Multi-Path Long-Term Vessel Trajectories Forecasting with Probabilistic Feature Fusion for Problem Shifting

TL;DR

The paper tackles long-term multi-path vessel trajectory forecasting from AIS data by fusing probabilistic route/destination features with deep learning in a phased framework. It introduces Route and Destination Probability Matrices on a hexagonal grid and integrates motion statistics into a deep auto-encoder–based trajectory reconstruction model that outputs the next $12$ hours from $1$–$3$ hours of input. The approach achieves $R^2$ exceeding $0.98$ and average/median errors of about $11$ km and $6$ km, respectively, surpassing state-of-the-art baselines and offering robust handling of complex route decisions. The work has practical impact for maritime safety and whale conservation (smartWhales) in the Gulf of St. Lawrence, showcasing the benefits of probabilistic feature fusion for open-water trajectory forecasting. Future directions include expanding route polygons, applying alternative losses, and broadening data accessibility for reproducibility.

Abstract

This paper addresses the challenge of boosting the precision of multi-path long-term vessel trajectory forecasting on engineered sequences of Automatic Identification System (AIS) data using feature fusion for problem shifting. We have developed a deep auto-encoder model and a phased framework approach to predict the next 12 hours of vessel trajectories using 1 to 3 hours of AIS data as input. To this end, we fuse the spatiotemporal features from the AIS messages with probabilistic features engineered from historical AIS data referring to potential routes and destinations. As a result, we reduce the forecasting uncertainty by shifting the problem into a trajectory reconstruction problem. The probabilistic features have an F1-Score of approximately 85% and 75% for the vessel route and destination prediction, respectively. Under such circumstances, we achieved an R2 Score of over 98% with different layer structures and varying feature combinations; the high R2 Score is a natural outcome of the well-defined shipping lanes in the study region. However, our proposal stands out among competing approaches as it demonstrates the capability of complex decision-making during turnings and route selection. Furthermore, we have shown that our model achieves more accurate forecasting with average and median errors of 11km and 6km, respectively, a 25% improvement from the current state-of-the-art approaches. The resulting model from this proposal is deployed as part of a broader Decision Support System to safeguard whales by preventing the risk of vessel-whale collisions under the smartWhales initiative and acting on the Gulf of St. Lawrence in Atlantic Canada.

Figures (14)

• Figure 1: Potential routes a vessel may take to enter the Gulf of St. Lawrence through the Cabot Strait. Whale and vessel symbols represent high vessel traffic in the region and the possibility of marine life encounters along the possible routes. Possible destinations include Quebec, New Brunswick, Nova Scotia, Newfoundland and Labrador ports. A vessel can anchor in one of these ports, continue into the St. Lawrence River, or exit the gulf through the Strait of Belle Isle. The figure illustrates that vessels can have a common start point but may take different routes, resulting in different trajectories.
• Figure 2: This study centers on the Gulf of St. Lawrence, with traffic policy zones defined by Transport Canada. These zones include voluntary seasonal slowdowns, dynamic shipping zones, and seasonal management areas, which are illustrated on the left-hand map. The regulated areas overlay main vessel traffic routes, such as the main shipping corridor between the Cabot Strait (between Cape Breton Island and Newfoundland) and the estuary of the St. Lawrence River routes (polygons 1--5). These areas cover the ports in the zone between New Brunswick and Québec (polygons 6--12) and the Jacques Cartier Strait between Anticosti Island and Northern Quebec (polygon 8). The visualization on the right-hand side displays vessel tracks (2015--2020) in the area, revealing inconsistent paths within shipping lanes.
• Figure 3: Gulf of St. Lawrence overlain with our hexagonal study grid. Hatched grid cells indicate hotspots for North Atlantic Right Whales (NARWs) o2022repatriation. Most vessels' trajectories intersect with the red route polygons at some point during their journey through the gulf. These polygons segment the gulf from north to south, covering major inbound and outbound movements inside the gulf. However, it is important to note that not all paths are required to cross a route polygon. Vessels whose paths cross the route polygons are generally easier to forecast, while those that do not pose intricate forecasting scenarios.
• Figure 4: The Gulf of St. Lawrence region has multiple paths with high vessel traffic. This figure shows four of them, which do not represent our dataset's complete set of trajectories but cover three major ones (1, 2, and 4) that cross a route polygon. Path 3, on the other hand, does not intersect with any route polygon. Instead, it represents tracks that interact with all major hotspots of NARW in the gulf.
• Figure 5: Probabilistic formulations of long-term trajectory forecasting. In the traditional method (a), the probabilistic matrix is utilized to determine the most likely whereabouts of a vessel by analyzing its movements concerning its starting point ($C_{i}$) on a grid using historical data. Our proposed approach (b) follows a similar method but with new features and a revisited understanding. In addition to historical data, the potential route depends on traversing the polygons. Essentially, our model employs route polygons as a type of memory to provide more precise guidance on the vessel's decisions at different moments in the past, aiming for a more reliable and effective probabilistic-based inference system.