A modified Ambient Ion Monitor-Ion Chromatograph was utilized to monitor the composition of water-soluble fine particulate matter (PM2.5) and precursor gases at the Bakersfield, CA, supersite during Research at the Nexus of Air Quality and Climate Change (CalNex) in May and June of 2010. The observations were used to investigate inorganic gas/particle partitioning, to derive an empirical relationship between ammonia emissions and temperature, and to assess the performance of the Community Multiscale Air Quality (CMAQ) model. The water-soluble PM2.5 maximized in the morning and in the evening because of gas/particle partitioning and possibly regional transport. Among the PM2.5 constituents, pNO3− was the dominant chemical species with campaign average mass loading of 0.80 µg m−3, and the mass loadings of pNH4+ and pSO42− were 0.46 µg m−3 and 0.53 µg m−3, respectively. The observed HNO3 (g) levels had an average of 0.14 ppb. Sub-ppb levels of SO2 (g) were measured, consistent with the absence of major emission sources in the region. Measured NH3 (g) had an average of 19.7 ppb over the campaign and demonstrated a strong relationship with temperature. Observations of ammonia were used to derive an empirical enthalpy for volatilization of 30.8 ± 2.1 kJ mol−1. The gas/particle partitioning of semivolatile PM2.5 composition was driven by meteorological factors and limited by total nitrate (TN) in this region. CMAQ model output exhibited significant biases in the predicted concentrations of pSO42−, NH3 (g), and TN. The largest model bias was in HNO3 (g), with an overprediction of an order of magnitude, which may be due to missing HNO3 (g) sinks such as reactive uptake on dust in the CMAQ framework.
Atmospheric PM2.5, particulate matter with an aerodynamic diameter of less than 2.5 µm, is important because it alters the radiative balance of the Earth by scattering solar radiation (direct effect) or by serving as cloud condensation nuclei (indirect effect). Atmospheric PM2.5 also affects remote and sensitive ecosystems through acid and nitrogen deposition, reduces visibility, provides surface area for heterogeneous chemical reactions [Finlayson-Pitts and Pitts, 2000; Jacob, 1999; Poschl, 2005; Seinfeld and Pandis, 1996], and affects human health [Mannucci, 2010]. Hence, knowledge of the chemical composition and physical state of atmospheric PM2.5 is important for a better understanding of both climate forcing and air quality. For semivolatile chemical species, the gas and particle phases are linked by thermodynamic equilibrium partitioning.
In North America, particulate ammonium (pNH4+), sulphate (pSO42−), and nitrate (pNO3−) can constitute as much as ~50% of PM2.5 mass [Pinder et al., 2007]. The rest of the mass generally consists of organic carbon, elemental carbon, and mineral cations [Poschl, 2005]. Gases such as ammonia (NH3 (g)), sulphuric acid (H2SO4 (g)), and nitric acid (HNO3 (g)) contribute to PM2.5 through chemical reactions and physical processes (e.g., nucleation and condensation). Gas phase sulphuric acid, an oxidation product of sulphur dioxide (SO2 (g)), has a low vapor pressure and forms particulate sulphuric acid, H2SO4(aq,s). Ammonia, directly emitted from agricultural and traffic sources [Reis et al., 2009], reacts preferentially with H2SO4(aq,s) to neutralize it even in the presence of significant HNO3 (g) concentrations. Nitric acid is an oxidation product of nitrogen oxides (NOx), which are emitted mostly by anthropogenic combustion processes. The formation of pNO3−(aq,s) is a thermodynamic equilibrium process and is governed by the absolute amounts and availability of total ammonia (TA ≡ NH3 (g) + pNH4+(aq,s)), total sulphate (TS ≡ H2SO4 (g) + pHSO4−(aq,s) + pSO42−(aq,s) = ~pHSO4−(aq,s) + pSO42−(aq,s)), total nitrate (TN ≡ HNO3 (g) + pNO3−(aq,s)), ambient temperature (T), and relative humidity (RH) [Mozurkewich, 1993; Stelson and Seinfeld, 1982]. PM2.5 constituents dry deposit at much lower rates than their gas phase analogues [Seinfeld and Pandis, 1996]. Hence, the atmospheric lifetimes and environmental impacts of these reactive nitrogen species will be affected by their physical states.
Research at the Nexus of Air Quality and Climate Change (CalNex) 2010 was a large-scale, collaborative, regional air quality and climate change assessment conducted by the National Oceanic and Atmospheric Administration, the California Energy Commission, and the California Air Resources Board (CARB) in California in May and June of 2010 [Ryerson et al., 2013]. The state of California was selected for the assessment because of its long-standing regional air quality problems due to frequent forest fires, heavy traffic, intensive crop and dairy farming, petroleum extraction and refining, and other substantial industrial practices (http://www.esrl.noaa.gov/csd/projects/calnex/whitepaper.pdf). California's San Joaquin Valley (SJV) is prone to air quality problems because the surrounding mountain ranges and predominant wind patterns pose a natural barrier to flow of pollutants out of the valley [Chow et al., 2006; Whiteaker et al., 2002]. Bakersfield is the largest urban area within Kern County, where this situation is especially pronounced. The SJV also represents one of the largest agricultural food production and animal husbandry regions in the state. Animal husbandry operations, which include cattle feedlots, dairies, chicken, and turkey farms, are major sources of NH3 (g) in the SJV [Chow et al., 2006]. The air quality in Kern County is generally poor, and the county represents one of the largest ozone, PM2.5, and PM10 nonattainment areas in the U.S. [Chow et al., 2006; Pusede et al., 2013].
