Deep Learning for Spatiotemporal Modeling of Illegal, Unreported, and Unregulated Fishing Events

Illegal, unreported, and unregulated (IUU) fishing encompasses a range of fishing practices that are considered environmentally harmful and that violate national and international regulations. IUU fishing is an ongoing challenge that undermines efforts to nurture sustainable fisheries. In the United States alone, over 13% of seafood imports in 2019 were estimated to be from IUU fishing. To combat IUU fishing, various means, including legislative, law enforcement, and technological, are used to track the movement of fishing vessels and fish catches globally. However, to effectively track the prevalence of IUU fishing events geographically and over time, a highly accurate model is necessary. Therefore, in our work, we present a spatiotemporal IUU fishing forecasting algorithm to predict IUU fishing events over time and in different geographic locations. We train and evaluate our model on disjoints subsets of a dataset of IUU fishing events associated with the Atlantic bluefin tuna (Thunnus thynnus) and southern bluefin tuna (Thunnus maccoyii) species in North America over the time period 1950–2014. Our model achieves a final mean absolute error of 0.085, and is able to capture the trend of IUU fishing event severity over time. Accurate forecasting of IUU fishing events can enable effective enforcement actions to advance sustainable fisheries.