Ship Voyage Analysis: Unraveling the Dynamics

of Turning Maneuvers using AIS Data

Name: Shahnoor
Student No. 2100893

Supervised by
Luigia Petre
Mikael Manngård

A thesis submitted in fulfillment of the requirements

for the degree MSc Information Technology, Computer Science

School of Information Technology

Åbo Akademi University
13/06/2023
Declaration

I, Shahnoor, declare that the work in this dissertation titled “Ship Voyage Analysis:
Unraveling the Dynamics of Turning Maneuvers using AIS Data” is carried out by
me. This work is submitted to Åbo Akademi for the award of a degree or educational
qualification. I also declare that the information published in this dissertation has
been obtained and presented in accordance with academic rules and ethical conduct.
Any information obtained from other sources has been properly referenced.

i
Acknowledgement

First of all, I express my gratitude to the family, who blessed me with the zeal and
enthusiasm to complete this research work successfully. I am extremely thankful to
my supervisors Luigia Petre, Senior university lecturer, Department of Information
technology, Computer Science, Åbo Akademi, and Mikael Manngård, Project Man-
ager, Novia University of Applied Sciences for their motivation and tireless efforts
to help me get deep knowledge of the research area and for supporting me through-
out the life cycle of my thesis. This work was done within the Applied Research
Platform for Autonomous Systems (ARPA) project funded by the Finnish Ministry
of Education and Culture.

ii
Table of Contents

1 Introduction 1
1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Research Aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Research Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 AIS Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5.1 What is AIS Data? . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5.2 Working of AIS . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5.3 Limitation of AIS Data . . . . . . . . . . . . . . . . . . . . . . 12

2 Literature Review 14

3 Methodology 19
3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Raw Data Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3 Data Wrangling and Preprocessing . . . . . . . . . . . . . . . . . . . 22
3.4 Feature Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.5 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.6 Data Analysis Methods . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.7 Wheel-Over Point and Turning Radius . . . . . . . . . . . . . . . . . 29
3.7.1 How to Determine the Wheel-Over Point . . . . . . . . . . . . 29
3.7.2 The Turning Radius of a Ship . . . . . . . . . . . . . . . . . . 31

4 Results 34
4.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

iii
 4.2 AIS Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

5 Discussion 40

6 Conclusion 42

Bibliography 48

iv
List of Figures

1.1 An example of AIS System Overview [1] . . . . . . . . . . . . . . . . 9

2.1 Application of AIS data in different areas [2] . . . . . . . . . . . . . . 15

3.1 The historical AIS dataset is enormous, as illustrated by a visualiza-

tion of around 200 million potential records. . . . . . . . . . . . . . . 21
3.2 On Marine Traffic website selection can be made according to vessel
segments and sub-segments [3] . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Basemap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.4 Location of Selected Vessels Journey . . . . . . . . . . . . . . . . . . 23
3.5 10-min detailed Vessel Journey and its location on Folium . . . . . . 24
3.6 Illustration of how the WOL, COT for a constant radius, turn can be
determined . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.7 Wheelhouse Poster [3] . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.8 Tuning circle of a ship/vessel . . . . . . . . . . . . . . . . . . . . . . . 33

4.1 800-1100 points taken from Viking Glory Journey present in AIS Data 35
4.2 Second Difference of Viking Glory Journey . . . . . . . . . . . . . . . 36
4.3 Fitted Circle when Viking Glory commences a turn . . . . . . . . . . 36
4.4 800-900 points taken from Viking Grace Journey present in AIS Data 37
4.5 Second Difference of Viking Grace Journey . . . . . . . . . . . . . . . 38
4.6 Fitted Circle when Viking Grace commences a turn . . . . . . . . . . 38
4.7 Finn Sky (Cargo Ship) Journey present in AIS Data . . . . . . . . . . 39
4.8 Second Difference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

v
List of Tables

1.1 Static Information in AIS . . . . . . . . . . . . . . . . . . . . . . . . 11

1.2 Dynamic Information in AIS . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Voyage related Information in AIS . . . . . . . . . . . . . . . . . . . . 12

2.1 Literature review summary . . . . . . . . . . . . . . . . . . . . . . . . 18

3.1 AIS data used throughout thesis . . . . . . . . . . . . . . . . . . . . . 25

vi
Acronyms

AIS Automatic identification system

IMO International Maritime Organization

GNSS Global Navigation Satellite System

VTS Vessel traffic services

SOLAS safety of life at sea

ECDIS Electronic Chart Display and Information Systems

SOG speed-over-ground

ROT rate-of-turn

OECD Organisation for Economic Co-operation and Development

PPO Proximal Policy Optimization

LIME Local Interpretable Model-Agnostic Explanations

LMTs Linear Model Trees

MMSI Maritime Mobile Service Identity

VHF Very High Frequency

GPS Global Positioning System

COG Course over ground

UTC Coordinated Universal Time

ETA Estimated Time Arrival

vii
OOW Officer on Watch

DBSCAN Density-Based Spatial Clustering of Applications with Noise

MSP Maritime Spatial Planning

TR Turning Radius

nm nautical mile

LOA Length Over-all

POI Point-of-overcoming the inertia

NAVSTAT Navigation Status

POSACC Position Accuracy

RAIM Receiver Autonomous Integrity Monitoring

COT Center of Turn

WOL Wheel Over Line

WOP Wheel Over Point

GRT Gross Register Tonnage

ARPA Applied Research Platform for Autonomous Systems

CSV Comma-Separated Values

viii
Abstract

The analysis of ship voyages using Automatic Identification System (AIS) data plays
a crucial role in enhancing our understanding of maritime behavior and improving
navigational safety. In this study, we focused on elucidating ship voyages by applying
the elastic trend filter to AIS data and analyzing turns using circle-fitting techniques.
Our primary objective was to gain insight into a ship’s navigational characteristics
and decision-making processes during turning maneuvers.

To achieve this, we identified turns in the ship’s trajectory, approximated their

radii by fitting circles to the data points, and extracted the underlying route using
the elastic trend filter. The consistent turn radii observed in the results indicated
the presence of a common navigational strategy. However, variations in turn radii
were also observed, suggesting instances of sharper turns made to avoid obstacles
or navigate through constrained spaces.

To assess the accuracy of our circle fitting method, we compared its precision to man-
ual measurements and previous investigations, demonstrating its reliability. These
findings have important implications for autonomous ship control systems, naviga-
tional regulations, and marine safety. Recognizing ship maneuvers and their associ-
ated radii can contribute to enhancing navigation procedures and optimizing route
planning.

By leveraging the analysis of AIS data, our study contributes to advancing trans-
parency and providing novel insights into the navigational behavior of ships. The
acquired knowledge holds significance in understanding and explaining ship voyages,
making a valuable contribution to this field of study.

ix
Chapter 1

Introduction

1.1 Overview
Shipping is one of the backbones of today’s globalized economy that is highly in-
terconnected. Over 70% of goods will be transported by sea by 2050, which is
expected to double the demand for freight transportation [4]. Passenger ships, fer-
ries, fishing vessels, and recreational boats also sail the world’s oceans in addition
to commercial transport ships. To ensure the continued operation of the increas-
ingly globalized market economy, as well as the health and safety of passengers and
marine ecosystems, it is necessary to ensure the safety and security of diverse mar-
itime traffic. Automatic identification system (AIS), designed by the International
Maritime Organization (IMO) in the 1990s to increase safety and security at sea, is
complementary to high-frequency radars [5]. A ship that is equipped with an AIS
transceiver broadcasts its position to nearby vessels and authorities on a periodic
basis based on the Global Navigation Satellite System (GNSS). On-shore Vessel
traffic services (VTS) rely on the AIS to guide and plan traffic as adjacent vessels
use positional data that aids the vessel in collision avoidance [6]. Ships that use AIS
transponders transmit regular information, such as their location, course, speed,
destination, and ship identifier, as required under the international convention for
safety of life at sea (SOLAS) [7, 8]. It is possible to collect this information over time
and analyze it in order to identify normal patterns of behavior [9]. The behavior of

1
a ship could indicate that it is being used for illicit activities if it exhibits anomalous
behavior. Tens of thousands of ships traveling between thousands of ports every day,
however, make it impossible to manually monitor the potential threat. Therefore, a
priority list for further investigation must be created using efficient and robust au-
tomatic data processing. In recent years, a variety of methods have been developed
to explain ship voyages and automatically detect anomalies in maritime AIS tracks
[10, 11, 12, 13, 14, 15]. Detecting anomalies within confined local areas is possible
through anomaly detectors developed in mentioned papers because vessel traffic is
to some extent predictable. The physical constraints of sea routes or mandatory
sailing routes also prevent vessels from acting arbitrarily. This makes such detectors
capable of detecting anomalies in vessel traffic patterns, like accidents or criminal
activities, and marking them as anomalies. Developed methods can also contain a
high number of false alarms, one of the common challenges in detecting anomalies
in a maritime environment.

