Modeling NH₄NO₃ Over the San Joaquin Valley During the 2013 DISCOVER-AQ Campaign

Abstract

The San Joaquin Valley (SJV) of California experiences high concentrations of particulate matter NH₄NO₃ during episodes of meteorological stagnation in winter. A rich data set of observations related to NH₄NO₃ formation was acquired during multiple periods of elevated NH₄NO₃ during the Deriving Information on Surface Conditions from Column and Vertically Resolved Observations Relevant to Air Quality (DISCOVER-AQ) field campaign in SJV in January and February 2013. Here NH₄NO₃ is simulated during the SJV DISCOVER-AQ study period with the Community Multiscale Air Quality (CMAQ) model, diagnostic model evaluation is performed using the DISCOVER-AQ data set, and integrated reaction rate analysis is used to quantify HNO₃ production rates. Simulated NO₃⁻ generally agrees well with routine monitoring of 24-hr average NO₃⁻, but comparisons with hourly average NO₃⁻ measurements in Fresno revealed differences at higher time resolution. Predictions of gas-particle partitioning of total nitrate (HNO₃ + NO₃⁻) and NHx (NH₃ + NH₄⁺) generally agree well with measurements in Fresno, although partitioning of total nitrate to HNO₃ is sometimes overestimated at low relative humidity in afternoon. Gas-particle partitioning results indicate that NH₄NO₃ formation is limited by HNO₃ availability in both the model and ambient. NH₃ mixing ratios are underestimated, particularly in areas with large agricultural activity, and additional work on the spatial allocation of NH₃ emissions is warranted. During a period of elevated NH₄NO₃, the model predicted that the OH + NO₂ pathway contributed 46% to total HNO₃production in SJV and the N₂O₅ heterogeneous hydrolysis pathway contributed 54%. The relative importance of the OH + NO₂ pathway for HNO₃ production is predicted to increase as NOx emissions decrease.

1. Introduction

The San Joaquin Valley (SJV or Valley) makes up the southern portion of California’s Central Valley and is formed by the coastal mountain ranges in the west, the Sierra Nevada mountains in the east, and the convergence of mountain ranges in the south at the Tehachapi mountains. SJV is about 400 km long and 60–100 km wide and includes parts or all of eight counties having a combined population of about 4.2 million (Calfornia Department of Finance, 2017). The Valley population is projected to increase rapidly in coming decades, by ~60% from 2016 to 2060 (Calfornia Department of Finance, 2017), which has implications for air quality planning (Hixson et al., 2012). SJV contains major cities such as Fresno (population ~520,000) and Bakersfield (population ~380,000), important oil and gas fields (CDOC, 2015; Gentner et al., 2014), and an extremely productive agricultural region (California Department of Food and Agriculture, 2016a). For instance, SJV had about 1.5 million dairy cows and produced about 36 billion pounds of milk (90% of CA total) in 2016 (California Department of Food and Agriculture, 2016b). The Valley is also a major north-south corridor for goods transport along Highway 99 in the east and Interstate 5 in the west. SJV’s terrain combined with pollutant emissions from the large population and economic activity leads to high concentrations of PM_2.5 (particulate matter with aerodynamic diameter ≤ 2.5 μm), particularly during periods of stagnant meteorology in winter months. SJV is in nonattainment of U.S. EPA’s primary national ambient air quality standards for PM_2.5 that are set to protect public health.

Air pollution in SJV has been studied for decades, and conceptual models of wintertime PM_2.5 formation in SJV have been developed, largely based on the 1995 Integrated Monitoring Study and the 2000/2001 California Regional PM₁₀/PM_2.5 Air Quality Study (CRPAQS; Herner et al., 2005, 2006; Pun & Seigneur, 1999; Watson et al., 1998; Watson & Chow, 2002). Briefly, high pressure systems over the Great Basin lead to subsidence temperature inversions over the Valley that limit daytime mixing heights from less than 400 to ~800 m for periods of days to more than a week. Radiation temperature inversions also form overnight and limit mixing of surface emissions to a ~30–50 m layer that is decoupled from the residual layer above. Primary carbonaceous particles are concentrated in the shallow nighttime surface layer. Between the radiation inversion and the subsidence inversion, air masses rich in oxides of nitrogen, largely from urban areas and major highways, mix with air masses rich in NH₃, largely from rural agricultural areas, in a valley-wide layer overnight. Ammonium nitrate (NH₄NO₃) forms in this layer and mixes to the surface in the morning when the radiation inversion breaks. Morning increases in NH₄NO₃at the surface therefore tend to coincide with decreases in carbonaceous PM_2.5 (Herner et al., 2006; Lurmann et al., 2006; Young et al., 2016). NH₄NO₃ makes up a large fraction of fine particle mass during major PM_2.5 episodes (e.g., L. W. A. Chen et al., 2007; Chow et al., 2006; Ge, Setyan, et al., 2012; Herner et al., 2005; San Joaquin Valley Air Pollution Control District [SJVAPCD], 2012). Persistent radiation fog also occurs in SJV in wintertime, and the chemistry of fine particles can be influenced by aqueous-phase processes (e.g., Collett, Hoag, & Rao, 1999; Collett, Hoag, Sherman, et al., 1999; Ge, Zhang, et al., 2012; Herckes et al., 2015; Jacob et al., 1986).

Air quality models have been used in combination with the CRPAQS data set to better understand air pollution processes in SJV. Overall, models did a reasonable job of predicting PM during CRPAQS (Kelly et al., 2011; Pun, Balmori, & Seigneur, 2009; Ying, Lu, Allen, et al., 2008; Y. Zhang et al., 2010) and were used to provide information on process rates, visibility impairment, and source apportionment and regional contributions to primary and secondary PM (J. Chen, Ying, & Kleeman, 2009, 2010; Ying, 2011; Ying & Kleeman, 2009; Ying, Lu, Kaduwela, et al., 2008; Ying, Lu, & Kleeman, 2009). Air quality models have also been used to understand the impact of precursor emissions on NH₄NO₃ (Blanchard et al., 2000; J. Chen et al., 2014; Kleeman et al., 2005; Livingstone et al., 2009; Pun, Balmori, & Seigneur, 2009; Pun & Seigneur, 2001; Stockwell et al., 2000). Generally, these studies found that NOx (NO + NO₂) emission reductions would be the most effective emission control for reducing NH₄NO₃ in SJV. Air quality management strategies based in part on NOx emission reductions, which are also important for reducing ozone in the Valley, have been implemented (e.g., SJVAPCD, 2012; SJVAPCD, 2016).

