This article was originally published in a journal published by

Elsevier, and the attached copy is provided by Elsevier for the

author’s benefit and for the benefit of the author’s institution, for
non-commercial research and educational use including without
limitation use in instruction at your institution, sending it to specific
colleagues that you know, and providing a copy to your institution’s
administrator.

All other uses, reproduction and distribution, including without

limitation commercial reprints, selling or licensing copies or access,
or posting on open internet sites, your personal or institution’s
website or repository, are prohibited. For exceptions, permission
may be sought for such use through Elsevier’s permissions site at:

http://www.elsevier.com/locate/permissionusematerial
 Author's Personal
ARTICLE Copy
IN PRESS
Reliability Engineering and System Safety 95 (2010) 573–589

Contents lists available at ScienceDirect

Reliability Engineering and System Safety

journal homepage: www.elsevier.com/locate/ress

Probability modelling of vessel collisions

Jakub Montewka , Tomasz Hinz, Pentti Kujala, Jerzy Matusiak
Aalto University, School of Science and Technology, Department of Applied Mechanics, Marine Technology, P.O. Box 15300, FI-00076 AALTO, Espoo, Finland

a r t i c l e in fo abstract

Article history: Among engineers, risk is defined as a product of probability of the occurrence of an undesired event and
Received 9 September 2009 the expected consequences in terms of human, economic, and environmental loss. These two
Received in revised form components are equally important; therefore, the appropriate estimation of these values is a matter of
16 January 2010
great significance. This paper deals with one of these two components—the assessment of the
Accepted 18 January 2010
Available online 25 January 2010
probability of vessels colliding, presenting a new approach for the geometrical probability of collision
estimation on the basis of maritime and aviation experience. The geometrical model that is being
Keywords: introduced in this paper takes into account registered vessel traffic data and generalised vessel
Marine traffic safety dynamics and uses advanced statistical and optimisation methods (Monte Carlo and genetic
Collision probability
algorithms). The results obtained from the model are compared with registered data for maritime
Minimum distance to collision (MDTC)
traffic in the Gulf of Finland and a good agreement is found.
Ship motion model
Gulf of Finland & 2010 Elsevier Ltd. All rights reserved.

1. Introduction encounter between two vessels at a distance greater than Dij

At present the most common approach used for static collision
probability assessment is rooted in the research carried in the
1970s by Fujii et al. [1,2] and MacDuff [3]. According to this, the
probability of a collision is defined as follows:
P ¼ NA PC
where NA is the geometrical probability of a collision course and
PC the causation probability, also called the probability of failing
to avoid a collision when on a collision course. A ship on a
collision course is called a collision candidate, which may end up
as a collision as a result of technical failure or human error. The
causation probability quantifies the proportion of cases in which a
collision candidate ends up as a collision. The value of the
causation probability for this analysis is adopted from a state-of-
the-art model based on a Bayesian Belief Network developed in
earlier research [4–6]. This paper focuses on modelling the
geometrical probability of a collision, and presents a novel
approach that takes into account ship manoeuvrability.
The Fujii’s approach, which has been commonly used in recent
decades, won the popularity among researches mainly due to its
simplicity and robustness. However it has also some drawbacks, it
lacks ship dynamics and it assumes that a collision happens when
two vessels come to a distance defined as the ‘‘collision diameter’’
(Dij) depicted in Fig. 1, which means almost physical contact. Such
an assumption may lead to an understanding that in any
 Corresponding author. Tel.: + 358 945123499, Mobile:+ 358401874777.
E-mail address: jakub.montewka@tkk.fi (J. Montewka).
they are able to avoid a collision, which in most cases is not true.
Therefore one may suspect that the model may not fully reflect
the real interaction between vessels at close quarters. Despite the
drawbacks the model was adopted by Pedersen, and with minor
modifications was used to determine the safety of navigation in
many European waters [7–9]. Hence in Europe it is mostly known
ð1Þ as the Pedersen model.
Another approach based on a ship domain theory was applied
to perform a risk analysis for a large suspension bridge by
Pedersen [10], in which the critical distance between two ships
was assumed a constant value, and only head-on encounters were
considered. A similar approach was used by Fowler and Sørgård
[11] in their model, which assumed a critical situation as a
situation when two vessels come to close quarters, crossing
within 0.5 Nm of each other, which was fixed regardless of the
crossing angle. A series of papers utilizing the ship domain
approach for ship-fixed object collision assessment was published
also by Gluver and Olsen [12].
A model for encounter probability estimation proposed by
Kaneko [13] defines a critical area of an optional form of a closed
boundary, around a ship, whose violation means collision. Kaneko
in his model recognizes two shapes of the critical area:
rectangular and circular, but again the size of the area is fixed,
and ship manoeuvrability is not considered.
In the last decade two trends became popular in the field of
marine transportation safety assessment; one in which methods
based on Markov processes are utilized [14,15], and another that
makes use of the Bayesian Belief Network [16–18]. However ship
dynamics in terms of manoeuvring abilities is rarely mentioned
there as well.

0951-8320/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ress.2010.01.009
 Author's
ARTICLEPersonal Copy
IN PRESS
574 J. Montewka et al. / Reliability Engineering and System Safety 95 (2010) 573–589
Fig. 1. Definition of collision diameter according to Pedersen [8]
At present it is very popular to apply models based on marine
traffic simulation, which are commonly known as fast time
numerical navigators (FNNs), in the assessment of collision and
grounding frequencies [19–23]. But whenever possible the
geometric approach is implemented to the FNN, simply because
of the speed and stability of the former, and similar to the
previous approaches it is a very common practise to consider the
critical distance between two ships as a constant value [22].
Although a definition of the critical distance between two
ships being on a collision course is often met in papers dedicated
to collision avoidance systems [24–26], it is very rarely trans-
ferred directly to collision and grounding assessment process.
In the approach presented in this paper, it is assumed that a
collision between two vessels becomes reality when the distance
between the vessels is not enough to perform efficient anti-collision
manoeuvres. The space and time required for a given vessel to
perform a given manoeuvre depends mostly on her hydrodynamics
and manoeuvrability features. For the first time the idea of ship’s
manoeuvrability implementation into a collision assessment model
was presented by Curtis [27], but his model was limited to one ship
type, which was a very large crude carrier (VLCC) and only
overtaking and head-on situations were considered.
The main intention of this paper is to obtain the minimum
distance between two vessels on collision courses that allows
them to perform efficient anti-collision manoeuvres, using a ship
in-plane motion model. It is assumed that these two vessels are
navigating in deep water, they start manoeuvres simultaneously,
and the rudder and engine settings are constant during the time of
the manoeuvres.
In the following chapters this critical distance is called the
minimum distance to collision (MDTC) and it is an input value for the
model that assesses the probability of a collision. The experiment
was conducted for 4 vessel types of different sizes, proceeding at
their maximum velocities. This finally produced 8 meeting scenarios.
Each scenario was repeated for 17 crossing angles.
The model is expected to work with data transmitted by a
ship-borne automatic identification system (AIS) as an input.
Therefore it has been tailored to be used with limited input
values. The model of ship dynamics that is implemented uses
approximations that allow an AIS message and information about
the ship type to be sufficient to give some realistic output.
The analysis presented in the paper concerns the most dense
sea area in the Gulf of Finland (GoF), in terms of marine traffic,
which is a junction of waterways between Helsinki and Tallinn.
2. Collision probability modelling
The collision probability prediction model presented in this
paper is based on a molecular collision model and is supported by
a model of ship dynamics. The molecular collision model is one of
the several models used in aviation for the probability analysis of
aircraft colliding [28]. Its application for marine purposes was
presented by MacDuff [3], but in his approach ship manoeuvr-
ability was not considered. This paper presents a two-dimensional
model, reduced from the original three-dimensional case for the
purposes of marine navigation. In this model the vessel is
represented as a particle surrounded by a disc of a given radius
(Fig. 2a), which constitutes a ‘‘no-go area’’ for other objects. Some
researchers define the vessel static domain in the similar way
[26,29], while others call it the bumper [30] or the guarding ring
[31]. The definition of the domain expresses it as the area around
the vessel that the navigator wants to keep clear of other vessels
or objects [32,33], and therefore a violation of the vessel’s domain
is not tantamount to a collision. In the presented model a
violation of the outer boundary of the disc is equal to a collision
(Fig. 2). A collision between two vessels (two discs) is described as
an overlap of these two discs. The occurrence of such an overlap is
equivalent to an event in which a point representing the centre of
one vessel enters the disc of which the radius equals the sum of
the radii of two original discs (Fig. 2b).
The diameter of the greater disc (Fig. 2b) is not fixed, and is
computed for each type of vessel and encounter individually, thus
it changes with situations. The value of that disc diameter is
considered the MDTC. If the distance between these two vessels
becomes less then MDTC, it means that a collision cannot be
avoided by any manoeuvres, and the vessels will collide. In other
words, as long as the disc of radius MDTC is not violated by
another vessel no collision will take place. This situation and
relations between MDTC, safe passing, and collision are presented
graphically in Fig. 3. The main factors relevant to determine the
MDTC value are: the vessels’ manoeuvrability, the angle of
intersection (labelled a in Fig. 3a), and the relative bearing from
one vessel to the other (angle labelled b in Fig. 3a). Recent
research has revealed that the vessel’s safety domain, which may
be comparable to MDTC, had a relatively low correlation with the
sea state and wind force [31]. Therefore the hydro-meteorological
conditions are not considered.
2.1. Model formulae
This chapter contains the formulae used in the model, and
presents results of the analysis of three main types of vessel
encounters: overtaking, head-on, and crossing.
Overtaking means a situation in which two vessels are
proceeding on the same route, lying on almost parallel courses,
with courses difference not exceeding 101.
Head-on means a situation in which vessels are lying on
almost reciprocal courses, and the course difference falls in the
range 1807101.
Crossing means a situation in which the difference between
two vessels’ courses falls in the range 10–1701.
Fig. 2. Representation of vessels as discs and definition of collision situation.
 Author's Personal
ARTICLE
IN PRESS
J. Montewka et al. / Reliability Engineering and System Safety 95 (2010) 573–589
Fig. 3. Relationships between MDTC, safe passing distance, and collision.
2.1.1. Vessels on parallel courses – overtaking
In this case, the vessels are navigating along the same route, on
mutually parallel courses, with the difference between their
courses not exceeding 10 degrees, but at different velocities. The
geometrical probability of a collision between two vessels while
overtaking also referred to as the numbers of candidates for
collision during overtaking, is expressed as follows:
Novertaking ¼ TO PO
where TO is the overtaking rate and PO the probability that the
vessels’ domains are violated during overtaking. The overtaking
rate is the number of vessels that will overtake another while on
parallel courses, irrespective of the passing distance. The over-
taking rate is not a collision frequency, unless the waterway width
is zero. It is calculated with the following formula [28]:
N2 0
TO ¼ E ½Vij 
2L
where N is the expected number of vessels in the waterway on
parallel courses, L the length of waterway, and E0 (Vij) denotes the
expected relative velocity of all pairs of vessels of types i and j.
The expected relative velocity is determined as follows [28]:
1 XN
E0 ½Vij  ¼ E½Vij 
N1 i;j
Z Z qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
E½Vij  ¼ ðVi2 þVj2 2Vi Vj cosyÞPv ðVi ÞPv ðVj Þdvi dvj
Vi Vj
where Vi is the velocity of a vessel of given group [m/s], Y the
angle of intersection, which is defined as the difference between
the courses of vessels in groups i and j, and Pv(V) denotes the
probability density function of velocity for a given type of vessel.
The probability that the vessel domains will be violated while
overtaking (PO) equals the probability of the event that two
vessels will pass each other at a distance less than the adopted
value and is expressed as follows:
 
