Real-time reporting of marine ecosystem metrics from active acoustic sensors
Marine autonomous vehicles (MAVs) carrying active acoustic sensors (echosounders) are being used for ecosystem research, but high data volumes are presenting challenges for data storage, processing and communication. One of the appeals of autonomous vehicles is directing them to regions of interest...
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2020
|
| Subjects: | |
| Online Access: | https://ueaeprints.uea.ac.uk/id/eprint/79527/ https://ueaeprints.uea.ac.uk/id/eprint/79527/1/2020BlackwellRPhD.pdf |
| Summary: | Marine autonomous vehicles (MAVs) carrying active acoustic sensors (echosounders) are being used for ecosystem research, but high data volumes are presenting challenges for data storage, processing and communication. One of the appeals of autonomous vehicles is directing them to regions of interest and receiving data in real-time, but current satellite networks have insufficient bandwidth for real-time acoustic data transmission. We seek solutions using data compression or summarisation. We first explore the use of generic, lossless data compression algorithms (e.g. ZIP) and find that they do not deliver the necessary reduction in data size. We then convert acoustic data to echograms and examine the role of colour palettes in echogram interpretation, but image compression is still unsatisfactory. Using echosounder data from the Southern Ocean ecosystem at South Georgia, collected by research vessels (which are easier to work with and more readily available than MAV acoustic data), we compute acoustic summary metrics and assess their correlation to independent ecosystem indices. There is a strong correlation between abundance and traditional krill density estimates (r = 0.83, p < 0.01) and location (centre of mass of acoustic backscatter) and chlorophyll (r = −0.7, p < 0.01) suggesting that acoustic summaries could be used as concise ecosystem descriptors. Aliased seabed is a corruption caused by acoustic reflections and its removal is an example of an acoustic processing step that is currently under-taken manually. We use modern machine learning techniques and develop a conventional algorithm to detect aliased seabed automatically in single frequency, split-beam echosounder data without the need for bathymetry. Finally, we demonstrate an unsupervised acoustic data processing system (RAPIDKRILL) that can transmit acoustically derived ecosystem indicators in real-time via the Iridium satellite network. The technology is fully autonomous, low-cost, and could be further developed for use on MAVs. |
|---|