Traditionally, before the era of Electronic Chart Display and Information Sys-
tems (ECDIS), route planning for ships was done with a ruler and compass. While a
voyage involves drawing lines on a chart, critical decisions have to be made through-
out. When navigating, a number of factors must be considered, such as weather
conditions, sea state, currents, traffic separation schemes, depths of water, and ma-
neuvering characteristics of the vessel.

Most of the studies have focused on using techniques to detect and restore the
anomaly AIS data based on route planning of a particular vessel, calculated from
machine learning models but several existing works ignored the changes in the ves-
sel’s direction in confined constraints due to the impact of size, speed or weather
conditions [16, 17, 18, 19]. To reduce the environmental impact and ensure the
efficiency, effectiveness, and safety of maritime transport, it is vital to understand
current maritime transport patterns, from how a single ship operates within a narrow
geographical context to how many ships operate within a wider geographical area.
Consequently, route plans can be decomposed into straight lines, circle sections, and
a few points of reference using simple rules and logic. This means reconstructing the
route plan and explaining the critical decision-making in an executed voyage might

2
be possible, given navigational AIS data. Taking into consideration the ship’s ability
to turn within a constrained space such as a narrow channel or harbor is something
that should not be overlooked, especially during route planning. Factors such as
a ship’s size, speed, and shape affect its maneuvering characteristics, including its
turning circle and stopping distance. It is imperative to understand these factors in
order to ensure safe and efficient navigation of ships in confined waters.

This study delves into the significance of understanding the underlying reasons be-
hind ship maneuvers, addressing a crucial gap in the current understanding of vessel
behavior. Even while data from the AIS is useful for learning about ship operations,
it largely focuses on what a ship is doing rather than the reasons for its particular
maneuvers. This research intends to uncover insights that can help to a more thor-
ough understanding of maritime operations by examining the reasons underlying
ship motions. Moreover, the development of automated methods for extracting voy-
ages and key maneuvers from AIS data has the potential to facilitate the creation
of meaningful datasets for researchers. When enormous amounts of AIS data are
generated during a journey, it is crucial to simply record the pertinent navigational
points in order to prevent needless data storage. The availability of stored ”typi-
cal voyages and key maneuver” data can be a useful resource for route planning in
educational contexts and developing hands-on training exercises for training simu-
lators, which is especially pertinent to institutions like Aboa Mare. Additionally,
having access to the ’normal’ maneuver patterns for a specific route makes it easier
to spot anomalous maneuvers, potentially improving Vessel Traffic Services (VTS)
capabilities.

For research purposes, AIS data is extracted from “The Finnish Transport Infras-
tructure Agency”. The data was collected using a web socket API on Google Cloud.
The thesis consists of four main chapters. Chapter 1 gives an introduction to the
general research area, previous research done in the same area, the problem state-
ment, the research aims and the scope. Moreover, it also gives an overview of AIS
data and its workings. Chapter 2 reviews the historical context and methods al-
ready in use from earlier studies to explain ship maneuvering. Chapter 3 gives a
review of the methodology used on AIS data. Moreover, it explains the research

3
framework which includes raw data acquisition, data wrangling and preprocessing,
feature analysis, dataset and the methods used to analyze the turning points in a
vessel’s journey. Chapter 4 shows the results obtained from methods implemented
on the AIS dataset. Chapter 5 discusses a more general and holistic picture of the
thesis and how it contributes toward the fulfillment of the research gap. Chapter
6 attempts to summarize the critical findings of the results and gives a holistic
overview of the nature of the research. Furthermore, it also explains the problems
and challenges that were faced during development and then provides suggestions
related to some future work in this field.

1.2 Problem Statement

According to the Organisation for Economic Co-operation and Development (OECD),
over 70-90% of the trade goods traded in the world are transported by ships [20].
As a result, maritime traffic and ship sizes are increasing. Because of high invest-
ment costs and geographical constraints, the infrastructure at ports and terminals
restrains the growth in the size of ships along waterways. Since both human and en-
vironmental impacts are involved in maritime logistics, the margin for error becomes
very small. An example can be taken by the grounding of the Ever-Given ship in
the Suez Canal in 2021 illustrating the severe consequences of human error in mar-
itime transport. The cause of maritime collisions has been evaluated statistically by
engineers using historical collision data [21, 15]. In each area over a given period of
time, there can only be a limited number of incidents collected on which statistical
analysis can be performed. With the help of land-based and satellite-based stations,
the AIS collects a large amount of maritime traffic data [22]. Insights into maritime
traffic behavior and route estimation can be gained from the existing AIS data, as
well as anomalous behavior can be detected.

Examining a voyage entails a thorough review of all the data points in the AIS
dataset that have been assigned to a specific ship. The research is guided by the
following questions: Based on the historical AIS data, how can maritime routing
patterns be discovered automatically and efficiently? To discover patterns and create

4
a network that represents maritime traffic, what algorithms might be applied? In
order to efficiently process huge amounts of maritime information, how can we design
and implement methods that can process huge amounts of maritime information?
By answering these questions, researchers will be able to derive valuable information
from the detailed analysis of voyage data, that helps advance maritime analytics,
optimize routes, and improve decision-making in a variety of areas, including vessel
traffic management, transportation planning, and maritime safety.

1.3 Research Aim

Explaining the voyage of a vessel can benefit a number of people in several ways.
Ship navigator officers plan the voyage manually which can be a time-consuming and
tiresome job, and sometimes human errors can be involved which can result in dis-
astrous events. With the help of my research, voyage planning can be autonomous,
and navigators who are unfamiliar with a sea area, and who don’t have information
about past experiences, would be able to plan a voyage easily due to the availability
of vessel routes taken in the past. This way they will be able to know what would
be the best practice in the considered area. In addition, avoiding technical disasters
can also help the global economy as a single canal blockage due to a vessel can cost
up to $9.6 billion of global trade [23].

I am trying to solve the problems explained in Section 1.2. Gaining a thorough

grasp of how ships react and function throughout various maneuvers is the goal of
ship maneuverability analysis. Researchers, naval architects, and maritime experts
can accomplish the following goals by studying ship maneuverability:

1. Safety Assessment: When evaluating the safety of ship operations, especially

during crucial maneuvers like turns, ship maneuverability is examined. Under-
standing a ship’s moving capabilities and limitations can help in identifying
potential dangers and risks, allowing for the formulation of safer navigational
plans and the application of appropriate safety precautions.

2. Port and Waterway Planning: Planners of ports and waterways must consider
ship maneuverability. Planning professionals may provide the best possible de-

5
 sign and layout for port infrastructure, including navigation channels, berths,
and turning basins, by examining how ships maneuver in constrained areas
like ports, canals, or locks. By doing this, congestion and delays are reduced,
and safe and effective vessel movements are guaranteed.

3. Collision Avoidance and Traffic Management: Systems for traffic management

and collision avoidance depend on an understanding of ship maneuverability.
Researchers can create advanced algorithms and models for assessing collision
risk, predicting trajectories, and optimizing traffic flow by studying ship be-
havior during maneuvers. This makes it possible to implement efficient traffic
management plans and create smart navigational systems to avoid collisions
and raise overall maritime traffic safety.

4. Environmental Impact Assessment: Assessing ship maneuverability aids in

determining how vessel operations impact the environment. Researchers can
assess the impacts of ship movements on sensitive coastal areas, marine ecosys-
tems, and air quality by observing turning behavior and maneuvering patterns.
The development of sustainable shipping practices, coastal zone management,
and environmental impact assessments all benefit from this information.

Overall, the goal of studying ship maneuverability is to increase marine safety, en-
hance port and waterway planning, create collision avoidance systems, and evaluate
how vessel operations affect the environment. Gaining knowledge of ship maneuver-
ability will help stakeholders in the marine sector make informed decisions, imple-
ment effective measures, and guarantee the efficient and safe movement of ships.

1.4 Research Scope

While the goal of this research is to offer insightful information on a ship’s movement
during turning maneuvers and to discover maritime routing patterns automatically
and efficiently, it is crucial to recognize the study’s inherent limitations. These
restrictions are caused by various factors, such as data availability, methodology
constraints, and contextual considerations. Contextualizing the results and ensur-
ing a fair interpretation of the research findings depends on being aware of these

6
limitations. When generalizing the findings and assessing the application of the find-
ings in realistic marine contexts, it is imperative to take the following constraints
into account.

1. Data Availability: AIS data quality and availability are crucial to the research
scope. In this study, the spatial and temporal limitations of the dataset used
may limit the analysis and findings. Moreover, the usage of specific dataset
can constrain the generalizability of results.

