The Community Multiscale Air Quality (CMAQ) model was used to predict O3-NOx-VOC chemistry for the Pacific Northwest and these results were evaluated by comparing to aircraft measurements of CO, NOy, O3, and VOCs collected during the Pacific Northwest field experiment in the summer of 2001 (PNW2001). The evaluation focused on three areas: 1) photochemical indicator values (O3/NOy), 2) accuracy of the emission inventory, and 3) VOC reactivity. The evaluation was performed for two modeling scenarios: a standard scenario and a reduced VOC scenario, which was developed based on the comparisons of measurements with the emission inventory. Results showed that model-predicted O3/NOy ratios were closely related to VOC-NOx sensitive conditions, with transitional values similar to those identified from previous studies. Peak O3 was associated with VOC-sensitive conditions, but these were not far from the transitional regime. The standard modeling scenario over-predicted peak O3 and the O3/NOy slope, indicating an overestimation of sensitivity to NOx, probably due to too much VOC in the emission inventory. The reduced VOC scenario resulted in better agreement with measurements in terms of peak O3 as well as O3/NOy correlations. Comparisons of observed CO and VOC to NOy ratios from the morning urban samples with those from the emission inventory also supported an overestimation of VOC in the standard scenario, with CO concentrations over-predicted by about 80% and the total VOC reactivity over-predicted by 30%. The standard modeling scenario substantially overestimated the reactivity from CO. The reduced VOC scenario showed generally good agreement with observations of the relative contributions to the total VOC reactivity.
 Tropospheric ozone is a major concern for air quality due to its adverse impact on human health [Lippmann, 1993] and ecosystems [National Research Council, 1991]. Ozone remains as the most troublesome of the criteria pollutants: more than 60 millions of people live in areas where the ozone air quality standard was exceeded in 2009 [http://www.epa.gov/airtrends/aqtrends.html]. It is well known that the photochemical production of ozone is a highly nonlinear system: in conditions with relatively high VOC/NOx ratios, production of O3 is limited by NOx availability, so it increases with increasing NOx concentrations and is less sensitive to changes in VOC concentrations; in conditions with relatively low VOC/NOx ratios, production of O3 is limited by radical availability and NOx inhibition, and therefore, O3 increases with increasing VOC concentrations and decreases with increasing NOx concentrations. For urban areas, ozone control strategies have been developed largely based on 3-D air quality model simulations, where model performance is usually judged by comparing observed and measured O3 levels. Due to the complex nonlinearity of ozone-VOC-NOx chemistry, however, predicted peak O3 level is not dependent on a unique set of VOC and NOx concentrations. Therefore, the model might predict incorrect sensitivities to changes in precursors even while correctly predicting O3 levels [Sillman, 1995]. In order to gain more confidence in model predictions and control policies, observation-based methods have been developed in recent years. The methods can be divided into two categories [Sillman, 1999]: those based on ambient levels of primary emissions of VOC and NOx, and those based on ambient levels of secondary reactants such as reactive nitrogen and peroxides.
 For methods involving secondary reaction products, one of the most important applications relates to the usage of photochemical indicators. The indicator concept was proposed by Milford et al. , Kleinman , and Sillman . They found that in 3-D photochemical models, the ratio of certain species would have different values related to NOx- and VOC-sensitive conditions. Based on steady state balance of radical initiation and termination, Sillman  developed indicator ratios including O3/NOy, O3/NOz, H2O2/HNO3, and other similar ratios related to ozone, reactive nitrogen, and peroxides. In comparison to these cumulative indicators, Tonnesen and Dennis [2000a, 2000b] developed local indicators for instantaneous ozone production rate using a slightly different approach based on radical propagation efficiency. Also based on a steady state balance, Kleinman et al.  and Kleinman  found that the sensitivity of instantaneous ozone production rate to NOx concentration, VOC reactivity, and radical production rate depends on the fraction of radicals that are removed by reactions with NOx.
 The general idea is that since the values of these measurable indicators could be related to NOx- or VOC-sensitive conditions, model predicted sensitivity could be evaluated by comparing to measured indicator values or ozone sensitivity could be evaluated directly from measurements. A number of studies have reported that ambient measurements of indicator values generally show expected behavior in various locations [Sillman, 1999]. Values indicating NOx-sensitive conditions were found in Atlanta [Sillman et al., 1995, 1997], and values indicating VOC-sensitive conditions were measured in Los Angeles [Sillman, 1995; Sillman et al., 1997] and Phoenix [Kleinman et al., 2005]. Measured indicator values were also compared with model predictions in Atlanta [Sillman et al., 1997], Los Angeles [Sillman et al., 1997], Nashville [Sillman et al., 1998], Paris [Sillman et al., 2003], and Mexico City [Sillman and West, 2009]. It was found that comparisons involving indicator values are more stringent than standard methods based only on ozone performance, and therefore, the uncertainties in ozone sensitivity from model predictions could be reduced.
 Elevated ozone levels occur periodically in the Puget Sound region, with one or more episodes occurring almost every summer [Barna and Lamb, 2000; Puget Sound Clear Air Agency, 2011]. With the continued rapid population growth and urbanization in the region, ozone concentrations in recent years have been very close to the 2008 8-h National Ambient Air Quality Standard (NAAQS). There is growing concern whether the region will maintain federal attainment status of the ozone NAAQS in the future, as the current 8-h standard might be further strengthened. In addition, there are a number of national parks and wilderness areas located relatively close to the Seattle area, and therefore it is important to understand pollution formation and transport in the region.
 In 2001, a field study (the Pacific Northwest 2001 field campaign; PNW2001) was conducted in August in the Puget Sound region. The goal of this study was to better understand ozone and aerosol formation in the region and to provide data sets for air quality model evaluation [Jobson et al., 2002; Elleman and Covert, 2009]. This is the first aircraft-based study and represents the most comprehensive gaseous and aerosol atmospheric data set for the region. The U.S. Department of Energy Gulfstream-I (G-I) aircraft was used to measure meteorological conditions, gas phase concentrations, and aerosol properties. Aircraft measurements also included morning “touch-and-go” urban profiles taken at Boeing Field in Seattle, during which the plane descended to a few meters above the runway, providing a low altitude sounding of entire depth of the PBL. Since air quality model evaluations for this region have been limited to relatively sparse ground measurements [i.e., Chen et al., 2008], the ambient data from the G-I flights provide a good opportunity for using observation-based methods to evaluate model predicted ozone chemistry and to have a better understanding of ozone formation in the region.
