Development of a quantitative risk assessment model for ship collisions in fairways

Safety Science 91 (2017) 71–83

Contents lists available at ScienceDirect

Safety Science journal homepage: www.elsevier.com/locate/ssci

Development of a quantitative risk assessment model for ship collisions in fairways Tian Chai a,b, Jinxian Weng c,⇑, De-qi Xiong a a

Environmental Science and Engineering College, Dalian Maritime University, Dalian 116026, China Navigation Institute, Jimei University, Xiamen 361021, Fujian, China c College of Transport and Communications, Shanghai Maritime University, Shanghai 201306, China b

a r t i c l e

i n f o

Article history: Received 28 October 2014 Received in revised form 30 June 2016 Accepted 20 July 2016

Keywords: Ship collision Frequency Quantitative risk assessment

a b s t r a c t This study develops a quantitative risk assessment (QRA) model to evaluate the risk of ship being involved in ship collisions which takes into account the frequency and consequence of all possible accident scenarios. Two accident consequence types including human life loss and oil pollution which is measured in terms of the volume of oil spilled are considered in this study. The proposed QRA model consists of a collision frequency estimation model, an event tree and consequence estimation models. The event tree comprises five intermediate events including ship type, ship size, loading condition, hull damage and survivability. Two ‘‘generic” mathematic models are developed to estimate the human life loss and oil pollution caused by ship collisions, respectively. A case study is finally created using the real-time ship movement data in the Singapore Strait from the Llyod’s Marine Intelligence Unit (Lloyd’s MIU) database. Results show that the container ship, bulk carrier and oil tanker are the three main ship types being involved in collision accidents. Although the passenger/RORO ship has the lowest frequency being involved in collisions, it will suffer the most serious consequence in terms of the human life loss once it is involved in an accident. Considering the relative high percentage of oil tankers involving in ship collisions and their severe consequences, focus should be placed on the tracking and management of oil tanker traffic. Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction Because of the increase of transportation demand, the marine traffic increases each year in many parts of the world and is expected to increase significantly over the next few decades. However, the significant increase in traffic demand could result in increased traffic movements in fairways. In general, the number of traffic movements in a busy fairway can be as high as 2000 per day (Yip, 2008) and this number is expected still to increase with the continuing growth of traffic demand. Such a large number of movements may result in high traffic density and the increase in the accident occurrence likelihood, especially in the busy fairway (Yip, 2008; Qu et al., 2011). Therefore, a number of maritime traffic control strategies have been implemented in order to enhance the navigational safety. For example, the Marmite Port and Authority of Singapore (MPA) has adopted the traffic separation scheme (TSS) to enable safer navigation in the Singapore Strait since 1981 (Qu et al., 2011). However, the effectiveness of these strate⇑ Corresponding author. E-mail address: [email protected] (J. Weng). http://dx.doi.org/10.1016/j.ssci.2016.07.018 0925-7535/Ó 2016 Elsevier Ltd. All rights reserved.

gies on the risk mitigation is still not known perfectly. Risk assessment is a key step toward evaluating the effectiveness of risk mitigation measures in the Singapore Strait. Although many researchers (e.g., Macduff, 1974; Fujii et al., 1974; Szlapczynski, 2006; Pietrzykowski, 2008; Wang, 2010; Qu et al., 2011) have developed various models for the navigational risk assessment, the majority of these models emphasized on the estimation of the occurrence frequency of navigational accidents. Obviously, it is inadequate to comprehensively assess the navigational risk only by evaluating the occurrence likelihood of navigational accidents. This is because the navigational risk is rendered by a broad range of accidents from frequent-minor to rare-major accidents. The quantitative risk assessment (QRA) technique is a formal and systematic approach to estimating the likelihood and consequences of hazardous events, and expressing the results quantitatively as risk to people or the environment. In other words, the QRA not only can provide an overall risk assessment but also has the capability of describing the relationship between the occurrence frequency and consequence (Si et al., 2012; Liwang et al., 2013). A large number of existing research works have

72

T. Chai et al. / Safety Science 91 (2017) 71–83

reported that ship collisions account for a substantial portion of the major types of accidents in fairways (Goossens and Glansdorp, 1998; Akten, 2004). For example, it was reported that a proportion of 45.5% ship accidents were collisions in the Bosphorus from 1953 to 2002 (Akten, 2004). In addition, the consequence of ship collisions in the fairway can be catastrophic for oil tankers and chemical/LNG/LPG ships which may cause serious environmental pollution and human life loss. For example, on 30 May 2010, an oil tanker and a bulk carrier collided in the eastern part of Singapore Strait, spilling an estimated 2500 tons of oil. Therefore, we are concerned about the ship collision risk in this study.

2. Literature review A number of studies have been conducted on the navigational risk assessment in the past. However, most previous studies emphasized on the estimation of the occurrence frequency of ship collisions, which is defined as the number of ship conflicts multiplied by a causation probability (Macduff, 1974). Since the causation probability for distinct water areas is assumed to be a constant under a particular accident scenario, the major focus has been placed on the estimation of the number of ship conflicts. Various mathematical models, such as ship domain models (Macduff, 1974; Tan and Otay, 1999; Fowler and Sorgard, 2000; Szlapczynski, 2006; Pietrzykowski and Uriasz, 2009; Wang, 2010) and speed dispersion model (Qu et al., 2011), have been proposed to evaluate the ship collision frequency. In addition, the computer simulation-based approach has been also applied to quantitatively examine various navigational safety issues. For example, Dand (2001) introduced the Permanent International Association of Navigation Congresses (PIANC) simulation approach to the water channel design. Zhang and Huang (2006) developed ship models to acquire pilot experience using the computer simulation approach. However, the simulation method is very time-consuming. It may also give rise to biased or inaccurate results because of a lack of practical criteria and incorrect interpretations of rules and seamanship. Compared with the collision frequency estimation, the literature on the collision consequence assessment is rather limited. Both mechanical models and simulation methods have been applied to estimate the ship collision consequence. For example, Servis and Samuelides (1999) analyzed the damages to the struck ship by using finite element techniques. This method can be used to assess the ship behavior under a specific collision scenario, and also to compare the survivability of different structure arrangement. It should be pointed out that the severity of the damage to the ship caused by the impact and the volume of oil spilled depend on the ship type, ship size, loading condition, hull damage and survivability (Van de Wiel and Van Dorp, 2009). Nevertheless, it should be pointed out that it is far not enough to comprehensively assess the navigational risk only by evaluating the occurrence frequency or the consequence of ship collisions. This is because there are a number of possible accident scenarios with distinct occurrence frequency and consequence once an accident is occurred. The quantitative risk assessment (QRA) technique is a formal and systematic approach to estimating the likelihood and consequences of hazardous events, and expressing the results quantitatively as risk to people or the environment. It is able analyze the potential accident scenarios, including consequences and initiating and controlling factors. In addition, the QRA model can provide an overall risk assessment via describing the relationship between the accident occurrence frequency and consequence (Meng et al., 2010).