Prior to CalNex 2010, several large-scale field studies were carried out in the SJV: Valley Air Quality Study (VAQS) in 1988 [Chow et al., 1992], SJVAQS/Atmospheric Utilities Signatures, Predictions, and Experiments in 1990 [Chow et al., 1996], Integrated Monitoring Study in 1995 [Chow et al., 1999; Solomon and Magliano, 1999], California Regional PM10/PM2.5 Air Quality Study (CRPAQS) in 1999 [Chow et al., 2006], and Central California Ozone Study in 2000. During the CRPAQS study, measurements of speciated PM2.5 were conducted at 38 sites in the SJV air basin representative of rural, urban, and boundary environments [Chow et al., 2006]. The measurements of PM mass loadings in the SJV were determined to be highest in the winter during periods of air stagnation, consistent with previous observations [Chow et al., 1992, 1993, 1996; Jacob et al., 1986]. Pollution episodes in urban areas of the SJV have been attributed to traffic emissions, residential wood combustion, and secondary ammonium nitrate, which accounted for 30–60% of PM2.5 in wintertime [Chow et al., 1996, 1999; Magliano et al., 1999]. Whiteaker et al.  determined that the particle phase composition in Bakersfield was dominated by carbonaceous particles and secondary ammonium, sulphate, and nitrate. The PM2.5 mass loadings in the SJV were generally lower in the spring and summer than in the winter [Chow et al., 2006].
According to CARB measurements from 2007 to 2012 (http://www.arb.ca.gov/aqd/-aqdcd/-aqdcddld.htm), the PM2.5 mass loadings in Bakersfield, CA, were consistently higher in the winter than in the summer (Figure 1). The inorganic constituents pNH4+, pSO42−, and pNO3− represented a large fraction of the total PM2.5 mass loadings throughout the year. Mass loadings of pSO42− did not vary greatly between seasons, but those of pNH4+ and pNO3− were higher in the winter, dominating PM2.5. The significant decrease in PM2.5 observed in summer could result from the hotter, drier conditions favoring a decrease in mass loadings of semivolatile pNH4+ and pNO3− [Li et al., 2014]. If this is the underlying cause of the observed seasonality, large concentrations of NH3 (g) and HNO3 (g) are expected to be present in the summertime. In the absence of precursor gas measurements, it is challenging to fully explain the factors leading to high wintertime and low summertime levels of PM2.5 in Bakersfield.
Walker et al.  showed that the GEOS-Chem model dramatically underpredicted the seasonality in pNH4+ and pNO3− mass loadings observed in the SJV. Comparison between NH3 (g) observed by the Tropospheric Emission Spectrometer and predicted by GEOS-Chem in SJV indicated a low bias in the model predictions throughout the year. However, Walker et al.  found that they could not resolve the model's low bias in wintertime pNO3− in the SJV, even by increasing ammonia emissions by a factor of 10, and suggested that the seasonality in mixed layer depths may not be properly represented in the model. Similarly, Heald et al.  concluded that the GEOS-Chem model significantly underestimated the NH3 (g) column observed by the Infrared Atmospheric Sounding Interferometer (IASI) over the SJV and the mass loadings of pNO3− in the region. Both analyses suggested that the seasonality of NH3 (g) emissions used in GEOS-Chem [Park et al., 2004] may be contributing to biases in the seasonality of PM2.5 in California. Schiferl et al.  used aircraft observations during CalNex to examine GEOS-Chem model sensitivity to emissions. Schiferl et al.  found that although NH3 (g) was underestimated by the model, it was in excess and PM2.5 formation was limited by gas phase acids in the SJV.
This work presents hourly measurements of inorganic PM2.5 composition (pNH4+, pSO42−, and pNO3−) and precursor gases (NH3 (g), SO2 (g), and HNO3 (g)). Nitrous acid and particulate nitrite measurements are discussed elsewhere (T. C. VandenBoer et al., Evidence for a nitrous acid (HONO) reservoir at the ground surface in Bakersfield, CA during CalNex 2010, submitted to Journal of Geophysical Research: Atmospheres, 2013). The measurements presented in this manuscript provide detailed information to aid in the interpretation of the long-term, filter-based PM2.5 monitoring by CARB, and the drastic reduction of semivolatile ammonium and nitrate concentrations during late spring/summer. An Ambient Ion Monitor-Ion Chromatograph (AIM-IC) was utilized to measure the water-soluble chemical composition of PM2.5 and associated precursor gases at the Bakersfield, CA, supersite during CalNex 2010 in May and June of 2010. We examine the relationship between ambient temperature and NH3 (g), and the formation and thermodynamic equilibrium partitioning of semivolatile inorganic constituents of PM2.5 such as pNH4+ and pNO3−. Another goal of this work is to evaluate the agreement between the field observations and the output of the Community Multiscale Air Quality (CMAQ) chemical transport model. The evaluation of the agreement will identify limitations and strengths of the model, which can help improve its performance and that of other chemical transport models. A more regional evaluation of CMAQ performance, which includes comparison to field observations at other CalNex supersites, is described in Kelly et al. .
