Optimal Ozone Control with Inclusion of Spatiotemporal Marginal

Article pubs.acs.org/est

Optimal Ozone Control with Inclusion of Spatiotemporal Marginal Damages and Electricity Demand S. Morteza Mesbah Department of Civil and Environmental Engineering, Carleton University, 1125 Colonel By Drive, Ottawa, Ontario K1S 5B6, Canada

Amir Hakami* Department of Civil and Environmental Engineering, Carleton University, 1125 Colonel By Drive, Ottawa, Ontario K1S 5B6, Canada

Stephan Schott School of Public Policy and Administration, Carleton University, 1125 Colonel By Drive, Ottawa, Ontario K1S 5B6, Canada S Supporting Information *

ABSTRACT: Marginal damage (MD), or damage per ton of emission, is a policy metric used for effective pollution control and reducing the corresponding adverse health impacts. However, for a pollutant such as NOx, the MD varies by the time and location of the emissions, a complication that is not adequately accounted for in the currently implemented economic instruments. Policies accounting for MD information would aim to encourage emitters with large MDs to reduce their emissions. An optimization framework is implemented to account for NOx spatiotemporal MDs calculated through adjoint sensitivity analysis and to simulate power plants’ behavior under emission and simplified electricity constraints. The results from a case study of U.S. power plants indicate that time-specific MDs are high around noon and low in the evening. Furthermore, an emissions reduction of about 40% and a net benefit of about $1200 million can be gained for this subset of power plants if a larger fraction of the electricity demand is supplied by power plants at low-damage times and in low-damage locations. The results also indicate that the consideration of temporal effects in NOx control policies results in a comparable net benefit to the consideration of spatial or spatiotemporal effects, thus providing a promising option for policy development.

1. INTRODUCTION In the U.S., the main portion of electricity (e.g., 37% in 2012) is generated by coal-burning power plants,1 placing them among the major sources of nitrogen oxides (NOx = NO + NO2) emissions, main precursors of ground-level ozone. The ozone formation potential of these precursors may vary significantly by location and time.2 Because of these temporal and spatial differences, health impacts of NOx emissions can vary up to 12 times in magnitude on a regional scale3 or up to 68 times on a national scale.4 Inclusion of such differences in policy design can result in improved public health,4−7 by simply redistributing emissions without a need for emission reduction. Furthermore, redispatching or shifting electricity generation from one time or location to another can reduce NOx emissions significantly and result in a reduction in ozone concentrations while meeting the electricity demand.8−10 Redispatching can be achieved through enforcement of an emission fee, which in turn raises the cost of generation for more polluting units. A previous study suggests that dispatching on the basis of local meteorological conditions and electricity demand may be an efficient method for control of surface ozone, comparable to © 2015 American Chemical Society

other emission control technologies such as installation of selective catalytic reduction (SCR) or selective noncatalytic reduction (SNCR).9 Redispatching has also been considered for introducing electric vehicles into the fleet11 or for reducing water consumption of power plants at regions affected by drought.12 An effective NOx emissions control policy aims to minimize the system-wide social cost that includes emission abatement costs for all polluters and the damage they impose on the environment and human health (external costs). The external cost caused by a polluter (hereafter damage) is calculated from the polluter’s marginal damage (MD), defined as the dollar value of damage for an additional ton of emissions.13 Cap-andtrade programs are designed to achieve lower social costs by providing cost-saving incentives for participants while capping the total emissions. For uniformly mixed pollutants with long Received: Revised: Accepted: Published: 7870

March 7, 2015 May 16, 2015 June 8, 2015 June 8, 2015 DOI: 10.1021/acs.est.5b01178 Environ. Sci. Technol. 2015, 49, 7870−7878

Article

Environmental Science & Technology

2. METHODOLOGY An adjoint model is combined with an optimization tool to include the MDs in the design of an improved NOx emissions control policy. Adjoint simulations are carried out to calculate the MDs, which are then used as inputs into the optimization model to predict the emission levels under different policies. The MDs in this work are defined for a damage function that includes the short-term ozone mortality in the U.S.22,27

