Surface Ozone Variability and Trend over Urban and Suburban Sites in Portugal

The surface ozone time series is analyzed for the seasonal and inter-annual variations and the trend in the following three categories (a) monthly mean (b) 8 hr monthly mean and (c) daily maximum monthly mean measured at 8 (urban/suburban) sites in Portugal for the period 2000–2010. The inter-annual variation of the monthly mean surface ozone time series showed an year to year variation with the highest value in May 2009 (95 μg m) at Monte Chãos and the lowest in Jan 2002 (17 μg m) at Paio Pires. The trend analysis of (1) original surface ozone time series and of (2) deseasonalized surface ozone time series for all the three data set categories and for all the 8 sites was performed and found to be statistically significant at 6 sites for monthly mean and 8 hr monthly mean and at 5 sites for daily maximum monthly meanusing the non-parametric Mann-Kendall test. The analysis of original surface ozone time series showed a positive trend at 6 out of 8 sites, but the results were not statistically significant for most of the sites due to the presence of the annual cycle masking the actual trend values. However, the analysis of deseasonalized surface ozone time series showed a statistically significant increasing trend at 7 out of 8 sites with high Z values. The positive trends found in the deseasonalized surface ozone time series were in the range 0.44 up to 1.42 μg m year for the monthly mean surface ozone time series (7 stations), 0.33 up to 1.43 μg m year in the case of the 8 hr monthly mean surface ozone time series (7 stations) and 0.66 up to 1.55 μg m year in the daily maximum monthly mean surface ozone time series (6 stations).


