Quantifying the contribution of inflow on surface ozone over California during summer 2008



[1] Transported pollution has been recognized as making a potentially strong impact on air quality in the western U.S., but large uncertainties remain in quantifying its contribution. Assessing the role of pollution transport in relation to local emissions and meteorology is especially important in light of possibly lower ozone standards and projected increases in transpacific pollution transport. We apply the Weather Research and Forecasting with Chemistry model to analyze the role of upwind pollution (“inflow”) to surface ozone over California during the Arctic Research of the Composition of the Troposphere from Aircraft and Satellites campaign in June–July 2008 over California. Comparisons of the model to surface and aircraft observations, ozonesondes, and satellite retrievals show an overall good agreement; a low bias (~5 ppb) in free tropospheric ozone is attributed to low ozone at the boundaries and likely places our estimated inflow contribution on the lower side. Most other studies applied sensitivity analyses, while we use a synthetic ozone tracer, which provides a quantitative estimate of the budget. We estimate that on average 10 ± 9 ppb of surface afternoon ozone over California is attributed to ozone and ozone precursors entering the region from outside. This contribution features a significant spatial and temporal variability. While in most high ozone events, transported pollution plays a small role compared to local influences, for some instances, the impact can be substantial. Omitting data impacted by wildfires, we estimate the 90th percentile of the relative contribution of O3INFLOW to 8 h ozone >75 ppb as 10%. Our results also indicate that inflow might have a stronger impact on surface ozone in less polluted compared to polluted areas.

1 Introduction

[2] The Environmental Protection Agency (EPA) national emission inventory estimates that from 1990 to 2010 ozone precursor emissions have been cut roughly in half, which has led to a decrease in extreme ozone events [Butler et al., 2011; Frost et al., 2006; Pozzoli et al., 2011]. However, several studies have shown that at the same time the baseline or background concentrations have been increasing [Hogrefe et al., 2011; Jaffe and Ray, 2007; Lefohn et al., 2010; Parrish et al., 2012; Parrish et al., 2009] and that reductions of surface ozone have been greater in the eastern U.S. compared to the western U.S. [Cooper et al., 2012]. This suggests that increasing background ozone entering the U.S. may in part be counteracting ozone reductions from domestic emission controls.

[3] A major component of this inflow is assumed to be due to Asian pollution reaching the lower free troposphere of the western U.S. [Jacob et al., 1999; Parrish et al., 2004; Zhang et al., 2008] and mixing down to the surface [Huang et al., 2010; Lin et al., 2012; Parrish et al., 2010, 2012; Pfister et al., 2011b]. Specifically during springtime and early summer, inflow might also be impacted by stratospheric intrusions mixing with Asian pollution [Langford et al., 2012; Lin et al., 2012b]. Evidence of long-range transport and of pollution and its impact on surface air quality (AQ) are not limited to ozone and ozone precursors but have also been observed for other pollutants such as dust, sulfate, or mercury [Husar et al., 2001; Jaffe et al., 2003; McKendry et al., 2008; Weiss-Penzias et al., 2007]. Advancing the understanding of transported pollution to local surface concentrations is crucial for assessing emission control strategies, estimating future air quality, and for risk and exposure modeling.

[4] Most modeling studies to date used global models with rather coarse resolution to estimate contributions of background and transported ozone to surface levels over the U.S., and estimates are generally derived by perturbing emissions from various source regions/sectors by a certain percentage [Lin et al., 2012; Zhang et al., 2011]. These estimates, however, are impacted by uncertainties due to model resolution. Advantages are gained by using higher-resolution regional-scale models, which are expected to better resolve the transport and mixing processes, specifically for complex terrain and urban regimes [Lin et al., 2010; Wild and Prather, 2006].

[5] California (CA) located at the U.S. West Coast is specifically susceptible to transpacific pollution transport, and despite aggressive pollution controls, large parts of the state continue to be in non-attainment for ozone and particulate matter standards (http://www.epa.gov/oaqps001/greenbk/index.html). This emphasizes the need for a detailed characterization of the impacts of baseline ozone on surface AQ and the quantification of the contribution of pollution inflow, especially during high pollution episodes. In this study we focus on the role of inflow of ozone (O3) and O3 precursors on surface ozone over California and concentrate on the time period from middle June to early July 2008, which overlaps with the CA phase of the NASA Arctic Research of the Composition of the Troposphere from Aircraft and Satellites (ARCTAS) mission. This time period also coincides with an extremely intense wildfire season with major fires starting around 20 June and lasting well into July. The fires significantly impacted surface ozone during the study period by changing the chemical regime and have a potential impact on the way incoming ozone and ozone precursors chemically interact. For this reason we attempt to separate out the fire impact where possible.

[6] Our estimates of the contribution of ozone from outside sources to the ozone budget over California are based on simulations with the regional Weather Research and Forecasting Model with Chemistry (WRF-Chem) including tracers for carbon monoxide (CO) and O3, which keep track of the inflow at the lateral boundaries of the regional domain. In contrast to most other modeling studies using sensitivity simulations with perturbed emissions or perturbed boundary conditions to derive estimates of the influence of transported background and pollution on surface concentrations, the tagging scheme applied here provides a measure of the actual budget term. A comprehensive evaluation of the modeled fields using surface in situ measurements, aircraft and ozonesonde data, and satellite tropospheric NO2 ensures detailed characterization of model performance and limitations.

[7] The paper is organized as follows. We start in section 2 with a description of the model setup and simulations including a discussion of the tagging schemes, followed by describing the observations used to evaluate the simulations and the respective evaluation results. Section 3 discusses the analysis and looks at the overall impact of transported pollution on surface ozone over CA as well as investigates the role of inflow during high ozone events. We close by summarizing our findings.

