Learning and predicting fishing activities from AIS data

Abstract

Marine life significantly impacts our planet by providing essential resources such as food, oxygen, and biodiversity. The surveillance of illegal fishing activities is critical for the sustainable management of marine resources. In this study, we investigate the use of two publicly available data sources to automatically detect fishing activities. One dataset is the Automatic Identification System (AIS) data, the other is the Catch Reports from the Norwegian fishing authorities. The data from fisheries along the Norwegian coast, specifically from vessels with a length of 15 meters or more, covers a period of 1 year and 11 months (January 2022 to November 2023). The AIS data was cleaned by removing duplicates and outliers, then interpolated using piecewise linear interpolation and resampling at a five-minute rate to turn the AIS data into time series of one-hour length. The time series were then tagged using the Catch Report data. The dataset is severely unbalanced with fishing and non-fishing tags. For better learning, the total data was resampled to create 100 balanced bootstrap sets, ensuring equal representation of fishing and non-fishing activities. This process resulted in a benchmark dataset containing about 30 million data points in about 2.5 million time series. We applied a range of classification methods, based on random forests and deep convolution networks. Three types of features were used, related to secant speed, distance to shore, and curvature. We achieved an accuracy ranging from about 92% to 93.7% depending on the features and classification methods used. The predictive capabilities of the classifiers are investigated and significance of the features are studied. The uncertainty of the classification was assessed using the bootstrap sets, providing robust evaluation metrics. Overall, this study demonstrates the effectiveness of leveraging AIS and Catch Report data and advanced data processing techniques for automatic and accurate monitoring of marine fishing activities.