atmospheric lifespans (e.g., carbon), an optimal cap on total emissions can limit the system-wide damage and minimize the social costs. However, a cap on total emissions does not necessarily limit the system-wide damage for short-lived pollutants such as NOx, whose MDs vary significantly by location and time. By differentiating emissions by MDs through exchange rate cap-and-trade policies4,5,7,14 or taxation,15−17 the system-wide impact of NOx emissions can be more effectively reduced. Both exchange rate and taxation policies provide higher emission reduction incentives for high-MD polluters than for low-MD emitters. For example, a fee per ton of emissions provides emission reduction incentives for high-MD polluters because their marginal abatement costs (MACs), or costs per additional ton of emission reduction, is more likely to be less than the imposed fee. The accurate determination of the MD for a NOx polluter then becomes crucial for determining the exchange rates or taxes required to minimize ozone concentrations and corresponding damages. The MD of a polluter is the derivative of the system-wide damage function with respect to emissions from the polluter. There have been various efforts to calculate MDs with differing levels of complexity. Some studies have used traditional sensitivity analysis methods that require running an air quality model for each additional source. This method is computationally expensive and has only been used for a limited number of sources.7,18,19 Simplified or reducedform models have been used to estimate MDs for a large number of sources.13,20,21 However, these simplified approaches do not account for all different physical and chemical processes influencing the fate of pollutants in the atmosphere. Recent studies have used backward (adjoint) sensitivity analysis to calculate MDs for a large number of sources.22 The adjoint method accounts for different physical and chemical processes included in the atmospheric model and computes MDs at comparatively low computational expense. The adjoint model is efficient for calculating the sensitivity of a desired function of outputs, such as ozone induced mortality, known as the adjoint cost function, with respect to numerous individual inputs, such as emissions from sources at different times and locations.22 Further details on the adjoint model and its mathematical formulations can be found elesewhere.23−26 This work is a continuation of our previous study on the application of adjoint models in policy instruments. Previously, we used the adjoint model to calculate exchange rates5 and location-specific MDs3, both of which were used to spatially differentiate between emissions, whereas the role of electricity demand was excluded. In this work, we add new dimensions by including temporal and spatiotemporal emission differentiation and demand-based redispatching strategies. The MDs are used within an optimization framework to investigate how setting variable NO x prices on the basis of temporal and spatiotemporal MDs would impact the redispatching strategy and the system-wide health damage. In comparison to previous studies on redispatching,8−11 this work does not model the electricity network but accounts for electricity demands. However, this study includes the spatiotemporal MD information to investigate the impact of redispatching strategies on ozone concentrations and the corresponding damage to human health. As such, MD information is used to set different spatial or temporal emission fees rather than setting one fee on the basis of hypothetical scenarios used in previous studies.8−10

MD =

∂TD ∂E

(1)

where (∂TD)/(∂E) is the derivative of total damage (TD) with respect to emissions (E). To estimate MD, a linearized form of TD is used (i.e., TD = ∑i(VSL M0i Pi∑t(1 − e−βΔCit) ≈ ∑i(VSL M0i Pi β∑tΔCit), where VSL is a value of statistical life, M0i is the baseline nonaccidental mortality rate at location i, which covers all grid cells in the U.S., Pi is the location-specific population, β is the concentration response factor, and ΔCit is concentration compared to baseline concentrations at location i and time t. Note that because of the small value of β linearization of the damage function entails negligible error. The optimization framework in this work is an extension of our previous study3 with a few notable differences. First, the MDs in this study are location- and time-specific, whereas the previous work did not consider temporal variability in MDs. Second, in the current optimization framework, the relationship between the power plants’ electricity generation and emissions has been included as an additional constraint within the optimization framework. This constraint has been added to account for power plants’ emission behavior in the electricity and emission markets where they have to supply electricity while meeting the emission requirements. 2.1. Polluters’ Behavior. Cap-and-trade systems are designed to minimize the system-wide emission reduction costs that govern the polluters’ behavior under such systems. The following optimization problem can be used to predict the distribution of polluters’ emissions leading to the minimum system-wide abatement cost for n polluters: Minimize: 24

n

∑ ∑ ci(eit )

(2a)

t=1 i=1

Subject to: 24

n

∑ ∑ eit ≤ ET

(2b)

t=1 i=1 24

n

24

n

∑ ∑ qit = ∑ ∑ (eit /R i) = QT t=1 i=1

qit ∈ [0, Gi]

t=1 i=1

(2c) (2d)

where eit is a variable representing the emissions from polluter i at hour t integrated over the ozone season; eq 2a is the objective function of the optimization problem and is defined as the summation of the source-specific abatement cost functions (ci) of n polluters (ci(eit) = MACieit). ET is the system-wide cap on emissions, qit is a variable representing electricity generation, Ri is the emission intensity (i.e., the ratio of emissions to generations) for polluter i and is a constant that depends on technical specification and control technology for each unit, QT 7871

DOI: 10.1021/acs.est.5b01178 Environ. Sci. Technol. 2015, 49, 7870−7878

Article

Environmental Science & Technology is the total electricity demand in the ozone season, and Gi is the hourly generation capacity of polluter i. Equation 2b ensures that the total emissions are less than the cap, whereas eq 2c ensures that the total electricity demand in the system is supplied by the power plants. Equations 2a−d extend our previous application of this approach, which included no consideration of temporal effects.3 Our previous study also did not address electricity demand and therefore the emission constraint was set equal to the total cap (as opposed to less than or equal as in eq 2b), an assumption that is not needed in the presence of a demand constraint. Equations 2a−d are designed to find a distribution of emissions regardless of their spatial and temporal impacts on human health. To account for spatial and temporal effects of emissions, a spatiotemporal social cost minimization problem can be defined as follows: Minimize: 24