INTRODUCTION
Surface ozone (O 3 ) is a highly reactive trace gas and an important greenhouse gas (Mickley et al., 2001) which contributes to global warming and climate change (Unger et al., 2006).It is also a volatile secondary photochemical air pollutant and an important photochemical oxidant.Surface O 3 affects adversely to the human health from its irritant properties and its induction of an inflammatory response in the lung.O 3 also has adverse effects on crop yields and tree growth (Felzer et al., 2007;Premuda et al., 2013).
Increase in emission of O 3 precursors such as nitrogen oxides (NO x ), volatile organic compounds (VOC's), carbon monoxide (CO) and non-methane hydrocarbons (NMHCs) (Saito et al., 2002) from traffic and industrial activities lead to increased production of surface O 3 over polluted regions (Kulkarni et al., 2010).This is of great importance since O 3 formed over source regions can then be transported over great distances affecting areas far from the source (Kulkarni et al., 2009;Kulkarni et al., 2011a, b).The build-up of surface O 3 has broad implications on atmospheric chemistry and plays a significant role in controlling the chemical lifetimes and the reaction products of many atmospheric species.During sunlight hours, photochemical oxidation of VOCs initiated by hydroxyl radicals (OH) produces organic peroxy radicals (RO 2 ), facilitating cycling of nitrogen oxide (NO) to nitrogen dioxide (NO 2 ) and formation of surface O 3 (Eqs.( 1)-( 4)) (Zhang et al., 2004;West et al., 2006). (1) As solar zenith angle (SZA) decreases, from early morning till afternoon, production rate of O 3 goes up and begins to accumulate in the atmosphere and reaches maximum concentration in the late afternoon.In the evening, with increasing SZA, O 3 concentration starts declining and, after sunset, reaches its minimum around late night/early morning.At night, the photochemical process ceases, terminating chemical NO x removal such as by the reaction of OH with NO 2 to form HNO 3 .Consequently, the large abundance of NO x near the surface level leads to the removal of O 3 (Eq.( 5)) (Zhang et al., 2004;West et al., 2006). (5) The Eqs. ( 1)-( 5) along with O 3 removal due to surface deposition (Pio et al., 2000) are primarily responsible for the diurnal variation of surface O 3 , with day time maxima and night time minima in the surface O 3 concentration.
Surface O 3 concentration in urban and suburban areas exhibits marked seasonal variability (Air Quality Expert Group-AQEG, 2009), with high concentration during the summer (due to high solar radiation and long sunlight hours) and low concentration during the winter (due to low solar radiation and short sunlight hours).However, in many rural environments the ozone seasonal peak may occur in spring, which may be due to the biogenic emissions in the rural regions leading to ozone formation.Apart from in-situ production of surface O 3 , long as well as short range transport of O 3 and of O 3 precursors, has an important impact on O 3 concentrations at both regional and local scales.On the other hand, transport of O 3 from the stratosphere to free troposphere, where mixing ratios are higher, and further into the boundary layer is another important source of surface O 3 (Jain et al., 2005).Furthermore, O 3 is formed in the free troposphere, especially in the northern hemisphere, mainly by the oxidation of methane (CH 4 ) and also of CO (West et al., 2006;Ghude et al., 2011a, b) which, under favorable conditions, gets transported into the boundary layer.Besides temporal variability and transport mechanisms, variability in the meteorological conditions contributes to large interannual variability of surface O 3 concentrations.
Surface O 3 concentration at any given location can increase or decrease over a period of time and may give rise to a trend.Therefore one of the main goals of surface O 3 concentration long term studies is the analysis of trends.The studies of seasonal and inter-annual variation as well as of trend analysis of surface O 3 concentration are essential for formulating policy decisions on air quality.
Recent studies have shown an increasing trend of surface O 3 over rural areas in Europe and over many urban and suburban sites all over the world.For example, Sicard et al. (2009) reported slight increase in annual averaged O 3 concentrations in rural areas between 1995 and 2003 over France.Brönnimann et al. (2002) found increase in the monthly mean values in Switzerland in the 1990s.AQEG (2009) reported positive trend in the annual mean daily maximum 8-hour mean surface O 3 at 48 sites out of 50 urban and roadside sites spread across the United Kingdom during the period from 1990 onwards and found that 18 of the 48 sites have statistically significant trend at 0.01(3 sites), 0.05 (7 sites) and 0.1 (8 sites) level of significance.The long term trend analysis of surface O 3 performed by Fernández-Fernández et al. (2011) for the period 2001-2007 over Iberian Peninsula, including one rural surface O 3 monitoring site (Monte Velho) in Portugal, located at the Atlantic coast, observed increasing trend in the south of the Iberian Peninsula (IB) whereas decreasing trends were observed in the northern IB.Similarly, Monteiro et al. (2012) reported negative trends over eastern and northern Iberian Peninsula and positive trends over Atlantic coast, including two surface O 3 monitoring site (Custóias and Paio Pires) in Portugal, for the period 2000-2009.
There have been various studies on the analysis of surface O 3 using different statistical techniques in different regions of Portugal, particularly over northern Portugal (Pio et al., 2000;Pereira et al., 2005;Alvim-Ferraz et al., 2006;Sousa et al., 2009a;Carvalho et al., 2010;Sousa et al., 2010).Pereira et al. (2005) observed surface O 3 exceedance of public information values and alert levels defined by air quality framework directive for Europe around Oporto region during the study period.Pio et al. (2000) carried out O 3 dry deposition measurement in the northern Portugal during 1994-1995 and observed prominent diurnal and seasonal patterns in deposition flux, dry deposition velocity and surface resistance, especially for the daytime period.Alvim-Ferraz et al. (2006) compared surface O 3 measurements from 1861-1897 with measurements from 2002-2003 period and found that these values are 147% higher than the ones obtained in the previous period 1861-1897, concluding that this increase is the result of enhanced photochemical production of O 3 due to the growth in anthropogenic emissions.Monteiro et al. (2005) developed and validated a numerical air quality operational forecasting system over Portugal during the summer 2003 and noted that still improvements are needed on the chemistry-transport model.Sousa et al. (2009b;2011) observed that occurrence of childhood asthmatic symptoms were significantly higher at sites with high surface O 3 values than at sites with low surface O 3 values.
Long term trend analysis of surface O 3 time series is performed for one rural site (Monte Velho) (Fernández-Fernández et al., 2011) and two suburban sites (Custóias and Paio Pires) (Monteiro et al., 2012).However, the same was not done for any site using deseasonalized time series which is the suitable approach for analyzingstatistically significant trends of a given environmental quantity varying periodically with time (Carslaw, 2005).Moreover this type of analysis was also never done for urban/suburban background sites used in this study, with the exception of Paio Pires.
The objectives of this work were: (1) to study seasonal and inter-annual variation of surface O 3 for the period 2000-2010 over urban and suburban sites in continental Portugal; (2) to analyze surface O 3 time series and deseasonalized surface O 3 time series for identifying possible existence of statistically significant trends over urban/suburban background sites in Portugal.

