ICESat-2水深提取方法及其反演应用

Water depth extraction of ICESat-2 and application to bathymetric inversion

相比于传统水深测量方法,结合主动激光卫星ICESat-2数据与被动光学遥感影像的主被动融合水深反演方法具有低成本、覆盖面积广、不受地域限制、无需地面实测数据等优点.研究选取澳大利亚大堡礁中部的John Brewer Reef和南部的Fitzroy Reef为实验区,开展主被动融合水深反演实验.针对ICESat-2光子点群信噪比低、光子点水深信息提取和改正难的问题,本文提出基于光子点群高斯分布特征的海面光子点群提取方法和基于深度自适应DBSCAN的海底光子点群提取方法,并顾及海面倾斜引起的入射角偏移,改进了Parrish水体折射改正几何模型.实验表明,基于光子点群高斯分布特征的海面光子点群提取方法和基于深度自适应DBSCAN的海底光子点群提取方法可以高效准确地提取出有效的海面、海底光子点群;经过地球物理改正和基于改进模型的水体折射改正后,水深变浅,且改正项与改正前水深呈线性关系,线性系数约为-0.2550.实验将提取并改正后的ICESat-2光子点水深与经大气校正、耀斑校正、空间域滤波等预处理后的Sentinel-2影像进行空间匹配构建水深反演样本集,并利用单波段模型(SB)、比值模型(BR)、Lyzenga多项式模型(LP)、二次多项式模型(QP)、三次多项式模型(CP)、支持向量回归模型(SVR)、多层感知器模型(MLP)、随机森林回归模型(RF)等水深反演模型对实验区全域水深进行反演.通过对比分析多种水深反演模型在不同地理区域和ICESat-2水深样本分布条件下的表现,研究为不同场景下的主被动融合水深反演策略提供了一定的科学依据.

Compared to traditional bathymetric technologies, satellite-derived bathymetry (SDB) combining ICESat-2 Lidar data and passive optical remote sensing imagery offers advantages such as low cost, wide coverage, no geographical restrictions, and no need for in-situ measurements. In this study, John Brewer Reef in the central part of the Great Barrier Reef (GBF) and Fitzroy Reef in the southern were selected as study areas to carry out bathymetric inversion. To address the low signal-to-noise ratio (SNR) of ICESat-2 photon groups and the difficulty in extracting and correcting water depth for every photon point, we propose an algorithm based on the Gaussian distribution characteristic to extract photons on the sea surface and an algorithm based on depth-adaptive Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to extract photons on the seafloor. Furthermore, we improved the geometric model of refraction correction proposed by Parrish to account for the shift of the incident angle caused by the inclination of the sea level. Our results show that our proposed methods can efficiently and accurately extract effective photon point groups on the sea surface and seafloor. After refraction correction based on the improved model, we found a strong linear relationship between correction and water depth before correction, with a coefficient of about -0.2550. To obtain the sample set required by bathymetric inversion, the optimal cloud-free Sentinel-2 remote sensing image after preprocessing such as atmospheric correction, deglint process, and spatial filtering is spatially matched with the extracted and corrected ICESat-2 water depths. We then use several different models including single band (SB), band ratio (BR), Lyzenga polynomial (LP), quadratic polynomial (QP), cubic polynomial (CP), support vector regression (SVR), multilayer perceptron (MLP), and random forest regression (RF), to invert underwater topography in the study area. By comparing and analyzing the performance of various bathymetric inversion models under different geographical conditions and ICESat-2 sample distribution, this study provides a scientific reference for the strategy of bathymetric inversion with active-passive fusion in different scenarios.