2.1 Bakersfield Ground Site
The measurement site was located at the Kern County Cooperative Extension (35.345°N, 118.960°W, and 115 m above sea level) in Bakersfield, CA. This is an urban industrial site situated in a large agricultural region. On a regional scale, the air basin is contained to the west by the Coastal Range (~ 50 km), to the east by Sierra Nevada (~ 25 km), and to the south by the Tehachapi Mountains (~ 25 km). These mountain ranges reduce air flow out of the region, which can lead to accumulation of air pollutants at the site [Whiteaker et al., 2002]. At the local scale, the measurement site was located 6 km southwest of the Bakersfield city center, 3.5 km northeast of the Bakersfield Municipal Airport, 0.8 km south of Highway 58, 7.1 km east of Highway 99, and 35 km east-southeast of Interstate 5. The Kern County Wastewater Treatment Plant was located within 1 km of the site. The data used in the analysis were collected between 21 May and 22 June of 2010.
2.2 Measurement Techniques
The inorganic composition of PM2.5 and associated precursor gases were measured by the AIM-IC. The AIM-IC provided continuous, near-real-time, hourly online measurements of the water-soluble inorganic constituents of PM2.5 and precursor gases with 3σ limits of detection (LOD) as low as 3 ng m−3. The following chemical components of PM2.5 and precursor gases were measured during the campaign: pNH4+, pSO42−, pNO3−, pCl−, pNO2−, pNa+, pK+, pMg2+, pCa2+, NH3 (g), SO2 (g), HNO3 (g), HONO (g), and HCl (g). The instrument consists of an AIM 9000D air sampler (URG Corp., Chapel Hill, NC) fitted with a 2.5 µm inertial impactor for particle size selection, a constantly generated parallel-plate wet denuder for collection of soluble gases, a particle supersaturation chamber for collection of soluble particles, and two ICS-2000 ion chromatographs (Dionex Inc., Sunnyvale, CA). The anion IC was equipped with a potassium hydroxide eluent generator cartridge (KOH EGC II), 4 mm TAC-ULP1 concentration, IonPac AG19 guard, and AS19 analytical columns, a 4 mm ASRS 300 suppressor and CD25A conductivity detector operated at 30°C. Anion chromatographic runs were performed with a gradient elution (1–85 mM KOH) at a flow of 1 mL min−1. The cation IC was equipped with a methanesulphonic acid eluent generator cartridge (MSA EGC II), 4 mm TCC-ULP1 concentration, IonPac CG12A guard, and CS12A analytical columns, a 4 mm CSRS 300 suppressor and CD25A conductivity detector operated at 30°C. Cation runs were performed with a gradient elution (5–80 mM MSA) at a flow of 1.2 mL min−1. Ambient air was sampled at 3 L min−1, and hourly averages of all species were reported in µg m−3 via Chromeleon 6.8 (Dionex Inc., Sunnyvale, CA) and URGAIM (URG Corp., Chapel Hill, NC) software.
An updated AIM-IC inlet configuration, described in Markovic et al. , was implemented for this study. The sample collection components of the instrument were installed in a separate housing unit, which was located on the second floor of the research tower (4.5 m above ground level (agl)). In this configuration, the gases and particles were collected into solution in the tower-mounted assembly to minimize sampling inlet losses for high-solubility and surface adsorptive gases, such as NH3 (g) and HNO3 (g). The collected samples were transported as dissolved analytes through ~20 m of shielded perfluoroalkoxy tubing to the analytical components of the instrument, which were housed in a trailer at the base of the tower. Offline calibrations of the AIM-IC were performed in the field before, during, and after the campaign by directly injecting multiple-ion standards into each IC, providing accuracy of 15% or better. Background values for each ion were determined by overflowing the inlet with zero grade air (Airgas West, Los Angeles, CA) before, during, and after the study. The AIM-IC inlet response times and sampling efficiencies for surface-active compounds like NH3 (g) and HNO3 (g) were evaluated by measuring zero air spiked with ~72 ppb NH3 (g) several times during the campaign, and by comparison (correlation coefficient of R = 0.87) to a collocated HNO3 (g) Chemical Ionization Mass Spectrometer [Crounse et al., 2006]. In addition, the AIM-IC (including the same inlet used in the field study) has been thoroughly characterized in the laboratory for collection of PM2.5 composition and precursor gases prior to and after the CalNex 2010 study [Markovic et al., 2012].
Temperature and relative humidity measurements were made by Vaisala HMP45C temperature and relative humidity probe (Campbell Scientific Inc., Logan, UT). The measurements were acquired from the research tower at 17.8 m agl. The uncertainty in temperature measurements was 0.3 K, and the uncertainty in relative humidity measurements was 0.03.
2.3 CMAQ Model
The CMAQ modeling system version 4.7.1 [Byun and Schere, 2006; Foley et al., 2010] was run during CalNex 2010 to support the measurement efforts. The model was utilized to predict gas and aerosol emissions, chemical transformation, transport, and deposition. CMAQ was applied with ISORROPIA inorganic chemistry [Nenes et al., 1998, 1999], Carbon bond 05 (CB05) gas phase chemistry [Yarwood et al., 2005], organic and sulphur oxidation aqueous phase chemistry [Carlton et al., 2008], and a secondary organic aerosol module [Carlton et al., 2010]. The Weather Research and Forecasting model (WRF), Advanced Research WRF core version 3.1 [Skamarock et al., 2008] prognostic meteorological model, was applied to generate gridded inputs for CMAQ [Baker et al., 2013].
