Response of Atmospheric Particulate Matter to ... - ACS Publications

Dec 21, 2007 - Atmospheric and Environmental Research, Inc., San Ramon, CA;. Tennessee Valley Authority, Muscle Shoals, AL; ICF International,...
0 downloads 0 Views 2MB Size
Download PDF
Environ. Sci. Technol. 2008, 42, 831–837

Response of Atmospheric Particulate Matter to Changes in Precursor Emissions: A Comparison of Three Air Quality Models B E T T Y K . P U N , * ,† CHRISTIAN SEIGNEUR,† ELIZABETH M. BAILEY,‡ LARRY L. GAUTNEY,‡ SHARON G. DOUGLAS,§ JAY L. HANEY,§ AND NARESH KUMAR⊥ Atmospheric and Environmental Research, Inc., San Ramon, CA; Tennessee Valley Authority, Muscle Shoals, AL; ICF International, San Rafael, CA; EPRI, Palo Alto, CA

Received September 17, 2007. Revised manuscript received November 2, 2007. Accepted November 9, 2007.

Three mathematical models of air quality (CMAQ, CMAQMADRID, and REMSAD) are applied to simulate the response of atmospheric fine particulate matter (PM2.5) concentrations to reductions in the emissions of gaseous precursors for a 10 day period of the July 1999 Southern Oxidants Study (SOS) in Nashville. The models are shown to predict similar directions of the changes in PM2.5 mass and component (sulfate, nitrate, ammonium, and organic compounds) concentrations in response to changes in emissions of sulfur dioxide (SO2), nitrogen oxides (NOx), and volatile organic compounds (VOC), except for the effect of SO2 reduction on nitrate and the effect of VOC reduction on PM2.5 mass. Furthermore, in many caseswherethedirectionalchangesareconsistent,themagnitude of the changes are significantly different among models. Examples are the effects of SO2 and NOx reductions on nitrate and PM2.5 mass and the effects of VOC reduction on organic compounds, sulfate and nitrate. The spatial resolution significantly influences the results in some cases. Operational model performance for a PM2.5 component appears to provide some useful indication on the reliability of the relative response factors (RRFs) for a change in emissions of a direct precursor, as well as for a change in emissions of a compound that affects this component in an indirect manner, such as via oxidant formation. However, these results need to be confirmed for other conditions and caution is still needed when applying air quality models for the design of emission control strategies. It is advisable to use more than one air quality model (or more than one configuration of a single air quality model) to span the full range of plausible scientific representations of atmospheric processes when investigating future air quality scenarios.

* Corresponding author phone: (925) 244-7125; fax: (925) 2447129; e-mail: [email protected]. † Atmospheric and Environmental Research. ‡ Tennessee Valley Authority. § ICF International. ⊥ EPRI. 10.1021/es702333d CCC: $40.75

Published on Web 12/21/2007

 2008 American Chemical Society

Introduction Mathematical models of air quality are routinely used to investigate future air quality by taking into account changes in pollutant emissions due to the growth and application of emission control technology. For example, the application of air quality models is required by the U.S. Environmental Protection Agency (EPA) to design effective emission control strategies to attain or make progress toward attainment of the National Ambient Air Quality Standards (NAAQS). In the case of particulate matter (PM), air quality models are used to develop emission reduction strategies to attain the annual and 24 h average NAAQS as well as to demonstrate progress toward improved atmospheric visibility in national parks and wilderness areas (1). Prior to their use to investigate the effects of emission control scenarios on air quality, models must be evaluated against ambient measurements of PM concentrations. However, satisfactory model performance does not guarantee that the models can correctly predict the effects of changes in pollutant emissions on PM concentrations because the ability of a mathematical model to calculate a state variable does not necessarily imply the ability to predict its derivatives with respect to input variables. Air quality is governed by a complex system of atmospheric processes, and the response of the concentrations of PM and its components to changes in precursor concentrations (or emissions) is nonlinear and, in some cases, “antagonistic” (2). A response is nonlinear when the relative change in PM mass or component concentration is different from the relative change in the precursor level; it is antagonistic when a reduction in a PM precursor leads to an increase in PM mass or component(s) concentrations. For example, a decrease in sulfur dioxide (SO2) emissions will lead to a decrease in sulfate concentration; however, the sulfate decrease may not be proportional to the SO2 decrease due to the nonlinear nature of the SO2/sulfate aqueous-phase chemistry (3). Furthermore, the decrease in sulfate concentration will typically be associated with an increase in gaseous ammonia concentration, which can lead to some increase in ammonium nitrate concentration. In some cases, the increase in ammonium nitrate may exceed the decrease in ammonium sulfate (4). A decrease in nitrogen oxides (NOx) emissions is generally associated with a decrease in PM nitrate concentration (5). However, it is possible in some cases where oxidant formation is sensitive to volatile organic compounds (VOC) that PM nitrate concentrations will actually increase because the increase in oxidant concentrations exceeds the decrease in nitrogen dioxide (NO2) concentrations (6, 7). A reduction in VOC emissions similarly leads to a reduction in PM organic compound concentrations but may lead to increases in sulfate and nitrate concentrations due to increases in oxidant concentrations and/or decreases in the formation of gaseous organic nitrates (8, 9). Oxidants such as O3 and OH are involved in the formation of a myriad of condensable products. Therefore, VOC and NOx, which are precursors to many oxidants, can potentially affect all secondary PM species. Different VOC and NOx chemical regimes lead to different amounts of volatile products relative to condensable products (e.g., organic nitrates vs HNO3), and can lead to linear, nonlinear, or even antagonistic responses. To date, the complex nature of the PM/precursor relationships has been reported for several specific areas of the United States based on results obtained from various air quality models. Recently, Thunis et al. (10) compared the response of PM10 to reductions in NOx and VOC emissions VOL. 42, NO. 3, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

