On the precision of predicting fishing location using data from the vessel monitoring system (VMS)

Abstract

Defining fishing grounds based on data from vessel monitoring systems (VMS) has been a widely researched topic in recent years. Much of the research has focused on filtering algorithms for identifying fishing locations from VMS point data, most often supplemented with either imputed or reported vessel speed information. This study compared the precision of categorizing fishing locations from VMS data either by the most wide-spread “speed rule” approach or by a probability model. Using data from Northeast U.S. Fisheries for fishing years 2010–2014, we showed that the traditional representation of fishing activities as derived by speed rules leads to a severe misrepresentation of fishing grounds for gears other than bottom otter trawl. Predictions based on probability models outperformed gear-specific speed rules in classifying VMS polls for sink gillnet and scallop dredge trips, without adding substantial computational effort. The probability models thus provide the largest improvements in gears with complicated fishing patterns, while controlling for issues such as fleet dynamics that historically have not been dealt with in the static speed rules but which can have significant impacts on the quality of predictions.