Abstract

The rapidly growing Cascadia region of the Pacific Northwest, consisting of western Washington, Oregon, and southwestern British Columbia, has experienced surface ozone concentrations that exceed federally mandated standards. A modeling system consisting of the prognostic meteorological model known as the Fifth-Generation Pennsylvania State University–National Center for Atmospheric Research Mesosale Model (MM5), the diagnostic meteorological model CALMET, and the photochemical air quality model CALGRID was developed to investigate ozone formation and transport in this region. To address both the complex topography within the model domain and the relatively sparse network of surface and upper-air meteorological observations, MM5 simulations were performed using 4D data assimilation and a relatively high-resolution inner domain (5-km grid). The MM5 results, however, failed to reproduce the observed wind patterns in some portions of the domain. As a result, it was necessary to employ the MM5 solution as the initial-guess wind field for CALMET (also with a 5-km grid). Objective analysis was applied within CALMET to interpolate the predicted winds with available surface observations. This method involved an iterative approach to find the optimal set of weighting factors within CALMET to merge the MM5 solution with the available meteorological observations.

The predicted ozone concentration patterns for a July 1996 event were very complex but generally showed areas of maximum ozone (130 ppb) occurring to the south and east of Puget Sound and within and to the south of the Portland area (170 ppb). Widespread ozone buildup does not occur over the course of the episode; rather, the maximum ozone concentration occurs each day downwind of each urban center. There was no evidence for recirculation of pollutants from one day to the next within an urban area. It also does not appear that emissions from one urban center influence the neighboring downwind urban area. The predicted ozone concentrations showed good agreement with observations at the monitors located along the Interstate Highway No. 5 corridor. Model performance was less good at three sites located in regions of complex terrain.

Introduction

The Cascadia region of the Pacific Northwest generally is not recognized for photochemical air quality problems. This area, however, which consists of the portions of western Washington, Oregon, and southwestern British Columbia bounded by the Pacific coast to the west and the Cascade range to the east, has a history of one or more ozone episodes occurring almost every summer, with ozone concentrations at or above both the 1-h National Ambient Air Quality Standard (NAAQS) of 120 ppb and the new 8-h NAAQS of 80 ppb. A 25-yr history of summertime ozone exceedances for Washington and Oregon is shown in Fig. 1. Locations along the Interstate Highway No. 5 (I-5) urban corridor, which runs along the western slopes of the Cascades, also have incurred exceedances of the particulate matter (PM) air quality standard (PM10; PM < 10-μm diameter) and existing PM2.5 (PM ⩽ 2.5 μm) data show that concentrations are below but very near the new PM2.5 annual standard. In addition, there is growing concern about the occurrence of regional haze that can obscure vistas of the peaks of the Cascades greatly. Many of these air quality problems may be exacerbated by the continued rapid population growth in the region.

Fig. 1.

Historical record of high ozone concentrations for 1-h and 8-h averages observed at monitoring sites within western Oregon and Washington (year: 70 = 1970).

Fig. 1.

Historical record of high ozone concentrations for 1-h and 8-h averages observed at monitoring sites within western Oregon and Washington (year: 70 = 1970).

To understand ozone formation in the Cascadia region better and to provide a foundation for modeling particulate formation in the future, a mesoscale photochemical grid modeling system has been assembled. This modeling system consists of the Fifth-Generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5) (Grell et al. 1994; Anthes and Warner 1978) and the CALMET–CALGRID modeling package (Yamartino et al. 1992; Scire and Robe 1997; Robe and Scire 1998); the CALMET diagnostic wind model and the CALGRID photochemical model originally were developed for use in California (hence the names). In this system, MM5 can be used to provide a detailed, three-dimensional (3D) simulation of the meteorological conditions that explicitly accounts for complex terrain and other effects. Because CALGRID was designed to use 3D winds from the diagnostic wind module in CALMET, the CALMET code can be used to reformat the MM5 output for use in CALGRID. CALMET also provides estimates of boundary layer parameters such as friction velocity and Monin–Obukhov length that are needed for the diffusion parameterization in CALGRID. If necessary, CALMET also can merge the MM5 winds with available wind observations to produce prognostic wind fields adjusted to take into account the observations.

This modeling system was applied to an ozone episode that occurred 11–14 July 1996 and represents the first attempt to model regional ozone in this rapidly growing area of the United States. Note that both CALMET–CALGRID and the Urban Airshed Model, Variable Grid Version (Douglas et al. 1998) have been used to model ozone concentrations in the Vancouver–Fraser Valley region of British Columbia (Jiang et al. 1996a, 1997; Hedley and Singleton 1997; Hedley et al. 1997). This area is included within the Cascadia modeling domain.

The Cascadia region is different from other parts of the United States for which ozone and secondary particulates are air quality concerns. Prevailing westerly winds sweep relatively clean Pacific air across the region; dense, coniferous vegetation contributes biogenic terpene emissions over a large area; a long north–south urban corridor links large metropolitan centers; and steep, rugged terrain bounds the eastern downwind edge. These features pose questions that must be answered to understand ozone formation and transport within the region. First, how important is the inflow of relatively clean air into the region in terms of minimizing ozone formation? Second, what is the role of widespread terpene emissions from the coniferous landscape in comparison with the relatively sparse but locally higher isoprene emissions? In turn, how important are biogenic emissions to rural ozone formation downwind of the city centers? Third, are the three urban centers interlinked so that one contributes to ozone downwind of a second? Fourth, the regional emission inventory is dominated by mobile sources, with relatively few major point sources. How does this feature affect ozone formation in comparison with other parts of the country in which industrial sources may be a more important part of the emission inventory? In turn, what control strategies will be most effective in maintaining the area as an attainment region in the face of continued urbanization and population growth? Last, how does the complex terrain affect regional flow patterns and the corresponding formation of ozone at high, pristine elevations in the rugged peaks of the Cascades? The latter question is of particular importance in view of the number of class I (a legal designation that affects regulation of adjacent emission sources) wilderness areas and national parks in the region.

All of these questions are not answered in this paper, but the basis for pursuing answers is outlined in terms of a photochemical modeling system adapted for application to the Cascadia region. The purpose of this paper is to present results of a modeling investigation of ozone formation in this area. The modeling system is described in section 2, a description of the 11–14 July 1996 ozone episode is given in section 3, and an evaluation of the model performance is presented in section 4. The paper concludes with a summary in section 5.

Development of the Cascadia modeling system

Several features of the Cascadia region create significant challenges for simulating the meteorological conditions and air quality accurately. The area is influenced strongly by mesoscale circulations that are generated or influenced by orographic effects and land–water interactions (Mass 1982; Chien et al. 1997; Chien and Mass 1997; Bond et al. 1996; Mass and Bond 1996). In addition, the network of surface and upper-air observational sites is relatively sparse. This lack of observational data, coupled with the complex topography, can strain the capabilities of a diagnostic wind model such as CALMET. To address these problems, results from the prognostic MM5 model operated in a nonhydrostatic mode can be used to account explicitly for topographical effects and to act as a supplement to the sparse meteorological network. Even with MM5 results, however, the simulated meteorological behavior can be imperfect. As a result, modeling the meteorological conditions and air quality of the July 1996 ozone episode has been an iterative process. The end result of this process was that the MM5 solution was merged with available wind observations through CALMET to yield wind fields that showed good agreement with observations and that also led to model results in good agreement with available ozone observations. The results of this iterative process are presented in the remainder of the paper.

