The objective of this research project was to develop an integrated cost-benefit analysis framework for ozone and fine particulate control, while accounting for variability and uncertainty. The framework includes air quality simulation, sensitivity analysis, stochastic multi-objective air quality management, and stochastic cost-benefit analysis. This framework has the potential for seven major uses: (1) to develop state implementation plan (SIP) control strategies; (2) to evaluate interstate transfer of pollutants and emissions; (3) to study trading across states/regions and/or pollutants; (4) to conduct cost-benefit analysis and risk management analysis; (5) to help set air quality standards; (6) to investigate the importance of uncertainty and conduct value of information analysis; and (7) to set research priorities.
Although the 1990 Clean Air Act is a federal law covering the entire country, the states do much of the work to execute the Act. For example, state air pollution agencies determine what controls and regulations will be used to meet the air quality standards. Under this law, the U.S. Environmental Protection Agency (EPA) sets limits on how much of a pollutant can be in the air anywhere in the United States. This ensures that all Americans have the same basic health and environmental protections. The law allows individual states to have stronger pollution controls, but states are not allowed to have weaker pollution controls than those set for the whole country. The law recognizes that it makes sense for states to take the lead in executing the Clean Air Act, because pollution control problems often require special understanding of local industries, geography, housing patterns, etc.
The federal standard for ozone and fine particulates falls under the regulatory authority of the EPA, but developing control strategies to achieve the standard tests is, in part, the responsibility of state governments. However, the process of developing strategies for ozone and fine particulates to meet National Ambient Air Quality Standards (NAAQS) has been a subject of considerable debate among the states and between states and the EPA, as well as in the research community. Among the many issues are: (1) the way the standards and demonstrations of attainment are formulated fails to consider meteorological and other uncertainties, resulting in an unrealistically tight de facto standard; (2) less-than-optimal emphasis on and incentives for finding cost-effective approaches to meeting standards; and (3) the lack of recognition that different control policies affect environmental endpoints differently. Currently, there lacks of a framework to integrate uncertain scientific information with policy analysis to inform decision-makers and to develop control strategies.
The proposed research has two primary objectives. The first objective is to develop stochastic source-receptor (S-R) coefficient matrices for ozone and fine particulate matter using an advanced air quality simulation model, the Urban-to-Regional Multiscale One Atmosphere Model (URM-1ATM), and an efficient sensitivity algorithm Direct Decoupled Method in Three Dimensions (DDM-3D).
In this study, URM-1ATM is used to account for the processes significantly affecting ozone and PM2.5 concentrations in the atmosphere, including atmospheric physics, gas and aerosol phase chemistry, cloud and precipitation processes, and wet and dry deposition.
The DDM-3D model is employed to calculate the local sensitivities of specified model outputs simultaneously with concentrations. It is an alternative to the brute force method, where multiple simulation runs are needed as explained in the previous paragraph. Instead, the DDM-3D simultaneously solves differential sensitivity auxiliary equations to calculate the sensitivity fields (temporal and spatial). The result of this method is the partial derivative of species concentrations with respect to minute changes in emissions. The major difference between the conventional brute force sensitivity approach and DDM-3D sensitivity analysis is the computational efficiency of the latter. The use of DDM-3D could potentially eliminate the need of a large number of individual perturbation runs.
Based on two very different meteorological episodes, the simulation outputs from URM-1ATM and DDM-3D are used to develop stochastic population-weighted and area-weighted S-R coefficient matrices for 1-hour daily maximum ozone and 24-hour daily PM2.5 average concentrations. Thirteen daily sensitivity measurements are used to quantify the variability and probability distributions of S-R coefficients. These S-R coefficients then are incorporated into an economic and stochastic multi-objective air quality management model for developing emissions control strategies for ozone and fine particulates.
