Abstract: This paper selected the Ebinur Lake as a study area and established multivariate stepwise regression, partial least square regression and support vector machine regression model based on the site measured data of apparent electronic conductivity and spectral with EM38 and ASD Fieldspec HH. In the three models, the apparent electronic conductivity measured with EM38 selected as independent variable and the measured soil soluble salt content as a dependent variable, and the selected spectral transformation form and characteristic bands have best correlation with the interpreted soil salinity data from EM38 used for model establishment. Results show that, firstly, the correlation coefficient of multivariate regression model reached up to 0.91, and indicated electromagnetic induction technology can be used for the indirect monitoring of soil salinity in this area; secondly, first-order differential transformation is better than the second-order differential transformation, the spectrum transformed with first-order differential can better predict soil salinity; thirdly, the spectrum transformed with first-order differential has good performance in all the three established models, and the model with support vector machine has higher accuracy than the partial least square and multivariate stepwise regression models. Accordingly, the support vector machine regression model established with first order differentiation transformation can be used for as a desirable monitoring model for arid lakeside wetland soil salinization.

Key words: soil salinization, electromagnetic induction technology, measured soil spectral