Surface ozone at a coastal suburban site in 2009 and 2010: Relationships to chemical and meteorological processes



[1] Air quality and meteorological measurements were conducted at the Chemistry and Physics of the Atmospheric Boundary Layer Experiment (CAPABLE) site during the summers of 2009 and 2010 in Hampton, Virginia. Significant differences in surface ozone mixing ratios were observed between the two years and are correlated with meteorological parameters such as temperature, humidity, and cloud cover. The number of exceedance days for ozone set by the U.S. Environmental Protection Agency within this region has been decreasing for the past decade, especially in urban areas. There were no exceedance days with respect to ozone in 2009, and there were four exceedance days in 2010. The four highest ozone daily maxima and the two exceedance days observed during the 2010 measurement period were coincident with sea breeze phenomena. In one case, surface ozone increased at a rate of 14.6 ppb h−1 with the passage of a sea breeze front. A comprehensive multilinear regression model as well as an operational forecast was unable to resolve the high ozone observed during sea breeze events. As the number of exceedance days per year within this region continues to decrease, accurately forecasting sea breezes may become more important for the forecasting of pollution events.

1. Introduction

[2] Tropospheric ozone (O3) is a criteria EPA-regulated pollutant, which causes respiratory diseases in humans and decreases photosynthesis rates in vegetation [e.g., Karnosky et al., 2007]. Tropospheric ozone is also a greenhouse gas, accounting for approximately 0.35 W m−2 radiative forcing since preindustrial times [Intergovernmental Panel on Climate Change (IPCC), 2007]. Multidecadal trends show that global ozone concentrations may be increasing [Vingarzan, 2004] even though NO2 column amounts, a significant precursor to O3, are suggested to be decreasing in most regions [Richter et al., 2005]. The lack of correspondence between O3 and NO2 trends illustrates the nonlinearity of O3 production to its precursor emissions [Cohan et al., 2005; Xiao et al., 2010].

[3] The accumulation of O3 at the surface in the eastern United States is largely driven by dynamic variability since precursor emissions are roughly constant day to day with the exception of large spills/accidents. Maximum daily O3 mixing ratios are correlated with temperature [Cox and Chu, 1993; Sillman and Samson, 1995; Walcek and Yuan, 1995; Camalier et al., 2007; Bloomer et al., 2009], although some cases show a lack of correlation [Luria et al., 2008]. Ozone mixing ratios have been shown to vary in proportion with temperature indirectly through higher downward solar radiation, increased biogenic and anthropogenic emissions, cloud cover and season [Guenther et al., 1993; Jacob et al., 1993; Sillman and Samson, 1995; Walcek and Yuan, 1995; Bloomfield et al., 1996; Olszyna et al., 1997; Aw and Kleeman, 2003; Rubin et al., 2006; Camalier et al., 2007; Mahmud et al., 2008; Bloomer et al., 2009, 2010; Ito et al., 2009; Jacob and Winner, 2009; Weaver et al., 2009; Steiner et al., 2010].

[4] Surface O3 mixing ratios are inversely correlated with wind speed owing to the lack of dilution of precursor emissions [Banta et al., 2005, 2011], and they can vary by wind direction depending on the locations of regional and point-source precursor emissions.

[5] Surface O3 mixing ratios may vary inversely to boundary layer (BL) height since precursor emissions emitted into a larger volume are better diluted. However, the conditions favorable to higher BL heights are also favorable for O3 production. The confounding processes have resulted in weak correlations between the two parameters [Rappenglück et al., 2008; Banta et al., 2011]. Coastal sites, in particular, are unique in that they are influenced by mesoscale phenomena such as the land/sea breeze (hereafter referred generally as sea breeze), marine BLs and internal convective BLs. The layering effects observed within the lowest layers of the atmosphere (<5 km) make it difficult to define the mixed layer height [Angevine et al., 1996], thus making it challenging to objectively quantify the relationship between BL height and surface pollutant concentrations.

[6] Sea breezes are driven by dynamic and thermodynamic surface forcings and have been known to dramatically change surface O3 mixing ratios over short time periods [Angevine et al., 1996, 2004; Gaza, 1998; Seaman and Michelson, 2000; Clappier et al., 2000; Ryerson et al., 2003; Banta et al., 2005, 2011; Darby, 2005; Rappenglück et al., 2008; Tucker et al., 2010]. Figure 1 provides a conceptual model of the important synoptic and mesoscale dynamics that initiate and propagate a sea breeze front and associated sea breeze circulation inland. Ozone precursors from terrestrial sources advect over marine surfaces, where they can concentrate within shallow, stable marine BLs. Once ozone is photochemically produced over water, it does not readily deposit to the surface because it is relatively insoluble. A typical O3 dry deposition velocity over land is 0.4 cm s−1 compared to 0.07 cm s−1 over water [Lenschow et al., 1982; Hauglustaine et al., 1994]. If the seaward directional component of the flow is weak, the buoyancy force created by the differential heating of the air over land (the temperature over land being relatively hotter than the temperature over water) can be enough to initiate an inland wind direction near the surface, or a sea breeze. During propagation of the sea breeze, the high O3 mixing ratios that built up over the water are advected onto land and coupled with in situ photochemical production of O3 from precursors that are being emitted into a more shallow and stable afternoon BL.

Figure 1.

Conceptual model of the processes prior to and after the development of a sea breeze propagating landward and how it impacts surface O3 concentrations. In these sea breeze cases the surface temperature over the land in the morning is approximately equal to or less than the surface temperature over the water. Owing to daytime heating, the surface temperature over the land becomes greater than the surface temperature over the water in the afternoon.

[7] The spatial extent of sea breeze fronts is determined by the thermal gradient between the land and the sea, the height of the circulation, the magnitude of the cross-shore horizontal wind component, the landward propagation of the sea breeze front and the strength and direction of the large-scale geostrophic flow [Miller et al., 2003]. Local variations in coastline curvature, topography and water-body dimensions can impact the geophysical relationships [Crosman and Horel, 2010, and references therein]. The timing of the initiation and propagation of the sea breeze front are critical in determining the impact that it will have on O3 concentrations. Fronts that form later in the day allow more buildup of precursors and photochemical production of O3, thus higher gradients are typically observed when the sea breeze front propagates inland [Oh et al., 2006]. Small spatial scales and high gradients in pollutant concentrations associated with sea breeze fronts make it difficult for air quality models to accurately represent the dynamic and chemical processes. Meteorological representation of the sea breeze front and the driving forces of its initiation and propagation limit the representativeness of air quality models [Zhang et al., 2007].

