Estimation of fish catch potential using assimilation of synthetic measurements with an individual-based model

Abstract

A large fraction of costs in wild fisheries are fuel related, and while much of the costs are related to gear used and stock targeted, search for fishing grounds also contributes to fuel costs. Lack of knowledge on the spatial abundance of stocks during the fishing season is a limiting factor for fishing vessels when searching for suitable fishing grounds, and with better planning and routing, costs can be reduced. Strategic and tactical decision-making can be improved through operational decision support tools informed by real-time data and knowledge generated from research. In this article, we present a model-based estimation approach for predicting catch potential of ocean areas. An individual-based model of herring migrations is combined with an estimation approach known as Data Assimilation, which corrects model states using incoming data sources. The data used to correct the model are synthetic measurements generated from neural network output. Input to the neural network was vessel activity data of over 100 fishing vessels from 2015-2018, targeting mainly herring. The output is the predicted normalized density of herring in discrete grid cells. Model predictions are improved through assimilation of synthetic measurements with model states. Characterizing patterns from model output provides novel information on catch potential which can inform fishing activity.