VI. MODELS AND MODEL EVALUATION
Air quality models are indispensable tools for studying some atmospheric processes, and for the development and evaluation of alternative air pollution management strategies. The earliest models used in air pollution research were emissions-based models (EBMs), i.e., photochemical Eulerian grid-type or Lagrangian models with requisite inputs of emissions data, horizontal and vertical boundary condition estimates, meteorology and atmospheric deposition submodules, and chemical transformation modules. All these required inputs are subject to considerable uncertainties. For this reason, EBMs had to be subjected to extensive evaluations before they were accepted for application and use in making air quality management decisions. Modeling advances in recent years have produced EBMs with great sophistication and improved accuracy. Nevertheless, uncertainties persist, especially regarding the quality of emissions inventories and meteorological inputs needed for some ozone non-attainment areas. Thus, emission uncertainties introduced by fugitive emissions, by automobile engine operation and traffic volume factors, and by factors affecting some VOC emissions from vegetation have resisted improvement efforts and are still substantial. Also, meteorological models simulating air flow reversal phenomena in coastal areas are still lacking in reliability.
In reaction to these persistent problems with existing EBMs, and in response to concerns expressed in the National Academy of Science's 1999 "Rethinking" report suggesting that EPA's VOC-control approach to attainment may be less effective than NOx control in some non-attainment areas, SOS researchers undertook the development of new, more reliable methods, based on direct observation of air concentrations of ozone and each of its many VOC and NOx precursors.
SOS scientists and engineers were convinced that:
- Effective ozone management strategies can only be achieved if the plans are based on reliable meteorological models and the amounts of ozone precursors actually observed to be present in the air rather than on amounts of precursors believed to be present in the air on the basis of often-inadequate emissions inventories; and
- A reliable air quality model must not only 'get the ozone right' but also 'get the precursors right' and the 'relationships between the ozone and the precursors right for the right reasons.'
The end-result of these efforts was development of:
- A series of observation-based methods of analysis and Observation-Based Models (OBMs) that were used to evaluate various existing EBMs, and
- Recommendations for complementary use of both OBM and EBM models and approaches in making air quality management decisions.
The success of the SOS' OBM studies in resolving the issue of relative impacts on ozone of VOC and NOx controls, led to the decision to compare relative impact data obtained by OBMs with those obtained by EBM methods. Such comparisons were extremely useful in that it is, in essence, an evaluation of the EBMs' predictive performance against real-world data.
Specific key findings in the OBM and EBM study areas are given in the two sections below. The OBM-related findings describe the various observational methods developed by the SOS team, and, also the application of such methods for studying the sensitivity of ozone production to VOC and NOx within various atmospheres (e.g., power plant plumes and urban plumes), for testing chemical mechanisms, and for developing improved emission inventory data.
A. Observation-Based Models (OBM)
The OBM-related part of the SOS program consisted of three components: A series of extensive field campaigns in the Atlanta and Nashville regions, extensive analyses of the field data toward development of OBM methods, and a comprehensive effort to evaluate and develop the analytical methods needed in the field campaigns. Key findings in the first two components are summarized below.
OBM1. SOS developed a series of observation-based methods and two major Observation-Based Models (OBMs), each of which uses in-situ atmospheric observations of NOx and VOC instead of emissions inventories to drive a photochemical model and thus infer whether ozone in a given locality is more sensitive to decreases in NOx emissions or to decreases in VOC emissions. When concentration fields from the OBM were used as inputs into the Urban Airshed Model, the two models predicted similar sensitivities to NOx and VOC. These observation-based approaches offer an independent and complementary method for evaluating alternative ozone pollution abatement strategies (Kleinman, 1994, 2000; Cardelino and Chameides, 1995; Kleinman et al., 1994, 1995, 1997, 1998).
OBM2. Biogenic VOC, primarily isoprene, represented a significant fraction of the total VOC reactivity in large parts of the SOS region and significantly decreased the efficacy of ozone management strategies that considered only anthropogenic VOC emission decreases as an ozone mitigation strategy (Cardelino and Chameides, 2000).