B1 þ B2
PO ¼ P x r ð6Þ
2
where x is the distance between two ships while overtaking and B
the breadth of a vessel of a given class.
2.1.2. Vessels on parallel courses — head-on
The number of collision candidates during head-on meetings is
calculated using the formulae presented in Eqs. (2)–(6). It is
assumed that the differences between the two vessel courses are
Copy
575
to be within 1807101 to consider such a situation as a head-on
meeting.
2.1.3. Vessels on crossing courses
To calculate the collision rate at the intersections of water-
ways, it is assumed that the vessels are entering the waterway
with a given velocity, which is modelled by a continuous
ð2Þ distribution, and with a given intensity, which is assumed to be
constant. The processes of the flow of vessels into waterways are
independent. The number of collision candidates is determined on
the basis of the following equation [28]:
X RE0 ½Vij li lj
Ncrosing ¼ ð7Þ
i;j
Vi Vj sina
where R is the MDTC value, l denotes the intensity of the vessels
entering the waterway, V is the velocity of the vessels according to
ð3Þ
type, and a the angle of intersection of the waterways. The number
of collision candidates depends on the vessels’ velocities and
dimensions, the traffic intensity, and the angle of intersection of
the waterways. The rate does not depend on the lateral distribution
of vessels across the waterway, whereas it is crucial in the case of
overtaking and head-on situations. Eqs. (6) and (7) are based on
different assumptions and are intended for different types of
ð4Þ
meeting. Eq. (7) is used in the case of crossing cases and includes
a MDTC value, which Eq. (6) does not. On the other hand the Eq. (6)
is used in the case of overtaking and head-on meetings and
ð5Þ
considers a spatial distribution of vessels across waterways, which is
not relevant in Eq. (7) as mentioned above. It should be understood
that these two equations are not to be used interchangeably, they
are independent. There is a strict range of angles they may be used
within, as was defined at the beginning of this chapter. Formulae
used beyond their angles’ range may produce wrong results.
2.2. Assessment of minimum distance to collision
To determine the value of MDTC and the factors that may
affect it, an experiment using the hydrodynamic model of ship
motion was conducted and several crossing-type meeting scenar-
ios were simulated. Analysis was carried out for the following:
 3 types of vessels (container carrier, passenger vessel, and
tanker);
 9 meeting scenarios, described in the next part of the paper;
 17 crossing angles varying from 101 (almost overtaking) to
1701 (almost head-on), with 101 increments (Fig. 4);
 Author's
ARTICLEPersonal Copy
IN PRESS
576 J. Montewka et al. / Reliability Engineering and System Safety 95 (2010) 573–589

Fig. 4. Vessel relative positions with three chosen crossing angles, before they start to manoeuvre.

Table 1
Turning-away manoeuvres.

Type of manoeuvre Vsl_B’s course Vsl_A to port, Vsl_B to stbd Both to port Both to stbd Vsl_A to stbd, Vsl_B to port

o1901–2701) No Yes Yes No

o2701–3501 4 Yes No No No

 4 types of manoeuvres conducted by both vessels to avoid

collision, as presented in Table 1.

The following assumptions were made:

 initially the vessels are proceeding at full sea speed;

 the vessels are fully laden;
 the vessels start their manoeuvres simultaneously;
 the settings of the steering gear and propulsion during the
manoeuvre are constant;
 the initial course of vessel A is always 3601 and vessel B
changes her initial course in each consecutive trial by 101;
 the initial relative bearing from vessel A to vessel B is 45 721
starboard;
 only the ‘‘turning away manoeuvres’’ performed by means of
rudder are considered;
 the influence of weather conditions is omitted;
 the curvature of the Earth is omitted because of the small
distances being considered.
The ‘‘turning-away manoeuvre’’ implies course alteration away
from each other to avoid collision and to shorten the time at close
quarters. The data presented in Table 1 define the turning-away
manoeuvres according to the initial course of vessel B.
The initial relative positions of the vessels are presented in
Fig. 4. The crossing angle number depends on vessel B’s course.
Number 1 represents vessel B on a course of 3501 and number 17
describes a meeting where vessel B’s course equals 1901.
The most intriguing question is if and how the types of
meeting vessels affect the value of MDTC. For this purpose the
following meeting scenarios were simulated, appropriate nomen-
clature was adopted, then used in Fig. 7:
 two container ships of the same size (cont_cont),
 two container ships of different sizes (cont_diff),
 two tankers of the same size (tanker_tanker),
 two tankers of different sizes (tankers_diff),
Fig. 5. Relations between value to be optimised (an initial position of vessel B at
time instant 0) and value of ei that is to be found a minimum.
 passenger vessel–tanker (tanker_pass),
 two passenger vessels (pass_pass),
 passenger vessel–container ship (pass_cont),
 ro-ro–container ship (ro-ro_cont),
 two ro-ro carriers (ro-ro_ro-ro).
The analysis was conducted for the traffic in the Gulf of Finland
(GoF); therefore the sizes of typical ‘‘GoF vessels’’ are considered.
A detailed analysis of marine traffic in this area is conducted in
Section 3, and typical ‘‘GoF vessels’’ are presented. The algorithm
for MDTC assessment is presented in Figs. 5 and 6. The
optimisation process whose aim is to obtain the minimum
distance between the initial positions of two manoeuvring
vessels (MDTC) is performed with the use of genetic algorithms.
The aim of the optimisation process was to find the value of B0,
which is the initial position of the centre of gravity of vessel B at
time instant 0, as presented in Fig. 5. An iteration algorithm was
used to determine the B0 value for a certain meeting scenario, by
 Author's Personal
ARTICLE Copy
IN PRESS
J. Montewka et al. / Reliability Engineering and System Safety 95 (2010) 573–589 577

