Journal of Geophysical Research: Atmospheres

Evolution of surface ozone in central Italy based on observations and statistical model



[1] Hourly and daily variations of surface ozone have been analyzed in relation to radon and meteorological parameters to explore its controlling mechanisms. Measurements in central Italy cover the years 2004 and 2005, showing a relevant role of transport in the ozone concentration variability. An analysis based on back trajectories shows that the site is affected by air masses originating from the west to northeast sector in about 74% of the days, suggesting that L’Aquila could be considered a background site. The background hypothesis is also supported by the rather low values of the following ozone quantities: maximum of monthly averages (39 ppbv, July), annual median of hourly data (29 ppbv), and annual average of hourly maxima recorded daily (49 ppbv). Only six hourly data recorded ozone above 90 ppbv in 2 years but never above 100 ppbv. The regression model reproduces measured ozone with accuracy in 67% of hourly observations and 74% of daily mean data. Here the model includes information from the following meteorological parameters: temperature, relative humidity, horizontal wind speed/direction, sun radiation, and radon concentration. A tracer like radon that tracks the dynamical changes of the lower atmosphere has a significant role in the model ozone prediction improvement, especially for hourly observations and for the synoptic component. In the first case (hourly observation), inclusion of radon data improves the regression model performance by 5% (from 62 to 67%); in the last case (synoptic component), the model accuracy increases by 3% (from 78 to 81%).

1. Introduction

[2] Ground-level ozone is one of the air pollutants of most concern in Europe and in North America due to dangerous effects on human health, ecosystems, and agricultural crops and materials [European Environment Agency (EEA), 2005; National Research Council, 1991]. Ozone is produced by oxidation of anthropogenic and biogenic volatile organic compounds, carbon monoxide, and methane in presence of nitrogen oxides and sunlight. Transport from the stratosphere is another source of ozone in the free troposphere, where the O3 lifetime strongly depends on the season: it is on the order of 200 days at midlatitudes in winter and less than about 5 days in summer. In the boundary layer, where dry deposition acts as the main O3 sink [Liu et al., 1987], the ozone lifetime ranges from hours up to a few days. The increase of anthropogenic emissions of O3 precursors led to an ozone rise over the past century [Wang and Jacob, 1998], while surface ozone trends in the last few decades are still a topic of debate due to sites where a decline is reported [Vingarzan, 2004, and references therein]. Even in presence of some mitigation of the emissions of ozone precursors, the target value of O3 level to protect human health was exceeded across large parts of Europe. This was observed especially over Mediterranean Europe during summer 2004 and summer 2003, when there were a record number of exceedances [EEA, 2005; Ordònez et al., 2005]. Besides emissions and photochemistry, the ozone concentration in any site is also strongly influenced by transport and meteorology: O3 values become particularly large during summer under persistent high-pressure meteorological conditions with high values of sun radiation and temperature in presence of precursor emissions. High relative humidity and wet and rainy weather are usually associated with low ozone concentrations due to a reduction of photochemical efficiency and an increase of ozone deposition on water droplet [Lelieveld and Crutzen, 1990]. The link between ozone time series variations and meteorological parameters, as well as forecast of O3 from meteorological data using statistical methods, has been well established in several works [Tarasova and Karpetchko, 2003; Gardner and Dorling, 2000; Lu and Chang, 2005; Gerasopoulos et al., 2005; Moody et al., 1995; Lu and Chang, 2005; Heo and Kim, 2004; Thompson et al., 2001; Davis et al., 1998].

[3] The role of transport in the surface ozone variability has been studied with models using wind data or dynamical tracers like radon, which is a natural radioactive gas emitted into the atmosphere by soils at a rate which is rather uniform and close to 1 atom/cm2/s. This value can be somewhat reduced by the presence of snow cover, whereas the effect of soil mixture on radon emissions is still poorly understood [Jacob and Prather, 1990]. Ice and water emit only negligible amounts of radon, so that close to coasts the actual concentration of radon in the atmospheric surface layer can be significantly affected by horizontal advection, depending on the direction of winds. On an inland site, like the one in the present work, the most important transport mechanism affecting the concentration of radon is the turbulent vertical mixing in the boundary layer. On a global scale, the net sink of this tracer is only its radioactive decay (with a well-established e-folding lifetime of 5.517 days). The radon local lifetime in the lowest layer of the troposphere, however, is normally much shorter, since the major removal process from the surface is the turbulent vertical mixing (typical lifetime of the order of few hours), except for stable nocturnal boundary layer conditions. In this case, a stable temperature inversion tends to suppress the turbulence, so that the concentration of a radon-like tracer (with a uniform soil production flux) rapidly increases. This is because its lifetime is longer with respect to daytime or nighttime unstable conditions, with active small-scale convective mixing. For the above reasons, radon is an ideal tracer to validate vertical mixing parameterizations in numerical models or to be used as a proxy for the boundary layer vertical mixing [Pont and Fontan, 2000; Perrino et al., 2001].

