Climatology of pure tropospheric profiles and column contents of ozone and carbon monoxides using MOZAIC in the mid-northern latitudes. (24° N to 50° N) from 1994 to 2009.

. The objective of this paper is to deliver the most accurate ozone (O 3 ) and carbon monoxide (CO) climatology for the pure troposphere only, i.e. exclusively from the ground to the dynamical tropopause on an individual proﬁle basis. The results (proﬁles and columns) are derived solely from the M easurements of OZ one and water vapour by in-service AI rbus air C raft programme (MOZAIC) over 15 years (1994–2009). The study, focused on the northern mid-latitudes [24–50 ◦ N] and [119 ◦ W–140 O 3 and CO tropospheric trends.

subsidence (Guttikunda et al., 2005) or boundary-layer venting (Agusti-Panareda et al., 2005;Auvray et al., 2005) and long-range transport (Cooper et al., 2010). In addition to O 3 , carbon monoxide (CO) is also involved in tropospheric photochemical processes: O 3 production takes place when CO and hydrocarbons are photo-oxidized in the presence of nitrogen oxides (NO x ). CO is a by-product of combustion from the BL and an excellent tropospheric air-mass tracer due to its rather long lifetime of ∼ 2 months on average (Yurganov et al., 2004).
Tropospheric O 3 distribution analysis started in the 1960s with soundings that were sparse in space and time (3-12 per month) over about 40 northern hemispheric sites (Logan, 1985(Logan, , 1994(Logan, , 1999. Fishman and Larsen (1990) later began tropospheric O 3 retrieval by remote sensing satellites and performed climatology analysis. From satellites, the O 3 / CO correlations were investigated recently to provide the strength of O 3 photochemical production and to show the continental outflow regions in the middle free troposphere (Zhang et al., 2006;Voulgarakis et al., 2011). Nevertheless, satellite observations still cannot replace in situ measurements because of their need for permanent calibration, their time dependence, low vertical resolution, cloud screening, etc. Since August 1994, the MOZAIC (Measurements of OZone and water vapour by in-service AIrbus airCraft; Marenco et al., 1998) instruments on board commercial aircraft have sampled the troposphere with high vertical resolution over about 50 airports and IAGOS (In-service Aircraft for Global Observing System) is the current ongoing programme (see the MOZAIC/IAGOS website 1 ).
The purpose and novelty of the present study is to produce, from the MOZAIC measurements, the first pure tropospheric climatology of O 3 and CO based on fully defined individual tropospheric profiles. The new methodology aims to improve the previous MOZAIC O 3 tropospheric climatology presented by Zbinden et al. (2006), which was restricted by the permanent 12 km limit of MOZAIC aircraft during ascent or descent. Here, all profiles are defined individually from the surface to the dynamical tropopause and include all sampled stratospheric intrusions. Furthermore, the study, based on such new profiles and their associated columns, encompasses a larger range of longitudes from Los Angeles to Japan in the northern mid-latitudes [24-50 • N].
The paper first briefly describes the MOZAIC data (Sect. 2) and explains the methodology (Sect. 3). The climatological results are presented in Sect. 4 for the monthly averaged tropospheric columns, the seasonally averaged tropospheric profiles and the boundary-layer, mid-and uppertropospheric partial tropospheric columns. To further highlight the usefulness of such a climatology, we compare the O 3 and CO tropospheric seasonal cycles from our analysis with those derived from spaceborne measurements at two 1 http://www.iagos.fr/ or via http://www.pole-ether.fr MOZAIC sites in Europe and Asia. Section 5 concludes our analysis.

MOZAIC data
The MOZAIC programme has collected numerous O 3 observations since 1994 by using instruments on board five commercial aircraft  throughout the troposphere and lower stratosphere. The O 3 is measured using the dual-beam UV absorption principle (Model 49-103 from Thermo Environmental Instruments, USA), with an accuracy estimated at ±[2 ppbv + 2 %] and a 4 s time response, i.e. < 50 m vertical resolution . Measurement quality control procedures have remained unchanged to ensure that long-term series are free of instrumental artefacts since the beginning of the programme. Instruments are laboratory calibrated before and after a flight period of about 6-12 months. The infrared CO analyser (Model 48CTL from Thermo Environmental Instruments, USA) included in the MOZAIC programme since 2001 measured CO with a ±5 ppbv (±5 %) accuracy for a 30 s response time (< 300 m vertical resolution) (Nédélec et al., 2003).
To characterize the vertical distribution over the troposphere, we selected the ascents and descents from the 4 s fullresolution data between August 1994 and March 2009. The results, focused on the northern mid-latitudes [24-50 • N] and on longitudes from Los Angeles to Japan [119 • W-140 • E], are based on 11 sites among those most regularly visited by the MOZAIC aircraft (see details in Table 1). To improve the sampling frequency of a few sites and to avoid wide data gaps in the time series, we have created clusters including data from neighbouring airports with the same seasonal cycles and monthly mean concentrations. For example, Germany is the cluster of Frankfurt and Munich, the most visited site (16 041 profiles). Selecting only Frankfurt would have left data gaps of two months in 2002 and six months in 2005. By adding Munich airport, which is close (500 km) to Frankfurt even though at higher surface altitude (500 m), we obtain continuous time series relevant for climatological studies. Japan is the cluster of Tokyo, Nagoya and Osaka airports on the south-eastern coast (≤ 500 km distance). Houston and Dallas airports form the USsouth cluster (250 km distance). In total, > 40 000 profiles are compiled here (Table 1), i.e. more than twice the number of the profiles used in a previous study .

