关键词: 土壤盐渍化, 高光谱, 单变量回归, 逐步多元线性回归, 主成分分析

Abstract: Taking Weigan River-Kuqa River Delta oasis in Xinjiang as research area, with the measured soil hyperspectral data and soil salinity as foundational data, the correlation of various spectral indices and soil salinity was analyzed, feature bands were selected, and three methods of stepwise multiple linear regression, univariate regression, and principal component regression were used to construct a hyperspectral monitoring model for soil salinity. The research indicated that based on stepwise multiple linear regression, the salinization remote sensing monitoring model utilizing logarithmic second-order differential spectral feature bands was best, with the highest stability and prediction accuracy, which could effectively estimate the soil salt content. The research results met the demand for salinization monitoring in arid regions, and provided a reliable reference for quantitative inversion of soil salinity in arid regions.

Key words: soil salinization, hyperspectral, univariate regression, stepwise multiple linear regression, principal component analysis