DATA AND METHODOLOGY
Hourly records of surface O 3 concentrations were obtained from 8 background sites representative of two categories: (A) urban background and suburban background close to urban centers i.e., to Lisbon metropolitan area and to Porto metropolitan area, and (B) suburban background far from urban centers, spread over continental Portugal (see Fig. 1).All of these stations belong to the Environmental Portuguese Agency (APA-Agência Portuguesa do Ambiente) network.The APA is responsible for air pollution measurements at various stations all over Portugal in different type of environment (urban, suburban and rural) and of influence (background, traffic and industrial).The categorization of both types (environment and influence) is done by APA.The air quality data measured at the stations are sent to a central database for public display.The data then undergoes a validation and quality check before being archived on the website for further use.APA measures various pollutants but primarily uses five main pollutants for the calculation of the index of air quality, which are CO, NO 2 , sulfur dioxide (SO 2 ), O 3 and aerosols (measured as PM 10 ).For further details about APA and its functioning please visit http://www.qualar.apambiente.pt/.
Table 1 provides a detailed list with information such as name of thestation, station code (used in this study), latitudelongitude-altitude coordinates, type of station (urban or suburban), temporal period and number of months of data used for trend analysis.Last column of Table 1 also shows, in parenthesis, percentage of number of months used for trend analysis with respect to the study period.Only days with more than 75% or with 18 hr of measurements (Air quality Directive 2008/50/EC) and months with more than 65% or 19 days of measurementsare used to compute monthly means for studying seasonal and inter-annual variations and trend analysis.
For simplification of analysis the aforementioned 8 sites can be broadly categorized into three regions namely, (a) Porto region: consisting of Ermesinde, VN da Telha and Leça do Balio monitoring sites, all within a radius of about 10 km, (b) Lisbon region: consisting of Beato, Alfragide and Paio Pires monitoring sites, all within a radius of about 15 km and (c) remaining sites: which include Teixugueira and Monte Chãos as depicted in Fig. 1.Lisbon is capital of Portugal and biggest metropolitan area with a population of 2,224,984 (Censes, 2011) with 0.58% growth since 2001.Lisbon region, comprising Lisbon Metropolitan area (LMA), is located in the western coast of Portugal (central part) close to the Atlantic Ocean with a complex coastline.To the north, there are Montejunto and Sintra hills reaching more than 400 m above sea level (a.s.l.).To the south, there are Arrábida hills (501 m) and the Sado estuary.The city of Lisbon is located at the northern side of Tagus river, near the Atlantic ocean.Tagus and Sado river (south of Lisbon Metropolitan area) valleys are the major orographic features (Barros et al., 2003).Lisbon and south of Lisbon has subtropical Mediterranean climate (Koppen-Geiger Climate Classification Csa) with mild winters and medium hot summers.Porto is the country's second biggest metropolitan area, which has a population of 1,816,045 (Censes, 2011) with 0.082% relativegrowth since 2001.Porto region, comprising Porto Metropolitan area (PMA), is located in the northwest of Portugal, at the cross-section of Douro River and Atlantic Ocean (Madureira et al., 2011).Porto and the north of Porto with some part of central Portugal has Mediterranean climate (Koppen-Geiger Climate Classification In this study, three different surface O 3 time series were computed such as: (a) monthly mean, (b) daytime 8 hr monthly mean and (c) daily maxima monthly mean.The importance of computing monthly mean surface O 3 values from hourly data helps in identifying seasonal variationsas well as trends and reflects the overall growth scenario (increasing or decreasing).On the other hand, the computing daily maximum monthly mean surface O 3 values focuses on short exposure at a high level (Qin et al., 2004) and the monthly mean of daytime (1000 hr-1700 hr) 8 hr monthly mean value emphases on longer exposure at a moderate level (USEPA, 1996).Moreover several studies have shown that the new 8 hr O 3 standard is more stringent than the 1 hr standard (Husar, 1996;Lefohn et al., 1998;Hogrefe et al., 2000;Velasco et al., 2000;Sather et al., 2001).Each surface O 3 time series is analyzed for seasonal and interannual variations and to compute the long term trends.
The trend analysis was done on the original surface O 3 time series and on the deseasonalized surface O 3 time series.In both cases, first the linear O 3 trend was derived from a simple linear regression analysis and secondly the statistical significance of the trend was tested with the nonparametric Mann-Kendall test.Simple linear regression is a technique in the parametric statistics that is commonly used for analyzing mean response of variable Y which changes according to the magnitude of an intervention variable X.It fits a straight line through the set of n points in such a way that makes the sum of squared residuals of the model as small as possible.The non-parametric Mann-Kendall test can be used to detect trends that are monotonic but not necessarily linear (Olofintoye and Sule, 2010) and is generally used in environmental science (AQEG 2009;De Leeuw, 2000).The null hypothesis in the Mann-Kendall test is that the data are independent and randomly ordered.The Mann-Kendall test does not require the assumption of normality, and only indicates the direction but not the magnitude of significant trends (McBean and Motiee, 2008).The Mann-Kendall test statistic S can be positive, negative or null.A very high positive value of S shows an increasing trend and a very low negative value shows a decreasing trend.However, it is necessary to compute the probability associated with S and the sample size, n, to statistically quantify the significance of the trend (Khambhammettu, 2005).The statistic test Z (which follow a normal distribution) has to be computed for a given level of significance, and in this study the 90% level of confidence (Z 0.05 = 1.68) was used.The trend is increasing if Z is positive and greater than the level of significance and decreasing if Z is negative and the absolute value is greater than the level of significance.If the absolute value of Z is less than the level of significance, there is no significant trend (Khambhammettu, 2005).

