Ozone Modeling for the Southern Appalachian Mountains Initiative

Talat Odman, James Boylan, James Wilkinson, Yueh-Jiun Yang and Armistead Russell

Georgia Institute of Technology, 200 Bobby Dodd Way, Atlanta, GA 30322-0512

Kevin Doty and Richard McNider

University of Alabama, Huntsville, AL

ABSTRACT (300 words)

As part of the Southern Appalachians Mountains Initiative (SAMI), the effects of emission controls on air quality are being assessed using the EMS-95 emissions model, RAMS meteorological model and the URM chemistry-transport model.  URM, originally a photochemical model, has been extended to include aerosol dynamics, heterogeneous chemistry, cloud processes and wet deposition in order to simulate how emissions controls impact not only ozone, but also visibility and acid deposition using a "one atmosphere" or integrated approach.  Using a triple nested grid, RAMS is being applied in the non-hydrostatic wet mode to provide more accurate precipitation and humidity fields to the URM. Advanced nudging techniques have been developed to improve humidity predictions given their importance to correctly simulating heterogeneous chemistry.  EMS-95 provides emissions of gas, aerosol and acid precursors. The multiscale URM grid, which covers the eastern United States, consists of a fine grid of 12 km over the southern Appalachians and surrounding areas, and successively coarser grids up to 192 km near the boundaries of the domain. Nine episodes, each seven to ten days long, are being modeled; the results will be used to develop seasonal and annual air pollution indices.

In this paper, URM is evaluated for its ability to accurately simulate ozone in two episodes. Both episodes,  May 24-29, 1995 and July 11-19, 1995, are summertime, higher ozone periods.  Performance was good over the lower elevation, urban sites as well as higher elevation, rural sites.  Overall daily normalized bias was low (generally less than 10%), with low normalized error (15-20%), suggesting that the model can be confidently used for assessing control strategy impacts. Finally, the simulation with the July 1995 meteorology is repeated with two different sets of emission estimates corresponding to 2010, The response of ozone to these emission scenarios is evaluated. There are significant differences between the responses at urban and rural sites, especially in nighttime ozone, which is estimated to increase at urban sites.

One of the capabilities of URM is to conduct direct sensitivity analysis.  While the model is simulating a particular episode, it also calculates the sensitivity of the results to emissions. In this case, the SAMI domain has been divided into eight source regions, and the source-impact relationships between ozone in the Southern Appalachians and emission sources in the various regions were studied.  There is significant variability in the impacts between elevated and low level NOx emissions, as well as between source regions.

INTRODUCTION (400-700 words)

Studies that have been conducted in national parks and national forest wilderness areas of the southern Appalachian Mountains have documented adverse effects to visibility, streams, soil, and vegetation.  Poor air quality in the region has been implicated as the major source of the adverse effects.  Beginning in 1990, the Federal Land Managers for Shenandoah National Park, Great Smoky Mountains National Park, and Jefferson National Forest/James River Face Wilderness Area made several adverse impact determinations in their review of proposed air permits for major new sources.  Although it is known that the air pollution levels which currently affect park and wilderness resources come from existing sources of pollution — large and small, mobile and stationary, near and distant—the relative contribution of each source type to the regional air pollution problem is not well quantified. 

The 1990 Clean Air Act Amendments (CAAA) require major emissions reductions for primary airborne pollutants, including sulfur oxides (SOX), nitrogen oxides (NOX), and volatile organic compounds (VOCs).  Although the reductions are expected to produce air quality improvements, it is uncertain whether the results will be enough to protect and preserve the ecosystems and natural resources of the Southern Appalachians, especially in Class I areas.

As part of SAMI, an Integrated Assessment is being built which will model and assess the environmental and socioeconomic responses to changes in air emissions, which result from various emissions management strategies.  One component of the Integrated Assessment is the atmospheric modeling of the air quality responses to emissions  controls.  Its purpose is to characterize the air pollution formation processes that affect air quality in the southern Appalachian Mountains.

SAMI's approach calls for the modeling of a limited number of past episodes that best characterize recent (1991-95) trends in ozone, visibility and acid deposition in Class I areas of the region. The goals are to assess the sensitivities to changes in emissions and to model the impact of future emission changes on air quality. Using data classification and optimization techniques,1 nine episodes, each 6 to 10 days long, were selected to represent the seasonal and annual air quality metrics most relevant to visibility, stream, and forest effects.. 