3. Objectives and contributions The objective of this study is to develop a QRA model in order to assess the ship collision risk including the frequencies and consequences of ship collisions. To achieve this objective, it is required to estimate the frequency and consequence of all possible accident scenarios. A collision frequency estimation model is first proposed to evaluate the frequency of a ship being involved in collisions. To reflect various accident scenarios triggered by a collision, an event tree comprising five possible intermediate events related to the collision is then built. Based on the event tree, the occurrence frequency of a particular accident scenario and its consequence can be evaluated. A case study is finally created using the real-time ship movement data from the Lloyd’s Marine Intelligence Unit’s (Lloyd’s MIU) automatic identification system (AIS). The contribution of this study is twofold. First, the proposed QRA model can be further embedded into a ship traffic simulation model which is used to generate ship movement trajectories in the future. The combination of the proposed QRA model and ship traffic simulations could be able to check the viability of new navigational safety strategy which will be implemented in fairways, by identifying whether both the estimated accident frequency and consequence exceed the acceptable risk levels or not. Second, the proposed QRA model can help shipping companies identify the risky areas of the fairway. 4. Quantitative risk assessment model formulation Once a ship is involved in a ship collision, there will be a number of possible scenarios with distinct consequences. These possible scenarios can be logically illustrated by a tree diagram in which all possible paths following a top node can be traced. The occurrence frequency of a particular scenario hence equals to the product of ship collision frequency and the occurrence likelihood of this scenario. In this study, two types of consequences are considered, including human life loss and environmental pollution which is measured in terms of the volume of oil spilled. Therefore, the major task of the model formulation is to estimate the ship collision frequency, build event trees and consequence estimation models. The output from the ship collision frequency estimation model will be used as the input to the event tree and consequence estimation models. 4.1. Ship collision frequency estimation According to the previous studies (e.g., Fujii et al., 1974; Mou et al., 2010), the ship collision frequency is equal to the number of ship conflicts multiplied by the probability of failing to avoid a collision for a given ship conflict. Namely,

f collision ¼ Nconflict  pcausation

ð1Þ

where f collision = the ship collision frequency; N conflict = the number of ship conflict; pcausation = the probability of failing to avoid a collision for a given ship conflict. Montewka et al. (2010) used the minimum distance to collision (MDTC) to determine the ship conflicts. In general, a ship conflict can be defined as an overlap of two ship domains, as shown in Fig. 1(a). Hereafter, the ship domain is expressed as the area around the ship that the navigator wants to keep clear of other ships or objects. It should be pointed out that the overlap of two ship domains is equivalent to an event in which a point representing the center of one ship enters the disc of which the radius equals

73

T. Chai et al. / Safety Science 91 (2017) 71–83

Ri

Rj

Ship i

Ship j

Ship i

R = Ri + R j

Ship i

(a)

(b) Fig. 1. Definition of ship conflict.

the sums of radii of two original discs shown in Fig. 1(b). One ship conflict example is shown in Fig. 2. It can be seen that Ship B is going to intrude into the disc at time t + DT. From Fig. 3, whether the ship i has a conflict with the ship j or not at time t can be determined by

( Iði; j; tÞ ¼

Therefore, the number of ships of type m which are involved ship conflicts during the entire analysis time duration T can be calculated by

Nm conflict ¼

NðtÞ T X X X X ðIðStype;i ¼ mÞ Iði; k; tÞÞ þ ðIðStype;j ¼ mÞ t¼0 k¼1

1; if 0 < 0;

Ltij

NðtÞ T X X  Iðk; j; tÞÞ

!

6 jV tij j  DT

ð2Þ

otherwise

where DT is the time interval,Iði; j; tÞ = 1 represents that the ship i has a conflict with the ship j from the time t to t þ DT, Ltij is the max!

imum sailing distance from the time t to t þ DT, and jV tij j is the relative speed of ship i over ship j from the time t to t þ DT. The !

detailed calculations for Ltij and jV tij j are given in Appendix A.

ð3Þ

t¼0 k¼1

where IðStype;i ¼ mÞ is 1 if the ship i belongs to the type m and it is equal to zero otherwise, IðStype;j ¼ mÞ is 1 if the ship j also belongs to the type m and it is equal to zero otherwise, N(t) is the number of ship conflicts from the time t to t þ DT. Note that the probability of failing to avoid a collision varies with the ship conflict type. In general, ship conflicts can be divided into the following three conflict types based on the course difference of two ships:

sel B

Ves

(i) Overtaking conflict. An overtaking conflict is defined as the conflict in which two ships are proceeding on the same route, lying on almost parallel courses. According to Montewka et al. (2010), the course difference of an overtaking should not exceed 10°. (ii) Head-on conflicts. A head-on conflict is a conflict in which ships are lying on almost reciprocal courses, and the course differences falls in the range between 170° and 190°. (iii) Crossing conflict. A crossing conflict is defined as a conflict in which the difference between two ships’ courses falls in the range 10–170° or 190–350°.

Ļ :

The number of ship collisions caused by the ship type m, m denoted by f collision , can hence be expressed by v er m;head m;cros ov er head f collision ¼ Nm;o conflict  pcausation þ N conflict  pcausation þ N conflict

se

es

V

m

lB

 pcros causation

R Vess

el A

Ļ:

ð4Þ

v er where Nm;o conflict is the number of overtaking conflicts involving the

ov er is the causation probability for the overtaking ship type m; pcausation

conflicts; Nm;head conflict is the number of head-on conflicts involving the Vessel

A

ship type m; phead causation is the causation probability for the head-on conflicts; N m;cros conflict is the number of crossing conflicts involving the ship type m, and pcros causation is the causation probability for the crossing conflicts.

Domain of Vessel A

4.2. Event tree building

Fig. 2. A ship conflict example.