 The objectives of this study are to: 1) evaluate model-predicted sensitivity and ozone chemistry by comparing to observed indicator values (O3/NOy), 2) evaluate the emission inventory using observed compounds versus NOy ratios in the morning urban profile, and 3) compare observed and model-predicted VOC reactivity. Measurements and photochemical modeling are described in section 2. In section 3, we present the results from comparing observed and predicted O3 and NOy levels, an examination of the behavior of modeled O3/NOy, and an evaluation of the model by comparing to observed O3/NOy values. Then from a slightly different perspective, we evaluate the emission inventory and compare measured and modeled VOC reactivity. Finally, we discuss the findings and uncertainties.
2.1. PNW2001 Field Campaign
 Details of the aircraft flights were described by Elleman and Covert . Briefly, the U.S. Department of Energy G-I aircraft made five flights during August 2001: 20 August (Monday), 26 August (Sunday), and 27 August (Monday). Both morning and afternoon flights were conducted on 20 and 27 August, while only afternoon flights were made on 26 August. Except for several vertical profiles and morning touch-and-go urban profiles at Boeing Field, the flights were typically at 600 m in the boundary layer west of the Cascades and 1500 m over the mountain range. The three flights on 26 and 27 August are included here for model comparison, as 20 August had very similar meteorological pattern to 27 August. As shown in the flight tracks (Figure 1), the morning flight was focused on measuring the Seattle urban area as well as U.S.-Canadian transport, while the afternoon flights were intended to sample the aged urban plume around Puget Sound as well as the background air mass.
 The meteorology conditions were representative of weak onshore flow which is frequently seen in the summer. Twenty-six August was characterized by isolated morning stratus clouds followed by sunny and warmer conditions with afternoon surface temperatures reached ∼27°C in the Puget Sound area. Even though these conditions didn't produce an exceedance ozone level, the light northwesterly winds (∼3 m/s) were consistent with typical flow patterns associated with high O3 concentrations in the region. Twenty-seven August was associated with cooler conditions and increased wind speeds. Afternoon temperatures reached ∼22°C, and wind speeds around noon ranged from 3 to 5 m/s at the surface.
 O3 was measured using a commercial UV absorption instrument (Thermo Environmental Instruments, model 49–100) that was calibrated before the field experiment. Uncertainty in the O3 measurement was estimated to be ±5%. Carbon monoxide was measured using a nondispersive infrared based monitor (Thermo Environmental Instruments, model 48C). The CO instrument was calibrated before and after each flight with a multipoint calibration. In-flight zeroes were performed in order to track changes in background due to changes in cabin temperature. Estimated uncertainty for a 1 min average was 30 ppb ± 15% of the measured mixing ratio. NO and NOy were measured with a commercial chemiluminescence instrument (Thermo Environmental Instruments, model 42S). The detector response to NO was calibrated before and after each flight with a multipoint calibration. NOy was converted to NO using a molybdenum catalyst at 350°C. Detection limits for NO were estimated to be 300 ppt. The uncertainty in the data was 300 ppt ± 15% of the measured NO mixing ratio. Instrument response times were 20–30 s for the CO, O3, and NO, NOy instruments. Data were sampled every second for these compounds.
 VOC measurements were made by using a proton transfer reaction mass spectrometer (PTR-MS) and by sampling into canisters for analysis by gas chromatography flame ionization detection (GC-FID). The inlet for the PTR-MS was a rearward facing PFA Teflon tubing inlet through which 1 SLPM was pulled through a low pressure drop mass flow controller by a diaphragm pump. The PTR-MS subsampled this flow at 300 sccm upstream of the mass flow controller. The PTR-MS was calibrated before and after each flight using a multicomponent compressed gas standard (Apel-Riemer Environmental) diluted with humid zero air. Humid zero air was made in situ by passing ambient air over a heated catalyst (1% Pt on alumina). In-flight zeros were performed by over flowing the inlet with humid zero air. The PTR-MS measured various compounds every 16 s, in which m/z 71 (methyl vinyl ketone and methacrolein), m/z 33 (methanol), m/z 45 (acetaldehyde), and m/z 59 (acetone) were used here for analysis. Detection limits were ∼50 –100 ppt depending on the compound. For methyl vinyl ketone (MVK) and methacrolein (MACR), data were only available for the two afternoon flights.
 Canisters were filled with a metal bellows pump sampling from a rearward facing stainless inlet. Canisters were pressurized in 10–15 s and returned to Washington State University (WSU) for analysis of C3-C12 hydrocarbons by GC-FID. Organics from a 500 mL air sample were cryogenically trapped on glass beads using liquid oxygen. The analysis was performed on a DB-1 column and mixing ratios were determined by referencing responses to a NIST certified standard of 2,2-dimethylbutane. Uncertainty in the canister measurements based on the precision of replicate analysis was the larger of 5% or 10 ppt. For sampling the morning urban profile at Boeing Field, canisters were filled sequentially every 20–30 s. For measuring the downwind aged plume, canisters were mostly filled every 4–5 min.
 Formaldehyde (HCHO) was measured by in situ scrubbing and derivatization by DNPH and analysis in flight by HPLC. Data were sampled every 90 s. This system has been described by Lee et al. .
 The Mesoscale Meteorological model Version 5 (MM5) was run for three one-way nested domains at 36, 12, and 4 km horizontal grid spacing to develop the 3-D meteorology fields. The domain centers on the Pacific Northwest with 38 vertical sigma levels. The MM5 physics options included the Kain-Fritsch cumulus parameterization scheme [Kain and Fritsch, 1990] at 36 and 12 km domain, the MRF PBL scheme [Hong and Pan, 1996], simple ice microphysics, and no shallow cumulus parameterization. Output from the 4 km MM5 simulation was used to provide 3-D meteorological fields for emissions and photochemical modeling.