[8] Larger-scale synoptic influences may contribute to the variability of surface O3 since O3 is somewhat long-lived in the troposphere. Persistent subsidence can lead to stable conditions that inhibit the venting of pollution [Gangoiti et al., 2002] and intra and intercontinental long-range transport of ozone, especially over water surfaces, have been observed [Angevine et al., 2004; Mao et al., 2006; Cooper et al., 2010].

[9] Relating surface O3 concentrations to these meteorological processes is difficult because of multiple interacting physical and chemical processes that cause surface O3 to be temporally and spatially variable. The number, location and sampling rates of continuous monitoring stations limit the extent to which measurements can effectively capture surface O3 variability. Satellites provide a solution to the need for spatially resolved measurements. Extracting tropospheric O3 concentrations from total column measurements is difficult since ∼90% of column O3 is located in the stratosphere. Recently, tropospheric O3 signatures from biomass burning [Jourdain et al., 2007], large metropolitan areas [Kar et al., 2010] and crop yield losses in the Midwestern United States [Fishman et al., 2010] were detected using satellite measurements. However, these measurements require significant temporal averaging to deduce a tropospheric/surface signal and the methods cannot capture the hourly variability observed at the surface. In addition, the retrievals of tropospheric signals from satellite measurements have been shown to be biased low when compared to integrated in situ balloon-borne ozonesonde measurements within the troposphere and the biases have been related to synoptic and mesoscale meteorological conditions [Doughty et al., 2011]. It is still unclear how tropospheric column O3 amounts relate to surface O3 variability. By better understanding local dynamic processes, we can improve modeling capabilities and thereby potentially improve satellite retrieval algorithms for existing and future satellite measurements.

[10] Statistical models have been employed to investigate and identify the relative contributions of certain meteorological parameters on surface ozone [Wise, 2009; Wise and Comrie, 2005; Camalier et al., 2007; Holloway et al., 2008; Davis et al., 2011]. Camalier et al. [2007] identified the most important meteorological variables contributing to the observed surface ozone variability and separated the results on the basis of geographic zones along the east coast of the United States. The important parameters in the Mid-Atlantic states, including Virginia, were daily maximum temperature, mean daytime relative humidity, transport direction and transport distance. These parameters, along with some others, explained ∼80% of the observed variability in surface ozone. Similarly, Davis et al. [2011] found that these few parameters can explain between 50 and 80% of the daily maximum 8 h average surface ozone concentration variability. In the work of Davis et al. [2011] the lowest R2 values are along the coastline, including Virginia Beach, Virginia. These results make the measurements of trace constituents and meteorology along a coastline especially important.

[11] Here, we report detailed analyses using surface and remote sensing air quality measurements collected in an urban/suburban coastal environment. Measurements were conducted as part of the Chemistry And Physics of the Atmospheric Boundary Layer Experiment (CAPABLE) site during the summers of 2009 and 2010 at the NASA Langley Research Center (LaRC) in Hampton, Virginia (37.1036°N, 76.3869°W; see Figure 2). This paper characterizes the important relationships between meteorological, chemical and anthropogenic processes on surface ozone.

Figure 2.

CAPABLE site in Hampton, Virginia.

2. Methods

2.1. Site Description and Historical Context

[12] The CAPABLE measurement site is urban/suburban with a population of one million in the metropolitan area that includes Hampton, Newport News, Norfolk and Virginia Beach, Virginia (Figure 2). Continuous measurements were made from 17 July to 1 October 2009 and from 16 June to 7 August 2010. During the summer, local time is in Eastern Daylight Savings (UTC–4); however, owing to the photochemical nature of ozone production, all times discussed here will be in local time (eastern standard time; UTC–5) unless stated otherwise. The proximity of the site to the mouth of the Chesapeake Bay/Atlantic Ocean (∼10 km to the east) makes the site ideal for studying how coastal meteorology affects air quality.

[13] Historically, Hampton's number of exceedance days of the EPA standard for O3 (8 h average above 75 ppb) is comparable to the rest of Virginia, which is mainly upwind of major urban centers (Figure 3). A general decrease in the number of O3 exceedance days in the Mid-Atlantic region spanning from 1990 to 2010 has been observed, especially in urban areas such as Baltimore and Washington DC. In particular, a sharp decrease in the number of ozone exceedance days in the region is observed after 2002. This decrease is concurrent with increased regulations on NOx emissions throughout the United States; subsequently reducing NOx levels throughout the troposphere by about 30% from 1997 to 2005 [Kim et al., 2006]. Hampton on average exceeded the EPA standard for O3 12 days per year over the past two decades. The most polluted years with respect to O3 occurred from 1997 to 1999, when the Hampton site observed 31, 28 and 30 exceedance days. The five most recent years (2006–2010) of the Hampton data record have been relatively unpolluted with respect to O3, averaging only 3 exceedance days per year.

Figure 3.

Number of O3 exceedance days defined as having an 8 h running mean of greater than 75 ppb for areas within the Mid-Atlantic region. Values presented are the averages of multiple air quality monitoring stations within each area. (Data courtesy of Maryland Department of Environment and the Virginia Department of Environmental Quality.)