2. Turning Point Extraction Methods: In this research, turning points from AIS
data are extracted using trend-filtering techniques. Even though trend filtering
can be an effective technique, choosing the right algorithm and parameter
settings can ensure improved accuracy and robustness. However, this research
does not cover alternative methods for identifying turning points or comparing
multiple algorithms extensively.

3. Limited Contextual Information: Positioning and navigation information are

primarily provided by AIS data, but certain contextual details may influence
ship behavior during turns. In the analysis, factors such as vessel cargo, in-
tended route, and navigational conditions are not directly taken into account.
Rather than looking at comprehensive contextual data, the research scope
focuses on ship trajectory analysis.

4. Simplified Turning Analysis: The research mainly focuses on identifying turn-

ing points and turning radii to characterize ship-turning behavior. The anal-
ysis may not include all aspects of ship maneuvering during turns, such as
acceleration profiles or particular turning methods used by different vessel
types, even though these measures offer insightful information.

5. Generalizability: The dataset and vessels used in the study could have influ-
enced the research findings. Different vessel types, sizes, or operational cir-
cumstances might demonstrate different turning behaviors that aren’t entirely
covered by the defined scope. Only three ships’ behavior is taken into account
in this study. Further validation and the use of a more varied dataset could
be necessary in order to generalize the results to the entire marine industry.

7
In conclusion, the main contributions of this thesis are summarized as follows:

1. Generate a benchmark route plan based on standard navigational rules and

practices.

2. Develop methods for reconstructing the route plans from historical AIS data.

3. This paper proposes different trends on AIS data using Trend filtering methods
such as Elastic Filtering.

1.5 AIS Data

In this study, as a first step towards explaining voyages and calculating radius, we
will develop methods for reconstructing the route plans from historical AIS data.
Before developing, it is necessary to know the AIS data and its guidelines. Moreover,
it is also important to have knowledge about wheel over point and turning radius
of a ship because when planning a passage, it is imperative that a ship adjusts its
course correctly and follows the new intended course precisely.

1.5.1 What is AIS Data?

A short-range coastal tracking system known as the Automatic Identification System

(AIS) is used by ships today. It serves both vessels and shore stations by providing
identification and positioning information. Navigation safety has been improved
by AIS since radar was introduced [24]. The system operates in the Very High
Frequency (VHF) maritime band and provides digital positional awareness. The
system provides additional information to assist situational awareness and assists
with identifying ships, target tracking, search and rescue operations, and simplifying
information exchange [25].

With the AIS, vessels continuously send their identities, positions, speeds, and
courses, as well as other relevant information, to all AIS-equipped ships in their
range. In addition to managing maritime traffic and reducing marine navigation
hazards, the system works in conjunction with a shore station to enhance maritime
safety. Figure 1.1 different routes AIS data can be transmitted [1].

8
 Figure 1.1: An example of AIS System Overview [1]

1.5.2 Working of AIS

In an AIS unit, automatic location and movement data are obtained by using the
vessel’s Global Positioning System (GPS) system or an internal sensor. A combi-
nation of this information and programmable information from the AIS unit (such
as Maritime Mobile Service Identity (MMSI) number, the name of the vessel, the
destination, and the cargo type) is then transmitted regularly at regular intervals
while receiving AIS information from other vessels. There is the option of using an
antenna splitter instead of the antenna that transmits from the VHF radio for the
AIS unit. It is important to use an active antenna splitter that is suitable for AIS
and VHF radios.

SOLAS requirements: Under the IMO Convention for SOLAS [26], all ships with
a gross tonnage over 300GT engaged on international voyages as well as all passenger
ships, regardless of size, must be equipped with AIS.

There are two types of classes in AIS:

1. Class A: This requirement applies to all passenger ships and vessels with a
displacement of 300 GT or more engaged on international voyages.

2. Class B: Designed for vessels that are not SOLAS-compliant and provides

9
 limited functionality. A pleasure craft is the most common vessel to use this
type of propeller.

There are two dedicated VHF channels or frequencies that AIS operates on [27]:

1. AIS 1: Works on 161.975 MHz- Channel 87B (Simplex, for the ship to ship)

2. AIS 2: 162.025 MHz-Channel 88B (Duplex for the ship to shore)

There are four different types of data that AIS devices exchange and transmit, i.e.
static, dynamic, voyage-related, and short safety-related messages.

1. Static Information
The data presented in Table 1.1 is broadcast every 6 minutes and on request
by a competent authority.

2. Dynamic Information
The data presented in Table 1.2 depends on speed and course alteration. This
data, typically, can be seen on the chartplotter. The AIS transceiver sends the
dynamic information every 2 to 10 seconds depending on the vessel’s speed
while underway, and every 3 minutes while the vessel is at anchor.

3. Voyage-related information
The data presented in Table 1.3 is broadcasted every 6 minutes when data is
amended or on request.

4. Short safety-related messages

These messages are in the form of free-format text messages which are ad-
dressed to one or many destinations or all stations located in the area. These
messages can be related to content like iceberg sightings, buoy missing, etc.

10
Information Description
MMSI number Vessel’s unique ID
Company issues a unique seven-digit number to
IMO number each vessel and it can refer to the ship owner or
company.
Name: Name of the vessel, max 20 characters.
Name and Call Sign Call Sign: International radio call sign is assigned
to each vessel during the process of its registration
Dimensions of ship Length and Beam are measured in meters
Ship is categorized on the basis of the cargo it
Type of ship transports and its size. The type of ship can be
passenger, reserved, cargo, tanker, etc.
Location of position Location of positioning system’s (e.g. GPS) an-
fixing antenna tenna onboard the vessel

Table 1.1: Static Information in AIS

Information Description
Navigation Sta- It indicates the status of the vessel “at anchor”,
tus (NAVSTAT) “under way using engine” or “not under command”
It indicates the vessel’s position with respect to
rate-of-turn (ROT)
right or left, 0 to 720 degrees per minute.
It indicates the speed of the vessel with respect
speed-over-ground
to 0.1 knot resolution from 0 to 102 knots - vessel
(SOG)
moving relative to the surface of the Earth
Position Accu-
Ship’s position with accuracy indication
racy (POSACC)
It indicates the vessel’s location with respect to:
Geographical Coordi-
Longitude – to 1/10000 minute
nates
Latitude – to 1/10000 minute
Course over ground
Relative to true north to 0.1 degree
(COG)
True Heading 0 to 359 degrees calculated using gyro compass
It is measured in Coordinated Universal Time
Position Timestamp
(UTC)

Table 1.2: Dynamic Information in AIS

11
 Information Description
Ship’s draught 0.1 meter to 25.5 meters
Types can include dangerous goods, harmful sub-
Type of cargo
stances, marine pollutants, etc.
Destination and Destination – max 20 characters
Estimated Time Ar- ETA at destination – UTC month/date
rival (ETA) hour:minute
Route plan consists of waypoints and legs. A way-
Route plan (Way-
point is a single coordinate within a route, at which
points)
a vessel stops or changes its course.

Table 1.3: Voyage related Information in AIS

1.5.3 Limitation of AIS Data

As AIS is a relatively new technology (circa 2000), long-term records are rarely kept
because the transmitted messages require so much physical space. In spite of the
fact that AIS only represents received data, the greatest limitation is what is not
received. Following are some of the key points to which AIS data is limited.

• AIS information may not be accurate. It is possible for ships to display the
wrong destination. They might have forgotten to enter their final destination
after leaving a port. It’s possible that they overlooked updating their naviga-
tional status as well. Due to the fact that it depends on active input from a
human, it is far too simple for this ”voyage data” to be incorrect.

• The AIS information could be misinterpreted. It is now the responsibility of

the receiver to appropriately interpret the data, assuming that the information
being transmitted by another vessel is accurate.

• Officer on Watch (OOW) could become over-reliant on AIS. The information

on the screen can appear uncomfortably reliable when utilized on an ECDIS.

• AIS is not available on all smaller ships. AIS is only required on the following
vessel categories (depending on Gross Register Tonnage (GRT)), according to
SOLAS:

1. Cargo ships >300 GRT on international voyages

2. Cargo ships >500 GRT on all voyages

12
 3. All passenger ships

• OOW should be aware that AIS may be disabled on certain ships, thus negat-
ing any information they might receive from them if AIS is installed.

• OOW should not assume that the information received from other ships might
be inaccurate or not of the same precision as information available on their
vessel.

However, collision avoidance should not solely be dependent on AIS information. A

vessel’s AIS system is merely a source of additional information for the OOW, and it
only serves to assist in navigating the vessel. Human expertise on bridges can never
be replaced by AIS and radar is the most reliable way to identify moving vessels..
In summary, AIS only contributes to better navigation safety through its assistance
to OOW/VTS. The installation of a standalone AIS system is just a matter of
plugging in a few cables and turning it on, since AIS is typically integrated with
ship bridge systems or multi-functional displays.

