Abstract: Abstract: Soil salinization is not only one of the most serious environmental problems in semi-arid and arid area, but it also leads to land degradation and productivity loss. At present, most studies on soil salinization pay much attention to the quantifying of the relationship between the saline soil salt content and integer differential transform of hyperspectral data. The integer differential method only focuses on the points in differential windows, thus if extending the integer calculus to fractional order, more information could be discovered due to the advantages of fractional differential method: it has memory and nonlocality. In this paper, the authors took the Delta oasis of Yutian in the southern rim of Tarim Basin in Xinjiang as the study area, and measured the spectral reflectance and the soil salt content in order to obtain the degrees of salinity in the study area. Firstly, the hyperspectral reflectance data were treated with 5 kinds of mathematical transform: root mean square, inversion, logarithm, inversion-logarithm, and logarithm-inversion. Secondly we calculated their 0-2nd order (interval 0.2-order) derivative by using the Grünwald-Letnikov fractional order differential formula and Java programming language, then computed the correlation coefficient between the salt content and the data of each mathematical transform and each order differential. Subsequently, we comparatively analyzed the varying trend between correlation coefficient curves and the influence of correlation coefficient on single bands treated by the differential method. The results showed that differentials could evidently increase the number of the bands highly significantly correlated with salt content (0.6-order＞first-order＞second-order＞0-order), the number followed increasing-decreasing trend (reaching the maximum at 0.6-order) with the increase of differential order. For spectral reflectance and other mathematical transform at 0.6-order, the numbers of the bands followed the order inversion-logarithm=logarithm＞root mean square＞inversion＞spectral reflectance＞logarithm-inversion. For the bands 2 444, 2 423, 2 142, and 2 005 nm, differential algorithm could significantly improve the correlation between salt content and spectra (and other mathematical transforms) of salinized soil, and all the maximum absolute values of correlation coefficient were obtained at the fractional order, corresponding to 0.6-order (logarithm-inversion transform corresponding to 0.4-order), 0.6-order (inversion transform corresponding to 0.8-order), 0.8-order, and 1.4-order respectively. In conclusion, from local to global, fractional differential had a better capacity than integer differential in lifting correlation. As the order increased, the correlation coefficient curves showed a gradual changing trend, and to some extent, capturing this trend could prevent information loss caused by big differences among the spectral reflectance, first-order, and second-order differential transform. We suggest here that further researches should be concentrated on physical meaning of fractional differential in hyperspectral data to provide theoretical basis to build and describe soil salinization quantitative inversion models.