Seasonal and Inter-Annual Variation
The major component of surface O 3 variation, both inter-annual and seasonal, is its annual cycle (spring/summer maxima and winter minima), which is primarily controlled at macro level by solar insolation cycle, in-situ production and transport, which includes both long and short ranges.
The annual cycle dominates in the monthly mean, the 8 hr monthly mean and the daily maximum monthly mean surface O 3 time series values at any given location around the globe and the same was also observed at all the sites used in this study (Supporting Information SI1-Figs.1(a), 1(b), 1(c): time series plots for all the 8 sites for each time series respectively).The high concentration of surface O 3 during spring/summer is mainly due to high solar radiation and longer daylight hours, while comparatively low concentration of surface O 3 during winter is due to low solar radiation and shorter daylight hours.Similarlythe type of environment, the type of influence (background, traffic or industrial) and pollutant transport also play a vital role.
The seasonal and inter-annual variation of monthly mean, 8 hr monthly mean and daily maximum monthly mean surface O 3 time series at all the 8 sites is shown in Fig. 2. Figs.2(a)-2(c) shows the 2D contour graphs of surface O 3 time series of monthly mean, 8 hr monthly mean and daily maximum monthly mean values where Y axis represents seasonal variation and X axis represents inter-annual variation for each month observed during the study period.Monthly mean (Fig. 2(a)), 8 hr monthly mean (Fig. 2(b)) and daily maximum monthly mean (Fig. 2(c)) surface O 3 time series show a clear cycle with minima in December-January and maxima in May-August periods with some inter-annual variation.
On close observation it can be noticed that for any of the three surface O 3 time series the seasonal variation (shown along Y axis) of the monthly mean (Fig. 2(a)), of the 8 hr monthly mean (Fig. 2(b)) and of the daily maximum monthly mean (Fig. 2(c)) surface O 3 time series shows a bimodal variation with a spring maxima around April-May and a summer maxima around August with a slight deep around June at all the sites.The similar pattern of seasonal variation of the monthly mean surface ozone was observed by Ferreira et al. (2004) for the Lisbon region.
The in-depth analysis of the bimodal variation observed in the three different surface O 3 time series used in our study, highlights that the monthly mean time series shows the absolute maxima during April-May period and relative The spring maxima of surface O 3 concentration in April-May period are probably due to more than one reason.Some of the known reasons can be the enhanced photochemistry after a winter accumulation of air pollutants (Penkett and Brice, 1986;Fernández-Fernández et al., 2011), Stratospheric-Tropospheric exchange of O 3 which further intrudes into the boundary layer (Atlas et al., 2003), long and short range transport (Carvalho et al., 2010) and variability of meteorological parameters (Kulkarni et al., 2013).The summer maxima of surface O 3 concentration in July-August seems to be due to high temperatures observed in the region during July-August period along with the variability in the number and in the magnitude of forest fires occurring in and around Portugal (Carvalho et al., 2011).Further, Ferreira et al. (2004) observed that high numbers of exceedances of the information threshold (180 µg m -3 -one hour) are usually registered in July and August period.This confirms that the observed bimodal variation in the seasonal cycle of surface O 3 with thespring maxima around April-May is mainly due to the dynamics of atmospheric processes in the atmospheric boundary layer and the summer maxima in July-August is due to the photochemical ozone generation (Tarasova et al., 2007;Zvyagintsev et al., 2008).Further, the spring relative maxima and the summer absolute maxima observed in the 8 hr monthly mean and in the daily maximum surface O 3 time series is representative of day time photochemistry due to high solar radiation and temperature leading to high numbers of exceedances during the summer, whereas the spring absolute maxima and summer relative maxima observed in the monthly mean surface O 3 time series is representative of the 24 hour average with longer nighttime period during spring compared to the summer time.In addition, Table 2 gives the statistical summary of minima, maxima and means (± Standard deviation) of monthly mean, 8 hr monthly mean and daily maximum monthly mean surface O 3 time series at all the 8 sites during the study period.In Lisbon region, at Beato, Alfragide and Paio Pires stations, the observed monthly mean, 8 hr monthly mean and daily maximum monthly mean values of surface O 3 for the study period are approximately the same (Table 2).The orographic features of Lisbon metropolitan region with Tagus and Sado river valleys trap the pollution and local circulation pattern may help toa better mixing of pollutants, which needs to be further validated.In Porto region, at the Ermesinde station the observed 8 hr monthly mean and daily maximum monthly meanvalues along with standard deviations are approximately the same (59 ± 17 µg m -3 ) which leads to the conclusion that surface O 3 concentration remains more or less the same over the period of 8 h on most of the days and there are very few peak of O 3 concentrations.However, at Vila Nova da Telha and Leça do Balio stations, which are suburban sites west of Ermesinde station and of Porto metropolitan area, lower 8 hr monthly mean surface O 3 values are observed compared to the daily maximum monthly mean surface O 3 values (Table 2).At Teixugueira and Monte Chãos, the observed maxima and mean values of monthly mean, 8 hr monthly mean and of daily maximum monthly mean surface O 3 concentration are more than over other urban/suburban sites (Table 2).The analysis of wind vectors was done (Supporting Information SI2-Fig.2) using the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA40 re-analysis data products (Uppala et al., 2005) with 1.125° horizontal resolution at 925 hPa over Portugal and adjacent area.The 925 hPa pressure level of the wind vectors was chosen as it approximately represents the 750 m altitude, which is well within the boundary layer height (for most of the days of the year as well as most of the time of the day).Secondly, the various stations used in the study have different topography and are located at different altitudes with different micro wind regimes.For large horizontal transport of the pollutants (in this case ozone precursors) emitted at the surface or near the surface, the pollutants gets elevated and are typically transported downwind far from the source within the boundary layer (Carvalho et al., 2011;Kulkarni et al., 2011b;Carvalho et al., 2012).The analysis of wind vectors shows that the typical prevailing wind conditions on the south western part of the Iberian Peninsula are north to south causing