SAMI’s atmospheric modeling approach is unique in combining gaseous, aerosol, and deposition processes in one integrated atmospheric modeling system. The major components include RAMS for meteorology, EMS-95 for emissions, and URM for the modeling of transport and chemistry. Here the most important features of the models will be highlighted and their recent modifications will be summarized. More in depth discussions and detailed descriptions of how the models are used in this project can be found in respective protocol documents for meteorological,2 emissions,3 and air quality modeling.4

Meteorological Modeling

A modified version of the Regional Atmospheric Modeling System5 (RAMS) version 3a was used in the SAMI meteorological simulations.  It was run with a system of three nested grids in a nonhydrostatic mode with only cloud and rain water activated.  The latter restriction was chosen because of computer memory and runtime limitations.  The nested grid structure used for the July 1995 episode consisted of 48-, 24-, and 12-km resolution grids.  The next episode modeled revealed undesirable interactions between the nested grid boundaries.   Therefore, a new configuration was chosen which consisted of grids of resolutions 96, 24, and 12 km and was used on all other episodes. Some of the more important modifications of the modeling system will be discussed next.

Modifications of RAMS

The National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) reanalysis data6 was chosen as the main data source for the meteorological simulations.  The initialization procedure was independent of the RAMS system whereby the NCEP/NCAR reanalysis data was transformed into a form which could be used to initialize the RAMS model and also provide nudging fields.  The majority of the initialization calculations were spent on decomposing the total wind field into its rotational and divergent components in a manner similar to Lynch7 and then adjusting the divergent component in a way to give a zero vertical velocity at the top of the model domain (~ 16 km) as in O’Brien.8

To partially compensate for not activating the ice microphysics the following changes were made: the autoconversion of cloud water to rain water was changed to a Kessler type formulation,9 and the precipitation terminal velocity and collection efficiency were made functions of temperature.  RAMS-3a has a modified Kuo scheme10 for the cumulus parameterization with a simple downdraft.  A convective inhibition11 (CIN) threshold was added to the cumulus parameterization so that, in general, convective precipitation was not allowed regardless of the convective available potential energy (CAPE) if the CIN had values of 20 J kg-1 or more.  Most of the changes dealt with the planetary boundary layer (PBL) calculations.  The original RAMS-3a model uses the Louis parameterization12 in determining the surface fluxes which eliminates the need for any iterative calculations to determine the Monin-Obukhov length.13  Beljaars and Holtslag14 provide evidence that the difference between the Louis approach12 and an exact calculation can be large in stable situations.  Therefore, an efficient iterative approach, which explicitly solves for the Obukhov length as a function of the surface Richardson number, has been implemented using the general recommendations of Beljaars and Holtslag.14  The transition of the surface fluxes to the free convection regime was implemented in a fashion similar to Beljaars.15  Attempts at simulating an episode early in the project (not included in this paper) made clear the problems created by the inability to initialize the soil moisture adequately.  As is well known for atmospheric simulations on the order of 10 days, both initial and boundary conditions are important.  To circumvent this problem a scheme was developed whereby the surface fluxes were constrained by observed surface analyses of temperature, moisture, and wind speed.  In and close to mountainous areas only the surface moisture was treated in this manner.  The reason for this was to allow the correct mountain-valley circulations to develop which would not be contained in the relatively coarse observed surface analyses.  This nudging technique was used on all the episodes after the July 1995 episode.

Discussion of Meteorological Modeling Results

Relative to a subset of about twenty five National Weather Service (NWS) sites in and close to the immediate SAMI region, the bias and standard deviation for various near-surface variables have been calculated.  The predicted values used in these statistical calculations are obtained from hourly model averages on the 12-km grid which are linearly interpolated in space to the observation site.  As an example, the biases of the temperature at two meters for the episodes in this paper ranged from +1.0o F to –3.9o F.  The standard deviations of the same variable ranged from 4.3o F to 5.6o F.  At least part of this temperature error in the Appalachians was due to remaining problems with clouds.  Although outside the scope of this paper, the latter involves not only the incorrect placement and depth of clouds, but also the parameterization of their effects (e.g., the liquid water path used in the short and long wave radiation calculations).  These errors were considered acceptable given that they are similar to other numerical model comparisons.  For the episodes after the July 1995 case for areas outside of the Appalachians the temperature biases were much less due to the impact of the nudging technique described above.  The timing and amount of precipitation are the most difficult variables for a meteorological model to simulate.  The July 1995 model results underpredicted precipitation due to a dry bias from the soil moisture problems mentioned earlier and also due to a minimum horizontal diffusivity, which is set too high.  Overall the precipitation timing and amounts were reasonable for the other episodes.  Objective measures of the precipitation performance are still in the process of being calculated. The interested reader is referred to https://www.saminet.orgah.edu/sami for detailed and most up to date results.

