Inversion Of Water Quality Parameters Of Miyun Reservoir Based On Multi-source Remote Sensing And Machine Learning

In recent years,although our country has done a lot of work in water pollution prevention and control,the water quality of many rivers and lakes is getting worse and worse,and has become a momentum of development.Water pollution in developed industrial areas is particularly serious.Miyun Reservoir is an important source of surface water in Beijing.It plays an important role in protecting the water security of the capital.However,in recent years,the water quality of reservoirs has been polluted to a certain extent by soil erosion in the upper reaches and agricultural and human activities in the surrounding areas.The water source area of Miyun Reservoir urgently needs further protection and control.How to monitor and grasp the degree of water pollution is the key to protection and control.Therefore,the accurate monitoring technology of water quality parameters of Miyun Reservoir is imminent.Therefore,the purpose of this study is to combine UAV remote sensing with machine learning to construct a more accurate multi-source inversion model of water quality parameters of Miyun Reservoir,which can effectively serve the water quality monitoring and protection of the reservoir water source,and provide support for the protection and supervision of the Miyun Reservoir water source.In this study,hyperspectral UAV remote sensing,hyperspectral satellite remote sensing and support vector machine(SVM),neural network(NN),Elastic Net,Bayesian Ridge Regression,Lasso,Linear Regression,Decision Tree Regressor and K-Neighbors Regressor were used.A total of twelve machine learning algorithms,Random Forest Regressor,ExtraTrees Regressor,Gradient Boosting Regressor and AdaBoost Regressor,were used to construct water quality parameters inversion models.Total nitrogen,total phosphorus,ammonia nitrogen and COD were inverted as four main water quality parameters affecting water pollution in Miyun Reservoir.The changes of water quality parameters of Miyun Reservoir in time and space scale and the effects of the implementation of water source supervision measures on the water quality of Miyun Reservoir are analyzed.The results show that:(1)Extreme Random Tree(ETR)algorithm is most suitable for the inversion model of total nitrogen,total phosphorus and COD based on high resolution hyperspectral data of UAV.The decision coefficient R2 is higher than 99%,and the gradient lifting algorithm(GBR)is most suitable for the inversion of ammonia nitrogen,and the decision coefficient R2 is higher than 99%.Extreme Random Tree(ETR)is most suitable for the inversion model of total nitrogen,total phosphorus and COD based on Landsat8.(2)After RFE dimensionality reduction,linear regression algorithm(LR)produces a lot of distortion,while decision tree regression(DTR),extreme random tree(ETR)and gradient lifting(GBR)algorithms can continue to maintain the model effect,the decision coefficient R2 is higher than 99%.(3)The results of inversion accuracy validation show that the relative errors of total nitrogen,total phosphorus,ammonia nitrogen and COD from UAV remote sensing data are 0.02318,0.00039,0.003427 and 0.45514 respectively,and those from Landsat 8 satellite remote sensing data are 0.2045,0.002683,0.017588 and 1.44225 respectively,which all have the ability to identify water quality classification.(4)The inversion results show that the total nitrogen concentration in the waters of HouBajiazhuang Village is the highest,and the relative fluctuation of total phosphorus in the waters of Bulaotun Village is relatively large on the grid scale.The water quality parameters of Baihe Dam are generally higher than those of Chaohe Dam.The concentration fluctuations of total nitrogen,total phosphorus and ammonia nitrogen occur in flood season,winter tillage,sowing and turning green.(5)According to the classification and inversion results of Surface Water Environmental Quality Standard GB3838-2002,the water quality of Miyun Reservoir is better,and it is in Class Ⅱ water all the year round,but the total phosphorus concentration is slightly higher.