831

over Europe among several air quality models. However, the response of PM concentrations to precursor emissions over the entire contiguous U.S. has not been systematically investigated. It is nearly impossible to evaluate the performance of air quality models for predicting the changes in air quality because this evaluation would require, in theory, ambient measurements of PM mass and component concentrations for a range of distinct emission levels under the same meteorological conditions. Nevertheless, it is possible to investigate whether different air quality models lead to similar or distinct results in terms of the response of air quality to changes in pollutant emissions. Such an analysis provides some useful indication of the possible range of PM responses to changes in precursor emissions due to various model formulations (i.e., uncertainties in the representation of atmospheric processes). Here, we compare the response of PM2.5 and PM2.5 chemical components (sulfate, nitrate, ammonium, and organic compounds) to changes in precursor emissions simulated using three air quality models. We discuss the similarities and discrepancies in the directions of the responses (increase, negligible change, or decrease) and the magnitude of the responses (relative change in a PM2.5 component due to a change in precursor emissions).

Materials and Methods The three air quality models included in this study are the EPA public release version of the Community Multiscale Air Quality (CMAQ) model (12), a version of CMAQ with the Model of Aerosol Dynamics, Reaction, Ionization, and Dissolution (MADRID) (13–15), and the Regional Modeling System for Aerosols and Deposition (REMSAD, ref 16). CMAQ version 4.2.2 (released in May 2003), CMAQ-MADRID (based on CMAQ version 4.2.1 with MADRID 1), and REMSAD version 7 represent a plausible range of scientific representations of atmospheric processes (e.g., different representations of the particle size distribution, different gas-phase and heterogeneous chemistry) relevant for the simulation of PM2.5 concentrations, as available in 2003. Although recent advances in our understanding of atmospheric processes-most notably, secondary organic aerosol (SOA) formation-have since then led to improvements in some specific aspects of the model formulations, the three models used here present a range of model formulations that is still quite representative of the current uncertainties associated with our understanding of atmospheric processes. The models are applied to simulate air quality over the continental United States (32 km horizontal resolution) and, with a finer (8 km) spatial resolution, over the southeastern U.S. (see Supporting Information (SI) Figure SI-1) for a twelveday episode, June 29 to July 10, 1999 (the first two-days are used for model spin-up and not for analysis). The same sets of meteorological conditions, boundary and initial conditions, and emissions inventory data are used as inputs to all models. Further details on the models and processing of input data are available elsewhere (ref 11 and references therein). Bailey et al. (11) present the results of the base case simulations and a comparative evaluation of the model performance for gaseous and particulate air pollutants. None of the models clearly emerges with the best overall performance for both particulate and gaseous species. All of the models tend to perform better for sulfate than for the other species. Differences in performance among the three models for sulfate, nitrate, ammonium, and EC are not great. Sulfate and nitrate are underpredicted by all models, EC is overpredicted by all models, ammonium is underpredicted by the CMAQ-based models but overpredicted by REMSAD. For the 8 km resolution simulations, the ranges of model performance in terms of the mean normalized error (MNE) 832

