Developing Decision Support with Bayesian Networks in Fisheries Surveillance

ABSTRACT

An application of Bayesian networks for decision support in surveillance of mackerel fisheries in the North Sea is presented. The application area involves assessment of each fishing vessel’s risk for acting irregularly with regard to release of caught fish, and then combining these risk assessments to suggest vessels on which to concentrate. The knowledge underlying the Bayesian model developed is acquired through extensive communication with the domain experts. Following the approach of information systems design research, we have developed a software tool and an understanding of the organisation’s needs in order to effectively realise use of the tool. The prototyping development process used is model-centric in the sense that development of the Bayesian networks drives the other activities in the project, such as establishing data sources and overall system requirements. Through further investigations and development of the Bayesian modelling approach presented here, including applying it to other risk areas, we have the potential to make both the operational practices in Norwegian fisheries surveillance more efficient as well as to improve the quality of their operational decisions. In our continuing efforts we will apply, evaluate and further improve on the work processes established during the development of this prototype.