As mentioned, there will be a number of possible scenarios with different consequences for the ship suffering a collision. In general,

74

T. Chai et al. / Safety Science 91 (2017) 71–83

Fig. 3. Relationships among ship positions, ship speeds and ship conflicts.

the occurrence probability of a particular scenario may vary based on the factors such as the ship type, ship size, and the severity of hull damage. For example, a ship which has more crew members and carries more oils may lead to a bigger number of fatalities and larger volume of oil spilled once it is involved in a collision, as compared with another ship which has less oils and crew members. The event tree is able to determine all possible accident scenarios and their corresponding occurrence probabilities. It should be pointed out that only the factors which can significantly influence the occurrence probability and consequence can be used as intermediate nodes of an event tree. It was reported that the occurrence probability significantly differs from the ship type and ship size (Li et al., 2012) while the human life loss and environment pollution (in terms of oil spills) are greatly affected by the hull damage, ship survivability and loading condition (e.g., PerezLabajos et al., 2009). Therefore, five intermediate nodes were taken into account for the event tree building in this study, which are ‘‘Ship Type”, ‘‘Ship Size”, ‘‘Loading Condition”, ‘‘Hull Damage” and ‘‘Survivability”, respectively. In general, the event tree is investigated from left to right. Hence, the event tree is started at the column of ‘‘a ship is involved in collisions” (top event) and terminated at the ‘‘Survivability” column, as shown in Fig. 4. Due to space limits, the event tree is decomposed into two sub-event trees, namely sub event tree (a) and sub event tree (b). Sub event tree (b) will continue from the intermediate event of ship size in the sub event tree (a). In the ‘‘Ship Type” column, we divide the ships into six types including oil tanker, container ship, bulk carrier, general cargo ship, chemical/LNG/LPG ship and passenger/RORO ship. In the ‘‘Ship Size” column, the factor of ship size is categorized into three types: small, medium and large. According to IMO (2008c), the smallsized oil tanker is defined as the oil tanker with the gross tonnage (GRT) less than 70,000 tons. For the oil tankers, the medium size is defined as 70,000 tons < GRT < 150,000 tons and the large size is defined as GRT > 150,000 tons. The ‘‘Loading Condition” can be divided into two groups: loaded and ballast. The ‘‘Hull damage” is also divided into two types depending on the severity of accidents: minor and critical. Since the human life loss is greatly affected by the condition whether the ship is sunk or not. In addition, the sinking speed could also influence the human life loss. Therefore, three possible alternatives including ‘‘stay afloat”, ‘‘sink slowly” and ‘‘sink rapidly” are taken into account for the ‘‘Surviv-

ability” column in this study. In addition, two types of consequences caused by ship collisions can be described as: (i) human life loss and (ii) oil pollution which is measured in terms of the volume of oil spilled, shown in the ‘‘Consequence Description” column. A particular scenario depends on the path, which is represented by a specific sequence of intermediate events in the event tree. Hence, the occurrence probability of a given scenario can be simply determined by multiplying the probabilities/percentages of all intermediate events on that path. According to the event tree structure in Fig. 4, there are a total of 6  3  2  2  3 ¼ 216 possible scenarios for the ship being involved in collisions. Let pseqr , r ¼ 1; 2; . . . 216, be the occurrence probability of the rth accident scenario. Mathematically, it can be expressed by:

pseqr ¼

5 Y

PðEk jEk1 Þ;

k ¼ 1; . . . 5

ð5Þ

k¼1

where Ek (k = 1, . . . , 5) is the intermediate event k along the path possessed by the rth scenario, PðEk jEk1 Þ is the probability that the event Ek is triggered by the event Ek1 along the path possessed by the rth scenario. Hence, the collision frequency of the rth accident scenario, r denoted by f collision , can be written as r

f collision ¼ f collision  Pseqr ;

r ¼ 1; 2; . . . 216

ð6Þ

4.3. Ship collision consequence estimation For maritime authorities, their major concern is how to enhance the safety of human life and preserve the environment caused by ship accidents (Ceyhun, 2014). Hence, two types of ship collision consequences are considered, including the loss of human life and the volume of oil spills from a ship collision. In general, the human life loss depends on the number of crew and passengers in the ship, hull damage and the survivability of the ship. It should be noted that these influencing factors may exhibit interaction effects on the human life loss. A ‘‘generic” human life loss model would ideally have a multiplicative format and account for the interaction between variables. Such a model can be represented by the following formula:

T. Chai et al. / Safety Science 91 (2017) 71–83

75

Fig. 4. Event tree structure for ship collisions.

Humanloss ¼ Nperson  aST  f HD  f SU

ð7Þ

Nperson ¼ Ncrew þ Npassenger

ð8Þ

where Humanloss = the number of fatalities for the ship involved in a collision (i.e., human life loss); N person = the number of persons which is the sum of the numbers of crew members and passengers in the ship; N crew = the number of crew members in the ship; N passenger = the number of passengers in the ship; aST = the average percentage of persons killed in the ship suffering a collision; f HD = adjustment factor for the hull damage; and f SU = adjustment factor for the ship survivability. In addition, a similar model is developed to calculate the volume of oil spilled from the ship collision:

Oilspill ¼ ðV bunker þ V loading  bLC Þ  bSU

ð9Þ

where Oilspill = the volume of oil spilled; V bunker = the volume of bunker oil carried by the ship; V loading = the oil volume carried by the ship; bLC = adjustment factor for the loading condition; bSU = adjustment for the ship survivability. 5. Case study 5.1. Data collection According to the requirement of the IMO’s International Convention for the Safety of Life at Sea (SOLAS), all larger seagoing ships (>300 GT) and all passenger ships are requested to be equipped with an automatic information system (AIS) on board after 2002. Through dedicated very high frequencies (VHF), AIS information can be transmitted between ships, from ships to shore and vice versa. In simple terms, AIS is a technology which makes

ships ‘‘visible” to each other. It can record information on ship behavior, including the effects of human behavior and ship maneuverability. In general, each AIS record consists of the following information for each ship at each reporting time (every 3–10 s): (i) (ii) (iii) (iv) (v)

MMSI (Maritime Mobile Service Identity) number; Latitude position; Longitude position; Speed over ground (SOG); Course over ground (COG).

To date, the AIS data has been widely used to evaluate ship collision probability (e.g., Goerlandt and Kujala, 2014). The Singapore Strait locates between the Strait of Malacca in the west and the South China Sea in the east, which links one of the largest ports to the rest of the world. The Strait has a high density of ship traffic. Although the Singapore Strait is of great importance to the global economy, it is not deep enough for some of the largest ships (mostly oil tankers). In addition, the Strait has substantial sections of narrower and shallower shipping lanes. Therefore, the safe navigation of ships in maritime transportation, especially in such narrow shipping fairways, is of utmost concern to Maritime and Port Authorities of Singapore. Therefore, we collected one month’s AIS data in the Singapore Strait from the Lloyd’s MIU database in this case study in order to assess the ship collision risk using the proposed QRA model. A total of 166,182 AIS records were collected from the 1st of July 2009 to the 31st of July 2009 to evaluate the ship collision frequency in the Singapore Strait between longitudes 103°210 E and 104°350 E. It should be pointed out that the raw AIS data do not contain the ship characteristics such as ship type and ship size. Actu-

