基于地形剖面拟合的浅海水深无原位光学遥感探测方法OA

Satellite-based single-photon measurements from ICESat-2(Ice,Cloud,and Land Elevation Satellite-2)are used to extract bathymetry data as training samples or constraints.In conjunction with the extensive spatial coverage provided by multispectral imagery,facilitating the development of a fused active-passive optical remote sensing water depth inversion model.This integrated methodology has emerged as a mainstream solution for efficient and large-scale water-depth remote sensing,particularly in regions where traditional in situ data acquisition is challenging or impractical.Despite the promising capabilities of this approach,the accuracy of water depth inversion is predominantly constrained by the precision of extracting bottom single-photon signals.Conventional clustering algorithms employed for photon signal extraction often suffer from limitations,such as the generation of discontinuous underwater photon point clouds or production of point clouds with excessive thickness.Such deficiencies compromise the spatial continuity and boundary clarity of the reconstructed underwater topographic profiles,adversely affecting the subsequent water depth inversion accuracy.To address these challenges,this study introduces a novel methodology based on terrain profile fitting for shallow water depth optical remote sensing detection without in-situ data.The proposed method begins by acquiring bottom single-photon signals using an advanced clustering algorithm.Subsequently,local weighted regression and spline interpolation techniques were applied to the extracted photon-point cloud data.This dual-step process facilitates the reconstruction of underwater photon point clouds and effectively compensates for data gaps,yielding a continuous and coherent underwater topographic profile that represents bathymetric features more accurately.Following the enhancement of the underwater topographic data,water depth correction was performed.This correction accounted for various environmental and sensor-specific factors,ensuring that the derived depth values were both accurate and reliable.The corrected water depth data were then integrated into Sentinel-2 multispectral imagery,leveraging the extensive spectral and spatial coverage of the imagery to further refine the bathymetric inversion process.In this integration phase,we constructed a water depth inversion model based on band ratio analysis,which exploited the spectral reflectance properties of water to determine water depth with higher precision.The efficacy of the proposed method was rigorously evaluated using experimental datasets from Sentinel-2 and ICESat-2 collected over three distinct study regions:Molokai Island,Crooked Island,and Quanfu Island.Using the results of the widely employed DBSCAN clustering algorithm,comparative analyses was conducted.Experimental results demonstrate that,for Molokai Island,the proposed approach achieved a coefficient of determination(R2)of 0.98 and root mean squared error(RMSE)of 0.70 m,compared with that of the ICESat-2 bathymetry validation set.Regarding Crooked Island,the model validation accuracy was further enhanced,with R2 reaching 0.94 and RMSE decreasing from 0.28 to 0.22 m,representing a 21％improvement.Similar improvements in water depth inversion accuracy were noted for Quanfu Island.These findings substantiate the feasibility of the terrain profile fitting-based method.By enhancing the continuity and clarity of underwater photon-point cloud topographic profiles,the robustness of water-depth inversion models can be significantly improved.Accordingly,this study offers robust technical support for optical active-passive shallow water depth inversion,thereby facilitating more precise and extensive bathymetric mapping in coastal and shallow-water environments.