Methodology
Our objective is to deliver a monthly mean pure tropospheric climatology of profiles and columns for O 3 and CO based on the ascent or descent phase of MOZAIC flights, strictly from the surface to the altitude of the dynamical tropopause z DT as defined by Hoskins et al. (1985) Zbinden et al., 2006). The dynamical tropopause criterion Atmos. Chem. Phys., 13, 12363-12388, 2013 www.atmos-chem-phys.net/13/12363/2013/ Table 1. Geographical context of the study over four continents (column 1) with the MOZAIC site labelling (column 2), the related airport or airports for a cluster (column 3), the airport geographic coordinates (column 4), the number of associated MOZAIC profiles (Nb P, column 5), the airport elevation (column 6) and total number of MOZAIC profiles included in this study (bottom line is more adapted than the lapse rate to capturing the tropospheric ozone trends (on sites where statistics are significant) and to distinguish the contribution of the stratospheric exchanges from the strict troposphere in further studies. Additionally, the dynamical tropopause has been already used in satellite/in situ comparison (Clerbaux et al., 2008;Hegglin et al. 2008;de Laat et al., 2009;Bak et al., 2013). Furthermore, some technical reasons reinforce the choice of using a dynamical tropopause instead of the lapse rate criterion as explained in Sect. 3.1. However, the troposphere may not be completely sampled by MOZAIC aircraft during individual ascent or descent due to a permanent ≈ 12 km limitation. The tropospheric layer frequently unvisited by MOZAIC was ignored or partially estimated in the previous study by Zbinden et al. (2006). For example, over Germany, the thickness of the tropospheric layer unvisited by MOZAIC is 0.8 km on average but might exceed 3.8 km over Japan in August. In this section, the methodology for deriving the pure tropospheric profiles and columns is explained, followed by the ozonesonde validation of these new products.

Pure tropospheric results assessment
At a specific site, a MOZAIC profile, MP(X, z, t), is defined for a molecule X, such as O 3 and CO at a given time, t, between z 0 and z top , i.e. the surface altitude and the highest altitude of the ascent or descent phase of the flight, with a 50 m resolution in z. The pure tropospheric profile, PTP(X, z, t), should simply result from the MP(X, z, t) without the stratospheric air above z DT at time t. To deliver consistent results between profiles, columns and satellite results, the O 3 volume mixing ratio at a given altitude z and time t is converted into a partial column of 50 m height resolution, expressed in Dobson units (DU) (see Zbinden et al., 2006, Appendix A) using its related measured temperature and pressure. Similarly, the CO volume mixing ratio is converted into molecules cm −2 . To set up z DT at time t, the potential vorticity pressure of 2 PVU is used, referring to the study by Thouret et al. (2006) with MOZAIC data when selecting only the cruise part of flights to document the upper troposphere-lower stratosphere. The potential vorticity pressures provided by the operational European Centre for Medium-Range Weather Forecast (ECMWF) analyses (T213) are interpolated for the specific aircraft positions and available with a 150 m vertical resolution on the MOZAIC database Zbinden et al., 2006). The methodology to assess the pure tropospheric profiles, explained just below, is illustrated in Fig. 1 with three typical cases selected over Germany ((a) 8 December 1994, (b) 10 May 2000) and Japan ((c) 16 August 1995). When z DT < z top , and only in this case, MOZAIC samples the entire troposphere (Fig. 1a). In all other cases, a tropospheric layer remains undefined between z top and z DT , , as shown in Fig. 1b and c. If the lapse rate criterion (WMO, 1957) was selected instead of a dynamical criterion, the thermal tropopause could have been fixed only at an altitude (z lr ) below z top -1 km. The different z top , z s and z DT are shown in Fig. 1 and additionally z lr in Fig. 1a. Consequently, using z lr instead of z DT , more profiles would be turned into uncompleted tropospheric profiles without any perspective to be completed. The particular MOZAIC vertical sampling leads to TP(X, z, t), the tropospheric profiles up to z DT or at least z top at time t, from which a preliminary climatology is calculated on a seasonal (s) basis, TP(X, z, s).
In Zbinden et al. (2006), we estimated at time t, using TP(X, z, s) above z top . Nevertheless, TP(X, z, s) is still strictly limited to z s , the ascent-or descent-phase maximum altitude within the season s, and is always less than 12 km. Consequently, is fully evaluated only when z DT < z s , hereafter noted s (Fig. 1b).
In this study, at time t, when z DT > z s (Fig. 1c), we additionally evaluated f between z s and z DT by using Mfit(X, z f , s), the best-fit line from MOZAIC data using a linear regression on TP(X, z, s), from 5 to 11 km for O 3 and from 8 to 11 km for CO. These limits in altitude were chosen to be as far as possible from the polluted BL but not too high to maintain significant sampling and, by the end, to be representative of the seasonal tropospheric amounts in the upper-tropospheric layers. For CO, similarly, the limits were fixed at higher altitudes but within a narrow layer because CO does not decrease linearly with altitude except above 7-8 km.
Our PTP(X, z, t) derivation can be summarized by the following three typical cases, where z = [z 0 , z DT ] at time t, PTP(X, z, t) = TP(X, z , t) + TP(X, z s , s) and using the equation Fig. 1a; (2) -if z top < z DT < z s , in case ( Table 2 provides the percentage of profiles corresponding to the 3 cases encountered at the 11 sites. We found more than 50 % of the profiles of Uaemi and USsouth belong to case (c), while less than 8 % for Europe. Note that the database encompasses domestic flights which connect two close airports, such as Abu Dhabi and Dubai (∼ 150 km apart) or Dallas and Houston (∼ 400 km apart). Consequently, cruise level is never reached and the flight is limited to a short ascent and short descent. MOZAIC data over Los Angeles, Eastmed, Atmos. Chem. Phys., 13, 12363-12388, 2013 www.atmos-chem-phys.net/13/12363/2013/ Beijing and Vienna do not include domestic flights, while they represent less than 1 % for Germany, USeast, USlake, Japan and Paris, 29 % of the MOZAIC traffic for USsouth, and 39 % for Uaemi, which partly explains Table 2. We did not discarded these profiles because they were documenting the highly variable BL and areas where no regular ozonesondes or carbon monoxide measurements exist. From PTC(X, t), we calculated the monthly times series (not shown) and finally the monthly averaged PTC m (X, t) as shown in Sect. 4.1. The z DT by month or season is not introduced in any of the PTC m (X, t) or PTP s (X, z, t) calculations; it is only provided in the figures as a guideline. Consequently, the delivered climatology results from profile and column tropospheric contents on an individual z DT basis, the cornerstone of this study. Figure 2a shows an example of the monthly averaged MOZAIC O 3 profiles, MP(O 3 , z, m), and the monthly averaged pure tropospheric profiles, PTP(O 3 , z, m), over Japan for September to November and February, when the monthly averaged z DT varies from 9.4 to 13.4 km. This example highlights that the profiles MP(O 3 , z, m) and PTP(O 3 , z, m) are identical until an altitude, z Ld , where they deeply diverge, located below the monthly averaged z DT . They show, from the surface to 1 km, a permanent strong positive vertical gradient and, above 1 km and up to z Ld , a sustainable negative vertical gradient. Above z Ld , MP(O 3 , z, m) returns to a positive vertical gradient as opposed to PTP(O 3 , z, m). Thus, z Ld illustrates the impact of depth penetration of the tropopause and stratospheric air contamination on a monthly basis; this altitude will be used later in Sect. 4.3.1.
Furthermore, the integral along the vertical of PTP(X, z, t) at time t provides a pure tropospheric column, PTC(X, t) for a molecule X, such as O 3 and CO, expressed in DU and in molecules cm −2 , respectively, with Moreover, the pure tropospheric columns can be decomposed into three partial columns over the boundarylayer (BLC), mid-troposphere (MTC) and upper troposphere columns (UTC). In this study, we have replaced the constant ceiling altitude of MTC (8 km as in Zbinden et al., 2006) by a variable altitude, z Ld , defined as the lowest altitude where MP(O 3 , z, m) diverges from PTP(O 3 , z, m). An example of the z Ld location over Japan shows variations from 6.0 km in February to 9.0 km in October (Fig. 2a). Thus the partial columns BLC(X, t), MTC(X, t) and UTC(X, t) are integrated over 0-2 km, 2 km −z Ld and z Ld −z DT , respectively.
The climatologies are given by month (m) for columns (Sect. 4.1), by season (s) for profiles (Sect. 4.2) and by month for the three partial columns considering the new monthly varying z Ld (Sect. 4.3).

