Minimizing Risks from Spilled Oil to Ecosystem Services Using

Aug 29, 2011 - Environmental Science & Technology 2016 50 (24), 13195-13205 .... Integrated Environmental Assessment and Management 2015 11 (10.1002/i...
6 downloads 0 Views 4MB Size
POLICY ANALYSIS pubs.acs.org/est

Minimizing Risks from Spilled Oil to Ecosystem Services Using Influence Diagrams: The Deepwater Horizon Spill Response John F. Carriger* and Mace G. Barron National Health and Environmental Effects Research Laboratory, Gulf Ecology Division, U.S. Environmental Protection Agency, 1 Sabine Island Drive, Gulf Breeze, Florida 32561, United States

bS Supporting Information ABSTRACT: Decision science tools can be used in evaluating response options and making inferences on risks to ecosystem services (ES) from ecological disasters. Influence diagrams (IDs) are probabilistic networks that explicitly represent the decisions related to a problem and their influence on desired or undesired outcomes. To examine how IDs might be useful in probabilistic risk management for spill response efforts, an ID was constructed to display the potential interactions between exposure events and the trade-offs between costs and ES impacts from spilled oil and response decisions in the DWH spill event. Quantitative knowledge was not formally incorporated but an ID platform for doing this was examined. Probabilities were assigned for conditional relationships in the ID and scenarios examining the impact of different response actions on components of spilled oil were investigated in hypothetical scenarios. Given the structure of the ID, potential knowledge gaps included understanding of the movement of oil, the ecological risk of different spill-related stressors to key receptors (e.g., endangered species, fisheries), and the need for stakeholder valuation of the ES benefits that could be impacted by a spill. Framing the Deepwater Horizon problem domain in an ID conceptualized important variables and relationships that could be optimally accounted for in preparing and managing responses in future spills. These features of the developed IDs may assist in better investigating the uncertainty, costs, and the tradeoffs if large-scale, deep ocean spills were to occur again.

1. INTRODUCTION The Deepwater Horizon (DWH) oil rig exploded on April 20, 2010 initiating the discharge of over 775 million liters of oil into the Gulf of Mexico (GOM) over a time span of approximately three months. The economic impacts on GOM coastal communities were immense with reduced employment opportunities and income.1 International concern over contaminated seafood resulted in over 20 million hectares of fishing closures and losses in tourism and recreation-related revenues. The damages from the spill to the public economic welfare and ecological condition of GOM are still being assessed through the Natural Resource Damage Assessment process (15 CFR Part 990). Though not fully accounted, losses to GOM communities have been recently estimated to be in the tens of billions of dollars.2 Massive resources were mobilized to contain the spill and prevent damages from accruing. The incidents from May to July of 2010 highlighted the challenges existing government and industry operations had in responding to a deep ocean oil spill.3 A diversity of counter measures were used including mechanical recovery, in situ burning, and shoreline booming. For the first time in history, oil dispersants were used in deep sea environments and over 6977 kiloliters were applied, in total, to disperse deep ocean oil. Many of the response measures applied in the DWH spill were based on knowledge of surface spill events (e.g., vessel spills) rather than information on deep sea releases. This article not subject to U.S. Copyright. Published 2011 by the American Chemical Society

Guidelines such as the U.S. Coast Guard funded selection guide4 and oil spill response manual5 did not consider deep sea spilled oil. A lack of understanding of risk-based outcomes in a spill of this magnitude and complexity resulted in high uncertainty in the response effort’s options and efficacy. Due to the lack of knowledge about adverse effects in deep ocean applications of dispersants, the decision to apply dispersants near the well was controversial and debated by experts.6 Dispersant usage increased oil availability in one region (water column) while preventing its appearance in another (shoreline).7 A presidential commission judged dispersant application to have reduced overall risks, particularly to shoreline habitats.8 Existing U.S. policy tools for addressing deep ocean spilled oil had limited ability for probabilistic inference, for capturing expert knowledge on the uncertainties, and for assessing outcomes for ecological system services (ES) and values that are held by the populace. Development of models that can consider multiple decision options under complex ecological and spill scenarios might allow for better comparisons of potential risks and costs, and ultimately spill response decisions. Framing social, Received: March 29, 2011 Accepted: August 2, 2011 Revised: July 25, 2011 Published: August 29, 2011 7631

dx.doi.org/10.1021/es201037u | Environ. Sci. Technol. 2011, 45, 7631–7639

Environmental Science & Technology

POLICY ANALYSIS

Figure 1. Example influence diagram for deep ocean that examines the impact of response decisions on costs of the action and exposure variables, which in turn create ecological impacts.