 The area, on-road mobile, non-road mobile, point, and biogenic emissions were processed at WSU using the Sparse Matrix Operator Kernel Emissions (SMOKE) processor (http://www.smoke-model.org/index.cfm). The emission inventory was based on information compiled from the Washington State Department of Ecology, Oregon Department of Environmental Quality, Idaho Department of Environmental Quality, Washington State University, the Western Regional Air Partnership, and Environment Canada. For on-road mobile sources, emissions were based on the Western Regional Air Partnership (WRAP) 2003 inventory with emission factors obtained from EPA MOBILE-6 [U.S. Environmental Protection Agency, 2003]. Other mobile sources such as railroad and commercial marine were also included. Point sources were based upon NEI96, with updates for 2001 for Washington and Oregon. Biogenic emissions from trees, plants, and crops were processed by the Biogenic Emissions Inventory System version 3 (BEIS3). The 1-km Biogenic Emissions Landcover Database, version 3 (BELD3), was used to generate normalized emissions, which were then used in BEIS3 along with meteorological data and speciation profiles to compute gridded and speciated hourly biogenic emissions. The emissions domain is the same as the photochemical modeling domain, which consists of 123 by 183 horizontal grids and encompasses the I-5 corridor of western Washington and Oregon as well as British Columbia (Figure 2). Table 1 summarizes a typical summer weekday emissions in the domain for the standard modeling scenario. For anthropogenic sources, total VOC as well as the main VOC components including alkane, alkene, aromatics, formaldehyde (HCHO), acetaldehyde (CH3CHO), acetone, and methanol are listed in the table. Similar to the previous modeling studies in the Pacific Northwest [Barna and Lamb, 2000; Chen et al., 2008], anthropogenic sources dominated the NOx emissions, while biogenic sources accounted for most of the domainwide VOC emissions.
Table 1. Summary of Typical Summer Weekday Emissions in the Modeling Domain, Based on the Standard Scenario
Anthropogenic, Base (tons/d)
 CMAQ version 4.4 was used for photochemical air quality modeling. Twenty-one layers were applied in the vertical with the surface layer about 30 m above the ground and 14 levels between the surface and 800 mb. The SAPRC99 photochemical mechanism [Carter, 2000] including aqueous chemistry and aerosol dynamics was employed. The Euler Backward Interactive (EBI) solver was used to solve the chemical kinetic equations. The piecewise parabolic scheme was used as advection scheme and eddy diffusion was used for vertical diffusion algorithms. The Models-3/CMAQ dry deposition model was applied to estimate the deposition velocities of gases and aerosols. CMAQ boundary conditions were based on GEOS-CHEM model [Bey et al., 2001] for most of the gaseous species and representative observations at Cheeka Peak on the Olympic Peninsula [Anderson et al., 1999] for aerosols. The only modification from Elleman's base case run was that the north and western boundary conditions for O3 were revised based on the observations at Trinidad Head and Cheeka Peak. The changes mainly affected the western boundary where the O3 concentrations near the surface were increased from 17 to 27 ppb. The model simulation included a spin-up period of 96 h and then 72 h of valid runs from 00 UTC 26 August until 00 UTC 29 August 2001.
Elleman and Covert  evaluated MM5 and CMAQ performance for the standard scenario. For meteorological parameters, MM5 was evaluated against surface measurements of winds, temperature, relative humidity, and sea level pressure from 200 stations. Satellite observation of clouds, vertical profiles from the operational wind profile at the Seattle office of the National Weather Office and from measurements made by the Gulfstream aircraft during PNW2001 were also used for comparison. The results showed that 2 m temperatures were overestimated by 0 to 5°C. The mean error was about 1 m/s for wind speed and 50° for wind direction, which are similar to the performance in the real-time forecasting run operated at the University of Washington (http://www.atmos.washington.edu/mm5rt). For both temperature and winds, there was a diurnal pattern in the model performance and the errors tend to be greatest in the early morning and less in the afternoon and early evening. For example, 2 m temperature was biased high by 3°C and 2°C on the early mornings of 26 and 27 August, but the bias reduced to 1°C at noon on both days and matched the observations well by the middle of the afternoon. Based on the observed profiles from the G-I aircraft, PBL height was overestimated by 450 m and 150∼300 m near Enumclaw, 40 km southeast of Seattle, at 23 UTC (16 PDT) on 26 and 27 August, respectively. This increased the mixing volume for the two days by approximately 40% and 20%. For CMAQ performance, Elleman found O3 and photochemistry were generally over-predicted on 26 August and under-predicted on 27 August. The gas phase precursors and meteorology parameters from the model and measurements were analyzed, but no clear conclusion could be drawn for what caused the errors in predicted photochemistry. Bias in meteorological variables such as temperature and modeled PBL heights were recognized by were not sufficient to explain the large variations in O3 performance due to similar performance on the two days. The normalized mean bias and normalized mean error for O3 are 28% and 62% for all available ground stations. This level of error is similar in magnitude to results from previous simulations using CMAQ in the Pacific Northwest [O'Neill and Lamb, 2005; O'Neill et al., 2006; Smyth et al., 2006].
2.2.2. Modeling Scenarios
 Two modeling scenarios were conducted. A standard scenario was based on Elleman's base case CMAQ run. A reduced VOC scenario was developed based on the comparisons of observed CO and VOC versus NOy ratios in the morning urban plume with those in the emission inventory. In the reduced VOC scenario, there were 30% reductions in most of the VOC species with the exception of 44% reduction in CO emissions and factors of 34 and 25 increases in methanol and acetone emissions, respectively. For each scenario, two sensitivity runs were conducted by decreasing either the anthropogenic VOC or NOx emissions by 30%. Details of the comparisons are discussed in Section 2.2.4 and 3.2.
2.2.3. Mapping Modeled and Observed Chemical Compounds
 When comparing observed and modeled VOCs, measured alkane, alkene, and aromatics were grouped into corresponding lumped model species in SAPRC99 chemical mechanism. Since the lumped species in SAPRC99 generally include more compounds than those being measured in PNW2001, the lumped groups from CMAQ output were multiplied by the estimated mole fractions of the compounds available from the measurements. These fractions were determined based on assumed composition of lumped species in the SAPRC99 mechanism, which was derived from an analysis of composition of hydrocarbons in a variety of urban areas of United States [Jeffries et al., 1989]. As the composition represents an average from a large number of cities, Carter [1994, 2000] used this ambient mixture of VOCs to derive the parameters for lump species in the base mechanism as well as reactivity scales of Carter. Detailed mapping between observation and model output is shown in Table 2.
Table 2. Mapping Modeled and Observed Chemical Compounds
Data based on assumed composition of lumped species in the SAPRC99 mechanism [Carter, 2000].
Data show mole fractions of the modeled compounds available from the measurements, which are estimated based on “% in SAPRC99 mixture.”
OH rate constant of lumped group recalculated based on measured VOC species, where the relative molar contributions are determined by “% in SAPRC99 mixture.”