Long Term Trend in Surface Ozone Concentration
In this sub-section long-term surface O 3 trend are analyzed for both surface O 3 time series and deseasonalized surface O 3 time series of monthly mean, 8 hr monthly mean and daily maximum monthly mean values at all the 8 sites used in this study.In order to obtain deseasonalized surface O 3 time series the major component of natural surface O 3 variations Tables 3-5 present the results of both the trend analysis for monthly mean, 8 hr monthly mean and daily maximum monthly mean surface O 3 time series and of the deseasonalized surface O 3 time series for the whole period of each station, listing the Mann-Kendall test statistic 'S', variance of S 'V', significance level 'Z' at 90% level of confidence (Z 0.05 = 1.68), magnitude of Trend 'T' (µg m -3 year -1 ) and the Trend significance 'TS'.All the three regions, except two sites i.e., Alfragide and Leça do Balio, show an increasing trend in the monthly mean and daily maximum monthly mean surface O 3 time series, and all regions, except one site i.e., Alfragide, show an increasing trend in the 8 hr monthly mean surface O 3 time series.However, at most of the sites, except for Alfragide, Teixugueira and Monte Chão sites, the trends are statistically not significant for the   3) indicating that the trends are statistically significant.Further analysis of negative trend at Alfragide during the study period reveals that there are in fact two different trend regimes for all the three deseasonalized surface O 3 time series, as shown in Fig. 4 (only for deseasonalized  et al., 2013) for the construction of the highway 'CRIL' (Circular Regional Interior de Lisboa).This may have resulted in the increased emission of NO X from vehicular exhausts, affecting O 3 mixing ratios due to titration effect of local high NO x emissions.By analyzing the NO and NO 2 time series (NO x = NO + NO 2 ) two different regimes can be observed (Supporting Information SI3-Figs.3(a (red [2001-2003] and blue [2004][2005][2006][2007][2008][2009][2010]) show the linear fits on the monthly mean surface ozone concentration.
analyzer was installed in 1998, which was operational until March 2004, and it was replaced with a new analyzer in March 2004 (Personal communication by Eng.João Matos, In-charge of APA Quality Control Laboratory).This may have had a minor role, but cannot be verified as there are no overlap measurements by the instruments, as it should have occurred.This situation also suggests that it is very important to regularly inspect and accordingly update the type of categorization of the monitoring sites, specifically regarding the influence (background, traffic and industrial) of pollution on the monitoring sites.In Porto region, all the three sites i.e., Ermesinde, Vila Nova da Telha and Leça do Balio show an increasing trend in the MM deseasonalized surface O 3 time series with positive Mann-Kendall statistic S (T = 0.51 µg m -3 year -1 and S = 1700), (T = 0.29 µg m -3 year -1 and S = 702) and (T = 0.08 µg m -3 year -1 and S = 232) respectively, during the study period.Due to comparatively high S and low variance of S at Ermesinde, the Z value is 3.35 which indicates the trend is statistically significant, whereas at Vila Nova da Telha and Leça do Balio, comparatively low S and high variance of S are obtained which results in low Z values, 1.38 and 0.46 respectively, indicating that the trends are not statistically significant.Teixugueira and Monte Chãos shows an increasing trend with very high positive Mann-Kendall statistic S (T = 1.32 µg m -3 year -1 and S = 2516) and (T = 1.42 µg m -3 year -1 and S = 2316) respectively, for the study period.Due to comparatively high S and low S variance at Teixugueira and Monte Chãos, the Z values are 4.95 and 4.77 respectively, which indicate that the trends are statistically significant.