Meteorological modeling—MM5

MM5 (Grell et al. 1994) was applied to the Cascadia region to provide an initial hourly wind field for CALMET. MM5 is a nonhydrostatic, primitive-equation, mesoscale model.

The MM5 simulations employed a nested grid system (shown in Fig. 2). Weather systems in the Cascadia region usually develop over the Pacific Ocean and migrate eastward or are dominated by high pressure from the north and east. To accommodate these synoptic features, the mother domain D1 extends from the central Pacific to the central United States and from Alaska to Mexico. The D1 domain has 100 × 76 grid points with a horizontal resolution of 45 km. The intermediate domain D2 covers Washington, Oregon, and Idaho, along with the southern portions of British Columbia and Alberta, and contains 76 × 76 grid points with a horizontal resolution of 15 km. The innermost domain D3 encompasses northwestern Oregon, western Washington, and southwestern British Columbia and has 148 × 76 grid points with a horizontal resolution of 5 km. The domain was specified to cover the Pacific coastline to the west and extend beyond the crest of the Cascades to the east, and contains the metropolitan areas of Vancouver, British Columbia, in the north; Seattle; and Portland toward the south. In the vertical dimension, 32 sigma layers were specified, with the first 8 layers existing within 1 km above ground level (AGL) and the top of the model domain extending to 100 hPa.

Fig. 2.

Three nested modeling domains used in MM5. The individual horizontal grid sizes for domains D1, D2, and D3 were 45 km, 15 km, and 5 km, respectively.

Fig. 2.

Three nested modeling domains used in MM5. The individual horizontal grid sizes for domains D1, D2, and D3 were 45 km, 15 km, and 5 km, respectively.

High-resolution, 30–arc second terrain and land use data, originally compiled by the U.S. Geological Survey (USGS) and then reformatted for use with the MM5 preprocessors, were used. This dataset was refined to account for specific northwestern land use classes as part of the MM5 forecast operations at the University of Washington (Mass and Kuo 1998).

One-way nesting was applied to the MM5 domains. The medium-range forecast (MRF) planetary boundary layer (PBL) scheme (Hong and Pan 1996), a multilayer soil temperature model, and the Kain–Fritsch cumulus parameterization scheme (Kain and Fritsch 1993) were employed in the MM5 simulation. The MRF PBL scheme is suitable for a high-resolution description of the boundary layer. The five-layer soil model responds to timescales less than the diurnal cycle to allow for rapid surface temperature changes. The Kain–Fritsch cumulus parameterization is applied during convectively unstable conditions when sufficient vertical motion and cloud depths prevail.

Four-dimensional data assimilation (FDDA) was applied to the mother domain. This technique applies Newtonian relaxation, or nudging, to the solution by adding artificial tendency terms to one or more of the prognostic equations to relax the model toward the gridded analysis or observations. Specifically, gridded analysis nudging (as opposed to observational nudging) was applied to the outer domain. Analysis nudging involves relaxing the entire solution toward gridded analyses that are based on synoptic observations from the 2.5°-resolution National Centers for Environmental Prediction Eta Model analysis and differs from observational nudging, which nudges the solution toward individual observations in the vicinity of those observations. The nudging coefficient was 2.5 × 10−4. Stauffer and Seaman (1994) conducted a study on the best application of FDDA methods and found that analysis nudging on the mesoalpha scale coupled with observational nudging on the finest scale offered the best simulation solution. Future MM5 simulations will incorporate both gridded analysis nudging and observational nudging.

The MM5 simulation covered 132 h from 1200 UTC on Wednesday 10 July 1996 through 0000 UTC on Monday 15 July 1996. The initial 12 h served as a model spinup time with only the mother domain running; the D2 and D3 domains were initiated together after the initial 12-h spinup period.

Meteorological modeling—CALMET

The CALMET–CALGRID model domain lies within the D3 MM5 domain and extends 370 km east–west by 660 km north–south, as shown in Fig. 3. The horizontal grid resolution is 5 km, resulting in a 74 × 132 grid mesh. Ten vertical layers, extending from the surface to 5000 m above the surface, were used (layer heights:20, 80, 160, 260, 410, 660, 1200, 2200, 3600, and 5000 m AGL). The topography within the domain is complex and includes a long and rugged coastline, the rolling extent of the I-5 valley, the heights of the Cascade and Olympic Mountains, and the Columbia River Gorge, as shown in Fig. 4. The terrain and land use data described above for the 5-km MM5 simulation also were employed in CALMET.

Fig. 3.

Map of the CALMET–CALGRID modeling domain with 5-km grid resolution. Surface ozone monitors are represented with a two letter name (e.g., “LS”), surface meteorological sites are displayed as squares, and rawinsonde sites are represented with open circles. The interstate highway system is shown as a dotted line.

Fig. 3.

Map of the CALMET–CALGRID modeling domain with 5-km grid resolution. Surface ozone monitors are represented with a two letter name (e.g., “LS”), surface meteorological sites are displayed as squares, and rawinsonde sites are represented with open circles. The interstate highway system is shown as a dotted line.

Fig. 4.

Terrain map of the Cascadia region with 100-m contour intervals. Vertical scale is exaggerated.

Fig. 4.

Terrain map of the Cascadia region with 100-m contour intervals. Vertical scale is exaggerated.

There are approximately 30 surface meteorological stations, primarily clustered along the I-5 corridor, and two upper-air stations within the inner domain. One upper-air station is located at Quillayute on Washington’s Olympic Peninsula, and the second is located in Salem, Oregon. Data from the surface meteorological sites were collected hourly, and rawinsondes from the two upper-air sites were dispatched daily at 1600 LST and 0400 LST (0000 UTC the next day and 1200 UTC, respectively).

CALMET originally was developed as part of an air quality modeling system that included CALGRID and CALPUFF, a puff diffusion model (Scire et al. 1995). CALMET consists of two components: a diagnostic wind module that generates a description of the surface and upper-level winds and a micrometeorological module that calculates various components of overland and overwater boundary layers (e.g., surface friction velocity, mixing depth, Pasquill–Gifford–Turner stability class). CALMET employs a terrain-following coordinate system and develops the wind field using a three-step process. First, an “initial-guess” wind field is created either by applying upper-air observations to the entire domain or through the introduction of prognostic model data, for example, an MM5 solution. Next, a“step-1” wind field is calculated by adjusting the initial-guess wind field for kinematic effects of terrain, slope flows, and terrain blocking. Last, observational data are introduced through an objective analysis procedure to create a “step-2” wind field. In the current applications, the CALMET objective analysis was used to interpolate the simulated MM5 wind fields with available observations; the relative weighting factor of the observations versus the step-1 wind field was specified by the user.

Two methods were used to introduce the 5-km MM5 wind field into CALMET. The first method treated the MM5 data as “pseudosoundings.” One pseudosounding was available for each CALMET grid point. The real observations were withheld from the solution and used for model validation. This approach resulted in CALMET merely passing through the MM5 wind field without modification except to format the data for use in CALGRID. The second method assigned the MM5 wind field as the CALMET initial-guess wind field. Again, one MM5 value was assigned to each CALMET node. The terrain corrections associated with the step-1 wind field were not invoked. The CALMET objective analysis was used to merge the predicted wind field with the observations. A small subset of the surface meteorological sites was withheld from the objective analysis process to be used for validation of the surface wind field. Note that, in either case, the MM5 data input into CALMET directly influenced the wind field prediction and had only an indirect effect on the CALMET boundary layer module.