The studies discussed above were largely based on PM_2.5 episodes that occurred one to two decades ago. Although these studies are still relevant, air quality has improved over time in SJV due to reductions in NOx and other emissions (e.g., McDonald et al., 2012; Pusede et al., 2014, 2016; Pusede & Cohen, 2012; Russell et al., 2012). The DISCOVER-AQ (Deriving Information on Surface Conditions from Column and Vertically Resolved Observations Relevant to Air Quality) campaign in January and February 2013 provides a rich data set for analysis of more recent wintertime PM_2.5 episodes in SJV. Peak PM_2.5 concentrations were lower during DISCOVER-AQ than CRPAQS, but NH₄NO₃ still made up a large fraction of fine particle mass consistent with the earlier study. The DISCOVER-AQ data set has recently been used to investigate PM_2.5 precursor emissions and formation processes in SJV (e.g., Miller et al., 2015; Parworth et al., 2017; Prabhakar et al., 2017; Pusede et al., 2016; Shephard & Cady-Pereira, 2015; Sun et al., 2015; Young et al., 2016; X. L. Zhang et al., 2016). In particular, Pusede et al. (2016) used the DISCOVER-AQ data set in combination with the historical monitoring record to interpret past trends and predict future trends in NH₄NO₃ in SJV. They found that NH₄NO₃ formation is limited by NOx emissions, both daytime and nighttime formation pathways are important, and the daytime pathway will become the dominant source of wintertime NH₄NO₃ if future NOx emission reductions of 75% are achieved.

Previous studies using the 2013 SJV DISCOVER-AQ data set were generally based on conceptual, analytical, and box modeling in combination with the measurements. Compared with earlier field campaigns, limited regional photochemical modeling has been done for SJV DISCOVER-AQ. Regional photochemical modeling is valuable because it provides comprehensive information on key processes in three dimensions across the entire region for the entire period and is constrained by relatively few assumptions. The lack of constraints in air quality modeling is advantageous for exploring alternative scenarios but requires that models be thoroughly evaluated to insure they adequately reflect ambient processes. Modeling PM_2.5 episodes in SJV is particularly challenging because it involves simulating meteorology in complex terrain under low wind speed and vertically stratified conditions, representing complex pollutant emissions distributions, and simulating daytime and nighttime chemistry that can be influenced by the mixing of urban and rural air masses. Reliable modeling of PM_2.5 in SJV is important, however, to help inform air quality management for the highly populated nonattainment area. Here a diagnostic model evaluation is performed for regional photochemical modeling of NH₄NO₃ in the Valley during January and February 2013 using the DISCOVER-AQ data set. Integrated reaction rate (IRR) and process analysis modeling is also used to quantify HNO₃ production rates, interpret model predictions, and contribute to the understanding of air pollution in the Valley.

2. Methods

2.1. Modeling

Photochemical grid modeling was performed with the Community Multiscale Air Quality (CMAQ; www.epa.gov/cmaq) model version 5.1 (Appel et al., 2017) on a domain covering SJV from south of the Tehachapi mountains to north of Sacramento and parts of the Sierra Nevada mountains in the east and Pacific Ocean in the west (Figure S1). The CMAQv5.1 simulations were configured with IRR and process analysis (Jang et al., 1995; Jeffries & Tonnesen, 1994; Kim et al., 2014) and covered the 10 January to 10 February 2013 period after six days of spin-up from a separate simulation initialized on 22 December 2012. Horizontal grid resolution of 4 km was used with 35 vertical layers that matched the vertical structure of the meteorological model. Chemical boundary conditions were developed from a CMAQv5.1 simulation that covered the contiguous U.S. and surrounding areas with 12-km horizontal resolution. NH₃ surface exchange was simulated with CMAQ’s bidirectional exchange parameterization (Bash et al., 2013; Pleim et al., 2013), and gas-phase chemistry was parameterized with the CB05e51 mechanism (Appel et al., 2017). Inorganic aerosol thermodynamics were simulated with ISORROPIA II (Fountoukis & Nenes, 2007) in metastable mode, where crystallization does not occur. Semivolatile inorganic particle components (i.e., NO₃⁻, NH₄⁺, and Cl⁻) in the Aitken and accumulation modes are assumed to be in bulk equilibrium with their gas-phase counterparts (i.e., HNO₃, NH₃, and HCl) in CMAQ, whereas diffusive mass transfer is explicitly simulated for semivolatile coarse-mode particle components (Kelly et al., 2010). Heterogeneous hydrolysis of N₂O₅ on Aitken- and accumulation-mode particles is based on Davis et al. (2008), and N₂O₅ hydrolysis on coarse-mode particles is based on Bertram and Thornton (2009) as described by Sarwar et al. (2012).

Gridded emission fields for CMAQ modeling were developed with the Sparse Matrix Operator Kernel Emissions (SMOKE) model (Houyoux et al., 2000) version 3.7. The emission modeling procedures used here are similar to those described in detail previously for national 12-km resolution modeling (U.S. Environmental Protection Agency [USEPA], 2017c). Point source emissions were based on 2013 continuous emissions monitoring data when available and state submitted data otherwise. Anthropogenic nonpoint source emissions were based on version 2 of the 2011 National Emission Inventory (NEI11v2; USEPA, 2016). Onroad mobile source emission totals by county were estimated by interpolating totals from 2011 and 2014 based on EMFAC2014 (www.arb.ca.gov/emfac/) modeling by the California Air Resources Board (CARB). The interpolated onroad emission totals were then temporally and spatially allocated using results of a MOVES2014a (Motor Vehicle Emission Simulator; www.epa.gov/moves) simulation according to a hybrid procedure described previously (USEPA, 2012b, 2017c). Offroad mobile source emission totals were also based on information provided by CARB. The Biogenic Emission Inventory System version 3.6.1 was used with the Biogenic Emissions Landuse Database version 4.1 to estimate biogenic NO and speciated volatile organic carbon emissions (Bash, Baker, & Beaver, 2016). NH₃emissions from livestock were based on NEI11v2 annual county totals allocated to hour of day according to Zhu et al. (2015) using 2013 meteorological data as described by USEPA (2017b). Fertilizer application used in CMAQ’s bidirectional exchange parameterization was based on a simulation with the Environmental Policy Integrated Climate (EPIC) model (Cooter et al., 2012). Spatial allocation of livestock emissions and fertilizer application was based on agricultural land use classes of the 2011 National Land Cover Database.

