Characterization of the seafloor at an underwater site with unexploded ordnance (UXO) is key to the ability to predict both the mobility of those munitions under different environmental conditions and the ability to detect and classify those munitions using sonar. Traditional seafloor measurement techniques such as grab sampling, collection of diver cores, and seafloor imaging using camera systems can be expensive and time-consuming when used to characterize a large site. The physics-based multibeam echosounder inversion developed and evaluated under this project provides quantitative measurements of sediment properties over large areas without the need for additional in situ sediment measurements. The objective of this work was to demonstrate the performance of the inversion using data collected for different sediment types, develop techniques to monitor the performance of the inversion and estimate the uncertainties in its output, and determine how best to utilize the inversion for wide-area sediment characterization.

To evaluate and improve the performance of the inversion, sonar data were collected along ten 200-m-long lines within Sequim Bay in 2019. The seafloors along these lines were composed of different sediment types, were at different depths, and had various topographic features. In addition to sonar data, ground truth measurements were collected at the center of each line such that each inversion output could be compared to a measured value. This dataset was also used to develop and evaluate quality control metrics which can be used to assess the performance of the inversion, and to enhance the algorithm such that it could provide uncertainty estimates for each of the inversion outputs. A second sonar dataset was collected in 2021 to assess the performance of the inversion when applied to survey data covering a large area. 

Using the 2019 Sequim Bay dataset, the inversion algorithm was shown to produce accurate estimates of the seafloor properties with two exceptions; the seafloor roughness and the mean grain size. While the inversion could provide a good estimate of the seafloor scattering strength, it was found that for near-nadir incidence angles the roughness spectral strength and exponent are correlated and only a combination of the parameters can be accurately estimated. This has the potential to produce significant uncertainties when using these parameters to predict scattering strength at shallow grazing angles. While the algorithm overpredicted the mean grain size for the soft sediments at the site, however, the volume scattering strength and the grain size were found to be correlated and this could provide an alternative approach for estimating the mean grain size. For the 2021 dataset, the algorithm was found to be robust when processing a large multibeam echosounder survey dataset with survey lines 1-2 km in length. Maps interpolated from this output were found to estimate the sediment properties to the same accuracy as the 2019 dataset inversion.

The inversion development is at the point where it is sufficiently robust that it could be used to process data collected with a Teledyne-RESON Seabat T-50 or T-20 echosounder and produce accurate maps for sediment types ranging from coarse sand to fine silt and for mixtures of these sediments. For seafloor types not captured in the 2019 and 2021 studies, such as those with cobble, coral, or seagrass, the inversion should raise quality control flags notifying the user that the inversion will not be accurate for these seafloors. While these seafloors could be captured in future improvements to the algorithm, there is a clear benefit to prioritizing the dissemination the Matlab code for UXO site monitoring and remediation. This can be done with the addition of documentation detailing the hardware needs, data collection protocols, and use of the Matlab code for users of Seabat sonars. The code can also be modified such that the Seabat sonar data pre-processing is moved to a self-contained module, such that users of different multibeam echosounders can write their own pre-processing module and invert data from their sonar.