A probabilistic approach for a cost-benefit analysis of oil spill management under uncertainty: A Bayesian network model for the Gulf of Finland.

Journal of Environmental Management xxx (2015) 1e11

Contents lists available at ScienceDirect

Journal of Environmental Management journal homepage: www.elsevier.com/locate/jenvman

Research article

A probabilistic approach for a cost-benefit analysis of oil spill management under uncertainty: A Bayesian network model for the Gulf of Finland €nninen c, Sakari Kuikka a Inari Helle a, *, Heini Ahtiainen b, Emilia Luoma a, Maria Ha a b c

Fisheries and Environmental Management Group (FEM), Department of Environmental Sciences, P.O. Box 65, FI-00014, University of Helsinki, Finland Natural Resources Institute Finland (Luke), Economics and Society, Latokartanonkaari 9, FI-00790, Helsinki, Finland Aalto University, Department of Applied Mechanics, Research Group on Maritime Risk and Safety, P.O. Box 12200, FI-00076, Aalto, Finland

a r t i c l e i n f o

a b s t r a c t

Article history: Received 29 December 2014 Received in revised form 27 March 2015 Accepted 28 April 2015 Available online xxx

Large-scale oil accidents can inflict substantial costs to the society, as they typically result in expensive oil combating and waste treatment operations and have negative impacts on recreational and environmental values. Cost-benefit analysis (CBA) offers a way to assess the economic efficiency of management measures capable of mitigating the adverse effects. However, the irregular occurrence of spills combined with uncertainties related to the possible effects makes the analysis a challenging task. We develop a probabilistic modeling approach for a CBA of oil spill management and apply it in the Gulf of Finland, the Baltic Sea. The model has a causal structure, and it covers a large number of factors relevant to the realistic description of oil spills, as well as the costs of oil combating operations at open sea, shoreline clean-up, and waste treatment activities. Further, to describe the effects on environmental benefits, we use data from a contingent valuation survey. The results encourage seeking for cost-effective preventive measures, and emphasize the importance of the inclusion of the costs related to waste treatment and environmental values in the analysis. Although the model is developed for a specific area, the methodology is applicable also to other areas facing the risk of oil spills as well as to other fields that need to cope with the challenging combination of low probabilities, high losses and major uncertainties. © 2015 Elsevier Ltd. All rights reserved.

Keywords: Cost-benefit analysis Bayesian network Oil spill Maritime safety Gulf of Finland Environmental valuation

1. Introduction Marine, coastal and freshwater ecosystems around the world have been altered by human activities for centuries, but the rate of change has accelerated in recent decades. Intensified impacts of drivers like habitat change, pollution, and overexploitation of species have resulted in adverse effects on biodiversity and ecosystem goods and services (Millennium Ecosystem Assessment, 2005). To counteract this trend, societies are eager to find means to restore and maintain the good status of ecosystems. In a world of limited resources, it is evident that environmental problems need to be combated as effectively as possible. Common approaches for assessing the economic efficiency of proposed environmental projects are cost-effectiveness analysis (CEA) and cost-benefit analysis (CBA). CEA aims at reaching a given target

* Corresponding author. E-mail address: inari.helle@helsinki.fi (I. Helle).

with minimum costs, while CBA compares monetized costs and benefits to assess the economic efficiency of a project or to identify the economically optimal level of action. Although the logic of CBA is fairly straightforward, i.e. to compare the expected gains with the expected losses, there are many issues that need to be addressed particularly when applying the method in the context of environmental problems (Hanley, 1992; Pearce, 1998). For instance, the valuation of environmental impacts is challenging, especially when non-market goods, such as recreation, biodiversity or landscapes, are involved. This challenge can be overcome, at least partly, by using economic valuation methods that quantify the benefits of environmental improvements (or the damages from deterioration) either by posing direct questions on willingness to pay (WTP) (Bateman et al., 2002) or by observing actual behavior (Bockstael and McConnell, 2007). Uncertainty poses another challenge for CBA analyses of environmental problems (Boardman et al., 2014; Pearce, 1998). Uncertainty originates from various sources: natural systems are stochastic by nature, and as they involve myriad of interacting

http://dx.doi.org/10.1016/j.jenvman.2015.04.042 0301-4797/© 2015 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Helle, I., et al., A probabilistic approach for a cost-benefit analysis of oil spill management under uncertainty: A Bayesian network model for the Gulf of Finland, Journal of Environmental Management (2015), http://dx.doi.org/10.1016/ j.jenvman.2015.04.042

2

I. Helle et al. / Journal of Environmental Management xxx (2015) 1e11

factors forming complex entities, the knowledge of the system is imperfect. The situation is even more complex if the occurrence of undesirable events is highly uncertain, and the potential outcomes are dependent on several varying factors and thus exhibit large range. In this case the decision-maker has to consider the ultimate question: “How much to invest in mitigation preparedness, if there is a possibility that the adverse impacts will never materialize?” This holds true for oils spills resulting from tanker accidents. These incidents can result in substantial costs and losses (Allo and Loureiro, 2013; Garza-Gil et al., 2006; Grigalunas et al., 1986; Loureiro et al., 2006). Direct costs result from offshore oil combating operations, shoreline clean-up activities, and logistic and treatment costs of recovered oil. In addition, oil spills usually inflict losses to fisheries and tourism sectors, and have negative impacts on recreational and environmental values (Alvarez et al., 2014; Carson et al., 2003; Loureiro et al., 2009; Loureiro and Loomis, 2013), some of which can be quantified in monetary terms. Although several studies have estimated the costs and monetary damages caused by oil spills, there are very few analyses that have examined the economic efficiency of improving oil spill preparedness by comparing the associated costs and benefits (Cohen, 1986). In recent decades, Bayesian networks (BNs) have gained popularity in the field of environmental research and management (e.g. Aguilera et al., 2011; Landuyt et al., 2013; Varis and Kuikka, 1999). BNs are models that describe the system with probabilistic variables and links between them. As BNs express uncertainty explicitly, they suit well for modeling problems where uncertainty has a fundamental role (Kelly et al., 2013). Bayesian networks can be extended to influence diagrams by adding decision and utility nodes into the networks (Howard and Matheson, 2005). Further, BNs are able to integrate various types of data (e.g. Uusitalo, 2007), including monetary data, and thus offer a potential tool also for cost-benefit analysis under uncertainty. Previously BNs have been applied to CBAs related to eutrophication management by Ames et al. (2005), Barton et al. (2005, 2008), to pesticide management by Henriksen et al. (2007), and to integrated pond management by Landuyt et al. (2014). In this paper we present a BN-based modeling approach that we use to conduct a probabilistic cost-benefit analysis of the management measures capable of reducing the harm from oil spills. Our study is focused on the Gulf of Finland (GoF), the easternmost basin of the Baltic Sea, but the methodology is applicable also to other marine and freshwater areas facing the risk of oil spills. The GoF has witnessed a multifold increase in the volume of transported oil since the early 2000s (Finnish Environment Institute, 2013), and, as part of the Baltic Sea, it has been designated as a Particular Sensitive Sea Area (PSSA) (Russian waters excluded) by the International Maritime Organization (IMO). As it is evident that a major oil accident in the GoF could result in substantial costs, our aim is to study whether there are economically reasonable management measures given the high uncertainty related to the frequency as well as the consequences of future tanker accidents. The primary aim of the model lies in the CBA, but the model can also be used to estimate the costs resulting from a single tanker accident. 2. Methodology: Bayesian networks and influence diagrams Bayesian networks are graphical models for reasoning under uncertainty (Jensen and Nielsen, 2007). Each variable is associated with a probability distribution describing the probability of the variable being in a certain state. Variables not dependent on any other variable have a single probability distribution, whereas variables dependent on one or more variables have a conditional probability table (CPT). A CPT describes the probability of each state