[4] Atmospheric dynamics is divided by meteorologists into different scales of motion: synoptic, seasonal, and short term. Filter methods are applied to meteorological data and ozone time series to divide the scales of motion into components since processes occurring at different scales are caused by different phenomena [Flaum et al., 1996; Rao et al., 1997; Tarasova and Karpetchko, 2003; Wise and Comrie, 2005; Vukovich, 1997]. In fact, the synoptic-scale component is influenced by the dynamics of large-scale meteorological occurrences, the seasonal scale component by the variations of the solar zenith angle, and finally the short-term component is attributable to day-to-day weather variability and short-term fluctuations in precursor emissions.

[5] The Mediterranean area has been recognized with particular interest for ozone studies because of the intense solar radiation and intercontinental transport of pollutants, especially during summer [Lelieveld et al, 2002]. Recently, the highest peaks of surface ozone in Europe were reported from Italy and Spain [EEA, 2005]. Nevertheless, only few systematic measurements are reported from the eastern Mediterranean [Gerasopoulos et al., 2005; Kouvarakis et al., 2000; Kouvarakis et al., 2002] and the western Mediterranean background sites [Ribas and Penuelas, 2004]. Measurements from Italy, that is, in the center of the Mediterranean region, are reported only from short intensive campaigns and very few systematic observations [Cristofanelli et al., 2006; Kalabokas et al., 1997; Thielmann et al., 2002; Nali et al., 2002].

[6] Background ozone is the O3 fraction measured in a site that is not attributable to anthropogenic local sources. Even though it is almost impossible to find an area completely unaffected by anthropogenic emissions, measurements of ozone in sites with minimum impact of human activities are very important to have a quantification of the background ozone [Vingarzan, 2004, and references therein]. The latter is important to properly set up controlling strategies of the anthropogenic emissions and for trend studies since it may be regarded as the lower boundary of the O3 concentration. Sources of background ozone are: (1) transport from the stratosphere, (2) long-range horizontal transport from distant polluted sites, and (3) in situ production from oxidation of biogenic methane and VOC reacting with natural NOx [Environmental Protection Agency, 1993]. Levels of background ozone change from site to site: Vingarzan [2004] reports a review of the measurements around the world, in Canada and United States. In the latter, observations in 11 parks show annual median between 13 and 47 ppbv and annual maxima from 49 to 109 ppbv. Despite the relevance of the background ozone, trends are not calculated from large part of Europe [EEA, 2005]. Ozone trends in Italy are calculated only from the background site of Mt. Cimone in the northern part of Apennines, where the median of ozone is 53 ppbv, with the monthly average between 40 ppbv in winter and 68 ppbv in summer during 6 years of observations [Cristofanelli et al., 2006; Bonasoni et al., 2000].

[7] In this work we analyze 2 years of continuous measurements of ozone, radon, and meteorological parameters from a small town in central Italy. Seasonal, diurnal, and hourly ozone evolution are analyzed together with back trajectories and meteorological parameters to find out the nature of our site. The temporal pattern of the ozone time series is analyzed with filter techniques. Another goal of our study is to explore the relative contribution of the above meteorological parameters in controlling surface ozone using a regression model. Improvements of the ozone prediction including radon data in the regression model are also discussed.

[8] This paper is organized as follows: in section 2, we describe the site where observations have been taken and the geographical location impact on the measurements followed by a short description of the instruments used and the seasonal variations of ozone and meteorological parameters. Analysis of hourly ozone data and description of the regression model and its sensitivity to radon data are described in the section 3. Section 4 is similar to the previous one, but for daily mean data, looking at the model improvement by adding solar radiation data as input. Analysis of different timescales in the time series of observed and modeled data is reported in section 5. The overall discussion of the results and the conclusions are summarized in section 6.

2. Measurements

[9] Ozone is continuously measured since 2003 at the University buildings at about 3 km northwest from downtown L’Aquila (42°22′N, 13°21′E), a small town of less than 70,000 people in the central part of Italy. The town of L’Aquila is located in the Aterno River valley at about 700 m above sea level (asl) between the Gran Sasso mountain chain, the highest peak of Apennines (2912 m asl), and the Sirente mountain chain (2348 m asl) (see Figure 1). The measurement site is far away from strong anthropogenic pollutant sources due to industries: the main pollution sources are traffic and public and private energy consumption from the town of L’Aquila. The site is characterized by cold winters: the mean temperature during the measurements period was 12.5°C, the coldest one was —13.9°C, and the hottest one was 36.2°C. The monthly mean temperature of December, January, and February was between 2.1° and 2.6°C, while the hottest month was July with a mean temperature of 22.3°C. Relative humidity ranges between 51% of July and 77% of January. Prevailing surface wind directions are east-southeast (120°) and west-northwest (290°). Five-day back trajectories calculated by the Hybrid Single-Particle Lagrangian Integrated Trajectory Model [Draxler and Hess, 1998] are used to classify the origin of air masses reaching our site. The model uses meteorological data from the US National Center for Environmental Prediction and has been run for each day of the 2 years of observations (2004–2005). Figure 2a shows a rose histogram of the air mass origin. Our site is mainly impacted by air masses that originated from the sector west to north and is located upwind from the town of L’Aquila, suggesting that our observations are relatively low impacted from local anthropogenic sources. Figure 2b shows the seasonal distribution of air mass origin deduced by the 5-day back trajectories, categorized in four sectors chosen looking at the air masses origin (see Figure 2a) and the topography of the site. The sector comprised between 240° to 60° (about west to northeast) accounts on annual average for 74% of the air mass origin with higher values from January to August (on average up to 80%) but less during the fall (on average 62.5%). From the sectors 60°–150° (northeast to southeast) and 150°–240° (southeast to southwest), air comes mainly during fall for 19 and 18.5% of the days, respectively, whereas during January–August, it is for only 9 and 11%, respectively. These features suggested that during the fall months, the impact of emissions from the town of L’Aquila and southerly winds could have a role in the ozone variability.