The Weather Research and Forecasting (WRF) model (Skamarock et al., 2008) version 3.7 was used to generate gridded meteorological fields for CMAQ and SMOKE. WRFv3.7 was applied with 35 vertical layers from the surface to 50 mb with higher resolution near the surface to better resolve the planetary boundary layer (PBL). The model was run from 10 January to 10 February 2013 with 19 days of spin-up and no reinitialization. Boundary conditions were developed from a 12-km simulation that covered the contiguous U.S. (USEPA, 2017c). Four-dimensional data assimilation of wind speed, temperature, and moisture was used to nudge the atmosphere above the PBL according to the iterative approach described by Appel et al. (2014). Key physics options used in the WRF simulation include the Pleim-Xiu land surface model (Pleim & Xiu, 2003), asymmetric convective mixing model version 2 (ACM2; Pleim, 2007), RRTMG short and longwave radiation parameterization (Mlawer et al., 1997), and Morrison two-moment microphysics scheme (Morrison et al., 2009).

2.2. Measurements

Measurements of NH₃, TNO3 (HNO₃ + fine particle NO₃⁻), NO, NO₂, NOy (oxides of nitrogen including NOx, HNO₃, HNO₄, HONO, NO₃ radical, organic nitrates, and N₂O₅), O₃, and HCHO made from the National Aeronautics and Space Administration (NASA) P-3B aircraft during daytime flights on 16, 18, 20–22, and 30–31 January and 1, 4, and 6 February 2013 are used to examine model performance. The aircraft flew 2–3 repeated circuits per day over SJV including vertical spiral trajectories with ~5-km diameters over six sites (i.e., Bakersfield, Hanford, Tranquility, Fresno, Huron, and Porterville). Flights did not occur at night, and N₂O₅was not measured during SJV DISCOVER-AQ. NH₃ was measured with a cavity ring down spectrometer (CRDS; G2103, Picarro Inc.) and a proton-transfer-reaction time-of-flight mass spectrometer (PTR-ToF-MS). Measurements from these instruments have been compared previously and were found to provide complementary information (Sun et al., 2015). Therefore, both the CRDS and PTR-ToF-MS measurements are used here. TNO3 was measured on the P-3B aircraft by thermal dissociation of ambient NOy species followed by laser-induced fluorescence of NO₂. Specifically, TNO3 was calculated as NO₂ measured in the 600°C channel minus that measured in the 400°C channel with correction for slight conversion of HNO₃ in the alkyl nitrate channel (Pusede et al., 2016; Womack et al., 2017). NO, NO₂, O₃, and NOy were measured with the National Center for Atmospheric Research four-channel chemiluminescence instrument. The NOy measurement likely includes some contribution from NO₃⁻ in sub-1-μm particles, although the amount of contribution is uncertain. Airborne size distributions of particles with diameters between 90 and 7,500 nm were measured with a laser aerosol spectrometer (TSI Inc.) calibrated with polystyrene latex spheres. Airborne measurements of aerosol composition by a particle-into-liquid sampler (PILS) and off-line ion chromatography (IC) analysis showed that nitrate constituted 53% of the water-soluble aerosol mass. HCHO was measured with difference frequency generation absorption spectroscopy (Weibring et al., 2006). P-3B measurements were acquired from Revision 4 merged files available in the NASA online database (NASA, 2017).

NH₃ was also measured from a mobile ground laboratory that sampled conditions across the Valley during transects on 21–22 and 25–31 January and 1, 3–5, and 7 February 2013. Mobile measurements were performed with an open-path, quantum-cascade laser-based sensor mounted on the roof rack of a sedan passenger car (Miller et al., 2014; Sun et al., 2014) as described previously (Miller et al., 2015; Sun et al., 2015). Mobile laboratory data were acquired from Revision 0 files available online (NASA, 2017). At the CARB Fresno-Garland site, water soluble inorganic ions including NO₃⁻, NH₄⁺, SO₄²⁻, and K⁺ from particles with diameters of 3.0 μm or less were measured with subhourly resolution during 19 January to 10 February using a PILS-IC instrument, and water soluble gases including HNO₃and NH₃ were collected with ~5–7 hr resolution using annular denuders and analyzed off-line by IC (Parworth et al., 2017). These data were acquired directly from the authors Parworth et al., (2017), although the data are also available online (NASA, 2017). Meteorology measurements collected by CARB were acquired from NASA, (2017), radar profiler measurements at Visalia were obtained from NOAA (2017), and 24-hr average PM_2.5 NO₃⁻concentrations at SJV monitoring sites were obtained directly from CARB, although routine monitoring data are also available online (USEPA, 2017a).

2.3. Model-Measurement Pairing

Model predictions were generally paired with measurements according to standard practice by extracting predictions from the grid cell containing the measurement and then averaging the hourly mean model output to the sampling period of the measurement. To match model predictions with P-3B and mobile laboratory measurements, the grid cell containing the measurement at each second was identified, and instantaneous hourly predictions from that cell were linearly interpolated to the time of the measurement. The paired 1-s data were then averaged to 10-s resolution for the boxplot comparisons below. For spatial comparisons of CMAQ predictions with mobile laboratory NH₃ measurements, medians of subcell median mixing ratios were used to ensure adequate grid cell coverage of the measurements and reduce the influence of near-source sampling as follows. First, the 4-km CMAQ grid cells were decomposed into 1-km subcells, and grid cells with measurements in at least four subcells were selected. Second, median NH₃ mixing ratios in each subcell were calculated from the 1-s paired model-measurement data. Finally, the median mixing ratio for a 4-km grid cell was calculated as the median of the subcell median values. For spatial comparisons of CMAQ predictions and P-3B measurements, mean or median mixing ratios were calculated from the 1-s paired model-measurement data over samples within the modeled PBL during 11–15 PST for grid columns with measurements on at least four days. Modeled PBL heights were well correlated with PBL heights estimated from measurements during P-3B spirals, but predicted values were moderately biased low (12–34%; Figure S2).