Pure Tropospheric O 3 validation
This subsection aims to validate the use of Mfit(X, z f , s) and ultimately our PTP(O 3 , z, t) and PTC(O 3 , t), with composite profiles combining TP(O 3 , z, t) from MOZAIC and an external in situ data set. The latter data came from the World Ozone and Ultraviolet Radiation Data Centre ozonesondes network (WOUDC hereafter) available for neighbouring areas: Wallops Island for USeast (936 sondes from 22 August 1994 to 27 March 2009 with 640 MOZAIC coincident profiles) Hohenpeissenberg for Germany (1823 sondes from 1 August 1994 to 30 March 2009 with 5127 MOZAIC coincident profiles) and Tateno for Japan (798 sondes from 10 August 1994 to 26 March 2009 with 402 MOZAIC coincident profiles). The effective height resolution of the vertical profile of an ozonesonde is 100-150 m, and the bias, the precision and the accuracy differ with ozonesonde types: electrochemical concentration cell (ECC sondes, Wallops Island), Brewer-Mast (BM sondes, Hohenpeissenberg) and carbon iodine (KC sondes, Tateno), as discussed in details by Smit and Team ASOPOS (2013). They indicate, between the surface and 15 km, that the bias varies from 0 to −7%, the precision from 3 to 10 % and the accuracy from 4 to 13 %. Thus, ozonesonde quality results depend on instrument type, launch conditions and altitude, while MOZAIC does not.
The data processing of WOUDC and MOZAIC is identical. However, to sample similar meteorological situations and improve accuracy, we selected only coincident profiles on both data sets (i.e. within a 24 h interval, denoted t ). Consequently, the sampling frequency is reduced. The coincident results by month are denoted m . Also, we assumed that z DT at time t was valid for both coincident data sets. Figure 2b shows the monthly averaged MOZAIC profiles, MP(O 3 , z, m ), and WOUDC profiles, WP(O 3 , z, m ), over Japan for the four months as provided in Fig. 2a. The best agreement between these monthly averaged profiles is found when z DT is greater than 12 km on a monthly average. This is an important result because the troposphere is not fully sampled with MOZAIC in such high z DT cases. After discarding the stratospheric air above z DT at time t , the monthly averaged pure tropospheric profiles were derived from MOZAIC and WOUDC as MPTP(O 3 , z, m ) and WPTP(O 3 , z, m ), respectively (Fig. 2c). Note that the grey-shaded rectangles of Fig. 2c highlight the layer unvisited by MOZAIC within the month and period and thus the greatest impact of Mfit on the monthly averaged profile climatology. Despite z DT uncertainties and/or co-location errors at time t , it is interesting to note that, as in Zbinden et al. (2006), both coincident tropospheric climatologies exhibit the same, almost straight, negative O 3 vertical gradient above 3 km (also clearly observed in summer over USeast but not shown).
Then, to validate the estimation of on the full MOZAIC data set, we derived a composite profile called MOZAIC-WOUDC pure tropospheric profiles, MWPTP(O 3 , z, t), by adding WPTP(O 3 , z , m ) to TP(O 3 , z , t) over with z = www.atmos-chem-phys.net/13/12363/2013/ Atmos. Chem. Phys., 13, 12363-12388, 2013 [z 0 , z DT ], z = [z 0 , z top ] and z = [z top , z DT ]. Only two cases needed to be considered: and Eq. (5) is used when z DT < z top , while Eq. (6) is used when z DT > z top . Before providing such a composite result, we checked the consistency of the MOZAIC and WOUDC coincident data sets between 2 and 8 km, MPTP(O 3 , z, m ) and WPTP(O 3 , z, m ). We selected the three most documented and distant sites (USeast, Germany, Japan). The altitude limitation was necessary to avoid the highly variable BL and to take into account the layers with the best MOZAIC sampling rate below z DT . We found that the correlation coefficient is r > 0.9 (Fig. 3) at all the sites. Moreover, when integrating the two MOZAIC and WOUDC data sets, the differences were −0.01 DU (−7 %), −0.003 DU (−2 %) and 0.008 DU (0.9 %) on average for USeast, Germany and Japan, respectively. Over Germany the high and regular sampling frequency of both data sets and the small z DT variability contributes to the best quality of results. Furthermore, a WOUDC O 3 excess occurred over USeast regardless of month, over Japan during May-September, and over Germany during all months except March (Table 3). The German range is 0-1 DU or 0-5 % and extreme values are −3.1 DU (14 %) in August over USeast and 1.7 DU (9 %) in February over Japan. Therefore, we consider that our estimate can now be validated by comparing the PTP(O 3 , z, m) with the composite result MWPTP(O 3 , z, m).
Finally, using all MOZAIC data sets, we validated the methodology by checking the consistency of TC (  U S e a s t s = 1 . 1 3 4 i = -0 . 0 0 8 r = 0 . 9 4 9 G e r m a n y s = 0 , 8 9 2 i = 0 , 0 1 9 r = 0 , 9 8 5 J a p a n s = 0 , 9 4 9 i = 0 , 0 0 7 r = 0 , 9 1 4 seasonal cycles for USeast, Germany and Japan with, in addition and as a guideline, the visualization of the MOZAIC unvisited tropospheric layer . is located between the altitude given by the solid green line, the monthly average of z DT in all cases, and by the dotted green line, the monthly average of z DT if z DT < z top and z top if z DT > z top . The three sites clearly show the same PTC(O 3 , m) and MWPTC(O 3 , m) seasonal cycles. The bias between MWPTC(O 3 , m) and PTC(O 3 , m) is less than 2 DU (6 %). Therefore, as the PTC(O 3 , m) and MWPTC(O 3 , m) correlation coefficient is r > 0.96, we conclude that the method is successfully validated for O 3 . The linear Mfit s , derived from MOZAIC for O 3 between 5 and 11 km, is particularly suitable for filling f . Consequently, PTP(O 3 , z, t) and PTC(O 3 , t) may be derived without the use of data external to the MOZAIC data set.
To summarize, in order to provide the pure tropospheric profiles PTP(O 3 , z, t) when z top < z s < z DT , we updated the methodology presented in Zbinden et al. (2006) by adding Mfit s to TP(O 3 , z, t). The methodology was validated over three different sites, each located on a different continent, by using WOUDC coincident data sounding. We concluded that the seasonal cycles of the composite MOZAIC-WOUDC tropospheric columns were positively biased, always by less than 2 DU (6 %), compared to the MOZAIC pure tropospheric ozone column. This result allows to use this methodology on sites not documented by sondes. PTP(O 3 , z, t) and PTC(O 3 , t) at time t are finally estimated without any data external to the MOZAIC data set, which is a major advantage. Additionally, to obtain purely tropospheric columns for CO, a similar methodology is applied as 90 % of the column amount resides well below 12 km, with a maximum weight in the lowermost troposphere. Hereafter, in the text and the figures that follow, we will use MP s (X), TP s (X) or PTP s (X) for averaged profiles at season s and TC m (X) or PTC m (X) for averaged columns at month m and for a given molecule X, O 3 and CO.