76

T. Chai et al. / Safety Science 91 (2017) 71–83

ally, these ship characteristics can be found from a ship database. Since each ship has a unique MMSI number, the ship characteristics were extracted from the ship database by means of the corresponding MMSI number. In addition, there are several records with inaccurate ship position and speed information though the majority of collected AIS records are accurate. Hence, the data processing method proposed by Qu et al. (2011) has been employed to remedy these inaccurate records including the incorrect speed and position related information in this case study. Fig. 5 shows the proportions of different ship types sailing in the Singapore Strait based on the collected AIS records. It can be clearly seen from the figure that container ships account for the biggest proportion (36.36%), followed by bulk carriers (20.50%) and then tankers (18.30%). The percentage of Passenger/RORO ships is quite small (4.77%) in the Strait. Fig. 6 provides the speed distribution of different ship types. It can be found that the container ship has the largest average sailing speed (11 knots), followed by the chemical/ LNG/LPG (10 knots) in the Singapore Strait.

ing the fact that the Maritime and Port Authority of Singapore has used the vessel traffic service (VTS) system to monitor the Singapore Strait, it is reasonable to assume that the causation probabilities for the crossing conflicts during the daytime and nighttime periods are respectively equal to 6.83  105 and 4.64  104 5 (i.e., pcros for the daytime conflicts and causation = 6.83  10 4 cros pcausation = 4.64  10 for the nighttime conflicts) in good weather 4 . conditions. In adverse weather conditions, pcros causation = 4.64  10

5.2. Input parameters

5.2.3. Consequence parameters In order to determine the ship collision consequences, Tables 4 and 5 report the input values of parameters for the consequence estimation. It should be pointed out that only the oil tanker is assumed to have the loaded oils. The values of loaded oils for the oil tankers can be taken from the previous IMO report (IMO, 2008c). Hence, the average loaded oil volumes (V loading ) are set to be 67,189 tons for a small-sized oil tanker, 124,970 tons for a medium-sized oil tanker and 291,633 tons for a large-sized oil tanker, respectively.

5.2.1. Ship collision frequency estimation parameters As mentioned above, the causation probability is used to compute the ship collision frequency. In general, the causation probability can be estimated in two ways which are the scenario approach and synthesis approach, respectively. In the scenario approach, the causation probability can be estimated from the available accident data at various locations and then transformed to the analysis area. The advantages of this approach are its simplicity and robustness. The causation probability depends on several functions related to traffic perception, communication and avoidance actions. It is also affected by the potential collision situation, weather conditions, etc. To have accurate accident probabilities, the causation probability should reflect the characteristics of the studied area. However, there are inadequate historical accident data to calibrate the causation probabilities in the Singapore Strait. A large number of previous studies have used the causation probabilities shown in Table 1 for different water areas because there is still no evidence to support that their values could significantly differ from water areas. In the Singapore Strait, there is no geographic area where the frequency of visibility less than 1 km is 3%. From the practical point of view, the risk assessment should be made based on the worst case. Therefore, we take the bigger same values for the overtaking v er and head-on conflicts in this case study: pocausation = 4.90  105 for 5 head the overtaking conflicts and pcausation = 4.90  10 for the head-on conflicts, which were also used in most previous studies. Consider-

5.2.2. Occurrence probabilities of intermediate events in the event tree To apply the proposed QRA model, we take a set of values for the percentages or probabilities associated with intermediate nodes. In this case study, the percentages of different ship types being involved in collisions (i.e., pðST 1 Þ; . . . ; pðST 6 Þ) are calculated based on the collision estimation results using the AIS data, as shown in Table 2. The other occurrence probabilities/percentages of intermediate events are taken from the previous IMO reports, as shown in Table 3.

5.3. Results and discussions Given all the parameters required by the proposed QRA model, the occurrence frequencies and consequences of all possible ship collision scenarios can be calculated. These results are discussed in detail below. 5.3.1. Collision frequency The total occurrence frequency of ship collisions in the Singapore Strait is 2.15/year, which implies that there will be one ship collision occurred in the Strait every six months. It should be pointed out that the causation probability values used in this case study are acceptable though they are from other studies. This is because the estimated 2.15 accidents per year is located in the 95% confidence interval [2.15  1.96 ⁄ 0.45, 2.15 + 1.96 ⁄ 0.45] = [1.27, 3.03]. Actually, the estimated collision frequency is quite close to the average actual frequency of 2.05/year which is calcu-

6.13% 20.50% 36.36%

18.30% 4.77%

General Cargo Chemical/LNG/LPG ship

13.94%

Bulk Carrier Passenger/RORO ship

Oil tanker Container ship

Fig. 5. Percentages of different ship types in the Singapore Strait.

77

T. Chai et al. / Safety Science 91 (2017) 71–83

Weibull(2.4898,9.6596)

Weibull(2.3243,12.635)

Probability density function

Probability density function

0.10 0.08 0.06 0.04 0.02 0.00

0

5

10

15

20

0.14 0.12 0.10 0.08 0.06 0.04 0.02 0.00

25

0

5

10

(a) Container ship Weibull(2.2456,9.2148)

Probability density function

Probability density function

0.08 0.06 0.04 0.02

5

10

15

20

0.12 0.10 0.08 0.06 0.04 0.02 0.00

25

0

5

10

Speed (knot)

Weibull(2.8557,11.502)

Probability density function

Probability density function

0.10 0.08 0.06 0.04 0.02 10

15

25

30

25

30

Weibull(1.7732,10.9631)

0.12

5

20

(d) Tanker

0.14

0

15

Speed (knot)

(c) Bulk carrier

0.00

25

Weibull(2.3721,9.0911)

0.10

0

20

(b) General cargo

0.12

0.00

15

Speed (knot)

Speed (knot)

20

25

0.10 0.08 0.06 0.04 0.02 0.00

0

5

10

Speed (knot)

(e) LNG/LPG

15

20

Speed (knot)

(f) RORO/Passenger

Fig. 6. Speed distribution of different ship types.

lated from historical accident records between the years of 1997 and 2004 in the Singapore Strait. This confirms the appropriateness of the used causation probabilities which were taken from the existing literature shown in Table 1. In addition, among these ship collisions, there is a biggest proportion of crossing collisions (1.36/year), followed by the overtaking collisions (0.55/year) and the head-on collisions (0.24/year). According to Table 2, the container ship, bulk carrier and oil tanker are the three main types of ships involved in collisions. Overall, the container ship is the most risky ship type in terms of the overall collision frequency, which is the sum of occurrence frequencies of overtaking, head-on and crossing collisions. Fig. 7 shows that this ship type has the biggest overtaking collision frequency. However,

oil tankers rather than the container ships have the highest headon collision frequency. The difference may be due to the ship navigation rules in the Singapore Strait. One of the rules requires that eastbound deep draught ships (i.e., oil tankers) should use the deep water routes. Since the deep water routes are adjacent to the westbound traffic routes, oil tankers would certainly have more chance of colliding head-on with westbound ships. In addition, another rule implies that eastbound tankers navigating in the deep water routes should avoid overtaking actions, which decreases the likelihood of oil tankers being involved in overtaking collisions. The spatial distribution of ship collisions in the Singapore Strait is presented in Fig. 8. It can be clearly seen from the figure that the most risky area is between longitudes 103°480 E and 104°060 E