The model domain extended vertically to 50 mb using 34 layers, most of which are in the boundary layer to better resolve mixing heights. The height of the first layer extended to 34 m. Horizontally, 4 km sized grid cells covered all of central and Southern California. A clean vertically variant concentration profile was used to supply initial and boundary conditions. Initial condition influence was minimized by not including the first 2 weeks of the model simulation in the analysis. The 2008 National Emissions Inventory (NEI) version 1 [U.S. Environmental Protection Agency, 2012] was processed with the Sparse Matrix Operator Kernel Emissions modeling system [Houyoux et al., 2000] to generate anthropogenic emissions for the modeling period. Where 2010 continuous emissions monitor data were available for specific point sources, that information was used to generate day-specific 2010 emissions. Other point, area, and mobile sources emissions were not adjusted from 2008 to reflect 2010. Biogenic emissions are estimated using the Biogenic Emissions Inventory System version 3.14 [Pierce et al., 1998] with day- and hour-specific temperature and solar radiation estimated with the WRF model. Day-specific fires were included in the modeling but did not impact the monitor locations [Bahreini et al., 2012]. In this work, the output of the CMAQ model was used for comparison to observations of inorganic chemical composition of PM2.5 and precursor gases at Bakersfield supersite. The results of the analysis are presented in section 3.3.
3.1 General Observations of NH3 (g), SO2 (g), and HNO3 (g)
The emissions of SO2 (g) and NOx in Bakersfield, CA, were attributed to oil field operations and traffic sources [Jacob et al., 1986]. Ammonia emissions in the SJV are among the highest in the U.S. and are strongly linked to animal husbandry operations [Goebes et al., 2003]. Remote sensing with the IASI, showed that SJV is one of the most prominent global hot spots of NH3 (g) with a total atmospheric column of > 1 mg m−2 [Clarisse et al., 2009, 2010]. Clarisse et al.  also concluded that the representative volumetric mixing ratios of NH3 (g) in SJV were highest in the spring and summer. During CalNex, Nowak et al.  demonstrated that the dairy operations in the California South Coast Air Basin are substantially larger emission sources of NH3 (g) than reflected in current emissions inventories. Similarly, we expect the emissions inventories to underestimate the contribution of dairy and beef farming to NH3 (g) emissions in the SJV.
The time series and diurnal plots of NH3 (g), SO2 (g), and HNO3 (g) mixing ratios measured during CalNex 2010 are shown in Figures 2a–2c and 3a–3c, respectively. Ammonia was the dominant PM2.5 precursor gas at the Bakersfield supersite, consistent with aircraft measurements in the SJV during CalNex [Schiferl et al., 2014]. Ammonia mixing ratios ranged from 0.9 ppb to 64.1 ppb with an average of 19.7 ± 10.0 ppb (Figure 2a). The lowest mixing ratios were observed from 21 to 28 May, during a clean and cool period that included sporadic rain events and the highest on 5–8 June during one of the warmest periods during the study (Figures S1 and S2 in the supporting information). The highest concentrations were observed in the early morning (06:00–09:00 PST), likely due to surface emissions into a shallow boundary layer, and the lowest in the evening (20:00–22:00 PST) (Figure 3a). The increases in ammonia mixing ratios as high as 64 ppb were likely due to regional agricultural emissions. According to the 2010 United States Department of Agriculture California Livestock Inventory (www.nass.usda.gov/ca), in the San Joaquin Valley, there were roughly 3.25 million cattle, of which 1.2 million were in Kern County. SJV is also home to some of the largest crop farming operations in the state of California and the nation (www.nass.usda.gov/ca), and significant amounts of NH3 (g) can be emitted from fertilized fields [Reis et al., 2009; von Bobrutzki et al., 2010]. In 2010, 7.67 t of synthetic fertilizer, most of which contained ammonium, were applied in the state of California (http://www.cdfa.ca.gov).
Sub-ppb average mixing ratio of SO2 (g) was observed during the campaign, consistent with the absence of significant regional emission sources such as coal-fired power plants. The SO2 (g) ranged from < LOD (0.005 ppb) to 5.0 ppb with an average of 0.70 ± 0.63 ppb, with the lowest values occurring from 21 to 28 May (Figure 2b). The SO2 (g) displayed a daytime maximum (07:00–19:00 PST) with a local maximum occurring between 08:00 and 09:00 PST (Figure 3b). Although the measurement site was located downwind of major highways and the downtown area, traffic emissions did not appear to be the major source of SO2 (g) at the site. The lack of the correlation between SO2 (g) and traffic tracers such as CO (R = −0.14) and NOx (R = −0.25) further supports this claim. The SO2 (g) mixing ratios observed at the site could be from oil refining or other industry-related upwind sources. The mixing ratios of HNO3 (g) ranged from < LOD (0.065 ppb) to 0.74 ppb with an average of 0.14 ± 0.10 ppb and were also lowest from 21 to 28 May (Figure 2c). The daytime maximum centered around 14:00 PST, and an early morning minimum occurred at 05:00 PST (Figure 3c). Processes thought to control the concentrations of precursor gases are discussed in more detail in section 4.