2.2. Measurements

[14] The Nittany Atmospheric Trailer and Integrated Validation Experiment (NATIVE;, a mobile platform housing in situ and remote sensing chemical and meteorological instrumentation, was deployed at LaRC to provide surface observations and ground validation for satellite and other remote sensing instrumentation during the summers of 2009 and 2010. The trailer has a 10 m extendable tower on which instrumentation was mounted. Sample air from 4 m above the surface was drawn through a heated (35°C) glass manifold at a rate of ∼60 L per minute. The residence time within the sample inlet and manifold is <2 s. The trace gas analyzers, except for the NOy analyzer, sampled from the manifold at their respective specified volumetric flow rates at atmospheric pressure. Ozone was measured using a UV-absorption spectrometer at 254 nm (TECO, Inc., Model 49C). The O3 analyzer was calibrated before and after each research campaign using an O3-generating ambient monitoring calibration system (Environics EN-9100) traceable to a NIST primary standard. The response of the instrument was monitored weekly using the output from an internal ozonator of a multigas calibration system (TECO, Inc., Model 146C) and relative uncertainties of typical ambient mixing ratios were ±2%. Nitric oxide (NO) and the sum of reactive odd-nitrogen species (NOy ≡ NO + NO2 + HNO3 + HONO + peroxynitrates + organic nitrates) were measured using a chemiluminescence reaction of NO with O3 to produce a detectable photon (TECO, Inc., Model 42C-Y). Total reactive nitrogen was converted to NO using an external molybdenum converter heated to 325°C located 25 cm downstream of the inlet and mounted to the tower 8 m above the ground. The NO-NOy analyzer was calibrated approximately 3 times per week and the mean uncertainty of NO and NOy within the range of measurements during the campaign was ±3%. Sulfur dioxide (SO2) was measured using a pulsed fluorescence SO2 analyzer (Thermo Scientific, Inc., Model 43C). Carbon monoxide (CO) was measured using an infrared absorption spectrometer (Thermo Scientific, Inc., Model 48C). A zero calibration on the CO analyzer was conducted every 3 h using an internal CO scrubber. The SO2 and the CO analyzers were calibrated weekly. The mean uncertainties of the SO2 and CO mixing ratios were ±5% and ±4%, respectively.

[15] A suite of meteorological instrumentation (R.M. Young, Inc.) provided continuous measurements of temperature, relative humidity, barometric pressure, wind speed and direction. The propeller anemometer was mounted on top of the 10 m extendable tower to avoid very local flow distortions from surrounding trees and buildings and to best characterize regional flow. Pressure, temperature and relative humidity were measured at the height of the trace gas sampling manifold inlet on top of the NATIVE trailer and 4 m above the ground. Wind directions during which the wind speed was less than 1 m s−1 were assigned values of “calm.” The in situ trace gas and meteorological measurements were collected continuously and averaged over 1 min.

[16] Vertical profiles of O3 were measured using balloon-borne ozonesondes with electrochemical concentration cell (ECC) devices (ENSCI Corp., Model 2Z). A 0.5% buffered solution of potassium iodide (KI) was used to react with O3, quantitatively producing 2 electrons per reaction, and resulting in a detectable current. Ozone concentration uncertainties are <10% below 30 km [Smit et al., 2007]. The average burst altitude of the sondes was 37 km and the vertical resolution of the measurements was ∼10 m. Each ozonesonde was equipped with a pressure-temperature-humidity (PTU) radiosonde (Vaisala, Inc., RS-80) that transmitted data to a NATIVE ground station. A total of 26 ozonesondes were launched in the two summer campaigns.

[17] A total of 38 rawinsondes (RS-92SGP, Vaisala) were launched during the 2010 campaign. The rawinsondes were equipped with a GPS receiver and measured temperature, pressure, dew point, wind speed and wind direction throughout the troposphere and lower stratosphere. The average burst altitude was 25.5 km and the vertical resolution of the measurements was ∼7 m.

[18] Three-dimensional wind speed and direction were measured using a portable wind profiling lidar (Windcube 70, Leosphere). The lidar has an altitude range between 100 and 1500 m with a 50 m vertical resolution. The lidar measurements were averaged every 5 min. Nominal accuracies for the wind speed and direction are 0.3 m s−1 and 1.5°, respectively.

3. Results

3.1. Site Chemistry

[19] The mean diurnal cycles of surface O3 mixing ratios and reactive nitrogen oxides as measured by NATIVE are shown for 2009 and 2010 during overlapping time periods (17 July to 7 August; see Figure 4). For comparison, the 2000–2008 hourly mean diurnal cycle is also shown (Figure 4a) using data from a nearby (11 km), long-term monitoring station (site id: 51-650-0004; Environmental Protection Agency Air Quality System) during June–August (JJA). The diurnal cycles are typical of an urban/suburban site. Ozone increases at sunrise owing to photochemical production from radical chemistry, photolysis of accumulated nighttime NO2 and entrainment from aloft into a growing BL. The growth rates of surface O3 mixing ratios calculated by averaging the daily slopes during both campaigns were 5 ± 4 ppb h−1 during the morning (06:00–12:00 LT) and −2 ± 5 ppb h−1 in the afternoon (12:00–18:00 LT), the latter not being significantly different from zero (p value = 0.12). The afternoon equilibration and subsequent decrease in surface O3 mixing ratio is expected since the amount of available energy used for the photochemical production of radicals decreases after solar noon [Ren et al., 2003]. The variability of the slopes can be attributed to meteorological variability, such as temperature, relative humidity, wind speed, wind direction, solar irradiance, emissions and synoptic and mesoscale phenomena, and is discussed in more detail below. Five days were omitted from the afternoon growth rate analyses owing to the evidence of surface conditions being impacted by sea breeze phenomena. The variability, as indicated by the ±1σ dashed lines in Figure 4 is the largest during midday. During the evening and nighttime hours, a shallow stable BL forms near the surface. Titration of O3 with fresh emissions of NO and O3 dry deposition to the surface depletes O3 in the lower (<500 m) layers of the troposphere during the night. The shapes of the diurnal cycles during both years are very similar, but the surface ozone mixing ratios were on average 15 ppb greater in 2010 compared to 2009. The mean daily maximum in 2010 was 59 ppb compared to 45 ppb in 2009. The 2009 campaign was much less polluted with respect to O3 compared to the 2000–2008 5mean, whereas 2010 was slightly more polluted. It is important to note that 2010 is more similar to the decadal mean than 2009; the implications of which are discussed below.

Figure 4.

Mean diurnal cycles of (a) O3 and (b) NOy surface mixing ratios during the CAPABLE summer campaigns of 2009 (blue) and 2010 (red). The 2000–2008 hourly means of O3 during June, July, and August from a nearby EPA monitoring site in Newport News, Virginia, are shown for comparison (dash-dot black line). Dashed lines encompass ±1σ of the 20 min binned averages. Only CAPABLE data during coincident measurement periods are shown (17 July to 7 August).

