzy Neural Networks (CANFIS) in predicting ESP by sand, clay, pH, SAR, EC as input variable. The second target is evaluation of performance of CANFIS model by selected variable of PCA model. Results showed that the performance of second model was acceptable Model 1 justifies 88% of the changes in the percentage of exchangeable sodium by entering all inputs. But CANFIS model with higher inputs selected by PCA model (principal component analysis) including sodium adsorption ratio, electrical conductivity, soil reaction has higher accuracy. So that the values of root mean square error and correlation coefficient in the test stage for the first model were 0.88 and 3.25 and the second model was 0.96 and 1.0, respectively. Conclusion: These results demonstrated the superiority of intelligent models in explanation of the relationship between ESP and other soil easily-measured properties. In order to model the soil retrieval properties such as cation exchange capacity and to achieve the most suitable model, it is necessary to pay attention to the number and most effective input variables. Because the main goal is to provide a model with a minimum number of inputs as well as inputs that are easy to measure and in a short time. The results of quantification of the importance of variables in the CANFIS model confirm the use of three characteristics of sodium adsorption ratio, electrical conductivity of soil saturated extract and soil acidity in modeling the percentage of exchangeable sodium. The results of this study can be generalized to soils of similar arid and semi-arid regions. Also, due to the ambiguity of soil-related phenomena or the approximate values of the measured values of different soil properties and the uncertainty in the data, the use of hybrid models such as CANFIS that use fuzzy sets, It can be useful in fitting soil transfer functions.Keywords: Easly-measured Properties, Hard-measured Properties, PCA All right reserved.