3.2 General Observations of pNH4+, pSO42−, and pNO3−
From the perspective of inorganic chemical composition, water-soluble PM2.5 mass loadings were dominated by pNH4+, pSO42−, and pNO3− (campaign average of ~70%), although the contribution of other chemical species (Cl−, pNO2−, pNa+, pK+, and pCa2+) summed to about 30%. The total PM2.5 mass loadings ranged from 0.3 to 7.4 µg m−3 with an average of 2.7 ± 1.3 µg m−3. Our PM2.5 measurements are consistent with CARB's long-term observations during the summer shown in Figure 1. The time series and diurnal plots of measured pNH4+, pSO42−, and pNO3− mass loadings are illustrated in Figures 2d–2f and 3d–3f, respectively. The pNH4+ mass loadings ranged from 0.04 to 1.62 µg m−3 with an average of 0.46 ± 0.25 µg m−3 (Figure 2d). Mass loadings of pNH4+ maximized between 06:00 and 10:00 and minimized between 14:00 and 18:00 PST (Figure 3d). The morning peak is associated with elevated levels of pNO3−, whereas the afternoon peak is associated with elevated levels of pSO42−. Particulate sulphate mass loadings ranged from 0.01 to 2.0 µg m−3 with an average of 0.53 ± 0.33 µg m−3 (Figure 2e). Sulphate had a daily maximum in the evening (19:00–22:00 PST) and a minimum in the early morning (03:00–06:00 PST, Figure 3e). Based on the flow patterns within the SJV, the increase in pSO42− mass loadings at the site throughout the day is likely the result of regional transport from upwind urban and industrial sources (e.g., Visalia and Fresno). Growth events of ultrafine particles at the site were observed earlier in the day and were dominated by the condensation of organics from traffic emissions rather than sulphate [Ahlm et al., 2012]. The mass loadings of pNO3− varied from < LOD (0.045 µg m−3) to 4.2 µg m−3 with an average of 0.80 ± 0.60 µg m−3. Particulate nitrate maximized frequently in the morning, but occasionally a small second maximum was observed in the evening (Figure 3f).
3.3 Comparison of AIM-IC Measurements to CMAQ Chemical Transport Model Output
3.3.1 Comparison of Measured and Modeled Precursor Gas Data
The average diurnal profiles for precursor gases predicted by CMAQ are overlaid with the observations in Figures 3a–3c, with error bars representing 1 standard deviation of the hourly average. The results of statistical analyses of the comparison are shown in Table 1. The statistical analysis of the agreement between the observations and predictions was accomplished by calculating mean bias (MB), mean error (ME), normalized mean bias (NMB), normalized mean error (NME), and the correlation coefficient (R), using the equations provided at the bottom of Table 1. These metrics are commonly used in the evaluation of model performance [Simon et al., 2012]. Statistical comparisons between the measurements and model yield negative values for MB and NMB when the model is predicting values lower than those observed and vice versa. The statistical evaluation was performed in instances when data points existed for both model and measurement data sets.
Table 1. Statistical Evaluation of Modeled and Measured Inorganic PM2.5 Mass Loadings and Precursor Gas Mixing Ratios in Bakersfield, CA, During CalNex 2010a
aM_mod = modeled mean, M_obs = observed mean, MB = mean bias, ME = mean error, NMB = normalized mean bias, NME = normalized mean error, N = number of observation-model data pairs used in the statistical evaluation, and R = correlation coefficient. The units for PM2.5 data are µg m−3 and for gas data are ppb, unless specified otherwise.
Mi = modeled data
Oi = observed data
CMAQ was able to predict the appropriate range of NH3 (g) mixing ratios, but the diurnal profile of observed NH3 (g) was poorly represented by the model (Figure 3a). The statistical comparison of measured and modeled NH3 (g) showed a MB of −1.1 ppb and NMB of −5.7%. CMAQ predicted the highest NH3 (g) mixing ratios at the site in the evening with the maximum around 20:00 PST, which was contrary to observations, which peaked around 07:00 PST. Although the model predicted the range of NH3 (g) mixing ratios at the site reasonably well, the correlation of the predictions to the measurements was poor, with the correlation coefficient of R = 0.23. The analysis described in section 3.4 and Figure 4 was carried out for the model data, and an enthalpy of 16.1 ± 2.5 kJ mol−1 was derived, suggesting that the temperature dependence of emissions in the model may need further investigation in the future.
CMAQ was biased low for SO2 (g), with an average modeled mixing ratio of 0.25 ppb, compared to the average measured mixing ratio of 0.70 ppb. The model predicted a diurnal profile with two maxima, which was consistent with the SO2 (g) observations (Figure 3b). The model predicted the morning maximum in SO2 (g) mixing ratios well, but it predicted the early evening maximum a few hours later than observed. Although CMAQ predicted lower daytime SO2 (g) mixing ratios, it was able to capture the evening decrease in the SO2 (g) mixing ratios. The model failed to predict the timing of the concentration minimum observed in the early morning from 04:00 to 05:00. The statistical comparison of measured and modeled SO2 (g) yielded a MB of −0.45 ppb and NMB of −64%. The correlation of the predictions to the measurements was poor with the correlation coefficient of R = 0.14. Since the SO2 (g) in the Bakersfield area is predominantly from nonpoint sources, the 4 km grid cells in the model output may be smearing out the traffic or oil-related emissions in close proximity to the site.
