Fine-scale simulation of ammonium and nitrate over the South Coast Air Basin and San Joaquin Valley of California during CalNex-2010

Authors

  • James T. Kelly,

    Corresponding author
    1. Office of Air Quality Planning and Standards, U.S. EPA, Research Triangle Park, North Carolina, USA
    Search for more papers by this author
  • Kirk R. Baker,

    1. Office of Air Quality Planning and Standards, U.S. EPA, Research Triangle Park, North Carolina, USA
    Search for more papers by this author
  • John B. Nowak,

    1. Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, Boulder, Colorado, USA
    2. Chemical Sciences Division, Earth System Research Laboratory, NOAA, Boulder, Colorado, USA
    Search for more papers by this author
  • Jennifer G. Murphy,

    1. Department of Chemistry, University of Toronto, Toronto, Ontario, Canada
    Search for more papers by this author
  • Milos Z. Markovic,

    1. Department of Chemistry, University of Toronto, Toronto, Ontario, Canada
    2. Now at NOAA and University of Colorado, Boulder, Colorado, USA
    Search for more papers by this author
  • Trevor C. VandenBoer,

    1. Department of Chemistry, University of Toronto, Toronto, Ontario, Canada
    2. Now at Department of Chemistry, Memorial University of Newfoundland, St. John's, Newfoundland and Labrador, Canada
    Search for more papers by this author
  • Raluca A. Ellis,

    1. Department of Chemistry, University of Toronto, Toronto, Ontario, Canada
    2. Now at Climate and Urban Systems Partnership, Franklin Institute, Philadelphia, Pennsylvania, USA
    Search for more papers by this author
  • J. Andrew Neuman,

    1. Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, Boulder, Colorado, USA
    2. Chemical Sciences Division, Earth System Research Laboratory, NOAA, Boulder, Colorado, USA
    Search for more papers by this author
  • Rodney J. Weber,

    1. School of Earth and Atmospheric Sciences, Georgia Institute of Technology, Atlanta, Georgia, USA
    Search for more papers by this author
  • James M. Roberts,

    1. Chemical Sciences Division, Earth System Research Laboratory, NOAA, Boulder, Colorado, USA
    Search for more papers by this author
  • Patrick R. Veres,

    1. Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, Boulder, Colorado, USA
    2. Chemical Sciences Division, Earth System Research Laboratory, NOAA, Boulder, Colorado, USA
    Search for more papers by this author
  • Joost A. de Gouw,

    1. Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, Boulder, Colorado, USA
    2. Chemical Sciences Division, Earth System Research Laboratory, NOAA, Boulder, Colorado, USA
    Search for more papers by this author
  • Melinda R. Beaver,

    1. Environmental Science and Engineering, California Institute of Technology, Pasadena, California, USA
    2. Now at National Exposure Research Laboratory, U. S. EPA, Research Triangle Park, North Carolina, USA
    Search for more papers by this author
  • Sally Newman,

    1. Division of Geological and Planetary Sciences, California Institute of Technology, Pasadena, California, USA
    Search for more papers by this author
  • Chris Misenis

    1. Office of Air Quality Planning and Standards, U.S. EPA, Research Triangle Park, North Carolina, USA
    Search for more papers by this author

  • This article was corrected on 23 JUL 2014. See the end of the full text for details.

Abstract

National ambient air quality standards (NAAQS) have been set for PM2.5 due to its association with adverse health effects. PM2.5 design values in the South Coast Air Basin (SoCAB) and San Joaquin Valley of California exceed NAAQS levels, and NH4+ and NO3 make up the largest fraction of total PM2.5 mass on polluted days. Here we evaluate fine-scale simulations of PM2.5 NH4+ and NO3 with the Community Multiscale Air Quality model using measurements from routine networks and the California Research at the Nexus of Air Quality and Climate Change 2010 campaign. The model correctly simulates broad spatial patterns of NH4+ and NO3 including the elevated concentrations in eastern SoCAB. However, areas for model improvement have been identified. NH3 emissions from livestock and dairy facilities appear to be too low, while those related to waste disposal in western SoCAB may be too high. Analyses using measurements from flights over SoCAB suggest that problems with NH3 predictions can influence NO3 predictions there. Offline ISORROPIA II calculations suggest that overpredictions of NHx in Pasadena cause excessive partitioning of total nitrate to the particle phase overnight, while underpredictions of Na+ cause too much partitioning to the gas phase during the day. Also, the model seems to underestimate mixing during the evening boundary layer transition leading to excessive nitrate formation on some nights. Overall, the analyses demonstrate fine-scale variations in model performance within and across the air basins. Improvements in inventories and spatial allocations of NH3 emissions and in parameterizations of sea spray emissions, evening mixing processes, and heterogeneous ClNO2 chemistry could improve model performance.

1 Introduction

U.S. Environmental Protection Agency's (U.S. EPA) integrated science assessment determined that PM2.5 (particulate matter with aerodynamic diameter < 2.5 µm) exposure has a causal or “likely to be causal” relationship with adverse health outcomes including mortality as well as with negative welfare effects [U.S. Environmental Protection Agency (U.S. EPA), 2009]. Accordingly, the national ambient air quality standards (NAAQS) for PM2.5 are currently set at 12 µg m−3 on an annual basis and 35 µg m−3 on a 24 h basis to protect human health. Despite much progress in recent decades [South Coast Air Quality Management District (SCAQMD), 2013; San Joaquin Valley Air Pollution Control District (SJVAPCD), 2012], PM2.5 design values (DVs) at monitors in the South Coast Air Basin (SoCAB) and San Joaquin Valley (SJV) of California remain well above the NAAQS levels. Note that a DV is a statistic that describes the air quality status of a given location relative to the level of the NAAQS. The highest 2010–2012 annual PM2.5 DV is 19 µg m−3 in SJV and 15.6 µg m−3 in SoCAB, while the highest 24 h PM2.5 DV is 59 µg m−3 in SJV and 37 µg m−3 in SoCAB [U.S. EPA, 2013]. At key monitors in SJV and SoCAB, ammonium (NH4+) and nitrate (NO3) made up more than 50% of total PM2.5 mass on highly polluted days in recent years [SCAQMD, 2013; SJVAPCD, 2012].