13
Chapter 2

Literature Review

In the past decades, AIS data has been extensively utilized in maritime research.
Since AIS is based on VHF signals, which have a limited range, rather than satel-
lites, most research has focused on narrow and congested waterways. It has been
identified by vessel operators that route planning sometimes involves long journeys
which can cause inappropriate cost estimation [11]. As a result, it is considered a
bad voyage plan. In [28], a data-driven methodology is proposed that can develop
an optimized route considering the ship’s dimensions, weather, and load conditions.
This method divides the voyage into open and local sea passages. On the basis of
speed, these passages are simplified to pattern nodes. After this process, two types
of algorithms are applied to the dataset i.e. K-Means algorithm for the classifi-
cation of pattern nodes on the basis of routes in the open sea and Density-Based
Spatial Clustering of Applications with Noise (DBSCAN) algorithm for the identi-
fication of connection points on the basis of routes in the local sea. It is possible to
determine the most navigated routes between two ports through a combination of
representative routes in open sea passages as well as link points in local sea passages.
Additionally, research based on the usage of AIS data to produce adequate infor-
mation for Maritime Spatial Planning (MSP) is currently being conducted at the
European level [29]. Integrating AIS with MSP offers a promising perspective for
European Marine Strategy Framework although AIS data was originally applied to
navigation safety; as data accessibility improved, it was used for diverse purposes.

14
These purposes include vessel traffic management, fisheries management, and in-
formed decision-making. The application of AIS data can be divided into three
stages, i.e., basic, extended, and advanced [2]. Approximately two to three applica-
tion fields are present in each stage, totaling seven (refer Figure 2.1). Furthermore, a
most recent application that has become famous, is whale-watching operations. To
estimate whale-watching efforts in areas with cetacean populations, big data analy-
sis based on AIS messages can be useful [30]. It is recommended that the responsible
authorities promote the installation of AIS transponders at least on all vessels au-
thorized to conduct whale-watching activities since the proposed methodology relies
heavily on the number of vessels equipped with AIS transponders.

Figure 2.1: Application of AIS data in different areas [2]

The system that learns the normal behavior of vessels, detects anomalies and pre-
dicts the motion using an artificial neural network trained with AIS data is explained
in [31]. The researchers have taken a few steps to analyze data which includes per-
forming data mining to extract motion patterns and then defining those motion
patterns. To simplify things, motion pattern is defined by kinematic and attribute
information with only one variable i.e., its origin. The kinematic information will in-
clude the ship’s location and velocity, both of them in two dimensions. The training
data used for anomaly detection is partitioned into two regions, one region corre-
sponding to the hypothesis (normal behavior) and the other to the anomaly. The

15
anomaly detector is proposed for vessel motion which is based on adaptive kernel
density estimation.

A novel approach to detect anomalies in AIS data based on the ship’s position
is introduced in [32]. To identify anomalies in AIS data accurately, parameters
like longitude, latitude, and speed have been taken into account [32]. In order to
satisfy the subsequent research and application based on trajectory, it is necessary
to delete the abnormal points in the raw data. The proposed method classifies
AIS abnormal points and processes them separately according to the longitude and
latitude, speed, acceleration, and direction information in AIS data. It is worth
noting that the proposed method only needs the AIS data of the ship itself and
does not need the support of the historical track data. In addition, the cubic spline
interpolation method is used to repair the trajectory after eliminating the abnormal
points, which further improves the continuity and integrity of the trajectory. The
results of processing actual ship trajectories show that the method proposed in this
paper can identify all kinds of trajectory abnormal points in AIS data effectively.

Another research paper gives an insight into the process going on behind black box
predictors by reasoning the solution to the automatic docking problem [33]. This
problem includes guiding the vessel from the open sea towards a particular des-
ignated point in a harbor area by taking into account the environmental factors,
non-linear motions, speed limit, and distance to other ships and obstacles. Re-
searchers have explained that in trajectory and collision avoidance, deep reinforce-
ment learning agents perform well. The agent was trained using Proximal Policy
Optimization (PPO) and Local Interpretable Model-Agnostic Explanations (LIME)
is applied to each data point once for every action. LIME is used to make approx-
imations related to the behavior of the predictor for a single prediction. Another
model Kernal SHAP is used, which is built on LIME framework and provides the
functionality of removal of features by sampling the background data and replac-
ing the missing features with random samples because setting the value to 0 is not
helpful. Moreover, Linear Model Trees (LMTs) were also applied to the dataset and
later advantages and disadvantages were calculated for each algorithm.

With the help of an unsupervised approach that can work on real-time systems,

16
[34] explain a framework that can learn AIS maritime traffic patterns automatically.
In addition to data collected by terrestrial AIS receiver networks, the proposed
methodology is also effective in situations where information is highly disrupted due
to spatial gaps in coverage or significant temporal differences due to satellite revisits.
By using Bayesian inference, the approach could be used to detect low-likelihood
behavior in real-time.

An overview of the AIS system is presented in [35], along with an assessment of data
quality and vulnerabilities for decision-making in maritime situational awareness
scenarios. In addition to improving safety and security at sea, AIS data quality
assessment would help detect AIS problems earlier. This would increase decision-
makers situation awareness, and improve AIS data quality assessment.

17
# Author Method Description Pros Limitation
This study uses adaptive kernel Successful prediction of
density estimation to build motion of vessel using Limited to the historic
1 B. Ristic
anomaly detector for vessel Gaussian sum tracking data analyzation
motion. filter
Construction of anomaly It is a possibility
detector on the basis of Abnormal data can that the detector
2 Shuguang Chen
different variables like be easily detected. only works on
speed, distance etc. several ships.
Critical analysis is done on The detection method
automatic docking problem Easy navigation guide of ship behavior
3 H. Rong
by considering the processes for vessels is only limited to
happen behind black box. one factor i.e. distance.
Unsupervised learning of
The tool supports route
AIS maritime traffic The method utilizes
planning, surveillance
4 Michele Vespe patterns automatically only AIS historic or
technology and vessel
using Bayesian inference real-time data.
position prediction.
method.
Risks model is not
Increased situational
defined in-case if the
Data quality assessment awareness of decision
assessment fails on
5 C. Iphar on the basis of false and makers and improved
some data, that model
spoofed messages. maritime safety and
can be followed to
security.
avoid further damages.
Data-driven methodology is
Reliable routes can be
introduced to identify most Solely relying on AIS
provided to support
6 Jie Cai navigated routes between two data can make predicted
decision-making at
routes using K-means and methods less reliable.
pre-fixture stage.
DBSCAN algorithms.
Incorporating AIS data
AIS data processing and Requires specialized
into maritime spatial
analysis can be used to knowledge of hardware
7 Le Tixerant M. information systems can
produce adequate maritime and software to handle
provide a clear way to
traffic density information. high volume of raw data.
facilitate dialogue in MSP
Future research
A review on usage of AIS applications are explained No novel approach
8 Dong Yang
data and its applications. that can be implemented is introduced.
on AIS data.
Sufficient AIS terrestrial
Methodology can be used stations need to be
Analyzing AIS data
to estimate seasonal installed for achieving
9 Dong Yang to make whale-watching
and annual trends in complete coverage and
operations more sustainable.
whale-watching effort. maximum reception of
broadcasted messages.

Table 2.1: Literature review summary

18
Chapter 3

Methodology

3.1 Overview
The approach used in this research to examine ship maneuverability during turn-
ing maneuvers is described in this chapter. To get insights into ship behavior, the
methodology includes data collecting, preprocessing, turning point extraction, turn-
ing analysis, and statistical tools. To analyze turning points in a vessel journey,
an analytical approach using elastic trend filters [36] is proposed. This approach is
particularly aimed at constructing a system that reflects vessel traffic. Additionally,
this approach aims to improve the practice of maritime voyage planning, which is
typically done manually by a ship’s navigation officer, because there are no methods
available for detecting critical maritime waypoints efficiently based on the analysis
of large amounts of historical data.

The purpose of this chapter is to provide an overview of how to prepare data for ap-
plying an Elastic Trend Filter. The chapter is divided into four main sections: raw
data acquisition, data wrangling and preprocessing, feature analysis, and dataset
and evaluation metrics. In section 3.2, the process of collecting and obtaining data
from various sources is discussed. It is crucial that the data collected in this stage is
of high quality in order for the algorithm to be accurate and useful. In section 3.3,
the process of cleaning, transforming, and organizing the raw data is presented.
Trend Filtering applications require high-quality data, which is a time-consuming

19
and complex process. The accuracy and reliability of the data are ensured through
techniques such as data imputation, normalization, and outlier detection. In sec-
tion 3.4, the most relevant variables or features are taken into account and it also
discusses the feature selection technique used for this study. In section 3.5, infor-
mation related to the dataset that has been used in this study is described. In
section 3.6, the methods used on the data to explain the vessel’s journey are ex-
plained in addition to how they are applied and why they are chosen for analyzing
ship’s turns.