Although it originally was intended to operate CALMET using the first method described above (a pass-through of the MM5 wind field without alteration), it became apparent that applying the objective analysis feature would be necessary in several portions of the domain. As a result, a series of CALMET simulations was conducted to obtain an optimal wind field; comparison with surface wind observations and surface ozone observations was used to judge the suitability of each iterative solution. Each CALMET iteration consisted of modulating the relative influence assigned to the surface observations (e.g., through parameters that regulate the radius of influence of the observation or extrapolation of surface winds to layers aloft).

Although this approach yields good model performance in terms of ozone predictions, the empirical nature of the iterative method makes it unsatisfactory. In a separate paper, results from a series of MM5 simulations aimed at identifying a more rigorous method for developing wind fields in complex terrain will be presented. The areas of emphasis include the use of a still-finer grid resolution (1 km), the application of a different PBL scheme, and the implementation of observational nudging in the inner MM5 domain.

Photochemical air quality modeling—CALGRID

The CALGRID photochemical air quality model, developed by Yamartino et al. (1992), has been used in numerous studies (Hedley and Singleton 1997; Hedley et al. 1997; Pilinis et al. 1993; Kumar et al. 1994). It currently is being used as the basis for a State Implementation Plan in the New England region (Schulman and Moore 1998). CALGRID employs several advanced features, including an advection scheme that prohibits negative concentrations and exhibits low numerical diffusion, modern boundary layer representations, a resistance-based dry deposition model for gases and particles, and the Statewide Air Pollution Research Center of California 1990 (SAPRC90) chemical kinetic mechanism. The SAPRC90 mechanism contains 54 chemical species and 129 reactions (Carter 1990). A lumped volatile organic compound (VOC) method is used to represent a class of VOC with similar structure and reactivity, as shown in Table 1. Jiang and coworkers developed a VOC emissions profile for southwestern British Columbia (Jiang et al. 1996b) and used this profile to develop a region-specific set of kinetic and mechanistic parameters within the SAPRC90 system. Comparison of the emissions profile for the Cascadia domain with the emissions in British Columbia showed similar distributions of VOC species. Hence, the region-specific kinetic mechanism developed by Jiang et al. (1996b) was employed in this application of CALGRID for the Cascadia region.

Table 1.

Explicit and lumped VOC included in the SAPRC90 chemistry mechanism.

Explicit and lumped VOC included in the SAPRC90 chemistry mechanism.
Explicit and lumped VOC included in the SAPRC90 chemistry mechanism.

In Washington, there are nine surface ozone monitors, located primarily along the I-5 corridor, and there are three monitors in Oregon in the vicinity of Portland (shown in Fig. 3). Ambient ozone concentrations are recorded hourly at these sites. During the summers of 1995–97, monitoring studies were undertaken to obtain speciated VOC ambient observations at as many as 10 locations within the region (Westberg et al. 1997). Tethered-balloon profile samples of VOC also were collected during two weeks in August 1996 at a location south of Tacoma, Washington. In 1996, only one nitrogen oxide (NOx) monitor, located in central Seattle, was in operation. In 1997 and 1998, the number of NOx monitors was increased so that both ozone and NOx were monitored at most of the existing ozone sites.

Ambient VOC measurements collected on the Olympic Peninsula, along with trace gas concentrations from the literature, were used to calculate the initial and boundary concentrations for CALGRID; these data are listed in Table 2. The concentrations shown in Table 2 generally are low and are representative of clean, background air. Thus, it is reasonable to assume these boundary concentrations can be applied along each edge of the modeling domain, which is bounded by the Pacific Ocean to the west and by sparsely populated regions along each of the inland sides. In addition, the domain’s horizontal extent was enlarged relative to the initial simulations (from a 60 × 114 to a 74 × 132 grid) to reduce boundary effects further. To minimize the influence of initial conditions, the first day of the simulation was considered to be a spinup period for the model.

Table 2.

Initial and boundary concentrations for CALGRID.

Initial and boundary concentrations for CALGRID.
Initial and boundary concentrations for CALGRID.

Emission inventory development

A detailed emission inventory was developed for the July 1996 ozone episode. The inventory consists of both area-source and point-source emissions. Twenty-one model species are specified in the emission inventory: nitric oxide (NO), nitrogen dioxide (NO2), sulfur dioxide (SO2), and carbon monoxide (CO), plus explicit and lumped VOC. On-road mobile source emission rates for Washington and Oregon were calculated using the U.S. Environmental Protection Agency (EPA) MOBILE5b model (U.S. EPA 1996) in combination with state Department of Transportation road-segment link and traffic density data. Other anthropogenic area emissions were calculated using a population surrogate and an appropriate activity factor taken from EPA source activity guidelines. Point-source emission rates either were determined directly using continuous emission monitoring data or were estimated using facility throughput or annual emissions data. A similar methodology was used to develop the anthropogenic area-and point-source emissions for southwestern British Columbia; emissions data for this area were provided by the National Research Council of Canada (Jiang et al. 1996c). Biogenic VOC emissions were estimated using a modified version of the Biogenic Emission Inventory System (Lamb et al. 1997), which incorporates new tree inventory data from the U.S. Forest Service (C. Geron 1997, personal communication) and a simple forest canopy model (Lamb et al. 1993).

A map of the anthropogenic VOC emissions from area and mobile sources is shown in Fig. 5a for 0800 LST on Friday 12 July 1996. The emission map reflects the distribution of roadways and population centers within the urban portions of the domain. Maximum VOC emissions equaled approximately 0.30 metric tons per hour per grid cell (each grid cell measures 5 km × 5 km), and maximum NOx emissions (not shown) equaled 0.12 metric tons per hour per grid cell. A similar map for the biogenic emissions of VOC for a midafternoon period (when biogenic emissions are greatest) is shown in Fig. 5b. In this case, the biogenic VOC emissions are distributed more evenly throughout the domain in comparison with the anthropogenic emissions. Higher emissions are evident in the southern portion of the domain; this pattern is due to the distribution of vegetation species and the warmer temperatures that predominate in this area. Maximum calculated biogenic VOC emissions were approximately 1.3 metric tons per hour per grid cell, and maximum NOx emissions from soils were very small, at less than 0.002 metric tons per hour per grid cell.

Fig. 5.

Spatial distribution of (a) area and mobile VOC emissions (metric tons per hour per grid cell, each grid cell measures 5 km × 5 km) from anthropogenic sources at 0800 LST on Friday 12 Jul 1996 and (b) biogenic VOC emissions (same units) at 1400 LST.

Fig. 5.

Spatial distribution of (a) area and mobile VOC emissions (metric tons per hour per grid cell, each grid cell measures 5 km × 5 km) from anthropogenic sources at 0800 LST on Friday 12 Jul 1996 and (b) biogenic VOC emissions (same units) at 1400 LST.