of the variable conditioned on every possible combination of the states of the parent nodes, i.e. the nodes on which the variable is directly dependent. If new information is fed into the network e.g. by instantiating one variable to a certain state, the states of the other variables are updated accordingly, based on the rules of the probability calculus and the Bayes' theorem. This propagation of new knowledge enables BNs to be used for both cause-to-effect and effect-to-cause reasoning. Various techniques can be used to quantify probability distributions within a BN. These include e.g. observed data, simulation results and expert knowledge (Uusitalo, 2007). A more detailed description of the methodology related to BNs can be found e.g. from Fenton and Neil (2013) and Jensen and Nielsen (2007). With decision and utility nodes BNs can be extended to influence diagrams (Howard and Matheson, 2005). This enhances their use as decision support tools. As influence diagrams calculate expected utilities related to different states of the system, they can be used to find the optimal combination of decisions under uncertainty. Further, influence diagrams enable the calculation of value of information (VoI; Raiffa and Schlaifer, 1961), which describes the expected increase in expected utility that could be achieved if new information was acquired before making a decision. BNs are applied in various fields of research including e.g. medicine (Forsberg et al., 2011), forensics (Biedermann and Taroni, 2012), social sciences (Haapasaari and Karjalainen, 2010), and engineering (Langseth and Portinale, 2007). They have gained popularity also in environmental management context, where BNs have been applied e.g. in fisheries (Kuikka et al., 1999; Levontin et al., 2011), water resources management (Bromley et al., 2005; Molina et al., 2010) as well as in other fields of ecology and environmental science (see e.g. reviews by Aguilera et al. (2011), Landuyt et al. (2013) and McCann et al. (2006)). In recent years they have been increasingly employed in studies related to maritime accidents and oil spill risk management (Aps et al., 2009; Carriger and Barron, 2011; Goerlandt and Montewka, 2014; Helle et al., 2011; €nninen and Kujala, 2012, 2014; Juntunen H€ anninen, 2014; Ha et al., 2005; Lecklin et al., 2011; Lehikoinen et al., 2013), and Montewka et al. (2013) have presented a BN for estimating the clean-up costs of oil spills. 3. Structure of the model The model includes altogether 55 variables relevant for the CBA: 2 decision variables, 40 random variables, and 13 utility variables. Further, as the calculation with utility variables is based on expected benefits and costs, additional 10 random variables were included in the model to demonstrate the uncertainty related to each cost and damage type. The model consists of the main model and one sub-model which is used to calculate the monetary damages to the environment (Fig. S1 in Supplementary Material). Marginal distributions and conditional probability tables for the variables were formed based on several resources and techniques such as existing statistics, expert knowledge, other models (simulation models, BN models as well as other types of models) and published papers, and the cost data were gathered from literature or the experts working with oil combating issues in the Finnish Environment Administration or in regional rescue departments. The description of variables, as well as the data and techniques used to populate the network, are presented in detail in Supplementary Material. The model was built with Hugin Researcher 7.6 software (Madsen et al., 2005; www. hugin.com). In the following we give a general overview of the model and the main assumptions related to it, after which decision and utility variables are described in more detail. A simplified representation of the model is presented in Fig. 1.



The model calculates the expected yearly costs and damages of oil spills given different management actions by taking into account the yearly probabilities of tanker accidents and the subsequent effects given there is an accident. The basic logic is thus similar as in Lehikoinen et al. (2015), who estimated the expected yearly amount of oil in the environment after tanker collisions. The approach can also be seen as an application of the widely used definition of risk that states that risk is a combination of the probability of an event and the adverse consequences of the event (e.g. Burgman, 2005; International Organization for Standardization, 2009). The management decisions in the model include one preventive measure, which has an impact on accident probabilities, and one post-spill measure affecting the efficiency of offshore oil combating. The spill-specific costs include direct costs originating from mechanical recovery at open sea and shoreline clean-up activities as well as the costs related to the storage, transportation and final treatment of oily waste. Further, data from a previously conducted contingent valuation survey (Ahtiainen, 2007) are used to calculate the marginal damages of spilled oil to the environment and coastal recreation. In the CBA, the benefits of the management actions are calculated as a decrease in these expected costs and damages. The model covers two types of tanker accidents: groundings and collisions with vessels, which are the most typical accident types in the GoF (Kujala et al., 2009). The number of tanker accidents as well as the size distribution of tankers is dependent on the future development of maritime traffic in the GoF. The number of spills per year is calculated by taking into account the probabilities that a tanker involved in an accident is loaded with oil, and the accident results in a leak. The accident type together with the size of the tanker has an effect on the subsequent spill size. After a spill, the amount of oil afloat is decreased by evaporation and mechanical recovery, which both are affected by environmental conditions such as wave height. The efficiency of mechanical recovery is based on the Finnish fleet of oil combating vessels. Oil that does not evaporate or is not recovered offshore is assumed to drift onshore. Mechanical and manual cleanup of shoreline continues until all oil

Fig. 1. A simplified representation of the model. The expected yearly costs of oil spills are calculated by combining the number of accidents with the cost/losses resulting from different types of accidents. In addition, the acquisition and maintenance costs of management options are taken into account. Rectangles represent decision variables, ellipses random variables (the exact states of which are uncertain) and hexagons costs and losses (V).

3

is removed. All these accident-specific factors are modeled as probabilistic variables, described in detail in Supplementary Material. The probability distributions for the variables when the model is in a basic state and no decisions are made are presented in Fig. S3 in Supplementary Material. 3.1. Management measures: decision variables The management options have two states, i.e. they are implemented or not. The implementation of the measures involves acquisition and possible maintenance costs, but also reduces the amount of oil in the environment, which results in lower clean-up costs and environmental damages. The preventive measure is an automatic alarm system (AAS) that would be merged with the present Vessel Traffic Service (VTS) system. The purpose of the VTS system is to prevent traffic jams and dangerous encounters of the vessels, and provide relevant information to the vessels. The new alarm system would give an automatic alarm if two ships are on a collision course. This improves the ability to detect situations that could lead to accidents, decreasing the probability of tanker accidents in the GoF. More information on the automatic alarm system can be found in Lehikoinen et al. (2015). The post-spill measure is the acquisition of a new oil combating vessel. Our analysis is based on the newest oil combating vessel in Finland, Louhi, which was put into operation in 2011. Thus, we were able to use the data on the already realized acquisition and maintenance costs for the description of the new vessel. 3.2. Costs and benefits: utility variables The model has 13 utility variables that describe the costs related to the implementation of the management measures, and the expected costs resulting from oil combating and clean-up operations and the losses of environmental values in the case of oil spills. In the CBA, the benefits of the implemented management measures are calculated as expected decreases in these costs and losses. For example, the automatic alarm system will decrease the expected amount of oil in the environment, which in turn results in lower expected oil combating costs and smaller environmental damages compared to the situation without the alarm system. All benefits and costs have been expressed as annual expected values by multiplying the costs of a single accident by the number of accidents per year. We chose to use expected values when calculating the costs and benefits, as this minimizes the information loss related to discretization of the variables and in order to avoid computational problems resulting from very large CPTs (see 5. Discussion and conclusions). The benefit and cost figures that originated from different years were made comparable by converting them into 2014 euros using the consumer price index from the Statistics Finland (2014). The cost data and other assumptions used in the calculations were based on previous published studies, assessments and reports related to the oil spill response operations in the Gulf of Finland (Table 1). In addition, expert knowledge was used when there was no published information available. 3.2.1. Purchase costs Alarm system cost is assumed to result only from the development of the alarm system, i.e. there are no additional operating costs. Based on an expert knowledge, the development cost was estimated to lie between 30 000 and 500 000 V, and the system was assumed to be in use for ten years (pers. comm. Tuomas Martikainen, Finnish Transport Agency). In order to get the lives of the management options commensurate with each other (30 years


4

I. Helle et al. / Journal of Environmental Management xxx (2015) 1e11 Table 1 Unit costs of oil combating operations. AeF refer to different classes of boats used in oil combating operations. Activity Shoreline operations Mechanical clean-up Manual clean-up Additional personnel cost (food etc.) Protective clothing Bags Booms Offshore Coastal Boats A B C D E F

Unit cost

Reference

110 V/h 20 V/h/person 20 V/d/person 50 V/person; 8 V/d/person 0.05 V (10 l); 0.2 V (100 l)

Partila Partila Partila Partila Partila

260 V/m (height 1.5 m) 49 V/m (average)

Finnish Environment Institute (2014) Finnish Environment Institute (2014)