3.2 Raw Data Acquisition

The dataset is provided by the Novia University of Applied Science in collaboration
with Aboa Mare and Turku University of Applied Sciences (Turku AMK) which
I have used throughout my thesis. The data is extracted and saved every minute
and data in this study is based on historical AIS records spanning from October
2022 to December 2022, which included more than a hundred thousand positional
records. The data is downloaded in Comma-Separated Values (CSV) format using
API developed by ARPA project by Novia and Turku AMK. In Original Data,
there are 12 variables. Those variables are MMSI, Longitude, Latitude, SOG, COG,
NAVSTAT, ROT, POSACC, Raim, Heading, Timestamp, Timestamp external. To
illustrate the vastness and coverage of the historical AIS dataset, Figure 3.1 shows
200 million AIS positional records. Analyzing vessel journeys requires taking into
account the time and feature variables. An important consideration when analyzing
multiple vessels’ journeys is their time frame. It can also help identify areas of
congestion and other issues due to patterns and trends in vessel movement. It is
also possible to gain valuable insights into a vessel’s journey by analyzing its specific
features in addition to its time variables. Under certain conditions or in certain areas,
a vessel’s size, speed, maneuverability, and draft can all have an impact on its ability
to navigate. Focusing on a specific vessel’s journey during a particular time frame
can help to provide a more detailed and focused analysis. Regarding the journey
of the ship ‘M/S Viking Glory’, focusing on its movements during the months of
October to December can provide a more focused analysis of its journey. This time

20
frame may be significant for several reasons, such as the weather conditions during
this period or specific events or activities in the area.

Figure 3.2 shows the real-time vessel tracking information. With the help of this
website, through AIS data, it is possible to monitor the location of ships, boats, and
other vessels at any time and anywhere in real-time. To explain vessel voyages, this
thesis incorporates vessel segmentation as part of its proposed approach. Another
important factor of this website is that it provides users with real-time vessel traffic
so that the fastest and safest maritime routes can be identified. This has helped me
analyze voyages critically. Blue markers in Figure 3.2 indicate ship location in real-
time, by clicking them further information related to the ship is shown, for example
vessel information, past track, route forecast, navigational status, speed/course,
draught, etc.

Figure 3.1: The historical AIS dataset is enormous, as illustrated by a visualization

of around 200 million potential records.

21
Figure 3.2: On Marine Traffic website selection can be made according to vessel
segments and sub-segments [3]

3.3 Data Wrangling and Preprocessing

First of all, duplicates and missing values are removed from the data so that suitable
results can be extracted. After that, three ships are selected for the analysis i.e.
FINN Sky (Cargo ship), Viking Grace, and Viking Glory. To obtain the data of the
related ships, the data is filtered out on the basis of their MMSI numbers. When this
step is done, normalization and feature engineering are applied to the dataset. To
analyze the position of the vessel and the rate at which it takes the turn, longitude,
and latitude variables are taken into account.

3.4 Feature Analysis

Figure 3.3 shows the position of three ships of interest using the longitude and
latitude variables obtained from AIS data, i.e. longitude and latitude. The longitude
and latitude variables are used to identify the location of ships and to visualize ship
routes and patterns. In Figure 3.4, each ship’s position is represented by a green
marker point on the map, and the lines connecting these points show the path taken
by each ship. Figure 3.5 shows a closer look on a 10min journey of a vessel and

22
pop-up indicates the MMSI belonging to that vessel. A detailed analysis of AIS
data can provide useful information about ship behavior, such as speed, course, and
destination. Using this data, shipping routes can be optimized, maritime safety and
security can be improved, and the environmental impact can be reduced.

Figure 3.3: Basemap

Figure 3.4: Location of Selected Vessels Journey

23
 Figure 3.5: 10-min detailed Vessel Journey and its location on Folium

3.5 Dataset
A ship’s position, speed, and other information can be shared with other ships and
shore stations through the Automatic Identification System (AIS). Ground stations
collect AIS data and is available for companies who can pay for the data. Tracking
the movement and behavior of ships can be done by analyzing this data. The dataset
is provided by the Novia University of Applied Science, located in Turku and Vaasa,
Finland. The university acquires AIS data from the nearby sea areas where ships are
regularly passing through. The original size of the data is 84MB and it comprises of
12 variables: mmsi, longitude, latitude, sog, cog, navstat, rot, posacc, raim, heading,
timestamp, timestamp external. Table 3.1 describes the above-mentioned variables
in AIS data with respect to its definitions.

24
 Variables Description
It is a unique 9-digit identification number that is
MMSI assigned to every vessel by the IMO. It is used for
identifying and tracking ships through AIS.
It’s a geographic coordinate that specifies the po-
Longitude sition of a ship on the Earth’s surface. It refers to
the east-west position.
It’s a geographic coordinate that specifies the po-
Latitude sition of a ship on the Earth’s surface. It refers to
the north-south position.
It is the speed at which the ship is moving relative
SOG
to the Earth’s surface.
It is the direction in which the ship is moving rel-
COG
ative to the Earth’s surface.
It is the status of the ship, such as ”underway using
NAVSTAT
engine”, ”at anchor”, ”moored”, etc.
ROT It is the rate at which the ship is turning.
It is the accuracy of the position reported by the
POSACC
ship’s GPS system.
Receiver Autonomous
It is a system used to ensure the integrity of the
Integrity Monitoring
GPS signal.
(RAIM)
Heading It is the direction in which the ship is pointing.
It is the time at which the position data was
Time stamp
recorded.
It is the time at which the AIS signal was received
Time stamp external
by a ground station.

Table 3.1: AIS data used throughout thesis

3.6 Data Analysis Methods

To decompose voyage data into simple, explainable segments, we propose the fol-
lowing two approaches:

1. Defining a ship navigational model based on information available in funda-

mental navigational rules.

2. Represent voyages in a small latent feature space: Represent the route in terms
of maneuvers (turn rudder, increase engine power etc.)

Motivated by the fact that traditional navigation consists of decomposing the route
into straight lines and circles segments, we propose to formulate the problem as a

25
piecewise regression problem where the goal is to decompose the realized path in
terms of interconnected circle arcs and straight lines.

The decomposition of a route into piecewise linear and arc segments can be seen as
a trend-filtering problem. Linear Detection of non-stationary components such as
outliers, level shifts, and measurement trends have been studied extensively in the
time-series analysis literature [37]. Trend estimation has been applied in a variety
of different fields, including macroeconomics[38], finance [39], and in biological and
medical sciences [40, 41]. Furthermore, in system identification, trend filtering has
been used to remove trends, outliers, level shifts, and slopes that would otherwise
deteriorate the identification accuracy [42, 43].

Several trend estimation techniques have been studied in the literature, including
Hodrick-Prescott (H-P) filtering [38], ℓ1 trend filtering [44], empirical mode de-
composition [16], smoothing splines [45], moving average filtering [46] and linear
programming with a fixed number of kink points [17]. However, if an underlying
dynamical system has generated the measured time-series, standard trend filtering
methods may produce poor results since they do not distinguish trends in data from
effects of the system dynamics or input signals [43]. Thus, if the data depend on a
known input signal, the input and dynamics of the system should be accounted for
when estimating the trends.

In signal processing, elastic trend filtering is used to extract smooth and sparse
trends from noisy data. In short, elastic trend filtering approximates a noisy signal
x with x̂ by solving the convex optimization problem [36]

minimize ∥x − x̂∥22 + λ1 ∥D2 x̂∥1 + λ2 ∥D2 x̂∥22 ,

x̂

where λ1 , λ2 > 0 are scalar parameters tuned to determine the sparseness and
smoothness of the estimate x̂, and D2 x̂ denotes the second difference of x̂. The
elastic trend filtering is a generalization of the H-P trend filter (when λ1 = 0) and
the ℓ1 trend filter (when λ2 = 0).

In applications where we want to remove high-frequency noise from a signal whilst

preserving its low-frequency features, trend filtering is particularly useful. In this

26
research, AIS data is analyzed using elastic trend filtering to extract the underlying
trajectory of a vessel from noisy position data. There are many sources of noise in
AIS data, including measurement errors, missing data, and other sources that can
obscure the true path of a vessel. These issues are addressed by elastic trend filtering,
which smooths the data while preserving the essential characteristics of the vessel’s
trajectory. A basic principle of elastic trend filtering is to minimize a cost function
based on both squared differences between the filtered data and the original data
and second-order differences between the filtered data (i.e., the curvature). As long
as these two terms are balanced, elastic trend filtering can extract the underlying
trends of a signal while avoiding overfitting.

