| The construction of operational capacity for water quality monitoring and early warning is the basis and basis for water environmental protection and refined water pollution control.Small and micro water bodies such as lakes,rivers,reservoirs,ditches and other widely distributed cities are close to human production and living areas and are greatly affected by human activities.The problem of pollution is particularly prominent.However,due to its large number and scattered distribution,it is difficult to monitor.Some small and micro water bodies have long lacked measured water quality data.It is of practical significance to find a simple,efficient and accurate water quality monitoring method suitable for small and micro water bodies.Compared with conventional chemical detection and satellite remote sensing,UAV multi-spectral remote sensing is most suitable for small and micro water quality monitoring,but there are problems of low accuracy and poor stability,and related research needs to be strengthened.In this study,Tiande Lake was used as the test area,and the P4 M multispectral imager carried by the DJI Phantom 4 UAV was used to obtain UAV images,and water samples were collected synchronously.The practical application effect of water quality monitoring in small and micro water bodies,including which water quality parameters can be monitored,how accurate the inversion is,and how to improve the inversion accuracy.The main contents and conclusions include:(1)Research on the applicability of P4 M for water quality monitoring in Tiande Lake: Correlation analysis is performed between the measured water quality data,single-band reflectance,and the combination of all mainstream bands,and the sensitive bands(band combinations)that meet the modeling conditions are screened.The results show that the optimal sensitive band combination of total nitrogen is ln[(b2-b3)/b5],and the correlation coefficient reaches 0.742,which meets the modeling requirements;the optimal sensitive band combination of total phosphorus is ln[b2/(b1-b4])],the correlation coefficient is 0.696,which meets the modeling requirements;the optimal sensitive band combination of ag(440)and ammonia nitrogen is ln[(b3-b5)/b5] and(b1-b5),and the correlation coefficients are only 0.362 and 0.38,which do not reach Due to the modeling requirements,the P4 M multispectral imager cannot effectively monitor CDOM and ammonia nitrogen in this water area due to its limited spectral resolution and insufficient ability to capture the spectral details of the water body.It is necessary to develop an unmanned aerial vehicle multispectral imager for water quality remote sensing.(2)Construction of the inversion model based on the measured data and determination of the best model: Build the empirical statistical model and machine learning model of TN and TP respectively,and test the pros and cons of the model through independent test set samples.The results show that the best model among the five inversion models of TN is the multi-hidden layer neural network model TN--Model 3,with a coefficient of determination of 0.82 and an average relative error(MRE)of 6.5%;The best model is the XGBoost model TN--Model 7,with a coefficient of determination of 0.75 and an average relative error of 8.1%.Compared with the statistical model,the machine learning model can better express the complex nonlinear relationship between water quality parameters and water body reflectivity.The performance is more precise.(3)Test the effectiveness of the collaborative model based on the entropy weight method to improve the inversion accuracy: In order to further improve the inversion accuracy,the idea of ensemble modeling was introduced,and the deterministic ensemble modeling method represented by the entropy weight method was selected to carry out Tiande Lake.For the ensemble modeling of TN and TP,the TN-EWM synergy model and the TP-EWM synergy model for retrieving the concentrations of TN and TP are constructed.The results show that the MRE of TN-EWM is 2.5percentage points lower than that of the best single model TN--Model 3,and it also solves the problem that the estimated value of TN-Model 3 is too high in the high concentration range of TN concentration greater than 2.5 mg/l.Problem;the MRE of TP-EWM drops by 3.7 percentage points compared to the best single model TP-Model7.The collaborative model can synthesize multi-model information,integrate and play the advantages of a single model,and avoid the insufficiency of a single model,thereby improving the inversion accuracy and stability.When the inversion accuracy of a single model is not high,an ensemble modeling method can be tried.(4)Analysis of the spatial distribution and driving factors of nitrogen and phosphorus elements in Tiande Lake: The collaborative model was applied to the preprocessed UAV images to make a thematic map of the spatial distribution of nitrogen and phosphorus elements in Tiande Lake.It was found that the distribution of pollutants was high in the west and low in the east.Analyze the reasons and give targeted water quality improvement countermeasures and suggestions. |
PDF Full Text Request