Predicting gear used in a multi-gear coastal fleet

Knowledge of the gear used in multi-gear fisheries is crucial for supporting fisheries management. Still, the high complexity and lack of data in the Portuguese multi-gear coastal fleet compromise this task. The present study developed a method to predict main fishing gear used in each fishing trip for the Portuguese multi-gear coastal fleet based on landing records (species caught, port, and month of landing). Landing records were used to predict gear (available for part of the fleet with electronic logbooks) using a machine learning model (random forest). This model was then applied to the remaining trips of the fleet, without electronic logbooks, to predict the gear used. A total of six gear types were considered: bivalve dredges, traps, gillnets, trammel nets, drifting longlines, and bottom longlines. The overall model prediction error was 14 %; bivalve dredges and longlines had the lowest errors, and trammel nets and gillnets were the highest. The study sheds new light on important aspects of the dynamics of this fleet, namely a decreasing trend in the use of longlines, poor electronic logbook coverage for some gear types, and greater diversity in the catches obtained with nets compared to other gear types.