2.2.4. Evaluation of the Emission Inventory
 Ratios of anthropogenic VOC, CO, formaldehyde, acetaldehyde, isoprene, acetone, and methanol with respect to NOy from the morning touch-and-go profile at Boeing Field on 27 August were compared to those with respect to NOx from the emission inventory. Also named as King County international airport, Boeing Field is a fairly representative sampling site of the Seattle urban emissions, as it is located within 1 km of the I-5 corridor and ∼8 km south of downtown Seattle. To compare with measurements, typical summer weekday emissions were extracted from a region that encompasses the Seattle urban area. In the inventory, only the values between 6 and 10 AM were included for comparison. Note that the emissions used here were different from these listed in Table 1, as the latter contains domain-wide daily emissions. Measurements with elevated NOy levels (>30 ppb) from the touch-and-go profile were assumed to be representative of fresh urban emissions and used to calculate the average measured ratios. For this period, there are two hundred data samples for CO and NOy, sixteen samples for acetone, methanol, acetaldehyde from the PTRMS, and six samples for hydrocarbons from canisters. Due to lower temporal resolution of the instrument, only one sample is available for formaldehyde. Therefore, the formaldehyde data should be viewed with some caution as it might not well resolve the urban profile. Since VOCs were generally collected with much less frequency than NOy, NOy concentrations were averaged over the VOC sampling periods to compute the ratios.
2.2.5. VOC Reactivity
 Here, we defined the total VOC reactivity (VOCR) as the sum of the OH loss frequency due to reactions with CO, anthropogenic hydrocarbons, oxygenated hydrocarbons (formaldehyde, acetaldehyde, acetone, methanol, methyl vinyl ketone, and methacrolein), and isoprene.
where ki is the OH rate constant and VOCi is the measured VOC concentration. OH rate constants come from Atkinson .
 For modeled VOC reactivity, the same compounds (CO, anthropogenic hydrocarbons, oxygenated hydrocarbons, and isoprene) were included in calculation. For lumped species, model output was multiplied by estimated mole fractions of the compounds available from the measurements. These fractions were determined based on assumed composition of lumped species in the SAPRC99 mechanism and were the same as these used in section 2.2.3. The OH rate constant of each lumped group was also recalculated based on measured VOC species, where the relative molar contributions were determined by the same method (see Table 2).
 Measured and predicted VOC reactivities were compared for selected time windows during the three flights, with these time windows shown as shaded areas in Figure 3. For the two afternoon downwind flights, the time windows were selected to cover the peak observed VOC reactivity with each spanning around 15 min. For the morning urban flight, the time window is about eight minutes and covers the touch-and-go profile. Within the time windows, there were four, six, and three canisters collected in the afternoon of 26 August, the morning and afternoon of 27 August, respectively. The observed reactivity was calculated based on the measurement frequency of the canister samples: CO was averaged over the time period when canister samples were collected, while acetone, methanol, acetaldehyde, methyl vinyl ketone, methacrolein, and formaldehyde were calculated by picking the closest data points coincident with the canister samples.
3. Results and Discussion
3.1. O3 and NOy
3.1.1. Aircraft Measurements Compared With Model Predictions
 For the afternoon flight of 26 August, the G-I aircraft flew mostly at 600 m during the two circuits around Puget Sound except for a vertical profile near Enumclaw around 15:30 PDT. As shown in Figure 1, elevated O3 levels up to 77 ppb, about 40 ppb above the regional background, were observed to the south and southeast of Seattle. These O3 concentrations are around the 90th percentile of recorded summer daily maximum values (year 2000–2010) for the Puget Sound area. More details of historical O3 levels in the area will be discussed in section 3.1.3. For the morning flight on 27 August, O3 concentrations were relatively higher in the early part of the flight south of Seattle (around 30 to 50 ppb) as the G-I was mostly between 1000 and 2000 m for this period and captured the concentrations from the residual layer. Most other locations showed levels around 30 ppb except that very low concentrations were measured from the touch-and-go profile at Boeing Field as a result of NOx titration. This is consistent with the elevated NOy levels (see Figure 3) in the profile due to the relatively shallow morning boundary layer and abundant rush hour emissions. For the afternoon flight on 27 August, O3 levels were generally lower than the previous day and close to regional average summer values. Background concentrations were around 30 ppb, while slightly higher levels around 40 to 50 ppb were measured at the northern part of Seattle, to the southeast of the city around Enumclaw, and to the west over the Cascade ranges.
 Measured O3 and NOy were compared with the model results from the standard and reduced VOC scenarios for the three flights (Figure 3). Here the model values were interpolated three-dimensionally in space to latitude, longitude, and pressure level of aircraft and in time to every second, so that they were matched in time and GPS location with the aircraft measurements [Elleman and Covert, 2009].
 For the afternoon flight on 26 August, both model runs captured the O3 plume to the southeast of Seattle around Enumclaw, but tended to over-predict to the north and east of the metropolitan area. For the peak O3 levels, there were large over-predictions in the standard scenario by around 20 ppb, while the reduced VOC scenario showed much closer agreement with the measurements. In terms of NOy, two model runs showed essentially the same results, which was also true for the two flights on 27 August. Both scenarios predicted the NOy background correctly and the peak levels fairly well except for missing a few spikes, probably because the plumes were too small to be resolved at 4 km grid resolution. For the morning flight on 27 August, both model scenarios showed some under-prediction of O3 levels for the early part of the flight when G-I was well above the morning boundary layer. Good agreement was reached for most of the other locations. The two model runs also predicted NOy levels relatively well, except for some under-predictions within the touch-and-go profile. This may partly due to the over-predictions of morning surface temperatures and the tendency to over-predict vertical diffusion by the MRF PBL scheme [Elleman and Covert, 2009]. For the afternoon flight on 27 August, the reduced VOC scenario predicted slightly lower O3 concentrations than the standard run. Both runs captured the background O3 levels well, but over-predicted NOy concentrations and correspondingly under-predicted the O3 levels to the east of Seattle. The overestimation was also true for primary pollutants such as NO (not shown), CO, and anthropogenic VOC, which are discussed in detail in Section 3.3.2 below. The improved prediction in O3 for the afternoon flight of 26 August is further illustrated in modeled versus observed concentrations (Figure 4). Model statistics performance also showed better agreement with observations, with the normalized mean bias reduced from 25% to 15%. Predictions of O3 showed fairly small changes for the two flights of 27 August.