hourly demand, which are then used as constraints on electricity generation under different optimization frameworks. The adjoint of the gas-phase community multiscale air quality (CMAQ) model version 4.5.1 is used to calculate the source-specific MDs.24,29 The CMAQ model is driven by meteorological inputs generated by the weather research and forecasting (WRF) model30 and emission inputs generated by the sparse matrix operator kernel emissions (SMOKE) model.31 The emission inventory used for the SMOKE model is based on the 2006 National Pollutant Release Inventory (NPRI) for Canada and the 2005 National Emissions Inventory (NEI) for the U.S., which are projected to the year 2007 on the basis of population and economic growth factors. The power plant emissions are based on the Continuous Emission Monitoring (CEM) data for 2007. To use the CEM data, we conducted additional SMOKE simulations for the power plants included in this study. Modeling is conducted over a North American domain with a 36 km grid resolution with 34 vertical layers, for a period corresponding to the ozone season (May−September, inclusive) of 2007. The performance evaluation of ozone simulations results in an 18% mean normalized error and a 1% mean normalized bias. The hourly electricity generation and emissions are taken from U.S. EPA clean air market data.32 The emission intensity for each power plant is calculated on the basis of the ratio of a plant’s total emissions to its total electricity generations during the ozone season. Cost estimation is for the short-term and is based on the US EPA Integrated Planning Model.33 The damage function in this paper is for short-term ozone mortality and is defined on the basis of a value of statistical life of $6.8 million34 and an epidemiological response factor of 0.051% estimated for average 8 h ozone.35 The International Classification of Disease (ICD)-10 code A-R36 is used to calculate spatially variable baseline nonaccidental mortality rates.

n

∑ ∑ ci(eit ) + Dit (eit ) t=1 i=1

(3a)

subject to same the constraints as eqs 2b−d. Equation 3a represents the spatiotemporal social cost of emissions and accounts for both abatement costs and health damage costs (Dit(eit) = MDiteit) from individual polluters. Equation 2c guarantees that power plants supply the total electricity demand, but it does not provide any restriction on hourly generation. Therefore, eq 3a, combined with eqs 2b−d, are referred to as flexible social cost minimization and allow power plants to change their hourly generation from one hour to another. This assumes flexibility in demand, meaning that consumers are willing to adjust to the revised damage-driven electricity production. Without such an assumption, the hourly electricity generation must meet the hourly electricity demand, which we refer to as demand-based social cost minimization. This optimization problem has a similar objective function as eq 3a, but eq 2c is replaced with eq 3b, shown below. n

n

∑ (eit /R i) = ∑ qit = QT , t = 1, ..., 24 i=1

3. RESULTS AND DISCUSSION The location of selected power plants and their average location-specific MDs are shown in Figure 1. The outputs of the adjoint model are location- and time-specific MDs for the simulation period. For the 5 month simulation period, each

i=1

(3b)

Equation 3b guarantees that the postminimization hourly system-wide electricity generation is equal to the prior electricity demand for hour t (Qt). 2.2. Case Study. A case study of 218 coal-fired electricity generating units in the eastern U.S. is conducted to examine the spatiotemporal effects of NOx emissions under different policies. These policies include: (1) the cost minimization (CMIN) policy, similar to cap-and-trade, which only accounts for abatement costs (eqs 2a−d), (2) a flexible social cost minimization (F-SCMIN) policy (eqs 3a and 2b−d) under which consumers and power plants are flexible in hourly electricity generation, and (3) a demand-based social cost minimization (D-SCMIN) policy on the basis of eqs 2b, d and 3a,b, under which power plants must supply the same hourly electricity demand. For each policy, a global optimization package (KNITRO 8.0)28 is used to solve the corresponding optimization problem and find the postoptimization emissions distribution. Our case study covers only a portion of electricity generation and demand. As the generation is equal to the demand at equilibrium, we use the actual generation by individual power plants and calculate the seasonal as well as

Figure 1. Average location-specific MDs for the selected power plants. Only part of the simulation domain that contains the studied power plants is shown. 7872


Article


Figure 2. Color-coded location- and time-specific MDs for the studied power plants. The dark blue cells of the matrix are locations and times with negative MDs. The graph on the right panel is the average of each row and represents the average location-specific MDs (MD(x, y)). The graph at the bottom is the average of each column and represents the average time-specific MDs (MDt).

location has (153 × 24 h) specific MDs. To calculate the average location- and time-specific MD, the model outputs are averaged as follows: MDt (x , y) =

1 ∑k ∑d ekdt (x , y)

in the Ohio River Valley have low MDs because of a low population density as well as high availability of NOx.37 Location- and time-specific MDs are shown in a color-coded matrix in Figure 2. Each row of this matrix represents the MD at a specific location for different hours. Although there is similarity in average location-specific MDs within some states, which supports the calculation of a single-state-wide MDs used in some studies,20,38 there is significant variability in average location-specific MDs for other states. For example, New Jersey contains three grid cells with average location-specific MDs of $11 500/ton, $15 100/ton, and $29 700/ton. The lowest value of the time- and location-specific MD shown in the color coded matrix in Figure 2 is −$8500/ton in Pennsylvania at hour 18. The highest value ($52 700/ton) occurs at noon in Pennsylvania. This latter grid cell also contains the highest ratio of maximum MD (hour 12) to minimum MD (hour 19) for single cell. This discrepancy in hourly MDs amounts to an avoided mortality per season for transfer of each 132 tons of NOx emissions from hour 12 to hour 19 during the ozone season. The general pattern of change for hourly MDs is also consistent for different locations. The highest average timespecific MD is $22 300/ton for hour 12, and the lowest average time-specific MD is $7400/ton at hour 19. Overall, hours 11− 14 are high-MD hours (with MDs greater than $20 000/ton), and hours 18−21 appear as low-MD hours (with MDs less than $10 000/ton). Hourly MDs increase from morning to noon, when the highest MDs occur in most locations, and reach a minimum in the evening. This pattern for hourly MDs exists because of the dependency of photochemical ozone formation reactions on the temperature and intensity of sunlight as well as the transport patterns to more densely populated areas. Different hourly-MD-based fees on power plants would affect the generation pattern. However, hourly generation is also a

