EVALUATING SATELLITE DERIVED BATHYMETRY IN REGARD TO TOTAL PROPAGATED UNCERTAINTY, MULTI-TEMPORAL CHANGE DETECTION, AND MULTIPLE NON-LINEAR ESTIMATION

Acoustic and electromagnetic hydrographic surveys produce highly-accurate bathymetric data that can be used to update and improve current nautical charts. For shallow-water surveys (i.e., less than 50m depths), this includes the use of single-beam echo-sounders (SBES), multi-beam echo-sounders (MBES), and airborne lidar bathymetry (ALB). However, these types of hydrographic surveys are time-consuming and require considerable financial and operational resources to conduct. As a result, some maritime regions are seldom surveyed due to their remote location and challenging logistics. Satellite-derived bathymetry (SDB) provides a means to supplement traditional acoustic hydrographic surveys. In particular, Landsat 8 imagery: 1) provides complete coverage of the Earth’s surface every 16 days, 2) has an improved dynamic range (12-bits), and 3) is freely-available from the US Geological Survey. While the 30 m spatial resolution does not match MBES, ALB, or SBES coverage, SDB based on Landsat 8 can be regarded as a type of “reconnaissance survey” that can be used to identify potential hazards to navigation in areas that are seldom surveyed. It is also a useful means to monitor change detection in dynamic regions. This study focused on developing improved image-processing techniques and time-series analysis for SDB from Landsat 8 imagery for three different applications: 1. An improved means to estimate total propagated uncertainty (TPU), mainly the vertical component, for single-image SDB; 2. Identifying the location and movement of dynamic shallow areas in river entrances based on multiple-temporal Landsat 8 imagery; 3. Using a multiple, nonlinear SDB approach to enhance depth estimations and enable bottom discrimination. An improved TPU estimation was achieved based on the two most common optimization approaches (Dierssen et al., 2003 and Stumpf et al., 2003). Various single-image SDB band-ratio outcomes and associated uncertainties were compared against ground truth (i.e., recent Lidar surveys). Several parameters were tested, including various types of filters, kernel sizes, number of control points and their coverage, and recent vs. outdated control points. Based on the study results for two study sites (Cape Ann, MA and Ft Myers, FL), similar performance was observed for both the Stumpf and the Dierssen models. Validation was performed by comparing estimated depths and uncertainties to observed ALB data. The best performing configuration was achieved using low-pass filter (kernel size 3x3) with ALB control points that were distributed over the entire study site. A change detection process using image processing was developed to identify the location and movement of dynamic shallow areas in riverine environments. Yukon River (Alaska) and Amazon River (Brazil) entrances were evaluated as study sites using multiple satellite imagery. A time-series analysis was used to identify probable shallow areas with no usable control points. By using an SDB ratio model with image processing techniques that includes feature extraction and a well-defined topological feature to describe the shoal feature, it is possible to create a time-series of the shoal’s motion, and predict its future location. A further benefit of this approach is that vertical referencing of the SDB ratio model to chart datum is not required. In order to enhance the capabilities of the SDB approach to estimate depth in non-uniform conditions, Dierssen’s band ration SDB algorithm was transformed into a full non-linear SDB model. The model was evaluated in the Simeonof Island, AK, using Lidar control points from a previous NOAA ALB survey. Linear and non-linear SDB models were compared using the ALB survey for performance evaluation. The multi-nonlinear SDB model provides an enhanced performance compared to the more traditional linear SDB method. This is most noticeable in the very shallow waters (0-2 m), where a linear model does not provide a good correlation to the control points. In deep-waters close to the extinction depth, the multi-nonlinear SDB method is also able to better detect bottom features than the linear SDB method. By recognizing the water column contributions to the SDB solution, it is possible to achieve a more accurate estimate of the bathymetry in remote areas