3.1.2. Modeled VOC-NOx Sensitivity and O3/NOy Values
 In order to evaluate CMAQ-predicted sensitivity using indicators, we first examined the behavior of model-predicted O3/NOy to see whether it indicates the VOC-NOx sensitivity of the system. For each modeling scenario, two sensitivity runs were conducted by decreasing either the anthropogenic VOC or NOx emissions by 30%. In Figure 5, we plot O3 reductions associated with either VOC or NOx controls against their base case O3/NOy ratios for each scenario. The model results are for 16 PDT on 26 August at layer 9 to match the aircraft measurements. Only the grid cells within a sub-domain which covers the peak O3 plume and the Seattle urban center are plotted. In the figure positive values were related to O3 reductions with reduced emissions, whereas negative values showed increased O3 values with emissions controls. It can be seen that ozone sensitivity was closely related to O3/NOy in both modeling scenarios: cells with O3/NOy greater than 8 were NOx-limited, and cells with values less than 6 were VOC-limited.
 Following the method of Sillman et al. , the grid cells were defined as VOC-sensitive if O3 in the 30% reduced VOC case was lower than both the 30% reduced NOx case and the base case by at least 2 ppb. The grid cells were NOx-sensitive if O3 in the 30% reduced NOx case was lower than both the 30% reduced VOC case and the base case by at least 2 ppb. If both the reduced VOC and reduced NOx cases resulted O3 reductions by at least 2 ppb relative to the base case, the grid cells were defined to have mixed sensitivity. Sillman et al. [1997, 1998, 2003] used percentile distribution of the indicator values to identify transitional values associated with ozone sensitivity. The 95th percentile of the O3/NOy ratios associated with VOC-sensitive locations and the 5th percentile ratios associated with NOx-sensitive locations were chosen as the transitional values. As shown in Table 3, the transitional values were computed accordingly for the two modeling scenarios and compared with the previous studies [Sillman and He, 2002]. It appears that for both modeling scenarios, the 5th percentile values for NOx-sensitive conditions were larger than the 95th percentile values for VOC-sensitive conditions. The results suggest that the transitional values were around 6–8, which is close to most of the previous studies despite the large differences in geographic locations, photochemical models, and chemical mechanisms.
Table 3. Distributions of O3/NOy Values From the Standard and the Reduced VOC Scenarios and From Previous Studiesa
3.1.3. Compare Observed and Modeled Indicator Ratios (O3/NOy)
 For the afternoon flight on 26 August, observed O3 versus NOy relationships were compared with predicted values from the standard and reduced VOC scenarios as shown in Figure 6. For each scenario, model grid cells were shown as VOC-sensitive, NOx-sensitive, or mixed sensitivity according to the criteria described above. To minimize the impact from changes in G-I altitude, the analysis here is based on the data around 3:00 – 4:30 PM (PDT), when G-I flew at 600 m except for the vertical profile near Enumclaw.
 The standard scenario predicted VOC-sensitive chemistry in the Seattle urban center and NOx-sensitive conditions downwind. Peak O3 was associated with VOC-sensitive conditions, but these were not far from the transitional regime. The model predicted a strong correlation between O3 and NOy in NOx-sensitive grid cells, with O3 increasing with increased NOy concentrations. Compared to the observed values, the predicted rate of increase in O3 with NOy was slightly higher and the O3 peak was substantially overestimated (95 ppb modeled versus 77 ppb measured). In VOC-sensitive grid cells, the model showed a large discrepancy with the observations.
 The reduced VOC scenario also predicted VOC-sensitive chemistry in the Seattle urban center and NOx-sensitive conditions in the outskirts, although the VOC-limited area extended slightly further downwind. Similar to the standard scenario, peak O3 was associated with VOC-sensitive and transitional locations. The model predicted lower O3 peaks and slightly lower rate of increase of O3 with NOy, which matched the observations much closer. The correlation between O3 and NOy in the VOC-sensitive regime was also in much better agreement with the measurements.
 No obvious correlation was found between observed O3 and NOy in the afternoon flight of 27 August due to little photochemical formation of O3. Suppressed O3 production could be contributed to combined effects of differences in wind speed and temperature from the previous day. Higher wind speeds have been generally related to lower O3 levels due to increased advection and dry deposition [Baertsch-Ritter et al., 2004; Dawson et al., 2007]. There is also a strong observed correlation between O3 concentration and temperature, where the dependence has been contributed to: 1) increased peroxyacetylnitrate (PAN) decomposition at warmer temperatures, and resulted higher NOx and HOx concentrations; and 2) increased biogenic emissions with increasing temperature [Jacob et al., 1993; Sillman and Samson, 1995]. Comparing these meteorological variables for the two days, midday surface wind speeds increased from ∼3 m/s on 26 August to ∼3–5 m/s on 27 August. Maximum surface temperatures also differed about 5°C, decreased from ∼27°C to ∼22°C. To further evaluate the temperature effects on regional O3 production, we examined the correlation between observed daily maximum temperature and maximum 1-h ozone concentrations at North Bend (for summer 2000–2010) and Enumclaw (for summer 2004–2010). Located at foothills of the Cascades 40 km west and southwest of Seattle, the two surface monitoring sites periodically experience elevated O3 concentrations in the summer. As shown in Figure 7, O3 concentrations are strongly correlated with temperature at both sites for temperature above 20°C. Linear fit using least squares regression for these records (T > 20°C) suggests O3 at the two locations have similar sensitivities to temperature, with slope of 3–4 ppb/°C. Therefore, temperature effects are likely to explain 15 to 20 ppb differences in O3 levels over the two days.