Emissions Modeling

The detailed inventories used in this project are prepared by the Pechan-Avanti Group. After checking the quality of these inventories, the Emission Modeling System16 (EMS-95) is used to generate gridded, time varying and speciated emission inputs to be used by the chemistry-transport model. EMS-95 combines the emissions in two categories: point and area source emissions. Area sources include low-level point sources, mobile sources, anthropogenic area sources, non-road mobile sources, and biogenic sources.  Biogenic emissions and emissions from mobile sources are estimated using U.S. EPA’s models:  Biogenic Emissions Inventory System, version 2 (BEIS2) and MOBILE5b highway vehicle emission factor model, respectively. In this project, the point source inventories are enriched with the day specific emissions data that are obtained from major utility companies in the SAMI region. Also, meteorological model results are used to generate the diurnal patterns of biogenic emissions.

Air Quality Modeling

The Urban-to-Regional Multiscale (URM) model is a three-dimensional Eulerian air quality model that accounts for the transport and chemical transformation of pollutants in the atmosphere.17  The model has evolved from the CIT (Carnegie/California Institute of Technology) airshed model.18 URM uses variable size grids in its horizontal domain to effectively capture the details of pollution dynamics without being computationally intensive.

In this project, the URM model has been expanded to include aerosol dynamics and wet deposition scavenging processes.  The ISORROPIA aerosol module19 and the Reactive Scavenging Module for acid deposition20, have been incorporated into the URM framework to produce an integrated “one atmosphere” airshed model. More detailed description of the aerosol and acid deposition modules can be found elsewhere.21

Brief Description of URM model

In this section, transport, chemistry and deposition processes of the URM model are briefly discussed.

Transport Processes

URM uses the two-dimensional Streamline Upwind Petrov-Galerkin (SUPG) finite element method for solving the horizontal advection equations.22, 23  The SUPG is a high-order accurate scheme, but is not monotonic or positive definite.  To avoid negative concentrations, the SUPG finite element solution is followed by application of a mass conservative diffusion filter.24  In this project, selective application of filtering in the streamline direction was abandoned in favor of an isotropic diffusion filter. For the solution of vertical advection URM uses the simple, first-order upwind differencing. To avoid mass conservation problems, the vertical velocities are adjusted by solving the continuity equation using the same numerical techniques.25  This adjustment has little effect on vertical transport.

Vertical and horizontal diffusion are treated using K-theory.  The values of the vertical and horizontal diffusion coefficients are obtained from the meteorological inputs.  Vertical diffusion is solved using an implicit finite difference scheme.  Horizontal diffusion is solved together with horizontal advection using the SUPG finite element method.

Early versions of URM did not incorporate a treatment for cloud processes. As part of this project, the Reactive Scavenging Module20 (RSM) has been incorporated into URM. The convective cloud component of RSM is based on a model developed by Scott.26 Convective precipitation is simulated with a two-cell, steady-state model. This approach allows the definition of several characteristics of convective clouds including large updrafts and vertical transport of low level air to upper levels.

Chemistry Processes

The gas-phase reaction kinetics are calculated by using the SAPRC chemical mechanism,27 which has been updated with a more accurate account of isoprene and some other species.28 For the solution of chemical kinetics, the model uses the hybrid method of Young and Boris.29 URM’s aqueous-phase chemistry is based on what is implemented in the RSM. It includes the reactions of S(IV) with H2O2 and O3 to form S(VI).  Hydrogen ion concentrations in cloud water and rain are calculated from an electroneutrality equation, based on the concentrations of anions such as sulfate and nitrate, and cations such as ammonium and other positive ions associated with crustal material. Lumped organic aerosol yields of Pandis et al. are used in URM.30  In this approach, gas phase reactions yield secondary organic aerosols.

