Study On Water Quality Inversion Of Harbin Section Of Songhua River Based On Multi-source Remote Sensing Data

Inland basin water quality monitoring technology is an important foundation for effectively carrying out inland basin water environment comprehensive management and water pollution prevention.China 's current water resources pollution situation is intensifying,water quality indicators of many rivers and lakes are decreasing,and water body eutrophication is becoming more and more serious.Traditional water quality monitoring methods are limited in terms of time,space and manpower.At this time,with the rapid development of remote sensing technology,from low spatial resolution sensors to medium and high spatial resolution,multi-spectral sensors to high-spectral sensors The development has realized the development from the qualitative identification of the water body boundary to the quantitative inversion of the monitoring of water quality indicators.In order to solve the deficiencies of the existing monitoring technology,improve the monitoring efficiency,expand the monitoring range,and reduce the labor cost,machine learning algorithms are now used to invert the concentration of the four water quality index parameters in the Songhua River Basin,including ammonia nitrogen,permanganate index,and chemical requirements.Oxygen content,total phosphorus.At the same time,a preliminary analysis of the spatial and temporal distribution of water quality in the Harbin section of the Songhua River Basin.In this paper,the Harbin section of the Songhua River Basin is used as the research area.Based on the measured hyperspectral data and the multi-source remote sensing data of the remote sensing satellite Sentinel-2A,GF-1 / WFV1 and Landsat8 / OLI,the quantitative inversion of the four water quality index parameters is studied.,First of all,the measured hyperspectral reflectance data of the water body obtained on July 9,2019 is normalized,first-order differential,and band ratio processing,and then the correlation between the spectral reflectance and the four indicators of water quality parameters is calculated,and the water quality is selected.The band with high index sensitivity establishes a traditional regression model.Secondly,the measured data of the permanganate index,total phosphorus,ammonia nitrogen and chemical oxygen demand indicators obtained in the three months of July 9,August 7,and September 25,2019 and Sentry-2A from the quasi-synchronous month,GF-1 / WFV1,Landsat8 / OLI satellite data extracted from the band spectral reflectance results for correlation analysis,and finally use the water quality sensitive band and water quality concentration data to build a traditional regression model,using the existing sampling point data to the regression analysis model After verification,it is concluded that the model accuracy cannot meet the actual application requirements.In order to improve the prediction accuracy of the model for water quality,this paper refers to the machine learning algorithm PSO-SVR,which uses the water quality sampling data of September 25 and simultaneous Landsat8 / OLI remote sensing data to construct a water quality inversion model,and finally obtains the inversion results,while using indicators R2,RMSE Using MAPE as the accuracy evaluation standard,the PSOSVR model inversion results are compared with the GA-SVR model and traditional linear regression model to calculate the accuracy of the results,and the PSO-SVR model is used to evaluate the root mean square of the best permanganate index.The error(RMSE)is 0.3419,the degree of fit is 0.9259,the average absolute percentage error is 0.0716,the root mean square error(RMSE)of the permanganate index evaluation index using the GA-SVR model is 0.4182,and the degree of fit is 0.9093,The average absolute percentage error is 0.0856,and the root mean square error(RMSE)of the best evaluation index of permanganate index using traditional regression model is 0.48,the fitting degree is 0.8802,and the average absolute percentage error is 0.0385.Using the PSO-SVR model for the best evaluation index of chemical oxygen demand,the root mean square error(RMSE)is 0.46617,the degree of fit is 0.7325,the average absolute percentage error is 0.23,and the GA-SVR model is used to evaluate the chemical oxygen demand.The root mean square error(RMSE)of the indicator is 4.20332,and the degree of fit is 0.7118;the average absolute percentage error is 0.2772.Using the traditional regression model,the root mean square error(RMSE)of the best evaluation index of chemical oxygen demand is 3.9490,the degree of fit Is 0.7292,and the average absolute percentage error is 0.25.The root mean square error(RMSE)of the best evaluation index of ammonia nitrogen using PSO-SVR is 0.1431,the fitting degree is 0.7118,the average absolute percentage error is 0.1669,and the root mean square error of the best index of ammonia nitrogen evaluation using GA-SVR model(RMSE)is 0.1590,the degree of fit is 0.67587;the average absolute percentage error is 0.1558 L,using the traditional regression model for the best evaluation index of ammonia nitrogen root mean square error(RMSE)0.1744 L,the degree of fit is 0.6210,the average absolute percentage The minute error is 1.2368.The root mean square error(RMSE)of the best evaluation index of total phosphorus using PSO-SVR is 0.01,the degree of fit is 0.8885,and the average absolute percentage error is 0.11.The best evaluation index of total phosphorus using the GA-SVR model is the mean square The root error(RMSE)is 0.01,the degree of fit is 0.88884;the average absolute percentage error is 0.11,the root mean square error(RMSE)of the best evaluation index of total phosphorus is 0.0125 using the traditional regression model,the fit is 0.8611,the average absolute The percentage error is 0.0956.The experimental results show that the accuracy of the three evaluation index values of the PSO-SVR model algorithm in the prediction accuracy of the inversion of the four water quality index concentrations is higher than that of the GA-SVR and traditional statistical regression models.