78

T. Chai et al. / Safety Science 91 (2017) 71–83

sponding frequency are shown on the abscissa and ordinate, respectively. Mathematically, the F/N curve can be expressed by

Table 1 Recommended causation probabilities from the literature. Conflict types

Causation probabilities

References

Remarks

Crossing conflicts

1:11  104

Pedersen (1995)

Without TSS*

9:5  105

Pedersen (1995)

With TSS

1:2  104

Macduff (1974)

1:3  104 8:48  105

Pedersen (2002) and Kujala et al. (2009) Otto et al. (2002)

6:83  105

Otto et al. (2002)

Overtaking and head-on conflicts

5:8  104

Otto et al. (2002)

4:64  104

Otto et al. (2002)

4:9  105

Macduff (1974), Pedersen (2002), Fowler and Sorgard (2000), Montewka et al. (2010) Karlsson et al. (1998)

2:7  105 8:4  105

* #

Good visibility Good visibility with VTS# zone Poor visibility Poor visibility with VTS zone

Fowler and Sorgard (2000)

In geographical areas where the frequency of visibility less than 1 km is 3%

TSS = Traffic Separation Scheme. VTS = Vessel Traffic Service.

Ship type m Oil tanker Container ship Bulk carrier General cargo Chemical/LNG/LPG Passenger/RORO

4692/year 7932/year 4992/year 1884/year 3312/year 1512/year

ð10Þ

where x is the consequence (i.e., human life loss) caused by a ship collision, and N is a given value; Pðx P NÞ can be described as the frequency that the consequence (i.e., human life loss), x, is not less than the number N. The expected human life loss is 7.25 persons/year and the expected oil spilled is 1143 ton/year in the Singapore Strait. Further, Fig. 9 depicts the relationship between the human life loss caused by ship collisions and the corresponding occurrence frequency. It can be clearly seen that the occurrence frequency decreases with the number of human life loss for all ship types. This implies that the occurrence likelihood for each ship type suffering a major collision is much lower than the probability of being involved in a minor collision. Hereafter, a major accident is defined as an accident causing a big number of human life loss (e.g., the number of human life loss >10) or the large oil spilled volume (e.g., oil spilled volume >1000 tons). In general, the passenger/ RORO ship has the biggest occurrence likelihood suffering the huge fatalities, as shown in Fig. 9. For example, the frequencies of an accident causing >25 human life loss are 3:24  106 /year for the oil tanker, 3:08  106 /year for the bulk carrier, 2:77  106 /year

Table 2 Number of ship conflicts in the Singapore Strait. v er N m;o conflict

F N ðxÞ ¼ Pðx P NÞ

for the general cargo ship, 2:14  106 /year for the chemical/

N m;head conflict

N m;cros conflict

m f collision

3648/year 2196/year 2592/year 324/year 1236/year 576/year

3420/year 3468/year 4128/year 1104/year 1548/year 1200/year

1.03/year 1.13/year 1.10/year 0.29/year 0.50/year 0.31/year

Note: v er N m;o = Number of ships of type m suffering overtaking conflicts per year. conflict N m;head = Number of ships of type m suffering head-on conflicts per year. conflict N m;cros = Number of ships of type m suffering crossing conflicts per year. conflict m

f collision = Number of ship collisions for ship type per year.

LNG/LPG ship, 1:81  106 /year for the container ship and 4:00  102 /year for the passenger/RORO ship, respectively. These results reflect that the frequencies of lower consequence events significantly vary with respect to different ship types. According to the F/N curve results shown in Fig. 9, one suggestion is that the major efforts should be made to avoid the passenger/RORO ships being involved in ship collisions in order to reduce the human life loss. Similarly, Fig. 10 shows that a serious accident causing a big volume of oil spilled is associated with the low occurrence frequency. The frequencies causing the oil spilled volume >1000 tons are 1:30  101 /year for the oil tanker, 2:98  103 /year for the general cargo ship, 8:07  102 /year for the container ship and

because of the biggest collision occurrence frequency. One possible reason for the high collision frequency might be that there are a large number of ship acceleration and deceleration maneuvers within this area. This is because the eastbound ships are requested to reduce their speed to 12 knots within this area while the westbound ships increase their speed from 12 knots if conditions are suitable. In addition, the traffic width in this area (specifically between longitudes 103°50´E and 104°00´E) is narrower than in other areas. Hence, the density of ships within this area is the highest in the entire Singapore Strait. The high ship density will increase the number of ship conflicts, suggesting that the headon collision frequency of this area is the biggest in the Singapore Strait. 5.3.2. Relationship between frequency and consequence of collisions The individual risk implies that the risk can be aggregated into a single number. For example, the expected human life loss per year is an individual risk which is a product of the occurrence frequency and occurrence consequence. However, note that a single number cannot address the relationship between the frequency and consequence. The F/N curve is the most common form to illustrate the relationship between the accident occurrence consequence and the corresponding occurrence frequency. The accident consequence (i.e., the number of persons being killed) and the corre-

4:00  102 /year for the passenger/RORO ship, respectively. These results show that the oil tanker could cause serious oil pollution more frequently than the other ship types. This is due to the fact that oil tankers usually carry more oils than other ship types. For the purpose of reducing oil pollutions, special strategy should be taken to reduce the risk of collision involving oil tankers. In addition, the comparison between the curves for the bulk carrier and chemical/LNG/LPG ship shows that the chemical/LNG/LPG ship has higher frequency (5:78  102 /year) than the bulk carrier (2:30  102 /year) suffering serious collisions that causes the large oil spilled volume (e.g., >1000 tons). Therefore, the maritime authorities should also take effective countermeasures to reduce the risk of the chemical/LNG/LPG ship being involved in collisions. It should be pointed out that the proposed QRA model is able to determine the effectiveness of navigational safety strategy when it is further embedded with a ship traffic simulation model. For example, extending and connecting the existing TSS could be a possible strategy to enhance the navigational safety in the Singapore Strait. The ship traffic simulation model is able to generate ship movement trajectories if the strategy is implemented. Based on these simulated ship trajectories, the proposed QRA model could be able to check the viability of this new navigational safety strategy by identifying whether both the accident frequency and consequence reduce or not.

