Fisheries advice is only as good as the data it is based on. One of the most informative ways of improving stock assessment reliability is increasing the consistency and length of the time-series used and ensuring complete spatial coverage of a stock. One method of achieving this is through combining data from multiple fisheries-independent surveys, reducing the need for specialised surveys targeting individual species. However, despite best efforts to standardise survey design, considerable differences persist between gears deployed, preventing reliable data integration.
Using snow crab (Chionoecetes opilio) abundance data from the Norwegian Snow Crab Survey in the Barents Sea, this study applies a novel machine learning approach of three successive random forest models, evaluated using spatial cross validation, to convert catches from Agassiz trawl and pot stations to a video standard.
Incorporation of environmental predictors increased the reliability of gear conversion factors when compared to a baseline regression model using station pairs within a 3NM distance and when compared to a regression using the expanded dataset. The subsequent inclusion of the converted Agassiz trawl and trap gear catches improved the accuracy of abundance predictions across the Barents Sea when compared to predictions based solely on data from video transects.
promotor/supervisor feedback
nothing yet
