Abstract
A numerical model for simulating the localization performance of a three- or four-hydrophone towed passive acoustic monitoring (PAM) array on multiple species clusters was developed that allows the Bureau of Ocean Energy Management (BOEM) to assess the localization efficacy of towed PAM arrays proposed in mitigation and monitoring plans for offshore wind farm development and other operations. Simulations of a 200-m aperture array were run for three marine mammal signal types of interest: sperm whale, right whale, and beaked whale. The ability to localize a marine mammal call using PAM systems commonly deployed for mitigation surveys conducted in support of offshore wind farm development requires several conditions and assumptions. In order to localize, the call must first be detected. A detection’s range will be dictated by the received amplitude of the signal, which in turn depends on the species, distance, and relative orientation to the receiver, as well as the noise conditions of the monitoring environment. The goal of this project was the development of an algorithm and user interface with the intent to enable BOEM personnel to input proposed array specifications and determine the theoretical localization capability for low-, mid-, and high-frequency cetaceans within 5 km of the array.
The parameters set by the project required the use of curved-wavefront localization (CWL), which is equivalent to hyperbolic localization in the time domain. Roughly speaking, a CWL becomes possible whenever a source is close enough to the array such that the apparent azimuth of the source will appear to vary along the array, when measured by pairs of adjacent hydrophones along the array. If expressed in terms of hyperbolic localization, CWL means that hyperbolas produced by pairs of hydrophones will eventually intersect and will not become asymptotically parallel at large ranges. The term “curvedwavefront localization” arises because, in order for an array to localize in range, the wavefront of a signal passing over an array cannot be a plane wave structure but must display curvature.
The fundamental approach of the algorithm development was to run multiple iterations of a marine mammal localization problem, randomly varying relevant parameters to estimate the bias and variance of the resulting range estimates. Each iteration generated a random realization of a marine mammal signal at each receiving hydrophone by randomly generating different source parameters, hydrophone positions, and the background noise time series. The simulated time series were then cross-correlated between hydrophones, and the desired localization algorithm applied. The process was repeated at all grid locations, and then additional iterations were conducted to build a distribution of range estimates from which the bias and variance of the range estimate can be computed for every grid point.
Two localization algorithms were applied, based on array geometries: a three-element case for the time-of-arrival algorithm, where the hydrophones can have any spacing, and a four-element case for the “cross-fixing” algorithm, where the hydrophones are divided into two pairs to form two short subapertures. For both approaches, the mean (unperturbed) positions of the hydrophone elements were used in the localization calculation. Localization algorithms that require a larger number of hydrophones were not modeled because the majority of existing systems do not use more than four hydrophones for localization during real-time mitigation monitoring, and because the determining factor for the range resolution of a system is the total aperture (i.e., maximum separation between array hydrophones). A maximum 5-km radius monitoring grid with 1-m resolution was used to evaluate localizations at each grid point of interest within the algorithm.
The algorithm was tested using a grid search method against the dataset published by Abadi et al. (2015, 2017). This dataset is key to the algorithm testing because it provides a peer-review standard that statistically correlates acoustic detections with visual observations. The algorithm was developed to enhance localization capabilities where no direct path exists between the source and receiver.
The main insight from these simulations was that the uncertainty in range estimation is dominated by the uncertainty in array element position, likely due to the fact that the specified noise levels are too low, having been derived for measurements from a stationary platform well away from vessel noise. The introduction of vessel noise in towed PAM array systems is expected to increase localization uncertainty when signals are detected. In consultation with BOEM, example background noise profiles from actual PAM operations will be explored to determine when background noise levels become a determining factor for PAM array performance, as a means to test simulations for conditions that may be present during offshore wind farm development.
The draft and final algorithm consists of a spreadsheet; a platform-independent, stand-alone software package; and a user manual. The spreadsheet contains the formulas from the study, so that the user can quickly determine whether a given towed array configuration has any chance of meeting the localization requirements of an application. The stand-alone software package is written in MATLAB but will be processed with the MATLAB compiler to generate a stand-alone application package that can be distributed on a royalty-free basis. The stand-alone application does not require MATLAB to be executable. The advantage of this approach is that the MATLAB compiler can generate versions compatible with most hardware platforms, including Microsoft Windows and Apple operating systems. It can also generate web-based versions and stand-alone versions on a desktop or laptop computer, so that BOEM could put the tool online for public use.