100 160 160 200 230 330

V/h V/h V/h V/h V/h V/h

Expert Expert Expert Expert Expert Expert

(2010) (2010) (2010) (2010) (2010)

and and and and and

references references references references references

estimation estimation estimation estimation estimation estimation

Open sea operations Combating vessels Halli Hylje Louhi Oili I Oili II Oili III Seili Merikarhu Tursas Uisko Air surveillance aircraft: Dornier 228 Auxiliary vessel

27 000 V/d 27 000 V/d 50 000 V/d 2900 V/d 2900 V/d 2900 V/d 11 000 V/d 102 720 V/d 102 720 V/d 102 720 V/d 6040 V/h 12 500 V/d

Finnish Environment Finnish Environment Finnish Environment Finnish Environment Finnish Environment Finnish Environment Finnish Environment Finnish Environment Finnish Environment Finnish Environment Finnish Environment Expert estimation

Waste treatment operations Storage: solid waste Processing: solid waste Storage: liquid waste Processing: liquid waste Transportation

35 V/t 340 V/t 20 V/t 150 V/t 2.7 V/km

Partila Partila Partila Partila Partila

for the new vessel), we used the roll-over method (Boardman et al., 2014) and assumed that the investment in the alarm system is repeated three times every 10 years. Vessel acquisition cost describes the acquisition price of a new combatting vessel calculated per year. According to the Finnish authorities, the acquisition price of the vessel was 48 million V and the vessel could be in use for 30 years (pers. comm. Heli Haapasaari, Finnish Environment Institute). As the purchasing costs of both the alarm system and the new vessel incur one-time at the first time period (year), they have been converted to annual values by dividing the purchasing cost by the annuity factor. We used 3.5 percent as the real discount rate to accord with the rates typically used for evaluating public projects (Boardman et al., 2014; HM Treasury, 2003). However, we also tested the sensitivity of the results with discount rates of 1 and 7%. 3.2.2. Maintenance costs The preparedness and maintenance of all 15 Finnish oil recovery vessels costs 3.4 million V and 400 000 V per year, respectively (pers. comm. Heli Haapasaari, Finnish Environment Institute). This amount was divided by the number of the vessels (15) and multiplied by the number of the vessels describing the fleet located in the GoF (nine, or ten, when the new vessel is in service). 3.2.3. Shoreline operations costs Shoreline clean-up costs include costs related to mechanical and manual removal of oil and the use of oil booms and boats in the vicinity of the shoreline. It was assumed that 80% of the waste in the shoreline is collected mechanically and 20% manually (Partila,

therein therein therein therein therein

(2010) (2010) (2010) (2010) (2010)

and and and and and

Institute Institute Institute Institute Institute Institute Institute Institute Institute Institute Institute

references references references references references

(2014) (2014) (2014) (2014) (2014) (2014) (2014) (2014) (2014) (2014) (2014)

therein therein therein therein therein

2010). All unit costs are presented in Table 1. The rate of mechanical clean-up, carried out with loading shovels equipped with oil recovery buckets, was set to 5 t/h (Partila, 2010), and cubic meters were transformed to tons by assuming the volume of a ton of waste to be 1.15 m3 (Halonen, 2007). Manual clean-up costs include the costs related to manual cleanup of oil, other personnel costs and bag costs. It was presumed that 700 people collect oil at the same time, and the total number of people involved is 1000 (Partila, 2010). However, the number of people working increases gradually, i.e. in the first day there are only 30 persons collecting oil, and the full capacity is reached in 20 days (Ministry of the Environment, 2011). The average collecting rate was set to 600 l/d/person (Partila, 2010). It was assumed that bags are half-filled with oily waste, and each container is equipped with two bags (Partila, 2010). The bags sizes of 10 l (0.005 m3 when half-filled), 100 l (0.05 m3) and 140 l (0.07 m3) were used. After an oil accident, booms are used by the regional rescue departments to prevent oil from stranding, to protect valuable areas and to deflect oil to areas where it can be recovered. The amount of booms and boats to be applied depends on the size of the spill. The number of booms in the GoF was obtained from Jolma (2009). Boats are used for various activities related to booming and transportation of combating personnel, and they can be classified into six classes (e.g. A is a small general-purpose boat for transporting people and light equipment, F is a seaworthy boat capable for oil recovery). The total number of boats was gained from Kinnunen and Lajunen (2010) and experts responsible for oil combating in the area. Based on the discussions with experts, we



assumed that all larger boats (DeF) are in use until oil strands, after which all small boats (AeC) and 50% of the larger boats are in use until manual shoreline clean-up activities are finished. 3.2.4. Open sea operations costs Vessel costs include the true daily rental prices of the fleet carrying out mechanical recovery, which range from 2900 V/d to 102 720 V/d depending on the vessel (Table 1, Finnish Environment Institute, 2014). The Finnish Border Guard has two Dornier 228 surveillance aircrafts, which are used to locate separate oil slicks and the areas of thickest slicks, i.e. areas where mechanical recovery is most efficient. It was assumed that one of the airplanes will fly 6 h/d as long as offshore oil combating operations are underway. The auxiliary tanker acts as a temporary depository, into which oil combating vessels can empty their tanks and thus avoid timeconsuming visits to the mainland. The average cost of the auxiliary vessel was set to 12 500 V/d according to an expert estimation €rvi, Finnish Environment Institute). (pers. comm. Jouko Pirttija 3.2.5. Waste treatment costs Waste treatment costs cover costs related to storage, transportation and final processing of oily waste. It was assumed that all the waste is stored before transportation, and material collected from shoreline is solid waste and material from open sea is liquid waste (Partila, 2010). The calculation of the transportation costs were based on the assumption that solid waste is transported with trucks having 30 tons cargo, and liquid waste with trucks of 25 tons cargo (Partila, 2010). Further, the transportation distance was set to 258 km, which was the average of the distances (Partila, 2010) presented as a round trip from the storage locations along the coast to the €ki in the southern Finland. Ekokem treatment plant in Riihima 3.2.6. Environmental damages Environmental damages describe the losses people experience due to the negative impacts spilled oil has on the environment, either in the sea or after stranding. The estimate presents the marginal damage of a ton of spilled oil and it is calculated in the sub-model “Damage”, which has seven random variables (Fig. 2). The calculation of marginal damage is based on a contingent valuation survey conducted in 2006 (Ahtiainen, 2007; Juntunen et al., 2013). With a sample of 360 respondents, the study examined the Finnish citizens' willingness to pay (WTP) for improvements in oil spill response capacity in the GoF. In the survey, respondents were given a description of the possible negative

Fig. 2. Sub-model “Damage”. The variable “Marginal damage” is used as an input in the main model.

5

effects of a major tanker accident and explained how the oil spill response capacity would be improved, after which they were asked whether they would be willing to pay in order to decrease the harm to nature and recreation possibilities by improving oil spill response capacity. The payment vehicle was a tax increase in the year 2007 (a lump sum), to be collected from all Finnish taxpayers, and dichotomous choice was used as the elicitation format.1 The questionnaire specified that the measures would improve mechanical oil recovery, which would reduce the oiled area both in the open sea and on the coast. The estimated WTP was used to derive the marginal damage of a ton of spilled oil based on following steps and assumptions: 1) There is a spill size that corresponds to the negative environmental impacts described in the contingent valuation survey; 2) The respondents assume a certain oil combating efficiency before any new investments (5e85%); 3) The respondents assume that new investments will improve the initial combating efficiency, and thus reduce oil and resulting harms in the environment, by a certain amount (30e70%); 4) The respondents assume that new investments will improve oil combating efficiency in a certain number of accidents (1, 6, 10, 15); 5) By combining the information of steps 1e4, a total amount of additional oil recovered over a given number of accidents can be calculated (i.e. the difference between recovered oil with and without improved efficiency); 6) The marginal value of a ton of recovered oil is calculated by dividing the aggregate WTP with the amount of oil calculated in step 5. The result can be interpreted as a marginal damage of a ton of oil that remains in the environment. To account for the uncertainty in the size of the spill in step 1, we applied a probabilistic Bayesian network model that describes the effects of major oil accidents of various sizes on several groups of organisms living in the GoF (Lecklin et al., 2011). As a BN, the model enables backward calculation, i.e. the spill size can be determined by setting the ecological effects into certain states and allowing the model to calculate backwards from effects to causes. We instantiated the variable “All studied groups are fully recovered within 10 years” into TRUE state, which corresponded well to the statement “The duration of the injuries to the animal and plant populations is 10 years at most, if another oil spill does not happen” in the WTP survey. As the original distribution in Lecklin et al. (2011) was discretized into fairly coarse classes, we fitted a log-normal distribution with appropriate parameters (mean mSS ¼ 3250, sdSS ¼ 7650) to be used in subsequent analysis (Fig. 3). The size of the spill was then simulated from this distribution using the number of accidents given in step 4. We assumed a uniform distribution for the variables in steps 2e4 (“Recovery efficiency (WTP)”, “Improvement factor (WTP)” and “Number of accidents (WTP)”), which represents our high uncertainty related to the variables. The aggregate WTP was estimated in Juntunen et al. (2013), who produced a probabilistic interpretation of the results of the original WTP survey (Ahtiainen, 2007). Thus, we used a gamma distribution for the WTP with parameters a and b (a ¼ m2WTP/sd2WTP and b ¼ mWTP/sd2WTP), and the values for mWTP and sdWTP were set to 122 000 000 V and 15 000 000 V, respectively (Juntunen et al., 2013). The simulation model was implemented and run with OpenBugs 3.2.2 (Gilks et al., 1994). We ran the model with 500 000