Deposition Processes

For dry deposition, URM uses the three-resistance approach based on the formulation of Wesely.31  For the aerosol particles, size dependent particle deposition velocities are calculated using data from NCAR.32  Removal by wet deposition is based on the parameterization in the Reactive Scavenging Module (RSM) incorporated in URM for the SAMI project.

Modeling domain and grid

The modeling domain covers the eastern half of the United States and is shown in Figure 1. URM differs from other air quality models in the way it provides multiscale modeling capability. While other models resort to grid nesting techniques, URM provides a single grid with variable

Figure 1. The multiscale grid used in the URM model: the urban areas are shown in green and the boundaries of the subdomains used in the sensitivity analysis are separated by bold lines.

resolution. The grid cell dimensions shown correspond to 192, 96, 48, 24, and12 km with the finest resolution (i.e.,12 km) cells over the southern Appalachian Mountains and the adjacent areas that are expected to directly influence the air quality in the region.  Coarser cells are placed in areas that are not expected to significantly contribute to the air quality in the SAMI region with the coarsest cells near the boundary of the domain.  The domain extends from the surface to a height of 12,867 meters and is divided into seven vertical layers.  The thickness of each layer from the ground to the top of the domain is given in Table 1. The use of finer resolution near the surface of the domain, as compared to the more coarse resolution aloft, allows capturing the steeper concentration gradients that typically exist in the near-surface troposphere and the

Table 1. Vertical structure of the URM model.


Coverage (m)

Thickness (m)























evolution of the mixing depths during the day. 

Initial and Boundary Conditions

URM requires initial (IC) and boundary conditions (BC) for both gaseous and aerosol species.  Initial concentrations of the modeled species are needed to initiate the model. Initial conditions can have a major impact on the modeled concentrations at the beginning of the simulation, but the impact usually diminishes as the simulation proceeds. Gaseous species, having relatively short lifetimes, quickly undergo chemical transformation and deposition, and their effects become negligible quicker compared to aerosol species. A two-day ramp-up period was used before each SAMI episode. This is generally sufficient to damp out the effect of initial conditions on gaseous species concentrations. However fine aerosols, that have smaller settling velocities and longer lifetimes, can persist for longer periods. Further, some aerosol species do not undergo chemical transformation and their only method of removal is through deposition and/or transport out of the domain. Therefore, the concentrations of aerosol species have a stronger dependence on initial conditions and extra care must be used in setting those conditions. 

The same principles apply to the boundary conditions. The lifetimes of gaseous species are not long enough for transport from the domain boundaries into the Southern Appalachian Mountains. However, the aerosol species can be transported considerable distances (e.g., hundreds of kilometers) and they can impact the concentration of aerosols in the Southern Appalachian Mountains. The IC/BCs for the gaseous species were derived using data from the Aerometric Information Retrieval System (AIRS) and NARSTO-NE data archives as well as data from specialized studies and smaller networks. The IC/BCs for the aerosol species were derived from the Interagency Monitoring of Protected Visual Environments (IMPROVE) measurements.

The gaseous species for which IC/BCs have been derived include SO2, NOx, VOCs and ozone. The conditions are based on measurements at a number of different sites within the modeling domain. The initial conditions are based on observations that correspond to the time and date that the model simulation begins. Boundary conditions are based on observations that vary spatially and temporally over the duration of the modeling episode. Because the monitoring network does not correspond to the modeling grid, it is necessary to interpolate the observed values to the computational nodes on the modeling grid. For all gas phase species, except SO2 and NOx, Dirichlet tessellation  (triangulated irregular network interpolation) is used to determine concentrations at every node on the modeling grid. The values at the top of the domain are set to free troposphere values. Linear interpolation from the ground level values to the top of the domain is used to derive the IC/BC concentrations for layers in between. The IC/BCs for SO2 and NOx are treated differently than the other gas phase species. Since the AIRS SO2 and NOx monitors are usually located in areas with high concentrations (e.g. downwind from power plant plumes), interpolation using Dirichlet tessellation overestimates the initial and boundary conditions. An inverse concentration, distance squared weighting of the natural log of the concentrations was used instead. This minimizes and localizes the impact of locally high SO2 and NOx observations. The IC/BCs for the other gas phase species in the SAPRC chemical mechanism are set equal to zero and are allowed to evolve during the simulation.