economic, and risk-based concerns might alleviate public distress and focus limited funds on priority measurements for better decision-making in a crisis.9 Unique to the spill response was the creation of a scenarios team comprising experts from diverse fields to evaluate potential consequences of the spill for decisionmaking.10 The team took into consideration multiple scenarios and uncertainties and considered interventions to avert ecological, economic, social, and other crises. More scenario building exercises with experts are being conducted and their output can be informative for future decision support tools. To assist scenario-building and decision support efforts, we investigated the usage of influence diagrams (IDs) 11 to incorporate uncertainty in ES impacts from management decisions during a deep ocean oil spill. Influence diagrams are probabilistic networks that explicitly model the decisions related to a problem and their influence on important variables and outcomes of value. The construction of IDs allows stakeholders, experts, and decision-makers to consider the important variables influencing prospects and the interdependencies between decisions, random variables and objectives. In order to examine how such an approach can better address uncertainties in future deep ocean spills, a retrodictive ID was developed for the DWH spill response efforts. The ID demonstrates how policy-makers and implementers can consider the conditional relationships between exposure variables, ecological impacts, and ES objectives in a risk-based decision problem.

2. BACKGROUND ON INFLUENCE DIAGRAMS Influence diagrams are graphical representations of how decisions interact with processes and outcomes of value in a problem. Within an ID, there is a directed acyclic graph containing nodes (related to random variables, decisions, or valued outcomes) connected by arcs (arrows). The presence of an arrow from one node to another indicates a direct influence of the originating node (the parent) on the receiver (the child). In the case of chance nodes, an entering arrow indicates a probabilistic or deterministic relationship with the parent. In the case of decision nodes, an entering arrow indicates an informational or “no forgetting” relationship with the parent, which signifies that the previous decisions and informational chance nodes would be known at the time the current decision is made. In the case of value nodes, an entering arrow indicates an influence of the parent on a potential outcome of value. Within each chance node are discrete state categories pertaining to outcomes. States for

chance nodes are based on an exhaustive and mutually exclusive set of outcomes and can represent Boolean, integral, or ordered values.12 Decision node states represent the actions that can be taken by a decision-maker or set of decision-makers at a time period. In most IDs, continuous variables are discretized. Edge effects, or the influence of binning data on the propagation of uncertainty, should be considered during this procedure.13,14

3. QUALITATIVE AND QUANTITATIVE MODELING CONSIDERATIONS WITH INFLUENCE DIAGRAMS Often, input from stakeholders, decision-makers, policy-makers, analysts, and experts will formally or informally be used to develop decision models for ecological risk management. Perceptions between and within groups will vary as will knowledge of certain aspects of the decision problem. Cause-effect diagrams can be important tools to reduce complexity and formally understand the beliefs of all of the parties involved in an environmental management problem. Properly encoding mental representations of the causal interaction of decisions, situational variables, and attributes of importance can be a beneficial precursor to building models that make inferences on decision outcomes.15 For eliciting ID model structures, Arentze and colleagues15 test and demonstrate a protocol that might be useful for interactions with stakeholders. However, representing the temporal resolution of concern in a probability network can be difficult.16 Within the context of the spill, a time and spatial scale of interest necessary to model the impact to ES would be represented by the ID. Following preliminary conceptual diagrams created using the drivers-pressure-state-impact-response (DPSIR) framework, IDs were constructed in Netica17 for the deep ocean, offshore, and onshore environments. The goal was to combine these three figures into a single ID to make inferences on ES risks for all three environments; however, they are each described individually to increase clarity. Additional information and background on IDs is given in the Supporting Information. Figure 1 details the important relationships that might be better understood to assess the effectiveness of response actions in mitigating risks to deep ocean ES. Important variables influencing the delivery of ES were the states of oil in each region. The decision node is shown in light blue and represents a set of options available to managers in the deep ocean region. Numbers in the decision node are expected utilities based on probability weighted subjective scales for stakeholder satisfaction. Higher utility is better. The chance 7632

dx.doi.org/10.1021/es201037u |Environ. Sci. Technol. 2011, 45, 7631–7639

Environmental Science & Technology

POLICY ANALYSIS

Figure 2. Example influence diagram for offshore that examines the impact of response decisions on costs of the action and exposure variables, which in turn create ecological impacts.