OBM3. SOS investigators also have tested and evaluated the value of a wide variety of 'chemical indicator species' which can be used to help determine if a given locality ozone is more sensitive to decreases in VOC emissions or to decreases in NOx emissions (Milford et al., 1994; Kleinman et al., 1994, 1995, 1997; Sillman, 1995a, 1995b, 1999; Sillman et al., 1997; Tonnesen and Dennis, 2000a, 2000b).
OBM4. A wide range of values was observed for H2O2 and for summed NOx reaction products (referred to as NOz, or NOy-NOx), with no correlation between high H2O2 and high NOz. There is a strong correlation between ozone and the sum NOz+2H2O2, which appeared to be virtually identical between Nashville and Houston (Sillman, 2004).
OBM5. Consistency between different methods of determining ozone production rate (P(O3))is an important test of theoretical understanding and measurement procedures for ozone. ozone precursors, photooxidation products, and peroxy-radicals (Frost et al., 1998).
OBM6. The SOS science community has developed two complementary ways of analyzing the local rate and sensitivity of ozone production to one or the other or both of its two major precursors - NOx, VOC, or both: 1) By means of a radical budget, and 2) By means of radical propagation efficiency. These methods yield insights and useful formulae. Generalizations provided the basis for development of reliable Indicator Ratio Methods (Kleinman et al., 1997; Tonnesen and Dennis, 2000a, 2000b).
OBM7. Photochemistry under NOx-sensitive conditions preferentially formed peroxides; by contrast, under VOC-sensitive conditions, NOz compounds are formed preferentially. The sensitivity of P(O3) to NO and VOC is given by a simple analytic function of 'LN/Q," the fraction of free radicals removed by reacting with NOx (Kleinman et al., 1997).
OBM8. Ozone is formed in a chain reaction. A simple version of these reactions is:
OH + VOC ® HO2
HO2 + NO ® OH + NO2
NO2 + hv + O2® O3 + NO
Numerical calculations show that ozone yield is maximized when the chain reaction length is long. Radical loss processes limit the length of the chain of reactions that lead to ozone formation. At low NOx, HO2 radicals combine to form peroxides; at high NOx, OH reacts with NO2 forming HNO3. The combination of these two loss reactions causes a maximum in P(O3) at a particular NOx concentration. The ratio of the peroxide to HNO3 production rate tells us whether the atmosphere is on the low or high NOx side of the maximum (Tonnesen and Dennis, 2000b).
The above two findings (OBM7 and OBM8) are, in essence, a mechanistic explanation, and also justification of one "indicator species" OBM method for determining whether ozone formation is VOC- or NOx-limited. This explanation/justification reinforces the validity of the indicator species method.
OBM9. EBM models should be evaluated by determining how well they perform for two specific indicator ratios: 1) ozone/NOz, and 2) H2O2/HNO3, both of which are closely linked with NOx- or VOC-sensitive chemistry. The evaluation should be based on measurements that are congruent with peak ozone concentrations during the ozone episode of interest. Evaluations of EBMs should be done for a series of plausible model scenarios that give different results for both NOx-sensitive and VOC-sensitive ozone episodes (Sillman et al., 1997).
OBM10. Afternoon NO concentrations in the SOS region typically fell to concentrations that were at or below the limit of detection of the instruments used in EPA's Photochemical Assessment Monitoring Stations (PAMS). As a result, it is not possible to determine from PAMS measurements, whether ozone was more sensitive to decreases in anthropogenic VOC emissions or to decreases in NOx emissions (Cardelino and Chameides, 2000).
OBM11. During TexAQS 2000, use of emissions estimates based on ambient observations compared to inventory emissions resulted in air quality forecast model (NOAA-FSL) results for ozone concentrations that agreed better with measurements from the NCAR Electra and from the surface regulatory network (NOAA, 2003).
