International Journal

on Marine Navigation

and Safety of Sea Transportation

Volume 6

Number 3

September 2012

381

1 INTRODUCTION

In recent years, the volume of maritime traffic has

significantly increased in the Gulf of Finland, espe-

cially because of the expansion of the Russian oil

exports from harbors such as Primorsk and

Vysotskiy. Up to the recent economic recession, the

volume of oil exported from Russia has increased

every year, and it is expected to keep increasing in

the future (Kuronen et al. 2008, Helcom 2010). With

this increasing traffic density, inherent risks such as

oil spills are of special concern due to the highly

vulnerably marine ecosystem of the Gulf of Finland

(Helcom 2010).

Analysis of historic shipping accidents show that

worldwide, groundings, collisions and fires are the

most common accident types (Soares 2001), while in

the shallow, island-littered waters of the Gulf of Fin-

land, groundings and ship-ship collisions are most

frequent (Kujala et al. 2009). This justifies the con-

cern of this paper with the risk of oil tankers in-

volved in ship-ship collision accidents.

The main driving idea of the model presented in

this paper is the societal trend towards science-based

risk-informed decision making, an idea supported by

organizations such as the IMO or IALA. In the mari-

time field, the Formal Safety Assessment provides a

framework for this aim.

2 OUTLINE OF RISK ASSESSMENT

METHODOLOGY

The risk assessment methodology is rooted in the

commonly accepted framework of the Formal Safety

Assessment (FSA) (Kontovas and Psaraftis, 2005).

The conceptual FSA-methodology is shown in Fig.

1. It starts with an identification of hazards, followed

by an analysis of the risk. Thereafter, risk control

options are defined, the effect of which should be

evaluated using the risk analysis method.

This should be followed by a cost-benefit analysis

and recommendations as to which risk control op-

tions to implement. It is therefore essential that the

risk analysis methodology is able to provide a relia-

Simplified Risk Analysis of Tanker Collisions in

the Gulf of Finland

F. Goerlandt, M. Hanninen, K. Stahlberg, J. Montewka & P. Kujala

Aalto University, School of Science and Technology, Department of Applied Mechanics,

Espoo, Finland

ABSTRACT: Maritime traffic poses various risks in terms of human casualties, environmental pollution or

loss of property. In particular, tankers pose a high environmental risk as they carry very large amounts of oil

or more modest amounts of possibly highly toxic chemicals. In this paper, a simplified risk assessment meth-

odology for spills from tankers is proposed for the Gulf of Finland, for tankers involved in a ship-ship colli-

sion. The method is placed in a wider risk assessment methodology, inspired by the Formal Safety Assess-

ment (FSA) and determines the risk as a combination of probability of occurrence and severity of the

consequences. The collision probability model is based on a time-domain micro simulation of maritime traf-

fic, for which the input is obtained through a detailed analysis of data from the Automatic Identification Sys-

tem (AIS). In addition, an accident causation model, coupled to the output of the traffic simulation model is

proposed to evaluate the risk reduction effect of the risk control options. Further development of the model is

needed, but the modular nature of the model allows for continuous improvement of the modules and the ex-

tension of the model to include more hazards or consequences, such that the effect of risk control options can

be studied and recommendations made. This paper shows some preliminary results of some risk analysis

blocks for tanker collisions in the Gulf of Finland.

382

ble evaluation of the effect of the risk reducing

measures.

The system risk is defined based on the definition

of Kaplan (1997) as a set of triplets:

{(s

, l

, c

)}, i=1, 2, 3,… (1)

Here, s

defines the context of the accident sce-

nario, l

the likelihood of the accident occurring in

that scenario and c

the evaluation of the conse-

quence in the scenario.

Fig. 1. General outline of FSA methodology

It is important to indicate that l

and c

are de-

pendent on the accident scenario s

, which is to be

seen as a multi-parameter set, i.e. a range of varia-

bles relevant to the evaluation of the accident proba-

bility l

and the consequence c

The risk analysis methodology is based on a sys-

tem simulation of the maritime traffic in a given ar-

ea. The overall flowchart, focusing on the risk of

ship collision, is shown in Fig. 2. The various mod-

ules of this model, insofar these are already availa-

ble, will be introduced below.

