Estimation Model Of Salt Content In Soils By Remote Sensing: A Case Study Of Pingluo County, Ningxia Province, China

Elevated concentrations of soluble salts at surface or near surface horizons are a major worldwide environmental problem, especially in the arid and semi-arid area. Excessive salt concentrations in soils accelerate land degradation processes and produce negative effects on sustainable agricultural development. Therefore, accurate identification of soil salinization and judging soil salinization degree are helpful for better management and usage of land and designation of saline soil improvement methods. Remote sensing with its advantages like rapid, macroscopic and dynamic has become an important means of monitoring soil salinization. Hyperspectral data can reflect the nuances of the ground objects and provide affluent information for quantitative analysis of various physical and chemical properties of the ground objects. Studying the basic features of spectrum of soil, correlations between hyperspectral spectrum reflectance and salt content of soil and choosing sensitive bands to soil salinity are the basic approaches of quantitative analysis of salt content of soils by remote sensing. Partial least-squares regression (PLSR) has the advantages of multiple regression analysis, principal component analysis and typical correlation analysis; recently, it is widely used in quantitative analysis of chemical and physical properties of ground objects. In the process of quantitative analysis of physical and chemical characteristics of surface objects, PLSR method not only help to lower the dimension of hyperspectral data but also help to remain the bands which are highly correlated with the dependent variable. PLSR also can reduce the error caused by high correlation between variables.After chemical analysis of collected soil samples, spectrum reflectances of soil samples were measured. Then correlation between spectrum reflectance and salt content of soil samples were analyzed. According to the correlation coefficients between spectrum reflectance and salt content, combine with its influence factors, soil salinity sensitive bands were determined. This paper introduced salinity estimation model by measured hyperspectral data and QuickBird multi-spectral image based on PLSR method. The main contents and conclusions are as follows:(1)Soil samples were collected synchronous or quasi synchronous with the QuickBird multi-spectral image, then soluble salt content, organic matter content, PH value of the samples were measured in the lab. Distribution of the three properties of saline soil mentioned above was analyzed by descriptive statistics method. The result shows that in the study area, soil salinity has a serious salinization degree of big differences, the content of soil organic matter is low, and PH value is high. According to the high PH value and previous studies, we can infer that there are several ions in the soil samples, such as Na+, SO42-, CI1-, Ca2+, CO32-, Mg2+. The results can provide the basis for choosing salt content estimation model's parameters.(2)Spectrum reflectance data measured in laboratory were analyzed to get the basic saline soil spectra, and variation rules of spectral reflectance with the variation of salt content and correlation between soil salinity and spectral reflectance were analyzed. The results show that reflectance of the soil samples sharply raised in 400-595nm, slowly risen in 595-800nm, more slowly risen in 800-1860nm, and reached a maximum reflectance at 1840nm. Saline soil spectra has 10 absorption peaks between 400-2500nm range, bands in 400-524nm,788-1383nm, 1499-1869nm,2094-2500 nm range are sensitive to soil salinity levels.(3)Two models were constructed respectively used selected sensitive bands and bands in 400-1000nm by measured spectrum reflectance data based on PLSR method. In the first modeling process R2=0.97, RMSEC=0.29, RMSEP=0.71. In the second modeling process R2=0.94, RMSEC=0.47, RMSEP=0.77. These results demonstrate that the PLSR method is a promising strategy for quantitative estimation of salt content. The result of second model indicate that estimating of salt content of soil using visible and near infrared bands of hyperspectral data is feasible. (4) We Analyzed relationship between measured spectrum reflectance and multispectral reflectance of the image, and four multispectral images band's contribution to the soil salinity estimation. We used four multispectral images band's reflectance, NDVI and RVI to construct a predictive model, thus up to overcome influences of multiple correlation between parameters, to the prediction accuracy and to reduce the influence of vegetation and soil moisture to the salinity estimation accuracy. The result is R2=0.94, RMSEC=0.51, RMSEP=1.22, that means the predictive model has an acceptable accuracy. This paper estimated the salt content of study area based on the above model and mapped the soil salinity of study area.For further research, factors that influence spectrum reflectance of soil and indirect features of soil salinization should be further studied and combination of remote sensing data with GIS data such as groundwater depth, salt content of groundwater, geographic features should be studied to improve the accuracy of soil salinity estimation.