1. INTRODUCTION

Several large urban areas throughout the world are plagued by severe air quality problems. One of the most common problems is photochemical smog or ozone which is formed under sunlight by chemical reactions between nitrogen oxides and volatile organic compounds. These primary pollutants are emitted by a wide range of emission sources, most notably mobile sources such as automobiles, stationary point sources such as power plants and oil refineries, and stationary area sources such as residential heating. A sophisticated three-dimensional Eulerian grid model is frequently used to predict the air quality impacts of existing or proposed emission sources, or to assess the effectiveness of emission control measures. Until recently, regional air quality modeling is computer-intensive and requires the use of a mainframe or a supercomputer. Large advances in computer technology in recent years have resulted in inexpensive personal computers and workstations approaching the speed of the supercomputers of a few years ago. This paper describes the development of a state-of-the-art modeling system named SMART (System for Modeling Atmospheric Release and Transport) which is suitable for a wide range of regional air quality impact analyses yet capable of running on these inexpensive computers.
 

2. PHOTOCHEMICAL AIR QUALITY MODELING

Photochemical grid models, also called airshed models or numerical grid models, are the most sophisticated methods available for estimating impacts from pollutant emissions and are frequently used in the development of emission control strategies in areas with severe air quality problems. Photochemical grid models are based on the numerical solution of a set of partial differential equations known as the atmospheric diffusion equations or species continuity equations. These differential equations express in mathematical terms all physical and chemical processes in the atmosphere that affect pollutant, including emissions, transport, diffusion, chemical reactions and pollutant deposition.

Among the photochemical grid models available today, the most widely used is the Urban Airshed Model (UAM). This model is recommended by the U.S. Environmental Protection Agency for ozone modeling in urban areas (U.S. EPA, 1991). This model has been applied for air quality management planning in several U.S. cities that have been designated as non-attainment of the Federal ozone air quality standard (0.12 ppm for one hour). In a typical UAM model application, the region to be simulated is divided up into a three-dimensional grid covering the region of interest. Horizontal grid cells are rectangular with constant length (normally from 2 to 5 kilometers). Four to six layers of varying thicknesses are often employed for vertical transport. Chemical reactions are simulated with the Carbon Bond Mechanism version IV (CBM IV) which contains over 80 reactions and over 30 species. Organic compounds are represented by 11 species including olefin, paraffin, toluene, xylene, formaldehyde, higher aldehydes, ethene, aromatic, methanol, ethanol and isoprene.

Modeling inputs to the UAM model are organized into 13 files. These inputs are derived from the air quality, meteorological and emissions databases which are specific both to the site and the ozone episode selected for modeling. These databases are normally gathered during intensive field studies.

Because the UAM accounts for spatial and temporal variations as well as differences in the reactivity of emissions, it is ideally suited for evaluating the effects of emission control scenarios. This is accomplished by first replicating a historical ozone episode to establish a base case simulation. Model inputs are prepared from observed meteorological, emission and air quality data for the selected modeling days. The model is then applied with these inputs and the results are evaluated to determine its performance. Once the model performance is deemed acceptable, the same meteorological inputs and a projected emission inventory can be used to simulate possible future emission scenarios. That is, the model will calculate hourly ozone patterns likely to occur under the same meteorological conditions as the base case.

Due of its long history of development (since the early 1970s), its widespread use and its approval by regulatory agencies, the UAM model has been selected as the basic photochemical grid model for the SMART modeling system. We have also developed an improved version of the UAM model which allows the use of a modeling grid with variable resolution to minimize emission dilution due to large grid cells and computer resources by employing fewer grid cells (Tran and Mirabella, 1992). It is also worth noting that the UAM model can be used for inert pollutants such as carbon monoxide and sulfur dioxide.
 

3. METEOROLOGICAL MODELING

A key input, which affects the accuracy of a photochemical model such as UAM, is the temporally and spatially-varying wind fields. For use in the simulation of pollutant transport and dispersion, three-dimensional wind fields need to be developed at every grid point of the modeling domain on a hourly basis. The required wind fields are frequently constructed by one of the following techniques: diagnostic wind models in which observed surface and aloft winds are interpolated throughout the modeling domain. Simple constraints (e.g., mass conservation and parameterizations of terrain effects) are then imposed upon these interpolated wind fields; prognostic models which are based on the numerical solution of the governing equations for mass, momentum, energy, and moisture conservation along with the thermodynamic state equations; and, four-dimensional data assimilation (FDDA) techniques which dynamically incorporate actual measurements in the predictions of a prognostic model.

Interpolative and diagnostic techniques are simple to use and only require modest computer resources. Their main disadvantage is the lack of physics in simulating complex phenomena (e.g. land-sea breezes, complex terrain) and their total reliance on input observation data. In areas with sparse or no observations, the accuracy of their predictions is seriously questioned. Pseudo-stations are often employed in these areas to remedy this shortcoming. This study used the Diagnostic Wind Model (DWM) which is part of the SMART modeling system.

Prognostic models have the potential of producing the most accurate wind fields, since they are designed to simulate the relevant physical processes without requiring a substantial amount of observational data. Only the specification of the large-scale flow and the initial state of the atmosphere is required in a model simulation. The disadvantages of prognostic models are that they require substantial computer resources and, more importantly, their predictions are not always consistent with actual observations. The SMART system includes the Colorado State University Mesoscale Model (CSUMM).