2 Model Simulations and Evaluation

2.1 Model Simulations

[8] We use the regional chemistry transport model WRF-Chem, version 3.2 [Grell et al., 2005] in this work. The model was set-up for a single domain centered over California (CA) covering most of the western U.S. at 12 km × 12 km (143W–95W and 25N–51N) with 27 vertical levels between the surface and 50 hPa. The analysis presented in this paper focuses on the center part of this domain over the larger CA region. The simulation spans the time period from 10 June–10 July 2008 with output every 2 h. Meteorological initial and boundary conditions are taken from the National Centers for Environmental Prediction (NCEP) Eta North American Mesoscale Analysis with analysis nudging for wind, temperature, and humidity applied. Spatially and temporally (6-hourly) varying chemical boundary conditions are provided by global model simulations from the Model for Ozone and Related Chemical Tracers (MOZART-4) [Emmons et al., 2010a]. For these simulations, the global model was run at a spatial resolution of ~0.7° by 0.7° (T170) and with 64 vertical levels up to 2 hPa. The meteorological fields for driving MOZART-4 were taken from National Centers for Environmental Prediction (NCEP)-Global Forecasting System (GFS) analysis. More details about the global model simulations and its evaluation are found in Pfister et al. [2011b]. WRF-Chem and MOZART-4 both employ the MOZART-4 gas-phase chemical scheme fully described in Emmons et al. [2010a]. The aerosols in the global MOZART-4 model include 12 bulk aerosol compounds. In WRF-Chem, the MOZART-4 gas phase chemistry is linked to the bulk aerosol scheme Goddard Chemistry Aerosol Radiation and Transport (GOCART)model [Chin et al., 2002].

[9] Anthropogenic and fire emissions, which are described in Pfister et al. [2011a], as well as domain topography for our study region are shown in Figure 1. The anthropogenic emissions are based upon the U.S. EPA's 2005 National Emissions Inventory (NEI) (version 3) (S. McKeen, personal communication, 2008) and over CA are replaced by the California Air Resources Board (CARB) 2008 inventory (A. Kaduwela and J. Avise, CARB, personal communication, 2008). Fire emissions have been estimated using the Fire Inventory from National Center for Atmospheric Research (NCAR) (FINN V1) [Wiedinmyer et al., 2011] with a diurnal profile following recommendations by the Western Regional Air Partnership (WRAP) (Report to Project No. 178–6, July 2005). The WRF-Chem online plume rise module [Freitas et al., 2007] is applied to distribute the fire emissions vertically. Biogenic emissions are calculated online following the Model of Emissions of Gases and Aerosols from Nature [Guenther et al., 2006].

Figure 1.

Anthropogenic and fire emissions (104 mol/km2) over the larger California region for 10 June to 10 July and domain topography (km).

2.2 Tagging Schemes for CO and Ozone

[10] We incorporated a set of passive tracers in WRF-Chem including CO tracers for different source types as described in [Pfister et al., 2011a]. In this study we use CO tracers for CO emitted from fires within the domain (COFIRE) and also CO from inflow (COINFLOW). The latter is based on tagging CO that enters the regional domain through the lateral boundaries. This method only applies to CO and does not consider CO photochemical production from hydrocarbons entering the domain, which, however, is estimated to be a small contribution [Pfister et al., 2011a].

[11] The ozone tagging or “XNOx” method [Emmons et al., 2012], until now, exclusively has been used in global models for identifying individual source contributions and for quantifying the ozone budget [Lamarque et al., 2005; Pfister et al., 2006, 2008; Hess and Lamarque, 2007; Emmons et al., 2010b; Brown-Steiner and Hess, 2011]. In this study the method is applied for the first time in a regional model. We provide a brief overview of the tagging scheme, but for greater detail the reader is referred to Emmons et al. [2012].

[12] The XNOx scheme tags emissions of nitrogen compounds (e.g., NO or NO2) from a given source to keep track of the amount of ozone produced from this source. A duplicate set of tracers for all odd nitrogen species (e.g., peroxyacetyl nitrate (PAN), HNO3, N2O5) is added to the chemical mechanism. The tagged nitrogen compounds are followed in a separate but identical chemical mechanism from their source through to the production of ozone. The loss of tagged ozone due to chemical loss (discussed below) and deposition occurs at the same rate as for the standard ozone. As such, the tagged species are affected by species of the standard mechanism, but they do not affect the standard chemistry. Emmons et al. [2012] showed that the tagged ozone equals the standard ozone to within 3% on average when all tropospheric sources are tagged and stratospheric input is turned off.

[13] Here we apply this scheme to keep track of the contribution of ozone within the WRF-Chem domain that is due to air masses entering the regional domain at the lateral boundaries (“inflow”). For this purpose, we define the model lateral boundaries as the “source” and tag ozone as well as all odd nitrogen species at the lateral boundaries of the regional model domain. The tagged ozone within the regional domain is termed as O3INFLOW and includes both transported ozone and ozone produced from tagged precursor species.

[14] Two basic features of the tagging scheme have to be considered when comparing to other studies and will likely lead to our estimates being on the low end when compared to observational-based estimates. One is that, as a result of the tagging of nitrogen species, ozone will lose its tagging when reacting with a non-tagged nitrogen molecule. In the standard chemistry, NO, NO2, and O3 react in a null cycle: NO + O3 - > NO2 + O2; NO2 + hν - > NO + O; O + O2 - > O3. In the tagged mechanism, these reactions become XNO + O3 - > XNO2 + O3; NO + O3A - > NO; XNO2 + hν - > XNO + OA; OA + O2 - > O3A, with XNO and XNO2 indicating tagged NO and NO2 and O3A and OA indicating tagged O3 and O, respectively. Thus, when a tagged O3 reacts with non-tagged NO, the tag is lost but at exactly the same rate as the total ozone (NO concentrations are the same for both reactions). For example, incoming ozone that reacts with NO from a source within the regional domain, which during daytime can rapidly return to ozone, will no longer be considered inflow. Another important consideration is that the tagging scheme keeps track of incoming nitrogen species resulting in ozone production but not of incoming hydrocarbons. For example, ozone produced from reactions involving nitrogen species entering at the domain boundaries and peroxy radicals resulting from oxidation of hydrocarbons from a source within the domain would be labeled as O3INFLOW. Ozone produced from reactions involving nitrogen species originating from within the domain and peroxy radicals resulting from the oxidation of hydrocarbons entering at the lateral boundaries would not be tagged. The complex nature of ozone chemistry makes it a challenge to define ozone source contributions, and this tagging method originating with NO can be considered a measure of the enhancement in photochemical production due to inflow ozone.