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 3, 2008

are as follows: 35–47% for sulfate, 85–92% for nitrate, 49–53% for ammonium, and 62–67% for EC. The three models show significant differences in the performance of PM2.5 OC (MNE of 50–75%, mean normalized bias of -75 to 37%). Although all models underestimate PM2.5, the model that overestimate OC (CMAQ) performs better for total PM2.5 than the models that underestimate OC (CMAQ-MADRID and REMSAD), indicating the likely presence of compensating factors. Model performance statistics for PM and species are compiled in Table 6 in Bailey et al. (11). Building on the base case simulations (11), we present here a comparison of the response of PM chemical components to changes in precursor emissions. The emission changes investigated here are 50% reductions in SO2, NOx, and VOC. For VOC, both anthropogenic and biogenic emissions are reduced. This approach results in the maximum response to VOC and allows the differences in model responses to be studied effectively; however, the results cannot be used to directly inform control strategy selection for anthropogenic sources. CMAQ and REMSAD are applied to both the continental domain and the southeastern U.S. domain. CMAQ-MADRID is applied only to the continental domain. A one-way nesting approach (i.e., the continental 32-km horizontal resolution simulation provides the boundary conditions for the southeastern U.S. 8-km resolution simulation, but the southeastern U.S. simulation does not affect the continental simulation) is used for CMAQ, while a two-way nesting approach (i.e., there is simultaneous two-way interaction between the southeastern U.S. domain and the continental domain) is used for REMSAD. This study represents the most comprehensive investigation to date of the response of PM concentrations to precursor emissions over the entire contiguous U.S. using several air quality models with the same set of atmospheric conditions.

Results Spatial Maps. We present the changes in the ground-level concentrations of PM2.5 sulfate, nitrate and organic compounds due to changes in SO2, NOx and VOC. A graphical format of presentation is used to illustrate the relationships between PM2.5 components and the major gaseous precursors. The results are presented with a detailed discussion for the CMAQ model, followed by comparative discussions for the other two models. Figure 1 presents the differences in CMAQ-simulated PM2.5 sulfate, nitrate, and organic compounds due to 50% reductions in SO2, NOx, and VOC emissions over the continental U.S. Over the entire domain, reduction of a precursor of secondary PM leads to a reduction in the corresponding PM component, i.e., reductions in SO2, NOx, and VOC led to reductions in sulfate, nitrate, and organic compounds, respectively. The antagonistic response of PM nitrate to NOx reduction that was simulated for some locations in the western U.S. (California Central Valley, Idaho) with box models (6, 7) is not present in these results. It is likely that such antagonistic responses, although theoretically possible, are limited in space and time. They were simulated for winter conditions (when PM nitrate concentrations are greater and oxidant concentrations are lower) and this set of simulations is for summer conditions. The chemical reactions important for the formation of nitrate, and hence the sensitivity of nitrate to precursors, differ under conditions of lower temperature and higher liquid water contents (e.g., fog) from those prevalent in summer. CMAQ predicts a reduction of SO2 emissions to be effective toward reducing PM sulfate throughout the continental U.S. The reduction in SO2 emissions leads to some decreases in PM nitrate concentrations in the Midwest but to increases in California and the eastern U.S. The decreases result from

FIGURE 1. Changes (µg/m3) simulated with CMAQ on July 7, 1999 over the continental United States of the atmospheric ground-level concentrations of PM2.5 sulfate (top row), nitrate (middle row), and organic compounds (bottom row) due to 50% reductions in SO2 (left column), NOx (middle column), and VOC (right column) emissions. a decrease in the PM liquid water content, which leads to less dissolution of HNO3 in the particles. Changes in particle size can also affect the heterogeneous reaction of N2O5. The increases result from the increased availability of gaseous NH3 to react with HNO3 to form particulate ammonium nitrate, as the concentrations of ammonium sulfate decrease (17). The nitrate response to changes in SO2 must be taken into account as it can potentially diminish the effectiveness of SO2 emission reduction on PM2.5 mass and in some cases can lead to increases in PM2.5 concentrations (see below). The effect of a reduction in SO2 emissions on PM organic compounds is negligible in CMAQ. The reduction in NOx emissions leads to reduced PM nitrate formation in parts of the Midwest and southern U.S. Because base case nitrate concentrations are fairly low during this summer episode, the absolute magnitude of nitrate reduction is limited. In addition, the reduction in NOx emissions decreases sulfate concentrations over the southern and eastern U.S., but leads to some isolated increases along the Gulf coast and in the Midwest. The influence occurs via the concentrations of oxidants. A decrease in NOx emissions may lead to a decrease in oxidant concentrations in areas that are NOx limited (i.e., most nonurban areas and some urban areas) but may lead to an increase in oxidant concentrations in areas where O3 formation is VOC sensitive rather than NOx sensitive (i.e., some urban areas) (18, 19). Effects on organic compounds show some similarities compared to those on sulfate. A reduction in NOx emissions leads to decreases in organic PM concentrations over most of the domain, due to reduction of oxidants over nonurban areas, which are NOx sensitive. Increases of organic compounds are simulated in some urban areas such as the Los Angeles basin, CA; Phoenix, AZ; Chicago, IL; Houston, TX; Baton Rouge, LA; and New York, NY. This result is consistent with those urban areas being typically VOC-sensitive for O3

