Research On The Inversionmethod Of Chlorophyll-a Of Inland Lakes Using Multispectral Sensors

Chlorophyll a is an important component of aquatic plankton in water,and its content in plankton is relatively fixed,which is convenient to be measured in the laboratory,Therefore,Chla is usually used as an indicator for monitoring water biomass and nutrient level.The research of Chla concentration inversion model has also become an important field of water color remote sensing.By calculating the concentration of Chla,the primary productivity of each region in the lake can be obtained.Traditional monitoring of chlorophyll content in lakes takes time and energy,meanwhile it is difficult to obtain regional chlorophyll distribution results.With the development of remote sensing technology and machine learning technology,the combination of remote sensing technology and machine learning technology provides a new method for the high efficiency and low cost monitoring of water quality in large lakes.In this paper,taihu lake was selected as the research area.Based on the measured data of chlorophyll a concentration in taihu lake and the remote sensing image of HJ1B satellite CCD sensor in the same period,the traditional empirical model and different machine learning methods were used to conduct the chlorophyll a concentration inversion in taihu lake.By analyzing and comparing the precision of inversion results of different models,the inversion method suitable for inland lake water monitoring is obtained.The main work and conclusions are as follows:(1)Establishment of chlorophyll a concentration model in taihu lake.Based on hj1b-ccd data and chlorophyll a sampling point data,two kinds of traditional statistical models such as simple linear regression model and stepwise multiple linear regression model,as well as three machine learning algorithms such as K-NN,SVM and RF,were used for Chla concentration modeling in taihu lake.The principle and construction process of each model are described in detail.(2)Comparative analysis of the precision of the chlorophyll a concentration inversion model in taihu lake.IDL language was used for the cross-validation of simple linear regression model and stepwise multiple linear regression model,and R~2=0.55,RMSE=0.007mg/L,R~2=0.72,RMSE=0.007mg/L.Parameter Settings were carried out for three different machine learning methods,and the optimal K value of k-nearest neighbor method was 4,and the random forest decision tree was 600.finally,R~2=0.76?RMSE=0.0054mg/L,R~2=0.80?RMSE=0.0050mg/L,R~2=0.88?RMSE=0.0036mg/L are obtained for k-nearest neighbor method,support vector machine and random forest inversion model.In this paper,the residual method is used for leave-one-out cross validation,and the root-mean-square error is used to evaluate the accuracy of each model,so as to eliminate the error and uncertainty caused by the random distribution of training samples and test samples to the greatest extent.The results show that the estimation accuracy of the simple linear regression model is the lowest,that of the stepwise multiple linear regression model is the lowest,and that the estimation accuracy of the random forest is the highest.The estimation accuracy of k-nearest neighbor method is close to that of SVM.