Photochemical models are often used to understand the processes that lead to elevated NH4+ and NO3 concentrations in SoCAB and SJV and the potential response of these pollutants to precursor emission reduction. Modeling studies have indicated that NOx emission reductions are effective for reducing NH4+ and NO3 concentrations in these areas [e.g., SCAQMD, 2013; SJVAPCD, 2012; Kleeman et al., 2005], and NOx emissions reductions are also considered desirable due to associated reductions in ozone during summer months [Pusede and Cohen, 2012; SCAQMD, 2013]. Some secondary organic aerosol (SOA) constituents such as organic nitrates observed in California could also decrease with decreasing NOx emissions [Rollins et al., 2012], although the overall impact of NOx emission reductions on SOA levels is an area of active research [e.g., Pye et al., 2013]. While reductions in NH4+ and NO3 following previous NOx emissions reductions point to the success of photochemical modeling for air quality management in California, simulating NH4+ and NO3 in SoCAB and SJV remains challenging due to the complex meteorology, emissions, and terrain in these regions.

Since SoCAB is often close to the center of the Pacific high-pressure system, SoCAB meteorology is frequently influenced by large-scale subsidence inversions that limit vertical ventilation of the basin, while mountain ranges limit horizontal ventilation. SoCAB is also impacted by local-scale meteorology including upslope-downslope mountain flows and land-sea breezes that transport air from the LA urban core into the eastern basin during the day with weaker return flows overnight [Jacobson, 2002; Lu and Turco, 1996]. The complex SoCAB flows transport diverse emissions including those from the western ports of LA and Long Beach, heavy-duty diesel vehicle activity associated with goods movement and the ports [Bishop et al., 2012], light-duty vehicle [Bishop et al., 2010], and other activity that supports a population of more than 15 million, and numerous dairy facilities in the eastern basin. Elevated NH4+ and NO3 concentrations occur in eastern SoCAB when NH3 emissions from dairy facilities are injected into air masses with aged NOx emissions from the LA urban core [Russell and Cass, 1986]. Multiday pollution episodes can also occur when pollutants from the previous nighttime residual layer are mixed to the surface in morning or when aged pollution in the Santa Monica Bay is reintroduced into SoCAB [Riedel et al., 2012; Young et al., 2012; Lu and Turco, 1996].

Air pollution processes in SJV are also driven by complex interactions between meteorology, emissions, and terrain. Severe wintertime PM2.5 episodes with high NH4+ and NO3 concentration develop in the valley under cool, stagnant, and humid conditions where air is constrained by mountain ranges in the horizontal and temperature (T) inversions in the vertical. Conceptual models of these episodes involve the trapping of primary PM2.5 emissions overnight in a shallow surface layer below a radiation inversion and the mixing of NH3 from daytime emissions in rural areas with NOx from urban areas in a valley-wide layer above the radiation inversion and below a persistent subsidence inversion aloft. The NH4+ and NO3 that forms aloft overnight are then mixed to the surface in the morning when the radiation inversion breaks [Herner et al., 2006]. In addition to having meteorology conducive to NH4+ and NO3 formation, SJV is impacted by diverse emissions associated with its population of about 4 million that is projected to increase substantially in coming decades [California Department of Finance, 2013; Hixson et al., 2012]. Also, large emissions of NH3 occur in SJV as a by-product of its agricultural sector. Seven of the top 10 agricultural counties in California's massive agricultural industry ($43.5 billion in 2011) are located in SJV, with Tulare, Fresno, and Kern Counties ranking as the top 3 counties in California in terms of livestock (cattle and calf) population [California Department of Food and Agriculture, 2013]. SJV is also a major corridor for goods movement and experiences substantial NOx emissions from heavy-duty diesel trucks that move goods between Mexico and Canada on I-5 and between major SJV cities on Highway 99. Although peak PM2.5 episodes occur in wintertime in SJV, this area is in nonattainment of the annual PM2.5 NAAQS and nonnegligible concentrations NH4+ and NO3 exist in all seasons.

Typical NH4+ and NO3 modeling assessments include a model performance evaluation where predictions are evaluated against data from routine monitoring networks such as U.S. EPA's Chemical Speciation Network (CSN) [e.g., Simon et al., 2012]. While these evaluations can demonstrate a model's ability to predict NH4+ and NO3 concentrations, they are generally of limited diagnostic value since CSN measurements have low time resolution (24 h), observations of precursor gases (i.e., NH3 and HNO3) and key meteorological variables are mostly unavailable, and network observations have limited spatial coverage. Thorough diagnostic evaluations can be performed using comprehensive data sets from intensive field campaigns to develop air quality models and provide additional support for their use in decision making. These evaluations are especially important for fine-scale simulations of secondary pollutants such as NH4+ and NO3 in complex environments such as SoCAB and SJV that resolve detailed air pollution processes under challenging conditions. Fine-scale modeling (i.e., 4 km horizontal resolution) is currently used for air quality management in California [SCAQMD, 2013; SJVAPCD, 2012] and in research studies focused on California [e.g., Angevine et al., 2012; Baker et al., 2013; Ensberg et al., 2013; Fast et al., 2012; Kelly et al., 2011; Ying, 2011], while national-scale assessments use coarser resolution (generally 12 to 50 km) for computational efficiency [Schiferl et al., 2014; Huang et al., 2013; U.S. EPA, 2012a; Walker et al., 2012; Heald et al., 2012].

In May–July 2010, the California Research at the Nexus of Air Quality and Climate Change (CalNex-2010) study was conducted to address issues related to air quality and climate change [Ryerson et al., 2013]. The vast CalNex-2010 data set includes airborne and ground measurements of NH3, HNO3, NO3, and NH4+ at hourly and subhourly time resolutions along with many complementary chemical and meteorology measurements in SoCAB and SJV. Here we evaluate the ability of a fine-scale photochemical model simulation to predict observed concentrations of PM2.5 NO3 and NH4+ at surface sites as well as NH3 and HNO3 along aircraft flight paths and at surface sites. Additionally, we evaluate predictions of many complementary chemical species and meteorological variables and conduct offline thermodynamic modeling to interpret gas-particle partitioning results.