formation (20–22), where a reduction on NOx emissions may increase oxidant concentrations, resulting in increased formation of PM organic compounds. Although CMAQ predicts VOC reductions to be effective toward lowering PM OC over much of the U.S., the reduction in VOC emissions leads to increases in sulfate concentrations over the southern and eastern U.S. Because reactive biogenic and anthropogenic VOC act as local sinks of oxidants, reductions in all VOC emissions, as implemented here, lead to increased OH concentrations over much of the domain. Increased OH concentrations accelerate the oxidation of SO2 to sulfate. In response to reduced VOC emissions, increases in PM nitrate concentrations are simulated in California, the South, and the eastern U.S. as greater OH concentrations also increase the kinetics of NO2 oxidation to HNO3. An additional effect of VOC reduction on nitrate occurs due to reduced formation of organic nitrates and greater availability of gaseous nitrogen compounds for the formation of inorganic nitrate. Patterns obtained with the 8 km resolution southeastern U.S. domain are similar to those obtained with the 32 km resolution continental domain (see SI Figure SI-2). However, we discern some differences that result from the use of a different spatial resolution (e.g., nitrate response to SO2 and VOC emission reductions and OC response to NOx emission reduction). The grid size may affect the local chemical regime (e.g., NOx-sensitive versus VOC-sensitive for the formation of O3 and other oxidants) and, consequently, the formation of PM. Differences in the MM5 meteorological fields at 32 and 8 km resolution (23) can also be a contributing factor for different predictions in both the base and sensitivity cases. Figure 2 shows a comparison of the changes in PM2.5 concentrations due to 50% reductions in SO2, NOx and VOC emissions among CMAQ, CMAQ-MADRID, and REMSAD over the continental U.S. (32 km horizontal resolution). VOL. 42, NO. 3, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

833

FIGURE 2. Changes (µg/m3) simulated with CMAQ (top row), CMAQ-MADRID (middle row), and REMSAD (bottom row) on July 7, 1999 over the continental United States of the atmospheric ground-level concentrations of PM2.5 due to 50% reductions in SO2 (left column), NOx (middle column), and VOC (right column) emissions. The three models show similar results for the SO2 emission reduction with PM2.5 concentrations being reduced over most of the eastern half of the U.S. and larger reductions occurring in the southeastern states. PM sulfate reductions as a response to reductions in SO2 emissions govern the PM2.5 response in all three models. A notable difference is the prediction of some PM2.5 concentration increases by CMAQ-MADRID in some areas such as northern Texas. These increases are due to the fact that CMAQ-MADRID simulates larger increases in nitrate and some increases in organic PM when SO2 emissions are reduced (not shown) compared to CMAQ and REMSAD. CMAQ-MADRID includes the treatment of heterogeneous reactions at the surface of particles. As sulfate concentrations decrease, the surface area of particles available for the heterogeneous reactions decreases and the reaction of HO2 radicals to form H2O2 decreases. As a result, HO2 and OH concentrations increase, which leads to increases in nitrate and SOA formation. This result highlights the fact that different model formulations can lead, in some cases, to slight differences in the response of PM2.5 precursor reductions. The 50% NOx emission sensitivity simulations lead to lower PM2.5 concentrations over the eastern U.S. and, to a lesser extent, over the western U.S. The decrease in PM2.5 is driven to a large extent by decreases in nitrate and to a smaller extent by decreases in sulfate. The decrease in PM2.5 is more widespread in the CMAQ-MADRID simulation than in CMAQ or REMSAD. CMAQ, CMAQ-MADRID and REMSAD show isolated increases in PM2.5 concentrations in some locations such as Chicago, IL (CMAQ and CMAQ-MADRID), Houston, TX (CMAQ and REMSAD), Gulfport, MS (CMAQ and CMAQMADRID) and Tampa, FL (CMAQ). Increased concentrations of sulfate and organic PM (see Figure 1) are due to increases in oxidant concentrations in areas that are VOC-sensitive for O3 formation. 834

