Research On Optical Remote Sensing Bathymetry Under Different Water Depth Control Points Situations

Bathymetric surveys are of great importance for submarine topography mapping and coastal construction projects.They are also of great significance for terrain surveys of islands and coastal zones,maritime navigation and marine management planning.Passive optical remote sensing technology is an important method for shallow water depth measurements.Because of the advantages of large area coverage,fast update,and low cost,it has become an effective supplement to traditional water depth measurement technology.Passive optical bathymetry models often require a certain number of measured water depth training points to participate in model regression or other calculations.In recent years,some new models which do not require the use of measured water depth have also been developed.In order to solve the problem of shallow water bathymetry under different water depth control situations,this paper develops water depth bathymetry models for no,sparse and dense water depth control.The capabilities of the models are analyzed and compared from different aspects such as the overall accuracy,the accuracy of different depths,and the different depth profiles.The main work and conclusions are as follows:(1)In the absence of measured water depth situation,the calculation of the green band diffuse attenuation coefficient in the current dual-band log-linear model adopts many priori formulas,and the performance of the model is not stable.In order to obtain the more accurate value of the green band diffuse attenuation coefficient,this paper used Hydrolight to simulate and obtained 30 sets of simulated values under different water components and water depths,and the average value is taken as the final result.The experimental results of the model using the simulated parameters show that the accuracy of the Dongdao Island and Oahu areas are good,the Mean Absolute Error(MAE)is within 2 m,and the Mean Relative Error(MRE)is 20.9 % and 15.9 %,respectively.The Root Mean Squared Error(RMSE)is 2.27 m and 1.98 m.(2)In the case of sparse measured water depth control,considering the relationship between water depth and the inherent optical characteristics of water column,this paper proposes an Inherent Optical Parameters Linear Model(IOPLM),which uses the blue and green bands of multispectral images to obtain a wide range water depth information.In the absence of a large number of actually measured water depths,an accurate water depth can still be obtained.Taking the Dongdao Island and Saipan Island as the experimental areas,the proposed model is compared with the widely used classic log-linear model and Stumpf model.The overall accuracy,accuracy of different water depth sections and the results of the water depth profiles show that the accuracy of the proposed model is the best,especially when the water depth is shallow,the MAE is reduced by more than 1.3 m at most.(3)In the case of a large amount of measured water depth data,this paper uses two machine learning methods—LASSO regression and BP neural network to carry out experiments.The water depth training points of different orders of magnitude were selected for experiments.The results of different accuracy indexes in three study areas show that the LASSO regression model have higher stability and better accuracy than the BP neural network.The MAE of the Dongdao Island,Oahu Island and Saipan Island is 1.05 m,0.95 m and1.67 m,respectively.Then,the results are carried out the residual error correction experiment,and the accuracy of the Dongdao Island and Oahu Island have been improved after the correction.