3.2. Comparisons of the Emissions Inventory
 Ratios of CO and VOC with respect to NOy from the morning urban touch-and-go profile on 27 August were compared to those with respect to NOx from the emission inventory as shown in Figure 8 and Table 4. The ratio of CO to NOy was overestimated by 80% (6.5 observed versus 11.6 in the emissions inventory). Overestimation of the CO to NOy ratio might be due to an overestimation of the CO emissions or an underestimation of the NOx emissions. As mentioned in Section 3.1.1, we found model predicted NOy levels were generally in good agreement with the observed concentrations. Recent studies on fuel-based emission inventory in the urban areas [Parrish, 2006] also suggest NOx emissions are reasonably accurate for the mid to late 1990s. Therefore, the overestimation of the CO to NOy ratio here is most likely due to an overestimation of the CO emissions. Compared to other data sets, the observed CO/NOy ratio from G-I (6.5) was close to the reported CO/NOx ratios in 2000 at Nashville (5.7 ± 0.4), TN, Atlanta, GA (6.5 ± 0.4), and U.S. urban (7.9 ± 0.1) [Parrish, 2006]. Applying time trends estimated from that study, the ratios would further reduce to 5.2 for Nashville and 7.4 for U.S. urban in 2001. The CO/NOx ratio from the Seattle emission inventory (11.6) was also close to U.S. EPA annual average inventory value (10.7 for year 1998) [Parrish et al., 2002]. This suggests that both the observation and the emission inventory for Seattle were close to the U.S. urban average, and CO emissions were substantially overestimated in the recent inventory. Similar results have been reported by Parrish et al.  and Parrish  where they found CO emissions in U.S. vehicular emissions were overestimated by about a factor of two in the most recent EPA estimates, as well as by Hudman et al.  where U.S. anthropogenic CO emissions were reduced by 60% to be consistent with observations. It should be noted that the uncertainty in the observed CO/NOy ratio from the G-I here was estimated to be 30% according to the uncertainty in the measurements, and this uncertainty doesn't change the general conclusion we reached.
Table 4. Comparisons of CO and VOC Versus NOy Ratios From Observations and Those Versus NOx Ratios From the Standard Scenario Emission Inventory
Reactivity-Weighted Compound/NOy (NOx)
 For anthropogenic VOC (the sum of alkanes, aromatics, and alkenes), the ratio to NOy was overestimated by about 25%. For its components, the emission inventory predicted the correlation of aromatics to NOy relatively well, but overestimated alkanes to NOy ratio by about 30%. With only propene being available from the observed data sets, the ratios of alkenes to NOy were fairly low in both measured and predicted values. The ratios of oxygenated compounds to NOy were significantly underestimated, ranging from a factor of 3 for formaldehyde to a factor of 30 for methanol. Similar underestimations were reported for Boulder [Goldan et al., 1995] and Boston/New York and Los Angeles [Warneke et al., 2007]. These oxygenated compounds are emitted in large quantities from urban areas, but the sources of these emissions are still not well understood [de Gouw et al., 2005; Warneke et al., 2007].
 As shown in Figure 8 and Table 4, the ratios of these compounds to NOy were also compared to the emission inventory in a reactivity-weighted manner by multiplying the absolute ratios with the correspondent OH rate coefficient. There was an even split between CO and anthropogenic VOC, each of which accounts for 40% of the total measured reactivity. The contribution from oxygenated compounds was about half of that from CO. For anthropogenic VOC, aromatics played a more important role than alkanes. Compared to the observed ratios, there were about 30%, 80%, and 30% overestimations in anthropogenic, CO, and the total VOC reactivity in the emission inventory. This result also supports that overestimation of VOC in the emission inventory might be the cause for overestimating peak O3 and O3/NOy slope in the standard modeling scenario. According to the findings here, we developed the reduced VOC modeling scenario by reducing most of the VOC species by 30% with the exception of reducing CO by 44% and increasing methanol and acetone by factors of 34 and 25.
 It should be pointed out the effects of background and secondary sources on observed mixing ratios of CO and oxygenated compounds were not taken into account. Based on the observations of clean air masses during the flight, the background concentrations were determined to be 70 ppb for CO, 0.47 ppb for methanol, 1.18 ppb for acetone, and 0.13 ppb for acetaldehyde, accounting for ∼25% of measured CO, 39–46% of methanol and acetone, and ∼25% of acetaldehyde within the touch-and-go profile. These levels of background CO would reduce previously derived CO/NOy ratio by ∼25% to 4.72, which is still within 30% uncertainty range in the measurements. For methanol, acetone, and acetaldehyde, taking into account background influences would bring underestimation in the emissions inventory down to factors of 20, 14, and 2.6, respectively, which doesn't change our general conclusion that these compounds are severely under-predicted. The influence of secondary sources is also likely to be small since the touch-and-go observation profile was measured relatively early in the morning (10 AM) with cool temperatures (∼17°C) when photochemical production was small. Observed data such as O3 from the afternoon flight further show lack of vigorous chemistry on the day. In terms of specific compounds, de Gouw at al.  found acetone was mostly attributed to primary as compared to secondary anthropogenic and biogenic sources, whereas the contribution from secondary anthropogenic sources for methanol was negligible. Previous studies also reported strong diurnal variations of primary carbonyl fractions at urban areas [Rappenglück et al., 2005; Garcia et al., 2006], with primary formaldehyde and acetaldehyde dominated over photochemical productions at 10 AM [Rappenglück et al., 2005]. In summary, taking into account the influence of background levels and secondary sources would reduce the observed emission ratios of CO and oxygenated compounds versus NOy. However, this does not change our general findings that CO emissions were substantially overestimated and oxygenated compounds were severely under-predicted.
3.3. VOC Reactivity
 For each time window during the three flights (see shaded areas in Figure 3), the maximum, minimum, and median values from the measurements were compared with these from the standard and reduced VOC scenario for the total VOC reactivity and its components (Figure 9) as well as for contributions to the anthropogenic VOC reactivity (Figure 10).
3.3.1. Measured VOC Reactivity
 As pointed out in the evaluation of the emission inventory, morning urban VOC reactivity were dominated by anthropogenic VOC and CO, each of which contributed to about 40% of the measured reactivity. The contribution from oxygenated compounds was about half of that from CO, while the contribution from biogenic sources was minor. For the total VOC reactivity, the values measured here had a value around 4 s−1, which is in between the median values reported in Phoenix (∼1.2 – 5 s−1, median = ∼2 s−1) and Philadelphia (∼2.6 – 11 s−1, median = ∼4.5 s−1) [Kleinman et al., 2005].
 For the two afternoon flights, CO was the largest contributor to the total VOC reactivity in the aged plume, accounting for about 30–40% of the measured values. Compared to morning urban profile, anthropogenic sources showed a reduced contribution (∼20% on 26 August and ∼30% on 27 August), while the contribution from isoprene increased (∼10–15%), as a result of both warmer afternoon temperatures and more abundant emission sources in the downwind area. The isoprene spatial distribution was highly variable, indicated by the large differences between the maximum and median values. The contribution from oxygenated compounds also increased in the aged plume, which accounted for about 31% and 26% of the total VOC reactivity on 26 and 27 August, respectively. The larger contribution from oxygenated compounds on the first day was consistent with higher O3 levels. Within these oxygenated species, formaldehyde, methacrolein, and methyl vinyl ketone have strong secondary sources from biogenic isoprene oxidation. Compared to the morning urban plume, the overall trend here was reduced values in anthropogenic sources due to loss from chemical reactions and dilution, and increased contributions from biogenic and oxygenated sources as a result of photochemical production of these compounds.