Fig. 2. General outline of FSA methodology

At present, the model is capable only to assess the

risk of ship-ship collision, which is the second most

important hazard in the Gulf of Finland, based on

the accident statistics of Kujala et al. (2009). The

methodology can in principle be extended without

too many difficulties to other accident types such as

ship grounding and fires.

3 TRAFFIC SIMULATION AND COLLISION

ENCOUNTER SCENARIO MODEL

The traffic simulation and collision encounter sce-

nario detection module is one of the core units of the

overall risk assessment model. The basic idea is to

simulate the traffic on a micro-scale. For each vessel

sailing in the area, the trajectory is simulated, while

assigning a number of parameters to this vessel.

These include departure time, ship type, length,

loading status, cargo type and ship speed, as illus-

trated in Fig. 3. The simulation of all vessels in the

area provides a traffic simulation and the subsequent

detection of the vessels which collide, assuming that

no evasive action is made, results in the definition

collision encounter scenarios.

Fig. 3. Generated data for each simulated vessel (traffic event)

The input for this model is taken from data from

the Automatic Identification System (AIS), aug-

mented with statistical data from harbors concerning

the traded cargo types. Details on how this simula-

tion and collision candidate detection is performed,

is given in Goerlandt and Kujala (2011).

As an illustration of the input for the simulation

model, Fig. 4 shows the departure time distribution

for vessels sailing from Helsinki to Tallinn. Fig. 5

shows the ship length distribution for tankers to

Sköldvik and Primorsk. Fig. 6 shows the average

ship speed distributions for all considered ship types.

This information is used as a first estimate of the

ship speed before the collision candidates are ob-

tained. After detection of a collision candidate in a

specific area, the speed is resampled from ship type

specific speed distributions by location, as shown in

Fig. 8. This speed is then updated in the collision

encounter scenario.

Table 1 shows the harbor-specific data for cargo

types of chemical tankers, for the port of Hamina.

The cargos carried by the simulated vessels are sam-

pled from this information, after a more in-depth

analysis of which trade routes represent which cargo

types. Fig. 7 shows the simulated traffic in the Gulf

of Finland, based on the input obtained from AIS.

383

Fig. 4. Departure time distributions, traffic from Helsinki to

Tallinn

Fig. 5. Length distribution of tankers to Sköldvik and Primorsk

Fig. 6. Speed distributions of vessels in the Gulf of Finland

In Table 2, an example of output obtained from

the collision encounter simulation model is shown.

This is to be interpreted as the accident scenario con-

text using the definition of Kaplan (1997) as pre-

sented in Section 2.

Fig. 7. Simulated traffic for one year

map: © Merenkulkulaitos lupa nro 1321 / 721 / 200 8

Fig. 8. Average speed of tankers in the Gulf of Finland and lo-

cal speed distributions, based on AIS data of 2006-2009

map: © Merenkulkulaitos lupa nro 1321 / 721 / 200 8

Table 1. Example of data concerning harbor-specific trade vol-

ume: port of Hamina, Finland (Hänninen and Rytkonen, 2006)

__________________________________________________

IMPORT PRODUCTS

Product Vol. [ton] Product Vol. [ton]

__________________________________________________

Butadiene 53926 Sulphuric acid 39492

Buthyl acrylate 12233 Styrene monomer 3380

Phenol 1038 Vinyl acetate 1457

Caustic Soda 78547 Methyl ketone 501

__________________________________________________

EXPORT PRODUCTS

Product Vol. [ton] Product Vol. [ton]

__________________________________________________

Butane 741 Methyl-butyl ether 83104

Isoprene 8271 Nonylphenol 48830

Methanol 762012 Propane 2839

Styrene monomer 9602 Vinyl acetate 457

Propylene 5897

__________________________________________________

COMMON ORIGINS COMMON DESTINATIONS

__________________________________________________

St. Petersburg Rotterdam, Antwerpen,

Teesport, Hamburg, Gdynia

__________________________________________________

Table 2. Examples of encounter scenarios obtained by the

model of Goerlandt and Kujala (2011)