1 In the pretesting of the survey instrument, respondents favored taxes as a payment vehicle over voluntary donations.


6


Fig. 3. The original distribution of the spill size from Lecklin et al. (2011) (blue columns) and a corresponding log-normal distribution used in our analysis (black line).

iterations, and the results were discretized and transferred to the sub-model “Damage”. The procedure thus produced probability distributions for the variables “Additional efficiency (WTP)”, “Additional recovery (WTP)”, “WTP” and “Marginal damage” (Fig. 4).2 4. Results The model can be used to study several questions related to costs of tanker accidents. First we present results of the cost-benefit analysis and the value-of-information analysis, after which we review the costs of a single accident. 4.1. Cost-benefit analysis The results show that based on the expected yearly utilities, the implementation of the automatic alarm system is an economically viable decision as the benefits exceed the costs, whereas the costs of the new vessel exceed the benefits (Table 2). The magnitudes of the expected benefits of the implementation of the alarm system, calculated as a decrease in the expected costs and damages resulting from accidents, differ between the cost factors and vary from as low as 28 V (Auxiliary vessel cost) to 280 000 V (Environmental damages). Further, the expected benefits are slightly different depending on what we assume about the new vessel, i.e. is it in service or not (Table 2). When we assume that the new vessel is in use, the benefits of the alarm system related to open sea operation are slightly greater (or the same) than in a situation when the new vessel is not in use. The benefits of a lower accident frequency are thus greater in a situation where offshore combating costs are higher. However, the situation is the opposite when the benefits related to shoreline cleanup activities are considered: the benefits of the alarm system are smaller with the new vessel, as smaller amount of oil reaches the shoreline causing cleanup costs.

Fig. 4. Probability distributions for the variables in the sub-model “Damage”.

Similarly to the alarm system, the expected benefits and costs of the new vessel are dependent on the cost factors, and range from 679 V to 830 000 V, and 993 V and 2 720 000 V, respectively. Again, the expected benefits and costs differ slightly depending on the implementation of the other management measure: regarding the offshore combating costs, the costs related to the new vessel are greater when the alarm system is not implemented (i.e. with higher accident frequencies) as the expected amount of oil to be recovered is higher. At the same time also the benefits related to shoreline operations are a bit higher, as the new vessel decreases the amount of oil ashore. Although the CBA suggests that the new oil combating vessel is not an economically efficient acquisition, the total expected benefits of the new vessel are approximately 2.9-fold compared to the benefits of the automatic alarm system (Table 2). This stems directly from the ability of the management measures to decrease the amount of oil afloat. Although the expected reduction of oil left in the sea and to be thus washed ashore is fairly small for both measures (4.7% for the new vessel and 1.9% for the alarm system), the cost of a ton of oil in the environment is so high that the expected benefits are significant. However, as the purchase cost of the new vessel is high, the measure seems to be less reasonable in economic terms. As mentioned before, the range of the expected benefits is large. The inclusion of the environmental damages expressed as people's willingness to pay for improved combating capacity has a clear effect on the results: without the variable, the automatic alarm system would not be an economically justifiable investment.

2 It is worth noting that the marginal damage (damage per unit of spilled oil) is constant, i.e. it does not depend on the magnitude of the spill.



7

Table 2 The expected yearly costs and benefits of the automatic alarm system (AAS) and the new oil combating vessel (NV) (V). The values are calculated separately for situations, where the other management measure is either implemented or not. Decision option

AAS ¼ implemented

State of the other option

NV ¼ No

Costs/Benefits

Costs

Acquisition cost (ASS) Purchase cost (NV) Maintenance cost (NV) Offshore activities Vessel cost Auxiliary vessel cost Air survaillence cost Treatment cost: open sea Shoreline activities Mechanical cleanup cost Manual cleanup cost Bag costs Boat cost Boom cost Treatment cost: shore

33 200

Benefits

AAS ¼ No

Costs

Costs

Benefits

AAS ¼ implemented Benefits

Costs

Benefits

33 200

Environmental damage Total Net benefit value

NV ¼ implemented NV ¼ implemented

33 200

2 720 000 264 000

2 720 000 264 000

17 200

17 000

1010

993

1770 28 167 518

2000 28 167 538

379 1230 241 1230 3760 8210

357 1160 227 1190 3760 7740

23 600

23 100

280 000

270 000

830 000

820 000

298 000 265 000

33 200

4.2. Value-of-information analysis To analyze the effects of uncertainty related to individual random variables on the results, we conducted a value-ofinformation analysis (VoI) based on the value of perfect information (Raiffa and Schlaifer, 1961). A variable with VoI has some state(s), which, if known exactly, result in a greater expected utility with a different decision compared to the maximum expected utility (MEU) gained with the optimal decision without any new information. The VoI analysis was conducted with the Hugin software. The decision on the automatic alarm system seems to be insensitive to the uncertainties related to the variables, as none of the variables have VoI. However, the situation is different with the decision about the new vessel, as there are 14 variables that have VoI, the range of values being 703e539 000 V (Table 3). The variables with the highest values describe the number of accidents (“Leaking TOT”, “Leaking Gs”, “Leaking TOs”), the size of the spill (“Leak TOT”, “Leak G”) and the variables dependent on the latter (“Amount of oil”, “Boom and boat usage”). Also “Marginal damage” has relatively high VoI. Thus, if we knew these variables more precisely, it could change the results of the CBA, i.e. also the new vessel could be an economically justified option.

Table 3 The results of VoI analysis related to the decision on the new vessel (V). State of AAS

No

Yes

Leaking TOT Amount of oil Leak TOT Leaking Gs Leaking TOs Boom and boat usage Leak from G Marginal damage Wave height Leak from C Leaking TTs Number of TTs Number of TOs Number of Gs

539 000 398 000 347 000 272 000 261 000 188 000 149 000 131 000 90 000 88 000 45 000 27 000 22 000 1000

526 000 387 000 335 000 271 000 250 000 173 000 148 000 124 000 87 000 81 000 43 000 25 000 20 000 1000