It is important to know that there are many different algorithms and implemen-
tations of elastic trend filtering, each with its own strengths and weaknesses [47].
There are several popular approaches, such as variation regularization or basis pur-
suit denoising. A preprocessing step is typically required to remove missing values,
outliers, and other sources of noise from AIS data before applying elastic trend fil-
tering. Using this filtered trajectory, one can estimate the vessel’s speed, course, and
turn rate, as well as predict its future trajectory. Three ships are chosen to apply
an elastic trend filter so turning points in the whole ship journey can be observed.
After getting smooth trends by applying a trend filter to the vessel’s trajectory, the
second difference of that trajectory is taken. The second difference of the position
refers to the acceleration. The rate of change (first difference) of a position x(k), at
time instance k, is

∆x(k) = x(k) − x(k − 1) (3.1)

and the second difference is

∆2 x(k) = ∆x(k) − ∆x(k − 1) = x(k) − 2x(k − 1) + x(k − 2). (3.2)

Significant changes in acceleration often accompany turning points or changes in

the ship’s movement, which is what explains the connection between the second
difference and turning points. By analyzing a ship’s second difference in position,

27
we can detect when it changes direction and experiences a change in acceleration. It
is possible to identify peaks or valleys in the ship’s acceleration data by examining
the second difference over time. A peak or valley indicates a significant change in
acceleration rate, thus indicating a potential turning point and zero acceleration
means that the ship is moving with constant velocity. A large positive/negative
peak or valley may represent a sharp turn or maneuver, whereas a smaller peak or
valley may indicate a smooth change. Analyzing the second difference of a ship’s
acceleration data can, therefore, guide the detection and characterization of turning
points, providing insights into the ship’s navigation and movement patterns.

The radius of turning points in a vessel’s journey is calculated using Circular arcs.
The radius of the turn can be estimated by fitting a circular arc to a portion of the
vessel’s trajectory. For estimation, the vessel’s trajectory is analyzed in segments
in order to find the turning points. It is important that these segments contain
the relevant data prior to, during, and after the turn. Essentially, this involves
measuring how far the vessel’s trajectory deviates from a circular path. To calculate
the radius, a circular arc is fitted to each turning point segment. This is done by
using the Equation 3.3 of a circle with radius r centered at (x0 , y0 )

r2 = (x − x0 )2 + (y − y0 )2 (3.3)

or equivalently in parametric form

x = x0 + rcos(θ) (3.4)

y = y0 + rsin(θ) (3.5)

An accurate representation of a vessel’s trajectory during a given turn is represented

by its radius. Calculating the radius of the vessel’s turns and fitting circle arcs
provide insight into its maneuverability and navigational behavior.

28
3.7 Wheel-Over Point and Turning Radius
Route planning is influenced by a ship’s ability to turn tightly after altering the
course and stay on the new intended course. Due to the inertia of the ship, it
typically takes some time for a ship to react when the wheel is turned. Therefore,
the wheel must be spun before the turning point of the ship. It is often referred
to as “the wheel over the position”. A ship must initiate a turn at the wheel-over
position in order to arrive safely on the new track.

Turning around at one point is impossible for large vessels. Due to this, it is impor-
tant not to overlook the ship’s ability to turn within constrained spaces. Making a
turn requires a vessel to know its wheel-over position.

3.7.1 How to Determine the Wheel-Over Point

In order to come onto the desired new track safely, a ship needs to commence a
turn at the Wheel Over Point (WOP). It becomes critical to determine the wheel-
over point if the cross-track error is minimal in narrow channels or confined waters.
Ship size, water depth, and speed all factor into determining the wheel-over point.
Other factors include the deadweight and speed of the ship. The most significant
component is likely the ship’s speed, which dictates how much inertia must be
overcome before the ship starts to turn. Following are the examples that should be
considered when determining the position of the wheel:

• Current course: 90-degree turn - It means that the ship is traveling in a

direction that is due east.

• Next course: 45-degree turn - It means the ship will make a turn to the right
and travel in a direction that is northeast considering the current course of
the ship 90 degrees True.

• Speed Over the Ground (SOG): 12 knots - It means the ship is traveling at a
speed of 12 nautical miles per hour relative to the surface of the Earth.

• Turning Radius (TR) (TR = SOG/ROT) = 1.0 nm - TR of a vessel is a

measure of the minimum radius of the circle that it can turn while maintaining

29
 a constant speed and ROT. If the TR is calculated as TR = SOG/ROT and
is equal to 1.0 nautical mile (nm), then it means that the vessel requires a
minimum distance of 1.0 nm to complete a turn of 360 degrees.

• Length Over-all (LOA): 235 meters - LOA describes the maximum length of
a vessel, including any protrusions or extensions. This measurement is taken
from the tip of the bow to the end of the stern and includes any extensions
such as the anchor, bowsprit, or bulbous bow. Knowing the LOA of a ves-
sel is important for a variety of reasons, including determining the vessel’s
docking requirements, cargo capacity, and ability to navigate through certain
waterways.

• Point-of-overcoming the inertia (POI): 1.5 x LOA = 352.5 meters also known
as the execution point wherein the ship starts to turn. (Such info is posted in
the wheelhouse poster [3] (see Figure 3.7 for reference)).

• Wheel Over Line (Wheel Over Line (WOL)): It refers to a navigational ma-
neuver carried out by a ship. In order to bring the ship’s heading or course
precisely over a given line or track, the ship’s wheel (helm) must be turned.
The ship can securely sail along planned paths or through limited spaces by
performing a WOL maneuver. It aids in keeping the ship on course and en-
ables it to navigate around any dangers or obstacles that may be present on
either side of the desired line.

Calculating the trajectory for a constant-radius turn consists of determining the

Center of Turn (COT) point and the WOL and WOP illustrated in Figure 3.6. The
resulting theoretical ship trajectory consists of piece-wise linear trajectories and
circle segments when the ship turns.

30
 WOL
Final track
COT
r
u
Initial track
WOP

Figure 3.6: Illustration of how the WOL, COT for a constant radius, turn can be
determined

Figure 3.7: Wheelhouse Poster [3]

3.7.2 The Turning Radius of a Ship

A vessel’s pivot point follows roughly a circular path when the ship is turning with
a constant rudder angle. A ship’s turning radius is determined by the radius of this
circle. As the ship is moving, it is affected by the depth of the water, and by the
length of the ship. The general rule is that the turning circle of a long ship will be
larger. This rule means that a longer ship will typically need a larger area to make
its full turn. A ship’s final diameter is determined by the type of rudder and the

31
steering effect it produces, with the distance between the rudder and the hull also
playing a significant role. It is more efficient for the rudder to turn a boat when the
clearance between the rudder and hull is smaller. As a result, the turning circle and
turning radius will be affected by the following factors:

• Structural design and length of the ship.

• Draught and trim of the ship.

• Size and motive power of the main machinery.

• Distribution and stowage of cargo.

• Even keel or carrying a list.

• Position of turning in relation to the available depth of water.

• Amount of rudder angle required to complete the turn.

When examining a ship’s maneuverability and how the ship handles different sit-
uations, it is important to consider both wheel-over-point and turning radius. It
is difficult for a ship with a large turning radius to maneuver in tight spaces, for
instance. Alternatively, ships with a small turning radius may be more agile in
rough weather, but less stable. Watchkeepers can know at each waypoint the pa-
rameters they must maintain to correctly negotiate a course change by calculating
and drawing wheel-over-points on charts. The OOW, for example, will know the
correct parameters to use for negotiating the turn because he/she knows the ROT
equals the ship’s speed divided by its radius of turn.

Taking into consideration the ship’s ability to turn within a constrained space is
something that should not be overlooked. Traditionally, route planning was done
with a ruler and compass pen. Given a fixed speed-over-ground u (SOG) and rate-
of-turn θ̇ (ROT), the turning radius can easily be calculated. For a fixed u, the time
t it takes to traverse a full revolution of a circle of radius r is t = 2πr/u. Likewise,
for a fixed ROT, the time it takes to traverse a full revolution is t = 2π/θ̇ (assuming
that θ̇ is given in rad/s). Combining the expressions gives the turning radius

u
r= ,
θ̇

32
cf. Figure 3.8.