Figure 1.

Map of Europe and of the central Italy with the Aterno River valley, between Gran Sasso chain and Sirente, where the measurement site of L’Aquila is located. Rome is about 100 km southwest from L’Aquila, and the Adriatic Sea is about 80 km east.

Figure 2.

(a) Rose histogram of air mass occurrence deduced by 5-day back trajectories. (b) Seasonal classification of air mass occurrence deduced by 5-day back trajectories categorized in four sectors: about northwest to northeast (330°–60°), about northeast to southeast (60°–150°), about southeast to southwest (150°–240°), and about southwest to northwest (240°–330°).

[10] Surface ozone is measured with an ozone photometric analyzer model 400A (Teledyne Advanced Pollution Instrumentation) that uses a 254-nm UV light signal passing through the sample cell where it is absorbed in proportion to the amount of ozone. Span and zero levels of the analyzed are checked monthly; the zero level is checked by switching a valve to allow sampling air to be scrubbed of ozone [US Environmental Protection Agency (EPA), 1979]. The cross calibration of the instrument is performed every 6 months using factory standards. A Teflon filter to remove aerosol particles from the inlet line of the analyzer is changed every 15 days. Ozone measurements are made every 6 s, and averaged data are stored on hourly base.

[11] Radon is measured with Silena model 5S instrument using a scintillation Lucas cell technique [Abbady et al., 2004]. Radon measurements are made on a 5-min base from which hourly averages are calculated. The filter in front of the inlet line is changed every 15 days. The meteorological measured parameters are temperature, relative humidity, wind speed/direction, sun radiation, and precipitation, with resolution of 5 s, from which hourly average data are calculated.

[12] The data series analyzed in this work are continuous measurements from January 2004 to December 2005; short gaps are due to the maintenance of the instruments or power supply failure. The annual cycle (2004–2005) of ozone and solar radiation is reported in Figure 3a; the monthly average maximum of ozone is observed in July (39 ppbv) even if the solar radiation is about 12 W/m2 lower than in June. The spring maximum of ozone due to stratosphere-troposphere exchange and to photochemical O3 production from precursors accumulated during the winter [Monks, 2000; Gerasopoulos et al., 2005] is not evident in our site by looking at Figure 3a. The reason is that this maximum is somehow filtered out by averaging data over 2 years. Actually, looking separately at monthly ozone averages of 2004 and 2005, spring maxima come out in March 2004 and April 2005. Monthly averages of temperature and relative humidity are reported in Figure 3b. Temperature is peaked in July, as for ozone, while relative humidity is exactly anticorrelated with temperature and ozone, showing a minimum in July. This confirms the idea that hot, sunny, and dry conditions are favorable for photochemical production of O3.

Figure 3.

(a) Annual cycle of ozone and sun radiation, (b) annual cycle of temperature and relative humidity, and (c) annual cycle of radon and wind speed.

[13] A qualitative picture of the transport influence on the ozone concentration can be deduced by comparing Figure 3a with Figure 3c, where monthly averages of wind speed and radon concentration are shown. In November, there is an abrupt increase of the wind speed from 0.6 to about 1.3 m/s, which keeps to rise also in December from 1.3 to about 1.6 m/s. This wind speed increase might represent one of the factors explaining the relatively constant ozone value from October to December. This feature could be a footprint of southerly air that brings more polluted air, hypothesis confirmed also by back trajectories (see above and Figure 2), or, in alternative, these very low values are just the background values of ozone in our site. The stability of the boundary layer is well described by the radon concentration (Figure 3c). With the exception of April, the ozone buildup in mid-late winter, spring, and summer is well correlated not only with increasing solar radiation and temperature but also with the level of stability of the boundary layer (more stagnant air conditions favor a better isolation of the local photochemically produced O3). The interruption of the ozone decline in November–December is also supported by the decrease of radon concentration during these months. With radon being an indicator of the boundary layer stability, its decrease suggests that more vertical mixing is taking place during these two months, which may favor downward transport of ozone-richer air.