Results
In this section, the methodology is applied to the 11 MOZAIC sites to derive three types of climatological products: (1) the monthly averaged pure tropospheric columns PTC m (X); (2) the seasonally averaged pure tropospheric profiles PTP s (X); and (3) the monthly averaged partial columns, BLC m (X), MTC m (X) and UTC m (X). Finally, PTC m (X) is compared to satellite results.

Pure tropospheric column seasonal cycles
The PTC m (O 3 ) and PTC m (CO) cycles are given in Figs. 5 and 6, respectively. In Fig. 5, at all sites, the TC m (O 3 ) cycles, the flat z top and the z DT are additionally provided. Over all months in Europe, the value of is < 1.5 km. In contrast, is up to 3 km in summer at the other sites, while in USsouth and Uaemi it is between 1 and 4 km over all months due to intense domestic traffic. The numbers of monthly O 3 profiles (1994( -2009( ) and CO profiles (2002( -2009 are given in Figs. 5 and 6, respectively. The best sampling rate is obviously over Germany. USeast, Paris, Vienna and Japan, still regularly visited, are sampled more than twice a week, which corresponds to the best ozonesonde sampling rate. Irregular visits over the 15 years lead to the lowest MOZAIC sampling frequency over Uaemi, Los Angeles and Eastmed. To further characterize the representativeness of sites, the figures also include statistics on inter-annual variability (IAV) with a box-and-whisker plot for the quartiles Q25, Q50 and Q75. No box means that only one month has been documented over the period.

