Water Quality Inversion Of Xingyun Lake Based On Multi-source Remote Sensing

Lake is a special natural complex on the surface of the earth,which plays an indispensable role in the field of human life and survival.Timely monitoring and evaluation of the ecological environment and pollution of the lake with the help of scientific research means is of great importance for the protection and construction of water ecology.In this paper,Xingyun Lake in Jiangchuan District of Yuxi City is taken as the research area for water quality inversion research.Combining the advantages of water quality sampling unmanned ship and satellite remote sensing technology for large-scale monitoring,an appropriate water quality parameter estimation model is established,and the water quality is evaluated and analyzed on the basis of inversion.The main research contents and conclusions of this paper include the following aspects:（1）The distribution characteristics of measured water quality parameters collected and processed by unmanned ships show that each water quality parameter of Xingyun Lake presents a trend change with certain spatial differences.The concentration change of most indexes was north>South>central.According to the standardized variance,except for the nutrient index parameters（total phosphorus and total nitrogen）and some physical and chemical index parameters（potassium permanganate index）,the other index parameters（chlorophyll a,blue-green algae,p H,dissolved oxygen,electrical conductivity,chemical oxygen demand,turbidity）have relatively large variation differences between regions,and the difference is obvious in the north.（2）Extract GF-1 and Sentinel-2 multi-spectral remote sensing data and measured water quality index data for correlation analysis,and finally use water quality sensitive band and water quality concentration data to build a linear and nonlinear model.Experimental results show that compared with single band,GF-1 multi-spectral band combination can improve the correlation between remote sensing data and water quality parameters.The fitting coefficients of most regression models of chlorophyll a and potassium permanganate index were greater than 0.5.The best inversion band combination of chlorophyll a and potassium permanganate index was Ln（B3-B1）,and the best inversion band combination of potassium permanganate index was（B1+B2+B3）/B3.The best inversion models of both were cubic polynomial function models.In Sentinel-2construction spectral index,The NDWI index was the highest correlation with chlorophyll a,and the optimal model constructed with this as the independent variable was the S-curve function model,and the MEWI index was the highest correlation with potassium permanganate index,and the model with the highest accuracy was the cubic polynomial function model.（3）By the optical properties of the reflectance of GF-5 hyperspectral data,the water body is classified twice.After classification,the hyperspectral is normalized,first-order differential,second-order differential,SG filtering and other spectral transformation forms,and then the multiple regression,partial least squares,support vector regression and other models are constructed.The experimental results show that,The correlation between various spectral reflectance and the concentration of water quality parameters has been significantly improved,and different categories of water bodies have their own variables with stronger correlation.In the multiple regression model,the coefficient of determination in all models is greater than 0.5,and the error of DO,COD_Mn,TP and other indexes in Class A water bodies is lower than that in class B water bodies.The error of other indexes in Class A water body is higher than that in class B water body.In the partial least squares model,the model determination coefficients of COD_Mn,TN and TP reach more than 0.9.Except p H,the error of other indexes in Class A water body is lower than that in Class B water body.Among them,COD_Mn,TN and TP have the best fitting effect.In the support vector regression model,the model determination coefficients of all water quality parameters are above 0.8,and the error in Class B water body is lower than that in class A water body,among which,Chl-a has the best fitting effect.（4）Based on multi-spectral and hyperspectral remote sensing data sources,a series of models were constructed and analyzed for the concentration of water quality parameters,and the root mean square error test（RMSE）,mean absolute percentage error（MAPE）,Nash efficiency coefficient（NSE）and other methods were used to compare and verify the accuracy of each model.The accuracy of SVR in p H,DO and Chl-a estimation models is higher than that of other models,while COD_Mn,TN and TP are better than PLRS models.On the whole,the inversion accuracy of water quality parameters is Chl-a>COD_Mn>TP>TN>p H>DO.Through water quality evaluation,it can be seen that the first principal component is mainly permanganate index,chlorophyll a and dissolved oxygen,representing the water oxygen consumption,while the second principal component contains total phosphorus,total nitrogen and other indicators,representing the degree of eutrophication of water.Among them,the contribution of permanganate index,chlorophyll a and total phosphorus index is relatively obvious,which is mainly related to eutrophication and nitrogen and phosphorus pollution of Xingyun Lake.