Dynamic prediction of soil salinization in an irrigation district based on the support vector machine

Abstract

Soil salinization has random characteristics because of the influences of natural and anthropogenic factors. Therefore, a study on the dynamic prediction model of soil salinity is important for irrigation water management in salinization irrigation districts. In the present paper, the theory of supporting vector machine was introduced in the dynamic prediction of soil electrical conductivity (EC) values. Based on groundwater depth, irrigation water volume, and evaporation data from 1991 to 2010 in the Shahaoqu subdistrict of the Hetao irrigation district in Inner Mongolia, a dynamic prediction model of soil EC value was developed. The results show that the fitted values, tested values and the predicted values of the model have little difference from the real values. The absolute value of the fitting mean relative error is 2.14%, that of the testing mean relative error is 3.48%, and that of the predicting mean relative error is 6.37%. Compared with the Artificial Neural Network (ANN) model, the results show that SVM predictors perform better in forecasting soil EC values than the ANN model. The feasibility of the application of SVMs to soil salinity forecasting was demonstrated and their performance in soil salinity data analysis was proven.