nodes are colored green, gray, tan, yellow, and red. The yellow node identifies the variable representing ecological impacts. The red nodes are the objective nodes or what we wish to control for receiving more beneficial outcomes. The tan nodes are stressorrelated variables such as oil or dispersant concentrations. The green chance nodes pertain to external processes important for their impact on any of the above variables. The gray nodes are nodes for which we have input evidence to see how probabilities might propagate throughout the diagram. Numbers next to the state of each chance node are probabilities (%) of that state occurring given the evidence entered into the network. The utility nodes are in pink and contain the subjective value scale on the costs of response actions and the impacts on ES. The output of the utility node is displayed in the numbers for the decision node. Hypothetical probabilities were assigned for conditional relationships in Figure 1 and scenarios examining the impact of different response actions on components of spilled oil were investigated. As can be seen in Figure 1, the dispersants, plume oil, and oil-dispersant mixtures could have some ecological impact in the deep ocean, jointly or in isolation. This would be reflected in a conditional probability table that underlies the ecological impacts node, where each state for each parent exposure compound would have an associated probability for each state of the ecological impacts node. For some relationships, the probabilities are more certain than others. For example, deep ocean ecological impacts are assumed to have explicit impacts on deep ocean ES. For others, they are less certain. The amount of deep ocean oil in sediment contains a flat distribution indicating high uncertainty (each state is equally likely due to lack of knowledge) about the amount of oil that might accumulate given a well that is partially on and possible decision responses. The blue node (“Deep ocean response”) is a decision node and has expected utilities for each potential action given the potential impacts on cost or ES. Deep dispersant application is the only response option considered in the ID, other than the no action alternative. The Well release node contains three states: full,

partial, and off to indicate the intensity of crude oil loadings to the deep ocean domain. Thus, deep ocean plume oil would be influenced by the amount of release from the well along with the deep ocean response. In Figure 1, the well release and response actions potentially influence the levels of different physical states of oil in the deep ocean environment. Findings are entered to indicate the well release is partially on and the conditional probabilities of the oil and dispersant variables are then estimated by the ID. The utility, or value that we are uncertain about receiving, is derived from both the actual cost of the response and the impacts to ecological services. This model recommends “No action” as the best decision in all cases when only deep ocean impacts are included in the ID. Ecological impacts from oil are likely to occur to benthic and pelagic species and deep sea benthic communities such as methane vent communities and corals. The ecological production functions impacted could affect the existence, scientific, and fisheries value of the deep ocean environment. Some of the deep ocean oil (plume oil, dispersed oil) reaches the offshore pelagic regions to be assessed as floating oil (surface slick) or oil in the water column. These variables are taken from Figure 1 and used as root nodes (or nodes that are not children of any other nodes) in the offshore problem domain (Figure 2). These deep ocean related variables are colored bright blue in Figure 2. Probabilities of offshore oil states are partially dependent on the deep ocean oil states so robust fate and transport data would be required to populate the ID in Figure 2. Sea state is assumed to be known at the time offshore decisions are made as indicated by an arc connecting the two variables. In the offshore region, the exposure considerations now include mousse and tarballs as important weathering states of oil. If any of these oil states were insignificant in their ability to cause offshore ecological impacts, they could be removed to make the model more tractable and parsimonious. Alternatively, they could be joined into one variable for simplification. The available offshore responses also differ from the deep ocean with burning and mechanical 7633

dx.doi.org/10.1021/es201037u |Environ. Sci. Technol. 2011, 45, 7631–7639

Environmental Science & Technology

POLICY ANALYSIS

Figure 3. Example influence diagram for onshore that examines the impact of response decisions on costs of the action and exposure variables, which in turn create ecological impacts.

recovery being options for removal, in addition to dispersant application. From the expected utility values indicated in the decision node, the conditions in this ID recommend mechanical recovery as the preferred option, consistent with current spill response guidance. In the offshore environment, the impacts on ES such as fisheries production should be explicitly considered in the ID. In Figure 3, oil components move toward the onshore region from the offshore region and response actions become focused on preventing shoreline, habitat, and wildlife oiling. Trajectory of the spill is viewed as an important situational variable influencing the importance that should be placed on the decisions. The trajectory is known at the time of the decision so an informational arc is placed between the onshore decision variables and the trajectory chance node. Spill responses available for the onshore region are booming and berming, or water diversion, with limited ability for mechanical recovery or in situ burning because of the advanced weathering state of the oil. Dispersant applications in near shore regions were not approved during the DWH spill. Thus, the onshore spill responses in the ID are focused on preventing the oil from reaching the shore and not removing it or enhancing biodegradation. As discussed in the next section, the actions in the decision nodes are simplified and policies with onshore dispersants might be considered in practice and the recommendations from the model compared with alternate policies. As was done in Figures 1 and 2, probabilities for Figure 3 were assigned for the relationships to represent our prior uncertainty about the problem. For example, offshore dispersed oil has a prior marginal probability distribution that indicates that high concentrations of dispersed oil offshore are more likely than moderate or low concentrations. The probability of different states of onshore dispersed oil are influenced by this and conditional probabilities would be filled out for the probability of whether the onshore dispersed oil will be low, moderate or high depending on the offshore dispersed oil, the trajectory, and the response decision taken. The ecological impacts are extended to tidal, estuarine, and coastal communities including seagrass habitats and oyster reefs. The services onshore are diverse and varied and include a high amount of beneficiaries such as the tourism industry, recreational users, and commercial fisheries. The ID recommended booming and berming as being the