2 Methods

Air quality was simulated over California from 4 May to 30 June 2010 with the Community Multiscale Air Quality (CMAQ) model version 5.0.1 (http://cmascenter.org/cmaq/). The model grid covered nearly all of California and Nevada as well as parts of the Pacific Ocean, Mexico, and Arizona with 4 km horizontal resolution and 34 vertical layers (see Figure S1 in the supporting information for model domain). Inorganic aerosol thermodynamics was simulated within CMAQ's modal aerosol formulations using ISORROPIA II [Fountoukis and Nenes, 2007], gas-phase chemistry was simulated with the Carbon Bond 05 chemical mechanism [Yarwood et al., 2005], and heterogeneous hydrolysis of N2O5 was treated according to Davis et al. [2008]. Chemical boundary conditions (3-hourly) were derived from a GEOS-Chem v8-03-02 global model simulation (http://acmg.seas.harvard.edu/geos/). The prognostic meteorological fields used to drive CMAQ were based on a simulation with the Weather Research and Forecasting (WRF; www.wrf-model.org) model version 3.4 using a grid consistent with that of CMAQ. WRFv3.4 was configured with the asymmetric convective model version 2 planetary boundary layer (PBL) scheme [Pleim, 2007] and the Pleim-Xiu land surface model [Pleim and Xiu, 2003]. A meteorological simulation for the same period and configuration used here but based on an earlier version of WRF (version 3.1) has been evaluated in detail previously [Baker et al., 2013].

Point source emissions for the modeling period were based on state-submitted 2010 emission totals and day-specific continuous emissions monitor data when available. Other anthropogenic, nonmobile source emissions were based on the 2008 National Emissions Inventory (NEI08) version 2 as processed using the Sparse Matrix Operator Kernel Emissions (SMOKE) model (http://www.cmascenter.org/smoke/). Mobile source emission totals were based on 2007 and 2011 emission totals provided by the California Air Resources Board (CARB) that were interpolated to 2010. Spatial and temporal allocations of mobile source emissions were based on SMOKE-MOVES (Motor Vehicle Emissions Simulator, http://www.epa.gov/otaq/models/moves) simulations. The scaling approach for combining CARB emission totals with SMOKE-MOVES allocations is described elsewhere [U.S. EPA, 2012b]. Biogenic emissions are estimated using the biogenic emissions inventory system version 3.14 with hour-specific T and solar radiation based on WRF output.

Information on the surface observation sites considered here is provided in Table S1, and site locations are illustrated in Figure S1 in the supporting information. Measurements of 24 h average inorganic PM2.5 concentration at the CSN sites were acquired from the CARB 2012 Air Quality Data DVD (www.arb.ca.gov/aqd/aqdcd/aqdcd.htm). Measurements from the CalNex-2010 study were obtained through online data resources for the Pasadena site (www.esrl.noaa.gov/csd/tropchem/2010calnex/Ground/DataDownload), the Bakersfield site (https://bspace.berkeley.edu/portal/), and the NOAA WP-3 aircraft (www.esrl.noaa.gov/csd/tropchem/2010calnex/P3/DataDownload). Summary information on the relevant CalNex-2010 measurements is provided in Table 1 along with references to primary publications. Additional CalNex-2010 measurements used in the supporting information are described there.

Table 1. CalNex-2010 Measurements Used Here
LocationMeasurementReferenceTechniqueSample Interval
  1. a

    Averaged to 1 h for comparison with CMAQ hourly average concentration predictions.

  2. b

    CMAQ instantaneous mixing ratios at each hour were interpolated in time to match 1 s aircraft measurements.

PasadenaNH3

Ellis et al. [2010]

Quantum cascade tunable infrared laser differential absorption spectroscopy1 h
PasadenaHCl and HNO3

Veres et al. [2008]

Negative ion proton transfer chemical ionization mass spectrometry (CIMS)1 mina
PasadenaNO3, SO42−, NH4+, Cl, Na+, and K+

Weber et al. [2001]

Particle into liquid sampling and ion chromatography (IC)10 mina
PasadenaPBL height

Haman et al. [2012]

Vaisala Ceilometer CL31 
PasadenaT, RH, wind speed, and wind direction Davis VantagePro2 weather station on Caltech Millikan Library 
BakersfieldNO3, SO42−, NH4+, Cl, NH3, SO2, and HCl

Markovic et al. [2012]

Ambient Ion Monitor-IC1 h
BakersfieldHNO3

Crounse et al. [2006]

CIMS0.5 sa
NOAA WP-3NH3

Nowak et al. [2007]

Protonated acetone dimer CIMS1 sb
NOAA WP-3HNO3

Neuman et al. [2002]

SiF5 CIMS1 sb

3 Results and Discussion

3.1 Evaluation Using CSN Observations

Predictions of 24 h average concentrations of inorganic PM2.5 components were evaluated against observations at CSN sites in SoCAB and SJV. Time series of modeled and observed NO3 concentrations at the sites are shown in Figure 1. Time series for NH4+, SO42−, Ca2+, Mg2+, Na+, and K+ are provided in Figures S2–S5, and a statistical comparison is provided in Table S3 for the metrics used in this study: normalized mean bias (NMB), normalized mean error (NME), root-mean-square error, and Pearson correlation coefficient (r).

Figure 1.

Comparison of 24 h average modeled and observed PM2.5 NO3 concentrations at CSN monitors in SJV and SoCAB: (a) Baker-5558Ca, (b) Fresno-1st, (c) Rvrside-Rubi, and (d) Ls_Ang-NMain.