Fig. 6. Block diagram of the algorithm for MDTC assessment.

finding a minimum value of ei as follows:
minðei Þ ) B0
ei ¼ Bi Ai
0:6LOA 4 minðei Þ 40:5LOA
where Bi is the position of the centre of gravity of vessel B at the
time instant i: (XB, YB, ti), Ai the position of the vessel A’s centre of
gravity at the time instant i:(XA, YA, ti), ei the distance between the
centres of gravity of two vessels at the time instant i, and LOA
denotes the average value of length over all of two vessels
involved. It was assumed that a distance between the centres of
gravity of two appropriate vessels that is larger than 0.5LOA
implies non-contact meeting. The idea of the above process is
presented graphically in Fig. 5, and values of ei for consecutive
time instants are presented.
ð8Þ
2.2.1. Results of numerical experiment of MDTC assessment
The results of the experiment are presented in Fig. 7, together
with the corresponding ‘‘collision diameter’’ values (CD) used in
Pedersen’s model [8]. The collision diameter and MDTC may be
considered equivalent in principle, and therefore a comparison of
results is performed, to show the discrepancies between these
two values. Fig. 7 consists of a number of curves, which represent
the meetings of different types of vessels. According to the legend
 Author's
ARTICLEPersonal Copy
IN PRESS
578 J. Montewka et al. / Reliability Engineering and System Safety 95 (2010) 573–589

Tanker_Tanker
5 Tankers_cd
Tanker_Pass
Tanker_Pass_cd
4 Pass_Cont
Pass_Cont_cd

MDTC (LOA)
Cont_diff
3 Cont_diff_cd
RoRo_RoRo
RoRo_cd
2
Cont_Cont
Cont_cd
Tankers_diff
1
Tankers_diff_cd
Pass_Pass
Pass_Pass_cd
0
10 30 50 70 90 110 130 150 170
Angle of intersection (deg)

Fig. 7. Values of MDTC obtained for all meeting scenarios, with corresponding values of collision diameters.

Table 2
Results of Kruskal–Wallis test for MDTC values. 6

Segment a p Hypothesis rejected

5
10–601 0.05 0.213 H1
70–1001 0.05 0.813 H1 4
MDTC (LOA)

110–1401 0.05 0.899 H1

150–1701 0.05 0.518 H1
3

the values of MDTC and CD are non- dimensional, expressed in 2

terms of the average length of the vessels involved. Depending on
the angle of intersection, the discrepancies vary. For angles in the 1
range 10–301 and 140–1701 the values of MDTC and CD do not
differ much but for the angles 90–1101 the highest discrepancy is 0
observed. This range of crossing angles is the most unfavourable 0 20 40 60 80 100 120 140 160 180
in the sense of the value of course alteration required to avoid a Angle of intersection (deg)
collision. 1_port_2_stb Both_to_port CD
Depending on the types of vessels engaged in the meeting,
Fig. 8. MDTC and CD values computed at 95% confidence level by Monte Carlo
there are slight differences in the MDTC values. To prove whether
simulations.
these differences are statistically significant, an appropriate
analysis was performed. A hypothesis was formulated that each
MDTC curve was drawn from the same population, and therefore In all cases that were analysed the test statistics had a p-value
there is no influence of the vessel type on the value of MDTC. To higher than the adopted a-value (p 40.05), indicating that the
prove this hypothesis statistical tests were conducted. null hypothesis cannot be rejected. Therefore the results allow it
At first the Kolmogorov–Smirnov normality test was per- to be assumed that all MDTC values are drawn from the same
formed. This test compares the empirical cumulative distribution population. Because of the limited survey sample, further analysis
function of sample data with the distribution expected if the data was based on the Monte Carlo simulation and both mean and
were normal. The test hypotheses are H0: distributions of the standard deviation values were obtained. Afterwards, both MDTC
random variable follow the Gaussian distribution, versus H1: and CD values at a 95% confidence level were calculated, and the
distributions do not follow the Gaussian distribution. The p-values results are presented in Fig. 8.
of this test were lower than the chosen a level (p o0.05) for 90%
of the observations, and therefore it was concluded that the
2.3. Ship dynamics modelling
analysed random variable’s distribution is non-normal.
In the next step a nonparametric Kruskal–Wallis test of the
A vessel’s manoeuvrability is estimated with the quasi-linear
equality of medians for two or more populations was performed;
modular hydrodynamic model of the vessel in-plane motion. Ship
it tests whether two or more independent samples come from
motion is calculated from the following generalised formulae:
identical populations. The test hypotheses are H0: the population
8
medians are all equal versus H1: the medians are not all equal. The >
> du ðX þXprop þmrvÞ
>
> ¼ resist
test does not require the data to be normal, but instead uses the >
> dt ðmX u_ Þ
>
>
rank of the data values rather than the actual data values for the < dv ðmru þ Yv v þ Yr r þYr_ r_ þ LÞ
analysis. In the present case the data were ranked by types of ¼ ð9Þ
>
> dt ðmYv_ Þ
vessels that are at close quarters, and therefore 8 groups were >
>
>
> dr ðNv v þNr r þNv_ v_ þ LLPP =0:5Þ
>
> ¼
created, and named according to the legend in Fig. 7. Then the : dt ðIzz Nr_ Þ
range of intersection angles was divided into four segments
(10–601; 70–1001; 110–1401; 150–1701), and a Kruskal–Wallis where u is the linear velocity along X-axis, Xresist the ship
test was run for each segment individually. The results of this test resistance (Eq. (10)), Xprop the propeller thrust (Eq. (11)), m
are presented in Table 2. denotes the ship mass, r is the angular velocity along axis Z-axis, v
 Author's Personal
ARTICLE Copy
IN PRESS
J. Montewka et al. / Reliability Engineering and System Safety 95 (2010) 573–589 579