 For anthropogenic components (Figure 10), aromatics played the most important role in the morning urban plume on 27 August. In the afternoon aged plume, the reactivity from aromatics was less than that from alkanes on 26 August and showed comparable values on 27 August. For all flights, alkenes only accounted for a small fraction of the anthropogenic reactivity due to limited data from the measurements.
 It should be noted that there were also other unmeasured compounds that also contribute to OH reactivity. Due to the detection limit of the instruments, part of the alkanes, aromatics, and the majority of the alkenes were not available from the canister measurements. In particular, ethene, C4 and C5 alkenes might be important due to their fast reaction rates with OH radicals. Zweidinger et al.  measured roadside concentrations including various light alkanes and alkenes. According to their study, the contributions from ethene, and C4 and C5 alkenes are about 40% of the total measured reactivity. Goldan et al.  referred to Zweidinger's work to quantify the impact of unmeasured light alkenes on the total reactivity and pointed out that the results from roadway study could be used as an upper limit for aircraft studies due to the fast loss rates of these compounds.
 In addition to anthropogenic sources, other unmeasured biogenic VOCs such as terpenes, and unmeasured oxygenated compounds might also contribute significantly to the total VOC reactivity. We estimated their role based on model predicted values. According to the speciation in SAPRC99 mechanism, the following compounds were examined: terpenes, lumped C3+ aldehydes (RCHO), methyl glyoxal, glyoxal, ketones with OH rate constants less and greater than 5 × 10−12 cm3 molec−1 s−1 (MEK and PROD2, respectively), and lumped isoprene products (ISOPROD). The combined OH reactivity of these unmeasured species ranged from 0.8 s−1 on the two flights of 27 August to 1.1 s−1 on the flight of 26 August, accounting for about 20% of total observed VOC reactivity. Among these species, lumped C3+ aldehydes (RCHO) showed the largest contribution (40–50%). The two lumped ketones (MEK, PROD2) also accounted for significant fractions (20–30%). Large spatial variability were found for terpenes, with their contributions up to ∼30%. Compounds with strong secondary sources from isoprene (MGLY, GLY, ISOPROD) played a minor role, as their combined contribution was less than 10%.
3.3.2. Modeled VOC Reactivity Compared With Measurements
 As shown in Figure 9, the standard scenario substantially overestimated the contribution from CO to the total VOC reactivity, while the reduced VOC run showed much better agreement with the measurements. For anthropogenic VOC, both scenarios showed some under-prediction on 26 August while matching the observed median values on 27 August relatively well. For formaldehyde and acetaldehyde, both scenarios showed generally good agreement with observations. For acetone and methanol, there were substantial increases in the predicted mixing ratios from the reduced VOC scenario where their emission rates were actually increased. However, the changes in VOC reactivity were small, due to their slow reactions with OH radicals. Biogenic sources such as isoprene were highly variable in space. It appears that both runs tended to underestimate the role of isoprene, especially in the afternoon of 26 August when biogenic emissions were favored by relative warm temperatures. Consistent with the isoprene bias, methacrolein and methyl vinyl ketone reactivities were also under-predicted for that flight. For the total VOC reactivity, the reduced VOC scenario showed better performance than the standard case in the two flights on 27 August. In the afternoon flight on 26 August, the standard scenario was in closer agreement with the measured values, but with the wrong reason: there were large compensating errors from overestimating CO and underestimating isoprene-related reactivity.
 For anthropogenic VOC (Figure 10) in particular, both model scenarios predicted the relative importance of alkanes and aromatics fairly well for each flight. In the afternoon of 26 August, most components of the anthropogenic VOC reactivity were somewhat under-predicted by both scenarios. Noted that the reduced VOC scenario captured peak O3 and O3/NOy correlation well, indicating the model's reliability in simulating the cumulative production of O3 and history of photochemistry in the air parcel. Therefore, the underestimation in anthropogenic reactivity, which is related to the instantaneous production rates rather than O3 concentration, is most likely due to local emission sources missing from the model inputs. In the morning of 27 August, the reduced VOC run showed some under-predictions of alkanes and aromatics, consistent with the slight bias in NOy within the touch-and-go profile discussed previously.
 In the afternoon flight on 27 August, the reduced VOC scenario performed better than the standard scenario. Nevertheless, for that flight even the reduced VOC run tended to overestimate anthropogenic VOC for its maximum values despite the good agreement in terms of median level. As mentioned previously, the overestimations were also seen for CO, NO, and NOy along with the underestimation of O3. The discrepancies between the model and measurements here are partly due to errors in model representation of photolysis rate. CMAQ predicts photolysis rate in two steps: 1) calculating clear sky rate for specified location and time; 2) interpolating clear sky rate to model grid and time, and correcting for cloud cover. The cloud cover parameters (cloud top, cloud base, cloud fractional coverage, and liquid water content) are internally diagnosed by CMAQ based on the relative humidity profile from MM5. Observations suggested Seattle had sunny skies for 26 August, and slightly reduced (∼10%) surface radiation for 27 August due to thin cirrus throughout the day. However, CMAQ predicted about 50% cloud fractional coverage around the region on 27 August, and the resulted photolysis rates were only about half of the clear sky values between noon and mid-afternoon. Consequently, the photochemistry on 27 August was largely underestimated. Another aspect of the model error is related to predicted wind speeds. In between late morning and early afternoon hours of 27 August, observations showed wind speeds around 3–5 m/s at the Seattle urban area, whereas model predicted fairly light winds (1–2 m/s) around the region. This could have large impacts on O3 simulation due to misrepresentation of pollutant advection and dry deposition.