3. Results and Discussion

NH₄NO₃ in fine particles is generally considered to be in thermodynamic equilibrium with NH₃ and HNO₃ for time scales of relevance to regional air quality modeling (e.g., Fountoukis et al., 2009; Meng & Seinfeld, 1996). Evaluations of model predictions of NH₃, HNO₃, NOx, and NOy are therefore relevant for understanding the model’s ability to simulate NH₄NO₃. In section 3.1, NH₃ predictions are compared with measurements from the mobile ground laboratory and the P-3B aircraft. In section 3.2, predictions of NOx, NOx/NOy, and TNO3 are compared with measurements from the P-3B aircraft. Routine network observations of NO₃⁻are also used to understand the model’s ability to simulate NO₃⁻ across the Valley. In section 3.3, the NH₄NO₃ system is considered at the Fresno site where a comprehensive data set allows for detailed investigation. Finally, in section 3.4, model predictions of HNO₃production rates are presented to contribute to understanding of the spatial and temporal patterns of nitrate production in the Valley. The term NH₄NO₃ is used here for convenience and is not meant to imply a solid phase state. For supersaturated conditions and for stable equilibrium conditions at relative humidities (RHs) greater than the mutual deliquescence RH (MDRH) of the inorganic system, NH₄NO₃ would partially or completely dissociate into NH₄⁺ and NO₃⁻ ions in aqueous solution (e.g., Kelly et al., 2008; Nenes et al., 1998; Wexler & Seinfeld, 1991). Since RH is often high in winter in SJV, CMAQ’s assumption that NH₄NO₃completely dissociates into aqueous solution at all RHs is generally a good one, except possibly during afternoon hours as discussed below.

3.1. Examining NH₃ Mixing Ratios

Average modeled NH₃ mixing ratios over SJV during 15 January to 5 February 2013 are shown in Figure 1a. Mixing ratios greater than about 7 ppb are predicted throughout SJV, and mixing ratios greater than 20 ppb occur in regions just west of Fresno, around Bakersfield, and a large portion of the eastern side of the Valley between Bakersfield and Fresno. The spatial patterns of elevated NH₃ mixing ratios follow the spatial patterns of NH₃emissions (Figure 1b) closely. NH₃ emissions occur primarily during daytime (Figure S3a) due to the combination of increased emission-related activity and conducive meteorology (e.g., Lonsdale et al., 2017; Zhu et al., 2015). On average, NH₃ deposition fluxes in the boxed region of Figure 1b were 43% of the emission fluxes during 10–16 PST, and vertical transport of NH₃ from model layer 1 was 55% of the emission fluxes (Figure S3b). This behavior is consistent with a previous study for the eastern U.S. (Dennis et al., 2010) and explains the correspondence in spatial patterns of NH₃ emissions and model surface layer concentrations in Figure 1.

Figure 1.

Average NH₃ (a) surface layer mixing ratios predicted by CMAQ and (b) gridded emissions with box defining region for discussion (see text) during 15 January to 5 February 2013. Markers for P-3B spiral locations are also shown.

Model predictions of NH₃ are compared with measurements from the mobile ground laboratory in Figure 2. These comparisons were done by matching CMAQ predictions in space and time with the measurements for all transects and then calculating the median modeled and measured mixing ratio by CMAQ grid cell from median values in 1-km subcells as described above. Model predictions are scaled by two in Figure 2a to better illustrate spatial patterns. The model underestimates mixing ratios considerably in regions where elevated values (> ~20 ppb) were measured (Figure 2b). Yet the model correctly estimates that NH₃ mixing ratios are elevated just southwest of Turlock, near Fresno, and in a region to the southeast of Hanford and that NH₃ mixing ratios are relatively low on the western side of the Valley. The model-measurement comparison is complicated by the nonuniform sampling and wide range of scales represented by the high-resolution ground measurements compared with the 4-km horizontal resolution of the air quality model. The qualitative conclusion of underestimated NH₃ in high emission regions based on the aggregated results in Figure 2 appears to be robust though based on additional NH₃evaluation discussed below. Also, previous comparisons of CMAQ and GEOS-Chem predictions with NH₃ measurements in SJV yielded similar conclusions as here for the May–June 2010 CalNex period (Kelly et al., 2014; Schiferl et al., 2014) and earlier periods (Heald et al., 2012; Walker et al., 2012). Model predictions of NO₃⁻ appear to be insensitive to the NH₃underpredictions in the high emission regions though. For instance, in a simulation with NH₃emissions doubled in the boxed region of Figure 1b, average NO₃⁻ concentrations changed by <5% in 93% of SJV grid cells having NO₃⁻ concentrations >5 μg/m³ and the maximum change was 13%.

Figure 2.

Median observed and 2× median modeled NH₃ mixing ratio by CMAQ grid cell over all mobile ground laboratory sampling transects and (b) difference in median values. See text for description of grid cell median calculations.

Median modeled NH₃ mixing ratios are compared with CRDS and PTR-ToF-MS measurements from the P-3B aircraft in Figure 3. Model results are scaled by three in the figure to better illustrate spatial patterns. Similar to results of the mobile ground laboratory comparison, the spatial patterns of model predictions are in general agreement with P-3B NH₃ measurements, but model predictions are too low in areas where elevated mixing ratios were observed. One location where relatively large underpredictions are evident is Hanford, which is located just outside of the high emission and concentration region in the model (Figure 1). NH₃ measurements from the CRDS and PTR-ToF-MS are in good agreement near the surface during morning and afternoon P-3B aircraft spirals over Hanford (Figure S4) and indicate that median modeled mixing ratios were underpredicted by a factor of 7–9 in the 0–900 m altitude range. For the mobile laboratory comparison in Figure 2, median NH₃ mixing ratios over Hanford were underpredicted by a factor of 5. Considering that Hanford is located just outside of a high emission region in the model, further examination of the spatial allocation of NH₃ emissions in this area is warranted. Modeled PBL heights were in reasonable agreement with empirical estimates at the Hanford site (normalized mean bias: −12%; Figure S2), and so errors in mixing height predictions are unlikely to explain the model-measurement differences.