[14] The presence of the mountain chain of Gran Sasso in the northeast side of the town of L’Aquila is a shield for transport of ozone and precursors from north. The rose diagram of hourly ozone data as function of the wind direction (Figure 4a) shows that north-northeast ozone advection is rare. This is something expected due to the obstacle represented by Gran Sasso for air coming from north-northeast. Figure 4b shows that ozone concentrations larger than 55 ppbv (i.e., the largest daily average value recorded in our site) are always found when air comes from east-southeast (120°) or west-northwest (290°), which are exactly the two sides where the Aterno River valley is open. These observations imply that transport of pollutants reaches L’Aquila only from those few industrial sites located 50–80 km southeast of L’Aquila (Bussi-Sulmona) or from the northeast (Rieti). The influence of the city of Rome, with high level of pollution [Kirchmayer et al., 2005; Acker et al., 2006], is directly prevented by the Sirente mountain chain, as well as the transport from more industrial and populated areas over the Adriatic Sea (Pescara-Chieti), in this case, shielded by the Gran Sasso mountain chain. It is possible, however, that some advective contributions from these two areas can take place trough northwest-southeast track.

Figure 4.

(a) Rose diagram of ozone concentration (0° is for northerly winds). (b) As in Figure 4a but for ozone mixing ratios larger than 55 ppbv.

3. Hourly Mean Ozone Variations

[15] The ozone annual cycle reaches the maximum in July as shown in the previous section. The hourly mean ozone maximum is also measured in July (92 ppbv on the 19th and 21st of 2004, both days at 18:00 hours local time, and 83 ppbv on the 24th of 2005 at 17:00 hours local time). The annual median of hourly data is 29 ppbv, and the annual average of hourly maxima recorded daily is 49 ppbv. Only six hourly data recorded ozone above 90 ppbv during the 2 years of observations and, in any case, never above 100 ppbv. The annual hourly median is comparable with the median range recorded in US background sites (13–47 ppbv), whereas the annual average of hourly maxima is just the lower boundary of what recorded in US (49–109 ppbv) [Vingarzan, 2004, and references therein]. The comparison with the background site of Mt. Cimone in the North Apennines (see above) shows that the maximum of monthly ozone averages is much lower (39 ppbv in L’Aquila compared with 69 ppbv in Mt. Cimone) [Bonasoni et al., 2000]. Using some available NO, NO2 measurements taken in our site, we estimate a low contribution of the photochemical net production of ozone. Looking at a typical summer day of 2005, when the measured rate of ozone increase is about 10 ppbv/hr between 8:00 and 12:00 in the morning, the estimated rate of net photochemical production is roughly 3 ppbv/hr, which is one third of the observed rate of increase. These observations together with the back trajectory analysis could support the hypothesis that our site is only partially impacted from local anthropogenic sources.

[16] The correlation coefficients between ozone concentration and meteorological parameters give an idea of the factors more strongly influencing the ozone variability. In Table 1, the correlation coefficients between ozone and radon, temperature, relative humidity, wind speed/direction, sun radiation, and precipitation are reported. Radon is highly negatively correlated with ozone with coefficients larger than 0.54 for all the seasons, confirming the radon ability to trace lower tropospheric vertical motions, which have in turn a big role in the variability of boundary layer ozone accumulation. Figure 5 is an example (24 September 2005) of a diurnal variation of ozone and radon which clearly shows their highly pronounced anticorrelation. The sudden ozone increase in early morning is closely correlated with an abrupt radon decrease associated to small-scale convective mixing taking place after breaking up of boundary layer thermal inversion. The rapid increase of ozone from 09:00 to 12:00 hours and the relatively constant values from 14:00 to 16:00 hours are also well anticorrelated with radon, clearly showing the strong role of vertical transport and its impact on the ozone variability. Another interesting feature is the secondary evening ozone maximum around 21:00 hours well reproduced by a secondary radon minimum. Secondary ozone maxima are reported at several locations [Salmond and McKendry, 2002] and are due to aloft O3 transported by turbulence and horizontal mixing: In our specific example, we have also observed a sudden change of wind direction at that time of the day from 100° to 300°, suggesting a local transport of air rich of ozone from the latter direction.

Figure 5.

Typical diurnal variation of ozone and radon.

Table 1. Correlation Coefficient Between Ozone and Meteorological and Physical Parametersa
Ozone and …YearWinterSpringSummerFall
  • a

    All the data reported in this table are hourly averaged.