Modeling is being continued for a total of nine episodes that will be used to develop seasonal and annual air quality indices to assess ozone, visibility and acid deposition problems in the SAMI region.  This paper focuses on two episodes: July 11-19, 1995 and May 24-29, 1995. Both are summertime, high ozone periods with significant contributions to the seasonal ozone index. Up to date modeling results for other episodes can be found at the following web site address: https://mesl.ce.gatech.edu/. Animations of hourly ozone levels over the SAMI region can be downloaded from that web site. Also available are the daily maximum ozone concentration plots such as the one shown in Figure 2, for May 26, 1995. These plots are intended to give a quick look at the ozone levels in the region in comparison to the current national air quality standard of 120 ppb. Table 2 compares each day during these episodes in terms of observed ozone levels at Great Smoky Mountain (GSM) and Shenandoah National Parks (SNP). The classification is based on cumulative ozone levels (W126) where Class 1 corresponds to 375 ppb or lower, Classes 2 and 3 are from 375 to 750 ppb and 750 to 1150 ppb, respectively, and Class 4 is assigned to W126 larger than 1150 ppb. These numbers slightly vary for each park

A comprehensive set of statistical calculations have been performed to evaluate the ability of the model to accurately estimate ambient ozone concentrations.  The observational data used in these calculations are obtained from the Aerometric Information Retrieval System (AIRS). Among the statistical measures to be examined are the mean bias, normalized bias, mean error, and normalized error.  The statistical calculations were done using the Modeling Analysis and Plotting System (MAPS) package.33  MAPS also produces plots showing the time variation of predicted and observed ozone levels at specific sites.

Daily averaged biases and errors

The daily mean bias (MB) is calculated as:


Table 2. Classification of simulated episode days according to observed cumulative ozone levels.


O3 Class at GSM

Ozone Class at SNP

May 24, 1995



May 25, 1995



May 26, 1995



May 27, 1995



May 28, 1995



May 29, 1995



July 11, 1995



July 12, 1995



July 13, 1995



July 14, 1995



July 15, 1995



July 16, 1995



July 17, 1995



July 18, 1995



July 19, 1995




where  is the model-estimated hourly ozone concentration at station i,  is the observed hourly average ozone concentration at station i, and N equals the number of hourly estimate- observation pairs drawn from all valid monitoring station data on the simulation day of interest. The daily mean error (ME) is calculated in a way similar to the mean bias:


Figure 2. Predicted daily maximum ozone concentrations in the SAMI region for May 26, 1995.



The normalized mean bias (NMB) and normalized mean error (NME) are defined as:



Since the normalized quantities can become large when  is small, a cut-off is used in conjunction with Equations (3) and (4). Whenever  is smaller than the cut-off value, that estimate-observation pair is excluded from the calculations. A cut-off value of 40 ppb is used in the calculations reported here.

There are several hundred AIRS stations within the modeling domain reporting data during each episode. However, some of these stations fall into course resolution cells where the model does not capture local gradients and the predictions are not as accurate. Since the performance over the SAMI region is of greater interest, only the seventy four stations falling into 12 km grid cells are used in the calculations reported here.

Table 32 shows a summary of the statistics for each day during the May 24-29, 1995 and July 11-19, 1995 episodes. Note that the normalized biases are within  with the exception of July 19, 1995. On this day, the normalized bias is –17.5% with a mean bias of about -13 ppb. The normalized errors are less than 24% and the mean error is the largest, 16 ppb, also on July 19. Observation of biases and errors on sequential days of each episode shows no evidence of growth in those statistics even though the maximum mean bias and mean error were observed on the last day of the July 11-19, 1995 episode. There is also no clear sign of an alarming systematic bias even though more days were biased low.

EPA recommends that the daily normalized bias fall within  for urban scale modeling applications. The normalized bias is plotted in Figure 23 for easy comparison with EPA’s recommended limits. Note that urban applications involve much smaller domains and shorter simulations, therefore models can afford finer grid resolution, typically 4 km. In general, urban scale models can remain within the guidelines much easier than regional scale models with coarser grid resolution. Unfortunately, there is no guidance on how to judge model performance in regional applications. Given that the domain of interest is much larger, the finest grid resolution is limited to 12 km, and that the only time the normalized bias went below -15 % was on the last day of a 9 day episode, URM’s performance in predicting ozone concentrations over the SAMI region can be viewed as acceptable.

