Summary: | Despite numerous techniques for measuring and estimating water depth, bathymetry in the nearshore zone is notoriously difficult to map. Dangerous sea states, noisy environmental conditions, and expensive survey operations, particularly in remote areas, contribute to the difficulties of obtaining data along the coast. Global datasets, derived mainly from satellite altimetry methods, do exist, but they have significant limitations nearshore. Numerous high-resolution datasets, conventionally acquired with acoustic and lidar techniques, also exist, but they cover only a small percentage of the world's coasts. Spaceborne data fusion employing multispectral satellite derived bathymetry (SDB) offers the potential to significantly reduce the global lack of nearshore bathymetry, coined the "white ribbon" by the hydrographic community, referring to the alongshore data gap on many nautical charts. A broad term, multispectral SDB spans a diverse spectrum of methods that have been used extensively in specific case studies, but the application of multispectral SDB on a global or regional scale is significantly limited by the availability of in situ reference depths needed to tune derived values. Additionally, many existing approaches only use a single multispectral image, which can result in significant errors or missing data if the image contains environmental or sensor noise, such as clouds, sediment plumes, or detector-edge artifacts. This dissertation presents two spaceborne empirical multispectral SDB methods to address shortcomings of existing SDB approaches and reduce the global shortage of nearshore bathymetry – (1) active/passive spaceborne data fusion combining MABEL/ICESat-2 and multispectral data and (2) state space modeling of Sentinel-2 and Landsat 8 multispectral data to generate gap-free models of relative SDB (rSDB) with corresponding uncertainty estimates. The recently launched ICESat-2 mission offers an opportunity for a completely spaceborne active-passive data fusion approach to nearshore bathymetry by potentially providing a global source of nearshore reference depths to tune empirical multispectral SDB algorithms. The main objectives of the ICESat-2 mission are to measure ice-sheet elevations, sea-ice thickness, and global biomass, but ICESat-2’s 532-nm wavelength photon-counting Advanced Topographic Laser Altimeter System (ATLAS) was first posited, then demonstrated capable of detecting bathymetry in certain nearshore environments. Presented in two studies conducted prior to ICESat-2’s launch, the active-passive approach is demonstrated with data from MABEL, NASA’s high-altitude ATLAS simulator system. The first study assessed the ability to derive bathymetry from MABEL and then evaluated the accuracy and reliability of MABEL bathymetry using data acquired in Keweenaw Bay, Lake Superior. The study also developed and verified a baseline model to predict numbers of bottom returns as a function of water depth. The second study completed the demonstration of the spaceborne active/passive data fusion method by synergistically fusing MABEL-derived bathymetry and Landsat 8 multispectral Operational Land Imager (OLI) imagery over the entire Keweenaw Bay study site using the Stumpf band-ratio algorithm. The study also assessed the spatiotemporal viability of the data fusion approach by characterizing the variability of global coastal water clarity as interpreted from Visible Infrared Imaging Radiometer Suite (VIIRS) Kd(490) data. The calculated SDB agreed with a high-resolution topobathymetric lidar dataset to within an RMSE of 0.7 m, and the spatiotemporal viability analysis indicated that the spaceborne active-passive data fusion approach may be viable over many regions of the globe throughout the course of a year. State space modeling of empirical multitemporal SDB overcomes limitations of single-image SDB by leveraging the bathymetric signal in multispectral time series to create gap-free models of relative SDB (rSDB) for an arbitrary date, enabling SDB for dates with noisy or no data. State space models (SSMs) are well established in many applications but are absent in empirical SDB literature. Consisting of a state equation, which relates consecutive state vectors, and an observation equation, which relates observations to the state vector, SSMs are typically solved using Kalman filtering techniques, which provide estimates of uncertainties along with state estimates. SSMs also provide a mechanism for data fusion by allowing an observation equation for multiple observed time series. The third study demonstrates a state space approach to empirical multispectral SDB by applying local level SSMs to Landsat 8 OLI and Sentinel-2 MSI rSDB time series, both separately and fused. A representative single-sensor SSM (Landsat 8) was transformed to SDB that agreed with a high-resolution topobathymetric lidar dataset to within an RMSE of 0.29 m, which indicates the promising performance of the state space framework. Internally consistent fused-sensor SSMs verified that state space modeling also offers a data-fusion method capable of incorporating time series from a diverse suite of multispectral sensors.
|