In ocean acoustics, many types of optimizations have been employed to locate acoustic sources and estimate the properties of the seabed. How these tasks can take advantage of recent advances in deep learning remains as open questions, especially due to the lack of labeled field data. In this work, a Convolutional Neural Network (CNN) is used to find seabed type and source range simultaneously from 1 s pressure time series from impulsive sounds. Simulated data are used to train the CNN before application to signals from a single hydrophone signal during the 2017 Seabed Characterization Experiment. The training data includes four seabeds representing deep mud, mud over sand, sandy silt, and sand, and a wide range of source parameters. When applied to measured data, the trained CNN predicts expected seabed types and obtains ranges within 0.5 km when the source-receiver range is greater than 5 km, showing the potential for such algorithms to address these problems.

Experimental measurements of the peak pressure and sound exposure level (SEL) from underwater explosions collected 7 km off the coast of Virginia Beach, Virginia are presented. The peak pressures are compared to results from previous studies and a semi-empirical equation that is a function of measurement range and charge weight, and are found to be in good agreement. An empirical equation for SEL that similarly employs a scaling approach involving charge weight and range is also presented and shows promise for the prediction of SEL in shallow water.

The Intensity Vector Autonomous Recorder (IVAR) is a system that records four coherent channels of acoustic data continuously: one channel for acoustic pressure and three channels associated with a triaxial accelerometer from which acoustic particle velocity is obtained. IVAR recorded the vector acoustic field in broadband signals originating from Signal, Underwater Sound (SUS) (Mk-64) charges deployed at 5–13-km range from the fixed IVAR site (mean depth 74.4 m) as part of the 2017 Seabed Characterization Experiment (SBCEX) designed to study the acoustics of fine-grained muddy sediments. Sufficient geometric dispersion at these ranges permitted unambiguous identification of up to four modes as a function of frequency for frequencies less than 80 Hz. From time–frequency analysis of the dispersed arrivals, a single mode ( $n$ ) and single-frequency ( $f_{i})$ properties are identified at peaks in the narrowband scalar field, with time dependence corresponding to mode group speed. At these time–frequency addresses, four quantities derived from the vector acoustic measurements are formed by coherent combination of pressure and velocity channels: first, modal phase speed; second, circularity, a measure of the normalized curl of active intensity; third, depth-dependent mode speed of energy; and fourth, vertical component of reactive intensity normalized by scalar intensity. A means to compute these quantities theoretically is provided, and a comparison of model results based on a notional geoacoustic representation for the SBCEX experimental area consisting of a single low-speed mud layer over a half-space area versus a Pekeris representation based on the same half-space shows a striking difference, with the field observations also clearly at variance with the Pekeris representation. A fundamental property of mode 2, observed at the IVAR location, is a change in sign for circularity and vertical reactive intensity near 37 Hz that is posited as a constraint observation for mode 2 that must be exhibited by any geoacoustic model that includes a low-speed mudlike layer applied to this location.

Underwater noise from impact pile driving is studied through measurements using a vertical line array (VLA) placed at range 120 m from the pile source (water depth 7.5 m) over which bathymetry varied gradually increasing to depth 12.5 m at the VLA. The data were modeled assuming the pile impact produces a radial expansion that acts as sound source and propagates along the pile at supersonic speed. This leads to the conceptualization of the pile as a discrete, vertical line source for which frequency- and source-depth-dependent complex phasing is applied. Dominant features of the pressure time series versus measurement depth are reproduced in modeled counterparts that are linearly related. These observations include precursor arrivals for which arrival timing depends on hydrophone depth and influence of a sediment sound speed gradient on precursor amplitude. Spatial gradients of model results are taken to obtain estimates of acoustic particle velocity and vector intensity for which active intensity is studied in the time domain. Evaluation of energy streamlines based on time-integrated active intensity, and energy path lines based on instantaneous (or very-short-time integrated) active intensity reveal interesting structure in the acoustic field, including an inference as to the source depth of the precursor.

Deep learning techniques have the potential to address real-time source localization and seabed characterization from passive ocean acoustic signals. The challenge grows when considering single-sensor cases that lack absolute travel time with no known source or environmental parameters. To address this, convolutional neural networks (CNNs) are trained to predict the source–receiver range and the seabed type on extracted and normalized 1-s pressure time series from explosive charges measured during the 2017 Seabed Characterization Experiment. To investigate the impact of ocean variability, two training data sets containing different sets of water sound-speed profiles (SSPs) are used. CNNs trained on the first synthetic data set, which contains a static ocean depth but significant variability in water SSPs, predict seabed class more accurately than the second data set from the measured data. Conversely, CNNs trained on the second synthetic data set, which contains fewer SSPs (similar to those measured during the experiment) and multiple ocean depths, predict the source–receiver range more accurately than the first data set from the measured data. The CNNs, with a degree of uncertainty, predict the seabed type and the source range when trained on synthetic data that have been stripped of source–receiver information, such as arrival time and amplitude. The CNNs trained to only predict the seabed type also make more accurate predictions on the seabed type when compared to those for multiple prediction types. This study highlights the importance of accounting for ocean variability and prediction uncertainties in training deep learning algorithms for marine applications.

Measurements of acoustic sea surface backscattering, wind speed, and surface wave spectra were made continually over a 24-h period in an experiment conducted in 26 m of water near the Dry Tortugus collection of islands off south Florida in February 1995. The backscattering measurements were made at a frequency of 30 kHz and a sea surface grazing angle of 20°; a time series of the decibel equivalent of this variable, called SS20, was studied in terms of its dependence on environmental variables. On occasion reliable estimates of scattering in the grazing range 15°–27° were also obtained during the 24 hours. The scattering data exhibited evidence, in terms of scattering level and grazing angle dependence, of scattering from near-surface bubbles rather than scattering from the rough air–sea interface. The scattering data were compared with a model for ${\sigma }_b,$ the apparent backscattering cross section per unit area due to bubble scattering, that is driven by a parameter, ${\beta }_I,$ equal to the depth-integrated extinction cross section per unit volume. Using an empirical model for ${\beta }_I$ based on data from a 1977 experiment conducted in pelagic waters, model predictions agreed reasonably well with the 1995 measurements presented here. Additional model–data comparisons were made using four measurements from a 1992 experiment conducted in pelagic waters. Finally, the 24-h time series of acoustic scattering exhibited a hysteresis effect, wherein for a given wind speed, there was a tendency for the scattering level to be higher if prior winds had been falling. A better understanding of this effect is essential to reduce uncertainty in model predictions.