| Water resources are already scarce,and unreasonable utilization in recent years has led to water pollution,making the water environment more and more serious.Eutrophication of rivers and lakes is the most prominent problem.As an important environmental variable affecting the distribution of algae communities,the timely monitoring and prevention of total nitrogen content have attracted wide attention.The traditional water quality monitoring method is time-consuming and laborious,and it is difficult to obtain the water quality condition of large scale.With the development and rise of satellite remote sensing technology,it effectively makes up for the limitations of traditional monitoring and can study water quality problems from a new perspective,greatly promoting the development of water quality monitoring.At present,for the remote sensing study of total nitrogen content of water quality parameters,most scholars have adopted the empirical method based on mathematical statistical theory,which is relatively mature and is the most suitable method for total nitrogen research among the three methods of water quality remote sensing monitoring.The remote sensing data source is mainly based on satellite-borne multi-spectral data.As the core city of the east coast and a famous water town,the water environment problem of Taizhou city should not be ignored.In order to respond to the ecological river and lake implementation plan of the municipal government,this paper decided to take the total nitrogen content of water quality parameters as the research object in order to comprehensively control the eutrophication of river and lake and strictly control the total nitrogen and phosphorus emissions.Due to the dense water network within the territory,it is impossible to conduct the study one by one,so the rivers with key location,great significance and certain representativeness,namely Yinjiang rivers,are selected as the research area.Combined with Landsat 8 OLI multi-spectral remote sensing image,the estimation model of total nitrogen content was studied.Based on the relevant principle of water quality remote sensing,using the theory of mathematical statistics,correlation analysis was made between the band data of remote sensing image and the measured water quality data in the SPSS software.According to the Pearson correlation coefficient,determine the sensitive band or band combination.Then the sensitive band or band combination and the water quality data are regressed to establish the corresponding function model;Using support vector machine SVM based on statistical learning theory,modeling in the Matlab software,introducing libsvm toolbox,kernel function type choose RBF kernel function,the grid method of optimization to determine the best penalty factor c and the kernel function parameter g.With different input data,the corresponding parameters c and g and the accuracy of the model are also different,and the type of support vector machine is e-SVR.These two modeling methods belong to empirical methods.The relevant research results show that:(1)The single band with a high correlation with the total nitrogen content was mainly the blue band b2,the green band b3 and the red band b4.Among them,b2 and b3 were significantly correlated at the level of P=0.01,and b4 was significantly correlated at the level of P=0.05.The band combination with the highest correlation is the combination of blue band and green band b3*(b3+b2).According to the correlation coefficient r,when 0.7≤|r|<1,it is strongly correlated.In this paper,the band does not reach the level of strong correlation.(2)In the empirical method,support vector machine was added to significantly improve improve the fitting degree of total nitrogen content and the spectral reflectance of remote sensing image.Its squared correlation coefficient R2 can be as high as 0.9 or more,reflecting the superiority of SVM in data regression processing.The predictive accuracy of SVM regression equation for total nitrogen content was significantly higher than that of single band linear regression equation and band combination linear regression equation,and the predicted value was closer to the true value.(3)Combined with Landsat 8 OLI satellite data,it is feasible to study and estimate the total nitrogen content of water quality parameters in Yinjiang rivers. |