Estimating Soil Salt Content Based On Hyper-spectral Data In Minqin Oasis

Minqin Oasis is located in the downstream of Shiyang River, a famous inland river which suffering from severe water shortage and land degeneration. Under the unique environments, rapidly and accurately monitoring of soil salt contents is foundational and important to soil salinization control and salinize soil utilization. Based on fields investigation, soil sampling and test, as well as soil spectral measures, soil salt characteristics and spectral analysis was carried out to promote quantitative remote sensing of oasis soil in dryland.Spectral data in this paper was measured in laboratory. Differential transformation, continuum-removal, stepwise multiple linear regression and other techniques were applied to study the quantitative relationships between soil salt content, conductivity and soil spectral reflectance to extract diagnostic bands for predicting soil composition. The main conclusions are as follows:(1) Diagnostic spectrums sensitive to soil salt are mainly located around 450～620 nm,1400 nm,1900 nm and 2200～2400 nm. However, these bands are also heavily influenced by water vapor, which limits soil salt retrieval from hyper-spectral.(2) There is strong multicollinearity among spectral datasets, which makes it difficult to establish stable regression models. In order to settle this problem, reciprocal, logarithmic and first order differential transformations were applied to modify these datasets respectively. Results of regression analysis indicate that reciprocal and logarithmic transformations didn't improve sensitive of soil reflectivity to soil salt, but differential transformation did it remarkably. So differential transformation can partially solved the multicillinearity in soil spectral.(3) Continuum-removal is an effective method in soil spectral analysis. Continuum-removal was applied to soil spectral reflectance data which was performed smoothing processing and which was not processed. Results indicate that the latter is superior to the former. The reasons are that continuum-removal depends on subtle variation of spectral curve and smoothing processing eliminates these subtle variations.(4) Tillage and cultivation can influence soil properties significantly. According to landcover and crops, soil samples were set up to wheat, sunflower, bare land, corn and other crops. Then quantitative relationships between spectral reflectance and content of salt, soil electric conductivity were studies on each group, and the regression equations with significance level 0.05 were given respectively. Results revealed that soil salts varied in different cropland and so did soil spectral reflectance.(5) Illumination impacts can be removed by the normalization of lab and in-situ measured datasets. Verified by Mann-Whitney U test, spectral bands of 550～709 nm and 915～1100 nm from both datasets have similar spectrum.