287 000 254 000

1090 3430 693 2470

3 000 000

861 000 2 140 000

1070 3360 679 2420

3 000 000

851 000 2 150 000

Yet, it is important to notice that the variables also exhibit true randomness. For instance, the number of accidents per year is always uncertain. Hence, the basic idea of the VoI analysis is not fully applicable to all of the variables. However, with the variable related to environmental damages (“Marginal damage”) new knowledge would probably be able to decrease uncertainty, as the calculation was based on uniform distributions. Thus, better knowledge of this variable may have an effect on the decision about the new vessel. 4.3. Costs of a single accident In addition to a CBA, the model can be used to analyze the costs of a single accident. This information is interesting as such, but it also offers a way to validate, at least partially, the results of the model. To give an overview of the functioning of the model, we present the total costs of clean-up and waste treatment operations for two accident scenarios. The scenarios are: 1) A 3000 m3 spill of heavy oil, and 2) a 30 000 m3 spill of medium oil. The first scenario corresponds approximately to the most recent large oil spill in the Baltic Sea, i.e. the Baltic Carrier accident in Denmark in 2001 (IOPC Funds, 2009) and the second scenario to the worst-case scenario for the GoF according to the Finnish Environment Administration (Hietala and Lampela, 2007). The calculations include the new combating vessel, already in operation in the GoF, in order to give as realistic result as possible. The expected costs related to open sea oil combating, cleanup and waste treatment operations of selected scenarios are presented in Table 4. The costs rise substantially as the spill size increases. This holds true especially for the costs of the shoreline activities and waste treatment operations. The general picture is clear: cleaning up the shoreline costs much more than combating oil at sea. As pointed out by Montewka et al. (2013), the validation of any model dealing with oil spills in the GoF is challenging, as, for the time being, there have not been any large-scale accidents in the area. Therefore, to validate our model, we compared our results to the results of other models designed to analyze the costs of oil spills. However, the majority of the models available either cover total costs (e.g. Psarros et al., 2011; Ventikos and Sotiropoulos, 2014; Yamada, 2009), in which case it can be unfeasible to


8


Table 4 The expected costs of different oil combating activities in two accident scenarios (V), and the comparison with other studies. Scenario

1

2

Spill volume (m3)

3000

30 000

Oil type

Heavy

Medium

1 500 000 4 000 000

1 500 000 14 700 000

500 000 2 700 000

3 700 000 29 100 000

2 000 000 6 700 000 8 700 000

5 200 000 43 800 000 49 000 000

5 700 000 8 700 000

23 400 000 74 600 000

Combating operations Open sea Shoreline Waste treatment operations Open sea Shoreline All cleanup operations Open sea Shoreline TOTAL Comparison to other studies Kontovas et al. (2010) a Shahriari and Frost (2008) b

a To convert tons to cubic meters, we used specific densities of 0.975 t/m3 (heavy oil) and 0.871 t/m3 (medium oil). The exchange ratio EUR/USD was set to 1.3257 (2010). b To convert tons to cubic meters, we used specific densities of 0.975 t/m3 (heavy oil) and 0.871 t/m3 (medium oil). The exchange ratio EUR/USD was set to 1.4708 (2008). The level of preparedness was set to 3 (the Baltic Sea).

determine the share of the combating and clean-up operations, or the required input includes parameters not present in our model (Etkin, 1999, 2000). The two models that seem suitable for our purposes (although the coverage of the term “clean-up costs” is somewhat vague) include the model presented by Kontovas et al. (2010), which is based on the weight of oil spill, and the model by Shahriari and Frost (2008), which has three input parameters, i.e. spill weight, oil density and the level of preparedness. In scenario 1, our model produces estimates similar to the model of Shahriari and Frost (2008), whereas in scenario 2 our results lie between the results of two other models (Table 4). In the latter the results of the three models differ substantially as the results of our model and Shahriari and Frost (2008) are approximately 2 and 3.5 times higher, respectively, than the estimates of Kontovas et al. (2010). It is important to notice that if we go beyond expected values, uncertainty related to costs is high in both scenarios, i.e. the probability distributions of the costs are wide. E.g. although the expected total costs in scenario 1 are slightly below 9 million V, there is 3.5% chance that the treatment costs of shoreline oily waste alone exceed 10 million V (see Table S2 in Supplementary Material for complete probability distributions). 5. Discussion and conclusions We have developed a probabilistic model for estimating the costs and damages of oil spills and applied it to a cost-benefit analysis of spill-related management measures in the Gulf of Finland. The results of our model suggest that an automatic alarm system integrated in the existing Vessel Traffic Service system would be an economically viable acquisition, whereas a new oil combating vessel would not succeed in being such. The result is explained by the low purchase price of the alarm system compared to the new vessel, as the latter is actually more effective in decreasing the expected amount of oil in the environment. There are some issues that should be taken into account when analyzing the results. First, it is evident that the results of the CBA depend on how extensively we can include costs and benefits in the analysis. For instance, if we exclude the losses in environmental values from the analysis, neither management option is

economically efficient. On the other hand, the model lacks some factors which could increase the benefits of both management options substantially. For example, the benefits to the fisheries and tourism sectors and the possible effects on the prices of real estates were excluded, as no monetary estimates for the effects were available for the study area due to the lack of major oil accidents in the GoF in the past. Further, as our model covers only groundings and collisions of oil tankers, involves only spills from cargo tanks and excludes winter accidents, it is evident that the benefits of a new combating vessel are not taken into account in full. Second, the purchase cost of the new vessel is allocated entirely to oil combating. Yet, new vessels are multipurpose vessels, which are used for various types of activities such as the laying of undersea cables, diving operations, and diverse maintenance measures (Finnish Environment Institute, 2011). Actually, if 20% or less of the cost is allocated to oil combating, the model shows that the benefits of acquiring the new vessel exceed the costs. However, there are also matters that would increase the superiority of preventive measures if taken into account. These include potential costs related e.g. to vessel damages, lost or damaged cargo, injured people and the loss of human lives. These costs will be avoided altogether if an accident is prevented. The results of the cost of a single accident show that the costs resulting from waste treatment operations can be multi-fold compared to the costs of open sea oil combating and shoreline clean-up activities, and thus it is important not to limit the inspection to direct clean-up costs when estimating the costs of oil spills (e.g. Montewka et al., 2013). The value-of-information analysis shows that the result of the CBA for the new vessel is sensitive to certain variables, especially those related to groundings and marginal environmental damages. With the latter, the result encourages studying the subject in more detail: in principle the VoI analysis reveals that it would be worth to pay approximately 128 000 V in order to know the “true” marginal environmental damages, as it might lead to higher maximum expected utility. With groundings the situation is more complicated, as the uncertainty present in the variables represents not only the lack of precise information but also true randomness. However, as our accident probabilities for groundings are based on statistics that cover a relatively short period of time (see Supplementary Material) the results of the VoI analysis suggest that our analysis would benefit from a better understanding about the dynamics of groundings in the GoF. Thus, the results can be used to focus the future research efforts. The cost estimates of single accidents produced by the model seem reasonable when compared to other models. Our model produces higher cost estimates than the model by Kontovas et al. (2010) but lower estimates than the model by Shahriari and Frost (2008), especially with large spill sizes. The low estimates of Kontovas et al. (2010) may be at least partly explained by the fact that the cost data used in their analysis do not necessarily represent actual costs but rather indicate the amount of money that has been agreed to be a reasonable compensation for the claimants (Kontovas et al., 2010). Further, the suitability of the model by Shahriari and Frost (2008) for oil spills occurring in the GoF may be somewhat limited as their model has a low geographical resolution, and different nations within a region share the same level of oil spill preparedness. Finland has a large fleet of oil combating vessels, and thus, in favorable weather conditions, oil is recovered mainly at open sea. This reduces the costs of expensive and slow shoreline clean-up. The importance of the open sea combating capacity can be tested in the model by assuming that there is only one day to recover oil before it reaches the shoreline. This has a significant effect: the expected total costs of oil combating, clean-up and treatment of a 30 000 m3 spill would now be over 115 million V