In Fig. 6, the hourly anthropogenic emission inventory summed over the entire domain shows distinct early-morning and late-afternoon peaks caused by rush-hour traffic during the weekdays, but these peaks are missing on the weekends. Daytime VOC emissions are higher on the weekends in comparison with the weekdays, and NOx emissions are reduced on the weekends when compared with the weekdays. The increased emissions of VOC on the weekend are due to increased recreational boating and use of lawn or garden equipment. The decrease in NOx emissions on the weekend is due to the absence of heavy-duty diesel, commercial, industrial, and construction equipment. A summary of the average daily emissions summed over the domain during the 4-day period is given in Table 3.

Fig. 6.

Time series of anthropogenic CO, VOC, and NOx emissions (metric tons h−1), summed over the entire domain.

Fig. 6.

Time series of anthropogenic CO, VOC, and NOx emissions (metric tons h−1), summed over the entire domain.

Table 3.

Summary of anthropogenic and biogenic emissions (metric tons per day) summed over the entire modeling domain.

Summary of anthropogenic and biogenic emissions (metric tons per day) summed over the entire modeling domain.
Summary of anthropogenic and biogenic emissions (metric tons per day) summed over the entire modeling domain.

The main features of the emission inventory include 1) a relatively small contribution of point sources to the total anthropogenic VOC emissions (8%); 2) a slightly higher contribution of area VOC emissions (51%) to the total anthropogenic emissions in comparison with mobile VOC emissions (41%); 3) an approximately 18% contribution of NOx from point sources relative to the total anthropogenic NOx emissions; 4) a much higher contribution of mobile NOx sources (70%) to the total anthropogenic NOx emissions in comparison with area NOx sources (12%); and 5) biogenic VOC emissions that account for 85% of the domainwide total VOC emissions, as compared with biogenic NOx emissions that account for only 10% of the domainwide total NOx emissions.

In the biogenic VOC emission inventory, the lumped species OLE3, which includes isoprene and terpenes, accounts for 70% of the biogenic total VOC emissions. The remainder is distributed among oxygenated VOC (25%), other olefins (4%), and alkanes (1%). It is important to understand that, although biogenic VOC dominate the total VOC emission inventory on a domainwide basis, there are two aspects of biogenic VOC that reduce their importance in urban ozone production: 1) anthropogenic VOC emissions occur in relatively concentrated areas in comparison with the very dispersed biogenic VOC emissions, and 2) the biogenic VOC emissions are not accompanied by any significant NOx emissions, but anthropogenic VOC emissions occur in combination with comparable NOx emissions. For this region, very high VOC:NOx ratios that are not conducive to ozone formation occur in the unpopulated, heavily vegetated regions along the coast and in the Cascades. In contrast, moderate VOC:NOx ratios (on the order of 4–10), which may be optimal for ozone production, occur throughout each urban area, with a few grids in the center of each urban area that exhibit ratios near 1.

The 11–14 July 1996 ozone episode

Four days of the July 1996 ozone episode were simulated: Thursday 11 July through Sunday 14 July 1996. Results are presented below for the modeled meteorological conditions and air quality. Results from all 4 days (including the Thursday spinup day) are presented in the time series, but only results from the last 3 days are used in the model performance evaluation.

Meteorological conditions

The meteorological conditions during this ozone episode were characterized by sunny conditions with above-normal temperatures. The synoptic conditions were quite similar to the patterns identified by Steenburgh and Mass (1996), who analyzed a long record of high ozone occurrences in Washington State. McKendry (1994) similarly classified the dominant synoptic conditions conducive to high ozone concentrations in the Vancouver, British Columbia, area. Pryor et al. (1995) also came to similar conclusions about the importance of selected classes of synoptic conditions. In these different studies, the authors all showed that ozone episodes occurred with the building of an upper-level ridge of high pressure over the west coast of North America and a thermal trough that developed from California northward along the Pacific coast. This pattern was evident during the July 1996 event; thus, it can be assumed that this event is representative, on a synoptic scale, of the conditions typically associated with high surface ozone concentrations in the Cascadia region.

Under these conditions, the flow pattern during midafternoon exhibits northwesterly winds through the Strait of Georgia, as shown in Fig. 7. After reaching Puget Sound, the flow veers to the south and follows the I-5 corridor. Near Portland, some easterly flow moving through the Columbia Gorge is apparent. Northerly and northwesterly flow is evident off the coast of Washington and Oregon. Typical wind speeds range from 3 to 5 m s−1 during the afternoon. During the evening and early-morning periods, wind speeds are lower, and flow patterns are less organized.

Fig. 7.

Surface wind patterns from MM5–CALMET for 1600 LST on Sunday 14 Jul 1996. Observed surface wind vectors are shown as heavy arrows. Circled wind vectors indicate sites withheld from the CALMET solution and used for independent model evaluation. Shading represents terrain height (m).

Fig. 7.

Surface wind patterns from MM5–CALMET for 1600 LST on Sunday 14 Jul 1996. Observed surface wind vectors are shown as heavy arrows. Circled wind vectors indicate sites withheld from the CALMET solution and used for independent model evaluation. Shading represents terrain height (m).

Midafternoon surface temperature and mixing height contours from CALMET are shown in Fig. 8. Surface temperatures range from 25°C in the northern portion of the domain to 35°C within the southern interior (Fig. 8a). Nighttime surface temperatures drop to 12°–15°C. Mixing height contours (Fig. 8b) display a sharp gradient along the coast and Puget Sound. Mixing heights off the coast are below 200 m and rise steeply just inland to approximately 1100 m. Note that, initially, mixing heights calculated in the CALMET boundary layer module overestimated the mixing heights inferred from the afternoon upper-air soundings at Quillayute and Salem. As a result, the maximum mixing height in CALMET was set to 1200 m throughout the simulation period.

Fig. 8.

(a) Surface temperature contours (°C) and (b) estimated mixing height contours (m AGL) from CALMET for 1600 LST on Sunday 14 Jul 1996.

Fig. 8.

(a) Surface temperature contours (°C) and (b) estimated mixing height contours (m AGL) from CALMET for 1600 LST on Sunday 14 Jul 1996.

Air quality

The evolution of ozone concentration on 14 July 1996 is presented in Fig. 9. The ozone patterns predicted for the other 3 days were similar and for brevity are not presented here. The highest predicted ozone concentrations occurred south of Portland during the late afternoon of Sunday 14 July 1996, with concentrations that exceeded 170 ppb. Simulated surface ozone contours for this day are shown in Fig. 9. At 0600 LST (Fig. 9a), ozone concentrations were low uniformly throughout the domain and generally ranged between 15 and 30 ppb. At 1000 LST (Fig. 9b), however, ozone production was evident downwind of the Vancouver, British Columbia; Seattle; and Portland urban areas and along the I-5 corridor. Simulated concentrations exceeded 70 ppb south of Puget Sound and south of Portland. Maximum ozone concentrations were simulated to occur during the midafternoon hours. At 1400 LST (Fig. 9c), ozone concentrations above 70 ppb were simulated downwind of Vancouver, but ozone concentrations reached 120 ppb south of Puget Sound and immediately south of Portland. At 1800 LST (Fig. 9d), maximum ozone concentrations remained above 130 ppb south of Puget Sound and increased to more than 170 ppb south of Portland. In the evening and throughout the night (not shown), NO scavenging of surface ozone was evident in the urban areas where surface concentrations decreased to less than 30 ppb.