The agreement between the modeled and the measured HNO3 (g) was the poorest of the three precursor gases. The model was biased high by nearly an order of magnitude; the average modeled HNO3 (g) mixing ratio was 1.1 ppb compared to the average measured HNO3 (g) mixing ratio of 0.14 ppb (Figure 3c). The statistical comparison yielded the highest NMB of the three precursor gases, at 740%. The shapes of diurnal profiles of the modeled and observed HNO3 (g) mixing ratios were similar. The model captured the midday maximum and also predicted a slight increase in the concentrations in the early evening. Although the absolute values of modeled HNO3 (g) mixing ratios were biased high, the correlation between the measured and the modeled HNO3 (g) mixing ratios was reasonable with the correlation coefficient of R = 0.59. This suggests that the model represented the factors driving the diurnal variability in the HNO3 (g) mixing ratios at the site reasonably well throughout the campaign. However, the model predicted higher daytime and nighttime HNO3 (g) mixing ratios and, like SO2 (g), failed to predict a decrease in HNO3 (g) levels observed at the site in the early morning.
3.3.2 Comparison of Measured and Modeled PM2.5 Composition Data
The comparison of measurements of PM2.5 composition acquired with the AIM-IC and modeled by CMAQ is shown in Figures 3d–3f and in Table 1. The mass loadings of pNH4+ predicted by CMAQ were generally higher than the observations. CMAQ was able to predict the background levels well, but the modeled maxima of pNH4+ were higher than the observed. The average measured mass loading of pNH4+ was 0.46 µg m−3 compared to the average modeled mass loading of 0.67 µg m−3. The comparison of diurnal profiles of modeled and observed pNH4+ shows that the CMAQ model predicted higher pNH4+ earlier in the morning, but similar mass loadings during the day and in the evening (Figure 3d). The comparison of modeled and measured pNH4+ led to a MB of 0.22 µg m−3, NMB of 47%, and R of 0.22.
The average measured pSO42− was 0.97 µg m−3 compared to the averaged modeled pSO42− of 0.53 µg m−3 (Figure 3e). The model predicted higher pSO42− independent of the time of day. Although CMAQ did not capture the progressive daytime increase in pSO42− mass loadings at the site, it was able to predict the evening increase in pSO42− between 17:00 and 20:00 PST. Poor model performance toward pSO42− led to the MB of 0.44 µg m−3, NMB of 84%, and the correlation coefficient of R = 0.11.
The agreement between the modeled and the measured pNO3− was the best of the three examined PM2.5 inorganics (Figure 3f). The average modeled mass loading of pNO3− was 1.0 µg m−3 compared to the average measured mass loading of 0.80 µg m−3. Compared to the observations, the CMAQ model predicted a morning maximum in pNO3− earlier in the day, but accurately captured a small evening peak of ~ 0.5 µg m−3 observed around 20:00 PST. The model predicted ~ 50% lower daytime mass loadings of pNO3− compared to the observations. The comparison of the two data sets yielded the lowest MB of 0.21 µg m−3 and NMB of 27%, but the highest correlation coefficient of R = 0.48. The diurnal profile of modeled pNO3− mass loadings was very similar to that of pNH4+, suggesting that the formation of ammonium nitrate dominated the morning formation of PM2.5 modeled by CMAQ. Given the poor model performance in predicting accurate levels of NH3 (g) and HNO3 (g), the reasonable predictions for pNO3− are somewhat surprising. The combination of overestimated HNO3 (g), along with underestimated NH3 (g) likely conspires to produce reasonable mass loadings, while the strong meteorological controls on pNO3− formation drive most of the correlation.
3.4 Influence of Temperature on NH3 (g) Concentrations
A very small fraction of TA is present in the particle phase, and ammonia has slow gas phase sinks. Therefore, the variability in NH3 (g) can be interpreted as resulting mainly from changes in emission or deposition. While gasoline-powered vehicles are known to emit NH3 (g) [Bishop et al., 2010; Livingston et al., 2009; Nowak et al., 2012], the diurnal cycle of NH3 (g) at the site is not consistent with traffic being the dominant source. Ammonia emissions from agricultural activity are known to depend on many factors including soil or manure nitrogen content, pH, temperature, and wind speed [Sommer et al., 1991]. Comparison of the temperature and NH3 (g) time series (Figure S3) indicates that observed ammonia levels tend to be higher on days with high temperature. This is to be expected, as high temperatures will favor evaporative losses of ammonia from sources such as fertilized fields and animal waste lagoons. The vaporization of ammonia from the pure liquid has an enthalpy of 23.35 kJ mol−1 [Linstrom and Mallard, 2003], and this equilibrium (equation (1)) may be of relevance in areas with very high concentrations of animal waste. The evasion of ammonia to the gas phase from aqueous solutions is governed by the Henry's law constant (equation (2)), with an enthalpy of 34.9 kJ mol−1 [Dasgupta and Dong, 1986]; however, depending on the pH of the solution, it may also be necessary to consider the enthalpy of acid dissociation [Bates and Pinching, 1950] (equation (3)).