The Riverside-Rubidoux monitor (Rvrside-Rubi) experienced the highest NO3 concentrations of the CSN sites with an observed mean ± standard deviation of 6.81 ± 3.47 µg m−3 (Figure 1c). Good correlation (r = 0.78) exists between model predictions and observations of NO3 at this monitor suggesting that the model generally captures the processes leading to NO3 formation. However, predictions are biased low during the period of elevated observed concentrations in June and the overall NMB is −39.8%. SO42− predictions are also well correlated with observations at this site (r = 0.8) with moderately low bias (NMB = −16.6%). Therefore, the underpredictions of NO3 are not due to reduced NH3 availability associated with SO42− overprediction (see section 3.2 for discussion of NH3 near Riverside). Riverside is known to experience high PM2.5 NO3 concentrations when aged urban NOx emissions mix with fresh NH3 emissions from dairy facilities in the area [Nowak et al., 2012]. The elevated NO3 concentrations predicted east of the LA urban core (Figure 2) are consistent with this conceptual picture and the gradient in observations between the LA (Ls_Ang-NMain) and Rvrside-Rubi sites. On 7 June, the peak modeled NO3 concentration of 11.88 µg m−3 occurs just north of the Rvrside-Rubi site in Fontana (Figure 2b) and is closer to the observed value at the site (11 µg m−3) than is the modeled value in the Rvrside-Rubi grid cell (7.25 µg m−3).

Figure 2.

Model predictions of 24 h average PM2.5 NO3 concentration (> 0.2 µg m−3) with overlay of CSN observations on two elevated concentration days: (a) 20 May and (b) 7 June. L = Ls_Ang-NMain, R = Rvrside-Rubi, B = Baker-5558Ca, F = Fresno-1st.

The mean of the NO3 observations at the Ls_Ang-NMain monitor between 8 May and 25 June is 3.69 ± 1.62 µg m−3. In contrast to results for Rvrside-Rubi, NO3 predictions at Ls_Ang-NMain are relatively unbiased (NMB = 2%) but are poorly correlated (r = 0.12) with observations. While the peak NO3 prediction occurs on the same day as the peak observation (7 June) and agrees within 13% of the observation on that day, NO3 is significantly overpredicted on 26 May (Figure 1d). SO42− is also overpredicted at Ls_Ang-NMain on this day (Figure S3) indicating that the NO3 overprediction is not due to an excess availability of NH3 associated with SO42− underprediction. Modeled NO3 was elevated over much of SoCAB between 24 May and 27 May (Figure S6), and this time period is discussed further below in the context of the Pasadena CalNex-2010 observations.

The model captured the observed spatial pattern of lower NO3 concentrations in SJV than SoCAB during the study period (Figure 2). At the Bakersfield CSN site (Baker-5558Ca), the mean modeled concentration was 0.72 µg m−3 while the observed mean was 1.02 µg m−3. Modeled NO3 was well correlated with observations at this site (r = 0.9). The overall low bias in model predictions at Baker-5558Ca (NMB = −31.6%) is driven largely by days after 13 June when observations were less than about 1 µg m−3, and model predictions were less than about 0.5 µg m−3 (Figure 1a). NO3- predictions were also biased low at the Fresno-1st site (NMB = −39.3%). Greater underprediction of the peak observed concentrations occurred in mid-May at Fresno-1st than at Baker-5558Ca. Differences in behavior at these sites could be due in part to the greater inflow in northern SJV of marine air, which enters the Central Valley through the Carquinez straight but typically does not reach Bakersfield in the south [e.g., Baker et al., 2013].

At the CSN sites, NH4+ predictions and observations are strongly correlated with 2SO42− + NO3 (2SN) on a molar basis (i.e., r ≥ 0.92; Table S4 and Figure S7) indicating that ISORROPIA II correctly captured the influence of acid condensation on NH4+ formation. Also, predicted and observed NH4+ molar concentrations were less than 2SN concentrations at all sites suggesting the possible influence of crustal cations and/or sodium on nitrate partitioning. The normalized mean difference (NMD) between NH4+ and 2SN was comparable for predictions and observations at Fresno-1st, Rvrside-Rubi, and Baker-5558Ca but was higher (i.e., NH4+ was too high relative to 2SN) for predictions (NMD = −13.7%) compared with observations (NMD = −26.9%) at Ls_Ang-NMain (Table S4). The excess modeled NH4+ could be related to underpredictions of Na+ that appear to impact gas-particle partitioning near LA (see section 3.4).

3.2 Evaluation Using NOAA WP-3 Aircraft Observations

Measurements of NH3 and HNO3 on the NOAA WP-3 aircraft provide snapshots of mixing ratios across SoCAB and SJV that are valuable for evaluating the spatial patterns of predictions of these NH4+ and NO3 precursors. Modeled and observed mixing ratios of NH3 along the 8 and 19 May flights are shown in Figure 3 for a select region of SoCAB and < 1500 m above ground level (agl). These daytime flights sampled air masses upwind and downwind of SoCAB dairy facilities and have been discussed in earlier studies [Neuman et al., 2012; Nowak et al., 2012]. The altitude range was selected to focus on air masses recently impacted by surface emissions and to minimize overplotting in the figures due to overlapping flight paths in the vertical. The spatial patterns of mixing ratio differences (Figure 3 (top)) demonstrate large underpredictions of NH3 to the east of Chino in the vicinity of Riverside (e.g., markers “(1)” and “(4)”), whereas overprediction tends to occur to the west of Chino and just south of the San Gabriel Mountains (e.g., marker “(3)”). The much higher observed peak NH3 mixing ratios on 19 May than 8 May are due to direct sampling of dairy facility plumes during aircraft approaches to the Chino airport on 19 May.

Figure 3.

Observed and predicted NH3 mixing ratios along select portions of NOAA WP-3 SoCAB flight segments and <  1500 m agl on 8 and 19 May 2010. (top) Modeled minus observed mixing ratio; (bottom) time series for predicted and observed mixing ratio on log scale. Numbers indicate locations downwind (1, 2, and 4) and upwind (3) of dairy facilities.