It is interesting to note that, when compared the trend results of the deseasonalized monthly mean surface O 3 time series (Table 3) with the ones obtained for the deseasonalized 8 hr monthly mean surface O 3 time series (Table 4) and for the deseasonalized daily maximum monthly mean surface O 3 time series (Table 5), the absolute values of the trend T and of significance level Z show significant changes at all the stations and the T standard error starts to increase (not presented in Tables 3-5) from monthly mean to daily maximum monthly mean values.At Beato, Teixugueira and Monte Chãos sites, the T and Z decrease from the monthly mean time series (Table 3) to the ones obtained for the 8 hr monthly meantime series (Table 4) down to the ones of the daily maximum monthly meantime series (Table 5) respectively.On the other hand, the standard errors increase from monthly mean [(± 0.15) (± 0.24) (± 0.24)] to 8 hr monthly mean [(± 0.18) (± 0.30) (± 0.25)] up to daily maximum monthly mean time series [(± 0.21) (± 0.34) (± 0.28)] respectively at Beato, Teixugueira and Monte Chãos.This shows that at Beato, Teixugueira and Monte Chãos sites there is an overall increase in the background O 3 and just not during daytime hours.At Paio Pires and Ermesinde sites, the T and Z increase from monthly mean (Table 3) to the ones obtained for the 8 hr monthly mean time series (Table 4) and for Paio Pires up to the ones of daily maximum monthly mean time series (Table 5).For Paio Pires and Ermesinde the standard errors increase from monthly mean [(± 0.30) (± 0.16)] to 8 hr monthly mean [(± 0.39) (± 0.23) up to daily maximum monthly mean [(± 0.46) (± 0.27)] respectively.This shows that at Paio Pires and Ermesinde, there is a higher increase of surface O 3 during daytime hours than the background O 3 .At Alfragide the magnitudes of T and Z increase from the ones obtained for the monthly meantime series up to the ones obtained for the 8 hr monthly mean series but decrease from the ones obtained for the monthly mean to the ones obtained for the daily maximum monthly mean time series whereas the standard errors decrease from monthly mean (± 0.24) down to 8 hr monthly mean (± 0.23) and increases from MM to daily maximum monthly mean (± 0.28).The trend T at Vila Nova da Telha and Leça do Balio are not statistically significant for all the three deseasonalized surface O 3 time series, so the change in T, Z and T standard error will not give clear conclusions.
It can be noted that PP, T and MC sites show higher T for monthly mean (Table 3), 8 hr monthly mean (Table 4) and daily maximum monthly mean (Table 5) surface O 3 time series and forthe deseasonalized surface O 3 time series compared to other 5 sites.As described in previous subsection and from Fig. 1, it is clear that Teixugueira, Paio Pires and Monte Chãos are located downwind of Porto and Lisbon metropolitan areas.Generally, sites located downwind regions of big metropolitan areas shows higher trend T due to regional impact and transport of O 3 and O 3 precursors (von Schneidemesser et al., 2014).
The European policy review (ACCENT) has concluded that there is strong evidence that background O 3 concentrations in the northern hemisphere have increased by up to 10 µg m -3 per decade over the last 20-30 years (Raes and Hjorth, 2006).The trends revealed by O 3 soundings for the middle and upper troposphere (Logan, 1999) broadly agree with those from the surface observations.Furthermore tropospheric O 3 values were compared over the Iberian Peninsula for the two periods (1979-1993 and 1997-2005) and a systematic increase in the number of months with higher tropospheric O 3 concentration has been observed during the second period with respect to the first suggesting increase in Tropospheric O 3 in the last decade (Kulkarni et al., 2011b).The increasing trend of background surface O 3 and of free tropospheric O 3 observed by various researchers (Logan, 1999;Raes and Hjorth, 2006;Kulkarni et al., 2011b) are considered to be driven by increasing emissions of manmade tropospheric O 3 precursor gases, particularly methane, VOCs, CO and NO x since pre-industrial times.Lifetime of surface O 3 is short (few hours to couple of days, depending on the location and time of observation) compared to aloft or free tropospheric O 3 (IPCC, 2001).The increasing trend of free tropospheric O 3 and of the background surface O 3 may be one of the reasons for positive trendsalso observed over Portugal, at the different sites analyzed in this study, all classified as background sites, independently of their type of environment being urban or suburban.