∑ ∑ (ekdt (x , y) MDkdt (x , y)) k

d

(4)

where MDt(x,y) is the location- and time-specific MD at the grid location (x, y) and hour t; ekdt(x, y) and MDkdt(x, y) are emissions and MD, respectively, at layer k, day d, hour t, and grid location (x, y). Note that MDt(x, y) is the MD at hour t averaged over 153 days. To investigate the temporal and spatial effects separately, two average MDs are introduced. The first is the average timespecific MD (MDt), which is calculated by averaging MDs over N grids with power plants (the number of grids in color shown in Figure 1) for a particular hour (eq 5a). The second is the average location-specific MD (MD(x, y)), which is calculated by averaging hourly MDs over 24 h for a specific location (eq 5b). MDt =

∑x , y MDt (x , y)

MD(x , y) =

(5a)

N ∑t MDt (x , y) 24

(5b)

The average location-specific MDs (Figure 1) exhibit specific patterns. Power plants in the southeast generally have higher MDs because of the governing NOx-limited atmospheric regime where ozone concentrations are highly and positively sensitive to NOx reduction; power plants in the northeast also have high MDs because this region is densely populated and has areas with NOx-limited atmospheric regimes. Power plants 7873


Article

Environmental Science & Technology Table 1. Differences under the Cost Minimization Policies Considered temporal specificity spatial specificity hourly demand constraint

CMIN

SPT-D- SCMIN

TMP-D- SCMIN

D-SCMIN

F-SCMIN

SPT-F SCMIN

TMP-F-SCMIN

no no yes

no yes yes

yes no yes

yes yes yes

yes yes no

no yes no

yes no no

policies. Table 1 summarizes different aspects of the policies examined in this study. Social cost minimization results in different total damages and costs under each policy (Table 2). All policies are then compared with the CMIN policy, which is equivalent to the cap-and-trade system minimizing the abatement costs (not social costs) for the short-term.

function of electricity demand, which is usually higher in the late afternoons and evenings when people return home and engage in cooking and other energy-intensive activities. High demand for electricity in the evening is beneficial for damage minimization because it shifts generation and corresponding emissions to evening hours when emissions are less harmful. Shifting emissions to low-MD hours lowers health damages, thus favoring damage minimization. In contrast, seasonal electricity demand fluctuations are not in favor of damage minimization because demand is high on hot summer days (usually late summer) when MDs are similarly high.39,40 It should be noted that the current cap-and-trade program in the eastern U.S. has no limitation on hourly or daily emissions and only caps the total NOx emissions during the ozone season. Therefore, higher demand on hot summer days results in disproportional use of emission quotas on days when emissions are more harmful, leading to an increase in health damages.8 One approach to alleviate this problem would be to split the ozone season into peak and off-peak periods with distinct emission quotas. There is no negative average time-specific or average location-specific MD for the power plants studied. This finding is consistent with the previous studies where some urban polluters had negative NOx PM-based MDs, but electricity generating power plants did not.20 Note that the time- and location-specific MDs reported are averaged over 153 days, and the daily values have more variability and occasionally include negative values. In fact, some of the studied electricity generating units, with positive average location-specific MDs, have negative MDs for some hours. (See the MD matrix in Figure 2.) The negative hourly values are observed in ten locations. For the majority of these locations, the negative MDs occur in hours 17−19. Exceptions are power plants in Indiana and Michigan, which have negative values for some hours in the morning and around noon. A negative MD indicates a NOxinhibited atmospheric regime along the emission trajectory where the ratio of ambient NOx to VOCs is high. Note that NOx availability in a particular location is not necessarily limited to NOx emissions in the same location and can include NOx emissions from upwind sources. The three sets of gradients shown in Figure 2 can be used to evaluate the temporal and spatial effects of NOx emissions redistribution. As explained earlier, social costs can be minimized using the location- and time-specific MDs, both with and without limitations on hourly electricity generation (D-SCMIN and F-SCMIN, respectively). To consider only spatial effects, the two social cost minimization problems are conducted with one average location-specific MD for all 24 h. Because these optimizations only account for spatial differences in MDs, they are referred to as the SPT-D-SCMIN and SPT-FSCMIN policies, for demand-based and flexible-minimization cases, respectively. Similarly, to account for only temporal effects of NOx emissions, social cost minimizations are carried out with the same average time-specific MD in all locations regardless of where the power plants are located. These social cost minimizations account only for temporal effects and are referred to as the TMP-D-SCMIN and TMP-F-SCMIN

Table 2. Abatement Costs and Health Benefits (Millions of Dollars) For Different Policies Compared to the CMIN Policya

total emissions (1000 ton) increase in abatement cost ($ million) decrease in seasonal damage ($ million) net benefit ($ million)

SPT-DSCMIN

TMP-DSCMIN

DSCMIN

FSCMIN

SPT-FSCMIN

TMP-FSCMIN

108

101

108

110

108

102

94

89

94

94

89

89

1263

1154

1286

1315

1248

1203

1169

1065

1192

1221

1159

1114

a

Under the CMIN policy, the system-wide abatement cost is $1515 million, total damage is $2610 million, and total emissions are 181 thousand tons.

