Learning fishing information from AIS data

• Published in: Proceedings of the 2nd ACM SIGSPATIAL International Workshop on Animal Movement Ecology and Human Mobility
• Main Authors: Pons Recasens, Gerard, Bilalli, Besim, Abelló Gamazo, Alberto, Blanco Sánchez, Santiago
• Other Authors: Universitat Politècnica de Catalunya. Departament d'Enginyeria de Serveis i Sistemes d'Informació, Universitat Politècnica de Catalunya. inSSIDE - integrated Software, Services, Information and Data Engineering
• Format: Other/Unknown Material
• Language: English
• Published: Association for Computing Machinery (ACM) 2022
• Subjects: Àrees temàtiques de la UPC::Informàtica::Sistemes d'informació; Ships > Automatic identification systems; Squid fisheries; Data mining; AIS data; Analytics; Environmental data; Automation; Data analytics; Fisheries; Fishing vessels; Potential values; Satellite data; South Atlantic; Spatial cells; Training dataset; Ships; Embarcacions de pesca; Calamars > Pesca; Mineria de dades; South Atlantic Ocean
• Online Access: http://hdl.handle.net/2117/385050; https://doi.org/10.1145/3557921.3565543

The Automatic Identification System (AIS) allows vessels to emit their position, speed and course while sailing. By international law, all larges vessels (e.g., bigger than 15m in Europe) are required to provide such data. The abundance and free availability of AIS data has created a huge interest in analyzing them (e.g., to look for patterns of how ships move, detailed knowledge about sailing routes, etc.). In this paper, we use AIS data to classify areas (i.e., spatial cells) of the South Atlantic Ocean as productive or unproductive in terms of the quantity of squid that can be caught. Next, together with daily satellite data about the area, we create a training dataset where a model is learned to predict whether an area of the Ocean is productive or not. Finally, real fishing data are used to evaluate the model. As a result, for blind movements (i.e., with no information about real catches in the previous days), our model trained on data generated from AIS obtains a precision that is 18% higher than the model trained on actual fishing data-this is due to AIS data being larger in volume than fishing data, and 36% higher than the precision of the actual decisions of the ships studied. The results show that despite their simplicity, AIS data have potential value in building training datasets in this domain. Peer Reviewed Postprint (author's final draft)