| The estuarine coast is a key hub linking the sea and the mainland,an important place for material and energy exchange between land and sea,and has complex and variable hydrodynamic conditions,rapid geomorphological evolution,complex sedimentation processes and a fragile ecological environment.Suspended sand concentration(SSC)is one of the most important parameters of marine waters,directly affecting the optical properties of estuarine waters such as transparency and turbidity,and also having an important impact on the change of erosion and siltation of estuarine shores and soil and water conservation in coastal areas.Therefore,it is of great scientific value to study the SSC variation characteristics of estuarine shores.As an important estuary in China,the Yangtze River estuary is rich in marine resources and the changes in its marine environment are closely related to human life.However,due to the complex and variable currents in the Yangtze estuary,the temporal rate of change of SSC is extremely high,especially in highly concentrated water bodies.The most traditional method of monitoring suspended sand concentration is to obtain field sampling data with fragmented temporal and spatial distribution,which is not only time-consuming and labour-intensive,but also has poor comparative accuracy,making it difficult to achieve monitoring of dynamic changes in suspended sand concentration in a large range of waters.Remote sensing technology can effectively make up for the shortcomings of traditional monitoring methods,and is widely used to monitor the dynamic changes of suspended sand concentration.Most of the current inversion studies use satellite multispectral as the data source,which has a small number of spectral bands and low resolution,and is difficult to reflect the spectral information of the features in a more comprehensive manner.In addition,there is an urgent need for high-resolution sensors for dynamic monitoring and the construction of high-precision inversion models to achieve fine inversion of suspended sand concentrations.To address the above problems,this paper aims to improve the accuracy of hyperspectral remote sensing inversion of the suspended sand concentration in the Yangtze River estuary,and uses the quantitative simulation data of the suspended sand concentration in the Yangtze River estuary.The FD-CARS-BPNN model framework is constructed from the whole technical process of"spectral pre-processing-feature band extraction-inversion model building",breaking the limitation of existing studies which only focus on a single problem of feature band extraction or inversion model building.The validity of the model was verified by airborne hyperspectral simultaneous monitoring experimental data,and the model framework was applied to the hyperspectral inversion mapping of the suspended sand concentration in the Yangtze River estuary to obtain its spatial distribution characteristics.The main work and conclusions are as follows.(1)Based on the radiation transmission mechanism of light in water bodies,this paper designs and carries out quantitative simulation experiments for different sand-bearing water bodies,and obtains measured spectral data and measured SSC data for water bodies with different suspended sand concentrations.Based on this data,the spectral characteristics of water bodies with different suspended sand concentrations are analysed,and five mathematical transformations of the original spectra are carried out to obtain the variation characteristics of different transformed spectra;at the same time,the responsiveness of the sensitive band to changes in suspended sand concentration is explored,and it is found that when the suspended sand concentration changes,the position of the sensitive band also changes.(2)Using quantitative simulation experimental data as the data source,this paper proposes a CARS-based method for feature band extraction based on spectral variation characteristics,and compares and analyses it with the correlation coefficient method,the continuous projection algorithm(SPA)and CARS-SPA to obtain the feature band data set,and finds that the distribution range of the feature band is concentrated in the near-infrared band of 810nm-900nm and the visible band of about 400nm,effectively avoiding the problem of singularity in the selection of the feature band based on the correlation coefficient method.(3)A hyperspectral inversion model framework for suspended sand concentration,FD-CARS-BPNN,was constructed based on the feature band data set and compared with univariate fitting models,multiple linear regression models and partial least squares models by calculating four indicators,R~2,RPD,RMSE and RMSE%.The results show that the inversion accuracy based on the FD-CARS-BPNN model framework is the highest.The model has improved R~2 by 1.15%,RPD by 61.64%,RMSE and RMSE%by 39.29%and 35.72%,respectively,compared with the univariate fitting model constructed based on the correlation coefficient method to extract the feature bands,and the inversion accuracy has improved significantly.(4)Based on the airborne hyperspectral simultaneous monitoring experimental data to verify the validity of the model proposed in this paper,it was found that the R~2 and RPD of the FD-CARS-BPNN model were 0.9647 and 2.7340,respectively,and the RMSE and RMSE%were 44.6865mg/L and 9.80%,respectively,with significantly higher inversion accuracy than other models,proving that the model framework is useful in the Yangtze River estuary suspended sand The FD-CARS-BPNN model framework was applied to the hyperspectral inversion mapping of the suspended sand concentration in the local area of the north port of the Yangtze River estuary,and the spatial distribution characteristics of the study area were obtained,which showed a higher SSC in the northern coastal area and a strip-like distribution in the direction of water flow in the lower SSC area. |