optimal solution in comparison with water diversion or no action taken. 3.1. Qualitative Considerations. Influence diagrams or Bayesian decision networks can decompose complex problems into underlying causal mechanisms for easier quantitative assessments of submodels than holistic empirical or subjective judgments would allow. In Figures 1 and 2, the spill responses in the offshore and deep ocean regions are expected to only directly cause ecological impacts through the application of dispersants. For the onshore region in Figure 3, the response of water diversion is expected to have a potential adverse impact on structure of onshore communities. The movement of dispersed oil is expected to impact offshore and onshore resources depending on the prior decisions. Thus, offshore impacts would only be relevantly affected by a deep ocean response that changes the amount of deep ocean plume oil or dispersed oil reaching shallow water zones. This level of specification is apparent in Figure 2. Additional considerations such as information available at the time of decisions (e.g., sea state may be known prior to an offshore response) became more apparent when the actual ID was constructed. Of the over 750 M l oil spilled during the DWH event, response actions resulted in estimated removals of 17% mechanically recovered, 5% burned, 3% skimmed, and 16% chemically dispersed.18 From the representation in the ID, knowledge of this and alternate scenarios would be required to fully parametrize the model and reduce knowledge gaps. Thus, the transport, weathering, and risks of oil throughout the system as well as the influence of decisions on each of those processes is required for the ID in this paper to be operational in future response tasks. Each node contains several states. Following the Robertson and colleagues5 relative ranking system for habitat, values of low, moderate, and high were assigned to the potential ecological impacts. For other variables, categorical states of low, moderate, or high were also input for demonstration purposes. Interpretation problems could exist for categorical states such as questions on what constitutes high, medium, or low. This could result in what Keeney19 considers a lack of understandability of what the measure for the variable signifies. In the current ID, continuous variables with natural scales could have discrete intervals for values over an exhaustive set of outcomes. This would increase understandability of what each state signifies but a continuous 7634

dx.doi.org/10.1021/es201037u |Environ. Sci. Technol. 2011, 45, 7631–7639

Environmental Science & Technology

POLICY ANALYSIS

Figure 4. Influence diagram for the Deepwater Horizon oil spill decision scenario indicating the expected utility of onshore responses when well release is fully on, and dispersants are applied in the deep ocean and offshore regions.

distribution without discrete states would convey even more information on the variable’s uncertainty. Influence diagrams can be set up with continuous distributions in other software tools with some missing functionality such as the ability to perform diagnostic inquiries. The decision nodes contained individual response options. In actuality, a blend of policy options must be assessed. For example, a deep ocean response would consist of varying degrees of dispersant application and an offshore response would consider the combined effects of mechanical recovery and surface burning options. This can be difficult to represent in an ID. Kotta and colleagues20 consider how this might be done in a Bayesian network by assessing blended scenarios for future climate change and eutrophication predictions to a bay in the Baltic Sea. In a

qualitative reconstruction of the Exxon Valdez spill response, Harrald and colleagues21 break the individual response decisions into separate nodes to allow greater consideration of information needs and impacts on decisions. Parameterizing the latter could be difficult as most ID software requires a sequence between decision nodes and the additional node states will introduce greater complexity.12 For demonstration purposes, we separated the response types into discrete states or categories to emphasize their importance in mitigating exposure events for each zone. The structure of IDs allows decomposition into submodels that would enhance communication, data gathering, and logistical considerations that could impact decisions. The qualitative relationships in Figures 1 3 were also modeled as a full ID with an example scenario shown in Figure 4. The model was 7635

dx.doi.org/10.1021/es201037u |Environ. Sci. Technol. 2011, 45, 7631–7639

Environmental Science & Technology

Figure 5. Impacts to ES and costs anticipated from evaluating decisions and evidence of states of nature for two different influence diagrams. See text for more explanation.

developed to illustrate how the understanding of important causal relationships between response treatments and ES outcomes by domain experts might be represented. 3.2. Using Influence Diagrams for Response Strategies. The DWH response ID is shown in its complete form in Figure 4 where the three models in Figures 1-3 were combined and a scenario with full release of oil from the well is displayed. The decisions and random variables for the three regions are modeled in a sequence where an onshore decision will be made while knowing what response decisions already occurred in deep ocean and offshore regions. Complexities from changing decisions over time can also be modeled in an ID through sequentially linking variables at different time periods. This is exhibited by the connections between the decisions in different regions (e.g., offshore decisions are made prior to onshore decisions) and the same can be done for future considerations of options and impacts in the deep ocean (after onshore responses are applied) by adding another time slice to the ID. For example, cleanup activities for residual oil on beaches could be considered. A February of 2011 report estimated that residual oil was most likely to cause adverse effects on scavenging shorebirds and nesting sea turtles and that cleanup activities are more likely to cause greater damages than residual oil on several beaches.22 Decision trees can be helpful analytic structures for scenarios with a sequence of decisions and events.23 Time spans need to be considered for the model, particularly for the weathering of oil, the time to effects, cost implications and time between responses.