[15] The O3INFLOW tracer includes ozone and ozone precursors from all natural (including lightning and stratospheric ozone) and anthropogenic nitrogen sources outside the regional modeling domain. Within the regional modeling domain, O3INFLOW undergoes transport and chemical processes but is not produced from sources other than from reactions including the tagged species. Since in this version of WRF-Chem the stratospheric ozone is controlled by the lateral boundaries, ozone from stratospheric intrusions within the regional domain would be labeled as O3INFLOW as well. The influence of stratospheric intrusions in the western U.S. on surface ozone can be significant during springtime but is expected to be less frequent and of smaller magnitude during summertime [Fiore et al., 2003; Langford et al., 2012; Lin et al., 2012b].

[16] The tagging method can lead to significantly different answers compared to “perturbation” methods where contributions from individual sources are estimated based on differences between two sets of simulations—one with base emissions and one with a perturbation in the considered emission source. A perturbation approach has been commonly used for assessing the importance of emissions from a certain region on downwind areas [e.g., Fiore et al., 2009; HTAP, 2010; Zhang et al., 2011; Lin et al., 2012], and in a recent study, it has also been applied to look at inflow in a regional model [Huang et al., 2013].

[17] However, as was shown in previous studies [e.g., Pfister et al., 2006; Grewe et al., 2010; Emmons et al., 2012; Stock et al., 2013], changing the emissions alters the chemical environment and in turn affects the rate of ozone production and loss and as such contributes to the differences seen. Using the perturbation method for assessing source contributions can have potentially large errors (up to a factor of 2) dependent on the degree of linearity of the chemical system [Grewe et al., 2010]. The tagging method, in contrast, quantifies contributions without changes to the chemical regime or the ozone production and loss terms. In view of this, if the objective is to explore the effect of a change in emissions or for estimating policy relevant background ozone defined as ozone in the absence of anthropogenic emission sources [McDonald-Buller et al., 2011], then the perturbation method is the appropriate method. If, however, the objective is to estimate actual source contributions or derive a budget, then the tagging scheme is the preferred method. Distinguishing the basic fundamentals of the tagging scheme from the perturbation method is crucial when interpreting results and comparing to other studies, and it is important to appreciate the range of ambiguity resulting from different approaches.

2.3 Evaluation With Surface Ozone

[18] Figure 2 shows a comparison of WRF-Chem simulated surface ozone to observations at monitoring sites in the EPA network. Surface data for O3 and NO were provided by CARB. We show time series as well as histograms for the modeled and observed mixing ratios and model-measurement differences starting in 19 June. In addition to total ozone, we also include modeled concentrations of O3INFLOW. Sites are separated into regions, Northern (N-CA) and Southern CA (S-CA), which provides a first-order separation into sites strongly influenced by fire and anthropogenic emissions and sites mostly influenced by anthropogenic emissions only. Sites are further separated into urban, suburban, and rural categories. The latter filtering is based on the EPA site classification, which occurs when the sites are established, so these classifications may be flawed due to changes in land use and urban expansion that have occurred since the site was initially established. The comparison is influenced by different numbers of measurement sites in the different categories with more O3 and fewer NO sites and more urban and fewer rural sites as indicated in the maps in Figure 2.

Figure 2.

Model evaluation for urban/suburban and rural sites in S-CA (top two rows) and in N-CA (bottom two rows). Each row shows (a) observed (black) and modeled (WRF-Chem; red–total O3; yellow–O3INFLOW) time series of surface ozone, (b) frequency distribution for daytime (solid) and nighttime (dashed) surface ozone, and (c) frequency distribution of model bias. Site locations are shown in the inset of graphs in Figure 2a.

[19] Overall, the model represents the temporal variations in ozone well and captures the differences between urban and rural areas. Rural areas in S-CA show a reduced diurnal amplitude compared to urban areas due to higher nighttime values resulting from less titration and similar daytime concentrations. For N-CA, which largely has been influenced by the fires, we find similar day-to-day variability for all site categories, and both observations and model show enhanced daytime and nighttime ozone at rural versus urban sites.

[20] The observed daytime surface ozone mixing ratios cover a wide range (Figure 2b), and the model simulates well the observed distributions, which are broader for urban compared to rural areas. During nighttime, the model distributions are shifted to too low values compared to the observations and we also find a lower correlation between the model and observations than for daytime reflecting the difficulties in models representing nighttime chemistry and atmospheric dynamics when there is little to no turbulence mixing air parcels within the boundary layer. As will be discussed later, the model also shows a high bias in NOx mixing ratios likely related to a high bias in emissions causing nighttime ozone titration to be overestimated. On average, the modeled ozone levels agree well with observations with a bias of 6 ppb or smaller. The largest differences are seen for the middle of the simulation period for S-CA when the model underestimates the high daytime peaks.

[21] Part of the difficulties of the model in representing sites in S-CA can be explained by challenges in simulating the complex flow patterns due to the topography and the influence of land-ocean circulations, but uncertainties in emissions and NOx concentrations are another consideration. Figure 3 shows a comparison of observed surface NO mixing ratios to model estimates. Despite the site categories being influenced by land use changes, we see that to a first order, the rural sites show lower concentrations compared to urban sites in S-CA, while urban and rural sites in N-CA are less distinct due to the influence of fires. The model reflects these different site behaviors and on average agrees well with observed concentrations. Given that NO is also directly emitted and can have strong concentration gradients near source regions, the comparison to modeled NO, however, is more impacted by model grid resolution and spatial uncertainties in the emission inventory compared to when comparing a secondary product like ozone. WRF-Chem does underestimate peak concentrations, which to a large part is related to the model grid resolution. Urban monitors, specifically, are often located near major roadways, and the high emissions and mixing ratios are smeared over the 12 km × 12 km model grid. Despite the model not capturing the peaks, we find that WRF-Chem on average has a high daytime model bias at urban and suburban areas (Figure 3). A high bias in NOx might lead to less efficient ozone production efficiency in the model in high NOx areas and explain some of the underestimation in high ozone concentrations. The high NOx bias in WRF-Chem during daytime is also confirmed by evaluation with aircraft (section 2.4) and satellite (section 2.5) data.