Radon−0.47 ± 0.02−0.58 ± 0.03−0.62 ± 0.03−0.55 ± 0.03−0.54 ± 0.04
Temperature0.43 ± 0.020.49 ± 0.040.44 ± 0.040.74 ± 0.020.43 ± 0.04
Relative humidity−0.38 ± 0.02−0.41 ± 0.04−0.25 ± 0.04−0.65 ± 0.02−0.58 ± 0.03
Wind speed0.55 ± 0.020.50 ± 0.040.63 ± 0.020.60 ± 0.030.49 ± 0.04
Wind direction0.17 ± 0.020.09 ± 0.050.14 ± 0.040.34 ± 0.040.02 ± 0.05
Sun radiation0.34 ± 0.020.32 ± 0.040.32 ± 0.040.33 ± 0.040.30 ± 0.05
Precipitation0.00 ± 0.020.04 ± 0.050.02 ± 0.04−0.06 ± 0.040.04 ± 0.05

[17] On summertime, the parameter showing the highest correlation with ozone (0.74), while during other seasons is around 0.44, is temperature, as well as considering annually based data. The high summertime correlation of temperature and ozone is something we expected, as well as the negative correlation with relative humidity that is low during spring (−0.28) and high during summer (−0.65), which is the season when local ozone production is more effectively controlled by photochemistry. No correlation at all between precipitation and ozone is observed even if an anticorrelation is sometimes reported in the literature [Fischer et al., 2004]. Our results do not support the hypothesis of ozone deposition on water droplets. The indirect O3 precipitation anticorrelation via decreased sun radiation is probably smoothed out in our site since precipitations usually cover short-time intervals and because photochemical net production is normally much less important than dynamics for the short-time ozone variability (see below). Sun radiation has a quite constant correlation in all the seasons with values ranging between 0.30 during fall and 0.33 during spring-summer. The influence of sun radiation is not so evident in the hourly data and is roughly independent from the season. The correlation coefficient is higher and its seasonal changes are more pronounced when looking at daily mean data (see below), which are something we may expect since short-term ozone fluctuations are primarily triggered by dynamics. The role of photochemistry and its controlling factors (sun radiation and temperature) is more clearly depicted by daily, weekly, or monthly averaged concentrations since an estimate of the ozone lifetime in our site is about 6 hours on average.

[18] The important role of dynamics on the ozone variability that comes out from correlation analysis between radon and ozone is confirmed by the correlation between ozone and wind speed. This is positive in every season, high during spring and summer (0.63 and 0.60, respectively), and around 0.50 for the other two seasons. Looking at Table 1, it seems that wind direction does not show a big influence on the ozone concentration: during summertime, the correlation with ozone reaches 0.34, while in the other seasons, it is always less than 0.17. Wind direction, however, plays a significant role in the ozone variability at our site, as shown in Figure 4b, especially when concentrations are high. The correlation coefficient is not a good parameter to quantify the influence of wind direction on ozone variations. As shown in Figure 4b, we are looking at the potential preferred directions that may influence the ozone evolution: this does not directly imply a high correlation coefficient because ozone concentrations may span in a relatively wide range for a fixed value of wind direction (see above discussion of Figure 4b).

[19] Looking at ozone concentrations more than 55 ppbv, we found that more than 71% of the events are for wind directions between 100° and 150° and between 270° and 320°. This fraction increases to 92% when restricting the analysis to wind speed above 3 m/s (see Figure 6). Figure 6 suggests a clear impact of wind and topography on the O3 variability in our site. For winds faster than 10 m/s, high ozone concentrations (more than 55 ppbv) take place in our measurement site only when air rich of ozone and precursors is advected from the two open sides of the Aterno River valley (100–150° and 270-320°). The number of high ozone concentration events with wind speed faster than 3 m/s normalized to the total number of events (regardless from the wind speed) shows a clear seasonal cycle: more than 50% in April–May compared with 40 and 35% in June–July and August–September, respectively. This seems consistent with a more pronounced dynamical variability in spring (see Table 1) and a more efficient photochemical ozone production during summertime.

Figure 6.

Histogram of wind direction for hourly ozone mixing ratio larger than 55 ppbv and wind speed above 3 m/s (0° is for northerly winds).

[20] The linear regression model to describe the ozone variations was built using a stepwise technique. The procedure involves the identification of initial parameters to model ozone and the iterative alteration of the model by adding or removing a predictor variable to maximize the regression coefficients between model and measurements [Darlington, 1990]. Using the stepwise technique, we exclude redundant variable predictors like precipitation and sun radiation and identify temperature, relative humidity, wind speed/direction, and radon as the best set of predictors for ozone. For each season, we built two models, with and without including radon data, in order to quantify the role of this tracer in the ozone variability analysis. In Table 2, the regression coefficients with errors are reported for year, winter, spring, summer, and fall time series. Another parameter reported in Table 2 is the index of agreement d2 defined as:

equation image

where Pi and Oi are modeled and observed ozone concentrations, respectively. The regression coefficient R2 indicates the observed ozone variability that can be explained by the model. The index of agreement is a measure of the degree to which model’s predictions are error-free; it is a standardized measure to estimate the error regardless of units. The index of agreement, widely used to test and compare model performances [Willmott et al., 1985; Gardner and Dorling, 2000], scales with the magnitude of the variables, retains mean information, and does not amplify outliers. It ranges from 0 to 1, with the latter score suggesting perfect agreement. The regression coefficient between hourly observed and modeled ozone (Table 2) shows that on annual base, the model explains 62% of the ozone variations, using as predictors the temperature, wind speed/direction, and relative humidity, and 67% also including radon in the model. The indices of agreement are 0.74 and 0.78, respectively. The analysis for each season shows that during summer, the model is not very sensitive to the inclusion of radon: it explains 76% of ozone variations with radon while 75% without. During spring, the inclusion of radon has the bigger effect: the model follows the ozone observations with an agreement of 75% with respect to 67% without radon. From this analysis, we can conclude that in our site, it is possible to track the hourly variations of ozone using meteorological parameters plus radon data with an accuracy ranging between 68% during winter and 76% during summer. The fluctuations of the ozone mixing ratio on annual base could be described by the following regression formula:

equation image

where Rn is the radon concentration (pCi/l), T is the temperature (°C), RH is the relative humidity (%), WS is the wind speed (m/s), and WD is the wind direction (degrees from north). The formula without radon changes as follows:

equation image

The hourly ozone time series reported in Figure 7a shows that the regression model reproduces quite well the ozone maximum but overestimates afternoon and nighttime ozone values up to more than 20 ppbv. The effect of including radon data in the regression model is clear from the Figure 7b, where the overestimation of afternoon-nighttime ozone is removed. The major role played by radon in the prediction improvement of afternoon-nighttime ozone comes out if we separate afternoon-nighttime data (16:30 to 09:30 hours) from daytime data (09:30 to 16:30 hours). Table 3 shows that in the afternoon-nighttime model on annual base, the regression coefficient and index of agreement increase when including radon data from 0.46 to 0.58 and from 0.56 to 0.70, respectively. On the other hand, daytime ozone values are completely unaffected by radon. This different behavior may be justified due to the different relative role of photochemistry with respect to transport during nighttime with respect to the day. In the late afternoon, photochemistry starts to slow down since solar radiation decreases and the variability of surface ozone accumulation is primarily due to meteorological conditions leading to higher or lower stability of the boundary layer. Radon is an ideal tracer of the surface mixing since its large nighttime accumulation can only take place in the absence of pronounced vertical motions. In turn, a stable nighttime boundary layer favors surface ozone depletion by inhibiting mixing with the free troposphere while surface dry deposition keeps going.

Figure 7.

(a) Measured and modeled ozone using as proxy the temperature, relative humidity, and wind speed/direction. (b) Measured and modeled ozone using as proxy the temperature, relative humidity, wind speed/direction, and radon.

Table 2. Regression Coefficients Between Measured Hourly Averaged Ozone and Meteorological Parametersa
Rn, T, RH, WS, WDT, RH, WS, WDRn, T, RH, WS, WDT, RH, WS, WDRn, T, RH, WS, WDT, RH, WS, WDRn, T, RH, WS, WDT, RH, WS, WDRn, T, RH, WS, WDT, RH, WS, WD
  • a

    For each season are reported coefficients with and without radon. The parameters used in the regression analysis are the following: radon (Rn), temperature (T), relative humidity (RH), wind speed (WS), and wind direction (WD). ΔR2 is the error of the correlation coefficient, and d2 is the index of agreement (see text).

Table 3. Regression Coefficients, With Error, Between Measured Hourly Averaged Ozone and Meteorological Parameters for Afternoon-Nighttime and Daytimea
Rn, T, RH, WS, WDT, RH, WS, WDRn, T, RH, WS, WDT, RH, WS, WD
  • a

    For both situations are reported the regression coefficients with and without radon and the index of agreement d2 (see text).

R20.58 ± 0.020.46 ± 0.030.59 ± 0.020.59 ± 0.02

4. Daily Mean Ozone Variations

[21] Daily means are calculated to find out the seasonal variations of ozone and the quantities affecting it in timescales of weeks or months. The month of July seems to be the most important in our site, from the point of view of ozone analysis, because hourly and monthly maxima are reported during this month. The same is true for daily mean values, at least for 2005: In this case, the 27 July was the day with the highest mixing ratio (57 ppbv), while the minimum was recorded on the 19 October (5.2 ppbv). During 2004, the maximum was recorded in May (56 ppbv) and the minimum in January (4.2 ppbv).

[22] Repeating the same analysis described in the previous section, we find that the correlation between ozone and radon is negative during winter, spring, and fall, with correlation coefficients ranging between −0.32 and −0.38 (see Table 4), while the correlation is positive during summer (0.30). The summer positive correlation may be explained with the strong persistency of anticyclonic meteorological conditions in our site. These conditions favor strong accumulation of radon during nighttime and high photochemical ozone production during the day: looking at daily mean values, it is reasonable to expect a positive correlation during summertime [Pont and Fontan, 2000].

Table 4. Correlation Coefficient Between Ozone and Meteorological and Physical Parametersa
Ozone and …YearWinterSpringSummerFall
  • a

    All the data reported in this table are daily averaged.