Nowak et al. [2012] estimated NH3 emissions from dairy facilities near Chino and from automobiles in the LA urban core based on downwind aircraft measurements. Their estimates indicate that automobile and dairy emissions may be of similar magnitude in the basin, but the dairy facility emissions have more temporal variability and are more concentrated spatially. Daily average NH3 emission totals based on the hourly average for 12–15 PST during 7 May to 30 June are compared with the Nowak et al. [2012] estimates in Table 2. These hours were selected to approximate the timing of the aircraft transects used in generating the Nowak et al. [2012] estimates. Modeled NH3 emissions from the agricultural sector in the relevant region of the eastern basin (28 t/d) are just below the lower end of the range estimated for dairy facilities by Nowak et al. [2012] (33–176 t/d). The modeled peak NH3 mixing ratio is a factor of 4 closer to the observed peak for the 16 May transect used by Nowak et al. [2012] in calculating the low-end emissions estimate than for the 14 May transect used in calculating the high-end estimate (Figure S8). Accurate characterization of the large magnitude and variability of NH3 emissions from dairy facilities present challenges for photochemical grid modeling of NH4+ and NO3 in eastern SoCAB. However, the NEI08 emissions used here for dairy facilities provide a much better representation of NH3 emissions in SoCAB than do the NEI05 [Schiferl et al., 2014], and model bias for NO3 predictions in Riverside is much smaller when using NEI08 than NEI05 emissions (not shown).

Table 2. Average NH3 Emissions (Metric Tons per Day)
SectorEastaWesta
  1. a

    Regions of SoCAB defined by Nowak et al. [2012].

  2. b

    Based on hours 12–15 PST for 7 May to 30 June of modeled NEI08 inventory used here.

Allb35.1843.01
Agriculture (includes dairy facilities)b28.051.07
Mobileb0.4410.19
Areab4.3229.98
Pointb2.361.76
Nowak et al. [2012] automobilesNA62 ± 24
Nowak et al. [2012] dairy facilities33 to 176NA

Emissions from automobiles in a western region of SoCAB were estimated to be 62 ± 24 t/d by Nowak et al. [2012]. Although modeled NH3 emissions from all sectors combined (43 t/d) are within the uncertainty of this estimate, modeled mobile emissions of NH3 are only 10.2 t/d in this region suggesting that mobile emissions of NH3 may be too low in the model. The area-source sector is the largest NH3 emission sector in this region of the western SoCAB, and the largest categories within this sector are related to waste disposal.

In Figure 4, observed and predicted mixing ratios of HNO3 are compared for the 8 May and 19 May flight paths just discussed for NH3. In contrast to results for NH3, mixing ratios of HNO3 are overpredicted to the east of Chino and underpredicted to its west and just south of the San Gabriel Mountains. Model bias for HNO3 predictions along these flight segments exhibits a negative correlation with that for NH3. When the magnitude of the difference between modeled and observed HNO3 mixing ratio is greater than 2 ppb, there is negative correlation (r = −0.63) between differences in modeled and observed mixing ratios for HNO3 and NH3. While HNO3 mixing ratios are impacted by factors including emissions, gas-phase chemistry, and meteorology, the negative correlation in errors for NH3 and HNO3 predictions suggests that overpredictions of HNO3 may be caused in part by insufficient NH3 in the model for forming NH4+ and NO3. The underprediction of NO3 in Riverside in early June discussed above could also be due in part to insufficient availability of NH3.

Figure 4.

Same as Figure 3 but for HNO3.

Differences between modeled and observed NH3 mixing ratios for the southern SJV flight on 16 June are shown in Figure 5. The model greatly underpredicts NH3 mixing ratios in areas near Hanford and Tulare with large agriculture and livestock NH3 sources. However, in western SJV and southeast of Bakersfield, the model overpredicts NH3 for this flight. This pattern could suggest a need for improved spatial allocation of NH3 emissions in SJV. Higher NH3 mixing ratios were observed along the southern SJV flight on 7 May than on 16 June. The 98th percentile of observed mixing ratios was 186 ppb for the 7 May flight, compared with 41 ppb for the 16 June flight (Figures S9 and S10). Model predictions are much less variable than the observations, and the modeled 98th percentile is 19 ppb for the 7 May flight and 17 ppb for the 16 June flight. Similar to the behavior in SoCAB, HNO3 tends to be overpredicted in highly concentrated NH3 plumes in SJV where NH3 is underpredicted. However, the model also tends to overpredict HNO3 near Bakersfield where NH3 predictions are relatively unbiased compared with aircraft observations (Figures S11 and S12).

Figure 5.

NH3 mixing ratio difference (modeled − observed) along select portions of SJV flight segments < 1500 agl on 16 June 2010. See Figure S10 for mixing ratios.

A qualitative comparison of the spatial patterns of peak mixing ratios observed over all flights and the spatial patterns of peak modeled emissions rates was also done (not shown). In SJV, NH3 emissions are distributed relatively evenly in the region where distinct peaks and sharp gradients in mixing ratios were observed. By contrast, in SoCAB, modeled emissions are highly concentrated near the large dairy facilities in the eastern basin where peak concentrations were observed. This qualitative comparison suggests that both the spatial allocation and magnitude of emissions require improvement in SJV, whereas the magnitude of emissions deserves more attention in SoCAB. Note also that the maximum modeled mixing ratios in the surface layer were less than about 21 ppb over all hours and grid cells in SJV and SoCAB on the flight days discussed above. Since this value is much less than the observed mixing ratios near major sources, the underpredictions of NH3 cannot be explained in terms of a slight mismatch in time or space of the modeled plume compared with the observed plume.

3.3 Evaluation Using CalNex-2010 Ground Site Data

Comprehensive data from the Pasadena and Bakersfield ground sites enable a thorough evaluation of model performance for NH4+, NO3, and precursors at key locations in SoCAB and SJV. The ground site data complement the 24 h average PM2.5 measurements at the CSN sites (section 3.1) and the short spatial snapshots of NH3 and HNO3 from the aircraft (section 3.2).