The second objective is to demonstrate how to conduct analyses of alternative ozone and PM2.5 reduction policies using this modeling approach. A stochastic multi-objective programming model is developed for this purpose. Alternative objective functions include optimization of the net social benefits and maximization of the reliability of satisfying certain air quality goals. The problem could be formulated as finding the optimal net social benefit, such that the air quality at a specific state/region satisfies the air quality goal with a prespecified reliability. These models could yield information on emission control strategies of primary precursors from individual sources, the approximated distribution of net social benefits, and the approximate ozone concentration and PM2.5 concentration distribution at various geographical locations. We conducted a few illustrative analyses using 19 states/regions of the eastern United States (an area that broadly matches the contemplated NOx trading region).
Both of these objectives are important for laying a stronger scientific foundation for future control strategy development and reducing and quantifying the uncertainties in SIPs, in proposed changes in the SO2 trading program and in the design of the EPA's NOx trading program for the Eastern United States.
Below, is a list of tasks that we accomplished and findings from three major research efforts under this research project. The major accomplishments of this research include the following:
· Improved the air quality simulation model (URM-1ATM) and the sensitivity analysis algorithm (DDM-3D).
· Conducted model performance evaluation using ozone concentrations predicted by URM-1ATM and monitoring data.
· Compared modeling results using different domain grid definitions between results for the Southern Appalachian Mountains Initiative (SAMI) and the project reported on here.
· Developed a method to efficiently conduct interstate and intrastate sensitivity analysis using DDM-3D, and conducted the necessary calculations for two episodes.
· Developed programs to construct S-R coefficient matrices using URM-1ATM/DDM-3D and PAVE outputs. We constructed both population-weighted and area-weighted S-R matrices.
· Developed stochastic S-R coefficients for ozone and PM2.5 based on 13 daily samples from 2 very different meteorological episodes. We fit normal distributions to these S-R coefficients and conducted the "goodness of fit" test. We also investigated the correlations within and across S-R coefficient matrices.
· Developed piecewise linear total cost functions for utility point source NOx control.
· Developed the stochastic health benefit coefficients for ozone and PM2.5.
· Developed stochastic multi-objective air quality management models using 5, 6, and 7 as inputs to conduct: (1) benefit/cost analysis; (2) equity of control analysis; and (3) tradeoff between benefit/cost and reliability analysis.
· Constructed a stochastic simulation module using Analytica to estimate the net benefit distribution.
· Identified issues for further analysis.
URM Model Performance. The July and May 1995 base-case episodes have been used to evaluate URM performance against ambient measurements. Excluding model ramp-up days, 8 and 5 days of simulation are available for evaluation of the July and May episodes, respectively. Data from the EPA Aerometric Information Retrieval System (AIRS) are used to evaluate model performance. EPA guidelines for urban scale ozone modeling are +/- 15 percent for Normalized Mean Bias (NMB), and +/- 35 percent for Normalized Mean Error (NME). This regional scale application resulted in an average NMB for ozone of -4.4 percent for the May episode and 3.1 percent for the July episode. The NME was 17 percent for May and 22.3 percent for July. These four values are well within the stated guidelines. These values are calculated from measurements at sites coinciding with either 24- or 48-km2 cells. Data were available for more than 400 sites for each episode.
The Interagency Monitoring of Protected Visual Environments (IMPROVE) network provides 24-hour averaged speciated aerosol data taken on Wednesday and Saturday of each week in our episodes, and these data are used to evaluate model performance for aerosols. Three days of data were available during the July episode (July 12, 15, and 19) with measurements from 18 sites in total, 12 of which coincided with 24- or 48-km2 cells. These sites resulted in an average NMB for PM2.5 of -9.34 percent and an average NME of 22.34 percent. Two days of data were available for the May episode (May 24 and 27) with measurements from 17 sites in total, 11 of which coincided with 24- or 48-km2 cells. These sites resulted in an average NMB for PM2.5 of 9.77 percent and an average NME of 28.64 percent. There are no EPA guidelines to indicate acceptable model performance for aerosols.