Table 32. Ozone performance statistics for May 24-29, 1995 and July 11-19, 1995 episodes.


Norm. Bias (%)

Norm. Error (%)

Mean Bias (ppb)

Mean Error (ppb)












































































Figure 23. Daily normalized biases for May 24-29, 1995 and July 11-19, 1995 episodes.

Hourly ozone plots at specific sites

The statistical measures above are useful in giving a general picture of model performance. However, it is also important to understand how the model performed at each site and how this performance varied throughout each day. Since SAMI’s main concern is for ozone in Class I areas, it is also important to discern between performance at low-elevation urban sites and high-elevation rural sites. For these purposes, hourly ozone plots, also known as time series plots, are used. The plots present the hourly estimates and observations throughout the simulation period. The model estimates are derived in three different ways. First the value for the cell that the station falls in is reported. Second, a weighted value is reported using bilinear interpolation from the nearest four grid cells to the monitor. Third, the value that is closest to the observation within the nearest four grid cells is reported as best. With the time series plot one can determine the model's ability to reproduce the peak estimation, the presence or absence of significant bias and errors within the diurnal cycle, and whether the "timing" of the estimated peak agrees with the observation.

While MAPS produces time series plots for all of the AIRS stations in the domain, 33 sites were identified for detailed analysis. The examples given here are from typical urban and rural sites. The interested reader is referred to the web site for other sites. Figures 34 and 45 show the hourly ozone estimates and measurements at South Dekalb, Georgia, a low-elevation urban site, for the July 11-19, 1995 and May 24-29, 1995 episodes, respectively. The figures include the two ramp-up days, which should be ignored. Note that URM underestimates the peaks by as much as 40-50 ppb on July 11, 14, 15 and 19. The peak underestimation is of the order of 30-40 ppb on May 27 and 29. On all other days the peak estimates are quite good. The diurnal variations and the timing of the peaks conform well with observations.

Figure 3. Observed and predicted ozone levels at South Dekalb, Georgia for July 9-19, 1995.


Figure 5 shows the ozone time series at Look Rock, Tennessee, a high-elevation rural site, for the July 11-19, 1995 episode. Note that if the model estimates were raised by about 10 ppb, they would match the observations very well. This was the case, i.e., the model predictions matched observations, when Dirichlet tessellation was used to initialize the NOx fields. Clearly, the new


Figure 4. Observed and predicted ozone levels at South Dekalb, Georgia for May 22-29, 1995.


Figure 5. Observed and predicted ozone levels at Look Rock, Tennessee for July 9-19, 1995.

weighting used in setting initial concentrations for NOx, is leading to a lowered background ozone because of lower initial and boundary conditions for NOx. This result has important implications for ozone modeling and will be investigated further. Since no data were available at this site for the May 24-29 episode, time series at Clingsman Dome, Tennessee, a nearby site which is also classified as high-elevation rural, are shown in Figure 6. There is good agreement


Figure 6. Observed and predicted ozone levels at Clingsman Dome, Tennessee for May 22-29, 1995.

between the model estimated and observed ozone at this site throughout the episode.


For a preliminary assessment of how ozone levels may respond to emission changes in the future, the simulation of the July 11-19, 1995 episode was repeated using two scenarios for the year 2010. These scenarios reflect the expected changes in VOC and NOx emissions according to the rules and regulations that are already on the books (OTB) or on the way (OTW). In Table 4, the domain wide total, antropogenic VOC and NOx emissions for these two scenarios are compared to the base case (i.e., July 1995) for a typical day. The antropogenic VOC emissions decrease by 20% in the 2010 OTB scenario and by 24% in the 2010 OTW scenario with respect to the base case, while the decreases in NOx emissions are 24% and 39%, respectively. Note that these are not spatially uniform reductions.