Fig. 9.

Surface ozone concentration contours (ppb) simulated with CALGRID for Sunday 14 Jul 1996 at (a) 0600, (b) 1000, (c) 1400, and (d) 1800 LST.

Fig. 9.

Surface ozone concentration contours (ppb) simulated with CALGRID for Sunday 14 Jul 1996 at (a) 0600, (b) 1000, (c) 1400, and (d) 1800 LST.

Surface ozone time series for eight of the monitoring sites are shown in Fig. 10 for both observed and predicted concentrations (a discussion of model performance for these time series is given in the following section). The first five sites are situated roughly in a line running from the northern portion of the Cascadia domain to the southern portion, and the last three sites are located in the Portland area.

Fig. 10.

Observed and simulated surface ozone concentrations (ppb) at eight monitoring stations: (a) Custer, (b) Getchel, (c) Lake Sammamish, (d) Enumclaw, (e) Paradise, (f) Mountain View, (g) Milwaukee, and (h) Carus.

Fig. 10.

Observed and simulated surface ozone concentrations (ppb) at eight monitoring stations: (a) Custer, (b) Getchel, (c) Lake Sammamish, (d) Enumclaw, (e) Paradise, (f) Mountain View, (g) Milwaukee, and (h) Carus.

The time series of the three sites between Vancouver and Seattle [Custer (CU), Darrington (DA), and Getchel (GE)] show moderately increased ozone concentrations;the Custer and Getchel time series are shown in Figs. 10a and 10b, respectively. Peak concentrations at each of the sites are approximately 50–60 ppb, with nighttime minimums equal to zero. In addition, there is no apparent increase in peak ozone values from the beginning of the episode to the end. All of these sites are in rural settings, and these monitors evidently are not impacted by ozone formed downwind of Vancouver.

In the Puget Sound area, the pattern is very different. At Lake Sammamish (LS) (Fig. 10c), located near the eastern edge of the Seattle urban area, observed ozone concentrations reached 60 ppb on Thursday, exceeded 80 ppb on both Friday and Saturday, and climbed to 100 ppb on Sunday. Nighttime concentrations decreased to zero. The highest observed ozone concentration in the Puget Sound area equaled 118 ppb at Enumclaw (EW) (Fig. 10d), a rural site located in the foothills of the Cascade Mountains approximately 40 km southeast of Seattle. Midday concentrations at Enumclaw increased daily throughout the simulation, and nighttime concentrations remain higher by between 15 and 25 ppb. The fact that nighttime concentrations at the northern sites and at Lake Sammamish decreased to zero can be attributed to the nearby presence of nighttime NO sources such as the I-5 freeway that can lead to rapid titration of atmospheric ozone. In contrast, Enumclaw is well removed from any immediate nighttime sources, and ozone levels remained high throughout the night.

The Paradise monitor (PA) (Fig. 10e), located on the southern flank of Mt. Rainier, is unique for the fact that it resides at an elevation of 1650 m above mean sea level (MSL). The observed ozone pattern here is different in comparison with that of the other sites within the Cascadia domain. Ozone concentrations showed little diurnal variation, and concentrations generally ranged between 40 and 70 ppb with only a weak late-afternoon maximum apparent. This pattern has been observed by other researchers at high-elevation monitors. For example, Logan (1989) reported ozone concentrations that ranged between 40 and 50 ppb at a Whiteface Mountain site (1500 m MSL) in New York, with little diurnal variation evident. Logan attributed this behavior to the fact that high-elevation sites are located above the nocturnal inversion, which isolates them from the effects of surface deposition. Aneja et al. (1991) also recorded similar ozone patterns at two high-elevation sites located at Mt. Mitchell State Park in North Carolina. At the first site (2006 m MSL), ozone concentrations ranged between approximately 50 and 60 ppb;at the second site (1760 m MSL), observed ozone concentrations were between approximately 40 and 50 ppb. Note that both Logan and Aneja et al. reported the occurrence of the daily ozone maximum as occurring near midnight; at the Paradise monitor, peak ozone concentrations were observed between 1800 and 2100 LST. At Pack Forest (PF) and Packwood (PW; not shown), which are in relatively remote locations near Mt. Rainier, the ozone pattern exhibited daytime peaks of less than 70 ppb and nighttime concentrations that remained above 20 ppb. Although some data were missing at Pack Forest, there is no evidence of an increase in ozone from one day to the next at the Pack Forest or Packwood sites.

In the southern section of the domain, all four of the sites exhibited increased ozone concentrations on Saturday and Sunday. At Sauvie Island (SI), the maximum observed value on Saturday equaled 94 ppb, which was slightly larger than the maximum recorded on Sunday. Mountain View (MV; Fig. 10f) is located in a busy suburban section immediately north of Portland and exhibited maximum ozone concentrations of 90 ppb on Saturday and 112 ppb on Sunday, and nighttime concentrations were close to zero. At the Milwaukee monitor (MH), located in a suburb south of Portland, observed ozone concentrations were much higher on Sunday in comparison with the previous three days; the highest ozone concentration recorded in the Cascadia domain was measured at this site, with a peak of 145 ppb observed on Sunday afternoon. At Carus (CA), which is a rural monitor located farther south, maximum ozone concentrations of 124 and 108 ppb were observed on Saturday and Sunday, respectively. At the Milwaukee monitor, the nighttime ozone concentrations were very close to zero, and at Carus, which is farther removed from urban emissions, the nighttime ozone concentrations remained high and show an increasing trend during the episode.

The simulated ozone concentration contours and time series clearly show that ozone patterns are complex within the Cascadia region. There is not a widespread buildup of ozone throughout the domain, but rather ozone concentrations tend to increase during the episode in a similar manner at stations located downwind of the urban centers. Daytime simulated maximum concentrations downwind of Puget Sound on Sunday reached almost 120 ppb at the monitoring sites and exceeded 130 ppb at other downwind locations. At the Milwaukee monitor near Portland, the maximum simulated ozone concentration exceeded 170 ppb on Sunday afternoon. It appears that there is little influence of one urban center upon the downwind neighboring urban center. It also appears that ozone formed on one day is not recirculated within the area through the night; rather, a predominate northerly flow along the I-5 corridor tends to push air from the north to the south. The increase in ozone from one day to the next is driven primarily by increasing temperatures during the episode.

At sites well removed from an urban influence, such as Darrington and Packwood, concentrations do not increase from one day to the next, and the effects of an ozone episode are not apparent at these monitors. At locations near urban sources, nighttime ozone concentrations decrease to near zero, and, at other stations away from urban areas, nighttime ozone concentrations remain in the 20 ppb range. At Paradise, the only high-elevation monitoring site, the ozone pattern exhibited only a weak diurnal cycle but did show a steady increase from 45 to 60 ppb over the 4-day period.