OBM12. Four-dimensional data assimilation coupling an inverse, error-minimizing algorithm with a photochemical air quality model (URM) was used to minimize differences between simulated and measured concentrations of gas-phase and aerosol species to assess biases in the emissions inventory for the eastern US. Anthropogenic SO2, high-stack point-source NOx, and biogenic VOC emissions were estimated reasonably well in inventories, while area-source anthropogenic NOx, anthropogenic VOC, NH3, and organic PM2.5 emissions may be significantly biased and require revisions from the base-case inventories (Mendoza-Dominguez and Russell, 2001a).
OBM13. Estimated adjustments to the Atlanta emissions inventory were made utilizing four-dimensional data assimilation coupling an inverse, error-minimizing algorithm with the CIT photochemical air quality model for a 1992 ozone episode. In order to match ozone model predictions, rural anthropogenic VOC emissions were approximately doubled, anthropogenic NOx and BEIS2 biogenic VOC emissions remained close to their base case value, and CO emissions were decreased. (Mendoza-Dominguez and Russell, 2001b).
B. Emissions-Based Models (EBM)
EBM1. Vertical profiles for isoprene and other VOC during SOS' 1992 Atlanta Intensive field campaign often had complex structures with local maxima at heights between 100 and 300 meters above the land surface. But the average of all vertical profiles often took the expected monotonically decreasing shape, suggesting that the complex vertical features sometimes observed are caused by transient meteorological phenomena that cancel out over longer periods of time. Because of these transient features, however, surface measurements of VOC and other ozone precursors often do not correlate well with average concentrations at 40 to 100 meters above the land surface (Andronache et al., 1994; Lawrimore et al., 1995).
EBM2. Large-scale eddies play a significant role in the transport and dispersion of airborne chemicals within the boundary layer in the SOS region. Because these eddies are not resolved by mesoscale and urban-scale chemical transport models, their effects are likely not well-simulated in these models. This, in turn, may lead to significant biases in the simulation of oxidant photochemistry by these models (McNider et al., 1993, 1998).
EBM3. Higher resolution models (with 1km rather than the usual 4 km grid squares and 1hr-long averaging times rather than typical summer day averaging times) were shown to greatly increase the accuracy and precision of both EBMs and OBMs (McNider et al., 1993).
EBM4. The Urban Airshed Model was found to be very sensitive to the method used to specify upper-air wind data (Al-Wali, 1996; Al-Wali and Samson, 1996).
EBM5. Numerical experiments with an enhanced version of the Urban Airshed Model (UAM-V) showed that model performance can be enhanced by increasing its vertical resolution in the lowest portion of the boundary layer (Sillman et al., 1995; Imhoff et al., 1995).
EBM6. Urban Airshed Model results with and without cloud transport were found to differ significantly. Model results with cloud parameterization tended to be in significantly better agreement with aircraft observations of ozone and ozone precursors than those obtained without cloud parameterization (Lin et al., 1994).
EBM7. The Urban Airshed Model had difficulty simultaneously reproducing observed isoprene and ozone concentrations in Atlanta in 1992 (Sillman et al., 1995).
EBM8. The Urban Airshed Model was able to reproduce observed urban relationships between ozone and NOy in Atlanta in 1992 (Imhoff et al., 1995).
EBM9. Incorporation of BEIS2 into the Urban Airshed Model (either UAM IV or UAM-V) led to significantly better agreement with observed ozone concentrations than use of these UAM models without BEIS2 (Guenther et al., 1993; Geron et al., 1994).
EBM10. SOS was one of the first air quality research programs to explore regional decreases in emissions of NOx and VOC. While the ROM model has perhaps been supplanted by higher-resolution, newer-generation models, the basic results shown in this study probably still prevail and are consistent with regional observations carried out under SOS for rural NOx sensitivity (Roselle and Schere, 1995).
EBM11. Several SOS studies demonstrated the power of vertical concentration profiles, indicator ratios, and ozone production efficiencies in diagnosing model outputs. Simply achieving an acceptable statistical performance for ozone is not sufficient to ensure that control actions taken based on model results will have the desired effectiveness (Sillman et al., 1995, 1998).