the linear velocity along Y-axis, L the lift forces on the rudder where n stands for propeller revolutions [min  1], D is the
(Eq. (12)), Lpp the length of the ship between perpendiculars, Izz is propeller diameter [m], and VA is the speed of advance of the
a mass moment of inertia, and Xu_ , Yv, Yr, Yr_ , Yv_ , Nv, Nr, Nv_ , Nr_ are propeller [m/s]. The values of KT(J) are derived from the literature
hydrodynamics derivates and are calculated using formulae [41] for a screw of the Wageningen B series. In some cases,
derived from the literature [34–36]. especially for low ship speed, Eq. (19) may give unrealistic results.
To prevent these abnormally high thrust values, a limiting value
0:5ru2 SCT
Xresist ¼  ð10Þ was established, which corresponds to the bollard pull value. The
ð1tÞ
following formulae are used to calculate the bollard pull,
where r is the water density [tonne/m3], S denotes the wetted depending on the direction of propeller revolution:
surface of the ship hull (Eq. (18)), CT the total resistance coefficient (
7:8ðPD DÞ2=3 for n 4 0
(Eq. (17)), and t is the thrust deduction factor (Eq. (16)). The TBP ¼ ð21Þ
propeller thrust is calculated with the following conditional 5:5ðPD DÞ2=3 for n o0
equations: where PD is the power of the ship’s main engine [kW], which is
(
T TBPn o 0 r T r TBP estimated by means of Eq. (23). The Propeller diameter is
Xprop ¼ ð11Þ estimated with the use of the following formula:
TBP n 4 04T 4TBP
D ¼ 0:6 Tmax ð22Þ
where T is the propeller thrust calculated by means of Eqs. (19)
and (20), and TBP denotes the bollard pull, obtained with the use of The vessel’s power is estimated on the basis of modified
Eq. (21). The lift force on the rudder is calculated by means of the Admiralty formula:
following formula: ra V b
PD ¼ ð23Þ
0:5rAR VR2 2pLðL þ1Þ ZD C
L¼ ð12Þ
ðL þ 2Þ2 where V is the vessel velocity [kn], ZD the propulsion efficiency, C
where r is the water density [t/m3], AR the rudder area [m2], VR denotes the Admiralty constant, and a and b are exponents
the velocity near the rudder [m/s], and L the aspect ratio estimated for each type of vessel individually.
calculated as follows: Each class of analysed vessels that was analysed was studied to
find values of ZD, C and exponents of r and V. The research was
b2 focused on the four main groups of ships that operate most
L¼ ð13Þ
AR frequently across the Gulf of Finland: these are tankers, ro-ro,
where b is the rudder length [m]. The rudder area is estimated on passenger vessels, and container carriers. For each class of vessel
the basis of the following formula [36]: the following data were extracted from a ship database: length,
"  2 # breadth, draught, service speed, displacement, and main engine
2 B power. For the purposes of this paper, two ship databases were
AR ¼ 0:01Lpp Tmax 1 þ50CB ð14Þ
Lpp studied. One database was provided online by the Japanese ship
classification society Nippon Kaiji Kyokai, from which data
where Tmax is the maximum draught of the vessel [m], CB denotes
concerning tankers, container carriers, and ro-ro vessels were
the block coefficient, and B the ship breadth [m]. The block
extracted. Another source of particulars on vessels was the
coefficient is calculated using the following formula by Lee et al.
bulletins issued by the operators of passenger vessels that cruise
[37]:
across Gulf of Finland. For each type of vessel analysed, a sample
4:7 of 60 vessels was extracted from the databases. The exception was
CB ¼ 2:11 pffiffiffiffiffiffiffi ð15Þ
4:03u= LOA the group of passenger vessels (20 vessels); the sample size was
limited by the actual number of vessels in operation. Afterwards
where u is the linear velocity of ship [m/s] and LOA the overall ship
the optimisation method based on genetic algorithms was used to
length [m]. The thrust deduction factor is estimated as follows
analyse the samples and to find the values of ZD , C, and exponents
[38]:
     of r and V. The aim of the optimisation process was to minimise
L B the error function (e) and maximise the prediction interval (O)
t ¼ 0:905 0:17 þ 0:2CB 0:015 OA þ 0:01 2:5 ð16Þ
B T simultaneously. These two are expressed as follows:
where T is the ship draught [m]. The total resistance coefficient is jPPD j
e ¼ 100 ½% ð24Þ
calculated using the following formula [39]: P
KT ðJÞ jNP NPD j
CT ¼ 8=p ð17Þ O ¼ 100100 ½% ð25Þ
J2 NP
where KT(J) is a non-dimensional thrust coefficient and J an
NP e r NPD rNP þ e ð26Þ
advance coefficient calculated with the use of Eq. (20). The wetted
surface of the hull is calculated by means of the following formula where PD denotes the predicted power of the ship’s main engine, P
[40]: the power of the sample vessel, taken from the ship database, NP
1:7 the number of sample vessels for which the engine power is
S ¼ r=B þ ðB=TÞ ð18Þ known (taken from ship database), NPD the number of vessels for
CB 0:2ðCB 0:65Þ
which the predicted power is within the adopted limit, and e is
where r stands for the volume of ship displacement [m3]. The the error function defined by Eq. (24). The results obtained are
model of propeller thrust was calculated using the following presented in Table 3.
formulae:
T ¼ KT ðJÞrn2 D4 ð19Þ 2.4. Validation of applied ship dynamics model

VA Although the ship dynamics model used in this study is well

J¼ ð20Þ
nD known and has been validated many times [35,42], it is justified
 Author's
ARTICLEPersonal Copy
IN PRESS
580 J. Montewka et al. / Reliability Engineering and System Safety 95 (2010) 573–589

Table 3
Approximate values of coefficients used for vessels’ power prediction.

Vessel type ZD C a (exponent of r) B (exponent of V) e (%) O (%)

Ro-ro 0.7 918.0 0.516 3.6 25 83.5

Tankers 0.7 910.5 0.516 3.6 25 73.0
Container carriers 0.7 945.0 0.516 3.5 25 92.0
Passenger ships 0.7 640.0 0.516 3.6 25 90.0

Table 4 Tanker
Main particulars of ships for which the model was validated. 15 350
14
Vessel type L B T DWT Power Speed 300
13
[m] [m] [m] [t] [kW] [kn] 250
12

Course [deg]
Speed [kt]
Ro-ro 199.4 23.4 6.08 7400 18,900 20.9 11 200
Tanker 183.0 32.2 11.0 40,000 8848 13.7 10 150
Container 220.0 32.24 10.5 42,500 26,270 22.2
9
carriers 100
8
7 50
140 6 0
1 21 41 61 81 101 121 141 161 181 201
120
Time [s]
100 Speed_est Speed_obsv Course_est Course_obsv
Time (sec)

80
Container
60 24
160
22
40 20 140
18 120

Course [deg]
20
Speed [kt]

16 100
0 14 80
0 20 40 60 80 100 120 140 160 180 12
60
Angle of intersection (deg) 10
40
1port_2stb_max Both_port_max 8
6 20
Both_port_min 1port_2stb_min
4 0
30 1 21 41 61 81 101 121 141 161 181 201
Time [s]
25
Speed_est Speed_obsv Course_est Course_obsv
20
Frequency

Ro-ro
15 16 500
15
10 450
14
5 13
400
Course [deg]
Speed [kt]

12
0 11 350
0 20 40 60 80 100 120 140 10
Time [sec] 300
9
8
Fig. 9. Time to closest distance between two vessels (ei, min) performing collision 250
7
avoidance manoeuvres.
6 200
1 21 41 61 81 101 121 141 161 181 201
Time [s]

to compare the results coming from the model with the results of Speed_est Speed_obsv Course_est Course_obsv
sea trials to assess to what extent the simplification and
Fig. 10. Estimated value of ship speed and course versus observed data.
approximation adopted in the present study influence the
accuracy of the ship motion model. Validation was carried out
for three ship types, the data for which are presented in Table 4.
In the experiment described in the previous section, whose
aim was to determine the MDTC values, the evasive manoeuvre turning circle are presented. These values were computed in the
was based on the turning circle performed by both vessels. MDTC determination process. The time was counted from the
The validation of simulated manoeuvres was carried out in the beginning of the taking of the anti-collision action until the
domain of time and the time period investigated was 200 s. In minimum distance between these vessels min(ei) was reached
Fig. 9 the histogram and maximum and minimum values of time (Fig. 5). According to these data the maximum time required for
required by vessels to evade a collision situation by performing a vessels to perform a collision avoidance manoeuvre is 130 s.
 Author's Personal
ARTICLE Copy
IN PRESS
J. Montewka et al. / Reliability Engineering and System Safety 95 (2010) 573–589
In the validation process two parameters were compared: the
ship’s course and her speed during circle turning. Data from the
sea trials of newly built ships were chosen as references. The
results are presented in Fig. 10.
The graphs in Fig. 10 show good agreement between modelled
values and the observed ones for vessels equipped with conven-
tional single-propeller engines. Differences in the vessels’ course
and speed occur for vessels equipped with twin-screws engines
(ro-ro). However, the differences that occur until the 100 s time
instant are acceptable in all cases. Time duration of most
manoeuvres carried out in this study did not exceed 100 s, and
therefore the results obtained by means of the model that is
presented may be considered to be in conformity with reality for
the analysis presented in this paper.
3. Marine traffic modelling in the Gulf of Finland
3.1. Marine traffic profile
The vessel traffic profiles over the area under analysis are
described using data derived from the AIS transmissions recorded
in March and July 2006. The registered data do not fully reflect the
existing traffic, mostly because of the limitations of the require-
ments for carrying AIS [43]. Regulation 19 of SOLAS Chapter V
requires AIS to be fitted aboard: all ships of 300 tonnes gross and
upwards engaged in international voyages, all ships of 500 tonnes
gross and upwards not engaged on international voyages, and all
passenger ships irrespective of size. Because of this, all ‘‘small
traffic’’ is not included in the present study, although a large
number of pleasure craft and fishing boats navigate in the Gulf of
Finland, especially during the summer. These boats, despite their
small dimensions, may, in some situations, complicate the traffic
and raise the already high risk of collision and grounding in the
area. For further research this kind of traffic should be estimated
as well. Another reason behind the difference between registered
traffic and the actual situation is the incompleteness of the AIS
information transmitted by vessels. Transmissions without a
Maritime Mobile Service Identity number (MMSI), latitude, or
longitude were not stored in the database. Thus, the total number
of vessels recorded in an area may be smaller than the actual
number by some per cent. The area in question is the junction of
two main waterways: one leads N–S and the other E–W. The N–S
stream consists mainly of passenger vessels cruising between
Helsinki and Tallinn, whereas the E–W stream consists of cargo
Fig. 11. Waterway junction that was analysed with counting gates and main traffic flows marked.
581
vessels bound for and from harbours located in the Gulf of
Finland. To estimate the number of vessels that arrive in and
depart from the Gulf of Finland, counting gate number 1 was
established along meridian 023.51E. To compute the traffic
volumes of the streams in the junction, another two counting
gates were established. Gate number 2 was established along
parallel 601N to count N–S traffic, and gate number 3 along the
meridian 0261E to count E–W traffic (Fig. 11).
The types of vessels and their percentage share of the traffic in
the Gulf of Finland are presented graphically in Fig. 12. The
diagram constitutes the results of 2 months of AIS transmission
recordings, carried out in March and July 2006 at counting gate
numbers 1 (all vessels) and 2 (passenger vessels only). The traffic
recorded in March is considered the winter profile of traffic,
whereas the traffic registered in July is a summer profile of marine
traffic in the Gulf of Finland.
In the diagram shown in Fig. 12 the group labelled ‘‘other’’
consists of tugboats, icebreakers, research vessels, support
vessels, sailing yachts, and vessels for which AIS transmission
was not complete. In the case of messages with an MMSI number,
but without vessel details, the missing information was extracted
from the external ship database. This external database was very
helpful in categorising vessels other than passenger ship and
Type of vessels operating in the Gulf of Finland
Bulk Other General cargo
3% 3% 20 %
Ro-ro
9%
Container
10 %
Tankers
13 %
Passenger
42 %
Fig. 12. Types of vessels operating in the Gulf of Finland.
 Author's
ARTICLEPersonal Copy
IN PRESS
582 J. Montewka et al. / Reliability Engineering and System Safety 95 (2010) 573–589