 The effects of errors in photolysis and wind speeds on O3 levels were examined in two sensitivity simulations (Figure 11). In the first scenario, wind speeds were increased by 150% for hours between 11 to 15 PDT on 27 August, so that the predicted values were generally consistent with measurements. The changes were applied for the entire model domain from surface to layer 12 (∼1500 m), which is approximately the predicted PBL height. In the second scenario, we made the same modifications for winds and also turned off cloud attenuation effects in CMAQ so that the clear sky photolysis rates were used. For the simulation with changes in wind speeds only, the model predicted substantially lower NOy levels in the urban center and areas immediate to the east of the city, and these concentrations were in better agreement with observations. Although O3 levels at these locations were slightly increased as a result of moderation of the NOx titration effects, the O3 concentrations were still largely under-predicted. For the simulation with clear sky photolysis rates combined with increased wind speeds, O3 increases around 10 ppb were found over large areas covering Puget Sound region and further downwind, with predicted concentrations comparing much better with measurements. Modeled NOy levels in this case were similar to the scenario with changes in wind speeds only. Based on these sensitivity simulations, we conclude that the overestimation of primary pollutants and underestimation of O3 on 27 August are due to combined model errors in transport and photolysis.
3.3.3. VOC Reactivity and NOy
 Correlations between VOC reactivity and NOy were also compared between the measurements and the two modeling scenarios for anthropogenic VOC and CO as shown in Figure 12 and Table 5. Anthropogenic VOC and CO reactivity both appeared to increase with increased NOy in the observed and predicted values. For CO, the reduced VOC run showed lower CO/NOy slopes and much better agreement with observations in all three flights. For anthropogenic VOCs, the reduced VOC scenario generally showed close or better performance for the two days.
Table 5. Linear Correlations for CO and Anthropogenic VOC Reactivity Versus NOy From Measurements and From the Standard and Reduced VOC Scenariosa
26 Aug, PM
27 Aug, AM
27 Aug, PM
26 Aug, PM
27 Aug, AM
27 Aug, PM
Data based on Figure 12. The linear correlation coefficients shown as r2. Slope (b) and intercept (a) determined for: VOCr = a + b[NOy]. For model results, data with NOy concentrations below the minimum observed levels are excluded from the fit.
 The standard modeling scenario overestimated peak O3 and the O3/NOy slope, which suggests an overestimation of sensitivity to NOx, probably due to too much VOC in the emission inventory. Comparisons of CO and VOC versus NOy ratios from the morning urban plume and those from the emission inventory also suggested an overestimation of the total VOC reactivity by about 30%. The reduced VOC scenario was developed based on emission inventory comparison, and resulted in better agreement with the measurements in terms of peak O3 as well as O3/NOy correlations. Comparisons of observed and predicted VOC reactivity also showed that the reduced VOC scenario performed generally better than the standard run. CO predictions were substantially improved with the adjusted emissions, and better agreement with the measured VOC reactivity was reached most of the time.
 In addition to the errors in the emission inventory, indicator ratios can also be substantially affected by NOy removal processes such dry and wet deposition, aerosol interactions as well as measurements errors. We found model predicted NOy concentrations reasonably accurate in both the urban plume and the background air mass. Therefore, the modeled errors in O3 and O3/NOy correlations from the standard scenario are mostly likely related to the errors in the emissions instead of those associated with NOy removal or measurements.
 CMAQ-predicted O3-NOx-VOC chemistry was evaluated by comparing to the measured values from the PNW2001 field campaign. Two modeling scenarios were evaluated: 1) a standard scenario based on the original run from Elleman and Covert , and 2) a reduced VOC scenario developed based on comparisons of observations with the emission inventory.
 We found that the O3/NOy ratio is a useful indicator for evaluating model-predicted ozone sensitivities. Modeled ratios appeared to be closely related to VOC-NOx sensitive conditions, with transitional values similar to those identified from previous studies. Peak O3 was associated with VOC-sensitive conditions, but these were not far from the transitional regime. The standard modeling scenario over-predicted peak O3 and the O3/NOy slope, which suggested an overestimation of sensitivity to NOx, probably due to too much VOC in the emission inventory. The reduced VOC scenario resulted in better agreement with measurements in terms of peak O3 as well as O3/NOy correlations.
 Comparisons of observed CO and VOC to NOy ratios from the morning urban profile with those from the emission inventory also supported an overestimation of VOC in the standard scenario. The results suggested: (1) substantial overestimation of CO emissions for ∼80%, (2) overestimation of anthropogenic and the total VOC reactivity by 30%, (3) large under-prediction of oxygenated compounds.
 In the morning urban plume, anthropogenic VOC and CO appeared to be the most important contributors to odd oxygen photochemistry, while the contribution from oxygenated compounds was about half of that from CO. In the afternoon downwind plume, CO and oxygenated compounds accounted for two thirds of the total measured reactivity. Anthropogenic sources showed a reduced contribution due to loss from chemical reactions and dilution. Biogenic sources such as isoprene also played an important role in the aged plume. Compared to the measured VOC reactivity, the standard modeling scenario substantially overestimated the reactivity from CO. The reduced VOC scenario showed generally good agreement with observations of the relative contributions of the total measured VOC reactivity.
 Our results showed that both CO and oxygenated compounds play an important role in the regional photochemistry, however, their emissions were not well quantified. This has profound implications for modeling regional O3 productions due to the sensitivity to VOC/NOx ratios. Therefore, better methods should be developed for updating the current CO emission inventory, and more measurements are needed to have a better understanding of the sources of oxygenated compounds.
 This study is based on measurements from three aircraft flights with limitations in the size of the samples as well as availability for certain useful species. For example, NO2 was not measured and previous studies [Tonnesen and Dennis, 2000a, 2000b; Arnold et al., 2003; Sillman and He, 2002] have shown that accurate measurement of NO2 can provide powerful probes (O3/NOx, O3/NOz, and NOz/NOy) for diagnostic evaluation of O3 production and cycling processes. Meanwhile, more complete measurements of VOC species including light olefins and terpenes which were mostly not available in this work, are also needed to better characterize VOC reactivity in the atmosphere and access their impact on photochemical production of pollutants. Therefore, measurements with additional key species at various atmospheric conditions and comparisons with model results will further improve our understanding of the regional photochemistry and its representation in air quality models.
 Support for this work was provided by the Boeing Endowment and NW-AIRQUEST consortium of air quality agencies in the Pacific Northwest. We thank Jeff Arnold, Rob Gilliam, Jared Bowden, and Adam Reff for helpful comments. We also thank PNW2001 investigators for sharing the field data. We acknowledge collaborators at the Washington State Department of Ecology, Oregon Department of Environmental Quality, Idaho Department of Environmental Quality, and Environment Canada for data support.