Ozone seasonal cycle
Considering all the sites, PTC m (O 3 ) varies from a European minimum of 23.7 DU in December to a Middle East maximum of 43.2 DU in July (Fig. 5). Below, we detail regional characteristics.
-The European PTC m (O 3 ) cycles exhibit homogeneous patterns with a small summer maximum and a weak amplitude (i.e. from peak to peak here and thereafter) associated with a positive west-east gradient. Paris and Germany behave similarly, and PTC m (O 3 ) are within 24.3 DU in winter and 35.6 DU in summer. Vienna differs slightly by reaching a summer maximum of 38.4 DU, probably due to its continental location and polluted air masses coming from the western part of Europe or the Po Basin (Baumann et al., 2001). As the z DT variability is the weakest among the European sites (but also among all sites), the results obviously highlight the impact of photochemistry due to local or remote emissions of O 3 precursors and long-range transport.
-The Asian PTC m (O 3 ) cycles, i.e. Beijing and Japan, vary from 25-26 DU to 40-41 DU, with a strong similarity from December to May. We point out a strong regional contrast in June, when Beijing reaches a maximum and Japan declines sharply due to incoming O 3poor maritime air during the summer monsoon, consistent with what Logan (1985Logan ( , 1999 reported from sonde analysis. In addition, z DT over Japan is 2.5-5 km higher than over Beijing from July to September. Interestingly, of all the sites studied, Japan has the lowest amounts of summer O 3 due to the impact of the monsoon. Thus, over Beijing, the higher CO compared to Japan (see Sect. 4.1.2), with probably higher NO x (Lamsal et al., 2011), is favourable to higher O 3 in July in spite of the summer monsoon. Note that Wang et al. (2012) reported far greater O 3 tropospheric columns, up to 38 DU in winter and 70 DU in summer, over Beijing using sondes launched at 14:00 local time (LT) from 2002 to 2010 and taking the 2 • C km −1 lapse rate criterion for the tropopause. Firstly, their yearly averaged 46 DU, given for a 0-9 km partial column with a 4-5 % yr −1 increase, is far greater than our monthly summer maximum. Secondly, their monthly tropopause is located between 9 km (winter) and > 14 km (summer), 2 km above ours. Thirdly, their 140 to 600 ppbv monthly averaged O 3 volume mixing ratios in summer, between 9 km and the tropopause, far exceeds the 180 ppbv MOZAIC maximum at the same altitudes, suggesting a possible stratospheric air contamination. These combined factors obviously lead to much higher tropospheric columns than the pure tropospheric columns presented in our analysis.
-The North American PTC m (O 3 ) cycles show different patterns, from bimodal over the south (Los Angeles and USsouth), slightly bimodal over USlake, to perfectly unimodal over USeast. The z DT variations probably contribute to the northern US cycle differences. The USeast cycle is close to that found in Europe despite a larger peak-to-peak range (24-40 DU (December-June)). The USlake cycle exhibits two maxima, in May and July (38 DU), with one local minimum in June (37 DU), surprisingly also detected at southern US sites. In the southern US, Los Angeles, rather poorly and irregularly monthly sampled (< 42 monthly profiles, March sampled only in 2005 and November only in 2004), shows two maxima (40 DU in May, 37 DU in July-August). The spring maximum, the biggest over the US, is related to the pollution transported from Asia (Jaffe et al., 2003(Jaffe et al., , 2007Parrish et al., 2004;Cooper et al., 2006Cooper et al., , 2010Brown-Steiner and Hess, 2011). In summer, the secondary peak appears rather small, despite a high z DT , likely due to the strong influence of subtropical Pacific air masses, as Oltmans et al. (2008) Cooper et al. (2006) and Li et al. (2005).
cycles show a high June-July maximum: 39.7 DU (Uaemi) and 43.2 DU (Eastmed). The Eastmed maximum is even larger than the Beijing maximum. Both appear to be in agreement with the summer extremes shown on the OMI/MLS 2 climatology produced by Ziemke et al. (2011). The Uaemi cycle is the flattest (11.9 DU amplitude) due to an extremely high and steady z DT (> 14 km, except in December) and thus is greater than 33 DU from January to October. In contrast, the Eastmed cycle amplitude is 16.7 DU, the highest among all sites, as-sociated with a 5 km z DT amplitude. The difference between TC m (O 3 ) and PTC m (O 3 ) is less than 2 DU in spring compared to 4 DU in summer, where in the latter case the tropospheric column height has a significant contribution. In May, the PTC m (O 3 ) is more than 5 DU higher than Germany and is even higher than Beijing. These findings suggest favourable photochemical conditions allowing this local O 3 production, as detailed further in Sect. 4.2.3 and Fig. 9. In May, z DT over Uaemi is 15 km (11 km This quasi-hemispheric overview of the PTC m (O 3 ) seasonal cycle shows an overall minimum in Europe, a summer maximum over the northeastern US, and a spring-summer maximum that is extreme in the Middle East and strong in Asia. The Asian summer monsoon results in an abrupt decline in June over Japan, unlike over Beijing, where pollutants such as CO, and probably NO x , are exceptionally high, leading to a sharp geographical contrast in O 3 . Furthermore, it is noteworthy that in summer, Japan exhibits the lowest pure tropospheric O 3 amounts of all the sites studied. The Asian-south-western US connection is highlighted in spring as Japan and Los Angeles PTC m (O 3 ) maximize. Over Uaemi, a high and rather steady z DT strongly impacts the O 3 seasonal cycle, as seen in late winter and early spring, in contrast with Europe, where photochemistry, local emissions and long-range transport predominate. The European cycle patterns, very similar to those of the northern US, confirm a common source of variance. The low German IQR is related to the higher sampling quality and, quite likely, the low z DT seasonal variability.
-The North American PTC m (CO) cycles show an April maximum, except for USeast (March), and a September minimum, except for USsouth (July). Over all these sites, sharp May-June CO depletion highlights the powerful OH cleansing efficiency regulated by NO x (Lamsal et al., 2010). The northern US cycles differ by less than 0.1 × 10 18 molecules cm −2 and exhibit patterns more comparable to Europe than to any other sites due to the CO lifetime and strong connection through westerly winds. We found a July secondary peak over the USeast, associated with a small IAV, contrary to USlake. The IAV results from only 5 yr (2002)(2003)(2004)(2005)(2006)   The extension of Alaskan fires down to Texas on 18 July 2004 was pointed out by Morris et al. (2006) in a MOZAIC case study. These fires from Alaska and Yukon territories likely also extended to the west coast of the US. The profiles of the CO 2004 anomaly in July at USeast and USlake are given in Fig. 7. As a result, at USlake, there is a 2.5-10 km positive anomaly up to 0.8 × 10 15 molecules cm −2 and a negative anomaly at USeast extending 2-5.5 km with a maximum at 3 km up to −1.6 × 10 15 molecules cm −2 . Thus, this suggests that USeast was more on the CO plume pathway (or branch of pathway) than USlake. This finding and difference appear to be in agreement with what MO-PITT has captured 7 . Fig. 7 also shows how negative and positive anomalies on profiles may be counterbalanced, leading to a lack of anomaly on PTC m (CO) in 2004. Thus, the IAV of the PTC m (CO) seasonal cycle cannot reveal such details. The two northern US sites show a June-October excess of CO (up to 0.3 × 10 18 molecules cm −2 ) compared to the southern US sites. USsouth shows a June-October-wide flat minimum, probably related to the pollution lifted under the influence of a semi-permanent anticyclone over Texas and the active summer convection that helps in lifting the surface pollution to the mid-troposphere (Li et al., 2005;Liu et al., 2006). The Los Angeles amplitude is 50 % greater than in the other US cycles. The winter-spring maximum is almost equivalent to that of the northern US sites, while the deep, narrow, low minimum in late summer probably suggests the impact of the depleted polluted air from Asia or clean air from the southern Pacific. Despite the poorest MOZAIC CO sampling, the seasonal cycle appears to be captured as well.
-The European PTC m (CO) cycles vary from 1.9 to 2.7×10 18 molecules cm −2 . The IQR over Germany are the steadiest, with for all of them less than 0.25 × 10 18 molecules cm −2 , and highlight the impact of the sampling frequency that is well illustrated by comparing the three European sites in June. The sharp depletion in May-June, already seen over the US, is evidence of the powerful OH cleansing efficiency regulated by NO x (Lamsal et al., 2011). Later, the seasonal cycle remains almost unchanged from June to November (Paris) or starts a very slow increase in August (other European sites), probably related to remote fires. Interestingly, wintertime shows higher amounts over Vienna than over Paris (10 %) or Germany because of its downwind position in Europe or the influence from the Po Valley.
-The Middle East PTC m (CO) cycles are both in the same range. They show the lowest PTC m (CO) in winter and the comparable weakest seasonal cycle amplitudes (1.8-2.3×10 18 molecules cm −2 ). Focusing on Dubai, Tangborn et al. (2009) compared MOZAIC to the assimilated SCIAMACHY CO data into the Global Modeling and Assimilation Office (GMAO), which includes the CO transport by the use of meteorological analyses from the Goddard Earth Observing System (GEOS) version 4. From this comparison, with favourable cloud-free conditions, they had to reduce the OH by 10 % and double the CO emissions in the model in order to lessen the total CO column differences to their MOZAIC independent data set from 1.0 to 0.6 × 10 18 molecules cm −2 . Nevertheless, their MOZAIC data set did not take into account the extremely large unvisited tropospheric remainder, which we estimated to be ∼ 10 % of PTC m (CO) as a yearly average performed in 2004 and greater than 0.3 × 10 18 molecules cm −2 in October, i.e. equivalent to the amplitude of the seasonal cycle. Over Eastmed, due to the spring low PTC m (CO) in May-June, CO does not appear to be the most important spring source of O 3 here, and thus the large amount of TC m (O 3 ) appears here to be more consistent with intense stratospheretroposphere exchanges (STE), NO x contribution or a long-range transport hypothesis. In summer, CO remains in the same range as over Europe, far less than in Asia.
-The Asian PTC m (CO) cycles exhibit the highest CO amounts. They are extremely favourable to O 3 production when combined with the highest NO x amount on the hemisphere scale as detected from satellites over China, during 1996China, during -2005, and related to human activity (van der A et al., 2008; Schneider and van der A, 2012). The Beijing CO cycle in Fig. 6 has a very different vertical scale of 2.5-9.0 × 10 18 molecules cm −2 . In fact, the minimum of its cycle is 3.35 × 10 18 molecules cm −2 in November, and this minimum exceeds the maximum of all the cycles of the 10 other sites studied (i.e. 3.05 × 10 18 molecules cm −2 in March in Japan). These highest values are associated with the largest inter-annual variability (see Beijing -in December with 10 profiles sampled) and we noted that Beijing in February was only sampled in 2003 (no box-and-whisker plot). The Japanese cycle peaks in March (3.0 × 10 18 molecules cm −2 ) reaches a minimum in August (2.3 × 10 18 molecules cm −2 ) and has an irregular IQR from 0.1 to 0.4×10 18 molecules cm −2 . Finally, regardless of season and prevailing winds considered, the air masses over Beijing are systematically CO-enriched by a factor of 1.2-2.8 compared to Japan.