tankers, whose AIS status was ‘‘cargo vessels’’ (container carriers, Although the number of vessels passing gate 1 is similar
general cargo, ro-ro, bulk carriers). Afterwards the mode and throughout the year, the spatial distribution of traffic is different,
maximum values of length, breadth, and draught in each class of which depends mostly on icing conditions in the area. As an
vessel were calculated, and the results are presented in Table 5. example, histograms of the winter and summer traffic are
The mode value denotes the frequently recorded values for a presented in Fig. 13. The differences in the distributions are
given class of vessel, so they may also be called typical parameters significant and cannot be omitted in risk analysis. The above
of vessels operating in the Gulf of Finland in the year 2006. relations hold true for the other gates as well.
The total number of vessels registered at gates 1 and 2 is For the purposes of safety assessment and traffic modelling the
presented in Table 6. There are no significant differences in the traffic spatial distributions may be approximated either by
number of vessels passing gate 1 during each season (winter, statistical distributions, or by histograms if the registered traffic
summer). However, such differences exist for gate 2, where fits any known distribution poorly.
the number of vessels in summer is marked with an asterisk and
for winter with double asterisks. The main reason for the
3.2. Marine traffic modelling
difference is that during wintertime the operations of high-
speed craft between Helsinki and Tallinn were suspended, and
According to the analysis of traffic composition shown in
therefore the number of passenger vessels on the above route was
Fig. 12, the following main groups of vessels are considered:
reduced.
container carriers, tankers, general cargo vessels, ro-ro, cruise
ships, and fast ferries. Marine traffic in the area under analysis is
assumed to consist of four main flows: east, west, north, and
south, while the north and south flows are assumed to contain
Table 5
Types and sizes of typical and maximal vessels operating in the Gulf of Finland in passenger vessels only. Each flow is modelled with the following
the year 2006. input parameters: overall number of vessels, type of vessels,
number of vessels of a given type, size of vessel of a given type,
Vessel type L_mode L_max B_mode B_max T_mode T_max speed of vessels of a given type, course of vessel, and position of
Ro-ro 162.0 195.0 20.6 28.1 6.7 10.0
vessel across the waterway. For modelling purposes most of these
Tankers 239.0 320.0 27.3 58.0 11.2 22.0 values are approximated by continuous distribution or by
Container carriers 124.5 192.7 22.5 25.4 8.7 10.5 histograms. The distribution of the features being analysed is
Passenger ships 171.3 266.2 28.7 36.0 5.0 8.5 chosen according to the results of a chi-square test. Those which
General cargo ships 85.0 173.5 12.5 27.6 5.3 10.9
fitted best (obtained the highest value of chi-square test) were
selected as inputs for the model. In some cases, if none of the
available distributions fitted then recorded discrete values were
taken into the model, by random sampling.
Table 6
Number of vessels passing gate 1 (E–W traffic) and gate 2 (N–S traffic), registered
in March (**) and July (*) 2006. 3.2.1. Vessel course and speed modelling
In Figs. 14–16 discrete values of vessels’ speed and courses are
Vessel type Gate number 1 Gate number 2
presented. Fig. 14 contains seven graphs, which describe the
East West North South speed of a given class of vessel approximated by continuous
distribution. In all the cases analysed the estimated distributions
Ro-ro 397 400 fitted the observed value well and were adopted for the modelling
Tankers 618 607 purposes.
Container carrier 470 464
Passenger ship 505 502 1140* 488** 1140* 488**
In Figs. 15 and 16 the distributions of vessels’ courses
General cargo ships 895 929 registered at gate 1 are presented. In the case of the East–West
Bulk carriers 153 145 flow shown in Fig. 15 the continuous distribution describes the
Total 3038 3047
analysed variable well and was chosen as input for the model. In
Fig. 16 the vessels’ courses in North and South flows are fitted to

60.0° 60.0°

59.9° 59.9°

59.8° 59.8°

59.7° 59.7°

59.6° 59.6°

59.5° 59.5°

59.4° 59.4°

59.3° 59.3°
1 1
59.2° 59.2°

59.1° 59.1°

59.0° 59.0°
22.7° 22.9° 23.1° 23.3° 23.5° 23.7° 23.9° 24.1° 24.3° 22.7° 22.9° 23.1° 23.3° 23.5° 23.7° 23.9° 24.1° 24.3°

Fig. 13. Traffic flow at gate number 1 in March (to left) and in July (to right) in 2006.
 Author's Personal
ARTICLE Copy
IN PRESS
J. Montewka et al. / Reliability Engineering and System Safety 95 (2010) 573–589 583

Container vessels
BetaGeneral (11.039; 3.3125; -0.44136; 21.329) Tankers
X > 20.11 Logistic (13.6663; 1.0927)
X > 10.78
97.5% 2.5% X > 9.41 X > 17.46
0.25 97.5% 2.5%
0.3

Number of vessels (*1000)

Number of vessels (*1000)

0.2 0.25

0.15 0.2

0.15
0.1
0.1
0.05
0.05

0 0
0 5 10 15 20 25 0 2 4 6 8 10 12 14 16 18
Velocity (kn) Velocity (kn)

General cargo vessels Ro-ro vessels

Norm (11.03; 2.28) BetaGeneral(13.828; 3.9325; -5.8282; 24.070)
X > 6.38 X > 15.77
97.5% 2.5% X > 10.72 X > 21.95
0.18 97.5% 2.5%
0.16
0.16
Number of vessels (*1000)

Number of vessels (*1000)

0.14
0.14
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04 0.04

0.02 0.02

0 0
2 4 6 8 10 12 14 16 18 20 4 6 8 10 12 14 16 18 20 22 24
Velocity (kn) Velocity (kn)
Passenger vessels - north- and southbound
Passenger vessels - east- and westbound Weibull (13.629; 33.042) Shift = -16.059
Logistic(17.5463; 1.2973)
X > 8.76 X > 20.46
X > 12.50 X > 22.60 97.5% 2.5%
97.5% 2.5% 0.2
0.25
0.18
Number of vessels (*1000)
Number of vessels (*1000)