79

T. Chai et al. / Safety Science 91 (2017) 71–83 Table 3 Input parameters for estimating the occurrence frequency. Parameter

Descriptions

p(ST1) p(ST2) p(ST3) p(ST4) p(ST5) p(ST6) p(SS1)

Percentage Percentage Percentage Percentage Percentage Percentage Percentage

*

oil tankers involving collisions container ships involving collisions general cargo ships involving collisions bull carriers involving collisions chemical/LNG/LPG ships involving collisions passenger/RORO ships involving collisions small-sized ships

p(SS2)

*

Percentage of medium-sized ships

p(LC1)

*

Percentage of ships with loading cargos/oils

p(HD2)

*

of of of of of of of

Value

*

Probability of ships involving a critical hull damage

p(SU2|HD1) p(SU3|HD1) p(SU2|HD2)

*

p(SU3|HD2)

*

* *

Probability of sinking slowly under the minor damage condition Probability of sinking rapidly under the minor damage condition Probability of sinking slowly under the critical damage condition

Probability of sinking rapidly under the critical damage condition

23.4% 26.1% 6.9% 25.0% 11.5% 7.1% 70% for oil tanker 30% for container ship 90% for general cargo 80% for bulk carrier 80% for chemical/LNG/LPG 85% for passenger/RORO 10% for oil tanker 40% for container ship 10% for general cargo 17% for bulk carrier 20% for chemical/LNG/LPG 12% for passenger/RORO 60% for oil tanker 50% for container ship 50% for general cargo 60% for bulk carrier 50% for chemical/LNG/LPG 0 for passenger/RORO 24.3% for oil tanker 50% for container ship 50% for general cargo 24.3% for bulk carrier 65% for chemical/LNG/LPG 50% for passenger/RORO 0 0 5.7% for oil tanker 13.3% for container ship 13.3% for general cargo 5.7% for bulk carrier 10.6% for chemical/LNG/LPG 17.3 for passenger/RORO 2.9% for oil tanker 6.7% for container ship 6.7% for general cargo 2.9% for bulk carrier 4.8% for chemical/LNG/LPG 9.7 for passenger/RORO

Source from IMO (2002, 2007a, 2007b, 2008a, 2008b, 2011).

5.3.3. Impact analysis As mentioned, the factors of hull damage, loading condition and ship sinking speed could influence the accident consequence. To quantitatively evaluate the effect of a specific factor (e.g., critical hull damage) on the average human life loss and oil spilled volume, the corresponding occurrence probability (e.g., p(HD2)) is increased by 1% while the occurrence probabilities for other factors remain unchanged. Fig. 11 clearly shows that different factors exhibit various influences on the average number of human life loss and oil spilled volume. More specifically, each 1% increase in the probability of critical hull damage could increase the number of human life loss and the volume of oil spilled from ship collision by 2.1% and 3.6% on average, respectively. Compared with the critical hull damage, the loading condition factor shows smaller effects on the number of human life loss (0.7%) and oil spilled volume (1.3%). The factor of ship sinking speed presents the biggest influences on the human life loss resulting from ship collisions. For example, Fig. 11(a) shows that there will be an increase of 8.1% on the human life loss if the probability of sinking rapidly is increased by 1%. However, the oil spilled volume is not significantly influenced by the sinking speed. It can be seen from Fig. 11(b) that the 1% increase in the probability of the

rapid sinking only results in an increase of 0.5% in the oil spilled volume on average. 6. Conclusions In order to evaluate the risk of a ship suffering ship collisions, we proposed a QRA model which took into account the occurrence frequency and consequence of ship collisions in this study. A collision frequency estimation model was built to estimate the occurrence frequency of ship being involved in collisions. Since a number of scenarios may be possible for the ship involved in a collision, this study utilized an event tree diagram to identify all possible scenarios. Five intermediate events comprising ship type (ST), ship size (SZ), loading condition (LC), hull damage (HD) and survivability (SU) were considered in the event tree analysis. In addition, we formulated consequence estimation models for the QRA model in an attempt to estimate the quantitative consequence including the human life loss and environmental pollution which is measured in term of the volume of oil spilled from the ship, respectively. In order to testify the proposed QRA model, a case study was finally created using one month’s real-time ship movement data in the Singapore Strait which were collected from the Llyod’s Marine Intelligence Unit (Lloyd’s MIU) database.

80

T. Chai et al. / Safety Science 91 (2017) 71–83

Table 4 Input parameters for estimating the collision consequence.

a b c d e f

Ship type

Ship size

Number of persons (N person )

Bunker oil volume (V bunker )

Oil tankera

Small size Medium size Large size

26 28

6450 tons 10,122 tons

32

19,828 tons

Container shipb

Small size Medium size Large size

20 24

1600 tons 3207 tons

30

4800 tons

General cargo shipd

Small size Medium size Large size

20 24

1600 tons 3207 tons

30

4800 tons

Bulk carrierc

Small size Medium size Large size

26 28

4800 tons 9600 tons

32

14,400 tons

Chemical/LNG/ LPGe

Small size Medium size Large size

26 28

4800 tons 9600 tons

32

14,400 tons

Passenger/RORO shipf

Small size Medium size Large size

2080 2728

4800 tons 7200 tons

4000

9600 tons

Source Source Source Source Source Source

from from from from from from

IMO IMO IMO IMO IMO IMO

(2008a). (2007a). (2002). (2011). (2007b). (2008b).

it has been found that the occurrence frequency decreases with the number of human life loss and the volume of oil spilled. In other words, the major accident is associated with the low occurrence frequency while the minor accident is more likely to be associated with the high occurrence frequency. Due to the relative high percentage of ship collisions involving oil tankers and their severe consequences, focus should perhaps be placed on the tracking and management of oil tanker traffic. From the practical viewpoint, the proposed QRA model could be used to evaluate the effectiveness of possible navigational safety measures in fairways. For example, more and more mega ships will start to use the congested fairways for the foreseeable future because of the scale advantages. In order to accommodate the increasing number of mega ships, one possible strategy is that the deep-waterway will be widened. With the help of a computer simulation model, the proposed QRA model could assess the impacts of this strategy on the navigational safety. In addition, the proposed QRA model could help shipping companies identify those areas with high risk in fairways. Although the proposed QRA model has already taken into various factors like ship type, ship size and visibility in estimating the ship collision frequency, the impact of ship type on the accident consequence is not considered in this study. Due to data limits, this study did not take into account the effects of ship size and navigational skills on the causation probability. Therefore, future research will be conducted to examine the effects of ship type on the human life loss and environmental pollution resulting from ship accidents. In addition, we will investigate the relationship between the causation probability and the factors like navigational skills in the future. Appendix A

Table 5 Recommended adjustment factors for the consequence estimation. Adjustment factor