Pure tropospheric profiles
To complement the tropospheric column climatology, the PTP s (O 3 ) and PTP s (CO) are provided to evaluate the vertical anomalies more precisely from a regional point of view. The North American, European, and the Asian and Middle Eastern sites are grouped together in Figs. 8, 10 and 11, respectively. The PTP s (X) are plotted up to the seasonal mean z DT between 9.5 and 15 km. This additional information along the vertical allows for further analysis, shedding light on the origin of the two chemical species. For this reason, the O 3 (top row) and CO (bottom rows, i.e. 0-2 km and 2-15 km) are on the same figure for each region.

North American profiles
The results are plotted in Fig. 8. Over the northern US, the PTP s (O 3 ) show the typical autumn-winter and springsummer seasonal dichotomy as previously described in Zbinden et al. (2006). This is the typical seasonal behaviour of the vertical pure tropospheric ozone distribution in northern mid-latitudes. It characterizes the photochemical activity leading to O 3 production. The PTP s (CO) presents a different dichotomy in the free troposphere with a winter-spring maximum and a summer-autumn minimum, the minimum probably being related to the higher amounts of OH at that time.
Over the southern US, the seasonal vertical distributions are slightly different. Over USsouth, between 1 and 4 km, the summer O 3 profile reveals the O 3 -poor monsoon flux impact. Among all the sites, we found the lowest amounts of CO here in the summer BL because the high OH amounts shortened its lifetime. Between 6 and 9 km, quantities of O 3 are enhanced, probably due to additional O 3 from lightning NO x production, especially in summer (Li et al., 2005). This counterbalances the low amounts of O 3 in the BL and explains why the tropospheric columns in summer are still maximized (Fig. 5). In autumn, the amounts of O 3 in the BL are higher than in winter and thus the typical dichotomy is strongly modified. Furthermore, at that time in the BL, the CO is the lowest of all the studied sites likely due to the influence of www.atmos-chem-phys.net/13/12363/2013/ Atmos. Chem. Phys., 13, 12363-12388, 2013  U S l a k e -1 0 . 3 , 1 1 . 6 , 1 0 . 7 , 9 . 5 k m  U S s o u t h -1 1 . 9 , 1 3 . 4 , 1 3 . 0 , 1 1 . 5 k m  L o s A n g e l e s -1 1 . 2 , 1 3 . 0 , 1 1 . 4 , 1 1 . 4 k m  U S e a s t -1 0 . 3 , 1 1 . 9 , 1 0 . 9 , 9 . 7  oceanic air masses. Thus, this suggests this BL high ozone results from very local production, in which biogenic emissions might interplay. Note that airborne measurements over the southern United States during the field campaigns Tex-AQS2000, ICARTT2004 and TexAQS2006 have allowed for quantifying the biogenic emissions, have shown great interannual variability (by a factor of 2) within the period 2002-2006, and have found that the emission inventories were overestimated by a factor of 2 (Warneke et al., 2010). The biogenic emissions contribution is a hypothesis to explain this higher ozone in the BL. Such characteristics are also observed over Los Angeles and Beijing (see below). With regard now to Los Angeles, the typical autumn-winter and spring-summer seasonal dichotomy is not found, and this change cannot be attributed to the poor sampling because Germany, when under-sampled consistently with Los Angeles in coincidence (within 24 h), still shows the seasonal dichotomy (Fig. 9). The Los Angeles profiles in Fig. 8 exhibit fine structures along the vertical, probably accentuated by the low MOZAIC monthly sampling rate. However, the spring O 3 spikes of +0.03 DU between 2 and 7 km could be indications of long-range transport from Asia (Jaffe et al., 2003;Parrish et al., 2004;Cooper et al., 2005;Neuman et al., 2012). Above 1 km, the summer O 3 profile is unusually close to the autumn-winter one (less than 0.02 DU difference), while the CO profile is extremely low, which might be related to air coming from the southern Pacific, as already studied by Oltmans et al. (2008) andNeuman et al. (2012). In contrast, below 1 km, the highest summer maximum of all the sites studied (0.28 DU) seems to be more in agreement with the lack of deep convection (Cooper et al., 2006). The winter CO below 0.5 km is higher than elsewhere in the US. Thus, to summarize from what MOZAIC has measured under these poor sampling conditions, the secondary peak of the PTC m (O 3 ), in summer, depends on a high z DT and on heavy local pollution as O 3 in the BL reaches 0.28 DU (maximum of the overall study) with high CO up to 3 × 10 16 molecules cm −2 .

