通过对新疆阜康500水库下游的盐渍化土壤实地定点取样和光谱测量, 利用光谱变换、 相关分析等方法, 定量探讨了不同人为干扰程度的土壤盐分、 水分与光谱反射率之间的关系, 并建立了土壤反射光谱与盐分含量之间的多元线性回归预测模型。 结果表明: （1)人为干扰程度与土壤盐分呈极显著正相关, 而与土壤水分呈极显著负相关, 相关系数分别为0961和-0929。 （2)在不同干扰程度与土壤光谱反射率的关系中, 重度干扰的土壤反射率比轻度干扰土壤的反射率高10%, 比未干扰高17%。 这是由于人为干扰破坏了土壤表面的少量植被及生物、 物理结皮, 土壤表层因缺乏保护, 水分会迅速蒸发, 并将土壤下部的盐分带到上部, 加之降水稀少, 盐分在表层聚集。 干扰程度越高, 结皮破坏越严重, 土壤积盐越多, 反射率越高。 （3)随干扰程度的不断增加, 土壤原始光谱反射率与盐分相关系数的两个最大值逐渐向近红外波段偏移（999, 876～979, 1 182～1 370和1 900 nm), 这预示着, 在近红外区土壤光谱反射率对盐分含量更为敏感。 （4)利用反射率R、 反射率一阶导数R′、 反射率R+水分分别建立了不同干扰程度的三类土壤盐分含量预测模型。 综合R2和RMSE判断模型精度, 在不同干扰程度下, 同类型的土壤含盐量预测模型中, 干扰程度越小, 模型精度越高; 而在相同干扰程度下, 不同类型的土壤含盐量预测模型中, 均以一阶导数R′建立的模型预测效果最优, R2均超过0983。 总体上, 模型精度提高了5%～10%, 表明原始光谱经过一阶导数变换处理, 可以去除部分线性背景值的干扰, 提高预测土壤含盐量的精度。

The reflectance of saline soil in the downstream of No.500 reservoir in Fukang, Xinjiang province was investigated. Through filed sampling and spectral test, using the method of spectral transform, correlation analysis and a quantitative analysis were conducted on the salt and water content of the soil under different disturbance degree. A multiple linear regression model was established between the soil reflectance and soil salinity content. The results show that: first, the human disturbance has a significantly positive correlation with the soil content while it has an extremely negative correlation with the water content. The correlation coefficients are 0961 and -0929 respectively. Secondly, it shows that those most heavily disturbed soil reflectance is about 10%higher than the slightly disturbed, while the slightly disturbed soil reflectance is about 17% higher than the undisturbed soil. The reason is that the soil surface of barren land with a small amount of vegetation, the biological creature and soil surface crust have been destroyed. The more the disturbance is, the greater chance the surface layer would be destroyed. Meanwhile, the surface layer of soil will be lack of the crust protective; the soil salinity of the bottom rises to the surface associated with the soil moisture will quickly evaporate. The salt is concentrated to the surface layer due to both little precipitation and a lack of protection of soil crust. Thirdly, the peak wavelength location of the spectrum is increased (999, 876~979, 1 182~1 370, 1 900 nm) while the soil is taken from undisturbed to heavily disturbed conditions, which means that with the increase of disturbance, the soil becomes more sensitive in the near infrared region. What’s more, the three different prediction models are established though the reflectance R, the reflectivity of the first derivative R′, the reflectance R+water. According to the R2 and the RMSE to comprehensive judge the accuracy of the model. It is found that among those established prediction models of the same soil salinity in the different levels of disturbance, the smaller the degree of human disturbance is, the higher the accuracy of model is. It is found that among all of those established prediction models, the one based on the derivative of R works the best, of which R2 is larger than 0983, model accuracy is improved by 5%~10% ,which means that through a derivative transformation, the linear noises in the original spectrum can be removed.