instead of 49 million V. Also Montewka et al. (2013) have estimated the clean-up costs of oil spills in the GoF. However, their model produces much higher cost estimates than our model: 10 600 000 V and 93 400 000 V for scenarios 1 and 2, respectively. However, a major part (98e99%, calculated from the mean values) of the costs of shoreline operations in the work of Montewka et al. (2013) result from the use of boats owned by the rescue departments. In our model the share is 13e44%, depending on whether or not the costs of oil booms are taken into account. We believe our results are realistic, as it has been estimated that within the first 28 days after a large-scale oil accident in the GoF, the boats comprise 8e55% of the daily shoreline costs (boats and clean-up activities included) (Ministry of the Environment, 2011). Our model has several advantages that make it a convenient tool to produce cost estimates and to conduct a CBA. In addition to the explicit handling of uncertainty, the causal structure of the model enables the inspection of various separate factors, such as season, wave height, oil type etc., on the costs. Further, as Bayesian networks consist of separate variables, they can be updated easily if new information becomes available, and also the inclusion of new utility variables related to e.g. fisheries and tourism is straightforward. Also some technical features inherent to Bayesian networks are convenient, e.g. the ability to use various kinds of information within a same network. However, BNs have also some relevant limitations regarding to our study. First, model validation is a challenge if the model aims at predicting the outcomes of management measures which have not been applied before (e.g. Barton et al., 2008), or deals with accidents that have not yet happened. This holds true also in our case, as the automatic alarm system would be a new innovation. However, we have attempted to validate the rest of the model comparing the results to other available models. Second, the discretization of the variables may have, at least a slight, effect on the results, as it may result in computational artefacts and the loss of information (Barton et al., 2008; Nash and Hannah, 2011; Uusitalo, 2007). This encourages applying as dense discretization intervals as possible. However, in our case this is challenging, as the range of the possible spill sizes is large, and thus the number of states per variable would be considerable. We have solved this dilemma by using fairly dense discretization in variables related to the amount of oil spill, recovered oil and waste amount. Furthermore, the expected utilities were calculated directly from the variables related to oil, oily waste etc. thus avoiding the loss of information related to the discretization of this additional layer of variables without ending up in computationally unfeasible situations. It is evident that the economic viability of the management measures depends crucially on the estimated environmental damages. While the damages were based on a state-of-the-art valuation study, they were not specifically designed to be used in a cost-benefit analysis. This was reflected in the wide distributions of the variables in the simulation model that estimates the marginal damage (see Supplementary Material). Due to the importance of environmental damages to our results, we suggest that future cost-benefit analyses on oil spill management should incorporate environmental damages, and oil spill valuation studies should estimate values that can be easily linked to quantifiable indicators of oil spills, such as the amount of oil spilled or the length of oiled coastlines. Although the model is developed for oil spill management purposes, the idea and the methodology are suitable also for CBAs in other fields where the probability of the adverse event is low, uncertainty plays a major role, and potential costs and losses can be high. Our approach also offers one option to carry out the Formal

9

Safety Assessment (FSA), a process aimed at enhancing maritime safety by using risk analysis in the combination of cost benefit analysis (International Maritime Organization, 2002). In addition to the management options included in our study, it would be interesting to apply the approach also to other measures that are able to decrease the risks related to maritime traffic in the Baltic Sea. These include e.g. the ENSI service (a navigational support system for oil €nninen et al., 2014) and the optimization of fairways in tankers, Ha order to minimize the probability of coastal pollution (see e.g. Lehmann et al., 2014; Soomere et al., 2011a, 2011b; Soomere and Quak, 2013). Finally, it is also important to note that a CBA based on expected utilities is not a comprehensive decision analysis and it cannot be used alone to seek the “best” alternative. Instead of expected utilities, a decision-maker can prefer e.g. to minimize (maximize) the chance of the worst (best) possible outcome (Morgan and Henrion, 1990). Further, in addition to economic efficiency, one might require e.g. that independent of the costs and benefits, the amount of oil left in the environment at any circumstances should not exceed a certain limit, in which case it is reasonable to invest in measures that help achieve this target. These, potentially opposing, objectives can be taken into account via multi-criteria decision analysis (Belton and Stewart, 2002). To conclude, the allocation of resources to mitigate negative impacts of human activities is not a trivial task, and a cost-benefit analysis can be used to assist the decision-making process. However, if uncertainties related to the topic at issue are high, they should be taken into account in a coherent and explicit manner. The developed approach can be used to find management measures that are profitable even in situations where uncertainties are high. Our study emphasizes the importance of the environmental values when estimating the costs and benefits of management measures. Acknowledgments This research was conducted as a part of the project “Protection of the Baltic Sea: Benefits, Costs and Policy Instruments (PROBAPS)”. The authors wish to thank the four Finnish ministries that financed the project: Ministry of Agriculture and Forestry, Ministry of the Environment, Ministry of Transport and Communications, and Ministry of Finance. We also thank all the experts, who provided us valuable information on oil combating: Heli Haapasaari €rvi (Finnish Environment Institute), Tuomas and Jouko Pirttija Martikainen (Finnish Transportation Agency) and Ville Estlander (Helsinki City Rescue Department). Olli-Pekka Brunila and Jenni Storgård (University of Turku) are thanked for providing the data on the oil types transported in the Gulf of Finland, and Jarno Vanhatalo (University of Helsinki) for helping with Fig. 3. Appendix A. Supplementary data Supplementary data related to this article can be found at http:// dx.doi.org/10.1016/j.jenvman.2015.04.042. References Aguilera, P.A., Fernandez, A., Fernandez, R., Rumi, R., Salmeron, A., 2011. Bayesian networks in environmental modelling. Environ. Model. Softw. 26, 1376e1388. Ahtiainen, H., 2007. The Willingness to Pay for Reducing the Harm from Future Oil Spills in the Gulf of Finland e an Application of the Contingent Valuation Method. Discussion Papers/University of Helsinki, Department of Economics and Management, 18, University of Helsinki, Helsinki. Allo, M., Loureiro, M.L., 2013. Estimating a meta-damage regression model for large accidental oil spills. Ecol. Econ. 86, 167e175. Alvarez, S., Larkin, S.L., Whitehead, J.C., Haab, T., 2014. A revealed preference approach to valuing non-market recreational fishing losses from the Deepwater Horizon oil spill. J. Environ. Manag. 145, 199e209.