European profiles
The European O 3 profiles (Fig. 10) present similar seasonal dichotomies and are comparable to the two northern US ones (Fig. 8). Nevertheless, we found the following: (1) O 3 excess all throughout the troposphere explains the higher PTC m (O 3 ) over Vienna compared to the other European sites; (2) the greatest spring CO amounts above 2 km over Paris are indications of long-range pollution transport by the westerly wind from the east coast of the US; and (3) during winter, in the BL, CO increases from west to east, suggesting strong contamination by dry CO-polluted air from central and eastern Europe, in agreement with Kaiser (2009). The influence of the Po Basin on Vienna might be another source of contamination, but it is currently difficult to quantify (Kaiser, 2009); (4) over Vienna, regardless of season, at the surface, O 3 and CO are at least 10 ppbv and 25-50 ppbv greater than at the other European sites (not shown), again evidence of higher pollution.

Middle Eastern and Asian profiles
The seasonal profiles for the Middle East (Eastmed and Uaemi) and Asia (Beijing and Japan) are given in Fig. 11. Interestingly, none of them has the typical seasonal O 3 and CO dichotomy as seen over Europe or the northern US. For O 3 , this is due to (1) a strong positive summer 1-7 km anomaly over Eastmed; (2) a spring profile closer to autumn-winter over Uaemi above 2 km compared to all the other sites; and (3) strong summer anomalies in Asia. Note the CO horizontal scale change for the 2-15 km range over Beijing and Japan and the 0-2 km range over Beijing (by a factor of 6).
Over Middle Eastern sites, very similar winter profiles above 1 km for O 3 and CO suggest a probable common source of variance. Thus, the PTC(O 3 ) seasonal cycle contrast in winter comparing the two sites (Sect. 4.1.1) is mainly due to z DT . The Eastmed autumn profile is above 3 km in the range of Europe and less than Uaemi, while below 3 km, Eastmed exceeds Uaemi and of course Europe. At that time and below 3 km, a minimum of O 3 and CO over Uaemi might be related to the impact of sea breezes at the surface (Eager et al., 2008) and OH efficiency. Lawrence and Lelieveld (2010) reported evidence of a broad summertime Middle East O 3 maximum around 400-500 hPa. The authors mentioned this O 3 anomaly is "in contrast to the ozone-depleted Asian airmasses observed in the upper troposphere during MINOS, and is not always observed in satellite retrievals (e.g., Fishman et al., 2003;Liu et al., 2006)" and that "the cause of this difference is not yet resolved". Therefore, it seems useful to comment on the MOZAIC summer profiles in more detail. At 6 km, at both Middle Eastern sites, the O 3 and CO amounts are similar and O 3 is identical to that of Beijing and greater than in Europe. Above 6 km, the O 3 profiles are still comparable but surprisingly lower than Europe or Beijing (by 0.02 DU at 9 km), which suggests STE are probably not the most predominant processes involved. Below 6 km, the O 3 profiles show a marked contrast, as follows: (1) over Eastmed, an intense O 3 maximum within 1-6 km, never reached elsewhere in our study, and (2) over Uaemi, a pronounced negative O 3 anomaly (by −0.06 DU at 3 km) with higher amounts of CO (+10 %) when compared to Eastmed. The strong O 3 and CO anomalies combined with high levels of H 2 O (not shown) over Uaemi exclude the impact of predominant STE and probably reveal an intense inflow of sea air and an influence of the remote Indian summer monsoon, in agreement with Li et al. (2001). Lastly, in spring, the Uaemi O 3 profile is unusually close to autumn and winter profiles, in particular at 3 km. These results are in good agreement with the summer extremes shown on the OMI climatology produced by Ziemke et al. (2011) and with the Persian Gulf region study by Lelieveld et al. (2009). The summer O 3 at 5-7 km agrees well with the model results from Liu et al. (2011) and the study of Lelieveld et al. (2009). Liu et al. (2011) emphasize that the geographic position of the Arabian anticyclone has a major influence on chemical transport. They show that the Middle East mid-tropospheric summer maximum is strongly related to Asian sources of pollution (> 30 %), to local production (23 %) and also to northern USA pollution (> 6 %) transported through the subtropical westerly jet that descends in this area. Over the Zagros Mountains, using spaceborne SAGE II 8 data, Kar et al. (2002) found a CO summer positive anomaly at 7 km over 1985-1990 and 1994-1999 and, in a 2003 spaceborne MO-PITT study, argued and explained it as thermal mountain winds venting the BL . We found a different G e r m a n y -1 0 . 2 , 1 0 . 6 , 1 0 . 6 , 1 0 . 2 k m  V i e n n a -1 0 . 1 , 1 0 . 6 , 1 0 . 5 , 1 0 . 1 k m s p r i n g s u m m e r a u t u m n w i n t e r P a r i s -1 0 . 3 , 1 0 . 7 , 1 0 . 5 , 1 0 . 5  notable summer positive (negative) CO anomaly at Uaemi within 3-6 km (at 7 km) and not over Eastmed, with the MOZAIC sampling conditions and period. Due to regional complexity, extensive work is needed in order to go further into the geographical and seasonal variabilities, and in future studies we deeply recommend the addition of the H 2 O climatology using MOZAIC/IAGOS to reinforce the hypothesis on processes involved and air mass origin. Over Asia (Fig. 11), a comparison of Japan and Beijing seasonal O 3 profiles reveals that (1) winter and spring profiles are very similar, with a difference of less than 0.01 DU; (2) the autumn Beijing profile exceeds Japan by 0.03 DU (only 0.01 DU) in the BL (at 8 km); and (3) the summer Beijing profile exceeds Japan by 0.10 DU (0.02 DU) in the BL (at 8 km); and thus in Japan, O 3 at 2 km is reduced to the minimum ever seen regardless of season. Interestingly, up to 4 km, Japan has less CO in all seasons than Beijing by a factor of 2-8, but above 4 km, the two profiles are similar (< 0.2×10 16 molecules cm −2 ), suggesting a common source of variance. The Beijing winter CO maximum is extremely large, exceeding 30 × 10 16 molecules cm −2 or 1800 ppbv in the BL (Japan is 4.2×10 16 molecules cm −2 or 220 ppbv, even less than Vienna). A comparison of the two sites highlights the prevailing wind mechanism and the predominant monsoon impact. Both sites show a summer O 3 depletion below 5 km, but the monsoon is (1) less efficient over Beijing, probably due to the high CO (10×10 16 molecules cm −2 at the surface) compared to Japan, and (2) so intense over Japan that below 4 km, O 3 summer amounts are less than in winter. The monsoon is the most important and powerful mechanism for reducing tropospheric O 3 on the hemispheric scale.
The pure tropospheric column and profile climatologies of O 3 and CO presented in this study are complementary, highlighting seasonal variability and vertical anomalies and reinforcing the hypothesis on the processes involved in the northern mid-latitudes. The pure tropospheric profiles obtained here are significantly different from atmospheric profiles above 6 km. The autumn-winter and spring-summer seasonal O 3 dichotomy characterizes the typical seasonal behaviour of the vertical pure tropospheric ozone distribution in Atmos. Chem. Phys., 13, 12363-12388, 2013 www.atmos-chem-phys.net/13/12363/2013/  -1 0 . 3 , 1 1 . 6 , 1 0 . 3 , 9 . 7 k m  U a e m i -1 4 . 9 , 1 5 . 7 , 1 5 . 1 , 1 4 . 1 k m  E a s t m e d -1 1 . 1 , 1 4 . 5 , 1 2 . 5 , 1 0 . 7 k m  J a p a n -1 0 . 7 , 1 3 . 1 , 1 2 . 3 , 9 . 6  the northern mid-latitudes. It characterizes the photochemical activity leading to O 3 production. The CO dichotomy, not so obvious and different to the O 3 dichotomy, is pointed out with a winter-spring maximum and a summer-autumn minimum in the free troposphere. The minimum is related to the higher amounts of OH at that time. We have singled out the monsoon as the most efficient regime for O 3 reduction, with a significant impact on the hemispheric scale in the northern mid-latitudes, below 6 km. Such a pure tropospheric climatology should be considered as a reference for validation remote sensing instruments or chemistry-transport model outputs.