Figure 3.

Same as Figure 2 but for surface NO mixing ratios.

[22] Uncertainties in comparing the modeled to observed concentrations also arise from uncertainties in the inflow. As shown in Figure 2, O3INFLOW can be as high as 20 ppb with a slightly higher impact in N-CA. Urban sites, in general, are estimated to have smaller concentrations for O3INFLOW compared to rural sites. The time series shown reflect the large diurnal and episodic nature of inflow with strongest influence in S-CA during the first and middle part of the simulated time period and in N-CA during the first and latter portion of the time period. As expected for a short-lived species like NO, the contribution of inflow on surface NO mixing ratios is minor (Figure 3).

[23] In addition to grouping ozone sites by only category, we also performed an analysis when sites are grouped by elevation and/or category (not shown here). In this case, we find that sites with similar elevation show higher O3INFLOW mixing ratios at rural versus urban sites, and sites of the same category show higher O3INFLOW mixing ratios at high-elevation versus low-elevation sites. This confirms the suggestion that both chemical regime and elevation play a role in determining the impact of inflow on surface ozone.

2.4 Evaluation With Aircraft and Ozonesonde Data

[24] The evaluation of WRF-Chem with DC-8 aircraft observations for O3, NOx, PAN, and HNO3 is shown in Figure 4. In addition, we list in Table 1 the statistics for these and also other chemical species measured on board the DC-8. We compare the model to the merged 1 min measurements (R13) of the four science flights conducted over CA during the Arctic Research of the Composition of the Troposphere from Aircraft and Satellites (ARCTAS-CARB) campaign (18, 20, 22, and 24 June) as well as the segment over CA of the DC-8 transit flight on 26 June (Figures 4a–4d). In 22 June, one segment was conducted off the coast of CA to measure pollution inflow across the Pacific entering the U.S. West Coast (“Boundary Conditions (BC) Flight”), and we treat these data separately. For more information about the campaign and instruments, we refer to Jacob et al. [2010].

Figure 4.

Modeled mixing ratios (red–total species; blue–tagged species) compared to DC-8 aircraft measurements (black) for O3. Using measured acetonitrile mixing ratios the flight data are grouped into data over CA (a) impacted by fires and (b) not impacted by fires. (c) The flight leg off the coast of CA is shown separately. Numbers on the right indicate the sample size per altitude bin.

Table 1. Mixing Ratios (ppbv) for Various Chemical Species Measured on the DC-8 Science Flights and Interpolated Model Dataa
 FireNo FireBC
 0–2 km2–8 km0–2 km2–8 km0–2 km2–8 km
  1. a

    Data are grouped into different altitude ranges and into observations impacted by fires (CH3CN > 0.2 ppbv), not impacted by fires (CH3CN < 0.15 ppbv), and the boundary conditions leg on 22 June.

  2. b

    in (parts per trillion).

O3 Obs.71 ± 1862 ± 1264 ± 1953 ± 1034 ± 1365 ± 19
O3 Model62 ± 13857 ± 659 ± 1750 ± 927 ± 1655 ± 11
NOx Obs.2.6 ± 70.1 ± 0.12.3 ± 50.05 ± 0.070.08 ± 0.10.03 ± 0.02
NOx Model2.4 ± 70.3 ± 0.15.4 ± 100.14 ± 0.170.05 ± 0.050.05 ± 0.03
PAN Obs.0.7 ± 0.70.5 ± 0.30.5 ± 0.40.15 ± 0.150.03 ± 0.030.20 ± 0.18
PAN Model0.4 ± 0.40.2 ± 0.050.4 ± 0.40.15 ± 0.070.02 ± 0.030.18 ± 0.09
HNO3 Obs.1.9 ± 3.10.12 ± 0.121.8 ± 2.00.26 ± 0.200.17 ± 0.110.15 ± 0.18
HNO3 Model2.5 ± 4.80.99 ± 0.094.0 ± 4.50.20 ± 0.270.08 ± 0.060.09 ± 0.07
HCHO Obs.6.4 ± 7.61.6 ± 1.32.6 ± 1.80.37 ± 0.50.36 ± 0.20.19 ± 0.1
HCHO Model2.6 ± 1.40.8 ± 0.72.0 ± 1.30.38 ± 0.360.22 ± 0.20.20 ± 0.1
HOx Obs.b36 ± 1427 ± 927 ± 1217 ± 717 ± 616 ± 5
HOx Modelb27 ± 1223 ± 724 ± 1623 ± 819 ± 1228 ± 6
C2H6 Obs.3.2 ± 3.11.5 ± 0.81.3 ± 1.40.6 ± 0.20.6 ± 0.20.7 ± 0.3
C2H6 Model2.2 ± 1.21.2 ± 0.62.2 ± 1.30.7 ± 0.20.8 ± 0.20.8 ± 0.2
C3H8 Obs.1.2 ± 1.20.4 ± 0.30.9 ± 1.70.1 ± 0.20.1 ± 0.10.1 ± 0.0
C3H8 Model0.3 ± 0.30.1 ± 0.10.5 ± 0.70.03 ± 0.030.03 ± 0.00.03 ± 0.0
CH3CHO Obs.1.6 ± 1.61.1 ± 0.70.69 ± 0.60.14 ± 0.20.04 ± 0.030.06 ± 0.05
CH3CHO Model0.4 ± 0.50.1 ± 0.040.38 ± 0.40.05 ± 0.040.01 ± 0.000.01 ± 0.00
Isopr Obs.b194 ± 3538 ± 23110 ± 58410 ± 441.3 ± 0.71.4 ± 0.6
Isopr Modelb217 ± 33413 ± 40160 ± 3065 ± 260.0 ± 0.00.0 ± 0.0

[25] Using measured acetonitrile (CH3CN) concentrations as a surrogate for fire influence, we separate the data over CA into two groups: observations over mainland impacted by fires (CH3CN > 0.2 ppbv) and those not significantly impacted by fires (CH3CN < 0.15 ppbv) [Pfister et al., 2011a]. As the selection is only based on the observations, erroneous simulations of the fire impact might also influence the comparison through errors in the simulated wind field, estimated emission rates, the timing of those emissions, and the injection height of the emissions. In addition to total model concentrations, we show in Figure 4 the respective inflow tracers (O3INFLOW, NOxINFLOW, PANINFLOW, HNO3INFLOW).