Figure 7 compares the ozone levels estimated with the 2010 OTB and OTW emission scenarios to the base case in Atlanta, Georgia. The peak ozone levels do not change significantly with the OTB scenario. The largest decrease corresponds to 7 ppb on July 17. An increase in nighttime ozone levels is more noticeable. This may be explained by the reduced NO levels scavenging less ozone at night. On July 12, the near zero nighttime ozone levels of the base case increase to 20 ppb. The larger reductions in the NOx emissions of the OTW scenario may be leading to more pronounced decreases in peak ozone levels. While there are no significant decreases for July 11, 14 and 19, the peak ozone is estimated to drop by as much as 10-15 ppb on other days. On the other hand, it is estimated that the nighttime ozone levels will significantly increase, up to levels of 35 ppb, leading to larger daily totals. This represents how a typical urban site may respond in the future.

The response shown in Figure 8 for the Great Smoky Mountain National Park is substantially different. At this typical high-elevation rural site, even with the OTB scenario, the peak ozone levels are estimated to drop significantly, by as much as 10 ppb. The benefits of the OTW scenario are more pronounced, of the order of 15-20 ppb. The nighttime ozone levels are also estimated to drop, of the order of 5 and 10 ppb with OTB and OTW scenarios, respectively. Thus, both scenarios are expected to lead to lower daily total ozone at high-elevation rural sites.


Table 4. Daily, domain wide total, antropogenic emissions for July 1995 base case and 2010 OTB and 2010 OTW scenarios.

Emission Scenario

VOC Emissions(tons)

NOx Emissions (tons)




2010 OTB



2010 OTW



Figure 7. Estimated ozone levels in Atlanta, Georgia for July 11-19, 1995 using the basecase and 2010 OTB and OTW emissions.


Figure 8. Estimated ozone levels in Great Smoky Mountain, Tennessee for July 11-19, 1995 using the basecase and 2010 OTB and OTW emissions.



As part of the SAMI project, the RAMS-EMS95-URM atmospheric modeling "system" is being applied to assess how emissions controls will impact ozone, particulate matter and acid deposition in the southern Appalachians.  The system was evaluated for its ability to accurately simulate the evolution of ozone in this region, with particular attention to correctly tracking ozone in both rural and more urban locations in the target region.  Typical performance over the two higher ozone episodes modeled showed low bias and error, both within the guidelines suggested for urban scale modeling.  It was found that the performance was better in the region with the finer grid resolution, suggesting that the larger grids, as expected, did not capture the finer scale structures locally.  

A preliminary assessment using future emission scenarios showed significant differences between the responses of low-elevation urban and high-elevation rural sites. The peak ozone estimates at urban sites did not decrease significantly with either scenario. Further reductions in NOx with the OTW scenario (19% over OTB) led to relatively lower peaks, but significantly higher nighttime levels, making the daily total ozone larger that the base case (i.e., July 1995). Both peak and nighttime ozone levels were estimated to drop at high-elevation rural sites. The OTW scenario led to larger decreases: 15-20 ppb in peak ozone and as much as 10 ppb in nighttime levels.


This project is funded by the Southern Appalachian Mountains Initiative. The authors thank the members of the SAMI Atmospheric Modeling Subcommittee for their constructive criticism, innovative ideas and invaluable help throughout this project.


1.      Deuel, H.P.; Douglas, S.G. Episode selection for the integrated analysis of ozone, visibility and acid deposition for the Southern Appalachian Mountains; Systems Applications International, Inc.: San Rafael, CA, 1998; SYSAPP-98/07r1;

2.      Norris, W.B.;  Doty, K.G. The Southern Appalachians Mountain Initiative: Meteorological Modeling Protocol; University of Alabama in Huntsville, 1998.

3.      Russell, A.G.; McNider, R.; Wilkinson, J.G.; Moody, J. Meteorological, Emissions and Air Quality Modeling for an Integrated Assessment Framework in Support of the Southern Appalachians Mountain Initiative: Emissions Modeling Protocol; Georgia Institute of Technology, 1998.

4.      Russell, A.G.; McNider, R.; Wilkinson, J.G.; Moody, J.; Odman M. T. Meteorological, Emissions and Air Quality Modeling for an Integrated Assessment Framework in Support of the Southern Appalachians Mountain Initiative: Air Quality Modeling Protocol; Georgia Institute of Technology, 1998.