A ship must be maintained on the planned course during the course of a voyage in
order to avoid a hazard or danger, especially when it travels through traffic lane
separations, passes shorelines, or passes rocks. Because of the limited availability
of sea room, cross-track distance should be kept to a minimum or zero. Using the
ship’s turning radius in conjunction with the wheel over position is a technique
that ensures the ship stays on track during and after a course alteration in order
to minimize cross-track distances and to keep the vessel safe from potential hazards
while making course changes.

u
r
θ

Circumference C = 2πr

Figure 3.8: Tuning circle of a ship/vessel

33
Chapter 4

Results

4.1 Overview
The examination of turns in the ship’s course was made easier by applying the elastic
trend filter to the AIS data. We learned more about the ship’s navigational qualities
and behavior by fitting circles to these turns. The findings from the analysis of turns
using the circle-fitting approach are presented in this section, along with a discussion
of their consequences.

4.2 AIS Data Analysis

In order to improve operations and ensure maritime safety, AIS data analysis plays
a crucial role. As the Viking Glory ship navigated through the water, AIS data
provided valuable insights into its navigation patterns. Applying an elastic trend
filter to the data is one innovative method of analysis. AIS data is smoothed by
using this filter, removing noise and irregularities to reveal the underlying trend. A
more accurate representation of the ship’s trajectory and movement is possible by
applying this filter. The elastic trend filter is applied to all three ships separately
and every ship’s journey is analyzed individually. Figure 4.1 shows the original
vessel trajectory in addition to the trend filter applied to the dataset. Figure 4.2
visualizes the second difference of the trajectory highlighting the positive peaks

34
and sharp edges. A turning maneuver is the next step after the AIS data has
been processed using the elastic trend filter. Calculating the radius of a turn is
commonly done by using the circle equation method. As a result, the ship’s path
during a turn is approximated by a circular arc derived from this mathematical
method. On the right of Figure 4.1, the red markers highlight the turning points in
a vessel’s journey, and these turning points are the main focus of the ship’s journey.
Using the circle equation, it is necessary to make certain assumptions in order to
calculate the turn radius. The ship’s speed during the turning maneuver is often
assumed to remain constant throughout the maneuver. The filtered AIS data points
corresponding to the turn are fitted to a circle based on this assumption (refer to
Figure 4.3). The calculated radius value is 0.9977. It is important to note that
the accuracy of calculating the radius when the ship takes a turn is dependent upon
many factors, including the quality of the AIS data, the data point’s time resolution,
and the assumptions used. While this approach offers useful estimates, it can also be
applied to analyze a ship’s navigation performance and understand its maneuvering
characteristics.

Figure 4.1: 800-1100 points taken from Viking Glory Journey present in AIS Data

35
 Figure 4.2: Second Difference of Viking Glory Journey

Figure 4.3: Fitted Circle when Viking Glory commences a turn

A similar approach is applied to two other ships in addition to the Viking Glory
vessel using an elastic trend filter and the circle equation to calculate the radius when
the ship commences the turn. Figure 4.4 shows the Viking Grace ship’s trajectory
along with the smoothing of the data, which is achieved using an elastic trend filter.

36
On the right side of Figure 4.4, red markers represent the turning points in the
ship’s journey. It is easier to notice the sharp turns in the ship’s journey by taking
the second difference of the trend-filtered data. Figure 4.5 shows large positive
peaks which means a sharp maneuver. Moreover, it also indicates that the change
in acceleration rate is significant. Now the radius of the turn can be calculated
using Equation 3.3, and the calculated value of radius is 0.0392 (refer to Figure 4.6).
Figure 4.7 shows the journey of a cargo ship i.e. Finn Sky Ship along with the
filtered data that helps smoothen out the journey. In this journey, there are no
such sharp maneuvers that could be considered to fit the circular arc. Figure 4.8
shows that the turn made in the journey is a rather smooth change indicating minor
maneuvers.

Figure 4.4: 800-900 points taken from Viking Grace Journey present in AIS Data

37
 Figure 4.5: Second Difference of Viking Grace Journey

Figure 4.6: Fitted Circle when Viking Grace commences a turn

38
Figure 4.7: Finn Sky (Cargo Ship) Journey present in AIS Data

Figure 4.8: Second Difference

39
Chapter 5

Discussion

This thesis investigates AIS data’s use in analyzing and explaining ship voyages. It
has been possible to gain valuable insights into ship navigational behavior, including
turning maneuvers, by applying techniques such as elastic trend filtering and circle
equation calculations. This chapter discusses the study’s findings, compares them
with previous studies, and outlines its advantages and disadvantages.

As a result of applying the elastic trend filter to different ships in AIS data, the
underlying trajectory of the ships were extracted, and turning maneuvers were de-
tected. It provided crucial information about the vessel’s navigational characteris-
tics when the turn radius during these maneuvers was approximated using the circle
equation method. The findings of my investigation provided crucial insights into the
ship’s turning behavior. It is observed that radii calculated during different maneu-
vers are consistent, for example, indicating a systematic approach to navigation by
the crew. Furthermore, the study identified instances of steeper turns, which could
be suggestive of avoiding obstacles or maneuvering in restricted regions.

I also compared my findings to other studies in the field of explainable ship journeys
utilizing AIS data, I saw some commonalities as well as some unexpected findings.
[18] examined vessel movement patterns in a similar manner, but utilized a different
process, such as graph evolution analysis. This study focuses on topological voyage
graph features to asses stationary behavior. Study [19], on the other hand, looked
at ship voyages from a broader perspective, encompassing not only turning maneu-

40
vers but also the arrival and departure of ships from a Korean port. Their findings
explained pattern analysis for speed over ground, course over ground, and ship po-
sition, which my study did not explore. As a result, combining the findings of our
study with those of study [19] may provide a more complete knowledge of ship voy-
ages and navigational decision-making. One of my approach’s primary advantages
is its simplicity and ease of implementation. The elastic trend filter and circle equa-
tion method are simple and efficient in computation, allowing for rapid AIS data
processing. Furthermore, the results of our research provide useful insights into
ship journeys, assisting in the study of navigational behavior and probable decision-
making aspects. However, it is critical to recognize the limitations of my approach.
To begin, the accuracy of the estimated turn radii is dependent on the assumptions
established, such as maintaining constant speed during maneuvers. Deviations from
these assumptions may cause radii to be calculated incorrectly. Furthermore, the
resolution of the AIS data and any data gaps may have an impact on the precision of
the results. Finally, our study was limited to turning maneuvers; a more thorough
examination that included other characteristics of ship journeys could provide more
comprehensive insight.

Several future directions can be identified based on the results and restrictions of this
study. The accuracy of the results could be improved by using additional filtering
techniques, such as the elastic trend filter, with varying settings or investigating
other techniques like Kalman filtering. Second, including additional variables in
the analysis, like speed, weather, and traffic volume, might give a more complete
picture of the factors affecting ship journeys. Finally, utilizing machine learning and
artificial intelligence approaches for anomaly detection and predictive analysis may
lead to new insights into and explanations for ship voyages based on AIS data.

41
Chapter 6

Conclusion

In this study, we primarily focused on the analysis of turns through the fitting of
circles and investigated the idea of explainable ship journeys using AIS data. We
were able to derive the ship’s underlying trajectory and locate turning maneuvers
by using the elastic trend filter on the AIS data. We calculated the radii of these
turns using the circle fitting approach, which gave us important information about
how ships navigate and make decisions. My research has shown how important it
is to comprehend ship turns and their radii in the context of maritime safety and
navigational techniques. A standardized approach to navigation is suggested by
the similar turn radii seen throughout numerous maneuvers, showing the existence
of navigational standards and patterns. Insights on specific navigational issues en-
countered by ships, such as obstacle avoidance or maneuvering in crowded places,
are provided by differences in turn radii, such as sharper turns with smaller radius.
Circle fitting method’s accuracy and dependability have been confirmed by compar-
isons with traditional measurements and previous investigations. This strengthens
reliability when using circle fitting techniques and AIS data to analyze ship voyages
and comprehend navigational behavior.

This study’s implications go beyond the confines of academia. The creation of

navigational rules can benefit from knowledge about constant turn radii, making
maritime operations safer and more effective. Maritime authorities can use this
data to recognize deviations from established standards and implement targeted

42
measures to reduce risks and enhance security during navigation. The knowledge
gathered from the analysis of turns can also be used to help create autonomous
ship control systems. The agility of autonomous or semi-autonomous vessels can be
improved, and safer and more predictable navigation can be promoted, by including
standardized turning patterns in their algorithms. There are restrictions to take into
account in any study. Turns were the only feature of ship travel that I examined; I
did not include changes in speed or route optimization. By adding more variables
and performing a more thorough examination of ship behavior based on AIS data,
future studies may build on this investigation.

In conclusion, the use of elastic trend filtering with circle fitting algorithms to ana-
lyze turns and the analysis of explainable ship journeys using AIS data have both
yielded substantial insights into ship navigational behavior. The results have ramifi-
cations for navigating rules, ship safety, and the creation of autonomous ship control
systems. This study expands the realm of explainable ship journeys and shows how
AIS data analysis has the potential to increase transparency and enhance naviga-
tional procedures in the maritime sector.