In all cases, total emissions are lower than the total cap on emissions (i.e., 204 000 tons). This implies that the actual emission cap was not stringent enough because additional net gains could be realized by reducing aggregate emissions. Hourly electricity generations under demand-based optimizations (SPT/TMP-D-SCMIN and D-SCMIN) and total generation under flexible demand optimizations (SPT/TMP-F-SCMIN and F-SCIM) are, however, a binding constraint. Note that the total electricity generation under the CMIN policy and the SCMIN policies are the same, but generation and corresponding emissions are transferred from high- to low-emissionintensity power plants. The main reason for differences in total emissions under the CMIN and SCMIN policies is the imposed additional external damage constraint on power plants under the SCMIN policies. Such extra costs will provide lowemission-intensity power plants in low-MD locations an advantage over high-emission-intensity units in high-MD locations. In other words, power plants with lower damage per unit of electricity generation ($/GWh) are among those that generate more electricity when the social cost is minimized. Shifting emissions from high- to low-emission-intensity power plants results in a significant emissions reduction and corresponding health benefits under the SCMIN policies (Table 2). One important finding from the comparison of different social cost minimizations is that the net benefit of policies that account for averaged spatial or temporal effects are comparable with those that account for spatiotemporal effects combined. Although using average MDs under the SPT/TMP-D/F7874


Article


Figure 3. (a) Hourly electricity generation. Hourly NOx emissions from the power plants studied during the ozone season under (b) flexible demand policies and (c) demand-based policies.

MDs. The variability in MDs is important because it provides more flexibility for transfer of emissions from high- to low-MD locations or times, resulting in a more substantial reduction in total damage. Results shown in Table 2 also suggest that although the assumption of flexible demand increases the net benefit, the increase is not significant. The TMP-D-SCMIN and TMP-FSCMIN policies encourage a shifting of emissions from high- to low-MD hours and differ only in the constraint on hourly electricity generation. This difference changes the distribution of hourly emissions (Figure 3c) from a shape similar to that of hourly electricity generation under the base case (Figure 3a) to a shape with peaks at high- and low-MD hours (Figure 3b). The resulting difference in net benefits ($48 million) is not significant because low-MD hours, which emissions are shifted to, are also high-demand hours (Figure 3a). The difference in total emissions between the two policies at hour 19 is about 2000 ton, and at hour 13, it is about 1000 ton (Figures 3b,c). Note that for the most part the net benefit under both the TMP-D-SCMIN and TMP-F-SCMIN policies is a result of a 44% reduction in emissions as compared to the CMIN policy (an average of 3000 ton per hour). We note that the 36 km grid resolution used in this study is too coarse to properly represent health effects at urban locations; however, because of the computational cost of

SCMIN policies results in a lower net benefit compared to that from the D/F-SCMIN policies, the additional benefits from inclusion of spatiotemporal damages are relatively small (i.e., 2−11% additional benefits, Table 2). This is because under social cost minimization, emission reductions are beneficial as long as the MAC is less than the additional benefit of emission reduction (MD). The MDs used under the SPT/TMP-D/FSCMIN policies are average location-specific and time-specific MDs. On the basis of our previous estimations, the average MAC is approximately 10 times smaller than the average location-specific MDs.3 Because these MDs are still higher than the MACs, social cost minimization under these policies results in more or less similar emissions reductions and net benefits when compared to those under the D/F-SCMIN policies. It should be noted that the costs of emissions reduction for different hours are assumed to be the same, and therefore the differences in hourly abatement costs are not reflected in the results. This is one of the limitations of this study because the opportunity cost of emissions reductions by reducing electricity level changes with fluctuations in electricity prices. The difference in outcomes for different SCMIN policies is also dependent on the variation in MDs under different policies. The ratio of the 95th to the 5th percentile values are 3.5 for the average location-specific MDs and 2.7 for the average timespecific MDs, whereas this ratio is 6.7 for the spatiotemporal 7875