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 3, 2008

The 50% VOC emission sensitivity simulations show quite different results among the three models. CMAQ predicts decreased PM2.5 concentrations over the entire domain with the largest decreases occurring in the southeastern U.S. CMAQ-MADRID predicts decreased PM2.5 concentrations in the southeastern U.S. (where CMAQ predicts the largest PM2.5 decreases), but predicts some increases in PM2.5 concentrations in some areas of the South (e.g., northeastern Texas, Arkansas, Missouri, and Virginia). REMSAD predicts mostly increases in PM2.5 concentrations with decreases being limited mostly to some areas in the southeastern U.S. (e.g., South Carolina and Alabama). As discussed above, reductions in VOC emissions lead to reduced organic PM but increased sulfate and nitrate. Therefore, the net result on PM2.5 concentrations depends on the relative changes in these various components. In the case of CMAQ, the decrease in organic PM prevails. CMAQ-MADRID shows some areas where the decrease in organic PM dominates and some areas where the increase in inorganic PM dominates. For REMSAD, which tends to underpredict PM OC in the base case (11), the simulated decrease in OC is small and the increase in inorganic PM dominates. These results show that it is inherently difficult to predict correctly the response of PM2.5 concentrations due to changes in VOC emissions because the responses of organic and inorganic PM components can be opposite (decrease and increase, respectively). The relative magnitudes of the predicted changes depend on the model formulation. The representation of VOC reactivities in REMSAD’s micro-CB-IV mechanism vs the RADM2 mechanism used in the CMAQ-based models most likely contribute to the difference in the predicted response as well as the difference in base case performance for PM OC (11). Relative Response Factors. EPA recommends the use of relative response factors (RRFs) to estimate the amount of emission reduction needed to attain the PM2.5 NAAQS (1).

The RRFs calculated here are the ratios of the concentration in the emission reduction simulation to the concentration in the base simulation. RRFs were calculated for each of the emission reduction scenarios for particulate sulfate, particulate nitrate, particulate ammonium, particulate organic compounds, and PM2.5 concentrations at six locations. These locations correspond to measurement sites where model performance was evaluated for the base simulations (11): two SEARCH sites (Jefferson Street in Atlanta and Yorkville), two Nashville-area SOS sites (Hendersonville and Dickson) and two IMPROVE sites (Great Smoky Mountains National Park and Sipsey Wilderness Area). Since these six sites are located within the fine spatial resolution domain of the southeastern U.S. (see SI Figure SI-1), RRFs are calculated (see SI Table SI-1) for the 1–10 July 1999 period for each of the sensitivity simulations and for CMAQ with both 32 km and 8 km resolutions, CMAQ-MADRID with 32 km resolution, and REMSAD with 8 km resolution. Of the three emission scenarios, the RRFs for sulfate are lowest for the SO2 emission reduction case, as expected. CMAQ, CMAQ-MADRID, and REMSAD predict similar RRFs (within 10%) in the range of 0.57-0.66 (with REMSAD predicting the lowest RRF). In all cases, a less-thanproportional reduction is predicted, because the SO2/sulfate chemistry is sensitive to the concentration of SO2 as well as oxidants such as H2O2 and O3 (3). For the NOx emission reduction case, CMAQ predicts the lowest RRFs in all cases but one, and REMSAD predicts the largest RRFs in all cases. At any given site, RRFs of all four model simulations are within 10% of each other. For the VOC emission reduction case, all RRFs are greater than one (except for the REMSAD RRF at Sipsey), indicating increases in PM2.5 sulfate when VOC emissions are reduced. REMSAD predicts weaker responses (RRFs closer to 1) than the other models. CMAQMADRID predicts the largest increases at Yorkville, Great Smoky Mountains, and Sipsey; but CMAQ (8 km resolution) predicts the largest increases at Atlanta, Dickson, and Hendersonville. Overall, differences due to the spatial resolution of CMAQ are small (0.02 or less in 14 cases out of 18); they are greater for the VOC emission reduction case than the SO2 and NOx cases. For the VOC case, the three Tennessee sites show significantly greater RRFs for the fine spatial resolution. Due in part to low and variable simulated nitrate concentrations for this summer episode, nitrate RRFs show significant spatial variability compared to sulfate RRFs. The RRFs for nitrate show significant differences among the models for the SO2 emission reduction case. For example, REMSAD and CMAQ with 8 km resolution predict nitrate decreases at Dickson and Hendersonville whereas CMAQMADRID predicts increases. CMAQ with 32 km resolution also predicts an increase at Dickson, but not at Henderson. REMSAD predicts decreases at Yorkville and Sipsey whereas all the other models predict increases. All models predict decreases in nitrate for the NOx reduction case; however, the RRFs differ widely. REMSAD typically predicts the lowest RRF (range of 0.20-0.32) and CMAQ-MADRID predicts the largest RRF (range of 0.45-0.73). All models predict PM2.5 nitrate to increase when VOC emissions are reduced at these six locations. In terms of the magnitudes of the RRFs, the models are in better agreement for the VOC emission reduction case than the NOx reduction case, although REMSAD predicts some significantly greater increases at Yorkville and Sipsey (RRFs > 2) that are not simulated in the other models. The RRFs for ammonium are related to those obtained for sulfate and nitrate since the ammonium is associated with sulfate and nitrate in PM. The RRFs are in better agreement among the models than those for sulfate and nitrate because of some averaging effects. For the SO2