3.3.1 Pasadena

The time series of observed and predicted HNO3, NO3, total nitrate (TN) (i.e., HNO3 + NO3), and the molar gas-phase fraction of TN (GFN) (i.e., HNO3/(HNO3 + NO3)) at the CalNex-2010 Pasadena ground site is shown in Figure 6. In section 3.1, overprediction of NO3 at the LA CSN site on 26 May and underprediction at the Riverside site in early June was identified. These time periods are shaded grey in Figure 6, and data from the Pasadena site are used here to investigate causes of the errors.

Figure 6.

Comparison of measurements and predictions at Pasadena for the nitrate system. (a) HNO3, (b) PM2.5 NO3, (c) TN: HNO3 + NO3, (d) GFN: HNO3/ (HNO3 + NO3). Shaded regions indicate episodes discussed in section 3.3.1.

In the May period, daytime PBL height, T, and RH are generally simulated well (Figures S15–S17), although RH is overpredicted during the night and underpredicted during the day. The modeled PBL height also appears to underrepresent the degree of mixing during the evening transition from unstable daytime to stable nighttime conditions on 24 and 25 May. The underestimate of mixing in the model in the evening on 24 May seems to cause overpredictions of NH3 and NO (Figures 7a–7c). As a consequence of the overprediction of NO, titration of O3 in the model causes underpredictions of O3 (Figure 7d). Oxidation of NO and NO2 by O3 at night leads to production of N2O5 that is converted to HNO3 via heterogeneous reaction with H2O on particle surfaces. Such nocturnal HNO3 production could be overestimated for the chloride-containing particles in Pasadena (Figure S29) in publicly released versions of CMAQ, which do not include the conversion of N2O5 to ClNO2 on particle surfaces [Sarwar et al., 2012]. HNO3 production could also be overestimated because the Davis et al. [2008] parameterization used in CMAQ does not account for reduced N2O5 reaction probabilities on particles with organic coatings [e.g., Riemer et al., 2009]. The overprediction of TN associated with underestimated evening mixing and excessive oxidation of NOx to HNO3 translates into a large overprediction of NO3 because TN is predicted to partition almost entirely to the particle phase overnight, whereas observed partitioning is more moderate (Figure 6d). This overall behavior seems to explain the differences in timing of peak NO3 concentration for the model and observations on 24–25 May and other days (Figure 6b). However, the ceilometer measurements used here have difficulty detecting the PBL height during the transition from highly convective daytime conditions to stable nighttime conditions [Haman et al., 2012]. Such limitations are evident in Figure7a, where the observed values drop to about 500 m at 18 PST but then increase to over 1000 m for most of 19–21 PST. This oscillation seems unrealistic.

Figure 7.

Time series of modeled and observed (a) PBL height and mixing ratio of (b) NH3, (c) NO, and (d) O3 on 24–25 May.

In contrast to the overpredictions during 24–26 May, NO3 is underpredicted in Pasadena during the first week in June. Ryerson et al. [2013] describe how synoptic meteorology led to warm temperatures, subsiding air, and a prolonged O3 episode in SoCAB during this period. Underpredictions of TN and O3 during this period suggest that the model did not capture the impact that meteorology had on air quality by enhancing multiday pollution carryover and photochemical oxidation. During 1–7 June, the daytime (11–17 PST) average NOx-to-NOy ratio was 0.51 for the observations (compared with 0.59 at other times), while it was 0.73 for model predictions (and 0.73 at other times). The underestimate of NOx oxidation in the model could be due in part to oxidant limitations associated with NOx overpredictions (Table S8).

The model did capture the relatively low PBL heights during 1–7 June (Figure S17). However, underprediction of mixing during the evening PBL transition appears to occur on some days, and daytime peaks appear to be slightly overpredicted, possibly due to differences in PBL height definitions between the model and observations. RH and T predictions also agree well with observed values during this period. However, wind speeds were overpredicted with a NMB of 246% for 1–7 June compared with a NMB of 208% at other times (Figure S24). The average observed wind speed at the Millikan Library in Pasadena was only 0.61 m s−1 during 1–7 June, but it was higher at National Weather Service sites in downtown LA (KCQT; 1.84 m s−1), Riverside (KRAL; 3.17 m s−1), and near the coast (KLAX, KTOA, and KSMO; 3.45 m s−1). Wind speeds were overpredicted at the downtown LA site (NMB = 55%) but were underpredicted at the Riverside (NMB = −11%) and coastal sites (NMB = −24%) over the 7 May to 30 June period.

A statistical evaluation of model predictions for species just discussed as well as for additional inorganic gas and PM2.5 components is provided in the supporting information (Table S8). NH3 mixing ratios are overpredicted at Pasadena (NMB = 137%), and NH4+ concentrations are generally underpredicted (NMB = −45%) during the brief period when measurements are available (Figure S27). Also, Na+, Cl, and HCl are underpredicted (NMB ≤ −76%) and poorly correlated with observations (R ≤ 0.36) suggesting that the sea-salt emissions parameterization did not adequately represent the emissions flux, the modeled wind speeds used to drive emissions were too low, or that sea spray particles dry deposited too quickly in the model [Kelly et al., 2010].

3.3.2 Bakersfield

The model was also evaluated against measurements at the Bakersfield CalNex-2010 site, where meteorology was warmer and drier with higher wind speeds than at Pasadena (Figures S14 and S37). Diurnal box plots of observed and predicted concentrations of HNO3, NO3, TN, and GFN at Bakersfield are shown in Figure 8. Similar plots for the ammonia system are shown in Figure S34. The model correctly predicted that HNO3 and TN mixing ratios peak at about 11 PST, but daytime mixing ratios are substantially overpredicted leading to a NMB of 205% for HNO3 (Figure 8a and Table S10). While NOx and NOy tend to be overpredicted at the Bakersfield site, the daytime ratio of NOx to NOy is greater for predictions (74%) than observations (69%). Therefore, the overpredictions of TN appear to be related to overpredictions of NOy (a moderately conserved tracer for NOx emissions) rather than excessive oxidation of NOx. Such NOy overpredictions could be related to underpredictions of PBL height; however, daytime PBL heights appear to be simulated well in SJV during this period [Baker et al., 2013] suggesting that an overestimate of NOx emissions could be the primary cause. While HNO3 is overpredicted at all hours of the day, NO3 is typically underpredicted except at hours 5–6 PST when TN partitions more readily to the particle phase (Figures 8b and 8d).

