ABSTRACT

This study examines the use of the Multi-Spectral Instrument (MSI) in Sentinel-2 satellite in combination with regression-based random forest models to estimate bathymetry along the extended southwestern Florida nearshore region. In this study, we focused on the development of a framework leading to a generalized Satellite-Derived Bathymetry (SDB) model applicable to an extensive and diversified coastal region (>200 km of coastline) utilizing multi-date images. The model calibration and validation were done using airborne lidar bathymetry (ALB). As ALB surveys are very expensive to conduct, the proposed model was trained with a limited and practically feasible ALB data sample to expand the model’s practicality. Out of the three different sub-models introduced using varying combinations of historical satellite imagery, the combined-band model with the largest feature pool yielded the highest accuracy. The results showed root mean square error (RMSE) values of 8% and lower for the 0–13.5 m depth range (limit of the lidar surveys used) for all areas of interest, indicating the model efficiency and adaptability to varying coastal characteristics. The influence of training sample locations on model performance was evaluated using three distinct model configurations. The difference between these configurations was less than 5 cm, which highlights the robustness of the proposed SDB model. The quality of the satellite imagery is a significant factor that influences the accuracy of the bathymetry estimation. A preliminary methodology incorporating spectral data embedded in Sentinel-2 imagery to effectively select the most optimal satellite imagery was also proposed in this study.

摘要

本研究检验了在 Sentinel-2 卫星中使用多光谱仪器 (MSI) 并结合基于回归的随机森林模型来估计佛罗里达州西南部延伸近岸地区的水深测量。在这项研究中，我们专注于开发一个框架，该框架会导致一个广义的卫星测深 (SDB) 模型，该模型适用于利用多日期图像的广泛而多样化的沿海地区（>200 公里的海岸线）。使用机载激光雷达测深 (ALB) 完成模型校准和验证。由于 ALB 调查的执行成本非常高，因此建议的模型使用有限且实际可行的 ALB 数据样本进行训练，以扩展模型的实用性。在使用历史卫星图像的不同组合引入的三个不同子模型中，具有最大特征池的组合波段模型产生了最高的准确性。结果显示，所有感兴趣区域的 0-13.5 m 深度范围（使用的激光雷达调查的限制）的均方根误差 (RMSE) 值为 8% 或更低，表明模型效率和对不同海岸特征的适应性。使用三种不同的模型配置评估训练样本位置对模型性能的影响。这些配置之间的差异小于 5 厘米，这突出了所提出的 SDB 模型的稳健性。卫星图像的质量是影响测深估计准确性的重要因素。