Radon−0.16 ± 0.11−0.38 ± 0.22−0.36 ± 0.180.30 ± 0.19−0.32 ± 0.46
Temperature0.27 ± 0.100.25 ± 0.24−0.17 ± 0.200.75 ± 0.090.28 ± 0.23
Relative humidity−0.15 ± 0.11−0.09 ± 0.25−0.02 ± 0.20−0.66 ± 0.12−0.35 ± 0.22
Wind speed0.48 ± 0.090.41 ± 0.210.62 ± 0.130.55 ± 0.150.46 ± 0.20
Wind direction0.28 ± 0.10−0.22 ± 0.240.25 ± 0.190.43 ± 0.17−0.09 ± 0.25
Sun radiation0.50 ± 0.080.36 ± 0.220.13 ± 0.200.69 ± 0.110.13 ± 0.25
Precipitation0.02 ± 0.110.12 ± 0.250.03 ± 0.20−0.22 ± 0.200.06 ± 0.25

[23] Summertime correlation between ozone and temperature is as high as for hourly data (0.75), whereas it is much lower during winter and fall (0.25 and 0.28, respectively) and even negative during spring (−0.17). This negative correlation, coupled with the pronounced springtime correlation between ozone and wind speed (0.62), and the low correlation coefficient between ozone and sun radiation (0.13) point out to the dominant role of transport during this season with respect to photochemistry, as in the previously discussed analyses of the hourly mean data. Finally, the seasonal cycle of the sun radiation-ozone correlation is quite similar to that of temperature-ozone.

[24] Even if the stepwise method finds the sun radiation to be an additional good proxy for ozone (contrary to the hourly mean time series case), the regression model has been forced to use the same proxy included in the hourly model. This choice is made: (a) to check the role of radon, (b) to have a model using only meteorological parameters, and (c) to separate the influence of sun radiation because it mainly represents the role of photochemistry. Table 5 suggests that a significant model improvement including radon data is only present during winter and fall, both in terms of regression coefficient and index of agreement. Comparison of Table 6 coefficients with those in Table 5 allows to estimate the importance of sun radiation in the daily mean regression model. The improvement is significant for the annual time series and for spring and winter. The large increase of both regression coefficients and indices of agreement during winter (and partially in spring) is an indication that during these seasons, dynamics plays an important role in our site, so that heat advection associated to meteorological synoptic perturbations makes local temperature to be not necessarily closely correlated with sun radiation, as may be the case in summer and fall. In the latter two seasons, meteorology is typically quite (i.e., more persistent conditions of high pressure), so that no additional information is gained from sun radiation beyond that from temperature.

Table 5. Regression Coefficients Between Measured Daily Averaged Ozone and Meteorological Parametersa
Rn, T, RH, WS, WDT, RH, WS, WDRn, T, RH, WS, WDT, RH, WS, WDRn, T, RH, WS, WDT, RH, WS, WDRn, T, RH, WS, WDT, RH, WS, WDRn, T, RH, WS, WDT, RH, WS, WD
  • a

    For each season are reported coefficients with and without radon. The parameters used in the regression analysis are the following: radon (Rn), temperature (T), relative humidity (RH), wind speed (WS), and wind direction (WD). ΔR2 is the error of the correlation coefficient, and d2 is the index of agreement (see text).

Table 6. As in Table 5 But Now Including Sun Radiation (SR) Data
Rn, T, RH, WS, WD, SRT, RH, WS, WD, SRRn, T, RH, WS, WD, SRT, RH, WS, WD, SRRn, T, RH, WS, WD, SRT, RH, WS, WD, SRRn, T, RH, WS, WD, SRT, RH, WS, WD, SRRn, T, RH, WS, WD, SRT, RH, WS, WD, SR

[25] Figure 8 presents two scatterplots of measured versus modeled daily mean ozone: as expected, the inclusion of sun radiation reduces the intercept from 20 to 13, and the slope rises from 0.32 to 0.55. The inclusion of the sun radiation measurements in the model produces such a big improvement because the ozone seasonal cycle, mostly due to the sun cycle, is not removed in our case, whereas in some literature works, this is done using sine and cosine functions [Tarasova and Karpetchko, 2003].

Figure 8.

(a) Scatterplot of daily mean measured ozone and ozone modeled using meteorological parameters and radon as proxy. (b) Scatterplot of daily mean measured ozone and ozone modeled using meteorological parameters, radon, and sun radiation as proxy.