[20] Total reactive nitrogen peaks in the morning just after sunrise from the increase in anthropogenic emissions related to the local morning rush hour and an undeveloped/developing BL (Figure 4b). As the BL grows through midmorning, mixing ratios of NOy become diluted. At night, the buildup of NO2 and the development of a stable nocturnal BL concentrate NOy near the surface. The shapes of the NOy diurnal cycles are similar during the 2009 and 2010 campaigns, differing by an average of ∼10%. The summer of 2009 shows slightly higher NOy surface mixing ratios in the morning, and slightly lower NOy surface mixing ratios throughout the afternoon. The possible reasons for these observed differences are discussed below.

[21] The characterization of the VOC-NOx regime [National Research Council (NRC), 1991] in this region can be useful to understand the local chemical production of O3. Speciated and/or total VOC measurements are limited since they are not part of the NATIVE standard payload. The EPA measures speciated VOCs as part of the Photochemical Assessment Monitoring Stations (PAMS); however, the closest PAMS site to the CAPABLE site was 34 km south of the CAPABLE site in Virginia Beach, Virginia (latitude 36.8419; longitude −76.1812; site ID 51-810-0008). These VOC data are averaged over 24 h and collected every sixth day. In 2010, the VOC species measured are limited; the most notable (owing to its OH reactivity) unmeasured VOC is isoprene.

[22] Sharp decreases in O3 mixing ratios at the CAPABLE site were observed when wind directions were from the land (i.e., 180–330°). In the absence of oxidative species in the atmosphere (e.g., RO2, HO2, OH; RO2 ≡ organic peroxy radical), O3 will decrease with fresh NO emissions owing to changes in the NO2/NO ratio affecting the quasi-photostationary steady state of O3. This titration occurs on timescales of about 2 min. The CAPABLE site was located 500 m northeast (48°) of a waste-to-energy steam plant (Hampton/NASA steam plant), a point emission source of NO. Given a mean wind speed of 2 m/s, the time it takes the plume to reach the site is about 4 min. The high NOx values within the plume compete with ambient VOCs for radical reactivity through the reactions:

display math
display math

ultimately decreasing the production rate of O3. The lifetimes (τ), conservatively calculated on the basis of upper limits of regional VOC and radical measurements [Whalley et al., 2010], of O3 during the day with respect to reactions with radicals (τHO2 = 1.5 days; τOH = 7 days), photolysis (τphot = 1 day), ozonolysis of VOCs (τ > 5 days; based on 2008 JJA values of isoprene from the Virginia Beach site) and surface deposition (τ > 15 min) are sufficiently long compared to the transport time of the plume to the measurement site so that destruction processes of O3 other than reaction with NO are negligible. While reaction (R1) is temporary during the day because of the photodecomposition of HONO, the lifetime with respect to this photolysis is on the order of 10 min, making this reaction a sink of OH radicals on the timescales over which the plume travels to the measurement site. Reformation of O3 through the photolysis of NO2 is a slower process (2–3 min) compared to the NO + O3 reaction, even under the highest irradiances observed at the site (Jno2 = 7 × 10−3 s−1). So the NOx-O3 cycle has time to reach equilibrium by the time it reaches the measurement site. Overall, the decrease in O3 production rate and negligible changes to the ozone destruction rates result in a net decrease in O3 under steady state conditions (i.e., production rate = destruction rate). Although useful information regarding the lifetime of O3 can be extracted from regional VOC measurements, it is difficult to assess whether this site is VOC limited or NOx limited owing to data unavailability, coarse spatiotemporal resolution and the proximity of the site to local point sources.

3.2. Meteorological Influences on Ozone

[23] Since persistent meteorological conditions can lead to the buildup of background O3 and NOy and it is not possible to distinguish between local photochemical production and transport using this data set, analyses of meteorological conditions from the National Centers for Environmental Prediction (NCEP) North American Regional Reanalysis (NARR) data set was conducted spanning both 2009 and 2010 measurement periods during June–August. This period includes the peak ozone formation season. The NARR data set is derived by using the NCEP Eta model at 32 km resolution with the Regional Data Assimilation System (RDAS) eight times daily [Mesinger et al., 2006]. Figure 5 shows the downward solar radiation and cloud fraction differences between 2009 and 2010. Daily means and maxima of downward solar radiation and cloud fraction were calculated for each data set using the averages of the four closest grid points to the CAPABLE site and considering only daytime NARR outputs (07:00–16:00 LT). Total cloud fraction was calculated by summing the nonconvective and convective cloud fractions. The top and bottom of the boxes in Figure 5 define the interquartile range, and the horizontal line in the center of the box is the median. The whiskers define the extreme values and no outliers (±3σ) were observed in this analysis. The median for downward solar radiation daily mean is significantly lower in 2009 than in 2010 at the 95% confidence level. The daily maxima of downward solar radiation are lower at the 75% confidence level. The confidence intervals were calculated using a bootstrap statistical approach with a median resample size of 10,000. The total cloud fraction median is statistically higher in 2009 than in 2010 at 75% confidence using the same bootstrap method. The lower downward solar radiation can possibly lead to a delayed development of the morning boundary layer, essentially trapping more of the morning rush hour NOx and leading to higher morning NOy values in 2009. However, without the simulations of a high-resolution chemical transport model, the cause of the observed differences in NOy between 2009 and 2010 is speculative.

Figure 5.

Daily mean and maximum averages calculated from the NARR data set of downward solar radiation and total cloud fraction for June, July, and August of 2009 and 2010 at Hampton, Virginia. The cloud fraction percentage is greater than 100% in the extreme values owing to the summation of the convective and nonconvective cloud fraction NARR output variables.

[24] Temperature has been correlated with surface ozone mixing ratios [Bloomer et al., 2009; Steiner et al., 2010]. Average daily maximum temperatures during JJA were higher in 2010 (32.3°C) than in 2009 (30.6°C). The mean temperatures in JJA during 2010 and 2009 were 24.9°C and 22.7°C, respectively. Precipitation totals in JJA were 150.6 mm in 2010 and 226.6 mm in 2009. The 50% higher precipitation in 2009 compared to 2010 is likely to provide additional sinks of ozone precursors, namely NOy through the wet deposition of nitric acid. While we cannot say that the surface ozone differences are caused by the differences in meteorological conditions shown in this analysis, we do show that the meteorological conditions in 2010 were more conducive to in situ photochemical production of ozone.