POLICY ANALYSIS

Often, time lags are implicit in probabilistic models as a lag would always occur between a cause and an effect. Outcomes for different temporal scales can be explicitly represented in the model through a parent node specifying durations (e.g., one month, one year) in the states.12 For the onshore response, a policy maker might receive information that the trajectory of the spill is known and there is a moderate sea state offshore (Figure 4). Deep dispersant application and offshore surface dispersant application are selected as the major policies for the deep ocean and offshore regions. When placed in the ID, the onshore recommended policy is booming/berming when the trajectory is nearshore or no action when it is offshore. Recommendations for other responses can be evaluated contingent on prior decisions and variables that are known to the decision-maker at the time of the decision. Additional scenarios are investigated with the complete model in the Supporting Information. Figure 5 illustrates the potential impacts to ES under two different scenarios. The first is a lower impact scenario where the deep ocean well is fully releasing oil, deep dispersion occurs, there are no offshore fisheries closures, sea state is at a calm level, mechanical recovery is implemented offshore, the trajectory is offshore, and booming/berming are implemented onshore. Contrasted with this is a higher impact scenario in which no action is taken in the deep ocean, there are high offshore fisheries closures, sea state is rough, no action is applied offshore, trajectory is nearshore, and water diversion is used onshore (Figure 5). From the lower to the higher impact scenario, the greater probabilities of higher ES risks can be witnessed on all of the variables. However, lower impacts on costs are more likely to occur for the deep ocean and offshore responses for the higher impact scenario. This lower impact on costs was not enough to make the higher impact scenario a preferred one when the overall utilities were compared between each network. Diagnostic inference can also be performed where desired (or undesired) conditions for nodes related to objectives (i.e., ecological systemrelated variables) might be specified and upstream impacts on the likely states of causal nodes (i.e., oil concentrations or field variables) are examined. 3.2. Some Quantitative Aspects. The quantitative aspects of IDs are useful for eliciting and comparing beliefs about uncertainties between stakeholder groups and the data that might be gathered to reduce that uncertainty. Influence diagrams can be used to represent how factors that cause risks can impact ecological variables and ES of concern to stakeholders, the possible strength of those relationships, and what uncertainties might exist. As with Bayesian networks, the construction process for an ID allows analysts and decision-makers to focus on the aspects of the problem important to stakeholders and focus on how they will be quantified. The ID allows multivariate relationships and inputs to a system, such as decisions, to be evaluated based on probabilistic measures of uncertainty. Sometimes, soft scoring methods are utilized for complex problems and probabilistic risk analysis is focused on lower level risks that are easier to empirically assess and have less drastic outcomes, a phenomenon known as the risk paradox.24 Probabilistic reasoning forces analysts to consider the magnitude and likelihood of adverse outcomes better than less formal methods and have been successfully applied to complex, mission critical risk management problems such as oil exploration and risks from nuclear power.24 Data needs, such as better understanding of the transport of oil from deep ocean to offshore to onshore environments, the 7636

dx.doi.org/10.1021/es201037u |Environ. Sci. Technol. 2011, 45, 7631–7639

Environmental Science & Technology adverse effects of different spill-related stressors to important receptors (e.g., protected species, fisheries), and the need for stakeholder valuation of the benefits that could be impacted by a spill, can be explicitly recognized by stakeholders and experts alike. Sensitivity analysis can be used to evaluate which inputs cause the greatest changes to expected utilities and decision recommendations in IDs.25 Value of information to decisions would be a novel way of considering sensitivity analysis in IDs.26 This might be addressed by asking whether a decision-maker would consider a different decision with or without certain or uncertain information that might be available. Thus, how decisions might change depending on data like ecological impacts, oil states, and field variables could be used to assess the sensitivity of this data to responses.27 Sensitivity analysis in the current ID might also be used to evaluate which oil components have a high degree of impact on ES predictions.

