Water Quality Parameters Estimation Of Guanhe River Using Airborne Hyperspectral Imagery

The problem of water environment in China is serious,and the monitoring of inland water quality is of great significance to national health and food safety.Airborne hyperspectral remote sensing has the characteristics of high spatial resolution and high spectral resolution,which provides an accurate and efficient method for the rapid monitoring of inland water.In this study,the Guanhe river is taken as the study area.Based on the preprocessing of hyperspectral images and the data analysis of the laboratory tested water quality parameters,the estimation algorithm of water quality parameters in bio-optical environment of inland water are studied and discussed,and the inversion models based on spectral feature selection and deep feature extraction are established in this dissertation.After that,the spatial distribution and water quality classification maps of water quality parameters are obtained using airborne hyperspectral images,which provide a novel way for the accurate estimation of optically inactive water quality parameters.The main work and conclusions of this dissertation are listed as follows:?1?The estimation model of water quality parameters based on spectral feature selection was established by studying the correlation between different spectral variables and water quality parameters.Through the preprocessing of the water spectra and analysis of correlation coefficients,it is found that the spectra after the first-order differential can effectively highlight the characteristics which are hard to be explicit in the original spectra,and there exist well-preserve esponses to the organic parameters COD_Mn,TN,TP near 600 nm,660 nm,700 nm,770 nm and 820 nm.In addition,band ratio is introduced for the establish between the spectral value and organic parameters COD_Mn,TN and TP because of its advantages of its advantages of eliminating the influence of external factors and the sensor itself.?2?Through the analysis of spectral features of water quality parameters,it is found that COD_Mn and TN are affected by chlorophyll a,and their absorption and reflection characteristics are consistent with chlorophyll a.The spectral characteristics of TP may be affected by the combination of chlorophyll a and suspended matter,showing the absorption and reflection characteristics different from those of COD_Mn and TN.However,for the optical inactivity water parameters with less content,the model accuracy cannot be guaranteed by using the method of spectral feature selection.?3?Two deep learning-based regression models:deep neural network regression?DNNR?and convolutional neural network regression?CNNR?were proposed to estimate water quality parameters.Compared with traditional regression models:Partial Least Squares Regression?PLSR?and Support Vector Regression?SVR?,the deep learning-based regression models have better accuracy,CNNR model obtains the best performance for all seven water quality parameters with the Rp2 greater than 0.6.Results demonstrate the robustness and accuracy of the proposed deep learning-based model.?4?The spatial distribution of water quality parameters concentration and the water quality classification map in research area show that the spatial distribution of pollutant concentration obtained by CNNR model is consistent with the distribution characteristics of regional pollution sources.Based on the laboratory test data and on-site investigation of the research area,it is found that the high value regions are distributed around two large industrial parks and residential areas.According to the water quality classification maps,the whole Guanhe river basin is heavily polluted,in which the water parameters TN and NH3-N seriously exceed the standard,the overall water quality reaches the grade of inferior class V,and the water eutrophication is serious.