[25] BL heights were defined using vertical potential temperature gradients from ozonesondes and rawinsondes. A total of 53 sondes were launched during the campaign. The top of the BL was designated as the height of the inversion nearest the surface having [Heffter, 1980]:

display math
display math

where inline image is the lapse rate of potential temperature and θtop and θbase are the potential temperatures at the top and the base of the inversion, respectively. Other methods to determine the BL height were tested, such as using the second derivative of equivalent potential temperature [Stull, 1988] and maintaining static BL heights of 2 km, but these methods failed to reliably define realistic heights, especially during stable conditions. The defined BL height can be influenced by the vertical resolution of the sonde data. Thus, the vertical profiles of potential temperature from the sondes were smoothed by block-averaging the data into 50 m bins. The average flight time of the sondes was 122 min. Surface O3 measurements were averaged during the flight-sampling period and compared to the measured BL height for each flight. A weak but significant correlation (slope = 0.03 ppb m−1, R2 = 0.13, p value = 0.01) was observed between the mean surface ozone and the BL height (Figure 6), in agreement with previous coastal observations [Banta et al., 2011]. The correlation coefficients and p values were similar when considering the median and the maximum surface ozone mixing ratios during each flight. The analyses can be influenced by the launch time of the sondes. Seventeen of the fifty-three sondes were launched prior to 11:00 local time (EDT), before the boundary layer has a chance to fully develop and before the daytime maximum in surface O3 is observed. When only the afternoon launches are considered (after 11:00 EDT), the correlation decreases and the slope is not significantly different from zero (slope = 0.03, R2 = 0.01, p value = 0.52).

Figure 6.

Surface O3 mixing ratio correlated with BL heights calculated from ozonesondes and rawinsondes (slope = 0.03 ppb m−1, R2 = 0.13, p value = 0.01).

[26] NATIVE surface O3, NO, NOy, CO and SO2 mixing ratios during the 2009 and 2010 campaign were clustered into 45° wind direction bins (Figure 7). North, east, south and west were centered around 0°, 90°, 180° and 270°, respectively. The highest O3 mixing ratios occurred when the wind direction was from the east, while the lowest O3 mixing ratios occurred when the wind direction was from the west. A bootstrap statistical sampling method was applied to the east and west bins to show that the mean mixing ratios in these bins are significantly different with 95% confidence (p value < 0.01). The distribution of O3 mixing ratios contrasts the distributions of CO, NO and NOy. CO, NO, and NOy distributions show the highest mixing ratios when wind directions are from the south and southwest, when the sampled air is more likely to be influenced by anthropogenic emissions. Sulfur dioxide mixing ratios do not change with changing wind direction except in the case of northerly wind directions, when higher SO2 mixing ratios are observed. High SO2 mixing ratios most likely originated from a power plant 13 km to the north-northeast (Yorktown Power Station, Yorktown, Virginia). Interestingly, the lowest observed ozone mixing ratios occurred during calm (wind speeds <1 m s−1) conditions. These analyses include all times of the day and calmer conditions were observed more during the night when a stable nocturnal BL persisted and O3 deposition and O3 titration by NO are more likely to become dominating loss processes at the surface. Figure 7 also illustrates the complexity of the relationship of O3 with other chemical species. The dissimilarities in the shape of the O3 histogram compared to the NOy and CO histograms may be indicative of transport processes driving surface O3 variability rather than in situ chemistry.

Figure 7.

Mean surface mixing ratios of (a) O3, (b) CO, (c) NO, (d) NOy, and (e) SO2 distributed by 45° wind direction bins using 1 min average data during the 2009 and 2010 CAPABLE campaigns.

[27] A possible explanation for the higher O3 values coming from the maritime wind direction sectors (northeast, east, southeast) is pollution from offshore ships, which can represent a significant portion of the total pollution budget in the marine BL [Eyring et al., 2005; Williams et al., 2009]. Since larger mixing ratios of SO2 and NOy were not observed when the wind direction was from these sectors, it is unlikely that the higher O3 mixing ratios observed are due to ship pollution.

[28] Ozone mixing ratio diurnal cycles clustered on the basis of wind direction and time of day are shown in Figure 8. Ozone measurements were binned and averaged over 20 min. Each bin was clustered into 45° wind directions (north centered about 0°) and separated on the basis of the sampling year. The mean diurnal cycles for each year's sampling period are shown in each subplot for comparison. The lack of surface deposition and NO titration from local anthropogenic emissions keep the nighttime maritime (northeast, east, southeast) surface O3 mixing ratios relatively high compared to the nighttime terrestrial (northwest, west, southwest, south) surface O3 mixing ratios. Surface O3 mixing ratios greater than the mean diurnal cycle for each year were observed when midday winds originated from maritime directions, especially in 2010. We cannot conclude that higher morning ozone mixing ratios lead to higher afternoon ozone mixing ratios from this analysis, especially since these profiles are not continuous in time but rather clusters of events. In fact, there was no significant correlation observed between morning O3 surface mixing ratios and daily maximum surface ozone mixing ratios (R2 < 0.1, p value = 0.88). This result suggests that the synoptic flow is less important than entrainment due to a growing BL and/or in situ photochemical production in this region, in contrast to what has been observed in other regions [Morris et al., 2010; Banta et al., 2011]. The higher maritime O3 mixing ratios are possibly the result of a combination of processes, such as surface deposition differences over land and water, anthropogenic emissions and land/sea breeze effects, the latter being discussed in more detail below. Ozone mixing ratios in terrestrial wind direction clusters (northwest, west, southwest, south) showed good agreement or are lower than the diurnal mean for each year.

Figure 8.

Surface O3 mixing ratios clustered on the basis of the time of day in 20 min binned averages (dots) and 45° wind direction sectors (subplots). Analyses were conducted similarly for the 2009 (blue) and 2010 (red) CAPABLE campaigns. Solid lines indicate the mean diurnal cycles for each year and include all wind directions. Clusters with a sample size less than 300 are not shown.

[29] The predominant wind direction during the sampling period was from the west and southwest, representing a combined 50% of the total 1 min averaged wind direction measurements (Figure 9). The third most prevalent wind direction was from the east (12%). When only daytime (10:00–18:00 LT) observations are considered, the percentage of wind directions from the east increases to 19% and all wind direction distributions with an easterly component (N, NE, E, SE) except southerly increases. The difference between total and daytime surface wind direction distributions suggest that this site is influenced by sea breeze dynamics.

Figure 9.