10


Ames, D.P., Neilson, B.T., Stevens, D.K., Lall, U., 2005. Using Bayesian networks to model watershed management decisions: an East Canyon Creek case study. J. Hydroinf. 7, 267e282. Aps, R., Fetissov, M., Herkul, K., Kotta, J., Leiger, R., Mander, U., Suursaar, U., 2009. Bayesian inference for predicting potential oil spill related ecological risk. Saf. Secur. Eng. III 108, 149e159. Barton, D.N., Saloranta, T., Bakken, T.H., Solheim, A.L., Moe, J., Selvik, J.R., Vagstad, N., 2005. Using Bayesian network models to incorporate uncertainty in the economic analysis of pollution abatement measures under the water framework directive. Water Econ. Stat. Financ. 5, 95e104. Barton, D.N., Saloranta, T., Moe, S.J., Eggestad, H.O., Kuikka, S., 2008. Bayesian belief networks as a meta-modelling tool in integrated river basin management - pros and cons in evaluating nutrient abatement decisions under uncertainty in a Norwegian river basin. Ecol. Econ. 66, 91e104. Bateman, I.J., Carson, R.T., Day, B., Hanemann, W.M., Hanley, N., Hett, T., Jones€ Lee, M., Loomes, G., Mourato, S., Ozdemiroglu, E., Pearce, D.W., Sugden, R., Swanson, J., 2002. Economic Valuation with Stated Preference Techniques: a Manual. Edward Elgar Publishing, Cheltenham. Belton, V., Stewart, T.J., 2002. Multiple Criteria Decision Analysis. An Integrated Approach. Kluwer Academic Publishers, Boston. Biedermann, A., Taroni, F., 2012. Bayesian networks for evaluating forensic DNA profiling evidence: a review and guide to literature. Forensic Sci. Int. Genet. 6, 147e157. Boardman, A., Greenberg, D., Vining, A., Weimer, D., 2014. Cost-benefit Analysis. Concepts and Practice, fourth ed. Pearson Education Limited, Essex, England. Bockstael, N., McConnell, K., 2007. Environmental and Resource Valuation with Revealed Preferences: a Theoretical Guide to Empirical Models. Springer, Dordrecht, The Netherlands. Bromley, J., Jackson, N.A., Clymer, O.J., Giacomello, A.M., Jensen, F.V., 2005. The use of Hugin((R)) to develop Bayesian networks as an aid to integrated water resource planning. Environ. Model. Softw. 20, 231e242. Burgman, M., 2005. Risks and Decisions for Conservation and Environmental Management. Cambridge University Press, Cambridge. Carriger, J.F., Barron, M.G., 2011. Minimizing risks from spilled oil to ecosystem services using influence diagrams: the Deepwater Horizon spill response. Environ. Sci. Technol. 45, 7631e7639. Carson, R.T., Mitchell, R.C., Hanemann, M., Kopp, R.J., Presser, S., Ruud, P.A., 2003. Contingent valuation and lost passive use: damages from the Exxon Valdez oil spill. Environ. Resour. Econ. 25, 257e286. Cohen, M.A., 1986. The costs and benefits of oil-spill prevention and enforcement. J. Environ. Econ. Manag. 13, 167e188. Etkin, D.S., 1999. Estimating cleanup costs for oil spills. In: 1999 International Oil Spill Conference, Seattle, Washinton, US. Etkin, D.S., 2000. Worldwide analysis of Marine oil spill cleanup cost factors. In: 23rd Arctic and Marine Oilspill Program (AMOP) Technical Seminar, Alberta, Canada. Fenton, N., Neil, M., 2013. Risk Assessment and Decision Analysis with Bayesian Networks. CRC Press, Boca Raton. Finnish Environment Institute, 2011. New Oil and Chemical Spill Response Vessel was Given the Name Louhi. Press Release. Finnish Environment Institute, Helsinki. Finnish Environment Institute, 2013. Maritime Accident Risks and Response Cases. http://www.ymparisto.fi/en-US/Waters/Environmental_emergency_response_ in_Finland/Marine_pollution_response/Maritime_accident_risks_and_ response_cases (Last accessed 30 Novermber 2014). Finnish Environment Institute, 2014. Oil Recovery Equipment in Finland. http:// wwwi.ymparisto.fi/oilspill/helcom/equipm.htm (Last accessed 30 November 2014). Forsberg, J.A., Eberhardt, J., Boland, P.J., Wedin, R., Healey, J.H., 2011. Estimating survival in patients with operable skeletal metastases: an application of a Bayesian belief network. Plos One 6, e19956. IOPC Funds, 2009. Annual Report 2008. Report on the Activities of the International Oil Pollution Compensation Funds in 2008. International Oil Pollution Compensation Funds, London, UK. Garza-Gil, M.D., Prada-Blanco, A., Vazquez-Rodriguez, M.X., 2006. Estimating the short-term economic damages from the Prestige oil spill in the Galician fisheries and tourism. Ecol. Econ. 58, 842e849. Gilks, W.R., Thomas, A., Spiegelhalter, D.J., 1994. A language and program for complex Bayesian modeling. Statistician 43, 169e177. Goerlandt, F., Montewka, J., 2014. A probabilistic model for accidental cargo oil outflow from product tankers in a ship-ship collision. Mar. Pollut. Bull. 79, 130e144. Grigalunas, T.A., Anderson, R.C., Brown, G.M., Congar, R., Meade, N.F., Sorensen, P.E., 1986. Estimating the cost of oil spills: lessons from the Amoco Cadiz incident. Mar. Resour. Econ. 2, 239e262. Haapasaari, P., Karjalainen, T.P., 2010. Formalizing expert knowledge to compare alternative management plans: sociological perspective to the future management of Baltic salmon stocks. Mar. Policy 34, 477e486. € O € e Toimintamalli suuren o €ljyntorjuntaoperaation koordiHalonen, J., 2007. SOK €ljyntorjunnasta vastaaville viranomaisille. Kymenlaakson nointiin rannikon o ammattikorkeakoulun julkaisuja. In: Sarja A. Oppimateriaali 15. Kymenlaakso University of Applied Sciences, Kotka. Hanley, N., 1992. Are there environmental limits to cost benefit analysis? Environ. Resour. Econ. 2, 33e59. €nninen, M., 2014. Bayesian networks for maritime traffic accident prevention: Ha

benefits and challenges. Accid. Anal. Prev. 73, 305e312. €nninen, M., Kujala, P., 2012. Influences of variables on ship collision probability in Ha a Bayesian belief network model. Reliab. Eng. Syst. Saf. 102, 27e40. €nninen, M., Kujala, P., 2014. Bayesian network modeling of Port State Control Ha inspection findings and ship accident involvement. Expert Syst. Appl. 41, 1632e1646. €nninen, M., Mazaheri, A., Kujala, P., Montewka, J., Laaksonen, P., Salmiovirta, M., Ha Klang, M., 2014. Expert elicitation of a navigation service implementation effects on ship groundings and collisions in the Gulf of Finland. Proc. Inst. Mech. Eng. Part O J. Risk Reliab. 228 (1), 19e28. Helle, I., Lecklin, T., Jolma, A., Kuikka, S., 2011. Modeling the effectiveness of oil combating from an ecological perspective e a Bayesian network for the Gulf of Finland; the Baltic Sea. J. Hazard. Mater. 185, 182e192. Henriksen, H.J., Kjaer, J., Brush, W., Jacobsen, L.B., Jensen, J.D., Grinderslev, D., Andersen, P., 2007. Environmental benefits and social cost e an example of combining Bayesian networks and economic models for analysing pesticide management instruments. Nord. Hydrol. 38, 351e371. Hietala, M., Lampela, K., 2007. Oil Pollution Preparedness on the Open Sea e Final Report of the Working Group. Finnish Environment Institute, Helsinki. Howard, R.A., Matheson, J.E., 2005. Influence diagrams. Decis. Anal. 2, 127e143. International Maritime Organization, 2002. Guidelines for Formal Safety Assessment (FSA) for Use in the IMO Rule-making Process. MSC/Circ.1023/MEPC/ Circ.392. International Organization for Standardization, 2009. ISO Guide 73: Risk Management e Vocabulary. Jensen, F.V., Nielsen, T.D., 2007. Bayesian Networks and Decision Graphs. Springer, New York. € ljyntorjuntavalmiuden Jolma, K., 2009. Kokonaisselvitys valtion ja kuntien o kehitt€ amisest€ a 2009e2018 (Report on the Development of the Governmental and Municipal Oil Combating Preparedness in 2009e2018). Finnish Environment Institute, Helsinki. €nen, J., Kuikka, S., 2005. How to model the oil Juntunen, T., Rosqvist, T., Rytko combatting technologies and their impacts on ecosystem: a Bayesian networks, CM 2005/S:2002. 2005 ICES Annual Science Conference Aberdeen, United Kingdom. €ntyniemi, S., 2013. A Bayesian approach to address Juntunen, T., Ahtiainen, H., Ma statistical errors and uncertainties in single binary choice contingent valuation. In: Juntunen, T., Steps towards Comprehensive Bayesian Decision Analysis in Fisheries and Environmental Management. Doctoral Dissertation. University of Helsinki Kelly (Letcher), R.A., Jakeman, A.J., Barreteau, O., Borsuk, M.E., ElSawah, S., Hamilton, S.H., Henriksen, H.J., Kuikka, S., Maier, H.R., Rizzoli, A.E., Delden, H., Voinov, A.A., 2013. Selecting among five common modelling approaches for integrated environmental assessment and management. Environ. Model. Softw. 47, 159e181. Kinnunen, J., Lajunen, T., 2010. Cleaning and Maintenance of Oiled Equipment for Worst Case Oil Spill Scenario. B.Sc. thesis. Kymenlaakson ammattikorkeakoulu. University of Applied Sciences, Kotka. Kontovas, C.A., Psaraftis, H.N., Ventikos, N.P., 2010. An empirical analysis of IOPCF oil spill cost data. Mar. Pollut. Bull. 60, 1455e1466. Kuikka, S., Hilden, M., Gislason, H., Hansson, S., Sparholt, H., Varis, O., 1999. Modeling environmentally driven uncertainties in Baltic cod (Gadus morhua) management by Bayesian influence diagrams. Can. J. Fish. Aquat. Sci. 56, 629e641. €nninen, M., Arola, T., Ylitalo, J., 2009. Analysis of the marine traffic Kujala, P., Ha safety in the Gulf of Finland. Reliab. Eng. Syst. Saf. 94, 1349e1357. Landuyt, D., Broekx, S., D'hondt, R., Engelen, G., Aertsens, J., Goethals, P.L.M., 2013. A review of Bayesian belief networks in ecosystem service modelling. Environ. Model. Softw. 46, 1e11. Landuyt, D., Lemmens, P., D'hondt, R., Broekx, S., Liekens, I., De Bie, T., Declerck, S.A.J., De Meester, L., Goethals, P.L.M., 2014. An ecosystem service approach to support integrated pond management: a case study using Bayesian belief networks e highlighting opportunities and risks. J. Environ. Manag. 145, 79e87. Langseth, H., Portinale, L., 2007. Bayesian networks in reliability. Reliab. Eng. Syst. Saf. 92, 92e108. € ma €, R., Kuikka, S., 2011. A Bayesian network for analyzing biological Lecklin, T., Ryo acute and long-term impacts of an oil spill in the Gulf of Finland. Mar. Pollut. Bull. 62, 2822e2835. €ntyniemi, S., Kuikka, S., 2013. Optimizing the recovery Lehikoinen, A., Luoma, E., Ma efficiency of finnish oil combating vessels in the Gulf of Finland using Bayesian networks. Environ. Sci. Technol. 47, 1792e1799. Lehikoinen, A., H€ anninen, M., Storgård, J., Luoma, E., M€ antyniemi, S., Kuikka, S., 2015. A Bayesian network for assessing the collision induced risk of an oil accident in the Gulf of Finland. Environ. Sci. Technol. 49 (9), 5301e5309. Lehmann, A., Hinrichsen, H.-H., Getzlaff, K., 2014. Identifying potentially high risk areas for environmental pollution in the Baltic Sea. Boreal Environ. Res. 19 (2), 140e152. Levontin, P., Kulmala, S., Haapasaari, P., Kuikka, S., 2011. Integration of biological, economic, and sociological knowledge by Bayesian belief networks: the interdisciplinary evaluation of potential management plans for Baltic salmon. Ices J. Mar. Sci. 68, 632e638. Loureiro, M.L., Loomis, J.B., 2013. International public preferences and provision of public goods: assessment of passive use values in large oil spills. Environ. Resour. Econ. 56, 521e534.