4. ECOSYSTEM SERVICE IMPACTS AND COSTS OF REMEDIATION AS TRADE-OFFS In order to quantify the importance of ES with stakeholders, a few potential approaches may be implemented. The first is a costbenefit analysis that relies on market-and nonmarket based estimates of ES benefits. This tool makes each ES commensurate on a dollar scale. However, problems with monetary valuation for all objectives, including nonuse ones, might create poor and wrong assumptions. The second is a quasi-economic approach that relies on eco-dollars or relative importance rankings against a service with a reference dollar value.28 The services are then scaled into a valuation index for a policy problem. This agglomeration requires direct input from stakeholders and would be helpful for aggregating the weight that stakeholders place on different benefits from ES. Considerations such as the type and range of ES that could be gained or lost in an oil spill should be incorporated into such an approach. The third consists of nonmonetary valuation techniques for quantifying ES which can be applied using theories from multiattribute utility theory (MAUT).29 Hajkowicz30 developed a MAUT index of ES that relies on natural units, not just dollars. Preparing such an index can assist in weighing trade-offs between ES with constructed measures for aesthetics, natural measures such as number of recreational visitors, and binomial outcomes such as presence or absence of an endangered species population. Robust techniques exist for constructing individual and MAUT functions for various forms of ES. The benefits of MAUT applications include the efficaciousness in which properly constructed measures for ES fundamental objectives can be transferred to commensurate value or utility scales, the strong theoretical basis for considering constructed preferences, and the ability to effectively capture trade-offs and other salient aspects of a decision problem.19 The disadvantages are the stringent assumptions that must be met when quantifying and combining values in an additive fashion,31 the difficulty stakeholders could have in accurately eliciting individual utility functions,32 and the analytical nature of the valuation process.33 The U.S. Forest Service’s CRAFTipedia site (http://craft. forestthreats.org/CRAFTiPedia/index.php/Main_Page) describes multiple methods and considerations for exploring and eliciting utility functions for IDs.

POLICY ANALYSIS

5. SUGGESTIONS FOR MAKING AN OPERATIONAL MODEL The ID format gives stakeholders and decision-makers the opportunity to decompose the problem into a set of clear variables of importance to better understand their relationships and the necessary data to evaluate the impacts of responses on ES. Reducing the uncertainty among these relationships should be done through targeted data collection that closes gaps and modeling that tests understanding. In addition, model components should be evaluated with data. Data needs include the fate and transport of oil from deep ocean to offshore to onshore environments, the ecological risk to important receptors (e.g., protected species, fisheries), and stakeholder opinion or weighting of the ES that could be impacted by a spill. Dialogue with stakeholders is necessary to effectively elicit opinions in a constructive manner and evaluate and communicate efforts for recovery. Data have been collected extensively since the spill occurred and are mostly publically available (http://www.restorethegulf.gov/news/maps-data). Stakeholder opinion would be necessary for building better value models while expert opinion would be needed for assessing uncertainties when direct measurements are infeasible. For Bayesian networks, protocols for data gathering for ecological risk assessment have been implemented by Pollino and colleagues,34 for complementing research and enhancing communication on landholder decision impacts by Ticehurst and colleagues,35 for water resource planning with stakeholders by Bromley and colleagues,36 and for exploring the beliefs of property managers on land use changes by Bacon and colleagues.37 As Gregory and colleagues38 describe, protocols have been developed for eliciting expert opinion that minimize biases (e.g., overconfidence, availability, representativeness) and may be more appropriate to implement than additional data gathering activities under certain scenarios. For additional information, Hora39 describes protocols for individual experts and Clemen and Winkler40 describes methods for aggregating probability distributions from multiple experts. The ID provides a framework for developing and implementing expert opinion elicitations and/or gathering additional data. Bayesian decision science approaches have been utilized in oil spill contingency planning and other areas.21,41,42 6. CONCLUSIONS Documents from the U.S. Coast Guard4 describe a series of responses to contain, remove, and inhibit oil from reaching sensitive resources in the vicinity of spilled oil. Conceptual guidance currently allows responders to follow a pathway to a best set of options and make rule-based decisions. The background paper that describes the reasoning behind these procedures and how the types of shoreline features might be weighed in management options is the “least regret” paper.5 Influence diagrams could assist in preparing decision support models for response emergencies prior to the spill or explicitly considering the values of resources at stake and risks from the spill. This article discusses a modeling framework for considering impacts of stressors from decisions and spilled oil on ecological variables as well as ecological services. The framework graphically represents the conditional influences among variables important for assessing ecological risks and trade-offs from the Deepwater Horizon response. Functional representations of management interventions on key exposure-related variables and risks to 7637

dx.doi.org/10.1021/es201037u |Environ. Sci. Technol. 2011, 45, 7631–7639

Environmental Science & Technology ecological features of concern can facilitate enhanced communication and rigorous accounting in spilled oil responses. Preventing events that lead to an oil spill, both massive and small, should be a future priority. However, understanding the value of ecosystem resources and the chain of events that could occur in future disasters would benefit managers and policymakers protecting the GOM and other complex ecological systems. Framing the DWH problem domain in an ID provided a retrodictive model of the trade-offs faced in the spill event. Future deep ocean drilling scenarios should consider the lessons from the DWH to better understand the fate, transport, and threats to ecologically diverse environments. The flexibility of the ID model environment allowed a variety of potential outcomes to be considered in the model. The IDs can conceptualize important variables and relationships that could be optimally accounted for in preparing and managing responses in future spills. Reducing uncertainties on costly prospects is targeted in any risk management endeavor and an ID framework might assist future regulators and decision makers in responding to oil spills and better understanding the potential outcomes. These features of the developed IDs will assist in better investigating the uncertainty in deep ocean spills, the costs from losing ES, and the necessary trade-offs for minimizing these losses if future large scale spills were to occur again. Constructing IDs to display how risks are assessed and how policy options are weighed could potentially improve the communication process between regulators and stakeholders and could make the plausibility of management policies more understandable to all concerned parties.