43
Bibliography

[1] How ais works. https://wwwcdn.imo.org/localresources/en/OurWork/

Safety/Documents/AIS/Resolution%20A.1106(29).pdf, 2007.

[2] Dong Yang, Lingxiao Wu, Shuaian Wang, Haiying Jia, and Kevin Li. How big
data enriches maritime research – a critical review of automatic identification
system (ais) data applications. Transport Reviews, 39:1–19, 07 2019.

[3] https://www.marinetraffic.com/en/ais/home/centerx:-71.619/
centery:-33.040/zoom:19.

[4] International Transport Forum. ITF Transport Outlook 2021. 2021.

[5] IMO IMO. Resolution msc. 74 (69): adoption of new and amended performance
standards. MSC 69/22/Add. 1, 74(May):20, 1998.

[6] Iala guideline no. 1028 on the automatic identification system (ais), part i op-
erational issues. 1, 2002.

[7] International Maritime Organization. Solas chapter v: Safety of navigation,

2002.

[8] A Resolution. 1106 (29)“revised guidelines for the operational use of shipborne
automatic identification systems (ais)”. IMO, London, 2015.

[9] Mélanie Fournier, R Casey Hilliard, Sara Rezaee, and Ronald Pelot. Past,
present, and future of the satellite-based automatic identification system: Areas
of applications (2004–2016). WMU journal of maritime affairs, 17(3):311–345,
2018.

[10] Dimitris Zissis, Konstantinos Chatzikokolakis, Giannis Spiliopoulos, and Marios

44
 Vodas. A distributed spatial method for modeling maritime routes. IEEE
Access, 8:47556–47568, 2020.

[11] Enmei Tu, Guanghao Zhang, Lily Rachmawati, Eshan Rajabally, and Guang-
Bin Huang. Exploiting ais data for intelligent maritime navigation: A compre-
hensive survey from data to methodology. IEEE Transactions on Intelligent
Transportation Systems, 19(5):1559–1582, 2017.

[12] Abdoulaye Sidibé and Gao Shu. Study of automatic anomalous behaviour detec-
tion techniques for maritime vessels. The journal of Navigation, 70(4):847–858,
2017.

[13] Maria Riveiro, Giuliana Pallotta, and Michele Vespe. Maritime anomaly detec-
tion: A review. Wiley Interdisciplinary Reviews: Data Mining and Knowledge
Discovery, 8(5):e1266, 2018.

[14] Etienne Martineau and Jean Roy. Maritime anomaly detection: Domain intro-
duction and review of selected literature. 2011.

[15] N Ferlansyah and S Suharjito. A systematic literature review of vessel anomaly

behavior detection methods based on automatic identification system (ais) and
another sensor fusion. Adv. Sci. Technol. Eng. Syst. J, 5:287–292, 2020.

[16] Azadeh Moghtaderi, Patrick Flandrin, and Pierre Borgnat. Trend filtering via
empirical mode decompositions. Computational Statistics & Data Analysis,
58:114–126, 2013.

[17] G Mosheiov and A Raveh. On trend estimation of time-series: a simple linear

programming approach. Journal of the Operational Research Society, 48(1):90–
96, 1997.

[18] Emanuele Carlini, Vinı́cius Monteiro, Amilcar Soares, Mohammad Etemad,

Bruno Brandoli, and Stan Matwin. Uncovering vessel movement patterns from
ais data with graph evolution analysis. 01 2020.

[19] Hyeong-Tak Lee, Jeong-Seok Lee, Hyun Yang, and Ik-Soon Cho. An ais data-
driven approach to analyze the pattern of ship trajectories in ports using the
dbscan algorithm. Applied Sciences, 11(2), 2021.

45
[20] Konrad Wolsing, Linus Roepert, Jan Bauer, and Klaus Wehrle. Anomaly detec-
tion in maritime ais tracks: A review of recent approaches. Journal of Marine
Science and Engineering, 10(1), 2022.

[21] Enmei Tu, Guanghao Zhang, Lily Rachmawati, Eshan Rajabally, and Guang-
Bin Huang. Exploiting ais data for intelligent maritime navigation: A compre-
hensive survey from data to methodology. IEEE Transactions on Intelligent
Transportation Systems, 19(5):1559–1582, 2018.

[22] Ais in maritime research. Marine Policy, 106:103520, 2019.

[23] https://www.bbc.com/news/business-56559073.

[24] Tadeusz Stupak. Influence of automatic identification system on safety of nav-

igation at sea. TransNav, the International Journal on Marine Navigation and
Safety of Sea Transportation, 8:337–341, 09 2014.

[25] https://wwwcdn.imo.org/localresources/en/OurWork/Safety/
Documents/AIS/Resolution%20A.1106(29).pdf.

[26] https://www.liscr.com/sites/default/files/SOLAS%20V_Reg19.pdf.

[27] https://www.marineinsight.com/marine-navigation.

[28] Jie Cai, Gang Chen, Marie Lützen, and Niels Gorm Maly Rytter. A practical
ais-based route library for voyage planning at the pre-fixture stage. Ocean
Engineering, 236:109478, 2021.

[29] M. Le Tixerant, D. Le Guyader, F. Gourmelon, and B. Queffelec. How can au-

tomatic identification system (ais) data be used for maritime spatial planning?
Ocean I& Coastal Management, 166:18–30, 2018. Maritime Spatial Planning,
Ecosystem Approach and Supporting Information Systems (MapSIS 2017).

[30] Delponti P Almunia J and Rosa F. Using automatic identification system (ais)
data to estimate whale watching effort. 03 2021.

[31] Branko Ristic, B. Scala, M.R. Morelande, and N. Gordon. Statistical analysis of
motion patterns in ais data: Anomaly detection and motion prediction. pages
1–7, 01 2008.

46
[32] Shuguang Chen, Yikun Huang, and Wei Lu. Anomaly detection and restoration
for ais raw data. 03 2022.

[33] H. Rong, A.P. Teixeira, and C. Guedes Soares. Data mining approach to ship-
ping route characterization and anomaly detection based on ais data. Ocean
Engineering, 198:106936, 2020.

[34] Michele Vespe, Ingrid Visentini, Karna Bryan, and Paolo Braca. Unsupervised
learning of maritime traffic patterns for anomaly detection. volume 2012, pages
1 –5, 05 2012.

[35] Clément Iphar, Aldo Napoli, and Cyril Ray. Data quality assessment for mar-
itime situation awareness. ISPRS Annals of Photogrammetry, Remote Sensing
and Spatial Information Sciences, II-3/W5:291–296, 08 2015.

[36] Juyoung Jeong, Yoon Mo Jung, and Sangwoon Yun. Elastic trend filtering.
International Journal of Nonlinear Sciences and Numerical Simulation, 0, 2022.

[37] Theodore Alexandrov, Silvia Bianconcini, Estela Bee Dagum, Peter Maass, and
Tucker S McElroy. A review of some modern approaches to the problem of trend
extraction. Econometric Reviews, 31(6):593–624, 2012.

[38] Robert J Hodrick and Edward C Prescott. Postwar US business cycles: an

empirical investigation. Journal of Money, Credit, and Banking, pages 1–16,
1997.

[39] Ruey S Tsay. Outliers, level shifts, and variance changes in time series. Journal
of Forecasting, 7(1):1–20, 1988.

[40] WILLIAM A Link and JOHN R Sauer. Estimating equations estimates of

trends. Bird Populations, 2:23–32, 1994.

[41] Sander Greenland and Matthew P Longnecker. Methods for trend estima-
tion from summarized dose-response data, with applications to meta-analysis.
American Journal of Epidemiology, 135(11):1301–1309, 1992.

[42] Lennart Ljung. System identification - Theory for the User, 2nd ed. Prentice
Hall PTR, 1999.

47
[43] Amir H Shirdel, Jari M Böling, and Hannu T Toivonen. System identification
in the presence of trends and outliers using sparse optimization. Journal of
Process Control, 44:120–133, 2016.

[44] Seung-Jean Kim, Kwangmoo Koh, Stephen Boyd, and Dimitry Gorinevsky. ℓ1
trend filtering. SIAM Review, 51(2):339–360, 2009.

[45] Christian H Reinsch. Smoothing by spline functions. Numerische Mathematik,

10(3):177–183, 1967.

[46] Denise R Osborn. Moving average detrending and the analysis of business
cycles†. Oxford Bulletin of Economics and Statistics, 57(4):547–558, 1995.

[47] Seung-Jean Kim, K. Koh, M. Lustig, Stephen Boyd, and Dimitry Gorinevsky.
An interior-point method for large-scale l1-regularized least squares. Selected
Topics in Signal Processing, IEEE Journal of, 1:606 – 617, 01 2008.

48