The Retrieval Model Of Soil Salt Content Based On Landsat 8

Soil salinization is one of the most important factors that impacting the global production of agriculture and the major reason for the soil degradation and the decline in agricultural production. With the increasing population, the pressure facing on the production of global foodstuff and the supply of water research is more and more serious. In our country, the soil salinization makes the situation of China's national condition even more intensified. Improving the research of the precision on the soil salinization to estimate the soil salt content and monitor the distribution of soil salinization and get the information about the scale, area, the distribution of geography, the degree of salinization is more and more important. Using proper math model to quantificate and simulate the data has very import theoretical and practical significance.The thesis' main purpose is to analysis the character of Shanyin and Ying county. Fristly, take some samples outdoors by the method of five points gathering. Measure the soil salt content in the samples, and make it accord with the sampling points. Download the corresponding LandSat 8 image accorded with the sampling period. Get the image of scale of research after the radiometric correction, atmospheric correction, geometric correction, mosaic, clip. By using the function of supervised classification in ENVI, the layer of farmland could be extracted in Shanyin and Ying county, and then extract the surface reflectance of each band of sampling points in this thesis. Use single factor correlation analysis to confirm the soil salt content and the surface reflection with much different forms and find out the sensitive band. Establish the model by using three different kinds of method, including multiple regression, geostatistics and BP neural network. With the aid of decision tree to classify the salinization, make the grade map of soil salt content in Shanyin and Ying county according to the criterion of salinization level, which would offer some guide for the relevant departments to soil reclamation, land investigation and the agricultural production. The main conclusions are as follows:1. With the help of stepwise regression for the seven bands, the bands and its changing forms that impact the soil salt content little gradually are excluded. The final model of soil salt content is: Y=5.2081g(1/B2)+48.454B6+9.38(1gB6)'-32.0752.By using the stepwise regression in the thesis, the equation of the soil salt content is obtained and apply it to invert the soil salt content. However, the accuracy rating is still very low according to the average mean error and the model has its boundedness. With the development and popularize of hyperspectral image, the inversion of soil salt content by hyperspectral image will be mainstream in the future.3.Estimating the value of soil salt content by the method of geostatistics is dissatisfactory. The accuracy of the geostatistics depends on large amount of sampling points. With the increasing spatial scale and less sampling points, the sensitiveness reduces or even lose during the local region, which is the main reason of bad estimating result.4.The highest accuracy among three models is BP neural network. Its ability of nonlinearity fitting is very strong. A conclusion could be made that inverting the soil salt content is not a process simply making a linear equation combined with different bands and its changing forms.