[26] Ozone for the BC flight, which mostly is attributed to inflow, is overall too low in the free troposphere. The model underpredicts the high ozone plume at 8–10 km during the BC flight which is due to the difficulties in global models in resolving the high concentrations observed in plumes transported over large distances and an underestimate in the mixing of stratospheric ozone with Asian pollution [Pfister et al., 2011b]. The low boundary conditions also lead to a low bias in the free troposphere (FT) for both fire and non-fire data, where above ~4 km altitude, most of the ozone in the model is explained by inflow. Compared to the no fire data, both observed and modeled fire data show enhanced ozone due to the additional ozone production within the fire plumes (Figure 4a).

[27] Generally, good agreement is found for other species, and the corresponding inflow tracers show a similar vertical contribution as ozone, i.e., values are increasing with altitude, and most of the BC mixing ratios are attributed to inflow. Near the surface, fire and BC flight data show good agreement for NOx (Figure 4b), but the model is too high for the no fire data suggesting an overestimate in anthropogenic emissions as was also concluded from the comparison to surface NOx measurements (section 2.3). This high surface bias in part might also explain the high bias in the FT. Aside from the fire plumes, PAN shows a fairly good agreement with the aircraft data (Figure 4c) while larger differences are found for HNO3 with the model mostly underestimating observations (Figure 4d). Possible reasons for this might be uncertainties in the modeled removal of HNO3 through precipitation scavenging, which represents a major part of the HNO3 budget and in the chemical processes (e.g., lack of aqueous phase chemistry). For the fire data, PAN and HNO3 are overall too low, which may be attributed to the model not representing the timing and location of the fire plumes correctly but also reflects the challenges in simulating emissions, dynamics, and chemistry of a highly variable source such as fires on a regional scale. The larger model uncertainties in fire plumes compared to no fire data are also seen for other species (Table 1). WRF-Chem is in fairly good agreement for HCHO, HOx, and isoprene but clearly low in C3H8 and CH3CHO for all data sets, likely resulting from a combination of a low bias in inflow and in emissions.

[28] Figure 5 evaluates the model performance of ozone, relative humidity, and temperature using data from ozonesondes launched at Trinidad Head. Trinidad Head is located at the coast in Northern CA and well suited to sample air entering the western U.S. from the Pacific [Oltmans et al., 2008]. Data are accessible through the NOAA Earth System Research Laboratory (ESRL) Global Monitoring Division (ftp://ftp.cmdl.noaa.gov/). Fifteen profiles taken between 20 June and 8 July were used in this evaluation. The model has a high bias of about 20% in relative humidity throughout the extent of the troposphere and a bias in temperature of up to 4 K near the surface. Model values near the surface for the Trinidad Head site have to be interpreted with caution, though. Comparison to observed surface winds (not shown here) indicate difficulties in the model in resolving coastal flows due to the limited spatial resolution. WRF-Chem simulates the shape of the vertical profile and also compares well in the vertical distribution of ozone with average differences below about 10 ppbv throughout the troposphere. Similar to what was shown for the aircraft data, the contribution from O3INFLOW is in the range of 10–20 ppbv near the surface and increases with altitude; for 500 hPa and above, most of the ozone is from inflow. The low FT model bias is in agreement with comparisons to the DC-8 aircraft data.

Figure 5.

Top row: WRF-Chem compared to ozonesonde data for Trinidad Head. Observations (black), model (red), O3INFLOW (yellow). Average profiles for O3 (left), relative humidity (middle), and temperature (right). Absolute difference is shown as dotted line. Bottom row: Example model profiles for O3 and O3INFLOW and wind direction for 4 July (left) and 28 June (right).

[29] Figure 5 also shows two individual profiles to demonstrate the changes in the ozone vertical distribution with changes in wind direction. When the wind comes from a more westerly direction, as is the case for 4 July, ozone levels increase fairly steadily with altitude and the major part of total ozone is attributed to inflow. In this case we also see an enhanced ozone plume with up to 120 ppbv entering from the boundaries. In contrast, a more southerly wind brings ozone from within CA to Trinidad Head with little contributions from inflow in the lower troposphere and in this case enhanced ozone levels just above the boundary layer.

2.5 Evaluation With Satellite NO2 Retrievals

[30] In the final part of the model evaluation, we compare WRF-Chem to tropospheric NO2 retrievals from the Ozone Monitoring Instrument (OMI). OMI aboard Aura observes the atmosphere in nadir view with a local equator crossing time of around 13:30 LT. We use the Royal Netherlands Meteorological Institute Derivation of OMI Tropospheric NO2 (DOMINO) product, which is estimated to have absolute and relative retrieval errors of about 1.0 × 1015 molecules cm−2 and 25%, respectively. For more information about the product, we refer to Boersma et al. [2011]. We use daily level-2 (Version 2.0) data with solar zenith angle < 80 degrees and cloud radiance fraction less than 30% (cloud fraction less than 0.1). We further exclude pixels corresponding to viewing angles exceeding 45 degrees (pixels at the edge of swath and/or with a field of view wider than about 50 km). For comparison with the model, we interpolate modeled NO2 profiles to the time, location, and vertical spacing of OMI data and apply the retrieval averaging kernels and air mass factor following the guidelines in the DOMINO User's Guide.

[31] Figure 6 shows OMI and modeled tropospheric NO2 columns averaged over a 0.25° × 0.25° grid for 12 June to 8 July. The model reflects the spatial distribution of NO2 with hot spots in the LA basin and Bay Area and enhanced values throughout the Central Valley. It overestimates NO2 in the major urban areas, such as the Bay Area and the LA basin but underestimates NO2 columns in the Central Valley. This spatial bias is in agreement with evaluation results from comparison to surface observations (Figure 3; detailed spatial distribution not shown here).