S-R Matrices. The methods developed in this research yield the first regional set of S-R matrices derived from a unified air quality chemistry model. The receptor states/regions of interest typically cover multiple simulation grid cells. Therefore, to derive S-Rs, we aggregated individual grid cell sensitivity values to a single receptor site value. The sensitivity used for aggregation is the change of pollutant concentration at the peak of the specified time scale (1-hour or 8-hour daily maximum for ozone, and 24-hour average for PM2.5). This S-R aggregation is performed on both a population-weighted and an area-weighted basis. Population-weighted S-Rs are needed for estimating health benefits from application of source controls. They give a better proxy for health effects than area-weighted measures. From a regulatory perspective, the population-weighted S-Rs also are more useful because they better apply to the urban areas. Of course, under the Clean Air Act, people living in rural areas are accorded the same level of protection as people in urban areas. The sensitivities are normalized as the change of pollutant concentration over a specific time scale per 1,000 tons of a precursor's emissions reduction by state and by source type. These sensitivity values represent the marginal reduction of emissions from the source region to ozone or PM2.5 reduction at a receptor site.
Although a per ton sensitivity comparison is useful for evaluating equivalent emissions reductions between states, states have very large differences in what they emit and therefore in how much they are able to reduce emissions. To account for this discrepancy, we also define and present what we term "control effects" to quantify the pollution reduction in receptor states that would be achieved by reducing emissions from the 19 states/regions combined. Control effects are calculated by multiplying the sensitivity matrix by a vector of 30 percent emissions reductions from each of the source states. We use population-weighted sensitivities in these control effects calculations because the results focus on policy.
Our results show that local emissions account for only about 25 percent of local ozone concentrations and PM2.5 concentrations. We also find that reducing SO2 emissions can be 10 times more effective than reducing NOx emissions in reducing fine particulate concentrations during summer episodes with high particulate matter concentrations.
Detailed matrices were derived for sulfates, nitrates, and PM2.5, as well as ozone. Although "bounce-back effects" on nitrates are observed when SO2 emissions are reduced, they generally are not significant enough to result in perverse effects on PM2.5. Along the East Coast states in May, NOx scavenging is a geographically broad enough phenomenon to make the ozone sensitivities to NOx negative for several state/region combinations.
The ozone sensitivity results are comparable with others in the literature based on simpler models. The PM sensitivity coefficients, however, are smaller. This result may be due, in part, to the episodes chosen for study.
Area and population-weighted sensitivity matrices tell very different stories. We found that population-weighted sensitivities exceed area-weighted sensitivities by as much as six times in some regions. When evaluating health damages, one should use population-weighted sensitivities to account for human exposure.
Figure 1 provides some detail on the effect of SO2 reductions in a given state of PM2.5 concentrations in all states in the domain.
Figure 1. May and July 24-Hour Average PM2.5 Sensitivity With Respect to Total SO2 Reduction (µg/m3 per 1,000 tons/day). Relative sensitivities for each state of the top six contributing state/regions and "all other states" for the effect of a unit reduction in SO2 emissions on 24-hour average area weighted PM2.5 concentrations for both the July and May episodes.
Cost-Benefit and Net-Benefit-Reliability Analyses. We present preliminary results from a few cases to show the types of outputs produced by the framework developed in this research project.
Case 1. Optimize Expected Net Benefit Without An Air Quality Constraint. In this first case, we examined the allocation of emissions reductions that maximizes net benefits for the entire study domain, without an air quality requirement for any state. In the results table, the cost column gives the NOx control cost for the individual state. The three benefit columns (ozone, PM, and ozone and PM benefit) in Table 1 show the total benefits occurring at a receptor state because of the NOx reductions over the entire study domain. Negative net benefits indicate states where emission reductions benefit other states, but aggregated reductions do not benefit these states enough to offset the state's control costs (AL, IN, WI, and WV). This could occur for a number of reasons, such as the state's location near the upwind modeling domain boundary (AL, WI), low population levels (IN, WV) or pollution levels in that state, and/or high emissions and therefore control costs (IN, WV). Because emissions reductions in those states are found to contribute overall to net benefits, one implication of these results is that states benefiting from these reductions should share the costs of emissions reductions in the upwind states as part of their own air quality plans.