Equations (2) and (3) can be used to describe the bidirectional exchange of ammonia over fertilized and seminatural surfaces [Massad et al., 2010; Zhang et al., 2010], such as those present throughout the SJV. A bidirectional algorithm can be used in CMAQ to predict the surface-atmosphere exchange of ammonia [Bash et al., 2013], but it was not implemented for this analysis. While information on many physical and activity factors is needed to develop accurate ammonia emissions inventories, most inventories do not explicitly include temperature dependence, despite the physical basis for doing so. The ammonia mixing ratios predicted by CMAQ using the conventional representation of emissions do not show the same correspondence with temperature that is seen in the observations (Figure S3).
Assuming that the depositional losses of NH3 (g) have a negligible temperature dependence, a Van't Hoff analysis (equation (4)) can be used to extract the empirical temperature dependence of ammonia emissions in the region.
where ∆Hvol is the empirically derived enthalpy of volatilization of ammonia from all sources in the region, R is the gas constant, and c is a constant that reflects other factors governing the NH3 (g) mixing ratio, which are assumed not to be temperature dependent. Air temperature is used as a surrogate for the surface temperature, which would be more appropriate for this analysis but would be challenging to estimate for all upwind emitting surfaces. The fit to the observations (Figure 4, top) has a correlation coefficient of −0.52 and can be used to derive a value for ∆Hvol of 30.8 ± 2.1 kJ mol−1. This translates to a 50% increase in ammonia mixing ratio for a change in air temperature from 293 K to 303 K. In CMAQ, the relationship is much weaker (R = −0.20) and the slope is shallower giving only a 25% increase for the same temperature change. The magnitude of the empirically derived enthalpy is consistent with both animal husbandry and fertilizer application contributing to NH3 (g) emissions in the region. The observed temperature dependence is in the range of that observed in a variety of agricultural systems, reviewed in Sutton et al. . This analysis suggests that the temperature dependence of NH3 (g) emissions may contribute to the discrepancy between the observations and model predictions of NH3 (g) in the region.
Figure 5 provides a comparison of the gas/particle partitioning of TA and TN between CalNex observations and CMAQ modeling, alongside CARB long-term particle measurements in Bakersfield in the June and December of 2010. The overprediction of modeled pSO42− could be the result of errors in the oxidation rate of SO2 (g), the pSO42− (or SO2 (g)) deposition velocity, the magnitude of SO2 (g) sources, or their geographical distribution. The combination of a high bias for pSO42− and a low bias for SO2 (g) suggest that errors in the oxidation rate may contribute to the model-measurement mismatch (Figure S4). Schiferl et al.  also found a high bias for pSO42− and a low bias for SO2 (g) when comparing GEOS-Chem predictions to aircraft observations in the SJV during CalNex. AIM-IC measured ~1.7 times as much total sulphur (SO2 (g) + pSO42−) as predicted by the CMAQ. This could be a consequence of the change in sulphur emissions by traffic and oil-related sources that are not reflected in the 2008 NEI inventory utilized by CMAQ. The overprediction of pNH4+ mass loadings by the model is partly the result of the overprediction of pSO42−. There is sufficient ammonia in the atmosphere to neutralize the particle sulphate. Therefore, every nmol of the overpredicted pSO42− in CMAQ will lead to overprediction of pNH4+ by 2 nmol (exactly for solid particles, or to within a few percent for deliquesced particles). It is important to represent sulphate in the model properly in order to accurately predict the mass loadings of pNH4+. The predicted pNO3− and pNH4+ levels are also influenced by the biases in predicted NH3 (g) and HNO3 (g) by the model.
Modeled TN levels at the site were higher than the AIM-IC observations. Thus, the model bias in HNO3 (g) mixing ratios cannot only be explained by inaccuracies in gas/particle partitioning. The high bias in the modeled HNO3 (g) could also partly result from inaccurate rates of production of HNO3 (g), and/or the modeled NOx and OH concentrations. If the dry deposition of HNO3 (g) to the ground surface was underrepresented in the model framework, CMAQ would predict higher HNO3 (g) mixing ratios compared to the observations. Using a rate coefficient of 9.2 × 10−12 cm3 molecule−1 s−1 [Mollner et al., 2010] and measured NO2 and OH at the site [Pusede et al., 2013], the instantaneous production rate of HNO3 (g) resulting from 3 ppb NO2 and 5 × 106 molecules cm−3 OH is 0.5 ppb h−1. This value is clearly much larger than the observed increases in HNO3 (g) (< 0.05 ppb h−1), implying large local sinks. We propose that a lack of proper representation of dust particle chemistry (CaCO3(s)/NaCl(s)-HNO3 (g) interactions) in the CMAQ model framework is partly responsible for the discrepancy between modeled and predicted HNO3 (g) mixing ratios, as mineral dust was ubiquitous at this location.
Based on the long-term measurements of PM2.5 presented in Figure 1, elevated summertime NH3 (g) and HNO3 (g) were expected at the Bakersfield site. Although the observed NH3 (g) mixing ratios were elevated, with the campaign average of 19.7 ± 10.0 ppb, the observed HNO3 (g) mixing ratios were 2 orders of magnitude lower (0.14 ± 0.10 ppb). The sum of HNO3 (g) and pNO3− measured in the late spring/summer was a factor of ~10 lower than the average pNO3− observed in the winter (Figure 5). This suggests that in addition to the volatilization of ammonium nitrate, other processes must be influencing the abundance of HNO3 (g) at the site in the summer. One possible process is the increased deposition TN in the summer because gas phase NH3 (g) and HNO3 (g), which dominate in the warm and dry summer conditions, have much larger deposition velocities than pNH4+ and pNO3−. Other important factors in the seasonality of the TN abundance in the SJV may include reduced vertical mixing and ventilation in the winter, and/or accelerated heterogeneous production of HNO3 (g) via N2O5 (g) in the winter.