EBM12. Recent applications of CMAQ for PM2.5 show that results can be sensitive to the parameterization of vertical mixing, and the minimum diffusivities that are often assumed. At present, there does not appear to be a universally optimal approach. This can be critical for current efforts by Regional Planning Organizations dealing with visibility assessments and related policy decisions (Nowacki et al., 1996).
EBM13. During 1996-99, SOS undertook a special project on seasonal modeling of regional air quality and developed a Seasonal Model for Regional Air Quality (SMRAQ). This simulation model used a non-hydrostatic version of the Multiscale Air Quality Simulation Platform (MAQSIP). SMRAQ had 22 vertical layers over a 36-km horizontal grid spanning the eastern US. Meteorological inputs were provided by the Mesoscale Meteorological Model (MM5) reinitialized every 5 days for the entire 4-month ozone season in 1995. Chemistry was provided by Carbon Bond Mechanism version 4.2 as in OTAG. The model was run with the then most current national emissions inventory of anthropogenic sources developed by the Ozone Transport Assessment Group (OTAG) and BEIS2 for natural emissions (Kasibhatla and Chameides, 2000).
EBM14. Model results were compared with actual measured values at 137 locations from Texas and North Dakota through Florida and Maine during the period from May 15-September 15, 1995. Across the 137 comparisons of modeled vs measured ozone concentrations, the SMRAQ model performed better in terms of simulating seasonal rather than episodic characteristics of the regional ground-level ozone distribution. These results suggest that a seasonal model of the SMRAQ type can be a valuable adjunct to the current paradigm of developing ozone control strategies mainly on the basis of episode-specific simulations (Kasibhatla and Chameides, 2000; Houyoux et al., 2000).
EBM15. During the development of SMRAQ, together with "NARSTO-Northeast" - a sister air quality research organization - the SOS and NARSTO-NE modeling teams evaluated the MAQSIP model by comparing its predictions of ozone concentration with the actual ozone concentrations measured during July 1995 at EPA's SLAMS, NAMS, and PAMS monitoring sites in all states east of 100th meridian. During this evaluation the SMRAQ model gave: 1) Very good predictions of ozone concentration at synoptic space scales throughout the eastern US, 2) Good predictions of the magnitude but only moderately good predictions of the amplitude of diurnal variability in ozone concentrations, and 3) relative poor predictions of the ozone concentrations on different high ozone episode days during July 1995. Since mesoscale characteristics of the multiple-day episodes observed during this one-month study period were not well replicated in the SMRAQ model, the SOS and NARSTO-NE modeling teams suggested that longer-term (seasonal) model simulations may be preferable for regulatory purposes than simulations based on modeling of only one or two selected ozone episodes (Kasibhatla and Chameides, 2000; Hogrefe et al., 2001).
EBM16. Use of the Direct Decoupled Method (DDM) for computing the effects of emission controls on ozone concentrations is subject to error due to the "non-linearity problem." This problem arises from the fact that the DDM method is accurate only for small changes in emissions, and is of consequence when controls of greater than about 50% are simulated. In each of the Georgia cities in which this study was made, local sources, on top of regional background sources had the major, if not dominating, role in the source-receptor relationships observed, but the make-up of other sources that impacted the cities was significantly different. On a ppb-of-ozone per ton of NOx basis, local, ground-level sources tend to have the greatest effectiveness (Odman et al., 2002).
EBM17. Results using the coupled LES chemical model show that the turbulence paradigm that most air quality models are based upon, i.e. first order closure, can be trusted even within the deep boundary layers of the Southeast. However, details on how the first order closures (K profiles) are formulated can make a difference in model results. Thus, additional research is needed to ensure that the K-profiles imposed on many current air quality models reflect the appropriate turbulent intensities and scales in real-world boundary layers (Herwehe, 2000).