I. Helle et al. / Journal of Environmental Management xxx (2015) 1e11 Loureiro, M.L., Ribas, A., Lopez, E., Ojea, E., 2006. Estimated costs and admissible claims linked to the Prestige oil spill. Ecol. Econ. 59, 48e63. Loureiro, M.L., Loomis, J.B., Vazquez, M.X., 2009. Economic valuation of environmental damages due to the Prestige oil spill in Spain. Environ. Resour. Econ. 44, 537e553. Madsen, A.L., Jensen, F., Kjaerulff, U.B., Lang, M., 2005. The Hugin Tool for probabilistic graphical models. Int. J. Artif. Intell. Tools 14, 507e543. McCann, R.K., Marcot, B.G., Ellis, E., 2006. Bayesian belief networks: applications in ecology and natural resource management. Can. J. For. Res. Revue Can. De Recherche For. 36, 3053e3062. Millennium Ecosystem Assessment, 2005. Ecosystems and Human Well-being: Synthesis. Island Press, Washington, DC. €ljyvahingoissa. Torjunnan Ministry of the Environment, 2011. Toiminta isoissa aluso €rjesta €minen, johtaminen ja viestinta € (Response actions during major marine ja oil accidents. Organizing and managing the response and communication). Ministry of the Environment, Helsinki. Molina, J.L., Bromley, J., Garcia-Arostegui, J.L., Sullivan, C., Benavente, J., 2010. Integrated water resources management of overexploited hydrogeological systems using object-oriented Bayesian networks. Environ. Model. Softw. 25, 383e397. € m, M., Kujala, P., 2013. A probabilistic model estimating oil Montewka, J., Weckstro spill clean-up costs e a case study for the Gulf of Finland. Mar. Pollut. Bull. 76, 61e71. Morgan, M.G., Henrion, M., 1990. Uncertainty. A Guide to Dealing with Uncertainty in Quantitative Risk and Policy Analysis. Cambridge University Press, Cambridge. Nash, D., Hannah, M., 2011. Using Monte-Carlo simulations and Bayesian networks to quantify and demonstrate the impact of fertiliser best management practices. Environ. Model. Softw. 26, 1079e1088. Partila, M., 2010. Separation Instruction Formation for Oily Waste Fractions after Tanker Accident. Master's thesis. Faculty of Technology, Lappeenranta

11

University of Technology. Pearce, D., 1998. Cost-benefit analysis and environmental policy. Oxf. Rev. Econ. Policy 14, 84e100. Psarros, G., Skjong, R., Vanem, E., 2011. Risk acceptance criterion for tanker oil spill risk reduction measures. Mar. Pollut. Bull. 62, 116e127. Raiffa, H., Schlaifer, R., 1961. Applied Statistical Decision Theory. Graduate School of Business Administration, Harvard University, Boston. Shahriari, M., Frost, A., 2008. Oil spill cleanup cost estimation e developing a mathematical model for marine environment. Process Saf. Environ. Prot. 86, 189e197. Soomere, T., Quak, E. (Eds.), 2013. Preventive Methods for Coastal Protection: Towards the Use of Ocean Dynamics for Pollution Control. Springer, Heidelberg. Soomere, T., Andrejev, O., Myrberg, K., Sokolov, A., 2011a. The use of Lagrangian trajectories for the identification of the environmentally safe fairways. Mar. Pollut. Bull. 62, 1410e1420. €e, B., 2011b. Modelling environmentally Soomere, T., Berezovski, M., Quak, E., Viikma friendly fairways using Lagrangian trajectories: a case study for the Gulf of Finland, the Baltic Sea. Ocean. Dyn. 61, 1669e1680. Statistics Finland, 2014. Annual Changes of Consumer Price Index. HM Treasury, 2003. The Green Book. Appraisal and Evaluation in Central Government. Uusitalo, L., 2007. Advantages and challenges of Bayesian networks in environmental modelling. Ecol. Model. 203, 312e318. Varis, O., Kuikka, S., 1999. Learning Bayesian decision analysis by doing: lessons from environmental and natural resources management. Ecol. Model. 119, 177e195. Ventikos, N.P., Sotiropoulos, F.S., 2014. Disutility analysis of oil spills: graphs and trends. Mar. Pollut. Bull. 81, 116e123. Yamada, Y., 2009. The cost of oil spills from tankers in relation to weight of spilled oil. Mar. Technol. Sname News 46, 219e228.


A Bayesian network for assessing the collision induced risk of an oil accident in the Gulf of Finland.

Toward integrative management advice of water quality, oil spills, and fishery in the Gulf of Finland: a Bayesian approach.

Gulf oil spill. Big test for bioremediation.

A Probabilistic Analysis of Muscle Force Uncertainty for Control.

A hierarchical Bayesian model averaging framework for groundwater prediction under uncertainty.

A full bayesian approach for boolean genetic network inference.

A mixture copula Bayesian network model for multimodal genomic data.

A Bayesian model for the analysis of transgenerational epigenetic variation.

NetGen: a novel network-based probabilistic generative model for gene set functional enrichment analysis.

Decision making under uncertainty: a quasimetric approach.

Prognostic factors for urachal cancer: a bayesian model-averaging approach.

A New Approach of Oil Spill Detection Using Time-Resolved LIF Combined with Parallel Factors Analysis for Laser Remote Sensing.

Correction: Order Under Uncertainty: Robust Differential Expression Analysis Using Probabilistic Models for Pseudotime Inference.

A unifying probabilistic Bayesian approach to derive electron density from MRI for radiation therapy treatment planning.

A probabilistic model for accidental cargo oil outflow from product tankers in a ship-ship collision.

A Hybrid Optimization Method for Solving Bayesian Inverse Problems under Uncertainty.

A hybrid dynamic Bayesian network approach for modelling temporal associations of gene expressions for hypertension diagnosis.

Network Modularity is essential for evolution of cooperation under uncertainty.

Meta-analysis for 2 x 2 tables: a Bayesian approach.

System analysis of waste oil management in Finland.

A bayesian approach to laboratory utilization management.

The Deepwater Horizon Oil Spill and the Mississippi Gulf Coast: Mental health in the context of a technological disaster.

Genealogical Working Distributions for Bayesian Model Testing with Phylogenetic Uncertainty.

Disease model: a simplified approach for analysis and management of human error: a quality improvement study.