| Shallow water depth is an important hydrological parameter and geospatial data for studying marine life,ecological environment and seabed topography in shallow water,which is very important for human production and life,scientific research and military defense.The acquisition of remote sensing water depth data can be divided into active and passive ways,both of which can be mounted on different platforms(such as satellites,manned/unmanned aircraft,manned/unmanned ships,etc.).The former actively transmits and then receives the returned electromagnetic wave energy to the water body,including sonar and lidar;When the latter is detected by remote sensing,the sensor receives and records the solar radiation reflected from the water bottom and water body,which can provide several or hundreds of bands for each pixel.Among them,passive satellite remote sensing has the advantages of wide coverage,short update period,no geographical restrictions,high efficiency and convenience.And with the development of water color remote sensing technology,water depth inversion based on satellite passive remote sensing data is an important way of shallow water sounding.According to whether the measured water depth data is used in the training and fitting of the model,multispectral water depth inversion can be divided into two types: water depth inversion algorithm with control points and water depth inversion algorithm without control points.Based on this,this paper uses World View-2 high-resolution multispectral satellite remote sensing as the image data source,and studies the water depth inversion with and without control points in the three research areas of North Island,Zhaoshu Island and Ganquan Island of Xisha Islands in South China Sea.The study area of this paper covers the shallow sea part of the above three study areas,especially the shallow ocean optical water area,that is,the area where the radiated energy reaches the bottom of the sea floor and can be reflected upward from the water surface [1],which is a "blue" area from the remote sensing image.Where the water quality is relatively clear,the deepest shallow sea can reach 20-30 m.The main work and conclusions are as follows:(1)In view of the strong spatial heterogeneity of data obtained from different geographic locations and the redundancy of the remote sensing spectral data in mathematically transformed spectra,this paper proposes a geographic weighted regression model based on principal component analysis(PCA-GWR).Firstly,PCA principal component analysis is used to analyze the reflectance data of each band after mathematical transformation.Then,the principal component factors obtained are used as the training parameters of PCA-GWR water depth inversion model,and the geographically weighted regression is calculated by combining the measured water depth values.Finally,the water depth inversion results of PCA-GWR model are compared with those of dual-band logarithmic ratio model,multi-band linear regression model and geographic weighted regression model(GWR).The results show that the inversion result of dual-band logarithmic ratio model is the worst,and the inversion accuracy of multi-band linear regression model is slightly better.However,in the water depth range of more than 15 m,the inversion water depth value of both models is lower than the actual water depth value.In this water depth range,GWR model and PCA-GWR model still have high inversion accuracy.The water depth inversion results of PCA-GWR model are better.The correlation coefficients of PCA-GWR model in Beidao and Ganquan Island are 0.96 and 0.98 respectively,which are 1.05% and 1.03%higher than that of GWR model respectively.The research shows that weighted regression of local parameters is more advantageous than global regression model to solve the problem of low accuracy of water depth inversion caused by the difference of water environment and sediment in different geographical locations.At the same time,PCA-GWR model can retain the main spectral information of water depth data,eliminate the interference of redundant information and enhance the potential difference of data.(2)The bathymetry performance of the P-DLA model without control points is verified,and the pixel sampling principle for solving four key parameters of the model is summarized.Aiming at the limitation that the widely used multispectral satellite image inversion shallow water depth model relies on the measured water depth data to participate in the training and fitting of empirical model,this paper uses the blue and green bands of World View-2 high-resolution satellite remote sensing image,in this paper,the blue and green bands of World View-2 high-resolution satellite remote sensing images are used to study and verify the effectiveness of a new algorithm for shallow water depth estimation independent of measured water depth points——the dual-band log-linear analysis model based on physics(P-DLA).Firstly,this paper introduces the physical mechanism of the P-DLA model and the four key parameters in detail,and then uses the spectral values of different types of sampling pixels extracted from multispectral images to solve all the unknown parameters of the model,which is used to calculate the shallow water depth.At the same time,this paper summarizes the pixel sampling principle for solving the four key parameters of the model to reduce the influence of random sampling on the accuracy of sounding results.The results show that the dual-band log-linear analysis model can still obtain good shallow water sounding results and can clearly reflect the underwater micro-topography characteristics and changes of islands and reefs,without the actual water depth data participating in the model training.(3)Carry out the contrast experiment of controlled and uncontrolled water sounding.In order to further compare the effectiveness of P-DLA algorithm without water depth control points,this paper uses the traditional dual-band logarithmic linear regression model(DLR)to carry out water depth inversion with control points in two study areas.The experimental results show that the RMSE of Ganquan Island water depth estimated by P-DLA model is 1.69 m and R is 0.91,which is 1.3% and 1.5%higher than DLR respectively.Using P-DLA model to estimate the water depth of Zhaoshu Island,RMSE is 1.74 m and R is 0.89,which is 18.9% and 5.8% higher than DLR respectively.Experiments show that the accuracy of P-DLA bathymetry without control points is roughly equal to that of DLR bathymetry inversion with control points,which shows that the dual-band logarithmic linear analysis model can realize effective bathymetry estimation without training the model with measured bathymetry data. |
PDF Full Text Request