Model performance evaluation

The performance of CALGRID and CALMET was evaluated using a set of statistics described by Lu et al. (1997), Hedley and Singleton (1997), and Hedley et al. (1997). The evaluation of CALGRID’s performance relative to the hourly surface ozone measurements recorded at the 12 monitors is presented first. Model performance statistics include the observed and predicted mean and standard deviation of ozone concentration, the slope and intercept of a linear regression applied to the predicted versus observed ozone concentration, the root-mean-square difference rmsd, the index of agreement I, the normalized gross error E, and the normalized bias B. The rmsd is defined as

 
formula

where pi and oi are the simulated and observed ozone concentrations, respectively, for measurement i. The total number of measurements is N. The skill level of the model is regarded as high if the rmsd is smaller than the standard deviation of the observation and if the standard deviation of the model output is comparable with that of the observations (Lu et al. 1997). The index of agreement is defined as

 
formula

where o denotes the average observed ozone concentration. A model index value of 1 indicates perfect agreement between predicted and observed concentrations, and an index of 0 indicates no agreement at all. The normalized gross error is defined as

 
formula

and the normalized bias is calculated as

 
formula

In addition to these statistics based upon observed and simulated concentrations paired in space and time, the maximum daily observed and simulated concentrations (paired in space but unpaired in time) also are of interest.

No effort was made to remove low ozone concentrations from the statistical evaluation; that is, both the nighttime and daytime performance are considered in the overall model assessment. The statistics are based upon the last 3 days of the simulation (the first day was regarded as a spinup day for the model) and are summarized in Table 4. Time series of observed and predicted ozone for 8 of the 12 monitors are shown in Fig. 10.

Table 4.

Performance statistics for the 12 surface ozone (O3) monitors shown in Fig. 3.

Performance statistics for the 12 surface ozone (O3) monitors shown in Fig. 3.
Performance statistics for the 12 surface ozone (O3) monitors shown in Fig. 3.

The three monitors located near the I-5 corridor in the northern portion of the domain (Custer, Getchel, and Lake Sammamish) show high indices of agreement, exceeding 0.8. These three sites also have comparable values for the simulated and observed standard deviations, and the rmsd for all three cases is well below the observed standard deviation. The difference between the average simulated and observed concentrations at Custer can be explained by overestimation of ozone at night at this site. Darrington, another site in the northern portion of the domain, exhibits much worse performance; the model index is 0.52, and there is poor agreement between the simulated and observed average concentration and standard deviation. This poor performance may be attributed to the very complex topography associated with this site, because Darrington is located at the confluence of two steep valleys within the Cascades.

Although the model index at Paradise, on the side of Mt. Rainier, is the lowest of the 12 monitors (I = 0.34), the general shape of the peculiar ozone time series measured at this site is duplicated relatively well by the simulated ozone curve (Fig. 10e), with simulated ozone concentrations showing a persistent offset of approximately 30 ppb relative to the observed concentrations.

Enumclaw, Pack Forest, and Packwood, located southeast of Seattle, have indices of agreement of 0.91, 0.63, and 0.56, respectively. At Enumclaw, the model performance is very good except during the midafternoon period when the peak observed ozone concentration is underestimated. At the Pack Forest and Packwood sites, daytime ozone is underestimated, although nighttime model results are well matched with the observations.

In the Portland area, the agreement between observed and simulated ozone at Sauvie Island is moderate, with I equal to 0.71. This result reflects the underestimation of ozone during most of the simulation. At the other three southern sites, the model performance is considerably better, with indices of agreement that range from 0.85 to 0.92. At these monitors, the model matches the maximum observed concentrations closely, although the timing of the peak is shifted slightly at the Milwaukee and Carus sites.

Overall, the air quality model performance is good at a number of the sites distributed throughout the domain. This good performance is particularly true at sites at which the highest ozone concentrations were observed: Enumclaw, Mountain View, Milwaukee, and Carus. This result is encouraging because it suggests that the emission inventory is not in serious error. Poor performance was evident at Darrington and Paradise, at which large terrain effects may have played a role. This also may be the case for Pack Forest and Packwood, both of which are located in the lower hills of the Cascades near Mt. Rainier. A scatter diagram of simulated versus observed ozone (paired in space and time) for all of the sites is shown in Fig. 11. In general, there is good agreement between simulated and observed values. At low concentrations (25 ppb), there is a tendency for underestimation of observed values. In contrast, the peak observed value is predicted almost perfectly, but there are several points at which the model greatly underestimates the observed ozone concentrations.

Fig. 11.

Simulated vs observed surface ozone concentrations for all monitoring stations during 12–14 Jul 1996; 1:1 line shown for comparison.

Fig. 11.

Simulated vs observed surface ozone concentrations for all monitoring stations during 12–14 Jul 1996; 1:1 line shown for comparison.

In the analysis of the meteorological simulation, four surface stations that were not used in the CALMET simulation were selected for use as an independent test of the meteorological model. The stations were selected in regions for which there were other observational sites located nearby, so that removing the selected stations would not compromise the overall results. Performance statistics for these stations are summarized in Table 5 for wind speed and wind direction and in Table 6 for temperature. The mean simulated wind speeds at these stations were higher than the observed means by 0.5 to 1.3 m s−1, and the standard deviation of the simulated wind speed also was higher than the standard deviation of the observed wind speeds by approximately 0.5 m s−1. The rmsd of the simulated wind speeds varied between 1.3 and 1.8 m s−1, which in each case was larger than the standard deviation of the observed wind speeds. The index of agreement for wind speeds ranged from 0.56 to 0.76, which suggests a moderate level of model performance.

Table 5.

Performance statistics (surface winds) for the four surface meteorological stations withheld from CALMET where ws is wind speed and wd is wind direction.

Performance statistics (surface winds) for the four surface meteorological stations withheld from CALMET where ws is wind speed and wd is wind direction.
Performance statistics (surface winds) for the four surface meteorological stations withheld from CALMET where ws is wind speed and wd is wind direction.
Table 6.

Performance statistics (temperature) for the four surface meteorological stations withheld from CALMET.

Performance statistics (temperature) for the four surface meteorological stations withheld from CALMET.
Performance statistics (temperature) for the four surface meteorological stations withheld from CALMET.

The difference between the accuracy of the model results at these independent sites and the accuracy of the model results at sites used as part of the CALMET solution is shown in Fig. 12. This figure shows a scatter diagram of simulated versus observed wind speed for the independent and dependent test sites. It is obvious that the simulated and observed wind speeds show relatively good agreement for the sites incorporated into the CALMET solution, and there is considerably more scatter for the independent sites. This result suggests that the incorporation of available surface observations is an important aspect in the development of an accurate wind field for air quality modeling.

Fig. 12.

Predicted vs observed surface wind speeds for all meteorological stations during 12–14 Jul 1996. Independent sites withheld from the CALMET solution are shown as circles; dependent sites used in CALMET objective analysis are shown as squares; 1:1 line shown for comparison.

Fig. 12.

Predicted vs observed surface wind speeds for all meteorological stations during 12–14 Jul 1996. Independent sites withheld from the CALMET solution are shown as circles; dependent sites used in CALMET objective analysis are shown as squares; 1:1 line shown for comparison.

The mean simulated wind direction agreed with the mean observed wind direction to within approximately 20°. Because of the circular characteristics of wind direction, the other statistics were not calculated for this parameter. The simulated temperatures agree with the observed mean temperatures to within approximately 1°C at three of the independent sites, but the mean observed and simulated temperatures differ by more than 2°C at the MV site. The standard deviations of simulated temperatures all are larger than the corresponding observational statistic, but the rmsd values all are less than the standard deviations of the observed temperatures. The model index of agreement is greater than 0.9 for each site. The scatter diagram of temperature using all four sites in Fig. 13 shows relatively good agreement between observations and model output, but there is a trend toward underestimation of the warmest observed temperatures.