Figure 3.

Median modeled and measured NH₃ within the modeled PBL during 11–15 PST by model grid cell for P-3B flights in January and February 2013.

3.2. Examining NOx, NOx/NOy, and NO₃⁻ Concentrations

Average model predictions of NOy during 15 January to 5 February are shown in Figure 4a. Relatively high mixing ratios are predicted over Fresno, Bakersfield, and northern SJV cities (e.g., Modesto and Stockton) and along Highway 99 between these cities. Average concentrations of fine particle NO₃⁻ (Figure 4b) are more uniformly distributed across the Valley than are mixing ratios of NOy, which are elevated in areas with high NOx emissions. The formation of NO₃⁻ requires the oxidation of NOx to HNO₃ and is promoted by the mixing of urban air masses with air rich in NH₃ from surrounding areas. These dependencies help explain the broader average spatial distributions of NO₃⁻ than NOy and NH₃ (cf., Figure 1a). Also, dry deposition velocities of HNO₃ and NH₃ are generally high compared with those of fine particle NO₃⁻ and contribute to sharper spatial gradients in NOy and NH₃.

Figure 4.

Average modeled (a) NOy (including fine particle NO₃⁻) mixing ratios in SJV with markers for P-3B spiral locations and (b) fine particle NO₃⁻ concentrations with markers for PM_2.5monitoring locations during 15 January to 5 February 2013.

NOx and NOy were measured during a series of P-3B spirals over sites including major cities in the north (Fresno) and south (Bakersfield), rural locations in the west (Tranquility and Huron), and the Hanford site discussed above. In Figure 5a, model predictions of NOx are compared with measurements during the aircraft spirals. The model correctly predicted that the highest mixing ratios occurred in Bakersfield and Fresno, relatively low mixing ratios occurred in Tranquility and Huron, and mixing ratios generally decreased with altitude. Yet NOx predictions were biased high in the 0–300 m bin at Bakersfield, Fresno, and Tranquility. The ratios of NOx-to-NOy are shown in Figure 5b, where modeled NOy was calculated by summing gas-phase NOy components and 20% of fine particle NO₃⁻. Size distribution measurements during the flights indicate that the majority of fine particle NO₃⁻ existed in particles with diameters less than 500 nm (Figure S5) and suggest that a significant, although unknown, fraction of fine particle NO₃⁻ was measured by the NOy instrument. NOx-to-NOy ratio comparisons based on modeled NOy with 0% and 100% of modeled NO₃⁻ included in the NOy calculation are provided in Figure S6. The model captured the general pattern of relatively high NOx-to-NOy ratios in urban areas (e.g., Bakersfield), where fresh NOx emissions comprise a large fraction of NOy, and relatively low NOx-to-NOy ratios in remote areas (e.g., Huron), where NOx oxidation products comprise a large fraction of NOy (Figure 5b). The model also captured the generally decreasing trends of NOx-to-NOy ratios with altitude. The overestimates of the NOx-to-NOy ratios in Fresno, Bakersfield, and Tranquility in Figure 5b suggest that the overpredictions of NOx in Figure 5a could be due in part to too-low modeled oxidation rates. However, the NOx-to-NOy evaluation is limited by uncertainty in the fraction of particle NO₃⁻ measured by the NOy instrument. Underpredictions of HCHO and O₃ during the aircraft spirals suggest that modeled oxidation rates may have been too low over the sites (Figure S7).

Figure 5.

Comparison of modeled and measured (a) NOy and (b) NOx/NOy mixing ratio distributions for 300-m altitude ranges for P-3B aircraft spirals (see Figure 4a for site locations). Boxes bracket the interquartile range (IQR), lines within the boxes represent the median, and whiskers represent 1.5 times the IQR from either end of the box.

In Figure 6, average TNO3 mixing ratios are shown for P-3B measurements at altitudes within the modeled PBL during 11–15 PST by model grid cell for cells with measurements on at least four days. Modeled TNO3 mixing ratios were generally biased high compared with the measurements, especially along Highway 99 between Fresno and Bakersfield. The relatively large TNO3 overpredictions between Bakersfield and Fresno resulted in weaker daytime gradients between the cities and surrounding areas for the model than were identified by Pusede et al. (2016). Modeled TNO3 was biased low relative to the ground site measurements in Fresno (see section 3.3).

Figure 6.

Average modeled and measured TNO3 within the modeled PBL by model grid cell over P-3B flights in January and February 2013 and (b) difference between modeled and measured TNO3.

In Figure 7, model predictions are compared with routine observations of 24-hr average fine particle NO₃⁻ at four sites spanning SJV from north (Modesto) to south (Bakersfield; see Figure 4b for site locations). A peak in the NO₃⁻ time series was observed at all sites on 22 January. The model performed well on this day for all sites except Bakersfield for which observations were underpredicted. On 3 February, high NO₃⁻ concentrations (>20 μg/m³) were observed in the south (Bakersfield and Visalia) and lower concentrations (<10 μg/m³) were observed in the north (Modesto and Fresno). The model underpredicted the NO₃⁻peaks on 3 February at Bakersfield and Visalia. The root-mean-square error (RMSE) for predictions increased in magnitude from north (1 μg/m³) to south (7 μg/m³), whereas correlation coefficients were high (r ≥ 0.78) at all sites. Additional model performance statistics are provided in Table S1. Model performance generally met benchmarks proposed by Emery et al. (2017).

Figure 7.

Comparison of 24-hr average PM_2.5 NO₃⁻ predictions of CMAQ with routine monitoring measurements at sites shown in Figure 4b.

