Study On Estimation Of Surface Soil Organic Matter And Total Nitrogen Content In Ugan-Kuqa Oasis Based On Hyperspectral Data

In order to study the estimation methods of soil organic matter and total nitrogen content,this paper selects the Ugan River-Kuqa River Oasis as the study area,and uses traditional mathematical transformation,continuous wavelet transformation and narrow-band and wide-band fitting methods to construct the soil organic matter And the inversion model of total nitrogen content.First,the soil organic matter and total nitrogen are classified according to the content,and then the hyperspectral curve changes of different grades are analyzed.Then,perform traditional mathematical transformation and continuous wavelet transformation on the original hyperspectral reflectance data in MATLAB environment,and analyze the correlation between the results and the soil organic matter and total nitrogen content data.Then through observation,select the band sensitive to soil organic matter and total nitrogen content,and use partial least square regression and support vector machine methods to establish an estimation model of soil organic matter and total nitrogen content of the measured spectrum.In addition,in order to achieve a large-scale estimation of soil organic matter and total nitrogen content,this paper fits the measured hyperspectral data with landsat8OLI remote sensing image data and constructs a partial least squares model to invert the soil organic matter and total nitrogen content.The main conclusions are as follows:（1）The content of soil organic matter and total nitrogen in the study area is relatively low.The content of soil organic matter varies from 1.15g·kg^-1 to 17.58g·kg^-1,while the content of total nitrogen varies from 0.06g·kg^-1Between 1.325g·kg^-1 and1.325g·kg^-1,the average values are 8.57g·kg^-1 and 0.58g·kg^-1,respectively.As the soil organic matter and total nitrogen content increase,its reflectivity decreases,and there is a negative correlation between the spectral reflectance and the soil organic matter and total nitrogen content.（2）The soil organic matter and total nitrogen content in the study area is relatively low,and the estimation is difficult.The correlation analysis between the original spectral reflectance and the reflectance after reciprocal logarithmic transformation and the soil organic matter and total nitrogen content found that there are very few wavebands.The correlation coefficient can pass the P<0.05 significance test,but none of them passed the P<0.01 extremely significant test.（3）Differential transformation of the original spectral reflectance is helpful to improve the correlation between the spectrum and the soil organic matter and total nitrogen content.Among them,the correlation between the spectral reflectance and the organic matter after the first-order differential transformation of the reciprocal logarithm The coefficient of determination can reach 0.28,and the coefficient of determination of correlation with total nitrogen can reach 0.27.Select the sensitive band that has passed the significance test,and use the method of constructing partial least squares and support vector machine to establish the estimation model.The effective band under the[Lg（1/R）]′transformation is the independent variable,and the organic matter is the dependent variable.The model is the best.The coefficient of determination and the root mean square error of the test set are 0.743 and 2.142,respectively,and the RPD is 1.967.The coefficient of determination and the root mean square error of the best model test set established with total nitrogen as the dependent variable are 0.701and 0.143,respectively,and the RPD is 1.963.（4）After the original spectral data undergoes wavelet transformation under different generating functions,it can be found that the transformation under the Bior1.3function is more effective in mining the sensitive information of the spectral data,soil organic matter and total nitrogen.Under various wavelet transformations,the spectral The correlation between the data and the soil organic matter and total nitrogen content is significantly improved,with the highest correlation coefficients being 0.38 and 0.37,which are 0.22 and 0.23 higher than the original spectral data.The sensitive wavelet coefficients after continuous transformation of various generating functions are selected as independent variables,and partial least squares and support vector machine estimation models of soil organic matter and total nitrogen content are constructed respectively.The best model of soil organic matter[lg（1/R）]’-CWT-SVM has a training set R² of 0.876,a test set R² of 0.831,a training set and a test set RMSE of 0.93 and1.21,and an RPD of 2.48.The best model with total nitrogen content as the dependent variable[lg（1/R）]’-CWT-SVM has training set and test set R² of 0.869 and 0.827,respectively.The training set and test set RMSE are 0.05 and 0.09,and its RPD Is 2.48.It shows that the original reflectance is processed by the reciprocal logarithm first-order differential,and then the wavelet transform is performed again and the model established by the support vector machine method can accurately predict the soil organic matter and total nitrogen content in the study area.（5）Fit the 1991 narrow bands of the hyperspectral data with the 7 wide bands of the Landsat8 OLI impact data,and perform mathematical conversion of the fitted band B.The converted fitted bands are respectively related to the soil organic matter and total nitrogen content.The correlation analysis between the results showed that the fitting band of the reciprocal logarithmic first-order differential transformation[lg（1/B）]’has a low correlation coefficient with the soil organic matter and total nitrogen content,so it was not selected.In the next step of the modeling calculation,among the other transformed fitting bands,the sensitive bands were selected and combined with the partial least square method to establish the regression model of soil organic matter and total nitrogen,and the models were compared.It was found that the first-order differential transformed The accuracy of the model established with the fitting band B′as the independent variable is better,in which the accuracy of the estimation model of soil organic matter content is 0.587,and the accuracy of the estimation model of soil total nitrogen content is 0.576.Finally,the band algorithm was used to spatially invert the soil organic matter content and total nitrogen content in the study area.