Fig. 13.

Predicted vs observed surface temperatures for all meteorological stations during 12–14 Jul 1996; 1:1 line shown for comparison.

Fig. 13.

Predicted vs observed surface temperatures for all meteorological stations during 12–14 Jul 1996; 1:1 line shown for comparison.

The performance of the meteorological model also was examined through comparison of the upper-air soundings at Quillayute and Salem with predicted vertical profiles, as shown in Figs. 14 and 15. At Quillayute, on the coast of the Olympic Peninsula, there is very good agreement between the observed and simulated wind speed, wind direction, and temperature profiles. At Salem, located south of Portland, the observed wind speed profile is missing data below 750 m AGL; it appears, however, that the simulated wind speed profile exhibits higher wind speeds than are observed up to approximately 1500 m AGL. Above that level, the simulated wind speed is less than the observed wind speed. The simulated wind direction at Salem shows good agreement with the observed profile, and the same is true for the vertical profile of temperature.

Fig. 14.

Observed (solid line) and predicted (dashed line) upper-air profiles at Quillayute on 1600 LST 14 Jul 1996 for (a) wind speed (m s−1), (b) wind direction (°), and (c) temperature (K).

Fig. 14.

Observed (solid line) and predicted (dashed line) upper-air profiles at Quillayute on 1600 LST 14 Jul 1996 for (a) wind speed (m s−1), (b) wind direction (°), and (c) temperature (K).

Fig. 15.

Same as Fig. 14 but at Salem.

Fig. 15.

Same as Fig. 14 but at Salem.

Summary

In the Cascadia region, ozone concentrations that exceed both the old and new NAAQS (120 ppb for 1-h average and 80 ppb, for 8-h average, respectively) have been observed in recent years. Several features of the Cascadia region are different than those in other areas typically evaluated in ozone modeling studies, including: 1) complex topographic and land–sea features (e.g., Puget Sound, the Cascade Range, the Columbia River Gorge); 2) an anthropogenic emission inventory that consists primarily of mobile sources, with relatively few large industrial point sources; 3) a biogenic emission inventory that is dominated by coniferous forests; and 4) a predominant inflow consisting of relatively clean maritime air.

The Cascadia modeling system represents the first attempt to model tropospheric ozone in the rapidly growing Pacific Northwest region of the United States. This modeling system, which consists of the MM5 prognostic meteorological model, the CALMET diagnostic wind and boundary layer model, and the CALGRID photochemical model, was used to investigate an ozone event in the Cascadia region that occurred on 11–14 July 1996. MM5 was employed with FDDA (gridded analysis) and a finescale grid (5 km) to address both the orographically influenced flows and the sparse network of surface and upper-air observations. It became apparent that the MM5 wind fields, even when generated at a 5-km grid resolution, failed to capture all of the complexities of the wind patterns in some areas of the domain. Thus, it was necessary to use CALMET to merge the MM5 solution with available observations in an iterative approach to obtain acceptable model performance with respect to simulated ozone concentrations. Further work is necessary to identify a more rigorous method for applying MM5 to air quality simulations in this region.

With the iterative MM5–CALMET approach used in this work, the ozone results show relatively good agreement with observations over most of the modeling domain. The model correctly simulates ozone at two sites between Vancouver, British Columbia, and Seattle, and good performance also is obtained at two sites southeast and downwind of Seattle. At a monitor north of Seattle, located at the confluence of two Cascade valleys, the model underestimated observed daytime ozone concentrations but overestimated observed nighttime concentrations. At a site located on the flank of Mt. Rainier at 1650-m elevation, the model results consistently were lower than observed concentrations (approximately 30 ppb less), but the relatively flat diurnal cycle was reproduced correctly. In the Portland area, where maximum ozone concentrations exceeding 145 ppb were observed, ozone predictions generally were in good agreement with observations.

The simulated ozone patterns within the region during this episode indicate that there is little influence of one urban area upon a neighboring downwind urban area. The patterns also indicate that there is not a widespread buildup of ozone over the course of the episode. High ozone concentrations occur downwind of each urban center during each day, but there is little evidence of recirculation of pollutants from one day to the next within an urban area.

Overall, the results of this first application of a regional photochemical modeling system in the Cascadia region of the Pacific Northwest show the complexities of modeling ozone formation in the presence of rugged terrain. Although the meteorological modeling required an iterative approach with constraints from observations in place, the fact that the model results showed good agreement with observed ozone concentrations suggests that the detailed emission inventory is not in serious error. Thus, a good basis for further investigation of photochemical air quality exists. Further work to develop better meteorological methods is needed, and sensitivity runs to develop a sense of model uncertainty should be completed. Finally, simulations of other ozone episodes are required to develop a more complete picture of photochemical air pollution within the Cascadia region.

Acknowledgments

This work was supported by the Washington State Department of Ecology and the Southwest Washington Air Pollution Control Authority. Support also was provided through Boeing endowment funds to Washington State University. The authors want to acknowledge the generous assistance of Dr. Donald Singleton and colleagues at the National Research Council of Canada in the initial stages of this work and Dr. Clifford Mass and coworkers for guidance in the installation and operation of MM5. We would also like to thank Chris Geron and Tom Pierce of U.S. EPA for providing assistance with the revision of the biogenic emission inventory.

