Uncertainty in the decision making process for the source control of priority substances

Tuomas Mattila, Finnish Environment Institute (SYKE)

Draft, 2.10.2008

In this working paper we attempt to model the propagation of uncertainty from the input data to the variables that affect decision making. The key issue in this analysis is "What is the decision to be made?" Here we try to simplify the complex issue, of finding the optimal means to reach the environmental quality standards set by the Water Framework Directive, into two questions posed to each alternative solution:

• Will this intervention increase the chances of meeting the quality standards?
• How cost effective is this intervention?

Based on these two questions we outlined the following causal diagram of the factors that affect the answer to these two questions (Figure 1). In the diagram the decision about the technological alternative (light green) will affect the cost efficiency and the probability of exceeding the EQS, but the outcome of that decision is affected by several other (uncertain) factors: baseline emissions, environmental and chemical properties and the ecotoxicological impact (represented by the EQS).   

Figure 1: A causal diagram of the factors that influence the decision making process for assessing a intervention. Blue describes calculated variables, cyan describes decision objectives and other col-ours describe sources of information (and uncertainty).

Example: assessing the removal of DEHP with improved landfill effluent treatment
In this example the concepts presented in the previous chapter are applied to a imaginary case of diethylhexylphthalate (DEHP) leaching from a landfill to a small coastal bay. The purpose is to illustrate how the incorporation of uncertainty analysis can improve the applicability and transparency of the analysis.

DEHP has been the most widely used plasticizer in PVC plastics. Due to health concerns it's use in children's toys has been banned in the EU since 2007 (www.dehp-facts.com). It is highly hydrophilic, with a logKow at about 7.5 (EQS Chemical fact-sheet). Half-lives are uncertain, being in the range of 2-3 weeks in water (US EPA 2002)

In this example it is assumed that 1 500 kg/a (1 245-1 810 kg/a) of DEHP is being leached from a landfill to a small bay with a surface area of 3 km2 and average depth of 3 m (1-5 m). The bay has a sediment depth of  4-7 cm and a water residence time of 1-6 days depending on the weather conditions. The waters are fairly turbid with solid concentrations of 30-40 mg/l and an organic carbon content of 10-20 %. The mean temperature in water is 10 °C ranging from 5-15 °C. 
It is assumed that the emissions could be reduced 20-30% by improving the treatment of effluents. It is also assumed that this treatment would have annual fixed and operating costs of 10 000 €/a. However also the cost estimate has uncertainties associated especially with the price of raw materials, amount of maintenance and the discount rate used. Based on a sensitivity analysis the costs are estimated to be 7 500-15 000 €/a.

The normal practice in environmental fate modeling would be to calculate the predicted environmental concentrations by using the most likely (median) values for chemical and environmental properties. After the calculations the sensitivity of the results would perhaps be studied by changing single parameter values at a time and recording the effects. Had this analysis been done without the uncertainty distributions, by using only the median values, the results would have been that due to the technological alternative, the EQS would have been met (with a margin of safety of 300 ng/L). The costs associated with pollutant removal would have been estimated as 50 €/kg removed. In order to keep the analysis relatively simple, the system is assumed to be in a chemical equilibrium (i.e. dissolved concentrations are equal in sediment and water, a level II fugacity model assumption).   
 
Here we expand this analysis by taking into account the probability distributions associated with each parameter and by performing a Monte Carlo simulation with the results. Several of the methods for uncertainty assessment have been described in the Environmental Fate Models section of this DSS. Here we will illustrate what the concepts mean in the framework of finding cost effective solutions to the source control of PSs. Of the various methods available, we will utilize Monte Carlo simulations, which can be considered as the state of the art tool for uncertainty assessments (Saltelli 2006). There are several software tools for making Monte Carlo simulations, the most popular being @risk, Crystal Ball, Analytica and Simlab.

By applying the above mentioned distributions to the system described in influence diagram (Figure 1) , the following distributions for the compliance to EQSs are obtained (Figure 2). Based on the probabilistic results, reduction of emissions by 20-30% resulted in an increase in the probability of complying with the EQS from 45% to 70%. However there is still a relatively high risk of non compliance (30%). Global sensitivity analysis, performed by correlating model output with input variables, revealed that the most influential parameter for determining the effect of reduction is the water phase residence time (Figure 3). Thus changes in the local hydromorphology (by dredging, constructing wetlands, dilution etc.) can have a more profound effect on the concentrations than what could be obtained with the management options. For example, decreasing the residence time below 2 days would result in reaching the EQS with 95% probability. This would not, however, not remove the substance from the environment, but only move it to somewhere else.

Figure 2: Cumulative probability distributions of the difference between predicted environmental concentrations (PEC) and environmental quality standard (EQS) with and without the proposed management option. It can be seen that the management option improves the probabilities of com-pliance but does not eliminate the risk.

Figure 3: A global sensitivity analysis of the Level II model results applied to DEHP pollution in a small coastal bay. The model output is not sensitive to the chemical half-life in water, but it is highly sensitive to the residence time of water.

Figure 4: The probability distribution for cost effectiveness of DEHP removal by treatment of land-fill effluents. The cost effectiveness depends on the local environmental conditions and emission intensities.

Figure 5: An increase from the probability of reaching the EQS from the baseline scenario (45% probability) to 75% will cost 11 k€ and to 90% will cost 21 k€. As the amount of risk decreases, the costs will increase.

References
Refsgaard, J.C., van der Sluijs, J.P., Hojberg, A.L., Van Rolleghem, P.A., 2007. Uncertainty in the environmental modeling process - a framework and guidance. Environmental modeling and software 22 1543-1556.