Another explanation for the low observed HNO3 (g) mixing ratios during the campaign could be the presence of reactive dust particles, which may be identified by coarse mode tracer cations such as pK+, pNa+, pMg2+, and pCa2+. In addition to being taken up by mineral dust particles, HNO3 (g) can undergo acid displacement reactions with calcium carbonate (CaCO3) and sodium chloride (NaCl) in particles to form Ca(NO3)2 or NaNO3. The majority of the mass loadings of these constituents are in the coarse mode and would not be sampled with the AIM-IC [Cwiertny et al., 2008; Sullivan et al., 2009; Vlasenko et al., 2006, 2009; Wu and Okada, 1994]. These acid displacement reactions can be very efficient. The heterogeneous uptake coefficient (γ) for the reaction of HNO3 (g) with CaCO3 ranges from 0.003 to 0.21 depending on ambient RH [Liu et al., 2008]. For the reaction with NaCl, γ can vary between 0.026 and 0.2 as a function of ambient RH, particle size, and particle composition variation [Liu et al., 2007]. Vlasenko et al.  determined that for the uptake of HNO3 (g) on Arizona Test Dust, γ increases from 0.022 to 0.113 with increase in ambient RH from 12% to 73%, showing that the actual dust samples uptake HNO3 (g) with efficiency similar to pure CaCO3 and NaCl substrates. Hence, the effects of dust chemistry on the thermodynamic equilibrium and the abundance of HNO3 (g) and TN at the Bakersfield site should be investigated further.
Simultaneous hourly measurements of water-soluble inorganic PM2.5 composition (dominated by pNH4+, pSO42−, and pNO3−) and associated precursor gases (NH3 (g), SO2 (g), and HNO3 (g)) were acquired with the modified AIM-IC system as the part of the CalNex 2010 study at the Bakersfield, CA, ground site. Relatively low mass loadings of the water-soluble inorganic PM2.5 with an average of 2.7 ± 1.3 µg m−3 were detected over the entire campaign, which was consistent with the long-term observational record collected by CARB at a nearby site. The mass loadings of PM2.5 measured by AIM-IC were found to maximize in the early morning and in the evening due to gas/particle partitioning, new particle formation, and regional transport. Among the PM2.5 inorganic constituents, pNO3− was the dominant chemical species by mass (hourly average = 0.80 µg m−3), but the mass loadings of pNH4+ and pSO42− were also significant with respective average mass loadings of 0.46 µg m−3 and 0.53 µg m−3. NH3 (g) was the dominant soluble gas phase species (average of 19.7 ppb) due to large agricultural emissions in the region. The observed NH3 (g) mixing ratios were strongly correlated with day-to-day variability in ambient temperature. Because pNH4+ was low compared to NH3 (g), this temperature dependence must be related to surface exchange of ammonia rather than to gas/particle partitioning. HNO3 (g) (average of 0.14 ppb) displayed a diurnal maximum around noon as a result of gas/particle partitioning of semivolatile NH4NO3, and possibly from local photochemical production. The low observed HNO3 (g) levels at the site suggest that significant sinks for HNO3 (g), such as reactive uptake on dust, could have existed at the site throughout the campaign and should be investigated further. On average, sub-ppb levels of SO2 (g) were detected during the campaign, consistent with the absence of major SO2 (g) emission sources in the region. In this work, we also evaluated the predictions of PM2.5 composition and precursor gases by the Community Multiscale Air Quality (CMAQ) modeling system. The model output exhibited significant biases in the predicted concentrations of pSO42−, NH3 (g), and TN. On average, the modeled mixing ratios of HNO3 (g) were an order of magnitude higher than the observations, which could be the result of missing HNO3 (g) mineral dust particle chemistry representation in the model framework. The model was also unable to predict the observed NH3 (g) temperature relationship, which may impact the timing of ammonia emissions. Given its poor performance in describing overall levels of HNO3 (g) and the factors controlling NH3 (g), the CMAQ model and/or the underlying emission inventories should be examined to improve confidence in future regulatory applications.
The authors would like to thank John Karlik and University of California's Kern County Cooperative Extension staff for help and support during CalNex 2010 campaign. Bakersfield supersite infrastructure was financially supported by the California Air Resources Board (CARB, contract 08–316 to University of California Berkeley, PI's Ron C. Cohen and Allen H. Goldstein). Funding for instrumentation was provided by the Canada Foundation for Innovation. The authors also thank R.J. Weber for providing the wind, temperature, and relative humidity measurements, CARB for providing the long-term speciated measurements of PM2.5 mass loadings, and Jeff Geddes for the assistance with the data analysis. M.Z. Markovic would like to acknowledge Centre for Global Change Science at the University of Toronto for the funding to participate in the Calnex 2010 campaign. T.C. VandenBoer acknowledges funding to participate in the CalNex 2010 campaign provided through and NSERC Alexander Graham Bell Doctoral Canadian Graduate Scholarship and the University of Toronto for a Special Opportunity Travel Grant. K. Baker would like to thank Chris Misenis, Lara Reynolds, Allan Beidler, James Beidler, and Chris Allen.