LWV21269 pelAcousticAI: automatic processing of echosounder data from pelagic fishing vessels and derived product to science and the fishing industry - BO-65-004-008

The Dutch pelagic fishing industry is actively engaged in data collection to expand the knowledge base on fish stocks to enable better management of fish stocks. In addition, the sector uses this data to develop more selective fishing methods. Through this integrated sustainability approach, the industry aims to maintain its "license to fish" in existing fisheries and create new fishing opportunities in potentially new areas. 

The objectives of this project are twofold: 1) to make an entirely new source of fish stock biomass information available to the scientific assessment process, based on high-resolution, long-term fish observations collected by downward looking active acoustic systems (so called echosounders) that are in use on the commercial fishing fleet, and 2) to provide fishermen with a science-based aid to fishing, in the form of real-time viewing of processed data (e.g. fish school metrics) and historical fish distribution maps. The current industry-driven data collection program is generating an ever-increasing wealth of echosounder and biological data that enables monitoring of fish stocks at a fine spatial and temporal scale.

However, there is a bottleneck in the processing of these data which hampers the ability to achieving the above objectives. First, the volume of data is increasing to the point that the capacity of traditional data processing methods is not appropriate. Second, the inherent variability of the data collected onboard commercial vessels (noise and varying setup) requires flexible data processing methods that are currently not available. In that context, this project will  develop data processing workflows using advanced machine learning to enable the routine processing of data onboard commercial trawlers. Sharing of the methods developed will be an essential part of the project since the information obtained from the commercial data is to be used for fish stock estimation and other scientific purposes.

Processing the wealth of data available at the fleet level, which is necessary to achieve the industry goal, requires flexible data processing methods that are not currently available. Problematic is the volume of data that is increasing to the point where the capacity of traditional data processing methods is no longer adequate. Another challenge is the inherent variability of the data collected on board commercial vessels, caused by the variability of the sensors, the noise characteristics of the vessels on which they are installed, and the variability in sensor setup and configuration by the user. In that context, this project will realise the routine processing of data on board commercial trawlers through the development of data processing workflows that use advanced machine learning. Transparency in the methods developed will be an essential part of the project, as the information obtained from commercial trawlers will contribute to sustainable fisheries through further improving fish stock assessments for data-rich stocks and making stock assessments possible for stocks that suffer from data deficiencies.