emission reduction case, the change in sulfate dominates but is compensated to some extent by the change in nitrate (lesser decrease than sulfate); therefore, in most cases, the ammonium RRF is greater than the sulfate RRF (both are less than 1). For the NOx and VOC emission reduction cases, the ammonium RRFs are similar to those obtained for sulfate. They can be slightly greater or smaller than the sulfate RRFs depending on the importance of the change in nitrate. The RRFs for organic compounds show negligible change for all three models for the SO2 emission reduction case. All models predict similar small decreases (RRFs > 0.9) in organic PM for the NOx emission reduction case. The only exception is REMSAD, predicting an RRF of 0.81 at Great Smoky Mountains. At most sites, CMAQ-MADRID shows a greater response of organic compounds to the change in NOx emissions than CMAQ or REMSAD. All models predict significant decreases for the VOC emission reduction case with CMAQ and CMAQ-MADRID typically predicting the largest decreases (RRFs ranging from 0.53 to 0.62 for CMAQ and from 0.52 to 0.67 for CMAQ-MADRID) and REMSAD predicting the smallest decreases at four sites out of six (RRFs ranging from 0.56 to 0.83). The RRFs for PM2.5 reflect the results obtained for the individual components. For the SO2 emission reduction case, all RRFs are less than one because the sulfate RRFs dominate the PM2.5 RRF values. REMSAD predicts the smallest RRFs whereas CMAQ with 8 km resolution predicts the largest RRFs (closest to 1). For the NOx emission reduction case, the two models with 8 km resolution, i.e., CMAQ and REMSAD, are in good agreement at all sites (RRFs within 0.01 or 0.02). CMAQ-MADRID always predicts the lowest RRF, implying a greater response to NOx reduction than the other models. For the VOC emission reduction case, PM2.5 RRFs can be greater than one in a few cases. CMAQ predicts the lowest RRFs and REMSAD typically predicts the smallest decrease or the largest increase in PM2.5. CMAQ-MADRID is closer to REMSAD than to CMAQ. Overall, the spatial resolution of CMAQ has little effect on RRFs. The PM2.5 RRFs as calculated using different resolutions in CMAQ are within 0.02 in all cases but three (within 0.03 for SO2 reduction at Hendersonville, within 0.04 for SO2 at Sipsey, and within 0.05 for VOC reduction at Sipsey).

Model Performance and Relative Response Factors The comparison of modeled concentrations with observations provides some indication on the ability of the model to simulate atmospheric processes (operational evaluation) but a dynamic evaluation is needed to provide information on the ability of the model to reproduce the response of concentrations to changes in emissions. Dynamic evaluations are, however, typically not feasible because of a lack of observations associated with the emission sensitivity scenario. In this case, the use of the predicted changes from a base case to an emission sensitivity case (e.g., via the use of RRFs) involves some implicit assumptions. First, there is the assumption of a correlation between the model’s ability to simulate the base case and the sensitivity case. Under this assumption, a model that performs better for the base case than the other models will most likely also have better predictions for the sensitivity case. Second, despite model performance, the response is subject to less uncertainty than the base or sensitivity simulations, because biases in both the base and sensitivity scenarios will cancel to some extent. It is of interest to investigate how the uncertainty associated with RRFs varies as a function of operational model performance. A range of RRFs due to different model formulations is used as a surrogate for the uncertainty in RRFs. Although good agreement among the models is not necessarily a guarantee that all models would predict the VOL. 42, NO. 3, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