EBM18. Consideration of air quality impacts across multiple episodes increased the importance of local sources. Regional modeling of source impacts on air quality showed that while long-range transport of pollutants and their precursors can be very significant during any one episode, when viewed across multiple episodes and meteorological conditions, the impact of local emissions became more important. Different long-distance source regions affected an area under different conditions, while local sources tended to have a role on a day-in/day-out basis. Thus, local controls will still have benefits, even in areas that are impacted from more distant sources (Odman et al., 2002).
EBM19. Ozone formation potentials of individual VOC were consistent between different "metrics" or "indices" of reactivity across wide geographical domains. As noted in other SOS-related research, VOC controls were potentially beneficial in many urban areas. However, individual VOC can have very different impacts on the amount of ozone formed. Currently, the regulatory structure developed by the USEPA treats all VOC as either reactive or non-reactive, not taking in to account the spectrum of variability among VOC with greater or lesser ozone production reactivity. Regional modeling showed that the use of relative reactivities led to consistent results for ozone reactivity, suggesting that such a scale should be used in regulatory policy setting (Hakami et al., 2004).
EBM20. Results from box model simulations run under conditions based on Houston's industrial regions suggest that emissions of as little as 100 pounds of light alkenes (ethylene, propylene, butenes, pentenes, butadiene) and aromatics can lead to >50 ppb enhancements of ozone concentrations per hour over a 1km2 area. Ozone productivities of alkane emissions are generally significantly lower than for alkenes and aromatics. The box model simulations also indicate much higher ozone productivities under conditions that involve high concentrations of both VOC and NOx, as opposed to conditions that involve high concentrations of VOC alone. (Daum et al., 2002)
EBM21. Rapid rates of NOy depletion imply that NOx may not be transported as far as is commonly assumed in current air quality models including UAM-IV, UAM-V, MM5, ROM, RADM (Nunnermacker et al., 1998, 2000; Daum et al., 2000a, 2000b).
EBM22. Photochemical models using the common Carbon Bond 4 chemical mechanism gave reasonably good estimates of ozone concentrations during rapid ozone formation episodes in Houston. These same models and mechanisms tended to over-predict NOz concentrations (and especially HNO3 concentrations) during rapid ozone formation episodes in Houston. These reasonably good estimates of ozone concentrations and over-predictions of NOz concentrations (and especially HNO3 concentrations) also were observed in recent applications of the Comprehensive Air Quality Model with Extensions (CAMx) photochemical model (Gillani and Wu, 2003b, 2003c).
EBM23. Chemical reactions with chlorine can increase ozone in Houston. Chemical reactions involving chlorine were incorporated into the Comprehensive Air Quality Model with Extensions (CAMx) photochemical model. An inventory of chlorine sources in the Houston urban area also was developed. Results from the chorine-included CAMx model indicated that estimated ozone concentrations were increased by 5-15 ppb compared to model results without chlorine chemistry. This was true in both the Ship Channel area and in other areas of Houston (Allen, 2003).
EBM24. Both temporal and spatial fluctuations in air turbulence and distances between major NO and VOC sources can change the effective rates of ozone formation reactions. Comparison of modeling results using both coarse-grid and fine-grid versions of the LES-Chem model show that:
- If NO and reactive VOC plumes are mixed by turbulent meteorological conditions soon after emission, rapid ozone formation will occur immediately and in close proximity to the NO and VOC emission sources; and conversely
- If NO and reactive VOC plumes occur under low-turbulence conditions, convergence of the two plumes can be delayed, and rapid ozone formation will occur at longer distances from emissions sources (Gillani and Wu, 2003a).
EBM25. The web-based Lagrangian Particle Model (LPM) developed at UAH has proven to be a useful tool for visualizing the transport and dispersion of atmospheric emissions. Since the model uses particles only as massless tracers, it is equally applicable to either gaseous or fine particulate emissions. UAH modified the LPM so that it can be operated through a web-page interface. When the run is completed, the user is given access to a temporal sequence of particle-position snapshots depicting transport and dispersion of particles released from the sources (http://texaqs.nsstc.uah.edu) (McNider, 2004).