When behaviour reveals activity: Assigning fishing effort to métiers based on VMS data using artificial neural networks

The identification of groups of vessels with the same exploitation pattern (e.g. gear used, fishing ground, target species) over time, usually referred to as a “metier”, is a major topic of fishery management. However, metier detection on fishing trips is still done using the incomplete or biased information present in declaration of landings (logbooks), under the assumption that the reported landings profiles reflect intended catches. Nowadays, fishing effort can be tracked at high spatial resolution using vessel monitoring systems (VMS). VMS potentially provides information about vessels fishing activity if the frequency of signals is natively high or appropriately interpolated. An artificial neural network is used to analyse interpolated VMS tracks and Vessel Register data to identify fishing activity. A multilayer perceptron network (MPN) was trained to recognize one among 15 possible metiers from a series of 33 variables: 12 in binary form for licensed gears, 6 probability classes for vessel speed, 3 for vessel heading and 7 for sea depth, respectively. The MPN was iteratively trained on subsamples of a large dataset corresponding to the activity of the Italian fishing fleet, for which information about metier was collected and validated by on-board observations by scientific operators, and then tested on other subsets of the data. The best architecture for MPN was identified and analysed. The mean percentage of correct predictions obtained on the test datasets was very high (>94%), confirming that VMS data can provide information on vessel activity. Overall, these findings suggest that this is a promising approach to assign fishing effort, resolved at single trip scale, to specific metiers, even giving independent assessment of fishing activity with respect to those provided by logbook and capture data.