Recommended values for the consequence estimation

aST *

aST aST aST aST

f SU *

f HD *

*

= 0.44 = 0.50 = 0.53 = 0.76

if if if if

ST = ‘‘oil tanker” or ‘‘bulk carrier” ST = ‘‘general cargo ship” or ‘‘container ship” ST = ‘‘chemical/LNG/LPG ship” ST = ‘‘passenger/RORO ship”

f SU = 0.32, if SU = ‘‘stay afloat” and ST = ‘‘oil tanker” or ‘‘bulk carrier” f SU = 0.0017, if SU = ‘‘stay afloat” and ST = ‘‘passenger/RORO ship” f SU = 0.033, if SU = ‘‘sink slowly” and ST = ‘‘passenger/RORO ship” f SU = 1, otherwise

Mou et al. (2010) reported that the majority of minimum distances to collision (i.e., radius of the disc) are approximately three times of the average ship length of the two ships. Hence, the domain radius of a particular ship i with respect to ship j can be computed as

Rij ¼ 3lij

ðA1Þ

where Rij is the domain radius of ship i with respect to ship j and lij is the average ship length between the ships i and j. !

Let ðxti ; yti Þ denote the position of a particular ship i at time t; v ti and ati denote the speed and course of ship i at time t, respectively.  tþDT tþDT  xi ; yi represents the probable position of ship i at time t þ DT. Similarly, ðxtj ; ytj Þ denotes the position of another ship j at

f HD = 0.811, if ST = ‘‘chemical/LNG/LPG ship” and HD = ‘‘minor” and SU = ‘‘sink slowly” or ‘‘sink rapidly” f HD = 1, if HD = ‘‘critical” f HD = 1, if ST = ‘‘passenger/RORO ship” and HD = ‘‘minor” f HD = 1, if HD = ‘‘minor” and SU = ‘‘sink slowly” or ‘‘sink rapidly” f HD = 0, otherwise

t, denoted by V tij , can be calculated by

bLC *

bLC = 2, if ST = ‘‘oil tanker” f HD = 0, otherwise

jV tij j ¼

bSU *

bSU = 0.001, if ST = ‘‘oil tanker” and LC = ‘‘loaded” and SU = ‘‘stay afloat” bSU = 1, if ST = ‘‘oil tanker” and LC = ‘‘loaded” and SU = ‘‘sink slowly” or ‘‘sink rapidly” bSU = 0, otherwise

Source from IMO (2002, 2007a, 2007b, 2008a, 2008b, 2011).

The results show that the container ship is the most risky ship type in terms of the collision frequency while the passenger/RORO ship has the lowest frequency suffering ship collisions. In addition,

!

v tj

and atj represent the speed and course of ship j at time   t, respectively. xjtþDT ; yjtþDT represents the probable position of time t;

ship j at time t þ DT. The relative speed of ship i over ship j at time !

!

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ! 2  ! 2 ! ! jv ti j cos ati  jv tj j cos atj þ jv ti j sin ati  jv tj j sin atj ðA2Þ

The distance between the ships i and j at time t, denoted by Dtij , can be computed by

Dtij ¼

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 ðxti  xtj Þ þ ðyti  ytj Þ

ðA3Þ

According to Montewka et al. (2010), there will be a traffic conflict only and only if the distance between two ships at time t þ DT is not greater than the radius of the disc shown in Fig. 3. Namely,

81

T. Chai et al. / Safety Science 91 (2017) 71–83

100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%

Overtaking accident

Head-on accident

Oil tanker Bulk carrier

Crossing accident

Container ship Chemical/LNG/LPG

General cargo Passenger/RORO

Fig. 7. Proportions of different ship types involving in each collision type.

Collision accident occurrence frequency (1/year)

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0 103°30'E 103°36'E

103°42'E

103°48'E

103°54'E

104°00'E

104°06'E

104°12'E

103°18'E

Longitude Fig. 8. Spatial distribution of ship collisions.

1.0E+00

1.0E+00

1.0E-02

1.0E-01

Frequency (1/year)

Frequency (1/year)

1.0E-01

1.0E-03 1.0E-04

Minor accident

Major accident

1.0E-05 1.0E-06 1.0E-07

Oil tanker Passenger/RORO General cargo

Bulk carrier Chemical/LNG/LPG Container ship

1.0E-02

Minor accident

Major accident

Oil tanker

1.0E-03

Bulk carrier Container ships General cargo ships Chemical/LNG/LPG Passenger/RORO

1.0E-08 1.0E+00

1.0E+01

1.0E+02

1.0E+03

1.0E+04

Human life loss (person) Fig. 9. Relationship between the occurrence frequency and the number of human life loss in ship collisions.

1.0E-04 1.0E+00

1.0E+01

1.0E+02

1.0E+03

1.0E+04

1.0E+05

Oil spill volume (ton) Fig. 10. Relationship between the occurrence frequency and the volume of oil spilled in ship collisions.

82

T. Chai et al. / Safety Science 91 (2017) 71–83

9.0% 8.0% 7.0% 6.0% 5.0% 4.0% 3.0% 2.0%

1% increase of p(HD2)

1.0% 0.0%

Loading condition

Critical hull damage

Sink slowly

Sink rapidly

(a) Percent changes in the human life loss 4.0% 3.5% 3.0% 2.5% 2.0% 1.5% 1.0% 0.5% 0.0% Critical hull damage

Loading condition

Sink slowly

Sink rapidly

(b) Percent changes in the oil spilled volume Fig. 11. Percent changes in the average number of human life loss and oil spilled volume with respect to the 1% increase in each factor.

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  2  2 xitþDT  xjtþDT þ yitþDT  yjtþDT 6 Rij

ðA4Þ

Assuming the ships i and j keep their current speed and course from the time t to t þ DT, the condition shown by Eq. (A4) is equivalent to rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  2  2 ! ! ! ! xti  xtj þ jv ti j sin ati DT  jv tj j sin atj DT þ yti  ytj þ jv ti j cos ati DT  jv tj j cos atj DT 6 Rij ðA5Þ

It should be pointed out that the condition shown in Eq. (A5) is also equivalent to rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi h !  ! i ! ! ! 2 2 ðDtij Þ þ jv tij DTj þ 2DT ðxti  xtj Þ jv ti jsin ati  jv tj jsin atj þ ðyti  ytj Þ jv ti jcos ati  jv tj jcos atj 6 Rij ðA6Þ

From Eq. (A6), we can obtain the following relationship

Ltij

¼

Dtij

cos htij

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  2 !  R2ij  Dtij sin htij 6 jv tij jDT

ðA7Þ

where Ltij represents the relative sailing distance of ship i with respect to ship j from time t to t þ DT, and htij is the angle between !

the relative speed V tij and the line connecting two ships, as shown in Fig. 3. 0 !2 !2 1 B B B t hij ¼ arccos B1  B @

!

