Predicting Catch-Per-Unit-Effort Using Semantic Trajectories and Machine Learning

Abstract

This work presents a framework for the prediction of catch-per-unit-effort (CPUE)—an important index in the assessment of fisheries resource exploitation—using three data sources from the North Adriatic region: fishing vessel tracking data (obtained from AIS), the associated daily landing reports (i.e., amount and species of catch per vessel), and the relevant environmental data. As a part of this framework, two high-level spatio-temporal representations of the data were constructed through the use of trajectory modelling and fusion of the data sources: a set of semantic trajectories of the fishing trips, and gridded spatio-temporal maps of CPUE. While both representations can have various applications in fisheries management, here they were used for the task of CPUE prediction. Our prediction results demonstrate the potential of Machine Leaning methods for this task. This framework could be used to facilitate data-driven and evidence-based policy making in other regions with intense fishing activities.