The mean modeled PBL height at 15 PST was 320 m during the 18–22 January period when elevated NO₃⁻ was simulated in Modesto and Fresno and was 490 m during the 1–5 February period when the model predicted lower NO₃⁻ concentrations (Figure S8). Wind speeds were also lower during the January period (mean: 1.4 m/s) than the February period (mean: 1.9 m/s). Compared with profiler measurements at Visalia, wind speeds were biased low in mid-January and were biased high near the surface in early February (Table S3). Considering that meteorological stagnation is central to the conceptual model of NO₃⁻formation and build-up in SJV, the relatively low NO₃⁻ concentrations simulated during the early February period are probably related to the greater transport and mixing in the model. The relatively large NO₃⁻ underpredictions at Bakersfield, where meteorology is more influenced by the convergence of mountain ranges to the south, may be attributed to challenges in simulating meteorology in complex terrain. Model performance statistics for meteorological variables are provided in Table S2. Model bias and error for temperature and wind speed met benchmarks proposed by Emery et al. (2001) at the Fresno site supporting examination of the NH₄NO₃ system in Fresno.

3.3. Examining the NH₄NO₃ System in Fresno

A relatively complete set of measurements for evaluating the NH₄NO₃ system were made during 19 January to 10 February 2013 at the CARB Fresno-Garland site. In Figure 8, predictions of fine particle NO₃⁻ are compared with PILS-IC measurements at this site. Two major NO₃⁻ episodes were identified in Fresno during the campaign from about 14–23 January and 29 January to 5 February (Young et al., 2016). During the first episode, the model overpredicted the peak NO₃⁻ concentration on 22 January (Figure 8). The modeled peak is due to overnight transport of NO₃⁻ from the south (Figure S10), where modeled production of HNO₃ was particularly high around Visalia during this period (see section 3.4). Modeled wind speeds were low in Visalia in reasonable agreement with observations (Figures S11–S13). However, observed winds at sites in SJV were relatively disorganized overnight compared with model predictions and suggest that the model overestimated transport of NO₃⁻ to Fresno on 22 January. In early February, the model underpredicted the elevated NO₃⁻ concentrations in Fresno. As discussed above, modeled wind speeds and PBL heights were relatively high across SJV during the February period, and modeled NO₃⁻concentrations were relatively low. Comparisons of predictions of SO₄²⁻, NH₄⁺, K⁺, and Cl⁻with PILS-IC measurements are provided in Figure S14. Underpredictions of the generally low measured Cl⁻ concentrations (mean: 0.4 μg/m³) are consistent with findings of studies of other parts of the U.S. (Kelly et al., 2010; Kelly et al., 2014; Kelly et al., 2016; Simon et al., 2010).

Figure 8.

Comparison of model predictions of fine particle NO₃⁻ with PILS-IC measurements at the Fresno-Garland ground site.

Distributions of hourly average modeled and measured NO₃⁻ concentrations in Fresno are shown in Figure 9 for the January and February episodes. Measured concentrations increase in the morning during both periods in a pattern consistent with mixing of NH₄NO₃ from the nocturnal residual layer to the surface during development of the daytime boundary layer (Parworth et al., 2017; Prabhakar et al., 2017; Young et al., 2016). The 75th percentiles of modeled concentrations increase in the morning during the 19–25 January episode, but median concentrations are relatively constant compared with the measurements. The morning increase in NO₃⁻ is also underpredicted during 30 January to 5 February. In the afternoon, measured NO₃⁻ concentrations reach a relatively constant level during the first period and decrease during the second period, whereas modeled concentrations decrease in the afternoon during both episodes (Figure 9).

Figure 9.

Hourly average modeled and measured NO₃⁻ distributions at Fresno ground site during two periods of interest. Boxes bracket the IQR, lines within the boxes represent the median, whiskers represent 1.5 times the IQR from either end of the box, and circles represent individual values less than and greater than the range of the whiskers.

In Figure 10, concentrations of TNO3 and NHx and the percentage of the total concentrations in the gas phase are shown during 19–31 January when model performance for NO₃⁻ was relatively good. The model is biased 27% low for TNO3 and 36% low for NHx during this period at the Fresno site. However, the model correctly predicts that most of NHx is in the gas phase and most of TNO3 is in the particle phase. This gas-particle partitioning behavior suggests that HNO₃ is the limiting precursor for NH₄NO₃ formation in SJV in both the model and ambient. Sensitivity simulations with reductions in NH₃ and NOx emissions were conducted and confirmed that HNO₃ is the limiting precursor in the model.

Figure 10.

Modeled and measured concentrations of (a) TNO3 and NHx and (b) percentage of total in the gas phase during 19–31 January at the Fresno ground site.

Although gas-particle partitioning is generally predicted well, the fraction of TNO3 in the gas phase is sometimes overestimated in the model (Figure 10b). The overestimates of partitioning to the gas phase appear to be driven primarily by meteorology (i.e., RH and temperature, T) rather than issues with particle composition predictions. The modeled gas-phase fraction of TNO3 is relatively high when RH is less than 50% and T is greater than 285 K (Figure S15a). The overpredictions of the gas-phase fraction of TNO3 under these conditions could be due in part to challenges in representing the particle phase state under low RH conditions. Recall that the model assumes that crystallization does not occur and inorganic components exist as ions in supersaturated solutions for low RH (e.g., RH < MDRH). Previous studies have found that this assumption yields lower predicted NO₃⁻concentrations compared with the stable equilibrium assumption for RH < ~50% (Ansari & Pandis, 2000; Fountoukis et al., 2009). To investigate the issue here, off-line simulations with ISORROPIA II were performed for cases of stable (i.e., including crystallization) and metastable (i.e., no crystallization) equilibrium using T, RH, and concentration inputs based on CMAQ output for hours where the sampling period average RH was <50%. These simulations confirmed that the phase state assumption influences partitioning predictions under the low-RH conditions in Fresno. For hours with RH between 37% and 54%, the average percentage of TNO3 in the gas phase was 50% for simulations based on the metastable assumption and 24% for the stable assumption. Segregation of results by time of day (Figure S15b) reveals that the overpredictions of partitioning of TNO3 to the gas phase occur in the afternoon. The overestimate of the decreasing trend in NO₃⁻ concentration in the afternoon in the top panel of Figure 9 could therefore be due in part to gas-particle partitioning prediction issues, which are sensitive to particle phase state assumptions under low RH conditions. Deposition rates of TNO3 are relatively large in the afternoon due to the relatively low atmospheric resistance of the convective boundary layer (Figure S16). The average simulated deposition velocity was 2.83 cm/s¹ for HNO₃ and 0.07 cm/s for accumulation mode particles during 12–17 PST for the 19–31 January period. Given the relatively high deposition velocity of HNO₃ compared with that of fine particle NO₃⁻, excessive partitioning of TNO3 to the gas phase could lead to excessive removal of TNO3 through HNO₃ dry deposition in the afternoon.

