Fusion of remote sensing and machine learning for high-resolution bathymetric modeling in shallow coastal zone

Abstract

In recent times, space-borne remote sensing techniques have emerged as a compelling and innovative alternative to traditional methods for extracting bathymetric data, which is essential for mapping the underwater topography of coastal zones. These modern techniques offer significant advantages, primarily their capability to efficiently acquire data over vast spatial areas and provide high-frequency temporal monitoring, which is crucial for dynamic coastal environments. One specific application of this technology, known as Satellite-Derived Bathymetry (SDB), utilizes multispectral images acquired by satellite constellations to survey shallow waters. A key benefit of SDB is its operational simplicity and cost-effectiveness, as it requires no mobilization of personnel or specialized equipment, thereby delivering rapid bathymetric data and generating substantial savings in both time and financial resources compared to conventional acoustic surveys. SDB methodologies can be broadly categorized as either passive or active. Passive SDB, which is the focus of this study, relies on analyzing the reflected solar radiation from the seafloor, which is captured by satellite-based optical sensors. The accuracy of this method is heavily influenced by water clarity and is generally limited to relatively shallow depths. In contrast, active SDB utilizes an emitted energy source, most notably airborne or spaceborne LiDAR (Light Detection and Ranging). While active methods primarily involve direct measurements via LiDAR, another class of active systems, such as satellite altimetry, indirectly infers bathymetry in the deep ocean by measuring sea surface height anomalies. This study, however, focuses on the passive approach, which is most suitable for coastal zones and the Göktürk-1 sensor. The Göktürk-1 satellite, a high-resolution Earth observation satellite launched in 2016, plays a pivotal role in this study. Its primary mission is to fulfill the intelligence and reconnaissance needs of the Turkish Ministry of Defence. However, its high-quality imagery is also made available to the public sector and academic communities, creating a unique opportunity for scientific research. The satellite's onboard sensor operates on the pushbroom principle and has a swath width of 20 kilometers at nadir, capturing multispectral images with a resolution of 2 meters and panchromatic images at a resolution of 0.5 meters. Despite its excellent technical specifications, numerous previous studies have focused exclusively on the satellite's performance in atmospheric sciences and land applications. Until now, there has been a notable research gap in the marine domain, particularly concerning its potential for SDB applications. This thesis aims to address this critical gap by conducting the first-ever comprehensive assessment of the performance of Göktürk-1 satellite images specifically for SDB. The field of SDB can be broadly categorized into two distinct methodological approaches: analytical and empirical. The analytical approach is based on physical models that simulate the complex interaction of light with the water column and the seafloor. In contrast, empirical methods are data-driven and rely on establishing a statistical relationship between the measured satellite-derived spectral values and known bathymetric data points. The success of an empirical model is directly tied to the quality and quantity of the reference data used for its calibration and validation. An increasingly popular and powerful subset of empirical methods is the use of machine learning algorithms. These advanced computational models, such as Support Vector Machines and various forms of neural networks, can better capture the complex, non-linear relationships between satellite imagery and water depth, often providing superior accuracy and robustness compared to traditional linear models. This study, acknowledging the practical constraints and the goal of developing a repeatable and implementable methodology for the Office of Navigation, Hydrography and Oceanography, maintains a specific focus on the empirical application of SDB, leveraging the power of these modern techniques. Historically, SDB has been an area of scientific inquiry for over 40 years. For most of this period, its practical adoption was limited because the spatial resolution of widely available open-access satellite images was often unsatisfactory for detailed hydrographic work. This limitation contributed to the International Hydrographic Organization (IHO)'s hesitation to officially recognize SDB as a valid method. The situation changed dramatically with the launch of the Landsat-8 and Sentinel-2 missions, which provided the global community with access to higher-resolution, open-access satellite imagery. This technological advancement significantly improved the quality and reliability of SDB results, paving the way for its potential integration into operational workflows. Despite the significant advancements and official recognition by the IHO, the practical application of SDB is not without its challenges. The accuracy and effectiveness of SDB are highly dependent on environmental and physical factors that can vary greatly from one location to another. Key limitations include the inherent optical properties of the water column, such as turbidity, and the presence of dissolved organic matter. These factors can significantly attenuate the light signal, limiting the maximum depth to which bathymetry can be reliably extracted. Other variables, such as variations in the seabed type and reflectance, also introduce complexities. Furthermore, external factors like atmospheric conditions, sun glint, and even tidal fluctuations must be meticulously accounted for. These site-specific complexities underscore why a generalized, one-size-fits-all approach to SDB is not viable and emphasize the critical need for localized studies like this one, which validate the methodology under the unique conditions of a specific coastal area and with a new satellite sensor. This long-standing barrier was finally removed in 2020 when the IHO, a global authority on hydrographic standards, published Edition 6 of the 'IHO Standards for Hydrographic Surveys'. This updated standard officially recognized SDB as a valid method for hydrographic surveys by specifying the horizontal and vertical accuracies that could be achieved. This landmark publication was a turning point, marking SDB's transition from a scientific curiosity to an officially sanctioned hydrographic tool. Before the commencement of this research, the Office of Navigation, Hydrography and Oceanography lacked the internal capability to extract bathymetry using the SDB methodology. The successful outcomes of this study have demonstrably built confidence in both the Göktürk-1 data source and the SDB application domain. As a direct result of this research, Office of Navigation, Hydrography and Oceanography is now poised to incorporate SDB into its official nautical chart production lines for the first time, representing a significant advancement in the country's hydrographic capabilities and a concrete validation of the thesis's findings. This achievement is a testament to the practical applicability of the research and its potential to enhance maritime safety and efficiency.