Table 1. Optimize Expected Net Benefit Without an Air Quality Constraint
NOx Reduction (tons/day)
Cost ($1,000/ day)
Ozone Benefit ($1,000/ day)
PM Benefit ($1,000/day)
Ozone & PM Benefit ($1,000/ day)
Net Benefit ($1,000/ day)
The results show that the optimal NOx reduction is about 60 percent of the baseline, and that net benefits are over $6 million per day. Interestingly, daily health benefits related to ozone reductions exceed those from PM2.5 reductions, in spite of the significantly greater potency of PM2.5 on health. This occurs because NOx emissions create very little PM2.5 relative to ozone. Ozone-based health benefits are $7.1 million per day, while PM2.5 benefits are only about 60 percent of this amount. On a per ton reduced basis, benefits are $2,400/ton NOx reduced as it affects ozone and only $1,500/ton reduced as NOx affects PM2.5. Costs are a little more than $5 million per day, or $1,700/ton NOx reduced. The largest net benefits are experienced in New York, which is not surprising given the weather patterns, which would cause New York to gain many benefits from reductions in emissions from other states, and the relatively low costs of control in New York. The latter occurs because relatively few tons are reduced in the optimal case (only 43 percent of baseline) and at a relatively low cost per ton (only around $700/ton reduced).
Case 2. Every State Net Benefit is Non-Negative. In Case 1, the net benefits across states have large variation. Some states have large positive benefits and some states have negative benefits. In the paper, we present four different methods to incorporate equity criteria in developing control strategies. Here, we only show one example. In Case 2, we show that the expected net benefit for each state is constrained to be non-negative.
For this case, we just add an additional constraint for every state/region and say that the net benefit for the individual state has to be at least greater than zero. The results of adopting the first equity criteria are in Table 2.
Table 2. Every State Net Benefit is Non-Negative
NOx Reduction (tons/day)
Cost ($1,000/ day)
Ozone Benefit ($1,000/ day)
PM Benefit ($1,000/day)
Ozone & PM Benefit ($1,000/ day)
Net Benefit ($1,000/ day)
When comparing results in Table 2 with results for Case 1, total net benefits are reduced from $6.5 million per day in the optimal case to $6.2 million per day. States that have negative net benefit in Case 1 cut back their emissions control to reduce the control costs. This difference is fairly minor because few states are in the negative net benefit category.
Case 3. Tradeoff Between Expected Net Benefit and Reliability Level Required to Reduce DEMD Peak Ozone Concentration From 96.3 ppb (Baseline Concentration) to 93.5, 94, and 94.5 ppb. In this case, we investigate the tradeoff between net benefits for the entire study domain and the reliability level of DEMD meeting ozone reduction from predicted baseline concentration 96.3 ppb to 93.5, 94 and 94.5 ppb. The selection of DEMD and these ozone goals are for illustrative purposes only. If one selects an upwind state instead or a more stringent air quality goal, it is likely that there will be no feasible solutions because of the small S-R coefficients toward the upwind state and/or the very low baseline emissions.
Instead of selecting a single state for this analysis, it is theoretically possible to select all states/regions to consider a given pollution reduction target jointly. However, technically, this problem is very difficult to solve because a joint probability distribution for the entire system is needed. Because each state has different baseline pollutant concentrations as well as the upwind state issue we mentioned above, the feasible range for emission reductions are different for individual states. In general, downwind states have a bigger range; so it is very possible that we will not be able to find a feasible solution for the entire system with the same reduction target. One option is to consider only downwind states jointly, because they are more likely to have big reduction ranges.