Article


generation under each policy, such that a higher portion of the generation is assigned to low-emission-intensity power plants located in low-damage regions. A key policy insight is that more power generation during evening peak hours may be beneficial even if units with higher emission intensities are used. Assuming flexibility in demand can improve the performance of the system but with less significance than temporal or spatial effects. The effect of flexibility in demand increases the net benefit by 5% under a TMP-F-SCMIN policy as compared to that of a nonflexible temporal SCMIN policy (TMP-D-SCMIN, Table 2). Flexibility in demand can be created to some extent by setting higher costs for electricity in high-MD hours so that consumers use less electricity during those times. The impact is not very significant because the demand for electricity is not price-elastic (i.e., the demand does not decrease by the same rate as the price increases). In the U.S., the range of price elasticity for residential electricity demand is between −0.20 and −0.35 in the short term and between −0.3 to −0.8 in the long term,44 meaning that for a 1% decrease in electricity consumption for a particular hour, the price needs to be increased by 3−5%. Our most significant finding is that consideration of temporal effects under the TMP-D/F-SCMIN policies is almost as effective as consideration of spatial effects under the SPT-D/FSCMIN policies or of spatiotemporal MDs under the D/FSCMIN policies. For example, the increase in net benefit by switching from the TMP-D-SCMIN policy to the SPT-DSCMIN policy is 10% and to the D-SCMIN policy is 12% (Table 2). Policymakers would need to decide whether the additional net benefit provides adequate motivations to switch to spatial or spatiotemporal policies considering the level of complexity and implementation challenges entailed. Under temporal policies, participants and decision makers are only exposed to 24 unique MDs as compared to many spatial MDs under spatial or spatiotemporal policies. To moderately increase system performance, inclusion of regional hourly MDs can also be considered. On an operational basis, emission redistribution can be achieved through electricity rate adjustments and shifting to different energy sources on the basis of the time of day. Besides, spatial differentiation of emissions from power plants in different regions can lead to equity considerations, potential litigations, and political conflicts between stakeholders and states.45 Temporal differentiation of emissions does not cause the same level of complexity and interregional inequities because the same temporal constraint is applied to all sources regardless of the location. In addition, the temporal MD regulations can be tailored to the specific consumption patterns of a region or state. This advantage can ease the path toward the implementation of damage-based economic instruments and more effective internalization of external costs of electricity generation.

adjoint simulations, evaluation of these policies at a higher spatial resolution was not practical. Furthermore, simulations at a coarser resolution tend to underestimate the health damages; as such, simulations at a higher resolution are likely to result in larger estimated benefits from inclusion MDs in policy instruments.41,42 Furthermore, our estimation of NOx health damages is through formation of O3 only because a full adjoint model for all aerosol process in CMAQ is not currently available. Exclusion of NOx marginal damage through generation of secondary inorganic PM is likely to be another source of underestimation in reported benefits. We use linear optimization as driven by adjoint-based MDs that are calculated at baseline emission levels. We assume that MDs remain constant despite changes in emissions.5 To account for a nonlinear damage function in the optimization, an iterative optimization process would have been required, which is infeasible for the number of policies we consider. Although the linear optimization process is quite fast, the calculation of MDs by the adjoint model is a computationally expensive process. Each adjoint simulation requires a forward run to drive the backward calculations and an adjoint run that takes 4−5 times longer than the forward simulation. In a previous work,5 we showed that on a relative basis (i.e., for calculation of emission exchange rates) MDs remained fairly constant over a sizable range of change in emissions. Further examination of the assumption of constant MDs (for the CMIN policy) reveals that MDs remained relatively constant even if the optimization was conducted iteratively (Figure S1, Supporting Information). Although some noticeable differences exist (indicating nonlinearity), the overall behavior is close to linear because MDs remain relatively unchanged. Nonlinearity in atmospheric chemical response often occurs in the case of a shift in the chemical regime.43 Figure S1 suggests that as expected the change in power plant emissions resulting from the investigated policy is not large enough to induce a significant shift in the overall atmospheric chemical regime. Note that although a linear damage function does not greatly impact the predicted emissions behaviors it causes an underestimation in total damages because most points in Figure S1 are above the one-on-one line. Benefit−cost analysis and sensitivity models used here have restrictions and limitations, such as being subject to uncertainty in the air quality model, abatement, and damage cost estimations.2,4 However, our study provides further evidence in support of the inclusion of time-based damage information in economic instruments for ozone control. Another limitation of the current work is that the optimization framework presented does not account for the electricity generation dynamics and network transmission constraints, which limits the extent to which redispatching is possible. However, this constraint is not believed to have a significant impact on the NOx redistribution patterns. A previous study on the subject suggests that the lack of sensitivity to transmission constraints is possibly due to spatial heterogeneity between low- and high-emission-intensity units within a single electricity generating zone, allowing for redispatching within the zone without a significant increase in the use of transmission lines.8 The SCMIN policies considered in this study result in a decrease in total damage by 44−50%, comparable reductions in total emissions, and a 6% increase in abatement costs as compared to those of the baseline CMIN policy. For all of the policies considered, the total electricity generation is the same, and the only difference is the distribution of electricity

■

ASSOCIATED CONTENT

* Supporting Information S

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.est.5b01178.

■

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Phone: 613-520-2600 ext. 8609. Fax: 613-520-3951. 7876