0.16
0.2
0.14
0.15 0.12
0.1
0.1 0.08
0.06
0.05 0.04
0.02
0 0
8 10 12 14 16 18 20 22 24 26 28 30 0 5 10 15 20 25
Velocity (kn) Velocity (kn)
Fastferries. north- and southbound
Logistic (33.8245; 1.1907)
X > 29.19 X > 38.46
97.5% 2.5%
0.25
Number of vessels (*1000)

0.2

0.15

0.1

0.05

0
24 26 28 30 32 34 36 38 40
Velocity (kn)

Fig. 14. Histograms of velocities for different types of vessels with distributions fitted.

the distribution. In the case of the North flow, the distribution fits in the distributions. Therefore this variable cannot be modelled by
well, but for the South flow, there is a significant difference. One the distribution, and a random sample from recorded data is used
may notice two peaks in the histograms, which are not considered as the input value.
 Author's
ARTICLEPersonal Copy
IN PRESS
584 J. Montewka et al. / Reliability Engineering and System Safety 95 (2010) 573–589

LogLogistic (9.5392; 66.072; 12.646) InvGauss(7.6505; 3.3633) Shift=+255.5136

X > 58.1 X > 99.4 X > 255.51 X > 283.59

97.5% 2.5% 100.0% 5.0%
Number of observations (*100 ) 7 0.25

Number of observations (*100 )

6
0.2
5

4 0.15

3
0.1
2
0.05
1

0 0
0 20 40 60 80 100 120 140 160 180 250 260 270 280 290 300 310 320 330 340
Course (degrees) Course (degrees)

Fig. 15. Histograms of vessels’ courses at gate number 1 with distributions fitted (eastbound traffic to the left, westbound traffic to the right).

BetaGeneral(22,273; 5,9173; 83,701; 218,35)

Logistic(367,2069; 5,4454) X > 207,65
X > 166,64
X > 346,0 X > 388,4 97,5% 2,5%
97,5% 2,5% 6
6
Number of observations (*100 )
Number of observations (*100)

5
5

4
4

3 3

2 2

1 1

0 0
290 300 310 320 330 340 350 360 370 380 390 400 130 140 150 160 170 180 190 200 210 220
Course (degrees) Courses (degrees)

Fig. 16. Histograms of vessels’ courses at gate number 2 with distributions fitted (northbound traffic to the left, southbound traffic to the right).

Table 7
Vessel types analysed and distributions used for ship length modelling in the E–W flow.

Vessel type Percentage share in E–W flow Mode (m) Maximum (m) Minimum (m) Distribution

Ro-ro 15 162.0 195.0 77.0 Triangle

Container carriers 17 124.4 203.0 47.2 Triangle
Passenger ships E–W 18 180.7 266.2 79.0 Triangle
Tankers 20 239.0 320.0 58.3 Triangle
General cargo ships 30 85.0 173.5 42.5 Triangle

3.2.2. Vessels’ length and breadth modelling 2. Data from gate number 2 were filtered and sorted to remove
Ship lengths and breadths are modelled for the E–W and N–S vessels that did not enter the crossing being analysed (ferries
streams separately; the percentage shares of the types of vessels from Helsinki to Stockholm, and vessels not bound for Tallinn).
and the distributions used are shown in Tables 7–10. These two streams were first described by continuous distribu-
tions and the results (number of collision candidates) were
compared with the results obtained from the model, which used
3.2.3. Modelling of lateral distribution of vessel across waterways the distribution of vessels across the waterways in the form of
The distribution of vessel positions across the waterways discrete values. The discrepancies were significant, and it was
differs and depends on season and geographical location (Fig. 13). decided not to model this variable by means of continuous
In this research summer traffic in the junction between Helsinki distribution but use discrete values as the input and random
and Tallinn is considered. The E–W stream is modelled on the sampling as the method of data acquisition. Histograms of
basis of the data obtained from gate number 3 (Fig. 11), whereas considered values with the best fit distributions (to present the
the N–S stream is modelled on the basis of data from gate number discrepancies) are shown in Figs. 17 and 18.
 Author's Personal
ARTICLE Copy
IN PRESS
J. Montewka et al. / Reliability Engineering and System Safety 95 (2010) 573–589 585

Table 8
Vessel types analysed and distributions used for ship breadth modelling in the E–W flow.

Vessel type Percentage share in E–W flow Mode (m) Maximum (m) Minimum (m) Distribution

Ro-ro 15 20.6 28.0 12.8 Triangle

Container carriers 17 22.5 25.4 9.0 Triangle
Passenger ships E–W 18 31.5 36.0 15.3 Triangle
Tankers 20 27.3 58.0 9.3 Triangle
General cargo ships 30 12.5 27.6 9.2 Triangle

Table 9
Vessel types analysed and distributions used for ship length modelling in the N–S flow.

Vessel type Percentage share in N–S flow Mode (m) Maximum (m) Minimum (m) Distribution

Passenger ships 48 119.4 242.4 119.0 Triangle/discreet

Fast ferries 52 82.0 119.0 37.5 Triangle/discreet

Table 10
Vessel types analysed and distributions used for ship breadth modelling in the N–S flow.

Vessel type Percentage share in N–S flow Mode (m) Maximum (m) Minimum (m) Distribution

Passenger ships 48 24.2 36.0 21.7 Triangle

Fast ferries 52 16.0 24.2 8.3 Triangle

9 9
8 8
7 7
6 6
Frequency [*10]
Frequency [*10]

5 5
4 4
3 3

2 2

1 1

0 0
7000 12000 17000 22000 7000 9000 11000 13000 15000 17000 19000 21000
Distance_W [m] Distance E [m]

Fig. 17. Histograms of vessels distance from the reference point in the W–E stream counted at gate 3 (west to left).

The distance between ships’ recorded positions and the fixed
reference point for each gate was computed as follows:
 at gate 2 the position of the reference point was: Latitude
60100.00 N, Longitude 024151.50 E; the distance was computed
along the parallel from this position towards the East,
 at gate 3 the position of the reference point was: Latitude
59149.50 N, Longitude 026121.60 E; the distance was computed
along the parallel from this position towards the North.
shown in Fig. 19, and each area was analysed separately. The
following main interactions between flows were analysed:
crossing situations in four crossing areas, named NE, NW, SW,
and SE; overtaking situations in four traffic lanes – North, South,
East, and West; and head-on situations between two pairs of
lanes: East–West and North–South. The length of the E–W leg was
80 Nm, and 30 Nm in the case of the N–S leg.
3.3. Marine accidents

3.2.4. Model output
The outputs of the model are: the number of encounters
(crossing, overtaking, head-on), the geometrical probability of a
collision for a specified type of encounter, and the mean time
between accidents (collisions). Because of the complexity of the
traffic over the junction, it was divided into four crossing areas, as
A thorough analysis of accidents for the whole Gulf of Finland
was conducted by Kujala et al. [6]; in their study a detailed
accident statistics for the period 1997–2006 were presented and
the results of theoretical models were compared with these
statistics. The present analysis is limited to the area of the
junction between Helsinki and Tallinn, and therefore only
 Author's
ARTICLEPersonal Copy
IN PRESS
586 J. Montewka et al. / Reliability Engineering and System Safety 95 (2010) 573–589

2 2.5
1.8
1.6 2

Frequency [*100]
Frequency [*100]
1.4
1.2 1.5
1
0.8 1
0.6
0.4 0.5
0.2
0 0
0 3000 6000 9000 12000 15000 18000 0 3000 6000 9000 12000 15000 18000
Distance_N [m] Distance_S [m]

Fig. 18. Histograms of vessels distance from the reference point in the N–S stream counted at gate 2 (north to left).

Table 11
Positions and dates of collisions in the junction that was analysed during the
period 1987–2007.