CONCLUSIONS
The seasonal variation of surface O 3 concentration in Portugal was characterized by bimodal variation in the annual cycle with spring and summer maxima, with a slight dip around June, and the winter minima.In the bimodal variation observed in themonthly mean surface O 3 time series, the absolute maxima occurs in the spring season and the relative maxima in the summer, while in the bimodal variation observed in the 8 hr monthly mean and in the daily maximum monthly mean surface O 3 time series, the relative maxima occurs in the spring and the absolute maxima in the summer.The observed change in the occurrence of absolute and relative maxima during the spring and summer in the monthly mean surface O 3 time series compared to 8 hr monthly mean and the daily maximum monthly mean surface O 3 time series can be attributed to: 1.The day time photochemistry, due to high solar radiation and temperature, is more intense during the summer than during the spring, leading to high numbers of exceedances and to the occurrence of absolute maxima in the summer in both the 8 hr and the daily maximum monthly mean surface O 3 time series.
2. The monthly mean surface O 3 time series is representative of 24 hour average of days with longer nighttime period during spring compared to the summer time.
All stations show large inter-annual variation in the surface O 3 concentration averaged over the study period, varying from 40 up to 62 (± 10-15) µg m -3 for monthly mean surface O 3 time series, from 57 up to 71 (± 13-20) µg m -3 for 8 hr monthly mean surface O 3 time series and from 60 up to 87 (± 14-22) µg m -3 for the daily maximum monthly mean surface O 3 time series.Suburban sites located away but downwind from Lisbon and Porto metropolitan areas, such as Teixugueira and Monte Chão, showed higher 8 hr monthly mean and daily maximum monthly mean surface O 3 concentrations, averaged over the study period, than sites (urban/suburban) located inside the two metropolitan areas.
A trend analysis showed that in general the annual cycle (summer maxima and winter minima) of surface O 3 concentration masks the actual trend values, sometimes even polarity, and their statistical significance due to low Mann-Kendall test statistic 'S' and high variance of S 'V'.This implies that the deseasonalization of original surface O 3 time series is very important before calculating trends.An increasing long term trend in the deseasonalized monthly mean, 8 hr monthly mean and daily maximum monthly mean time series of surface O 3 was observed at 7 out of 8 stations located inthe different parts of Portugal used in this study.Out of 8 stations, 6 stations (5 positive and 1 negative) show statistically significant trends in the monthly mean and 8 hr monthly mean surface O 3 time series, whereas 5 stations (4 positive and 1 negative) show statistically significant trend in the daily maximum monthly mean surface O 3 time series with high Z values.Further the suburban sites (PP, T and MC) located downwind region of big metropolitan areas (Lisbon and Porto) show higher trend values due to transport of O 3 and O 3 precursors.The two different trend regimes observed at Alfragide shows the effect of land-use changes and the influence of development activities on the monitoring sites and warrants regular inspection and updating of the classification of the measuring station, specifically regarding the influence (background, traffic and industrial) of pollution on the monitoring sites.