Wind rose plots for (a) the entire CAPABLE 2010 campaign and (b) the daytime (10:00–18:00 LT) only in percentage frequency for the respective data sets.

3.3. Multilinear Regression Model

[30] A multilinear regression model was constructed using 41 predictive daily meteorological variables summarized in Table 1 to explain the variance of daily maximum ozone observations. Daily maximum ozone was chosen as the response variable rather than the maximum 8 h mean of surface ozone so that correlations between meteorological variables and surface ozone during sea breeze events were not dampened. Meteorological conditions describing the temperature, wind field, humidity, pressure, stability and synoptic conditions at the surface and higher isobaric surfaces were chosen as predictor variables. These variables were chosen on the basis of previous studies [Camalier et al., 2007; Davis et al., 2011]. Surface measurements were collected on board NATIVE, while upper air data were collected from the NARR data set. Each variable measured is assumed at the surface unless it is noted otherwise. In some cases, variables during specific times of the day rather than daily means, maxima or minima are used. Morning, afternoon, midday and nighttimes are defined as 06:00–09:00, 12:00–16:00, 10:00–16:00, and 23:00 (previous day) to 04:00 LT, respectively. Since the predictor variables are often correlated with other predictor variables, a stepwise regression process was used to select the most significant predictor variables to be used in the model. In the stepwise regression, variables were iteratively added to the model until no additional variables contributed significantly to the explained variance of the daily maximum ozone. Separate regression models were constructed for 2009 and 2010. The statistics from the regression models are presented in Table 2. In 2009, cloud cover, daily minimum relative humidity and mean afternoon wind speed were used to construct the model. The variables explained 48% of the overall variance in daily maximum ozone. Individually, cloud cover explained 29%, daily minimum relative humidity explained 24% and mean afternoon wind speed explained 18%. The overall explained variance is less than the summation of the individual components owing to correlations between the components. In 2009, the significant meteorological conditions based on the regression model can be classified as ozone inhibiting; that is, the predictor variables are inversely correlated with daily maximum ozone. On the basis of the model intercept, if the cloud cover, minimum relative humidity and mean afternoon wind speed were zero, the predicted daily maximum ozone would be 97 ppb, which is high enough to contribute to an exceedance. In 2010, the significant predictor variables are mean wind speed, mean temperature, mean afternoon easterly component (u) of the surface wind, and mean afternoon temperature. The variables explain 54% of the observed variance in daily maximum ozone. The individual contributions to the daily maximum variance are 21, 3, 9 and 4% for the mean wind speed, mean temperature, mean u and mean afternoon temperature, respectively. The temperature at 925 hPa was a significant contributor to the variance of the response variable (6%), in agreement with previous studies [Holloway et al., 2008]; however, the surface temperature was used in the regression model instead owing to the availability of the measurements and the lack of significant change in the overall explained variance of daily maximum ozone. Although humidity variables were not explicitly included in this regression model, the negative coefficient for the mean temperature predictor variable most likely reflects the importance of humidity on ozone production since nighttime temperatures are less under less humid conditions, which acts to decrease the daily mean temperature. The differences in the regression models for each year highlight the differences in the driving meteorological forces for each year, ultimately leading to the observed differences in the amount of surface ozone for each year. The significant predictor variables for each year are in agreement with previous studies for this area [Camalier et al., 2007].

Table 1. Daily Meteorological Variables Considered in the Formulation of a Multilinear Regression Model to Predict Daily Maximum Surface Ozonea
  • a

    Surface values were measured on NATIVE, while upper air values were taken from the NARR data set. Variables that do not indicate a height value are from the surface.

  • b

    Time of previous day.

Temperaturemaximum temperature
 mean temperature
 minimum temperature
 morning (06:00–09:00 LT) mean temperature
 afternoon (12:00–16:00 LT) mean temperature
 diurnal temperature change
 925 hPa temperature; 12:00 UTC
 850 hPa temperature; 12:00 UTC
 700 hPa temperature; 12:00 UTC
 500 hPa temperature; 12:00 UTC
 850 hPa; surface temperature difference; 12:00 UTC
Windmean wind speed
 mean wind direction
 mean morning (06:00–09:00 LT) U component
 mean morning (06:00–09:00 LT) V component
 mean afternoon (12:00–16:00 LT) U component
 mean afternoon (12:00–16:00 LT) V component
 mean morning (06:00–09:00 LT) wind speed
 mean morning (06:00–09:00 LT) wind direction
 mean afternoon (12:00–16:00 LT) wind speed
 mean afternoon (12:00–16:00 LT) wind direction
Humiditymean dew point
 mean relative humidity
 minimum relative humidity
 maximum relative humidity
 mean midday (10:00–16:00 LT) relative humidity
 mean night (23:00b–04:00 LT) relative humidity
 maximum dew point
 minimum dew point
 maximum water mixing ratio
 850 hPa dew point
 24 h change in 850 hPa dew point
Pressuremean pressure
 850 hPa geopotential height; 12:00 UTC
 700 hPa geopotential height; 12:00 UTC
 500 hPa geopotential height; 12:00 UTC
Stabilitymaximum planetary boundary layer height
 maximum rate of increase in planetary boundary layer height
Synopticmean incoming solar radiation
 mean total cloud cover
 total precipitation
Table 2. Regression Model Statistics and Predictor Variables for 2009 and 2010a
YearVariablesCoefficientsInterceptRMSER2 Adjusted
  • a

    Daily maximum ozone was the response variable.

2009mean total cloud cover−
 minimum relative humidity−0.4   
 mean afternoon wind speed−7.5   
2010mean wind speed−
 mean temperature−2.1   
 mean afternoon u-component−4.7   
 mean afternoon temperature2.4   

[31] The regression models were used to predict the daily maximum ozone observations for each year resulting in a mean bias (model-observation) of 0.37 ppb and −0.10 ppb and a mean absolute error of 6.7 and 6.7 ppb for 2009 and 2010, respectively. The largest absolute error for 2009 occurred on 27 August, concurrent with the highest observed ozone values during the 2009 campaign. The model under-predicts the observed daily maximum by 29 ppb (32%). It is unknown if a sea breeze occurred on this day on the basis of available measurements and the definition criteria of a sea breeze described below. The largest absolute error for 2010 occurred on 7 July, a day in which a sea breeze occurred. The model under-predicted the observations by 24 ppb (21%). Similarly, the observed daily maximum surface ozone concentration was the highest of the 2010 campaign. If the coefficients from the 2010 model are used to predict 2009 observations or if the coefficients from the 2009 model are used to predict 2010 observations, the integrity of the models are greatly diminished. The mean bias and absolute error for the 2010 model applied to the 2009 observations are 14.6 ppb (28%) and 16.0 ppb (31%), respectively. The mean bias and absolute error for the 2009 model applied to the 2010 observations are −8.4 ppb (13%) and 12.3 ppb (19%), respectively. These results highlight the different processes driving the variability of surface ozone for both years as well as indicate that more data are needed to formulate a more robust model that can be applicable across many years.

