Real-time reporting of ecosystem metrics from acoustic sensors on gliders

The assessment of marine pelagic ecosystems poses a number of methodological challenges for sampling: the requirement for high spatial and temporal resolution, concurrent biological and environmental information, and behavioural responses to sampling equipment. Active acoustic techniques are now routinely used to resolve the high resolution distribution of marine organisms (from zooplankton to fish and larger organisms), typically from large research ships. The ability of Autonomous Underwater Vehicles (AUVs) such as underwater gliders to carry active acoustic sensors pertinent to ecosystem research has only recently been explored. However, due to their high data volume creation, these sensors currently store data locally for retrieval and analysis once the platform is recovered.
One of the large appeals of gliders is directing them to regions of interest and receiving data in real-time. The current simple echosounder integrated into gliders generates 256 byte strings per ping (ping rate of 0.25 - 1 Hz). Newer, more complex wideband echosounders an order of magnitude more. These data need processing and constraining into metrics (acoustic area backscattering strength, vertical distribution, aggregation) onboard the glider, which can then be transmitted back. Thus enabling ecosystem descriptors to be transmitted back to shore from the glider/AUV in real-time and realizing these platforms capabilities for ecosystem research relevant to both fisheries management and impact assessments.
This PhD project will work with state-of-the-art acoustic instruments to develop on-board processing capabilities to realize this challenge.
When acoustic data are displayed in echograms, aggregations and scattering of zooplankton and fish are evident forming diverse spatial patterns such as schools, shoals and diffuse clouds. Echotrace classification techniques enable this complex information to be simplified. This project will use existing acoustic datasets from wideband and narrowband echosounders to develop advanced compression methods that will allow the pertinent information to be transmitted down the low-bandwidth channel and hence influence the mission. Given that onboard processing is limited, our initial approach will be to state-of-the-art machine learning methods such as Deep Neural Networks (currently being used by us in lip-reading) which are trained expensively (offline) but need little run-time computation. Our aim is to develop a compression hierarchy in which the most needed information is sent first, followed by the nuances.
The student will work with glider and acoustic instrument manufacturers to implement the developed metrics and processing capabilities into a glider deployment. BAS and UEA deploy gliders in a number of environments (e.g. North Atlantic, Antarctic) and it is envisaged the student will use one of these opportunities to implement their technique.