5. Synoptic, Seasonal, and Short-Term Components

[26] Meteorological and physical parameters that control the ozone variability have different timescale variations, suggesting the separation of each component to find the relative weight. Timescale separation is usually done by several filter techniques: one of them (Kolmogorov-Zurbenko) is widely used for the efficiency in a broad range of frequencies and for its nonsensitivity to gaps in the dataset [Eskridge et al., 1997; Rao et al, 1997; Tarasova and Karpetchko, 2003]. This filter uses the running average technique applied recursively to the time series of ozone and meteorological/physical parameters: the output of a previous filter step is used as input for the next step. The recursive use of the filter guarantees the timescale separation and the noise suppression [Rao et al, 1997]. Three components were identified: short term, less than 11 days; synoptic variation, between 11 days and 2 months; and finally the seasonal component [Tarasova and Karpetchko, 2003; Vukovich, 1997]. From the correlation coefficients of the ozone short-term, synoptical, and seasonal components with meteorological parameters, radon and sun radiation (all filtered as ozone to find the three timescale components), we see that synoptic and short scales are the ones showing highest correlation (see Table 7). Radon reaches correlations with ozone larger than −0.70 in the synoptic scale and −0.60 in the short scale, which are significantly higher than −0.47 for hourly nonfiltered data. The short-term component is dominated also by temperature, with significant roles of radon and wind speed. The seasonal component is dominated by wind speed, and the other components have weaker roles than nonfiltered data. Temperature accounts for only 21% of the ozone variations. A regression model, built as for nonfiltered data, reproduces the synoptical component of ozone in the 81% of the measurements, a bit less excluding radon from the model (78%). From the seasonal component model, we see an important role of radon since its inclusion in the model improves the agreement with measurements from 61 to 69%. A picture of the model measurement comparison is presented in Figure 9, where the synoptical and seasonal components for the first 8 months of 2004 are reported. The comparison for the short-term component (not shown) suggests that 73% of the measurements are well reproduced by the model (see Table 8). A test to check the seasonal influence of sun radiation on ozone is made excluding it from the model: in this case, no large differences are present in the regression coefficients of the short and synoptic components, whereas a drop from 69 to 57% of the regression coefficient is calculated in the seasonal component. This is consistent with the qualitatively picture reported in Figure 3, pointing out the role of sun radiation as ozone driver on seasonal scale.

Figure 9.

Synoptic and short-term components of measured ozone with modeled synoptic and short-term ozone.

Table 7. Correlation Coefficient Between Ozone and Meteorological and Physical Parametersa
Ozone and …ShortSynopticSeasonal
  • a

    All the data are filtered to divide the short component from the synoptic and seasonal components.

Radon−0.60 ± 0.02−0.71 ± 0.02−0.39 ± 0.03
Temperature0.69 ± 0.010.71 ± 0.020.21 ± 0.03
Relative humidity−0.46 ± 0.02−0.45 ± 0.02−0.18 ± 0.03
Wind speed0.48 ± 0.020.66 ± 0.020.51 ± 0.03
Wind direction0.02 ± 0.020.18 ± 0.030.22 ± 0.03
Sun radiation0.35 ± 0.020.34 ± 0.030.31 ± 0.03
Precipitation−0.00 ± 0.02−0.02 ± 0.030.06 ± 0.03
Table 8. Regression Coefficient, With Error, Between Ozone and Meteorological and Physical Parameters for the Three Components (Short, Synoptic, and Seasonal) With and Without Radon and the Index of Agreement d2 (See Text)
Rn, T, RH, WS, WD, SRT, RH, WS, WD, SRRn, T, RH, WS, WD, SRT, RH, WS, WD, SRRn, T, RH, WS, WD, SRT, RH, WS, WD, SR
R20.73 ± 0.010.70 ± 0.010.81 ± 0.010.78 ± 0.010.69 ± 0.020.61 ± 0.02

6. Conclusions

[27] In this paper, 2 years of continuous ozone observations, combined with measurements of meteorological and physical parameters, has been analyzed using a statistical model to study the mechanisms that control the ozone variability in our site.

[28] A dynamical analysis conducted by means of 5-day back trajectories suggests that our site is mainly impacted from air mass originated from the west to northeast sector. The observed ozone levels are rather low, suggesting that our site could be included in the central Mediterranean background sites, although some additional more in-depth analyses would be needed to definitively support this hypothesis (i.e., evaluation of O3-NOx-VOC photochemistry).

[29] We show that in our site, both horizontal advection and surface vertical mixing have the dominant role in controlling the ozone variability for all seasons and all timescales. As expected, in summertime, local photochemical production of ozone is found to be more important than in the rest of the year due to high sun radiation values and meteorological more stable conditions.

[30] A regression model reproduces hourly and daily mean ozone measurements in about 67 and 74% of the cases, respectively. Separation of different timescale components shows that, for scales up to 2 months, the regression model analysis can reproduce measured ozone in more than 80% of the cases.

[31] Inclusion of observed data for a dynamical tracer like radon in the regression model improves the forecast in all seasons and for all kinds of time averages; the improvement is more important for hourly than for daily averaged data. It is shown that the radon role is relevant in reconstructing the afternoon-nighttime hourly ozone observation in the regression model. We found that, in this case, inclusion of radon data increases the regression coefficient and index of agreement from 0.47 to 0.58 and from 0.56 to 0.70, respectively. The reason for these findings is that the concentration of (soil-emitted) radon is well anticorrelated with the level of vertical mixing in the atmospheric surface layer, which has been proved to be one of the most important controlling factors of the hourly ozone variability in our site.


[32] We thank the Center of Excellence CETEMPS for partially supporting this research. Many students of the Environmental Science and Physics courses of the University of L’Aquila have given some contributions in the early stage of instrument setup and data acquisition. We also thank three anonymous reviewers for their suggestions that helped us to substantially improve this paper.