The disagreement between the observations and the predictions of a prognostic model remains, however, a serious challenge. A four-dimensional data assimilation (FDDA) technique can be used to minimize the discrepancies between observations and prognostic model outputs. It involves the incorporation of observed data directly into the prognostic model simulation through additional forcing terms in the governing equations (Stauffer and Seaman, 1990). Using the method of Newtonian relaxation or nudging, the prognostic model predictions can be nudged towards either gridded analyses or individual observations during a period of time surrounding the observations. Gridded wind fields for use in the data assimilation can be generated by a diagnostic wind model. We have implemented the FDDA approach in the prognostic CSUMM model, which relies on the diagnostic DWM model for gridded analyses of observed data.
 

4. MODEL PERFORMANCE EVALUATION

The above meteorological and ozone modeling techniques were compared recently for a Southern California Air Quality Study (SCAQS) ozone episode in the Los Angeles air basin (Tran and Mirabella, 1992). In another study, we have applied them to an actual three-day ozone episode which occurred in the coastal areas of the Santa Barbara Channel during the South Central Coast Cooperative Aerometric Monitoring Program (Tran and Murphy, 1993). Wind fields generated by each technique were used by the photochemical grid model Urban Airshed Model (UAM) to predict ozone concentrations. Comparison of predicted and observed ozone concentrations provides a basis for determining the accuracy of the wind modeling techniques.
 

Simulation of SCAQS August 26-28, 1987 Episode

This episode was used in the development of the Air Quality Management Plan for the Los Angeles air basin. The selected modeling domain covers most of the South Coast Air Basin and part of Ventura County and is divided into 65 x 36 squares at 5-km resolution. Since the selected episode occurred during an intensive field program, extensive wind measurements are available for model evaluation: surface winds from 53 sites and upper-air measurements from 14 locations. To provide an objective comparison, these measurements were used directly in the DWM and MMFDDA simulations.

Examination of the vector plots of the predicted wind fields shows that, in general, all models were able to reproduce the gross features of the land-sea breezes. As expected, the diagnostic DWM fields (surface and aloft) are fairly uniform in the mountainous areas where wind measurements are sparse. The prognostic CSUMM model tends to overpredict the observed wind speeds, especially at night and in the early morning hours. CSUMM performed, however, much better than DWM in reproducing the flow patterns observed at the surface and aloft. The MMFDDA model was able to correct the wind speed overprediction of CSUMM.

The UAM ozone simulations started at 1500 PST on August 26, 1987 and ended at midnight August 28, 1987. Table 1 summarizes the performance statistics obtained with various wind fields. For each modeling day, this table shows the ratio of the predicted peak to the observed maximum (paired in space and time), the normalized mean bias and the normalized gross error. Among the wind models tested, the prognostic techniques (CSUMM and MMFDDA) performed best in predicting the ozone peak of 24 pphm observed on 8/27/87 at Redlands. The CSUMM winds result in the worst prediction of the ozone peak of 29 pphm observed on 8/28/87 at Glendora. The best prediction of this peak was obtained with the MMFDDA wind fields. The magnitudes of normalized bias and gross errors are similar for all wind models.
 

5. COMPUTER BENCHMARKS

As part of the SMART system development, the Fortran codes of all the above models have been ported and tested on various MS-DOS personal computers (PCs) and UNIX workstations. Execution times for the CSUMM and UAM tests are reported in Table 2. Test results show that a Pentium-based system offers great improvement in floating point calculations over the 486 processor. The Pentium performance is comparable to that of low-end UNIX workstations and is only a factor 3 to 4 times slower than the fastest workstations. Hence, it is a viable platform for running these models. With an average cost of around $3000, a Pentium PC is many times cheaper than a UNIX workstation. It is also noted that system software such as the Fortran compiler, GIS and graphics packages are also much cheaper for a PC than an UNIX workstation.
 

6. CONCLUSIONS

A state-of-the-art modeling system suitable for regional air quality impact analysis was developed. The SMART system includes the photochemical grid model UAM recommended by regulatory agencies. The system includes improved versions of UAM that allow the use of nested grids and source-specific ozone analysis. Both diagnostic and prognostic models are available for wind field modeling. The prognostic models (CSUMM and MMFDDA), which require only a minimal amount of meteorological measurements, allow the system to be applied to regions with complex flows and sparse monitoring data. With these prognostic models, SMART is capable of emergency forecasting both the meteorology and dispersion of accidental releases of hazardous materials. The system is also capable of running on an inexpensive Pentium PC system.
 

7. REFERENCES

Stauffer, D.R. and N.L. Seaman, 1990: Use of Four-Dimensional Data Assimilation in a Limited-Area Mesoscale Model Part I: Experiments with Synoptic-Scale Data. Monthly Weather Review, vol. 118, pp. 1250-1277.

Tran K.T. and V.A. Mirabella, 1992: Comparison of Modeling Techniques for Generating 3-D Wind Fields Used in Photochemical Grid Modeling. Sixth Conference on Mountain Meteorology, American Meteorological Society, Portland, September 1992.

Tran K.T. and T.M. Murphy, 1993: Wind Field and Photochemical Grid Modeling in the Santa Barbara Channel. 86th Annual Meeting, A&WMA, Denver, Colorado, June 1993.

U.S. Environmental Protection Agency, 1991: Guidelines for Regulatory Application of the Urban Airshed Model. EPA Publication No. EPA-450/4-91-1013. Office of Air Quality Planning and Standards, Research Triangle Park, NC.Table 1. UAM Ozone Performance with Different Wind Fields for the SCAQS 1987 Episode
 

Recent benchmarks on additional computers

Intel Pentium 133               109                      96                    MS-DOS w/ 32 MB RAM
Sun UltraSparc 143             46                      32                    Solaris 2.5.1 w/  96 MB RAM