Figure 6.

Average tropospheric NO2 column for 12 June to 8 July from OMI/DOMINO and WRF-Chem and their difference (Model Observation).

3 Results: Contribution of Inflow to Surface Ozone

[32] The fairly good representation of the model to a range of different observations gives confidence in the model ability to simulate the spatial and temporal distribution of ozone over CA. In this section, we analyze the model results, examine how the import of ozone and ozone precursors influence surface ozone over CA, and look into its role during high ozone episodes.

3.1 Contribution of Inflow to Surface Ozone Over CA

[33] Figure 7 shows the distribution of mean total and tagged surface ozone, averaged for 19 June to 9 July 2008, local afternoon. Highest surface ozone is found in the Central Valley and downwind of LA; the largest variability is estimated for the southern basin, around the Bay Area and in Northern Calfornia; the latter is attributed to the wildfires, but this region also is estimated as having the largest contribution from inflow of up to about 15 ppbv and a variability of the same magnitude. The state of Nevada (NV), which is at high elevation and has fewer local pollution sources, shows the strongest influence of 25 ppbv and higher. In general, we do find the largest O3INFLOW at higher elevations in part because these areas are more exposed to free tropospheric air and also because these are less polluted environments where ozone production might be more efficient compared to when ozone and ozone precursors enter a strongly polluted environment (to be discussed in section 3.2).

Figure 7.

Modeled estimated mean and standard deviation of total O3 and O3INFLOW and COINFLOW (ppbv) as well as percentage contribution of O3INFLOW (%) (local afternoon).

[34] In their analysis of the ARCTAS-CARB time period, Huang et al. [2013] similarly find the largest influence of transported pollution in the northern part of CA, but in contrast to our findings, they did not find enhanced contributions over NV. This is explained by different models, setup, and metrics, but the discrepancy is also caused by the different methods used to estimate contributions from inflow. While our estimate is based on the actual chemical regime and considered a budget term, the values derived by Huang et al. [2013] were extrapolated from sensitivity simulations with a 50% reduction in boundary conditions for all species, hence include sensitivity to hydrocarbons, and are also subject to nonlinearities in ozone chemistry. While also not directly comparable to our estimate, the study by Zhang et al. [2011] produced comparable spatial patterns for the policy relevant background (estimate of surface ozone over the U.S. in the absence of North American anthropogenic emissions) and estimates the contribution of intercontinental pollution and anthropogenic methane to ozone as ranging from 11–13 ppbv across the U.S. Focusing on the Asian component of the inflow, Lin et al. [2012] estimate the contribution of Asian pollution during May–June 2010 by turning off Asian anthropogenic emissions in a global chemical transport model. Similarly, they find the greatest impact in the high-elevation regions over the western U.S. and estimate a comparable spatial influence as in our study.

[35] The strong dependence of O3INFLOW on the chemical regime into which it is mixed is shown by comparison to the spatial distribution of COINFLOW, also included in Figure 7. CO, in contrast to ozone, is much more dynamically driven, and COINFLOW gradually decreases when transported through the northern and western boundaries into the domain. In addition to chemical processes, the vertical distributions of ozone and ozone precursors and CO differ and will be discussed in section 3.2.

[36] The mean surface ozone concentration in the local afternoon over CA is on the order of 58 ± 21 ppbv, and it is estimated that 10 ± 9 ppbv (20 ± 21%) of this is due to inflow (Table 2). The standard deviation is of similar magnitude as the mean value and reflects the highly variable nature of inflow on spatial and temporal scales. Huang et al. [2013] estimate a mean sensitivity of nearly 3 times our budget term, but given the two different approaches, they are not directly comparable. Also given in Table 2, the 90th percentile of O3INFLOW in the afternoon over CA is calculated as 24 ppbv (corresponding to 53% of total ozone), which implies that 10% of the time inflow accounts for over 53% of surface ozone over CA. For nighttime, the absolute concentrations of O3INFLOW are reduced (7 ± 7 ppbv) but given the overall lower surface ozone mixing ratios at night, the relative contribution is higher (24 ± 21%).

Table 2. Statistics of Total and Tagged Surface O3a
Mean ± StdMedianMean ± StdMedian10th–90th1st–99th
  1. a

    Absolute and relative contribution averaged over CA and for 18 June to 10 July 2008.

WRF_12 km58 ± 215610 ± 9 (20 ± 21%)7 (13%)<1–24 (1–53%)<1–36 (<1–81%)
WRF_12 km34 ± 18347 ± 7 (24 ± 21%)5 (17%)<1–17 (2–56%)<1–28 (<1–84%)

[37] The longitude-altitude distributions of total ozone and O3INFLOW as well as of COINFLOW and the average height of the planetary boundary layer (PBL) are shown in Figures 8a and 8b for cross sections at 34N (roughly latitude of LA), 38N (roughly latitude of the Bay Area), and 41N (roughly Trinidad Head). We include two different times of the day: 00 UTC (~local afternoon; Figure 8a), and 08 UTC (~local midnight; Figure 8b). These 3 week average plots are not intended to demonstrate specific events of entrainment of free tropospheric air into the boundary layer but rather give an indication of the vertical structure of inflow.

Figure 8.

Longitude-altitude cross sections for O3 (top row), O3INFLOW (middle row), and COINFLOW (bottom row) at 34°N (about latitude for LA), 38°N (Bay Area), and 41°N (Trinidad Head) for (a) 00 UTC (local afternoon) and (b) 8 UTC (local night). The dotted black line indicates the average PBL height. Average for 18 June to 8 July.

[38] At all latitudes, we see a sharp vertical gradient in ozone and O3INFLOW over the ocean with lowest mixing ratios in the marine boundary layer and mixing ratios increasing with altitude. COINFLOW, in contrast, shows the largest values over the ocean surface and is more well mixed in the free troposphere because of differences in chemistry, deposition, and the longer chemical lifetime of CO. The CO tracer only refers to CO that has been transported into the domain and does not refer to CO produced through photochemical reactions of hydrocarbons entering the regional domain (section 2.2). Lower OH concentrations within the marine boundary layer compared to the free troposphere will lead to less destruction of COINFLOW near the ocean surface. The transport of CO into CA using the CO tracer method in WRF-Chem has been discussed in Pfister et al. [2011a].