’ ASSOCIATED CONTENT

bS

Supporting Information. Additional examples of oil spill scenarios using the integrated deep ocean, offshore, and onshore influence diagram shown in this paper. This material is available free of charge via the Internet at http://pubs.acs.org.

’ AUTHOR INFORMATION Corresponding Author

*Phone: 850-934-9226; e-mail: [email protected].

’ ACKNOWLEDGMENT We thank Eric D. Johnson and Susan H. Yee for helping with initial conceptual frameworks and DPSIR diagrams that assisted in establishing cause-effect linkages for the ID. This research was supported by the U.S. EPA and while reviewed according to EPA guidelines, it does not necessarily reflect EPA policy. Mention of trade names or commercial products does not constitute endorsement by the US EPA. This is contribution No. 1420 from the U.S. EPA Gulf Ecology Division, Gulf Breeze, FL. ’ REFERENCES (1) Levy, J.; Gopalakrishnan, C. Promoting ecological sustainability and community resilience in the US Gulf Coast after the 2010 Deep ocean Horizon Oil Spill. J. Nat. Resour. Policy Res. 2010, 2 (3), 297–315. (2) National Commission. Deep Water: The Gulf Oil Disaster and the Future of Offshore Drilling, Report to the President; National Commission on the BP Deepwater Horizon Oil Spill and Offshore Drilling, 2011; www.oilspillcommission.gov.

POLICY ANALYSIS

(3) Hagerty, C. L.; Ramseur, J. L. Deepwater Horizon Oil Spill: Highlighted Actions and Issues; Congressional Research Service, 2011; www.crs.gov. (4) Sea Inc. Selection Guide for Oil Spill Applied Technologies: Volume 1- Decision-Making; Scientific and Environmental Associates, Incorporated: Cape Charles, VA, 2003. (5) Robertson, S.; Steen, A.; Pavia, D.; Walker, A. Marine oil spill response options: the manual. In International Oil Spill Conference Proceedings; American Petroleum Institute Publication No. 4651: Washington, DC, 1997. (6) Schmidt, C. W. Between the devil and the deep blue sea: dispersants in the Gulf of Mexico. Environ. Health Perspect. 2010, 118 (8), A338–A344. (7) National Research Council. Oil Spill Dispersants: Efficacy and Effects, Committee on Understanding Oil Spill Dispersants Efficacy and Effects; Ocean Studies Board; Division on Earth and Life Studies; The National Academies Press: Washington, DC, 2005. (8) Cho, A. Government chided for poor planning and communication. Science 2010, 330, 302–303. (9) Bohnenblust, H.; Slovic, P. Integrating technical analysis and public values in risk-based decision making. Reliab. Eng. Syst. Saf. 1998, 59 (1), 151–159. (10) Machlis, G. E.; McNutt, M. K. Scenario building for the Deepwater Horizon oil spill. Science 2010, 329, 1018–1019. (11) Shachter, R. D. Evaluating influence diagrams. Oper. Res. 1986, 34 (6), 871–882. (12) Korb, K.; Nicholson, A. Bayesian Artificial Intelligence, 2nd, ed.; CRC Press: Boca Raton, FL, 2011. (13) Uusitalo, L. Advantages and challenges of Bayesian networks in environmental modelling. Ecol. Model 2007, 203 (3 4), 312–318. (14) Kuhnert, P.; Hayes, K. How Believable Is Your BBN?, 18th World IMACS/MODSIM Congress, Cairns, Australia, 2009. (15) Arentze, T. A.; Dellaert, B. G. C.; Timmermans, H. J. P. Modeling and measuring individuals’ mental representations of complex spatio-temporal decision problems. Environ. Behav. 2008, 40 (6), 843–869. (16) Barton, D. N.; Saloranta, T.; Moe, S. J.; Eggestad, H. O.; Kuikka, S. 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. 2008, 66, 91–104. (17) Norsys. Netica 4.16. Norsys Software Corp.: Vancouver, BC, 2010. (18) FISC. Oil budget Calculator, Deepwater Horizon, technical documentation; Federal Interagency Solutions Group (FISC), 2010. (19) Keeney, R. L. Value-Focused Thinking: A Path to Creative Decisionmaking; Harvard University Press: Cambridge, MA, 1992. (20) Kotta, J.; Aps, R.; Orav-Kotta., H. Bayesian inference for predicting ecological water quality under different climate change scenarios. WIT Trans. Ecol. Environ. 2009, 127, 173–184. (21) Harrald, J.; Marcus, H.; Wallace, W. The Exxon Valdez: An assessment of crisis prevention and management systems. Interfaces 1990, 20 (5), 14–30. (22) OSAT-2. Summary Report for Fate and Effects of Remnant Oil in the Beach Environment, Operational Science Advisory Team (OSAT-2); Gulf Coast Incident Management Team, 2011. (23) von Winterfeldt, D.; Fasolo, B. Structuring decision problems: a case study and reflections for practitioners. Eur. J. Oper. Res. 2009, 199 (3), 857–866. (24) Hubbard, D. W. The Failure of Risk Management; John Wiley & Sons, Inc.: Hoboken, NJ, 2009. (25) Bielza, C.; Gomez, M.; Shenoy, P. P. Modeling challenges with influence diagrams: constructing probability and utility models. Decis. Support Syst. 2010, 49, 354–364. (26) Shachter, R. D. An ordered examination of influence diagrams. Networks 1990, 20, 535–563. (27) Shachter, R. D.; Kenley, R. Gaussian influence diagrams. Manage. Sci. 1989, 35 (5), 527–550. 7638