835

predicted response needs to be confirmed for other conditions and caution is still needed when applying an air quality model for the design of emission control strategies for PM reduction. It is advisable to apply more than one air quality model to improve the confidence in both the base case and sensitivity results of the model simulations. In addition to different air quality model representations, various emission and meteorology models can also form an ensemble that provides information on the range of performances and possible responses. FIGURE 3. Uncertainties in RRFs of key response variables to changes in precursor emissions. Each individual vertical bar represents a range of RRFs predicted by the different models at the following sites: Atlanta, GA; Yorkville, GA; Dickson, TN; Hendersonville, TN; Great Smoky Mountains National Park, NP; and Sipsey Wilderness, AL in that order from left to right. Each horizontal bar represents the range of the mean normalized errors obtained by the different models for the PM2.5 component. response correctly, large discrepancies among the models will necessarily imply incorrect responses predicted by at least some of the models. Figure 3 presents the range of RRFs predicted by the models for the response of a PM2.5 component to a change in the emissions of the corresponding precursor (i.e., SO2 for sulfate, NOx for nitrate, and VOC for OC), as a function of model operational performance for the PM2.5 component. The RRFs are presented for the six locations mentioned above, and model performance is presented as a range for the four model simulations. Model performance is best for sulfate, then OC, and worst for nitrate; there is no overlap in model performance. There is a clear relationship between model operational performance and the range of RRFs. RRFs for sulfate are very consistent at all six sites (implying low uncertainty in the sulfate response, less than 10%) among the four model simulations whereas the RRFs for nitrate show large uncertainties (more than 100% of the lowest RRF). The RRF uncertainties for OC fall in the middle. Figure 3 also presents similar information for the range of RRFs predicted by the models for the response of a PM2.5 component to a change in the emissions of a compound that is not the direct precursor, here the response of sulfate, nitrate, and ammonium to VOC emissions. (VOC emissions were selected because they led to the more sensitive response across the PM2.5 components, see SI Table SI-1.) Despite generally larger uncertainties in the indirect responses than in the direct responses, it appears that there is also a relationship between model operational performance and the RRF uncertainties. Large uncertainties are seen for the nitrate RRFs, whereas lower uncertainties are seen for the sulfate and ammonium RRFs, which is consistent with better model performance for sulfate and ammonium than for nitrate. These results show that even when models produce largely similar performance in base case simulations, significantly different results may be obtained in terms of the sensitivities of PM components to changes in precursor emissions depending on the model formulation and, in some cases, the model spatial resolution. However, it appears that operational model performance for a PM2.5 component may provide some useful indication on the reliability of the RRF for a change in emissions of a direct precursor, as well as for a change in emissions of a compound that affects this component in an indirect manner, such as via oxidant formation. For a given model, concentrations and sensitivities are functions of meteorological regimes (including seasons), chemical regimes, and emission characteristics. The relationship between model performance and reliability of the 836

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 3, 2008

Acknowledgments Funding for this work was provided by the Electric Power Research Institute (EPRI), the Tennessee Valley Authority, the Midwest Ozone Group and Southern Company.

Supporting Information Available Maps of modeling domains, CMAQ PM concentration changes over the southeastern United States and Table of RRFs. This material is available free of charge via the Internet at http://pubs.acs.org.