__________________________________________________

Location Time Origin Type‡ Speed

[long | lat] [m.h:m] Struck Striking [kn]

__________________________________________________

24.60|59.82 01.05:10 Hamina C P V

loc

†

22.31|59.34 03.08:47 Sköldvik GC GC V

loc

27.96|60.17 03.13:32 Kotka OT GC V

loc

24.10|59.55 04.21:10 St. Petersb GC OT V

loc

25.23|57.53 06.09:05 Vyborg P GC V

loc

29.11|59.95 07.14:13 St. Petersb GC GC V

loc

__________________________________________________

† V

loc

is the local speed distribution for the relevant ship types

‡ Type: C = chemical tanker, P = passenger vessel, GC = gen-

eral cargo ship, OT = oil tanker

384

4 COLLISION SCENARIO AND WEATHER

MODEL

While the collision encounter scenario model is able

to partly define the accident context, this is insuffi-

cient to accurately define either the likelihood of the

accident l

or the consequences c

As a first concern, it should be noted that an en-

counter scenario, which depends only on the nature

of the maritime traffic flows, is not equivalent to the

actual collision scenario. In particular, due to possi-

ble evasive maneuvers made prior to collision, es-

sential parameters such as vessel speed and collision

angle may deviate significantly from the encounter

conditions. This has an important effect on the eval-

uation of the consequences c

, as can be evaluated by

inspecting the collision energy models of Zhang

(1999) or Tabri (2010).

Several authors have proposed models for the pa-

rameters relevant to the collision scenario, usually

based on accident statistics. Some of these proposals

are briefly described in Table 3. However, at present

no reliable model exists linking the encounter sce-

nario and the collision scenario. This has been inves-

tigated by Goerlandt et al. (2011) using a compari-

son of the hull breach probability for various

collision scenario models, based on a collision ener-

gy model by Zhang (1999) and a criterion for the

critical energy the ship hull can withstand before

breach of the double hull. The results of the local

probability of oil spill resulting from the various col-

lision scenarios from Table 3, is shown in Fig. 9.

Table 3. Impact scenario models available in literature

__________________________________________________

Impact model by Rawson (1998)

__________________________________________________

Collision angle: U(0,180)

striking

: Truncated bi-normal N(5,1) | N(10,1)

struck

: Idem as V

striking

Collision location: U(0,180)

__________________________________________________

Impact model by NRC (2001)

__________________________________________________

Collision angle: N(90,29)

striking

: W(6.5, 2.2)

struck

: E(0.584)

Collision location: B(1.25,1.45)

__________________________________________________

Impact model by Lützen (2001)

__________________________________________________

Collision angle: T(0, α

enc

, 180)

striking

: Below .75V

enc

: U(0, .75V

enc

)

Above .75V

enc

: T(.75 V

enc

, V

enc

)

struck

: T(0, V

enc

)

Collision location: Empirical distribution, see Lützen (2001)

__________________________________________________

† U: uniform | N: normal | W: weibull | E: exponential | B: beta

| T: triangular distribution

For a proper formulation of the accident context,

a weather model, capable of predicting the factors

which are needed in the evaluation of the accident

likelihood and consequences, is needed as well.

These factors include wind velocity, sea state and

visibility. At present, this weather simulation mod-

ule has not been implemented in the presented mari-

time accident assessment methodology.

In terms of the parameters defining the accident

context, denoted s

in the formulation of Kaplan

(1997), the weather model adds certain parameters

to the values obtained from the collision scenario

model, as given in Table 2. These weather-related

factors affect the likelihood of the accident l

and the

effectiveness of response to oil spill.

The collision scenario model adds certain param-

eters such as which is the striking and struck ship

and the location of the collision along the struck ship

hull. In addition, this model should modify certain

parameters such as the collision angle and vessel

speed of striking and struck vessel, which have an

important contribution to the consequence assess-

ment, i.e. c

in the Kaplan-nomenclature of Eq. 1.