Date Latitude Longitude Type of Type of

[N] [E] vessel traffic
involved

02.02.1996 601000 251000 Reefer Summer

15.11.2001 591500 241480 Ro-ro Summer
25.3.2003 591480 241300 Ice Winter
breaker

Drawing inferences about navigational safety from the acci-

Fig. 19. Junction that was analysed with the four main crossing areas marked.
accidents that happened there are considered. Data about marine
accidents in the Gulf of Finland were extracted from the HELCOM
accidents database, which covers the period between 1987 and
2007. According to this database, in the junction between Helsinki
and Tallinn there have been 3 collisions during the last 20 years.
The positions of these collisions are listed in Table 11 and a
graphical presentation is provided in Fig. 20.
It may be noticed that two collisions happened during the
winter season (in the years 1996 and 2003); one of them involved
an icebreaker, but there is no additional information. According to
the archive reports concerning the ice cover in the Gulf of Finland
issued by the Finnish Meteorological Institute, at the beginning of
March 2003 the maximum ice extent was reached, and the whole
Gulf of Finland was covered with ice; in 2001 the ice cover started
to develop in the Gulf of Finland in late December; in February
1996 the ice cover extended from the east as far as the longitude
of Helsinki, and the rest of the Gulf was ice-free. Therefore it may
be assumed that the collision that takes place in the presence of
ice shall be considered a winter traffic collision; otherwise it is a
summer traffic collision. Thus the collision that involved the
icebreaker in 2003 was a winter traffic collision, whereas the
other two due to lack of ice were considered summer traffic
collisions.
The positions of all collisions that happened in the period
under analysis are marked in Fig. 20. The collisions that happened
in the area of the junction between Helsinki and Tallinn are
labelled with exclamation marks; other collisions are labelled
with flags.
dent statistics only or carrying out validation of models that
assess the probability of a collision on the basis of number of
accidents only, which are very rare, is quite difficult and highly
uncertain. Therefore it seems justified to analyse the safety of
navigation on the basis of the numbers of both accidents and
near-miss situations, which may better reflect the collision hazard
[44]. In their report Berglund and Huttunen [45] analysed the
meeting situations in the Gulf of Finland for summer traffic (May,
July, July) in 2006, 2007, and 2008. They provided data concerning
near misses for crossing, overtaking, and head-on encounters. A
near-miss situation was defined there as a meeting of two vessels
being on collision courses at a distance of less than 0.3 Nm. In this
paper the number of near-miss situations is roughly equivalent to
the number of collision candidates computed by the model
presented. In the analysis an assumption, in common use about
blind navigation was made, which means that vessels do not
perform any manoeuvres to avoid each other. In the case of real
encounters vessels are able to steer, and therefore it may be said
that the near misses are controlled, and the number of near-miss
situations is affected by human factors; therefore the assumption
of blind navigation is not met. Therefore, the numbers of near
misses in the case of blind navigation may be expected to be even
higher than those presented in the report. Abridged results of the
analysis are collected in Table 12.
4. Results
Using the model described above, the numbers of encounters
per month for three types of vessel meeting situations are
obtained. Calculations were performed on the basis of the
summer traffic profile, based on AIS data recorded in July 2006.
The modelled values are compared with the number of encoun-
 Author's Personal
ARTICLE Copy
IN PRESS
J. Montewka et al. / Reliability Engineering and System Safety 95 (2010) 573–589 587

Fig. 20. Positions of ship–ship collisions in the Gulf of Finland between years 1987 and 2007.

Table 12 Table 13
Number of near misses, according to type of encounter, summer traffic (May–July) Monthly number of encounters observed and estimated.
in the Gulf of Finland in 2006 [45].
Data source Type of meeting
Encounter/Number Head-on Crossing Overtaking
Crossing Overtaking Head-on
Total 8.0 419.0 943.0
Per month 2.8 142.7 321.1 Berglund and Huttunen [45] 142.7 321.1 2.8
MDTC based model 169.0 407.0 6.8
Pedersen 46.8 146.4 6.4
Pedersen_MDTC 104.3 146.4 6.4