!

j v ti j sin ati j v tj j sin atj !

jV tij j



xtj xti Dtij

!

þ 2

!

j v ti j cos ati j v tj j cos atj ! jV tij j



ytj yti Dtij

C C C C C A

ðA8Þ

References Akten, N., 2004. Analysis of shipping casualties in the Bosporus. J. Nav. 57, 345–356. Ceyhun, G.C., 2014. The impact of shipping accidents on marine environment: a study of Turkish seas. Eur. Sci. J. 10 (23), 10–23. Dand, I.W., 2001. Approach channel design: the PIANC approach. In: Proceedings of the International Workshop on Channel Design and Vessel Maneuverability, Norfolk, VA. Fowler, T.G., Sorgard, E., 2000. Modeling ship transportation risk. Risk Anal. 20, 225– 244. Fujii, Y., Yamanouchi, H., Mizuki, N., 1974. Some factors affecting the frequency of accidents in marine traffic. II – The probability of stranding. III – The effect of darkness on the probability of collision and stranding. J. Nav. 27, 239–247. Goerlandt, F., Kujala, P., 2014. On the reliability and validity of ship–ship collision risk analysis in light of different perspectives on risk. Saf. Sci. 62, 348–365. Goossens, L.H.J., Glansdorp, C.C., 1998. Operational benefits and risk reduction of marine accidents. J. Nav. 51, 369–381. IMO, 2002. Formal Safety Assessment of Bulk Carrier Safety. International Maritime Organization, London, Document No.: MSC75/5/2, February 12, 2002. IMO, 2007a. Formal Safety Assessment FSA-Container Vessels. International Maritime Organization, London, Document No.: MSC83/INF.8, July 3, 2007. IMO, 2007b. Formal Safety Assessment FSA-Liquefied Natural Gas (LNG) Carriers. International Maritime Organization, London, Document No.: MSC83/INF.3, July 3, 2007. IMO, 2008a. Formal Safety Assessment FSA-Crude Oil Tankers. International Maritime Organization, London, Document No.: MEPC 58/INF.2, July 4, 2008. IMO, 2008b. Formal Safety Assessment FSA-RoPax Ships. International Maritime Organization, London, Document No.: MSC85/INF.3, July 21, 2008. IMO, 2008c. Formal Safety Assessment FSA-Crude Oil Tankers. International Maritime Organization, NY. IMO, 2011. Formal Safety Assessment FSA of General Cargo Ships. International Maritime Organization, London, Document No.: MSC2088/INF.8, 2011. Karlsson, M., Rasmussen, F.M., Frisk, L., 1998. Verification of ship collision frequency model. In: Gluver, H., Olsen, D. (Eds.), Ship Collision Analysis, Proceedings of the International Symposium on Advances in Ship Collision Analysis. Balkema, Rotterdam, pp. 117–121. Kujala, P., Hanninen, M., Arola, T., Ylitalo, J., 2009. Analysis of the marine traffic safety in the Gulf of Finland. Reliab. Eng. Syst. Saf. 94, 1349–1357. Li, S., Meng, Q., Qu, X., 2012. An overview of maritime waterway quantitative risk assessment models. Risk Anal. 32 (3), 496–512. Liwang, H., Ringsberg, J.W., Norsell, M., 2013. Quantitative risk analysis-ship security analysis for effective risk control options. Saf. Sci. 58, 98–112. Macduff, T., 1974. Probability of vessel collisions. Ocean Ind. 9, 144–148. Meng, Q., Weng, J., Qu, X., 2010. A probabilistic quantitative risk assessment model for the long-term work zone crashes. Accid. Anal. Prev. 42, 1866–1877.

T. Chai et al. / Safety Science 91 (2017) 71–83 Montewka, J., Hinz, T., Kujala, P., Matusiak, J., 2010. Probability modeling of vessel collisions. Reliab. Eng. Syst. Saf. 95, 573–589. Mou, J.M., Tak, C., Ligteringen, H., 2010. Study on collision avoidance in busy waterways by using AIS data. Ocean Eng. 37, 483–490. Otto, S., Pedersen, P.T., Samuelides, M., Sames, P.C., 2002. Elements of risk analysis for collision and grounding of a RoRO passenger ferry. Mar. Struct. 15, 461–474. Pedersen, P., 1995. Collision and grounding mechanics. In: Proceedings of the WEGEMT’95. The Danish Society of Naval Architects and Marine Engineers, Copenhagen, pp. 125–157. Pedersen, P., 2002. Collision risk for fixed offshore structures close to high-density shipping lane. Proc. Inst. Mech. Eng. Part M: J. Eng. Marit. Environ. 216, 29–44. Perez-Labajos, C., Blanco, B., Azofra, M., Achutegui, J.J., Eguia, E., 2009. Injury and loss concentration by sinkings in fishing fleets. Saf. Sci. 47 (2), 277–284. Pietrzykowski, Z., 2008. Ship’s fuzzy domain – a criterion for navigational safety in narrow fairways. J. Nav. 61, 499–514. Pietrzykowski, Z., Uriasz, J., 2009. The ship domain – a criterion of navigational safety assessment in an open sea area. J. Nav. 62, 93–108. Qu, X., Meng, Q., Li, S., 2011. Ship collision risk assessment for the Singapore Strait. Accid. Anal. Prev. 43, 2030–2036.

83

Servis, D.P., Samuelides, M., 1999. Ship collision analysis using finite elements. In: SAFER EURORO Spring Meeting. Si, H., Ji, H., Zeng, X., 2012. Quantitative risk assessment model of hazardous chemicals leakage and application. Saf. Sci. 50, 1452–1461. Szlapczynski, R., 2006. A unified measure of collision risk derived from the concept of a ship domain. J. Nav. 59, 477–490. Tan, B., Otay, E.N., 1999. Modeling and analysis of vessel casualties resulting from tanker traffic through narrow waterways. Naval Res. Logist. 46, 871–892. Van de Wiel, G., Van Dorp, J.R., 2009. An oil outflow model for tanker collisions and groundings. Ann. Oper. Res. http://dx.doi.org/10.1007/s10479-009-0674-5. Wang, N., 2010. An intelligent spatial collision risk based on the quaternion ship domain. J. Nav. 63, 733–749. Yip, T.L., 2008. Port traffic risks – a study of accidents in Hong Kong waters. Transport. Res. Part E: Logist. Transport. Rev. 44, 921–931. Zhang, H., Huang, S., 2006. Dynamic control of intention priorities of human-like agents. In: Proceedings of the 17th European Conference on Artificial Intelligence, ECAT 2006, Italy.