[39] Both O3INFLOW and COINFLOW show values increasing toward higher latitudes. Air masses entering inland get lofted over the high mountain terrains, where the high PBL during the afternoon allows for entrainment into the PBL and mixing down to the surface. At night, COINFLOW and O3INFLOW feature a similar distribution as during the daytime, but with the collapse of the PBL, the air aloft is now separated from the surface. Total ozone shows a residual high ozone layer just above the PBL from the previous day.

3.2 Air Quality Relevant Statistical Analysis

[40] Here we take a closer look at the role of pollution inflow to surface air quality. Figure 9 illustrates the relationship between local afternoon surface NOx mixing ratios, taken as a surrogate for the degree of pollution, and tagged tracer contributions to reflect whether the impact in inflow changes between polluted and less polluted regions. In addition to O3INFLOW, we also show COINFLOW, which in contrast to ozone is impacted much more by transport and less by chemistry. We limit the analysis to areas below 500 m in surface elevation; the conclusions, however, do not change if all data are applied or if we limit the analysis to only data not influenced by fire plumes using the model CO from fires as filter.

Figure 9.

(b–d) Total O3, O3INFLOW, and COINFLOW as function of NOx for daytime over CA. Only (a) grids with elevation < 500 m have been selected. The black symbols represent individual data points, while red symbols and bars indicate mean and standard deviation for bins of NOx mixing ratios, respectively.

[41] At low NOx, the absolute contributions of total O3 increase with increasing NOx, reaching a maximum at about 2 ppbv of NOx. At higher levels of NOx, total O3 decreases reflecting a more efficient ozone production in NOx-limited versus NOx-saturated regimes. O3INFLOW shows a steady decline with increasing NOx suggesting that the influence of inflow is relatively stronger in less polluted regions. To first order, transport does not seem to account for this relationship as COINFLOW tends to increase with increasing NOx. Even though these statistics are impacted by a dominance of data points in the low emission range with varying meteorological and dynamical influence, the results strongly suggest an overall reduced influence from inflow of ozone and ozone precursors in highly polluted regions. The likely reason for this is less efficient ozone production and possible ozone titration under high NOx conditions, but more comprehensive analysis and modeling work are needed to confirm and quantify this hypothesis. Rural areas, with lower NOx levels, may be preferentially affected by inflow ozone, but these regions have fewer monitors and thus high ozone incidents remain undocumented. It further is of interest for future predictions where it is estimated that emissions over the U.S. will likely decrease while at the same time emissions from Asia go up, leading to an increased inflow of ozone and ozone precursors into regions with reduced NOx.

[42] Figure 10 provides a qualitative model estimate of the contribution of inflow on high ozone events, specifically exceedances of the 8-hour national ambient air quality standard of 75 ppbv. Over the considered time period, most exceedances are simulated in N-CA, in part due to the fires, with exceedances also throughout the Central Valley, the Bay Area, and the South Coast Basin (Figure 10a). Ozone violations also stretch into Nevada, though these are not considered in the following statistics. The number of violations calculated from surface observations at the EPA monitoring sites shows a similar spatial pattern (Figure 10e), but overall, the model underestimates the number of exceedances, which is in agreement with the model underestimating the tail of the distribution discussed in Figure 2 and also shown in Figure 10f illustrating the frequency distribution of observed and modeled 8 h ozone at surface monitoring sites.

Figure 10.

(a) Number of exceedance days with modeled daily 8 h maximum O3 > 75 ppbv. (b–d) Mean, maximum, and standard deviation of 8 h O3INFLOW (%) for the date and time of modeled daily 8 h exceedances. (e) Number of exceedance days from observed daily 8 h maximum O3 > 75 ppbv; black dots indicate sites without exceedance. (f) Frequency distribution of observed (black) and modeled (red) 8 h O3 at monitoring sites. (g) Mean (red triangles), median (blue diamonds), standard deviation (red bars), 10th–90th percentile (thick black bars), and 1st–99th percentile (thin black bars) of 8 h O3INFLOW (%) as function of 8 h surface O3 over CA. (h) The same as Figure 10g but showing absolute 8 h O3INFLOW (ppbv) as function of observed 8 h surface ozone at monitoring sites.

[43] The O3INFLOW contribution during modeled exceedance events is on average less than ~5% or less but is highly variable and can reach values of up to 10–15% (Figures 10b–10d). The average modeled contribution of O3INFLOW over CA to the 8 h maximum on days with an 8 h maximum > 75 ppbv is 4 ± 3% with a median of 3%. Assuming a 65 ppbv threshold, the relative contributions increase and we estimate a mean contribution of 6 ± 5% and a median of 5%. While on average inflow makes up for a relatively small part of high ozone events, its significance can increase in some cases. This is reflected in Figure 10g showing a statistical analysis of the contributions of O3INFLOW to 8 h surface ozone for all of CA.

[44] O3INFLOW has the largest relative impacts on moderate surface ozone concentrations and its impacts lessen toward higher ozone mixing ratios. Statistics for locations with monitoring sites (not shown here) look similar as statistics for entire CA but do not extend up to the highest end of surface ozone values and show slightly smaller contributions. This can be somewhat explained in that the majority of the monitoring sites are located in urban regimes, where, as was shown above, inflow might contribute less efficiently. We calculated percentiles of the percentage contribution of inflow on surface ozone and find that for 8 h ozone mixing ratios larger than 75 ppbv, the 99th and 90th percentiles are 13% and 8%, respectively. This demonstrates that inflow can have significant contributions during high ozone events. This is even more pronounced for a lower limit; for 8 h ozone greater than 65 ppbv the 99th and 90th percentiles increase to 21% and 12%, respectively. Similar statistics are calculated if data points influenced strongly by fire plumes (COFIRE > 20 ppbv) are omitted. The analysis is then more strongly weighted by data in S-CA, and the 99th and 90th percentiles are 15% and 10% respectively for a 75 ppbv threshold (25% and 15% for a 65 ppbv threshold, respectively).