3.4. Sea Breeze Influences on Surface Ozone Concentrations

[32] The cool temperatures and measurement period, since the initiation of the sea breeze is largely driven by the thermal gradients between land and sea and this gradient is the largest in the late spring and early summer, contribute to the lack of sea breeze events observed in 2009. Therefore, the 2010 campaign will be the focus of the remaining sea breeze analyses and discussions.

[33] In 2010, the site was impacted by sea breezes on 13% (6 of 48) of the days within the measurement period, and possibly up to 21% (10 of 48). Sea breeze phenomena were characterized by (1) a midday shift from westerly to easterly in the near-surface winds, (2) a ∼180° shear in wind direction within the lowest (<2 km) part of the troposphere, (3) an increase in absolute humidity, and (4) the lack of mesoscale to synoptic-scale fronts as defined by NCEP surface analyses ( On four of the ten suspected days that a sea breeze occurred, evidence of a sea breeze exists, but the events cannot be confirmed owing to one or more of these criteria not being met.

[34] Figure 10 shows the typical sea breeze impacts on vertically resolved wind speed, wind direction, surface O3 and surface dew point for 20 June 2010. Morning wind directions on 20 June 2010 were from the west and northwest throughout the lowest kilometer of the troposphere. The wind direction shifted around solar noon, first transitioning to the northeast and then to the southeast in a layer from the surface to ∼600 m. The wind shift was coupled with relatively calm conditions (<4 m s−1) throughout the lowest kilometer. A sharp increase in the dew point temperature was observed at 12:00 LT. A coincident surface O3 increase was observed at the initiation of the sea breeze event. Ozone mixing ratios increased from 62.5 ppb prior to the sea breeze frontal passage (12:00 LT) to a daily maximum of 94.7 ppb at 15:53 LT. The surface ozone mixing ratio growth rate during this period was 9.1 ppb h−1, which greatly contrasts the −2 ppb h−1 average on days in which no sea breeze occurred.

Figure 10.

Sea breeze example on 20 June 2010 showing the impacts on (a) wind direction and surface O3, (b) wind speed, and (c) dew point. The vertical dashed line indicates the sea breeze frontal passage. Wind measurements are from a wind lidar.

[35] The maximum observed surface O3 mixing ratio during the campaign occurred on 7 July 2010 (Figure 11). A 1 min averaged mixing ratio of 114 ppb was observed at ∼15:50 (LT). Sea breeze dynamics caused a shift in wind direction, from west-southwest in the early morning, transitioning to northerly and finally east-southeasterly as observed by the surface wind anemometer. From the initiation of the sea breeze to the peak value in surface O3 mixing ratio, O3 increased at a rate of 14 ppb h−1. Coincidentally, the dew point increased. The increase in surface O3 was enough to cause an exceedance of the 75 ppb, 8 h average standard set by the EPA.

Figure 11.

Impact of sea breeze dynamics on (top) wind direction (black dots) and surface ozone mixing ratio (red line) and (bottom) temperature (black line) and dew point (green line) on 7 July 2010. The initiation of the sea breeze is indicated (vertical dashed line).

[36] Figure 12 shows the differences in the diurnal cycles of surface ozone mixing ratios for the six sea breeze events compared to days in which no sea breeze was observed. The four highest daily maximum surface O3 mixing ratios are concurrent with sea breeze events. On 2 of the 4 days (6–7 July 2010), the initial high ozone mixing ratios coupled with the impacts of the sea breeze recirculating ozone photochemical precursors and photochemically aged air masses induced local surface ozone mixing ratios that exceeded the EPA standards. The maximum 8 h running mean of surface ozone mixing ratios on 6 and 7 July 2010 was 86 and 91 ppb, respectively. On both days, a sea breeze event was observed. On 6–7 July 2010, the measurement site was influenced by a persistent anticyclone, causing clear skies, relatively calm winds (<3 m s−1) and high maximum temperatures (36° and 37°C). These conditions are favorable for O3 photochemical production and buildup. In comparison, the average maximum temperatures on days in which no sea breeze was observed was 32 ± 4°C. It should be noted that while sea breeze events can contribute to the probability of an exceedance of the current EPA standards, four out of six sea breeze events did not result in an exceedance (Figure 12). However, the 2 exceedance days were coincident with sea breeze events. These analyses show that sea breeze events must be coupled with preexisting and persistent meteorological and chemical conditions in order to exceed EPA standards in this region.

Figure 12.

Diurnal cycles of surface ozone during days in which a sea breeze was observed (colored) and days during which no sea breeze was observed (gray) during the 2010 measurement period.

[37] The persistent high O3 mixing ratios during the 7 July 2010 pollution episode are evident in the elevated layers of the atmosphere. Ozone mixing ratios in elevated layers are more conserved than in surface layers because of the lack of surface deposition and titration from fresh emissions of NO. Figure 13 shows the vertical profiles from 15 launches conducted during 2010 in Hampton, Virginia. Five of the launches occurred prior to sunrise and have depleted ozone (<30 ppb) within 500 m of the surface. Three ozonesondes were launched on days in which a sea breeze was observed (see colored data in Figure 13). Ozone mixing ratios on 7 July 2010 of greater than 70 ppb were observed in a layer 1–6 km above the surface. These mixing ratios, when compared to other O3 mixing ratios from ozonesondes launched during this campaign, are among the highest O3 mixing ratios observed. The high O3 mixing ratios aloft observed on the morning and afternoon of 7 July show the buildup of O3 through persistent favorable photochemical production conditions since high surface O3 resulting in an exceedance was observed on the day prior. High elevated O3 mixing ratios mixed down into the BL as it grew in the midmorning, in part, contributing to high afternoon O3 mixing ratios and an O3 exceedance of EPA standards. These results highlight the conditions favorable for increased chemical production of ozone are also conditions more favorable for the occurrence of sea breeze circulations.