Fig. 1 .
Fig. 1.Map of Portugal with location of Air Quality (ozone) monitoring sites (red dots) used in this study.

Fig. 2
Fig. 2(a).Seasonal and year to year variation of monthly mean surface ozone concentration at different sites during the period 2000-2010 (2001-2010 for A and PP).

Fig. 2
Fig. 2(b).Seasonal and year to year variation of 8 hr monthly mean surface ozone concentration at different sites during the period 2000-2010 (2001-2010 for A and PP).

Fig. 2
Fig. 2(c).Seasonal and year to year variation of daily maximum monthly mean surface ozone concentration at different sites during the period 2000-2010 (2001-2010 for A and PP).

(
seasonal component) must be subtracted from each of the three surface O 3 time series (Supporting Information SI1-Figs.1(a), 1(b), 1(c)).The surface O 3 time series (Supporting Information SI1-Figs.1(a), 1(b), 1(c)) and the deseasonalized surface O 3 time series (Figs.3(a), 3(b), 3(c)), respectively of monthly mean, 8 hr monthly mean and daily maximum monthly mean surface O 3 concentrations were used to compute linear trends over study locations (all 8 sites) as was explained in Section 2.

Fig. 3
Fig. 3(a).Deseasonalized time series of monthly mean surface ozone concentration at different sites during the period 2000-2010 (2001-2010 for A and PP).Straight lines show the linear fits on the monthly mean surface ozone concentration.

Fig. 3
Fig. 3(b).Deseasonalized time series of 8 hr monthly mean surface ozone concentration at different sites during the period 2000-2010 (2001-2010 for A and PP).Straight lines show the linear fits on the 8 hr monthly mean surface ozone concentration.

Fig. 3
Fig. 3(c).Deseasonalized time series of daily maximum monthly mean surface ozone concentration at different sites during the period 2000-2010 (2001-2010 for A and PP).Straight lines show the linear fits on the daily maximum monthly mean surface ozone concentration.
) and 3(b)) in the periods (a) 2001-2003 and (b) 2004-2010, as in the case of surface O 3 at the Alfragide site (Fig. 4).The second reason may be due to the change of the surface O 3 analyzer at the Alfragide site.The first surface O 3

Table 1 .
Location of ozone monitoring sites, Lat; Long; Altitude, type, study period, No. of months data used for trend analysis (% w.r.t period).No. of months of data used for trend analysis.

Table 2 .
Minimum, Maximum and Mean surface ozone concentration of monthly mean, 8 hr monthly mean and daily maximum monthly mean values at each site.

Table 3 .
Summary of monthly mean trend (µg m -3 year -1 ) analysis and trend significance using Mann-Kendall test of ozone time series and deseasonalized ozone time series for each site.

Table 4 .
Summary of 8 hr monthly mean trend (µg m -3 year -1 ) analysis and trend significance using Mann-Kendall test of ozone time series and deseasonalized ozone time series for each site.

Table 5 .
Summary of monthly mean daily maximum trend (µg m -3 year -1 ) analysis and trend significance using Mann-Kendall test of ozone time series and deseasonalized ozone time series for each site.