Predicting vessel activities from how they move

This is the title of the case study that should take up no more than a few lines.

nasa
Global fishing is massive, in the order of $250 billion annually, feeding and supporting 10s of millions of people. Regulation is key for sustainability, but monitoring is difficult. 
This is the space for a nicely written and engaging synopsis that will entice people to read.

Client: Secretariat of the Pacific Community (SPC) and the Fisheries Forum Agency (FFA)
Date: 2016-ongoing.

Client: The London Fire Brigade
Date: September 2018

Fishing is a heavily regulated industry in many territorial waters, or be subject to international agreements with regards the high seas. This is to ensure future sustainability of fisheres and protection or recovery of endangered species. Enforcement of any such regulation is clearly difficult given the extent of ocean and number of vessels, with illegal fishing possibly accounting for $23 billion (FAO, 2018). Monitoring of fishing is multi-facted including physical observers, enforcement officers & vessels, movement tracking, satellites, cameras, complex modelling and more. Here we deal with the acivities of registered fishing vessels in the South Pacific. 

What we did

Knowing where, when and what people are catching is important for many reasons. The catch compositions are needed for stock assessment and management, and extracting resources from a nation's territorial waters has financial implications - a fee may be levied. Due to this, legal fishing fleets are required to be tracked remotely (GPS/VMS) and record their activities. Concerns about the veracity of recorded activities leads to exensive analysis of the real-time location data of vessels. In short, what activities are vessels likely to be engaged in based on behaviour?

We have built predictive models of vessel behaviour classes based on very large ground-truthed datasets for vessels operating within the South Pacific (17 Pacific nations under the umbrella of the FFA). Vessel movements and various auxillary data provide real-time probabilistic predictions of 1000s of vessels 24 hours a day.

Some technical bits

A variety of machine-learning and statistical classifiers are fitted and compared, either naturally multi-class, or an aggregate of binary classifiers. In addition to the basic spatio-temporal data from VMS, many movement variables are derived and combined with auxillary data. Observer data provide ground-truthed behaviours for both the fitting and validation of model predictions.