Article

Environmental Science & Technology Present Address

(17) Tong, D. Q.; Mauzerall, D. L. Spatial variability of summertime tropospheric ozone over the continental United States: Implications of an evaluation of the CMAQ model. Atmos. Environ. 2006, 40, 3041− 3056. (18) Tong, D. Q.; Muller, N. Z.; Mauzerall, D. L.; Mendelsohn, R. O. Integrated assessment of the spatial variability of ozone impacts from emissions of nitrogen oxides. Environ. Sci. Technol. 2006, 40, 1395− 1400. (19) Wang, L.; McDonald-Buller, E.; Allen, D. T. Photochemical modeling of emissions trading of highly reactive volatile organic compounds in Houston, Texas. 2. Incorporation of chlorine emissions. Environ. Sci. Technol. 2007, 41, 2103−2107. (20) Fann, N.; Fulcher, C. M.; Hubbell, B. J. The influence of location, source, and emission type in estimates of the human health benefits of reducing a ton of air pollution. Air Qual., Atmos. Health 2009, 2, 169−176. (21) Levy, J. I.; Baxter, L. K.; Schwartz, J. Uncertainty and variability in health-related damages from coal-fired power plants in the United States. Risk Anal. 2009, 29, 1000−1014. (22) Pappin, A. J.; Hakami, A. Source attribution of health benefits from air pollution abatement in Canada and the United States: an adjoint sensitivity analysis. Environ. Health Perspect. 2013, 121, 572− 579. (23) Errico, R. What is an adjoint model? Bull. Am. Meteorol. Soc. 1997, 78, 2577−2591. (24) Hakami, A.; Henze, D. K.; Seinfeld, J. H.; Singh, K.; Sandu, A.; Kim, S.; Byun, D.; Li, Q. The adjoint of CMAQ. Environ. Sci. Technol. 2007, 41, 7807−7817. (25) Sandu, A.; Daescu, D. N.; Carmichael, G. R.; Chai, T. Adjoint sensitivity analysis of regional air quality models. J. Comput. Phys. 2005, 204, 222−252. (26) Wang, K. Y.; Lary, D.; Shallcross, D.; Hall, S.; Pyle, J. A review on the use of the adjoint method in four-dimensional atmosphericchemistry data assimilation. Q. J. R. Meteorol. Soc. 2001, 127, 2181− 2204. (27) Anenberg, S. C.; Horowitz, L. W.; Tong, D. Q.; West, J. J. An estimate of the global burden of anthropogenic ozone and fine particulate matter on premature human mortality using atmospheric modeling. Environ. Health Perspect. 2010, 118, 1189−1195. (28) Byrd, R.; Nocedal, J.; Waltz, R. KNITRO: An integrated package for nonlinear optimization. In Large-Scale Nonlinear Optimization; Di Pillo, G., Roma, M., Eds.; Proceedings of the Workshop on Large Scale Nonlinear Optimization held in Erice, Italy, June 22−July 1, 2004; Springer: New York, 2006; Vol. 83, pp 35−39. (29) Byun, D.; Schere, K. L. Review of the governing equations, computational algorithms, and other components of the Models-3 Community Multiscale Air Quality (CMAQ) modeling system. Appl. Mech. Rev. 2006, 59, 51−77. (30) Skamarock, W.; Klemp, J.; Dudhia, J.; Gill, D. O.; Barker, D. M.; Wang, W.; Powers, J. G. A description of the advanced research WRF version 2; NCAR Technical Note NCAR/TN-468+STR prepared for Mesoscale and Microscale Meteorology Division, National Center for Atmospheric Research: Boulder, CO, 2005. (31) Sparse Matrix Operator Kernel Emissions Modeling System (SMOKE) v2.4 User’s Manual; Carolina Environmental Program: Chapel Hill, NC, 2007. (32) United States Environmental Protection Agency. Clean air market, data and maps. http://ampd.epa.gov/ampd/. (33) United States Environmental Protection Agency. Chapter 5: Emission Control Technologies. In Documentation for EPA Base Case v.4.10 Using the Integrated Planning Model; EPA #430R10010; Clean Air Markets Division, Office of Air and Radiation, U.S. Environmental Protection Agency: Washington, DC, 2010. (34) United States Environmental Protection Agency. Guidelines for Preparing Economic Analyses; EPA #240-R-10-001; National Center for Environmental Economics, Office of Policy, U.S. Environmental Protection Agency: Washington, DC, 2010.

S.M.M.: Agriculture and Agri-Food Canada, Eastern Cereal and Oilseed 8 Research Centre, Ottawa, Ontario K1A 0C6, Canada. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS This research was funded by the Natural Sciences and Engineering Research Council of Canada.

