The role of scientific studies in building consensus in environmental decision making: a coral reef example

Abstract

We present a new approach for characterizing the potential of scientific studies to reduce conflict among stakeholders in an analytic-deliberative environmental decision-making process. The approach computes a normalized metric, the Expected Consensus Index of New Research (ECINR), for identifying where additional scientific research will best support improved decisions and resolve possible conflicts over preferred management actions. The ECINR reflects the expected change in agreement among parties over preferred management actions with the implementation and consideration of new scientific studies. We demonstrate the ECINR method based on a preliminary application to coral reef protection and restoration in the Guánica Bay Watershed, Puerto Rico, focusing on assessing and managing anthropogenic stressors, including sedimentation and pollution from land-based sources such as sewage, agriculture, and development. Structured elicitations of values and beliefs conducted at a coral reef decision support workshop held at La Parguera, Puerto Rico, are used to develop information for illustrating the methodology. The ECINR analysis was focused on a final study group of seven stakeholders, consisting of resource managers and scientists, who were not in agreement on the efficacy and respective benefits of reducing loadings from three sources: sewage, agriculture, and development. The scenario assumed that loadings would be reduced incrementally from each source through a series of management steps, which would be ranked in order of maximizing anticipated benefits. An examination of whether beliefs exhibited greater confidence and coherence between stakeholders when informed by plausible study results followed. The results suggest that new scientific research would be generally likely to bring people who initially disagreed to agree. Seventy-five percent of the hypothetical research results were projected to result in more agreement among the stakeholders. However, there can be situations where prior beliefs may be too different from the study results to shift perspectives enough to result in more agreement. Furthermore, in a few cases, hypothetical research results were projected to lead to more conflict among stakeholders. Priority research, according to the seven stakeholders, would be to quantify loadings from agriculture and sewage, and not loadings from development, since it is predicted to make little difference in the outcome. Assuming the stakeholders are conflict-averse, they would likely opt for research on sewage loadings as the highest priority. Though preliminary, these results suggest that ECINR can provide useful insights into the social implications of a research program.

Appendices

Appendix 1: Blank elicitation form

1.1 Glossary

Drivers :

Socioeconomic sectors that drive human activities (Waste disposal, agriculture, construction, fisheries, tourism)

Ecosystem Services :

The products of ecological functions or processes that directly or indirectly contribute to human well-being (clean air and water, food and fiber, erosion and flood control, habitat and biodiversity, climate stability, and esthetic enjoyment)

Hydroseeding :

A planting process which utilizes a slurry of seed and mulch, which is transported in a tank, either truck- or trailer-mounted and sprayed over prepared ground in a uniform layer

Impacts :

Effects of environmental degradation on ecosystem functioning, affecting the quality and value of ecosystem services

Management and policy options :

A number of alternatives that are under the control of and from which one or a combination of several of them (to be implemented as a strategy) can be chosen

Pathogen :

Microorganisms (e.g., bacteria, viruses, or parasites) that can cause disease in humans, animals, and plants

Pressures :

Human activities that stress the environment (discharge, boating activities, climate change, land use/land cover change, coastal erosion)

Riparian :

Of or relating to or located on the banks of a river or stream

States :

Reflect condition of the natural and living phenomena (such as air, water and soil parameters and growth, survival, and reproductive parameters)

Strength or magnitude of the relationship (between variables) :

The degree to which one variable is associated with or can cause a change in a second variable (i.e., between decisions and outcomes)

Toxics :

Poisonous chemicals

Uncertainty :

Inability to predict outcomes due to random variability (for example, streamflow is sometimes high and sometimes low) or incomplete scientific knowledge regarding causal relationships (for example, how does a given concentration of sediments in the harbor affect coral reef growth rates)

1.2 References

SPARROW: http://water.usgs.gov/nawqa/sparrow/

Questions in the face-to-face elicitation

1. How would you rate the following outcomes in relation to one another? (e.g., a 1 for tourism and a 2 for fish indicates that fish health is twice as important as tourism health).

• tourism -

• fish -

• coral -

2. What percentages of the total loadings (nutrient and sediment) to the Guánica inland water system comes from development, agriculture, and sewage, respectively? (percentages should sum to 100 %)

• development -

• agriculture -

• sewage -

3. How sure are you that the lagoon will work (i.e., be effective in reducing loadings that enter the Bay)?

• I am _% sure that the lagoon will work

4. What are the probabilities that the following sets of environmental stressors would produce: (a) good/bad coral reef health; and (b) good/bad fisheries health, respectively? (percentages should sum to100 %)

• stressors for coral reef health:

• water quality (WQ),

• ocean warming/acidification (OW)

• marine protection areas (MPA)

• Stressors for fisheries health:

• coral reef health (CR)

• ocean warming/acidification (OW)

• marine protection areas (MPA)

Example 1: If water quality is considered to be most responsible, followed by ocean acidification/warming, and then marine protection areas (considered useless in this example), and no synergism is assumed, the following probabilities could apply:

• 25 % WQ/OW/MPA

• 20 % WQ/MPA

• 25 % WQ/OW

• 5 % MPA/OW

• 20 % WQ

• 0 % MPA

• 5 % OW

Example 2: If water quality combined with ocean warming/acidification and MPAs is thought to be the most important set of stressors contributing to coral health, followed by water quality and ocean warming/acidification, and then followed by water quality and MPAs, and assuming synergism among the various factors, the following probabilities could apply:

• 50 % WQ/OW/MPA

• 30 % WQ/OW

• 10 % WQ/MPA

• 4 % MPA/OW

• 5 % WQ

• 1 % MPA

• 2 % OW

a. Probabilities that these sets of stressors lead to good/bad coral reef health:

• % that it’s all 3 (WQ/OW/MPA) -

• % that it’s these 2 (WQ/MPA) -

• % that it’s these 2 (WQ/OW) -

• % that it’s these 2 (MPA/OW) -

• % that it’s only 1 factor (WQ) -

• % that it’s only 1 factor (MPA) -

• % that it’s only 1 factor (OW) -

b. Probabilities that these sets of stressors lead to good/bad fisheries health:

• % that it’s all 3 (CR/OW/MPA) -

• % that it’s these 2 (CR/MPA) -

• % that it’s these 2 (CR/OW) -

• % that it’s these 2 (MPA/OW) -

• % that it’s only 1 factor (CR) -

• % that it’s only 1 factor (MPA) -

• % that it’s only 1 factor (OW) -

Appendix 2: Explanation of BBN

Our BBN was designed to represent the current situation of coral reefs stressors and management in the Guánica Bay Watershed, Puerto Rico from the viewpoint of stakeholders. Based on elicitations and discussions at the workshop in Puerto Rico, we developed a model in Netica (Norsys 2010) based on the DPSIR/DL framework (Rehr et al. 2012) that summarizes the essential components involved in coral reefs management. We included management options, environmental processes, and ecosystem services outcomes. Each node in Fig. 2 represents a particular variable that is part of coral reefs management. Each arrow in Fig. 2 represents a causative link between two nodes. In this explanation of the model, we use the BBN and inputs for Participant A.

At the lower right of the diagram is the endpoint of the BBN: Benefits. Benefits is continuous variable distributed over ten intervals, and a function of Tourism, Fisheries, Coral Reef Health and Coral Eco Services (ecosystem services), the four inputs or outcomes of interest to stakeholders that influence the level of benefits. Generally, the greater the inputs, the greater the resulting benefits. However, the total benefits are weighted according to the elicited values stakeholders place on each outcome in relation to each other. For example, Participant A believes that coral health is twice as important as tourism and fisheries and has the following equation for benefits (the particular values applied to the weightings are set to correspond to values used throughout the model and will be discussed later in this document):

$$\, Benefits = 150\,*\,Tourism + 150\,*\,Comm.\,\& \,\text{Re} c\,Fishing + 300\,*\,Coral\,\text{Re} ef\,Health\,*\,CoralEco\,Services$$

The following is the data table in Netica for Benefits for Participant A:

The first input into Benefits, Tourism, is a discrete variable and can be either low or high. Lagoon WQ, Comm. & Rec Fishing, and Coral Reef Health are three inputs that influence the level of tourism. With improvements in these inputs come improvements in tourism. The following is the data table in Netica for Tourism for Participant A:

The second input into Benefits, Comm. & Rec Fishing, is a discrete variable and can be either poor or good. Coral Reef Health, Marine Protect (MPAs), FishLinks, and Ocean Warm/Acid are four inputs that influence the level of Comm. & Rec Fishing. As coral reef health improves and if MPAs are applied, the probability that fishing will be good tends to increase. As ocean warming/acidification increases, the probability that fishing will be good tends to decrease. FishLinks is a discrete variable that contains elicited probabilities (that sum to 100 %) that varying sets of environmental stressors (coral reef health, MPAs, and ocean warming/acidification) will produce poor or good fisheries. FishLinks can be influenced by a node Fisheries Research, a discrete variable that allows for the possibility of testing the effects of different research outcomes for FishLinks, and updating prior probabilities based on new evidence. The probability that fishing will be good is adjusted by the elicited inputs into Fish Links, which can take into account the belief that there is synergism among variables. For example, Participant A’s inputs into FishLinks are shown in the table below:

The higher the probability placed on a set of stressors that contains coral reef health and MPAs, the higher the probability that fishing will be good. The higher the probability placed on a set of stressors that contains ocean warming/acidification, the lower the probability that fishing will be good tends. The following is the data table in Netica for Comm. & Rec Fishing for Participant A:

The third input into Benefits, Coral Reef Health, is a discrete variable and can be either poor or good. Bay & Water Quality, Marine Protect (MPAs), Coral Links, and Ocean Warm/Acid are the four inputs that influence the level of coral health. As bay water quality improves and if MPAs are applied, coral reef health increases. As ocean warming/acidification increases, the probability that coral reef health will be good decreases. Coral Links is a discrete variable that contains elicited probabilities (that sum to 100 %) that varying sets of environmental stressors (MPAs, ocean water quality, and ocean warming/acidification) will produce poor or good coral health. Coral Links can be influenced by a node Coral Effects Research, a discrete variable that allows for the possibility of testing the effects of different research outcomes for CoralLinks, and updating prior probabilities based on new evidence. The probability that coral reef health will be good is adjusted by the elicited inputs into Coral Links, which can take into account the belief that there is synergism among variables. For example, Participant A’s inputs into CoralLinks are shown in the table below:

The higher the probability placed on a set of stressors that contains ocean water quality and MPAs, the higher the probability that coral reef health will be good. The higher the probability placed on a set of stressors that contains ocean warming/acidification, the lower the probability that fishing will be good tends. The following is the data table in Netica for Coral Reef Health for Participant A:

Coral Reef Health is multiplied by Coral Eco. Services, a discrete variable with probabilities set at 25 % that they are low, 50 % that they are medium, and 25 % that they are high. Coral Eco. Services can be influenced by a node Coral Eco. Services Research, a discrete variable that allows for the possibility of testing the effects of different research outcomes for Coral Eco. Services, and updating prior probabilities based on new evidence.

Marine Protect (MPAs), an input into both Comm. & Rec Fishing and Coral Reef Health, is one of the five management options included in the model. Marine Protect (MPAs) is a discrete variable and can be either applied (Yes = 100 %) or not (No = 100 %). Implementation of MPAs is believed to increase the probabilities that coral reef health and fishing are good.

Ocean Warm/Acid, an input into both Comm. & Rec Fishing and Coral Reef Health, is a discrete variable and can be either high or low. Left to chance these probabilities are set at 50 % that it is low and 50 % that it is high.

Bay & Ocean Water Quality, an input into Coral Reef Health, is a discrete variable and can be either poor or good. Inland Water Quality and Lagoon WQ (water quality) are the inputs that influence the level of bay water quality. As the probability that inland and lagoon water quality are good increase, the probability that bay water quality will be good also increases. Lagoon WQ is a discrete variable and can be either none (if the node is not activated), poor, or good. This node is only activated when the management option, Restore Lagoon, is implemented. Restore Lagoon is a discrete variable and can be either applied (Yes = 100 %) or not (No = 100 %). Restoring the lagoon is believed to increase the probability that bay and ocean water quality will be good if inland water quality is not too poor. The following is the data table in Netica for Lagoon WQ for Participant A:

The following is the data table in Netica for Bay & Ocean Water Quality for Participant A:

Inland Water Quality, an input into Lagoon WQ and Bay & Ocean Water Quality, is a discrete variable and can be either poor or good. Total Load is the main input into Inland Water Quality. The following is the data table in Netica for Inland Water Quality for Participant A:

Total Load, the total pollution load, is a continuous variable distributed over six intervals (very low, low, moderately low, moderately high, high, and very high) and a function of individual loading sources (SewLoad, AgLoad, and DevLoad) and their associated hypothetical reductions (SewRed, AgRed, and DevRed) (management options), as shown in the following equation:

$$Total\,Load = SewLoad \times \left( {1 - \frac{SewLoad}{100}} \right) + AgLoad \times \left( {1 - \frac{AgLoad}{100}} \right) + DevLoad \times \left( {1 - \frac{DevLoad}{100}} \right)$$

SewLoad, AgLoad, and DevLoad are discrete variables and can be low, medium, high, or very high. Loading distributions for the individual sources were computed over the low, medium, high, and very high in a manner that minimized variance based on stakeholders’ prior beliefs. As the distributions tend toward the very high, the total load tends toward the very high. Loading values included in the model are relative (and therefore unitless) though roughly scale to mg/L concentrations of suspended solids in unpolluted source waters (very low = 0–25; low = 25–50), moderately polluted source waters (moderately low = 50–125; moderately high = 125–250), and highly polluted source waters (high = 250–500; very high = 500–750). The range of values used for TotalLoad of 0–750 was thought to allow for a more realistic distribution (with six intervals from low to very high) than would a smaller range. Since the analysis is of a comparison of benefits, the actual units used are not important.

SewLoad, AgLoad, and DevLoad can be influenced by the nodes Sew Load Research, Ag Load Research, and Dev Load Research, which are discrete variables with four possible outcomes each, and which allow for the possibility of testing the effects of different research outcomes, and updating prior probabilities based on new evidence. The following table in Netica for SewLoad shows the likelihood functions (false +/false- rates) for Participant A (they indicate that the research is nearly perfect, with the probability that the correct inference is made equal to 94 %):

The management options, SewRed, AgRed, and DevRed, are discrete variables and can be set to a 0 % (None), 40, 70, or 90 % reduction. The following is the first and last parts of the lengthy data table in Netica for TotalLoad for Participant A: