Abstract
Seychelles high seas tuna fleets have a high AIS use with a transmission frequency considerably higher than that of VMS. However, AIS has far fewer transmissions than VMS and many more gaps in transmission longer than a few hours. The spatial coverage of the AIS data is good for Seychelles longline vessels, with acceptable coverage over the core fishing grounds. By contrast, AIS data are deficient for purse seiners and supply vessels with most data only present around ports due to the switch-off behavior linked to the piracy threat.Consistent with data coverage, AIS seems to be very useful in describing the spatiotemporal patterns of the longline fishery and for identifying fishing hotspots. The GFW neural net algorithm predicts well the fishing operations for longliners but predictions for purse seiners are not informative. Metrics for effort at the scale of 5° x 5° squares, such as those typically used by tuna regional fisheries management organizations (RFMOs) for longline fisheries, are well correlated between logbooks and GFW algorithms. Thus, GFW is able to accurately distinguish fishing from non-fishing activities for longliners. However, the frequent breaks in transmission, perhaps due to issues with AIS reception, lead to consistent underprediction by AIS and GFW algorithms of the "true" patterns shown using VMS and logbook data. The increased satellite coverage observed between 2016 and 2017 resulted in improved GFW algorithm performance in deriving estimations of longline fishing effort.The relationships between GFW predictions of longline fishing and effort could be useful in data-poor fisheries where poor collection and management systems may prevent the reporting of spatial effort to the RFMO. In such cases, the availability of AIS or VMS data combined with information on the number of hooks deployed per operation may enable predictions of gridded effort, which would improve compliance with the Conservation and Management Measures.