Figure 13.

Ozone profiles measured using balloon-borne instrumentation. Ozonesondes launched on days concurrent with sea breeze events are highlighted in colored dots (red, blue, green). Ozonesondes launched on days when no sea breeze was observed are indicated by the gray dots. Only data below 10 km are shown. Times listed indicate the respective launch times and are in local time (eastern standard time).

3.5. Operational Forecasts and Observations Comparison

[38] It is of interest to see how operational forecast models perform in this complex coastal environment. Forecasts from the National Air Quality Forecasting System (NAQFS) were archived during the campaign. NAQFS uses meteorology from the 12 km NAM output. The 12Z model run from the prior day was used to compare to hourly averages from the NATIVE measurements. Overall, the forecasts were biased high by 8.0% (Table 3). The model correctly predicted the only 2 exceedance days during the campaign. However, the forecast issued a false alarm on 3 days (5.7%). If only the sea breeze days are considered when comparing forecasts and observations, the mean bias decreases to 1.6%.

Table 3. Statistical Skill Values for the NAQFS Operational Forecast of O3 and the Surface Measurements of O3 From NATIVE
StatisticOverallSea Breeze DaysNon–sea Breeze Days
Number of points12731441129
RMSE (ppb)12.011.912.0
Mean bias (ppb)
Normalized mean bias (%)
Mean absolute error (ppb)
Normalized mean absolute error (%)20.717.121.3

[39] These results do not indicate that the operational forecasts are able to better predict the processes occurring on sea breeze days. Most likely, the increase in surface ozone concentrations observed during sea breeze days is compensating for the high model bias. Figure 14 shows the difference between forecasted binned averaged hourly values and binned averaged hourly observations as measured by NATIVE during sea breeze days and non–sea breeze days. The model bias is significantly positive throughout most of the day during non–sea breeze days and peaks at 14:30 UTC (10:30 local time). As shown in Figure 14, 14:30 UTC is a highly transitional time when ozone is quite variable day to day most likely due to local source emissions, photochemistry and boundary layer growth. On days with a sea breeze, the differences between forecasts and observations are more negative compared to a non–sea breeze day except for the 12:30 and 13:30 UTC bins. This shift to more negative biases reduces the overall mean bias during sea breeze days, but the mean absolute errors for sea breeze and non–sea breeze days (9.0 and 9.2 ppb) are similar.

Figure 14.

Diurnal cycles of differences between NAQFS O3 forecasts and NATIVE O3 surface observations during sea breeze (red) and non–sea breeze days. Data were bin averaged every hour. The error bars represent the standard error of the mean for each bin.

4. Conclusion

[40] Air quality and meteorological measurements during the summers of 2009 and 2010 were conducted in Hampton, Virginia, using a combination of surface and balloon-borne in situ and remote sensing instrumentation. The objective of the experiment was to characterize the relationships between dynamic and chemical forces and surface ozone variability at this complex coastal site.

[41] Differences in 2009 and 2010 air quality, especially surface O3 mixing ratios, were observed. The differences in the mean diurnal cycles are most likely caused by differences in the critical meteorological conditions that are favorable for O3 production. In both years, higher O3 mixing ratios were observed when wind directions were from maritime source regions. Clustering events based on wind direction and time of day showed that under maritime wind directions, O3 mixing ratios remained higher than average during nighttime and midafternoon. At night over water, the O3 mixing ratios remained high (∼40 ppb) owing to the lack of surface O3 deposition and O3 titration with NO as compared to terrestrial wind directions. There was little correlation observed between morning (presunrise) O3 mixing ratios and daily maximum O3 mixing ratios, in contrast to observations in other regions (e.g., Houston). In the afternoon, O3 mixing ratios are not correlated with boundary layer height.

[42] A regression model using 3–4 variables described daily maximum surface ozone variability relatively well (R2 = 48–54%). The model identified meteorological variables driving the variability of surface ozone consistent with previous studies. The model had good agreement with observations when used in a predictive capacity for the respective years, but the integrity of the model decreased dramatically when applied to different years.

[43] During sea breeze events, surface O3 mixing ratios increased in the midafternoon, leading to higher than average midafternoon concentrations. During one sea breeze episode, O3 mixing ratios increased at a rate of 14 ppb h−1, leading to an O3 exceedance and the highest observed surface mixing ratio during the 2010 campaign. This growth rate greatly contrasts with the average growth rate of −2 ppb h−1 measured during the afternoon when no sea breeze was observed. The four highest daily maximum O3 mixing ratios occurred on days when sea breeze events were observed. The EPA standard for O3 was exceeded on two days during the 2010 measurement period (6–7 July). On both days, a sea breeze was observed.

[44] Operational forecasts of surface ozone are biased high in this region. The increased ozone observed during sea breeze events compensates for the forecasts high biases, leading to overall lower mean biases and absolute errors during sea breeze events even though the controlling processes are most likely not adequately captured in the forecast model.

[45] The negative trend in exceedance days shows the effectiveness of increased regulations on anthropogenic atmospheric pollutants, such as NO2. As the likelihood of exceedance days continue to decrease and the standards for O3 become more stringent, the impact of sea breezes on the probability of an exceedance will likely become magnified. Where as the occurrence of a sea breeze does not necessarily lead to an ozone exceedance, all ozone exceedance days observed in 2010 were coincident with sea breeze events. In a region where the numbers of exceedance days are decreasing, sea breeze events will likely become increasingly more important toward the probability of an exceedance occurring. The discrepancies shown between both the multilinear regression model output and the operational forecasts with the surface observations highlight the inability to predict high-ozone events during sea breeze events. Thus, in the Mid-Atlantic region, our understanding of the dynamics of sea breezes and the ability to forecast these phenomena will become increasingly important to forecasting air quality.


[46] This work was supported by the NASA Aura Validation Program (NNG05G062G) and a NASA Geo-CAPE grant (NNX09AF93G). The authors would like to thank Jack Fishman and Doreen Neil (NASA Langley Research Center); Anders Jensen and David Doughty (PSU); and Laurent Sauvage, Alexander Sauvage, and Adrien Quesnel (Leosphere) for their support and contributions.