ters observed [45], number of encounters computed by means of

Pedersen’s model [8], and number of encounters obtained with
modified Pedersen model in which the collision diameter values
were replaced with MDTC values. The results of this experiment Table 14
for all four crossing areas according to Fig. 19 are shown in Number of encounters, probability of collision, and collision recurrence for
junction of waterways for summer traffic profile.
Table 13.
In the next step the probability of a collision is computed on Probability of Number of Time between
the basis of the number of encounters obtained and the causation collision per candidates per collisions in
probability values adopted. This paper uses two causation year year years
probability values are commonly accepted [6–8]: 1.3  10  4 for
Crossing
crossing courses and 4.9  10  5 for parallel courses. The obtained Crossing area
values of the probability of a collisions and the time between NE 0.057 440 17.6
collisions for each crossing area are presented in Table 14. NW 0.077 593 13.0
The differences between the estimated and observed values SW 0.054 415 18.5
SE 0.075 576
are presented in Fig. 21. They are expressed as a percentage of
observed values according to the formula Total 0.263 2024 3.8
  Overtaking
NEST NOBS Traffic lane
D¼  100 ð%Þ ð27Þ
NOBS E 0.120 2568 8.0
W 0.144 2304 8.8
where NEST denotes the number of encounters estimated by N 0.009 180 113.0
means of the appropriate model and NOBS is the number of S 0.009 180
encounters observed. Total 0.282 5232 3.5
For all the meeting cases analysed the presented model Head-on
overestimates the number of encounters in comparison with the Traffic lane
observed values in the case of crossing situations by 19%, whereas E–W 0.004 82 249.0
N–S 0.033 667
Pedersen’s model underestimates this number by almost 70%. The
number of encounters computed by a modified Pedersen model, Total 0.037 749
with an MDTC module applied, is still 27% less than the observed
Grand total 0.582 8005 1.7
value.
In the case of overtaking the presented model overestimates
the result by 27%, while, on the contrary, the results given by
Pedersen’s model are underestimated by 54%.
The numbers of head-on meetings determined by the MDTC
based model and the Pedersen’s model are similar, and in both The overestimation of the results given by the MDTC based
cases the results are overestimated, by 140% and 128%, respec- model seems natural. In real traffic, ships have the opportunity to
tively. In both cases only E–W traffic was considered, and N–S avoid each other; therefore the assumption of blind navigation,
traffic was left out. The assumption was made that vessels which is the principle of the geometrical model, is not met.
cruising in the N–S flow, which are passenger liners, are operated Because of this discrepancy one may expect models to give larger
by highly skilled crews, and head-on meetings do not occur (and results than those that can be gained by the observation of real
are controlled if they do). traffic.
 Author's
ARTICLEPersonal Copy
IN PRESS
588 J. Montewka et al. / Reliability Engineering and System Safety 95 (2010) 573–589
Pedersen_MDTC -54
-27
-54
Pedersen
-67
Proposed_model 27
19
-100 -50 0 50 100
Δ (%)
Crossing Overtaking Head on
Fig. 21. Differences between numbers of encounters estimated and observed.
5. Conclusions
This paper presents a new approach for collision probability
modelling, which is an important part of risk analysis process. The
innovative use of a ship motion model to determine the Minimum
Distance To Collision may be recognised; this approach meets the
demands expressed in recent works and the obtained results are
very promising. The number of encounters in the waterways
crossing between Helsinki and Tallinn for the summer traffic
obtained by means of the MDTC based model is presented and the
estimated values are very close to the observed values. However
the difference between the observed and modelled periods
between collisions should be noted. The reason for this is the
causation factor used in this study, which seems not to be
appropriate for application in the MDTC based model, whereas it
is suitable for Pedersen’s model. Although the probability of a
collision obtained by means of Pedersen’s model [6] is very close
to the observed value derived from the accident statistics, the
number of collision candidates shows significant differences from
the observed values.
The causation probability value used in the study was derived
from the literature, and does not take into account specific
characteristics of the area under study. The area under analysis is
under constant surveillance by VTS centres and the traffic is
separated by Traffic Separation Schemes; these factors, among
others, may significantly reduce the causation probability. At
present research is being carried out at Aalto University that aims
to redefine the factors that affect the causation probability. They
will help to tailor the appropriate causation probability model for
the model proposed.
The model presented here is based on MDTC values, computed
with the assumption that two vessels take action designed to
avoid collision. The analysis of MDTC in the case of single-ship
manoeuvres is the aim of the next step of this research.
Another essential factor that should be taken into account in
further research is the variation in marine traffic according to the
time of the day. In the model presented here intensity of ships is
assumed to be constant during the day, and peak hours are not
taken into account, whereas they exist both in the N–S and the E–
W streams. Although this factor would not change the total
number of collision candidates, it might help define the high-risk
periods during the day.
The model presented should also be verified in other sea areas
for which near-miss data are available. As more AIS data become
available, successive analyses for the years 2008 and 2009 are
expected to be carried out.
Acknowledgements
129
D.Sc. Erik Sonne Ravn from the Danish Maritime Safety
Administration in Copenhagen is thanked for his help in
evaluating the traffic flows in the Gulf of Finland on the basis of
129
the AIS data.
D.Sc. Jaros"aw Artyszuk from the Maritime University of
Szczecin in Poland is thanked for his invaluable comments on
ship motion model adaptation.
143 The authors also appreciate the financial contributions of the
following entities: the EU, Baltic Sea Region (this research was
carried out within EfficienSea project), the city of Kotka, and the
Finnish Ministry of Employment and the Economy.
150
References
[1] Fujii Y, Yamanouchi H, Mizuki N. On the fundamentals of marine traffic
control. Part 1: probabilities of collision and evasive actions. Electronic
Navigation Research Institute Papers 1970;2:1–16.
[2] Fujii Y, Yamanouchi H, Mizuki N. Some factors affecting the frequency of
accidents in marine traffic. II—the probability of stranding and III—the effect
of darkness on the probability of collision and stranding. Journal of
Navigation 1974;27:239–47.
[3] MacDuff T. The probability of vessel collisions. Ocean Industry 1974:144–8.
[4] Papanikolaou A. Risk-based ship design. Berlin: Springer; 2009.
[5] Hänninen M, Kujala P. The effect of causation probability on the ship collision
analysis. Accidents probabilities in selected areas of the Gulf of Finland. In:
Weintrit A, editor. Marine navigation and safety of sea transportation.
London: Taylor & Francis; 2009. p. 267-72. Gdynia, Poland: Balkema; 2009.
267–72.
[6] Kujala P, Hänninen M, Arola T, Ylitalo J. Analysis of the marine traffic safety in
the Gulf of Finland. Reliability Engineering and System Safety
2009;94(8):1349–57.
[7] Otto S, Pedersen PT, Samuelides M, Sames PC. Elements of risk analysis for
collision and grounding of a RoRo passenger ferry. Marine Structures
2002;15(4–5):461–74.
[8] Pedersen PT. Collision and grounding mechanics. The Danish Society of Naval
Architects and Marine Engineers 1995:125–57.
[9] Søfartsstyrelsen. Risk analysis for sea traffic in the area around Bornholm.
Konges Lyngby: COWI; 2008.
[10] Pedersen P. Collision risk for fixed offshore structures close to high-density
shipping lanes. Proceedings of the Institution of Mechanical Engineers Part
M: Journal of Engineering for the Maritime Environment 2002;216(1):29–44.
[11] Fowler TG, Sørgråd E. Modeling ship transportation risk. Risk Analysis
2000;20(2):225–44.
[12] Gluver H, Olsen D. Ship collision analysis. Balkema: Rotterdam; 1998.
[13] Kaneko F. Methods for probabilistic safety assessments of ships. Journal of
Marine Science and Technology 2002;7:1–16.
[14] Tan B, Otay E. Modeling and analysis of vessel casualties resulting from
tanker traffic through narrow waterways. Naval Research Logistics
1999;46(8):871–92.
[15] Smalko Z, Smolarek L. Estimate of collision threat for ship routes crossing. In:
Proceedings of the marine traffic engineering conference 2009, Maritime
University of Szczecin, 2009.
[16] Eleye-Datubo AG, Wall A, Saajedi A, Wang J. Enabling a powerful marine and
offshore decision-support solution through bayesian network technique. Risk
Analysis 2006;26(3):695–721.
[17] Trucco P, Cagno E, Ruggeri Grande O. A bayesian belief network modelling of
organisational factors in risk analysis: a case study in maritime transporta-
tion. Reliability Engineering and System Safety 2008;93(6):845–56.
[18] Ren J, Jenkinson I, Wang J, Xu DL, Yang JB. An offshore risk analysis method
using fuzzy bayesian network. Journal of Offshore Mechanics and Arctic
Engineering 2009;131(4):41101–12.
[19] Hara K, Nakamura S. A comprehensive assessment system for the maritime
traffic environment. Safety Science 1995;19(2–3):203–15.
[20] Merrick JRW, van Dorp JR, Mazzuchi T, Harrald JR, Spahn JE, Grabowski M.
The Prince William sound risk assessment. Interfaces 2002;32(6):25–40.
[21] Merrick JRW, van Dorp JR, Blackford JP, Shaw GL, Harrald J, Mazzuchi TA. A
traffic density analysis of proposed ferry service expansion in San Francisco
Bay using a maritime simulation model. Reliability Engineering and System
Safety 2003;81(2):119–32.
[22] Gucma L. Evaluation of collision probability of ships in the Baltic Sea by means
of simulation model and statistical data. In: Terje A, Vinnem JE, editors. Risk,
reliability and societal safety. London: Taylor & Francis; 2007. p. 2689–92.
[23] Uluscu ÖS, Özbas- B, Altıok T, Or _I, Yılmaz T. Transit vessel scheduling in the
strait of Istanbul. The Journal of Navigation 2009;62(01):59–77.
[24] Hilgert H, Baldauf M. A common risk model for the assessment of encounter
situations on board ships. Deutsche Hydrographische Zeitschrift 1997;49(4):
531–42.
 Author's Personal
ARTICLE Copy
IN PRESS
J. Montewka et al. / Reliability Engineering and System Safety 95 (2010) 573–589
[25] Tran T, Harris C, Wilson P. A vessel management expert system. Proceedings
of the Institution of Mechanical Engineers, Part M: Journal of Engineering for
the Maritime Environment 2002;216(2):161–77.
[26] Sz"apczyński R. A unified measure of collision risk derived from the concept
of a ship domain. Journal of Navigation 2006;59(03):477–90.
[27] Curtis R. A ship collision model for overtaking. Journal of Navigation
1986;37(04):397–406.
[28] Endoh S. Aircraft collision models. Master’s Thesis. Massachusetts Institute of
Technology, Cambridge, 1982.
[29] Pietrzykowski Z, Uriasz J. The ship domain—a criterion of navigational safety
assessment in an open sea area. Journal of Navigation 2009;62(01):93–108.
[30] Fujii Y, Yamanouchi H, Matui T. Survey on vessel traffic management systems
and brief introduction to marine traffic studies. Electronic Navigation
Research Institute Papers 1984;45:1–48.
[31] Kao S, Lee K, Chang K, Ko M. A fuzzy logic method for collision avoidance in
Vessel Traffic Service. The Journal of Navigation 2007;60(01):17–31.
[32] Fujii Y, Yamanouchi H, Tanaka K, Yamada K, Okuyama Y, Hirano S. The
behaviour of ships in limited waters. Electronic Navigation Research Institute
Papers 1978;19:1–14.
[33] Goodwin E. A statistical study of ship domain. The Journal of Navigation
1975;28:328–44.
[34] Artyszuk J. Ship motion mathematical model analysis, in aspect of benefits
for Integrated Navigation System algorithms. PhD thesis, Maritime University
of Szczecin, Szczecin, 2000.
[35] Brix J. Manoeuvring technical manual. Hamburg: Seehafen Verlag; 1993.
589
[36] Molland AF, Turnock SR. Marine rudders and control surfaces. Oxford:
Butterworth-Heinemann; 2007.
[37] Lee G, Surendran S, Kim S. Algorithms to control the moving ship
during harbour entry. Applied Mathematical Modelling 2009;33(5):
2474–90.
[38] Soung YJ, Rhee K. New prediction method on the stopping ability of diesel
ships with fixed pitch propeller. International Shipbuilding Progress
2005;52(2):113–28.
[39] Artyszuk J. Ship manoeuvring auxiliary differential equations for optimal
sea trials programme design.In: Proceedings of the marine traffic engi-
neering conference 2005. Swinoujscie, Poland: Maritime University of
Szczecin; 2005.
[40] Bertram V, Schneekluth H. Ship design for efficiency and economy, 2nd ed.
Oxford: Butterworth-Heinemann; 1998.
[41] Dudziak J. Ship theory. Gdansk: Fundacja Promocji POiGM; 2008 [in Polish].
[42] Abkowitz M. Stability and motion control of ocean vehicles organization,
development, and initial notes of a course of instruction in the subject.
Cambridge: Massachusetts Institute of Technology Press; 1969.
[43] I.M.O. SOLAS: International Convention for the Safety of Life at Sea. London:
International Maritime Organization; 2003.
[44] Inoue K, Kawase M. Innovative probabilistic prediction of accident occur-
rence. In: Marine navigation and safety of sea transportation. London: Taylor
& Francis; 2007. p. 31–4.
[45] Berglund R, Huttunen M. Analysis of crossing ship traffic in the Gulf of
Finland. Espoo: VTT Technical Research Centre of Finland; 2008. [in Finnish].