Fig. 9. Results of location-specific spill probability according

to algorithm in Fig. 9 and (Eq. 4), impact models: see Table 5,

map: © Merenkulkulaitos lupa nro 1321 / 721/ 200 8, taken

from Goerlandt et al. (2011).

385

5 ACCIDENT CAUSATION MODEL

The accident causation model gives a probability of

a collision accident occurring in a given context, in

terms of the system risk definition by Kaplan (1999),

this is the scenario specific likelihood of accident l

This accident causation module from Fig. 2 is

constructed using the methodology of the Bayesian

Belief Network (BBN). The model is shown in Fig.

10, and is discussed in more detail in (Hänninen and

Kujala 2009, Hänninen and Kujala 2011).

The model is rooted in expert opinion, accident

and incident data, with the understanding that some

parameters are taken directly from the output of the

simulation model, in particular the traffic encounter

scenario model and the weather model. For instance,

the values for the nodes for encounter type, ship

types and sizes, time of year, daylight condition and

whether or not the encounter location is in a VTS ar-

ea can be derived from the encounter scenario mod-

ule as explained in Section 3. The visibility and

weather conditions could be derived from the weath-

er model as discussed in Section 4.

Table 4 gives an overview of the groups of nodes

in the Bayesian Network, giving a number of exam-

ples of some nodes in these groups. The parameters

which are directly taken from the traffic and weather

simulation models are marked in italics.

The accident causation model is an important el-

ement in the study of the risk control options, as dis-

cussed in more detail in Section 7.

Table 4. Node groups in the Bayesian model with examples

__________________________________________________

Visual detection Management factors

__________________________________________________

Visibility Safety culture

Other ship size Maintenance routines

Bridge view Bridge resource management

Daylight

__________________________________________________

Navigational aid detection Human factors

__________________________________________________

Radar detection Stress

AIS installed Competence

AIS signal on radar screen Situational assessment

Collision avoidance alarms Familiarization

__________________________________________________

Support Evasive actions / overall

__________________________________________________

VTS vigilance Encounter type

Pilot vigilance Give way situation

Other internal vigilance Time of year

Weather

Ship type

__________________________________________________

Technical reliability

__________________________________________________

Steering failure

Radar functionality

AIS functionality

__________________________________________________

Fig. 10. Causation probability model

386

6 HULL BREACH PROBABILITY AND SPILL

SIZE

In terms of collision consequences c

, the focus of

this paper is limited to the probability of spills from

oil tankers. The environmental or socio-economic

damage or implications for oil combating operations

is at present not considered.

The hull breach probability can be determined

based on a comparison of the available deformation

energy in the collision scenario, compared to the en-

ergy which the ship structure can withstand before

the inner hull is breached.

For the available deformation energy, a number

of models is available. Zhang (1999) proposed a rel-

atively simple analytical model, assuming rigid bod-

ies and 2-dimensional ship motions. Brown (2002)

proposed a simplified model taking the interaction

between inner mechanics (i.e. the structural defor-

mation) and the outer mechanics (i.e. the ship mo-

tions in a collision) into account, limited to 2-

dimensional ship motions. Tabri (2010) proposed a

full 6 degrees of freedom model, coupling inner and

outer mechanics and taking the sloshing of liquids in

a tank into account.

For the ship structural energy, methods such as

finite element calculations, e.g. as proposed by Eh-

lers (2010) could be used. In Goerlandt et al. (2011),

a simple criterion based on regression of available

ship structural data is proposed.

The methodology to compare the available de-

formation energy and the hull structural strength is

outlined in detail in Goerlandt et al. (2011). Also the

work by Klanac et al. (2010) uses a variation of this

approach to assess the hull breach probability.

For the oil spill size, a number of models has

been proposed in the literature. Examples are a

probabilistic extension of the IMO-tanker design cri-

teria as proposed by Montewka et al. (2010). A re-

lated methodology has been proposed by Smailys

and Cesnauskis (2006). A simple oil volume outflow

model based on statistics of tank sizes has been pro-

posed by Gucma and Przywarty (2007). While these

models have their merits, they are very simplified

and do not take the detailed information from the ac-

cident scenario si into account.