Figure 8.

Comparison of diurnal distributions of measurements and predictions at the CalNex-2010 Bakersfield site for the nitrate system during 22 May to 28 June. (a) HNO3, (b) PM2.5 NO3, (c) TN: HNO3 + NO3, (d) GFN: HNO3/ (HNO3 + NO3). Circles indicate data more than 1.5 times the interquartile range from either end of the box.

Modeled NH4+ has a diurnal pattern similar to NO3 and was underpredicted at all hours except for roughly 5–7 PST (Figure S34). NH4+ was strongly correlated with 2SN in both measurements (r = 0.84) and predictions (r = 0.91) and was generally less than 2SN for the observations (NMD = −18.1%) and predictions (NMD = −29%). Differences in the relationship between NH4+ and 2SN for the model and observations could be related to different impacts of crustal cations on particle acidity. NH3 mixing ratios are underpredicted at Bakersfield for most hours except for those during and just after the modeled evening PBL transition (18–22 PST) when they are overpredicted. Too early stabilization of the modeled PBL could be the reason for this overprediction rather than the temporal allocation of NH3 emissions, since NOx is also overpredicted at this time, ozone is titrated, and NH3 emissions from the agricultural sector peak earlier in the day (~14 PST) (Figures S35 and S36).

3.4 Gas-Particle Partitioning

Differences in observed and predicted partitioning of TN between the gas and particle phases at the CalNex-2010 ground sites are evident in Figures 6d and 8d. Here we conduct 0-D ISORROPIA II simulations to explore possible causes of these differences. ISORROPIA II accounts for the impact of crustal cations and has been used successfully to interpret observations previously [Fountoukis et al., 2009; Zhang et al., 2002]. The metastable branch of ISORROPIA II, where salts do not crystallize and supersaturated solutions exist for RHs below the mutual deliquescence RH, is used in all calculations unless otherwise noted. This branch was selected based on the RH observations in Pasadena and because it is the approach used in the CMAQ model under investigation. Note that the phase states of inorganic salts in ambient particles are often uncertain due to deviations from laboratory measurements of deliquescence for bulk solutions [Robinson and Stokes, 1970] and efflorescence for levitated droplets [e.g., Chan et al., 1992]. In any event, gas-particle partitioning calculations for NH4+ and NO3 are often insensitive to phase state (e.g., see Langridge et al. [2012], and the Bakersfield discussion below).

At the Pasadena site, the model partitions too much of TN to the particle phase at night and too much to the gas phase during the day (Figure 6d). Some of the disagreement between predictions and observations in the early morning (3–6 PST) is due to the sensitivity of the chemical ionization mass spectrometry (CIMS) measurement to particle nitrate, wherein some NO3 volatilizes in the inlet and is measured as gas-phase HNO3. In Figure 9, GFN predictions of ISORROPIA II are shown as a function of TA (i.e., NH3 + NH4+), and TN based on simulations where values of T, RH, SO42−, Na+, K+, Ca2+, Mg2+, and total chloride (i.e., HCl + Cl) are fixed at the average values of the model predictions (Figure 9 (left)) or observations (Figure 9 (right)) for two time periods: 14–16 PST (Figure 9 (top)) and 0–3 PST (Figure 9 (bottom)). Since Ca2+ and Mg2+ were not measured, predicted values were used for these ions in calculations for the “observed-conditions” cases. The markers in Figure 9 correspond to modeled (open circles) and observed (squares) mixing ratios of TA and TN at matching hours.

Figure 9.

ISORROPIA II predictions of molar gas fraction of nitrate, HNO3/ (HNO3 + NO3), as a function of TA and TN based on average (left) modeled and (right) observed conditions at Pasadena for T, RH, SO42−, TCL, Na+, Ca2+, K+, and Mg2+ during (top) 14–16 PST and (bottom) 0–3 PST. Markers correspond to observed and predicted TA and TN concentrations at matching hours. Note: Ca2+ and Mg2+ were not measured, and so modeled values were used for these ions in observed-conditions cases.

For daytime conditions (Figure 9 (top)), GFN predictions of ISORROPIA II for “modeled conditions” (Figure 9 (left)) indicate a broader region of high partitioning to the gas phase than those for observed conditions (Figure 9 (right)) (see below for explanation). This pattern is consistent with the daytime overprediction of GFN in Figure 6d. Notice that the markers for model predictions overlap a region of high GFN (red region) for the modeled-conditions case but overlap a region of moderate GFN (orange region) for the observed-conditions case. For hours where measurements are available, predictions tend to underestimate the observed TN during the day; however, errors in TN and TA have a smaller impact on GFN than do errors in other aspects of the system during daytime hours.

For the nighttime period (Figure 9 (bottom)), the gradients in GFN as a function of TA and TN are similar for conditions based on modeled (Figure 9 (left)) and observed (Figure 9 (right)) values. However, TA mixing ratios are overpredicted, and the markers for predictions overlap a region of lower GFN than the region overlapped by markers for observations. The primary cause of underpredictions of GFN during the nighttime hours therefore appears to be overprediction of TA. These overpredictions could be caused in part by the apparent underpredictions of mixing in nighttime PBL (e.g., Figure 7a).