dx.doi.org/10.1021/es201037u |Environ. Sci. Technol. 2011, 45, 7631–7639

Environmental Science & Technology

POLICY ANALYSIS

(28) Jordan, S. J.; Hayes, S. E.; Yoskowitz, D.; Smith, L. M.; Summers, J. K.; Russell, M.; Benson, W. H. Accounting for natural resources and environmental sustainability: linking ecosystem services to human well-being. Environ. Sci. Technol. 2010, 44 (5), 1530–1536. (29) Keeney, R.; Raiffa, H. Decision Making with Multiple Objectives: Preferences and Value Tradeoffs; John Wiley and Sons: New York, 1976. (30) Hajkowicz, S. Multi-attributed environmental index construction. Ecol. Econ. 2006, 57 (1), 122–139. (31) Hajkowicz, S.; Young, M.; Wheeler, S.; MacDonald, D. H.; Young, D. Supporting Decisions: Understanding Natural Resource Management Assessment Techniques; CSIRO Land and Water Resources and Development Corporation: Adelaide, SA, Australia, 2000. (32) McDaniels, T.; Gregory, R.; Fields, D. Democratizing risk management: successful public involvement in local water management decisions. Risk Anal. 1999, 19 (3), 497–510. (33) Gregory, R.; Slovic, P. A constructive approach to environmental valuation. Ecol. Econ. 1997, 21 (3), 175–181. (34) Pollino, C. A.; Woodberry, O.; Nicholson, A.; Korb, K.; Hart, B. T. Parameterisation and evaluation of a Bayesian network for use in an ecological risk assessment. Environ. Modell. Software 2007, 22 (8), 1140–1152. (35) Ticehurst, J. L.; Curtis, A.; Merritt, W. S. Using Bayesian networks to complement conventional analyses to explore landholder management of native vegetation. Environ. Modell. Software 2011, 26 (1), 52–65. (36) Bromley, J.; Jackson, N. A.; Clymer, O. J.; Giacomello, A. M.; Jensen, F. V. The use of HuginÒ to develop Bayesian networks as an aid to integrated water resource planning. Environ. Modell. Software 2005, 20 (2), 231–242. (37) Bacon, P. J.; Cain, J. D.; Howard, D. C. Belief network models of land manager decisions and land use change. J. Environ. Manage. 2002, 65 (1), 1–23. (38) Gregory, R.; Failing, L.; Ohlson, D.; McDaniels, T. L. Some pitfalls of an overemphasis on science in environmental risk management decisions. J. Risk. Res. 2006, 9 (7), 717–735. (39) Hora, S. C. Eliciting probabilities from experts. In Advances in Decision Analysis: From Foundations to Applications; Edwards, W., Miles Jr., R. F., von Winterfeldt, D., Eds.; Cambridge University Press: Cambridge, UK, 2007. (40) Clemen, R. T.; Winkler, R. L. Aggregating probability distributions. In Advances in Decision Analysis: From Foundations to Applications; Edwards, W., Miles Jr., R. F., von Winterfeldt, D., Eds.; Cambridge University Press: Cambridge, UK, 2007. (41) Aps, R.; Fetissov, M.; Herk€ul, K.; Kotta, J.; Leiger, R.; Mander, € Suursaar, U. € Bayesian inference for predicting potential oil spill U.; related ecological risk. WIT Trans. Built Environ. 2009, 108, 149–159. (42) Klemola, E.; Kuronen, J.; Kalli, J.; Arola, T.; Hanninen, M.; Lehikoinen, A.; Kuikka, S.; Kujala, P.; Tapaninen, U. A cross-disciplinary approach to minimising the risks of maritime transport in the Gulf of Finland. World Rev. Intermodal Transp. Res. 2009, 2 (4), 343–363.

7639

dx.doi.org/10.1021/es201037u |Environ. Sci. Technol. 2011, 45, 7631–7639