3.4. Examining HNO₃ Production

Previous studies and the current modeling indicate that the limiting precursor for NH₄NO₃formation in SJV is HNO₃. Understanding chemical production of HNO₃ is therefore important for understanding NH₄NO₃ formation. HNO₃ production during daytime when OH levels are high is typically dominated by R1:

$${\text{NO}}_2+\text{OH}\to {\text{HNO}}_3.$$ (R1)

At night, when OH mixing ratios are low and photolysis of NO₃ radical is negligible, heterogeneous hydrolysis of gas-phase N₂O₅ with particle-phase H₂O is important. This process has been represented with the overall formula (Sarwar et al., 2012; Sarwar et al., 2014).

$${\text{N}}_2{\text{O}}_5{\text{+H}}_2\text{O(p)+Y\hspace{0.17em}}{\text{CI}}^{-}\text{(p)}\to {\text{Y(HNO}}_3{\text{+CINO}}_2)+2(1{\text{-Y)HNO}}_3$$ (R2)

where Y is the yield of ClNO₂ (Bertram & Thornton, 2009; Roberts et al., 2009). O₃ is an important oxidant in the production of N₂O₅ at night:

$$\text{NO}+{\text{O}}_3\to {\text{NO}}_2{\text{+O}}_2$$ (R3)

$${\text{NO}}_2{\text{+O}}_3\to {\text{NO}}_{\text{3}}$$ (R4)

$${\text{NO}}_2{\text{+NO}}_3\leftrightarrow {\text{N}}_2{\text{O}}_5$$ (R5)

Hourly 75th percentile HNO₃ production rates based on IRR results for R1, R2, homogeneous hydrolysis of N₂O₅ with water vapor, and heterogeneous hydrolysis of organic nitrates over Fresno are shown in Figure 11a for model layers 1, 5, and 7 during 17–22 January. Reaction of NO₂ with OH (R1) dominates HNO₃ production in all layers during daytime. Overnight, heterogeneous N₂O₅ hydrolysis (R2) dominates production in layers 5 and 7. This HNO₃ can condense to form fine particle NO₃⁻ and increase surface NO₃⁻ concentrations in the morning as the daytime boundary layer develops (e.g., Figure 9). In the surface layer overnight, R1 and R2 contribute significantly to HNO₃ production over Fresno in the model. OH mixing ratios that drive R1 are typically low at night because photolysis reactions important for OH production are negligible. The primary source of OH in the model at night is the reaction NO + HO₂ → OH + NO₂. This reaction is important in the model surface layer over Fresno because of the substantial NO emissions and the limited vertical mixing at night. HO₂ sources in the model that do not directly depend on sunlight include reactions of organics with O₃ and NO. Measured increases in surface NO₃⁻ concentrations in Fresno in the morning suggest that production in the ambient surface layer over Fresno is relatively small compared with production aloft. Therefore, there is evidence that modeled HNO₃production in the nighttime surface layer over Fresno is too high. Also, O₃ mixing ratios in the surface layer are overestimated at the Fresno site overnight during this period (Figure S17a). Observations indicate that O₃ is almost entirely depleted at the site on most nights due to the high NOx levels and reactions such as R3 and R4. NOx mixing ratios are lower in the model than the ambient overnight and enable partial recovery of O₃ mixing ratios following decreases during the evening rush hour when NOx emissions are high.

Figure 11.

Hourly 75th percentile HNO₃ production rates by chemical pathway over Fresno for the (a) base simulation and (b) simulation with K_z,min = 0.01 m²/s for model layers 1, 5, and 7 during January 17–22. N2O5homog: homogeneous gas phase reaction of N₂O₅ + H₂O; NTRhyd: heterogeneous hydrolysis of organic nitrates; N2O5hyd: heterogeneous hydrolysis of N₂O₅; and OH_NO2: reaction of OH + NO₂.

The apparently excessive production of HNO₃ in the model surface layer over Fresno at night appears to be due to overpredictions of O₃ mixing ratios. The cause of high O₃ mixing ratios in the surface layer in the model is vertical transport from higher layers. To test the impact of vertical mixing at night on the production of HNO₃ over Fresno, a sensitivity simulation was conducted where CMAQ’s parameterization for the minimum eddy diffusivity (K_z,min) was replaced by a fixed K_z,min of 0.01 m²/s in all grid cells as is done in ACM2 in the WRF model. The K_z,min change reduced vertical mixing of species overnight over Fresno, because ACM2 in CMAQ uses higher K_z,min values in urban areas (USEPA, 2012a). O₃depletion in the surface layer was nearly complete overnight during 17–22 January in the simulation with reduced vertical mixing (Figure S17b), and HNO₃ production in the surface layer was significantly reduced compared with the base simulation (Figure 11b). However, the underestimate of the morning increase in NO₃⁻ was not resolved by reducing K_z,min. Modeled advection of the nocturnal residual layer from Fresno to the south likely contributed to the underestimate of the morning increase in NO₃⁻ in the model. In a simulation with increased CO emissions in Fresno grid cells, the largest impacts on CO mixing ratios aloft at night were to the south of Fresno during this period (Figure S18). Too much overnight transport of NO₃⁻ from the residual layer over Fresno to the south is consistent with underestimates of NO₃⁻ at the surface in the morning.

