Background With the development of spatial information science, the hyperspectral remote sensing becomes more and more important in nowadays. Studying on spectral characteristic of soil water content is an important work, for it is the base of monitoring remote sensing. This study aims to investigate the spectral characteristic of soil water content and the relationship between the hyperspectral data and the soil water content, to explore a rapid and accurate method for estimating soil water contents, and to establish the hyperspectral remote sensing monitoring model for saline soil water in the arid area of the Ebinur Lake Watershed.

Methods This study took saline soil with different water contents in Ebinur Lake Watershed as the research object, used the spectral reflectance transformation and multivariate statistical analysis methods (MSAM) to analyze the spectral characteristic of soil water content, and built models.

Results The result showed that with the increase of soil water content, the reflectance of soil declined. In a certain range, the longer the wavelength, the higher the correlation between the soil spectral reflectance and soil water content; the soil spectral reflectance at the wavelength of 1 937 nm (r=-0.636) had the highest correlation with water content. After 8 transformation of soil spectral reflectance, the correlation of logarithmic first order differential sensitive band of 2 357 nm was the best (r=-0.808 6). Subsequently, MSAM were applied to analyze the correlation between spectrum and salinized soil with different water contents, then the spectral sensitive band of the soil was screened, and the correlation model was established. The result indicated that the model set up by Logarithm First Order Differential at the wavelength of 2 024 nm and 2 357 nm and the Root Mean Square First Order Differential at the wavelength of 1 972 nm and 2 357 nm was the best, the correlation coefficient r was 0.894 and 0.865 respectively. Based on the above established models, the authors constructed a new model, and the correlation r of which was 0.926, increased 0.032 against the Logarithm First Order Differential model, and increased 0.061 compared to the Root Mean Square First Order Differential model.

Conclusions Therefore, the estimation model of soil water content is feasible. In addition, this study provides a new model for the indoor hyperspectral estimation of soil water content in Ebinur Lake Watershed, and it also provides a theoretical and technical reference for the hyperspectral quantitative estimation of soil water content, which has certain guiding significance for hyperspectral remote sensing.