[45] Last, Figure 10h shows similar statistics as Figure 10g, but now we plot the absolute values of O3INFLOW as function of observed surface ozone at monitoring sites. Assuming that the model underestimate of high ozone events is largely due to uncertainties in local processes and not to inflow (Figure 10f), these statistics provide insight into the role of inflow of ozone and ozone precursors on high ozone events. On average, O3INFLOW accounts for 5 ppbv or less when 8 h ozone is larger than 65 ppbv or 75 ppbv but can reach ~10–15 ppbv. The upper 90th percentile is calculated at 13 ppbv for 8 h ozone greater than 65 ppbv and at 11 ppbv for 8 h ozone greater than 75 ppbv.

[46] While not directly comparable, the results are in the range of estimates by Fiore et al. [2002] who, by tagging odd oxygen production and loss, estimate that the maximum influence of Asian and European emissions on surface afternoon ozone in the range 50–70 ppbv over the U.S. during summer 1995 is up to 14 ppbv and declines above 70 ppbv. Lin et al. [2012], in contrast, find that the influence of Asian emissions is increasing with higher values of surface ozone and estimate that 53% of events with daily 8 h maximum ozone above 75 ppbv would not have occurred in the absence of Asian emissions. The latter study focused on the time period of May–June 2010, when Asian transport is stronger compared to later into the summer. They also focused their analysis on Southern CA and Arizona and based the estimates on sensitivity simulations with Asian emissions turned off.

[47] While the various studies differ in their specific outcome as a result of different methods, metrics, models, time periods, and study regions, all of them conclude that even though ozone exceedances are predominantly due to local influences, transported ozone might account for a non-negligible and occasionally significant contribution. Similar to Lin et al. [2012], we estimate the number of high ozone events that would have not occurred without the contribution of ozone from outside the model boundaries (i.e., we calculate the frequency of high ozone events for modeled 8 h surface ozone with and without contribution of O3INFLOW). We find for entire CA that 26% of the exceedances of a 75 ppbv 8 h ozone threshold (and 32% of exceedances for a 65 ppbv threshold) would not have happened without contributions from O3INFLOW; omitting fire-impacted data points, these estimates increase to 35% (46%) since many of the high ozone events during this time period were caused by emissions from the fires.

4 Summary

[48] Interpretation of pollution concentrations is a complex problem due to the interplay of sources, transport, and chemical processes. Modeling studies and synthetic model tracers can shed light into this question and provide valuable information about the importance of different processes. In this study we focus on the role of pollution transport on surface ozone over California during the ARCTAS-CARB field campaign in June–July 2008. This period was strongly influenced by wildfires and might not be fully representative for a typical summer. We include a tagging scheme for ozone into the regional chemical transport model WRF-Chem and keep track of ozone that is due to inflow at the domain boundaries (O3INFLOW), which allows us to quantify the contribution of inflow of ozone and ozone precursors to the surface ozone budget. This approach is different from other studies that used perturbation simulations to assess the sensitivity of surface ozone to inflow. Given the nonlinearity in ozone chemistry, these two techniques were shown to lead to quite different estimates of the contribution of inflow to total ozone. Which of these is the appropriate method to use depends on the objective. The major difference is that the tagging scheme does not alter the current chemical regime and derives an actual contribution or budget term, while the perturbation technique alters the chemical environment and provides an estimate of the sensitivities to changes in a specific source term.

[49] The model simulations are evaluated by comparison to available in situ measurements from the EPA surface monitoring network and the DC-8 aircraft as well as ozonesonde launches and OMI tropospheric NO2 retrievals. This evaluation showed generally good agreement with the observations. A low bias in free tropospheric ozone was caused by too low boundary conditions, which might lead to a low bias in our estimates of the role of inflow. The model represents well the distributions of observed mixing ratios and changes between polluted and less polluted regimes but tends to underestimate high ozone peaks, specifically in Southern California, which is attributed to low ozone production efficiency and strong titration as a result of too high NOx levels.

[50] We estimate an average contribution of O3INFLOW to surface afternoon ozone over CA of 10 ± 9 ppbv (20 ± 21% of total ozone). Inflow not only plays a role during low and moderate ozone events but can also impact high ozone events. We find that for 8 h surface ozone larger than 75 ppbv, inflow contributes more than 8% to total ozone in 10% of the cases and with more than 13% in 1% of the cases. These estimates increase if a lower limit of 65 ppbv is considered (12% and 21%, respectively), or if data impacted by fires are omitted (9% and 15% for a 75 ppbv limit).

[51] Various aspects of the analysis support the hypothesis that pollution transport from afar more effectively contributes to surface ozone in less polluted versus polluted regions. This might have important implications in rural areas, where monitoring is sparse and high ozone events might remain undocumented, and when future reductions in local emissions are considered, but further studies are needed to confirm this assumption.

[52] As was shown from this and previous studies, pollution transport is an important factor in understanding and correctly modeling local surface air quality. Modeling studies assessing the contributions of a highly variable source are subject to a number of uncertainties, and, specifically when conducted at high spatial resolution, are limited to selected time periods only. Uncertainties range from representing local emissions and chemistry, transport, and mixing processes up to uncertainties in boundary conditions. Global models commonly used for boundary conditions have underestimated high pollution levels in transported plumes [Rastigejev et al., 2010], which potentially underestimates the degree to which long-range pollution contributes to high surface ozone events. Improved representation of inflow is a prerequisite for improving upon a quantitative estimate of long-range transport to local air quality, and the growing field of assimilating trace gas retrievals from satellites proves a valuable step in this direction.


[53] The authors like to acknowledge Stu McKeen (NOAA ESRL) for support in using the EPA-NEI emissions inventory and the ARCTAS-CARB science team for providing aircraft measurements. We further acknowledge Mary Barth, Xiaoyan Jiang, and three anonymous reviewers for providing highly valuable input to the manuscript. This research was supported by the NASA AQAST project (grant NNX11AI51G). NCAR is operated by the University Corporation of Atmospheric Research under sponsorship of the National Science Foundation.