■

REFERENCES

(1) Annual Energy Outlook 2010 with Projections to 2035; Technical Report DOE/EIA-0383(2010) prepared for the U.S. Energy Information Administration, Office of Integrated Analysis and Forecasting, U.S. Department of Energy: Washington, DC, 2010. (2) Committee on Tropospheric Ozone; Commission on Geosciences, Environment and Resources; Division on Earth and Life Studies; National Research Council Rethinking the Ozone Problem in Urban and Regional Air Pollution; National Academy Press: Washington, DC, 1991. (3) Mesbah, S. M.; Hakami, A.; Schott, S. Optimal ozone reduction policy design using adjoint-based NOx marginal damage information. Environ. Sci. Technol. 2013, 47, 13528−13535. (4) Muller, N. Z.; Mendelsohn, R. Efficient pollution regulation: getting the prices right. Am. Econ. Rev. 2009, 99, 1714−1739. (5) Mesbah, S. M.; Hakami, A.; Schott, S. Improving NOx cap-andtrade system with adjoint-based emission exchange rates. Environ. Sci. Technol. 2012, 46, 11905−11912. (6) Mauzerall, D.; Sultan, B.; Kim, N.; Bradford, D. NOx emissions from large point sources: variability in ozone production, resulting health damages and economic costs. Atmos. Environ. 2005, 39, 2851− 2866. (7) Nobel, C. E.; McDonald-Buller, E. C.; Kimura, Y.; Allen, D. T. Policy analysis accounting for spatial variation of ozone productivity in NOx emission trading. Environ. Sci. Technol. 2001, 35, 4397−4407. (8) Martin, K. C. Implementing a time-and location-differentiated capand-trade program: flexible nitrogen oxide abatement from power plant in the eastern united states. PhD Thesis, Department of Technology, Management, and Policy, Massachusetts Institute of Technology, MA, 2008. (9) Sun, L.; Webster, M.; Mcgaughey, G.; McDonald-Buller, E. C.; Thompson, T.; Prinn, R.; Ellerman, A. D.; Allen, D. T. Flexible NOx abatement from power plants in the eastern United States. Environ. Sci. Technol. 2012, 46, 5607−5615. (10) Alhajeri, N. S.; Donohoo, P.; Stillwell, A. S.; King, C. W.; Webster, M. D.; Webber, M. E.; Allen, D. T. Using market-based dispatching with environmental price signals to reduce emissions and water use at power plants in the Texas grid. Environ. Res. Lett. 2011, 6, 044018. (11) Thompson, T. M.; King, C. W.; Allen, D. T.; Webber, M. E. Air quality impacts of plug-in hybrid electric vehicles in Texas: evaluating three battery charging scenarios. Environ. Res. Lett. 2011, 6, 024004. (12) Pacsi, A. P.; Alhajeri, N. S.; Webster, M. D.; Webber, M. E.; Allen, D. T. Changing the spatial location of electricity generation to increase water availability in areas with drought: a feasibility study and quantification of air quality impacts in Texas. Environ. Res. Lett. 2013, 8, 035029. (13) Muller, N. Z.; Mendelsohn, R. Measuring the damages of air pollution in the United States. J. Environ. Econ. Manage. 2007, 54, 1− 14. (14) Førsund, F.; Nævdal, E. Efficiency gains under exchange-rate emission trading. Environ. Resour. Econ. 1998, 12, 403−423. (15) Montgomery, W. D. Market in licences and efficient pollution control program. J. Econ. Theory 1972, 5, 395−418. (16) Tietenberg, T. Tradeable permits for pollution control when emission location matters: What have we learned? Environ. Resour. Econ. 1995, 5, 95−113. 7877


Article

Environmental Science & Technology (35) Zanobetti, A.; Schwartz, J. Mortality displacement in the association of ozone with mortality: an analysis of 48 cities in the United States. Am. J. Respir. Crit. Care Med. 2008, 177, 184−189. (36) Bell, M. L.; Mcdermott, A.; Zeger, S. L.; Samet, J. M. Ozone and short-term mortality in 95 US urban communities, 1987−2000. JAMA, J. Am. Med. Assoc. 2004, 292, 2372−2378. (37) Duncan, B. N.; Prados, A. I.; Lamsal, L. N.; Liu, Y.; Streets, D. G.; Gupta, P.; Hilsenrath, E.; Kahn, R. A.; Nielsen, J. E.; Beyersdorf, A. J.; et al. Satellite data of atmospheric pollution for U.S. air quality applications: Examples of applications, summary of data end-user resources, answers to FAQs, and common mistakes to avoid. Atmos. Environ. 2014, 94, 647−662. (38) Muller, N. Z. Linking policy to statistical uncertainty in air pollution damages. B.E. J. Econ. Anal. Policy Contrib. 2011, 11, DOI: 10.2202/1935-1682.2925. (39) He, H.; Stehr, J. W.; Hains, J. C.; Krask, D. J.; Doddridge, B. G.; Vinnikov, K. Y.; Canty, T. P.; Hosley, K. M.; Salawitch, R. J.; Worden, H. M.; et al. Trends in emissions and concentrations of air pollutants in the lower troposphere in the Baltimore/Washington airshed from 1997 to 2011. Atmos. Chem. Phys. Discuss. 2013, 13, 7859−7874. (40) He, H.; Hembeck, L.; Hosley, K. M.; Canty, T. P.; Salawitch, R. J.; Dickerson, R. R. High ozone concentrations on hot days: The role of electric power demand and NOx emissions. Geophys. Res. Lett. 2013, 40, 5291−5294. (41) Thompson, T. M.; Selin, N. E. Influence of air quality model resolution on uncertainty associated with health impacts. Atmos. Chem. Phys. 2012, 12, 9753−9762. (42) Punger, E. M.; West, J. J. The effect of grid resolution on estimates of the burden of ozone and fine particulate matter on premature mortality in the United States. Air Qual., Atmos. Health 2012, 29, 997−1003. (43) Hakami, A.; Odman, M. T.; Russell, A. G. Nonlinearity in atmospheric response: A direct sensitivity analysis approach. J. Geophys. Res. 2004, 109, 1−12. (44) Alberini, A.; Filippini, M. Response of residential electricity demand to price: The effect of measurement error. Energy Econ. 2011, 33, 889−895. (45) EME Homer City Generation L.P. v Environmental Protection Agency, et al., Respondents San Miguel Electric Cooperative, et al., Intervenors; case no. 11-1302 (US Court of Appeals for the District of Columbia Circuit, decided August 21, 2012).http://www.cadc. uscourts.gov/internet/opinions.nsf/ 19346B280C78405C85257A61004DC0E5/$file/11-1302-1390314. pdf.

7878


Optimal Ozone Control with Inclusion of Spatiotemporal Marginal

Recommend Documents