REFERENCES

REFERENCES
Aneja, V. P., S. Businger, Z. Li, C. S. Claiborn, and A. Murthy, 1991:Ozone climatology at high elevations in the southern Appalachians. J. Geophys. Res., 96, 1007–1021.
Anthes, R. A., and T. T. Warner, 1978: Development of hydrodynamic models suitable for air pollution and other mesometeorological studies. Mon. Wea. Rev., 106, 1045–1078.
Bond, N. A., C. F. Mass, and J. Overland, 1996: Coastally trapped wind reversals along the United States west coast during the warm season. Part I: Climatology and temporal evolution. Mon. Wea. Rev., 124, 430–445.
Carter, W. P. L., 1990: A detailed mechanism for the gas-phase atmospheric reactions of organic compounds. Atmos. Environ., 24A, 481–510.
Chien, F.-C., and C. F. Mass, 1997: Interaction of a warm-season frontal system with the coastal mountains of the western United States. Part II: Evolution of a Puget Sound convergence zone. Mon. Wea. Rev., 125, 1730–1752.
Chien, F.-C., C. F. Mass, and Y.-H. Kuo, 1997: Interaction of a warm-season frontal system with the coastal mountains of the western United States. Part I: Prefrontal onshore push, coastal ridging, and alongshore southerlies. Mon. Wea. Rev., 125, 1705–1729.
Douglas, S., H. P. Deuel, and J. L. Haney, 1998: Use of the UAM-V modeling system process analysis and source attribution features to assess model performance and response for urban and regional-scale applications. Proc. 10th Joint Conf. on the Applications of Air Pollution Meteorology, Phoenix, AZ, Amer. Meteor. Soc., 400–403.
Grell, G. A., J. Dudhia, and D. R. Stauffer, 1994: A description of the Fifth-Generation Penn State/NCAR Mesoscale Model (MM5). NCAR Tech. Note NCAR/TN-398 + STR, 122 pp. [Available from National Center for Atmospheric Research, P.O. Box 3000, Boulder, CO 80307.]
.
Hedley, M., and D. L. Singleton, 1997: Evaluation of an air quality simulation of the Lower Fraser Valley. Part I: Meteorology. Atmos. Environ., 31, 1605–1616.
Hedley, M., R. McLaren, W. Jiang, and D. L. Singleton, 1997: Evaluation of an air quality simulation of the Lower Fraser Valley. Part II:Photochemistry. Atmos. Environ., 31, 1617–1630.
Hong, S. Y., and H. L. Pan, 1996: Nonlocal boundary layer vertical diffusion in the medium-range forecast model. Mon. Wea. Rev., 124, 2322–2339.
Jiang, W., D. L. Singleton, M. Hedley, and R. McLaren, 1996a: Sensitivity of ozone concentrations to VOC and NOx emissions in the Canadian Lower Fraser Valley. Atmos. Environ., 31, 627–638.
Jiang, W., D. L. Singleton, M. Hedley, and R. McLaren, 1996b: Modification and evaluation of the CALGRID chemical mechanism for ozone studies of the Lower Fraser Valley, British Columbia. Report No. PET-1360-95S, Institute for Chemical Process and Environmental Technology, 81 pp. [Available from National Research Council Canada, Ottawa, ON K1A 0R6, Canada.]
.
Jiang, W., D. L. Singleton, M. Hedley, and R. McLaren, 1996c: Processing the Lower Fraser Valley Pacific 93 emission inventory for UAM-V applications: Area and mobile sources. Report No. PET-1385-96S, Institute for Chemical Process and Environmental Technology, 111 pp. [Available from National Research Council Canada, Ottawa, ON K1A 0R6, Canada.]
.
Jiang, W., D. L. Singleton, M. Hedley, R. McLaren, T. Dann, and D. Wang, 1997: Comparison of organic compound compositions in the emissions inventory and ambient data for the Lower Fraser Valley. J. Air Waste Manage. Assoc., 47, 851–860.
Kain, J. S., and J. M. Fritsch, 1993: Convective parameterization for mesoscale models: The Kain–Fritsch scheme. The Representation of Cumulus Convection in Numerical Models, K. A. Emanuel and D. J. Raymond, Eds., Amer. Meteor. Soc., 246 pp.
Kumar, N., A. G. Russell, T. W. Tesche, and D. E. McNally, 1994: Evaluation of CALGRID using two different ozone episodes and comparison to UAM results. Atmos. Environ., 28, 2823–2845.
Lamb, B., D. Gay, H. Westberg, and T. Pierce, 1993: A biogenic hydrocarbon emission inventory for the U.S.A. using a simple forest canopy model. Atmos. Environ., 27A, 1673–1690.
Lamb, B., B. Hopkins, H. Westberg, and P. Zimmerman, 1997: Evaluation of biogenic emission estimates using ambient VOC concentrations in western Washington. Proc. Workshop on Biogenic Hydrocarbons in the Atmospheric Boundary Layer, Charlottesville, VA, Amer. Meteor. Soc., 53–56.
Logan, J. A., 1989: Ozone in rural areas of the United States. J. Geophys. Res., 94, 8511–8532.
Lu, R., R. P. Turco, and M. Z. Jacobson, 1997: An integrated air pollution modeling system for urban and regional scales. Part II: Simulations for SCAQS 1987. J. Geophys. Res., 102, 6081–6098.
Mass, C. F., 1982: The topographically forced diurnal circulations of western Washington State and their influence on precipitation. Mon. Wea. Rev., 110, 170–183.
Mass, C. F., and N. A. Bond, 1996: Coastally trapped wind reversals along the United States west coast during the warm season. Part II: Synoptic evolution. Mon. Wea. Rev., 124, 446–461.
Mass, C. F., and Y.-H. Kuo, 1998: Regional real-time numerical weather prediction: Current status and future potential. Bull. Amer. Meteor. Soc., 79, 253–263.
McKendry, I. G., 1994: Synoptic circulation and summertime ground-level ozone concentrations at Vancouver, British Columbia. J. Appl. Meteor., 33, 627–641.
Pilinis, C., P. Kassomenos, and G. Kallos, 1993: Modeling of photochemical pollution in Athens, Greece. Application of the RAMS–CALGRID modeling system. Atmos. Environ., 27B, 353–370.
Pryor, S. C., I. G. McKendry, and D. G. Steyn, 1995: Synoptic-scale meteorological variability and surface ozone concentrations in Vancouver, British Columbia. J. Appl. Meteor., 34, 1824–1833.
Robe, F. R., and J. S. Scire, 1998: Combining mesoscale prognostic and diagnostic wind models: A practical approach for air quality applications in complex terrain. Proc. 10th Joint Conf. on the Applications of Air Pollution Meteorology, Phoenix, AZ, Amer. Meteor. Soc., 223–226.
Schulman, L. L., and G. E. Moore, 1998: Photochemical grid modeling of four high ozone episodes in the New England domain:CALGRID vs. UAM-IV. Proc. 10th Joint Conf. on the Applications of Air Pollution Meteorology, Phoenix, AZ, Amer. Meteor. Soc., 505–509.
Scire, J. S., and F. R. Robe, 1997: Fine-scale application of the CALMET meteorological model to a complex terrain site. Paper No. 97-A1313, Proc. 90th Annual Air and Waste Management Meeting, Toronto, Canada, Air and Waste Manage. Assoc.
Scire, J. S., E. M. Insley, R. J. Yamartino, and M. E. Fernau, 1995: A user’s guide for the CALMET meteorological model. Earth Tech, 193 pp. [Available from Earth Tech, Inc., 196 Baker Ave., Concord, MA 01742.]
.
Stauffer, D. R., and N. L. Seaman, 1994: Multiscale four-dimensional data assimilation. J. Appl. Meteor., 33, 416–434.
Steenburgh, W. H., and C. F. Mass, 1996: Synoptic and mesoscale circulations during high ozone episodes over western Washington. Final Report for the Puget Sound Clean Air Agency, 67 pp. [Available from Puget Sound Clean Air Agency, 110 Union St., Seattle, WA 98101.]
.
U.S. EPA, 1996: MOBILE5b Model and user’s guide. Office of Mobile Sources, National Motor Vehicle and Fuels Emission Laboratory, 216 pp. [Available from National Motor Vehicle and Fuels Emission Laboratory, 2565 Plymouth Rd., Ann Arbor, MI 48105.]
.
Westberg, H., B. Hopkins, E. Allwine, B. Lamb, and P. Zimmerman, 1997: Speciated VOC measurements in western Washington during the summer of 1996. Washington State Department of Ecology, 70 pp. [Available from Washington State Department of Ecology, P.O. Box 47600, Olympia, WA 98504-7600.]
.
Yamartino, R. J., J. S. Scire, G. R. Carmichael, and Y.-S. Chang, 1992: The CALGRID mesoscale photochemical grid model. Part I: Model formulation. Atmos. Environ., 26A, 1493–1512.

Footnotes

Corresponding author address: Dr. Brian Lamb, Laboratory for Atmospheric Research, Dept. of Civil and Environmental Engineering, Washington State University, Pullman, WA 99164-2910.