Because of meteorological variability, we expect that the reliability-net benefits curve would be downward sloping. More emissions would need to be reduced to meet the air quality goal with higher reliability requirement. Because the benefit function is linear and marginal costs increase with greater emissions reductions, net benefits fall as reliability increases.
The results in Figure 2 bear out this expectation. Indeed, the net benefit can go from positive to negative as the reliability requirement gets higher and higher. Meeting a given air quality goal can even become infeasible for reliability requirements above a certain level. In this illustrative case, the maximum reliability for DEMD to meet air quality goals of 93.5, 94, and 94.5 ppb is 67 percent, 83 percent, and 93 percent, respectively. This solution is found by treating the reliability parameter as an unknown variable, moving it to the objective function, and maximizing it.
Figure 2. Tradeoff Between Expected Net Benefit and Reliability Level Required to Reduce DEMD Peak Ozone Concentration From 96.3 ppb to 93.5, 94, and 94.5 ppb
Another interpretation of Figure 2 is that given the same reliability requirement, as the air quality target gets more stringent, the costs and benefits increase, but as the former increases faster than the latter, net benefits fall. The tradeoff curves shift unfavorably from right to left.
A third interpretation of this analysis could be useful for setting air quality standards. Draw a horizontal line parallel to the x-axis and across the three tradeoff curves. On this "iso net benefit" line, net benefits are constant. Thus, the same value of net benefits can be achieved using different combination of the air quality goal and the reliability requirement.
This project demonstrated a method to develop stochastic S-R relationships, and used those findings for an initial assessment of developing optimized control strategies. The results of this research show the potential of the approach, and provide some very important results. However, due to project constraints, only two episodes were studied. Although those episodes were chosen to represent very different meteorological conditions, they still do not span the whole range of conditions that lead to elevated ozone levels, and certainly do not include periods leading to high particulate matter in the winter. The number of episodes should be increased; this was done for the SAMI project, and could be extended to this project relatively efficiently.
Second, the costs were only available for electric utility sources of NOx. As shown, much of the benefits from emissions controls are derived from lower PM, much of which is sulfurous or carbonaceous in nature. The type of analysis conducted here should be extended to include other PM species. These initial results suggest that significant benefits can be found between differing strategies.
Publications and Presentations:
|Journal Article||Krupnick A, Shih J-S, Bergin MS, Russell AG. Integrated cost-benefit analysis for ozone and fine particulate control under variability and uncertainty. Journal of Environmental Economics and Management. |
|Journal Article||Bergin MS, Shih J-S, Boylan JW, Wilkinson JG, Krupnick A, Russell AG. Inter- and intra-state NOx and SO2 emissions impacts on ozone and fine particulate matter. Environmental Science and Technology. |
|Journal Article||Bergin MS, Boylan JW, Shih J-S, Wilkinson JG, Krupnick A, Russell AG. Multi-scale grid definition impacts on regional, three-dimensional air quality model predictions and performance. Environmental Science and Technology. |
air, ambient air, ozone, particulate matter, risk management, public policy, cost-benefit analysis, variability and uncertainty, stochastic modeling, stochastic simulation, multi-objective programming, decision-making, tropospheric ozone, National Ambient Air Quality Standards, ambient ozone reduction, benefits assessment, ecological assessment, ecosystem valuation, environmental equity, environmental values, health valuation models, integrated modeling, ozone abatement, photochemical model, phototchemical modeling, public policy, stakeholder, tropospheric ozone destruction. , Air, Economic, Social, & Behavioral Science Research Program, RFA, Scientific Discipline, Ecology and Ecosystems, Economics, Economics & Decision Making, decision-making, tropospheric ozone, National Ambient Air Quality Standards, ambient ozone reduction, benefits assessment, cost/benefit analysis, ecological assessment, ecosystem valuation, environmental equity, environmental values, health valuation models, integrated modeling, ozone, ozone abatement, photochemical model, phototchemical modeling, public policy, stakeholder, tropospheric ozone destruction