Monitoring and enforcement of marine reserves can be challenging in remote parts of the world, where many of the last near-pristine waters are found.
To help meet this challenge, an American independent non-governmental organization whose mission is to serve the public interest by "improving public policy, informing the public, and stimulating civic life", partnered with the Satellite Applications Catapult, the UK innovation and technology centre created to promote research and development collaboration through the exploitation of space. Together, they have developed a system that enables government officials and other analysts to identify and monitor unlawful activities in global waters, particularly illegal, unreported, and unregulated fishing (IUU).
The Catapult launched their Virtual Watch Room in January 2015. This cutting-edge technology merges satellite tracking and imagery data with other sources of information, such as fishing vessel databases and oceanographic data, to help monitor seas across the globe.
As part of this programme of activities, the Catapult sought the help of the Smith Institute to automate the process of classifying vessel behaviour as part of its ongoing activity to end IUU fishing.
The Smith Institute worked alongside the Catapult staff and fishing analysts to develop algorithms for detecting vessels' fishing patterns from AIS (Automated Identification System) data.
AIS is primarily a collision avoidance system. It allows ships to view marine traffic in their area and to be seen by that traffic. Shipboard AIS transceivers have a horizontal range that is highly variable, but typically only up to about 74 kilometres (46 miles). They reach much further vertically - up to the 400 kilometres orbit of the International Space Station (ISS). Therefore, it can be picked up by satellites passing overhead with a receiver attached.
AIS data from vessels contains GPS data at irregularly spaced times. This could be up to 10 minutes apart during steady steaming, but more frequent during manoeuvres. In the data used in this project, some vessels were registered as engaged in fishing, and the majority were not. The scope of the project was to detect patterns in behaviour in this data that are characteristic of long line fishing (long-lining).
Long-lining is a method of commercial fishing used to catch a range of species. Long-lining consists of a vessel travelling at reduced steady speed (less than 5 knots) in a straight line for a distance up to 30 miles (while it lays out the hook lines), followed by a stationary stage often as long as 2 or 3 days, followed by the same straight trajectory to collect the hook lines.
Image courtesy of the Fisheries Research and Development Corporation
The Smith Institute started the project by constructing mathematical models and filters to detect the necessary features in the AIS data. This took the form of an extended Kalman filter to detect speed, heading and turn rate from GPS data. Feature extraction and classification of the data was then performed. This identified whether the behaviour was characteristic of fishing or not. The resulting automated classifier relies on transitions between different speeds to detect characteristic long-lining behaviour. The process we developed is similar to what is called TMA (Target Motion Analysis) in a military context.
We ran our models on test data to evaluate their performance in terms of speed, correct identification rates, and false alarm rates. As a result, over 90% of vessels received the expected classification. Moreover, a list of anomalous vessels was provided to the Catapult. The data set was not expected to be completely accurate: some of the vessels not registered as fishing may be fishing illegally. The aim of our system was to direct the attention of an analyst to those vessels whose behaviour is most closely aligned with fishing.
In addition, the automated classifier revealed an unexpected difference in fishing behaviour in different regions, strongly distinguishing between Asian and European long-liners.
Remotely monitoring fishing activity is the only practical way of combating illegal fishing in inaccessible marine reserves. The automated classifier of fishing patterns developed by the Smith Institute is an aid to analysts in the decision-making process of identifying ‘suspicious’ fishing behaviour. It focuses analysts’ attention to the most important vessels, helping them use their time most effectively.
The Catapult was informed of those vessels whose behaviour did not correspond with their stated activities in the fishing vessels database.
Although the Smith Institute's tool was not ultimately used for the Virtual Watch Room platform, it will no doubt be used in other maritime tracking activities moving forward.
The Catapult asked the Smith Institute to develop an automated classifier to help identify vessel’s fishing behaviour. The Smith Institute rose to the challenge and produced some very creative software, and although it has not been deployed in the Virtual Watch Room to date, I am sure it will be helpful and benefit a number of projects in the future.Nick WiseHead of Application Solutions at Satellite Applications Catapult