5.      Pielke, R.A.; Cotton, W.R.; Walko, R.L.; Tremback, C.J.; Lyons, W.A.; Grasso, L.D.;. Nicholls, M.E.; Moran, M.D.; Wesley, D.A.; Lee, T.J.; Copeland, J.H. Meteor. Atmos. Phys. 1992, 49, 69-91.

6.      Kalnay, E.; Kanamitsu, M.; Kistler, R.; Collins, W.; Deaven, D.; Gandin, L.; Iredell, M.; Saha, S.; White, G.; Woollen, J.; Zhu, Y.; Chelliah, M.; Ebisuzaki, W.; Higgins, W.; Janowlak, J.; Mo, K.C.; Ropelewski, C.; Wang, J.; Leetmaa, A.; Reynolds, R.; Jenne, R.; Joseph, D. Bull. Amer. Meteor. Soc. 1996, 77, 437-471.

7.      Lynch, P. Mon. Wea. Rev. 1989, 117, 1492-1500.

8.      O'Brien, J. J. J. Appl. Meteor. 1970, 9, 197-203.

9.      Kessler, E. Meteor. Monogr. 1969, 32, 1-84.

10.  Kuo, H.L. J. Atmos. Sci. 1974, 31, 1231-1240.

11.  Colby, F.P. Mon. Wea. Rev. 1984 112, 2239-2252.

12.  Louis, J.F. Bound.-Layer Meteor. 1979, 17, 187-202.

13.  Monin, A.S.; Obukhov, A.M. Akad. Nauk SSR Geofiz. Inst. Tr. 1954, 24, 163-187.

14.  Beljaars, A.C.; Holtslag, A.M. J. Appl. Meteor. 1991, 30, 327-341.

15.  Beljaars, A.C. Q. J. R. Meteorol. Soc. 1994, 121, 255-270.

16.  Wilkinson, J.G.; Loomis, C.F.; McNally, D.E.; Emigh, R.A.; Tesche, T.W. Technical Formulation Document: SARMAP/LMOS Emissions Modeling System (EMS-95)., Alpine Geophysics, Pittsburgh, PA 1994; AG-90/TS26 & AG-90/TS27.

17.  Kumar, N.; Odman, M.T.; Russell, A.G. J. geophys. Res. 1994, 99, 5385-5397.

18.  McRae, G.J.; Seinfeld, J.H. Atmos. Environ. 1982, 16, 679-696.

19.  Nenes, A.; Pandis, S.N.; Pilinis C. Aquatic Geochemistry, 1998, 4, 123-152.

20.  Berkowitz, C.M., Easter R.C. Scott B.C. Atmos. Environ. 1989, 23, 1555-1571.

21.  Boylan, J., Wilkinson, J.G., Yang, Y.-J., Odman M.T., Russell, AG. this conference 2000.

22.  Odman, M.T.; Russell, A.G. Atmos. Environ. 1991, 25A, 2385-2394.

23.  Odman, M.T.; Russell, A.G. J. Geophys. Res. 1991, 96, 7363-7370.

24.  Odman, M.T.; Russell, A.G. Atmos. Environ. 1993, 27A, 793-799.

25.  Odman, M.T., Russell, A.G. In Air Pollution Modelling and its Applications XII; Gryning, S.-E.; Batchvarova, E., Eds.; Plenum Press: New York, NY, 2000.

26.  Scott, B.C. User’s Manual for the Convective Cloud Module Version 1.0; Pacific Northwest Laboratory, Richland, WA 1987; PNL-6188.

27.  Carter, W.P.L. Atmos. Environ. 1990, 24, 481-518.

28.  Carter, W.P.L. Atmos. Environ. 1995, 24, 481-518.

29.  Young, T.R.; Boris, J.P., J. phys. Chem. 1977,   81, 2424-2427.

30.  Pandis, S.N.; Harley, R.A.; Cass, G.R.; Seinfeld, J.H. Atmos. Environ. 1992, 26A, 2269-2282.

31.  Wesely, M.L. Atmos. Environ. 1989, 23, 1293-1304.

32.  National Center for Atmospheric Research, Regional Acid Deposition: Models and Physical Processes, Boulder, CO,1982.

33.  McNally, D.E.; Tesche, T.W. Modeling Analysis and Plotting System User Manual; Alpine Geophysics: Pittsburgh, PA, 1991.


Air quality, ozone, modeling, SAMI, model performance, sensitivity analysis