On the other hand, the model proposed by van de

Wiel and van Dorp (2009) is capable of predicting

the size of both a cargo oil spill and of a bunker oil

spill using a number of variables determined in the

collision accident scenarios. Such variables are the

vessel sizes, speeds, collision angles and collision

location along the struck ship hull. Thus, this model

provides a good match with the information of the

accident scenario information si. However, as indi-

cated in Section 4, there exists a significant uncer-

tainty concerning the validity of the available mod-

els for the collision scenarios.

The model is based on a combination of collision

energy calculations, used to determine the damage

length and width, and a limited reference ship data-

base. Based on this information, it is assessed

whether or not the hull is breached, and in the case it

is, how much oil will flow out of the ship. The mod-

el can also be used for estimating the spill in case of

grounding.

Since chemical tankers have a significantly dif-

ferent structural arrangement, the above mentioned

methods can not directly be used for estimation of

spill sizes of this vessel type. Also consequence

evaluation based on structural damage for other ves-

sel types is at present not available. However, the

principle behind the methods proposed by Ehlers

(2010) and Tabri (2010) can be used to get reliable

results for these accident types.

7 OVERALL RISK ASSESSMENT:

APPLICATION

The application of the risk assessment methodology

is to be done by modifying the values for the risk

control options in the model to evaluate the effect on

the risk level. With an estimate of the cost of imple-

mentation of the risk control options and the saved

cost due to the reduced risk, an informed decision

can be made.

A number of risk control options are related to the

accident likelihood l

. For instance, the VTS vigi-

lance, pilot vigilance, safety culture, navigator com-

petence, navigational equipment and aids to naviga-

tion are taken into account in the accident causation

model, as described in Section 5. Also the ship rout-

ing affects the accident likelihood, which in princi-

ple can be studied by modification of the traffic

streams in the traffic simulation model, resulting in

less and/or safer encounters.

Other risk control options affect the severity of

the consequences in case of an accident. Examples

of these are the speed limits in local sea areas and

the encounter situation, which directly affect the

available collision energy. Also the structural

strength of the ship hull is an important factor in the

severity of the consequences. The accident response

effectiveness in terms of number, location and

equipment of the available oil response or search

and rescue fleet, can also be studied based on the

risk maps produced in the risk analysis step.

It should be noted in this context that estimating

the accident costs is a difficult task in itself due to

the highly complex nature of the studied system. For

instance, for an oil spill due to collision, apart from

the spill size, the ecological and socio-economic

387

value of the environment in which the spill may oc-

cur, should be considered.

It may therefore be more feasible to study the rel-

ative risk reduction of the measures as such, and

comparing these to the costs of the risk control op-

tions. This will also lead to a decent risk-informed

decision, if a certain expertise is available to inter-

pret the results of the risk assessment.

8 CONCLUSION AND FUTURE WORK

It should be clear that the evaluation of the risks re-

lated to maritime traffic in the framework of a For-

mal Safety Assessment is a very wide and laborious

task, not in the least because of the multidisciplinary

nature of the studied system. Such fields as logistics,

maritime engineering, systems analysis, operations

research and environmental modeling should be

combined in an overall FSA-framework.

The maritime system simulation methodology

starts from the premise that the likelihood and con-

sequence of each relevant accident type can be cal-

culated based on situational information, as suggest-

ed by Kaplan (1997). The aim of determining each

of the modules building up the model for maritime

system risk in a scientifically sound manner is to be

seen as an attempt to rationalize the decision making

process in risk related matters.

It is clear that even though the scope of the cur-

rent model is rather limited (only the probability of

collision of ships in open waters and the conse-

quences in terms of oil spill size are included as yet),

and even within these models certain improvements

could be made (e.g. the collision scenario model

linking encounter scenario to actual impact condi-

tions), the modular nature of the model allows for

gradual improvement and extension of the models to

include additional hazards, risk analysis blocks or

risk control options.

