| Bohai Bay is heavily eutrophicated.Chlorophyll concentration is an important indicator of water quality detection.It can be used to detect water pollution and eutrophication.The remote sensing method has the characteristics of real-time and wide range for monitoring the chlorophyll concentration in the sea area.Therefore,it is necessary to establish an effective remote sensing inversion method for monitoring the chlorophyll concentration in Bohai Bay.The main work and research results of this paper are as follows:(1)Data preprocessing.The in-site measurement data of CTD were pretreated to obtain the surface chlorophyll concentration at the sampling position.GOCI and MODIS data synchronized with the measured chlorophyll concentration data in Bohai Bay were downloaded from relevant websites.Preprocessing remote sensing data,including atmospheric correction,geometric correction,image clipping and so on.(2)Linear regression and BP neural network models are established using GOCI remote sensing data.Twenty data were used to build the model and twelve data were used to validate the model.Stepwise regression analysis was used to establish the linear regression model.Thirty band combinations were constructed as variables.The linear regression equation was established by introducing three variables through significance test.By setting different input data of input layer and the number of hidden layer nodes,29 kinds of BP neural network structures are designed.All models were trained and validated one by one,and the BP neural network of GOCI with the highest accuracy is selected.The in-site data are used to verify the linear regression model,BP neural network model and three operational algorithms of GOCI.The results show that the accuracy of the linear regression model and BP neural network model established based on the in-site data is better than that of the operational algorithms.The inversion accuracy of BP neural network algorithm is the best among the chlorophyll inversion models based on GOCI data.The BP model takes 8 bands of GOCI as input data,and the number of hidden layer nodes is 13.And its absolute percentage difference is 27.93%.(3)Linear regression and BP neural network models are established using MODIS remote sensing data.Twenty-two data were used to build the model and thirteen data were used to validate the model.Twenty-four multi-band combinations were constructed as variables,and a linear regression equation was established by introducing one variable through significance test.By setting up different input data and the number of hidden layer nodes,18 BP neural network structures are constructed.All BP model is trained and validated one by one,the BP neural network of MODIS with the highest accuracy is selected.The linear model based on MODIS,BP neural network model and business/operational algorithm are validated with measured data.The results show that the inversion accuracy of BP neural network model is best.The comparison of GOCI and MODIS inversion methods show that the operational algorithm can not be simply used for the inversion of chlorophyll concentration in the Bohai Bay,the BP neural network algorithm has the ability of non-linear regression and has great advantages over linear regression.The best model for the inversion of chlorophyll concentration in the Bohai Bay established in this paper is the BP neural network model based on GOCI data.(5)An improved bilinear interpolation method was proposed to process the inversion results of BP neural network model based on GOCI data.This method can resample the invalid values of inversion results,increase the number of valid data,ensure the continuity and integrity of inversion data,and avoid the loss of a large number of valuable information. |