Literature Cited (1) EPA, 2006. Guidance on the Use of Models and Other Analyses for Demonstrating Attainment of Air Quality Goals for Ozone, PM2.5, and Regional Haze. http://www.epa.gov/scram001/ guidance_sip.htm; accessed December 2006. (2) NARSTO, 2004. Particulate Matter Science for Policy Makers: A NARSTO Assessment, P.H. McMurry, M. Shepherd and J. Vickery, eds., Cambridge University Press: Cambridge, United Kingdom. (3) Seigneur, C.; Saxena, P.; Roth, P. M. Computer simulations of the atmospheric chemistry of sulfate and nitrate formation. Science 1984, 225, 1028–1030. (4) Ansari, A.; Pandis, S. N. On the response of atmospheric particulate matter concentrations to precursor concentrations. Environ. Sci. Technol. 1998, 32, 2706–2714. (5) Stockwell, W. R.; Watson, J. G.; Robinson, N. F.; Steiner, W.; Sylte, W. W. The ammonium nitrate particle equivalent of NOx emissions for wintertime conditions in Central CA’s San Joaquin Valley. Atmos. Environ. 2000, 34, 4711–4717. (6) Pun, B.; Seigneur, C. Sensitivity of particulate matter nitrate formation to precursor emissions in the CA San Joaquin Valley. Environ. Sci. Technol. 2001, 35, 2979–2987. (7) Stockwell, W. R.; Kuhns, H.; Etyemezian, V.; Green, M. C.; Chow, J. C. The Treasure Valley secondary aerosol study II: modeling of the formation of inorganic secondary aerosols and precursors for southwestern ID. Atmos. Environ. 2003, 37, 525–534. (8) Meng, Z.; Dabdub, D.; Seinfeld, J. H. Chemical coupling between atmospheric ozone and particulate matter. Science 1997, 277, 116–119. (9) Pai, P.; Vijayaraghavan, K.; Seigneur, C. Particulate matter modeling in the Los Angeles Basin using SAQM-AERO. J. Air Waste Manage. Assoc. 2000, 50, 32–42. (10) Thunis, P.; Rouil, L.; Cuvelier, C.; Stern, R.; Kerschbaumer, A.; et al. Analysis of model responses to emission-reduction scenarios with the CityDelta project. Atmos. Environ. 2007, 41, 208–220. (11) Bailey, E. M.; Gautney, L. L.; Kelsoe, J. J.; Jacobs, M. E.; Mao, Q.; Condrey, J. W.; Pun, B.; Wu, S.-Y.; Seigneur, C.; Douglas, S.; Haney, J.; Kumar, N. A comparison of the performance of four air quality models for the Southern Oxidants Study episode in July 1999, J. Geophys. Res., 2007, 112, D05306, doi: 10.1029/ 2005JD007021. (12) 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. Rev. Appl. Mech. 2006, 59, 51–77. (13) Zhang, Y.; Pun, B.; Vijayaraghavan, K.; Wu, S.-Y.; Seigneur, C.; Pandis, S. N.; Jacobson, M. Z.; Nenes, A.; Seinfeld, J. H. Development and application of the model of aerosol dynamics, reaction, ionization and dissolution, J. Geophys. Res., 2004, 109, D01202, doi: 10.1029/2003JD003501. (14) Pun, B. K.; Griffin, R. J.; Seigneur, C.; Seinfeld, J. H. Secondary organic aerosol: II. Thermodynamic model for gas/particle partitioning of molecular constituents. J. Geophys. Res. 2002, 107, 4333–4347. (15) Pun, B.; Wu, S.-Y.; Seigneur, C.; Seinfeld, J. H.; Griffin, R. J.; Pandis, S. N. Uncertainties in modeling secondary organic

(16) (17)

(18) (19)

aerosols: three-dimensional modeling studies in Nashville/ western TN. Environ. Sci. Technol. 2003, 37, 3647–3661. ICF. User’s Guide to the Regional Modeling System for Aerosols and Deposition (REMSAD), Version 7. (03–077); ICF Consulting/ SAI: San Rafael, CA, 2003. Vayenas, D. V.; Takahama, S.; Davidson, C. I.; Pandis, S. N. Simulation of the thermodynamics and removal processes in the sulfate-ammonia-nitric acid system during winter: Implications for PM2.5 control strategies, J. Geophys. Res., 2005, 110, D07S14, doi: 10.1029/2004JD005038. Blanchard, C. L.; Fairley, D. Spatial mapping of VOC and NOxlimitation of ozone formation in central CA. Atmos. Environ. 2001, 35, 3861–3873. Zhang, Y.; Vijayaraghavan, K.; Seigneur, C. Evaluation of three probing techniques in a three-dimensional air quality model. J. Geophys. Res., 2005, 110, D02305, doi: 10.1029/ 2004JD005248.

(20) Kleinman, L. I.; Daum, P. H.; Imre, D. G.; Lee, J. H.; Lee, Y.-N. Ozone production in the New York City urban plume. J. Geophys. Res. 2000, 105, 14495–14511. (21) Kleinman, L. I.; Daum, P. H.; Lee, Y.-N.; Nunnermacker, L. J.; Springston, S. R. Sensitivity of ozone production rate to ozone precursors. Geophys. Res. Lett. 2001, 28, 2903–2906. (22) Pun, B. K.; Seigneur, C.; White, W. Day-of-the-week behavior of atmospheric ozone in three United States cities. J. Air Waste Manage. Assoc. 2003, 53, 789–801. (23) Zhang, Y.; Liu, P.; Pun, B.; Seigneur, C. A Comprehensive performance evaluation of MM5-CMAQ for the summer 1999 Southern Oxidants Study episode—Part I: evaluation protocols, databases, and meteorological predictions. Atmos. Environ. 2006, 40, 4825–4838.

ES702333D

VOL. 42, NO. 3, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

837