Consequently, the remaining work is still very

significant before any proper conclusions can be

made. Firstly, other hazards (ship grounding, fire)

should be included. Secondly, a weather model

should be coupled to the accident scenario genera-

tion. Thirdly, for ship collisions, the consequences

for other ship types (chemical tankers, passenger

vessels), should be determined in terms of economic

loss due to structural damage or loss of human life.

There is also significant work to be done in the un-

derstanding of accident causation, and for various

accident types, there is a lack of consequence mod-

els.

ACKNOWLEDGEMENTS

The authors appreciate the financial contribution of

the European Union and the city of Kotka. This re-

search is carried out within the EfficienSea project

and in association with the Kotka Maritime Research

Centre.

REFERENCES

Brown, A.J., 2002. Collision scenarios and probabilistic colli-

sion damage. Marine Structures, 15(4-5):335-364.

Ehlers, S. 2010. Material relation to assess the crashworthiness

of ship structures. Doctoral Dissertation, Aalto University,

School of Science and Technology.

Goerlandt, F., Kujala, P. 2011. Traffic simulation based ship

collision probability modeling. Reliability Engineering and

System Safety doi:10.1016/j.ress.2010.09.003.

Goerlandt, F., Ståhlberg, K., Kujala, P. 2011. Comparative

study of input models for collision risk evaluation. Ocean

Engineering – manuscript under review.

Gucma L., Przywarty M.: The Model of Oil Spills Due to Ships

Collisions in Southern Baltic Area. TransNav - Internation-

al Journal on Marine Navigation and Safety of Sea Trans-

portation, Vol. 2, No. 4, pp. 415-419, 2000

Hänninen M., Kujala P.: The Effects of Causation Probability

on the Ship Collision Statistics in the Gulf of Finland.

TransNav - International Journal on Marine Navigation and

Safety of Sea Transportation, Vol. 4, No. 1, pp. 79-84, 201

Hänninen, M. & Kujala, P. 2011. Influences of variables on

ship collision probability in a Bayesian belief network

model. In prep.

HELCOM. 2010. Maritime Activities in the Baltic Sea – An in-

tegrated thematic assessment on maritime activities and re-

sponse to pollution at sea in the Baltic Sea Region. Balt.

Sea Environ. Proc. No. 123.

Kaplan, S., 1997. The Words of Risk Analysis. Risk Analysis,

17(4), 407-417.

Klanac, A., Duletic, T., Erceg, S., Ehlers, S., Goerlandt, F.,

Frank, D. 2010. Environmental risk of collision for en-

closed seas: Gulf of Finland, Adriatic and implications for

tanker design. 5th International Conference on Collision

and Grounding of Ships, Espoo.

Kontovas, C., Psarafitis, H. 2005. Formal Safety Assessment –

Critical review and future role. Msc. Thesis, National Uni-

versity of Athens.

Kujala, P., Hänninen, M., Arola, T., Ylitalo, J. 2009. Analysis

of the marine traffic safety in the Gulf of Finland. Reliabil-

ity Engineering and System Safety 94(8):1349-1357.

Lützen, M., 2001. Ship collision damage. PhD thesis, Tech-

nical University of Denmark.

Montewka, J., Ståhlberg, K., Seppala, T., Kujala, P. 2010. El-

ements of risk analysis for collision of oil tankers. In G.

Soares (Ed.), ESREL 2010, London. Taylor and Francis.

National Research Council (NRC), 2001. Environmental Per-

formance of Tanker Designs in Collision and Grounding.

Special Report 259, The National Academies Press.

Smailys, V., Cesnauskis, M. 2006. Estimation of expected car-

go outflow from tanker involved in casualty. Transport Vil-

nius 21(4):293-300.

Tabri, K. 2010. Dynamics of ship collisions. Doctoral Disser-

tation, Aalto University, School of Science and Technology.

van de Wiel, G., van Dorp, J.R. 2009. An oil outflow model for

tanker collisions and groundings. Annals of Operations Re-

search: doi: 10.1007/s10479-009-0674-5.

Zhang, S., 1999. The mechanics of ship collisions. PhD thesis,

Technical University of Denmark.