To identify which differences in modeled and observed conditions caused the differences in GFN predictions for daytime conditions (Figure 9 (top)), we conducted additional ISORROPIA II sensitivity simulations where select values in the modeled-conditions case were set to the values in the observed-conditions case. In Figure 10, ISORROPIA II predictions of GFN for the daytime observed-conditions case are compared with those for the base modeled-conditions case and the sensitivity modeled-conditions scenarios at observed and modeled levels of TA and TN. GFN predictions of ISORROPIA II are greater than 0.5 under both modeled and observed conditions, but predictions for the base modeled-conditions case are much higher than those under observed conditions (NMB = 31%). When the Na+ concentration for the daytime modeled-conditions case (4 nmol m−3) is increased to match the observed value (40 nmol m−3), more of TN partitions to the particle phase and better agreement occurs (Figure 10, blue markers; NMB = 10%). This behavior is consistent with the nature of Na+ as a nonvolatile cation. Differences in average RH during daytime for the model (RH = 43%) and observations (RH = 53%) also impact GFN, and good agreement between ISORROPIA II predictions occurs when RH and Na+ values in the modeled-conditions case are set to observed values (Figure 10, orange markers; NMB = 2.5%). The sensitivity run with increased Na+ concentration also yielded lower NH4+ concentration, and this behavior is consistent with the overprediction of NH4+ relative to 2SN noted in section 3.1 for the LA CSN site. Therefore, improving predictions of Na+ should improve predictions of gas-particle partitioning for both TN and TA. Na+ predictions could be improved by tailoring the model representation of the coastal surf zone as well as the size distribution and magnitude of the sea spray emissions flux to conditions along the SoCAB coast.

Figure 10.

ISORROPIA II predictions of molar GFN at observed and predicted TA and TN levels based on average observed and modeled conditions during 14–16 PST as in Figure 9. Markers correspond to the base modeled-conditions case and sensitivity cases where select values for the modeled-conditions case (i.e., Na+, RH, RH and Na+, and T) were set to observed values.

Similar analysis was conducted to understand gas-particle partitioning at the Bakersfield site, where GFN was overpredicted at all hours (Figure 8d). For nighttime hours (0–3 PST), the lower average RH for the model (RH = 44%) than the measurements (RH = 52%) contributed to the overprediction of GFN (Figure S45). Underpredictions of TA at some hours also appear to contribute to the GFN overpredictions at night (Figure S46 (bottom)). For daytime hours in Bakersfield, the RH was low (e.g., mean observed RH = 22%), and the particle ionic strength exceeded the cap of 100 m in ISORROPIA II (Figure S47). Although predictions of GFN for the metastable branch of ISORROPIA II are likely compromised under such conditions, they are in general agreement with those of the stable branch of ISORROPIA II for observed and modeled TA and TN levels (Figures S46 and S48 (top)).

Errors in predictions of gas-particle partitioning of TN in Bakersfield could also be due to errors in predictions of crustal cations and the lack of treatment of the effects of organic acids and amines on inorganic aerosol thermodynamics. For instance, overpredictions of GFN (and HNO3) might be explained by underpredictions of carbonate-containing dust, since carbonates can enhance condensation of gas-phase acids including HNO3 [e.g., Kelly and Wexler, 2005]. However, the presence of additional carbonates could worsen overpredictions of the gas fraction of TA at some hours by causing more of the total NHx to exist in the gas phase. Note that dust concentrations are likely to be relatively low and RH conditions high enough to be well represented by the metastable branch of ISORROPIA II during humid and stagnant wintertime PM2.5 episodes in SJV.

4 Conclusions

The fine-scale CMAQ simulation captured the basic features of NH4+ and NO3 formation in SoCAB and SJV. For instance, NO3 predictions were well correlated with CSN observations in Bakersfield and Riverside, and elevated NO3 concentrations were predicted downwind of the dairy facilities in eastern SoCAB. However, the examination of differences in model predictions and measurements indicates areas where increased understanding will improve model performance. First, inventoried NH3 emissions from dairy and livestock facilities in SoCAB and SJV appear to be too low and do not capture the apparently large variability in emissions from these facilities. Additionally, the spatial allocation of NH3 emissions in SJV may require development. Underpredictions of NO3 concentration in Riverside may be caused in part by NH3 emission deficits. Conversely, NH3 emissions may be too high near Pasadena, as NH3 mixing ratios were overpredicted there. Considering that NH3 emissions from automobiles in the LA area appear to be too low, the overpredictions of NH3 mixing ratios in Pasadena suggest that emissions related to other major sources in this region (i.e., waste disposal-related area sources) could be overestimated. Offline ISORROPIA II simulations suggest that overpredictions of TA at Pasadena can cause too much partitioning of TN to the particle phase at night, while underpredictions of Na+ can cause too much partitioning of TN to the gas phase during the day. Also, underpredictions of Cl could impact HNO3 production via heterogeneous N2O5 reactions at night. Therefore, sea spray emissions parameterizations should be revisited for the SoCAB coast, although underpredictions of wind speed over water and too rapid particle deposition could also contribute to sea salt-related prediction errors. Underpredictions of mixing during the evening PBL transition appears to concentrate pollutant precursors and lead to excess nitrate formation via the N2O5 pathway on some nights in Pasadena. This behavior highlights a need for better representations in WRF of urban heat island effects or other features that influence the evening PBL transition. Overall, the analyses demonstrate variations in model performance within and across the air basins and suggest that improvements in areas of emissions, meteorology, and chemistry modeling could improve fine-scale simulations for these highly populated NAAQS nonattainment areas.

Acknowledgments

Discussions with Prakash Bhave, Rohit Mathur, and Jon Pleim of U.S. EPA (ORD) contributed to this work. The authors thank Allen H. Goldstein, Robert J. Weber, and Ronald C. Cohen of UC Berkeley for providing meteorology, O3, and organic nitrate data at Bakersfield. We also thank Barry Lefer of the University of Houston for providing ceilometer and NOx data at Pasadena and Paul O. Wennberg of Cal Tech for providing HNO3 data at Bakersfield. The authors also recognize the contributions of Chris Allen, Allan Beidler, James Beidler, Alison Eyth, Rich Mason, Norm Possiel, Alexis Zubrow, and Tyler Fox, and we thank the three anonymous reviewers for helpful suggestion.

Erratum

  1. The citations for “Schifer et al.” have been corrected to “Schiferl et al.”

Ancillary