HNO₃ production integrated over model layers 1–20 is shown in Figure 12a for SJV grid cells during 17–22 January. R1 is the dominant production pathway in urban areas with large NOx emissions such as Fresno and Bakersfield in the model. The R2 pathway is dominant in semiurban and rural areas along Highway 99, particularly around Visalia and in northern SJV. HNO₃ production in SJV peaks in model layer 6 (160–240 m; Figure 12c). R1 is productive in the middle of the daytime boundary layer due to the combination of relatively high OH and NO₂, and R2 tends to be most productive in the nocturnal residual layer due to the combination of high N₂O₅ and aerosol surface area (Riemer et al., 2003). Overall, the model estimates that R1 contributes 46% and R2 contributes 54% to total HNO₃ production for the 17–22 January period when the model predicted elevated NO₃⁻. This apportionment is similar to model estimates from previous episodes (Ying & Kleeman, 2009). In early February, when the model underpredicted NO₃⁻ concentrations, the modeled boundary layer was deeper during the day and production occurred over a wider range of altitudes (Figure 12d). The R2 pathway was relatively weak in the model in the area between Fresno and Bakersfield in early February (Figure 12b) compared with 17–22 January (Figure 12a).

Figure 12.

HNO₃ production integrated over layers 1–20 for SJV model grid cells during (a) 17–22 January and (b) 30 January to 4 February and integrated over SJV grid cells by model layer during (c) 17–22 January and (d) 30 January to 4 February.

Pusede et al. (2016) predicted that HNO₃ production from R1 would increase relative to R2 with decreasing NOx emissions. To explore the sensitivity of HNO₃ production to NOx levels in the model, a sensitivity simulation was conducted with NOx emissions reduced by 40%. In this simulation, R1 contributed 49% to integrated HNO₃ production during 17–22 January (i.e., production from R1 was enhanced relative to R2 compared with the base simulation). Decreases in NOx emissions lead to increases in OH mixing ratios in urban areas and along major highways in the model and thereby increase the percent contribution of R1 to total HNO₃ production relative to that of the base simulation. This behavior is qualitatively consistent with predictions of Pusede et al. (2016), although that study focused on the entire winter period rather than the multiday episode considered here. A wide range of N₂O₅heterogeneous reaction probabilities (i.e., the fraction of gas-particle collisions that result in net removal of N₂O₅ from the gas phase, γ) have been used in previous studies of NO₃⁻formation in SJV (e.g., Prabhakar et al., 2017; Ying & Kleeman, 2009), and studies have reported that the influence of organic coatings (not treated here) on γ can be important (e.g., Riemer et al., 2009) To explore the sensitivity of HNO₃ production to γ and the ClNO₂ yield (Y in R2), three additional simulations were conducted with γ scaled by 0.5 and 1.5 and with Y = 0. Total HNO₃ production decreased by 11% relative to the base case when γ was reduced by 50% for the scenario in Figure 12a. A 24% reduction in HNO₃ production from R2 was partially compensated for by a 5% increase in production from R1. Total HNO₃production increased by 6% relative to the base case in the simulation with a 50% increase in γ. Setting the yield of ClNO₂ to zero had negligible impact on HNO₃ production consistent with the generally low concentrations of Cl⁻ in SJV, although Cl⁻ was underpredicted (Figure S14). A summary of HNO₃ production in SJV for the base and sensitivity simulations is provided in Table S4, and γ values predicted over the P-3B spiral sites during 17–22 January are shown in Figure S19.

4. Conclusions

This study demonstrates that regional photochemical grid models are capable of simulating NH₄NO₃ formation and build-up during major recent PM_2.5 episodes in SJV. For example, routine measurements of NO₃⁻ were generally predicted well at sites in SJV, including days where 24-hr average NO₃⁻ reached 20 μg/m³. Gas-particle partitioning predictions were in good agreement with measurements in Fresno and indicate that the model correctly predicts that NH₄NO₃ formation is limited by HNO₃ availability. Modeled chemical production of HNO₃ via daytime and nighttime pathways was generally consistent with reports from previous studies and conceptual models of NO₃⁻ formation in SJV. During a period of elevated NH₄NO₃, the model predicted that the OH + NO₂ pathway contributed 46% to total HNO₃ production in SJV and the N₂O₅ heterogeneous hydrolysis pathway contributed 54%. This result highlights the importance of nighttime chemistry in NO₃⁻production in SJV.

Despite generally favorable model performance, the DISCOVER-AQ data set provided insights on areas where additional work could improve NH₄NO₃ modeling for SJV. First, additional study on meteorological modeling of the major stagnation events that drive PM_2.5episodes in the Valley would be valuable, particularly for southern SJV where the terrain is more complex than in central and northern SJV. Challenges in simulating meteorology in southern SJV could help explain the better NO₃⁻ model performance for Fresno and Modesto than Bakersfield. Also, work toward improving the simulation of diurnal patterns of vertical mixing would be valuable, because the coupling and decoupling of processes in the surface layer from layers aloft influences HNO₃ production and the diurnal profiles of NO₃⁻at the surface. Additional evaluation of the degree to which urban-nonurban transport of NO₃⁻ occurs in the ambient would also be helpful because predictions suggest that this transport can be important. Improvements in meteorological modeling are likely necessary to improve performance against the hourly average NO₃⁻ measurements in Fresno. Second, additional work on NH₃ emission and air quality modeling is warranted based on underpredictions of NH₃ in emission source regions where very high mixing ratios were measured. Although the NH₃ underpredictions do not appear to have a large impact on NO₃⁻ predications (because NO₃⁻ is generally HNO₃-limited during wintertime episodes), NH₃ levels are too low in the model in source regions and warrant further study. Improvements in the spatial allocation of NH₃ emissions are also warranted, especially near Hanford. Third, there is evidence that gas-particle partitioning predictions under low-RH conditions could benefit from additional study. Although the overall impact of gas-particle partitioning issues may be minor due to the generally high RH during SJV PM_2.5 episodes, the potential for premature removal of TNO3 via rapid deposition of HNO₃ when the gas-phase fraction is overestimated in afternoon makes this an area of interest.

Another topic for future investigation is on HNO₃ production in the nocturnal residual layer over urban and surrounding areas. Although this pathway is central to the conceptual model of NO₃⁻ formation in SJV, measurements that can directly constrain nighttime HNO₃production aloft over SJV are extremely limited. Researchers have made progress by using indirect methods to infer characteristics of the nocturnal residual layer based on measurements over urban areas on the previous day and following morning, but direct measurements of the key species at night over urban and surrounding areas would be valuable.