Partial tropospheric columns
Here, we investigate the interest of characterizing z Ld , the altitude from which PTP m (X) and MP m (X) diverges, to calculate the BL, MT and UT partial tropospheric columns (BLC, MTC and UTC, respectively) for USeast, Germany and Japan.

Divergence between MP and PTP
The limit, z Ld , is defined by month, and examples are given in Fig. 2a. z Ld varies by season and site due to either tropopause altitude variations or the occurrence of stratospheric air detection. The z Ld is higher in summer than in winter and generally higher at southern sites than at northern sites. For example, the 6 km minimum of z Ld , observed in late winter over USeast, is an indication of high residence frequency of the polar jet stream and a low tropopause. In August, over USeast the z Ld is 9.5 km; therefore the MT ceiling fixed at 8 km, as used in Zbinden et al. (2006), is not satisfactory. We suggest replacing it with the monthly varying one, z Ld , defined as the lowest altitude where MP m (O 3 ) differs from PTP m (O 3 ) by 0.001 DU. In this way, UT, the layer between z Ld and z DT , strongly influenced by the stratospheretroposphere transients, will be more faithfully represented by month.
The z Ld seasonal cycle varies in a 6-12 km range (Fig. 12). Due to low variability of the z DT over Germany, z Ld varies less (6.6-8.1 km) than USeast (6.0-9.5 km) or Japan (6.0-11.9 km). In Thouret et al. (2006) the tropopause layer has been defined between z DT − 15 hPa and z DT + 15 hPa, with z D T z L d z DT fixed at 2 PVU. Our UTC thickness, shown only with respect to altitude in Fig. 10, is in the range of 115-170 hPa, 52-214 hPa and 45-194 hPa for Germany, USeast and Japan, respectively. Therefore, in contrast to the tropopause layer, our UTC is much thicker and excludes the stratospheric air above z DT . Moreover, our UTC is thicker in winter than in summer (Fig. 12) and z Ld highlights the lowest winter tropopause position in the northern mid-latitudes. This finding is noteworthy because the seasonality of the deepest stratospheric intrusions, the one that stayed more than four days in the troposphere, is characterized by a winter maximum and a summer minimum (Stohl et al., 2003). Finally, as dynamical and photochemical processes are different in the BL, MT and UT, we give the O 3 and CO partial tropospheric columns over USeast, Germany and Japan using either the steady altitude fixed at 8 km or a monthly varying z Ld used as the MT ceiling (Fig. 12). The partial tropospheric columns using z Ld have been evaluated strictly for those three most documented sites in order to present significant results, and not elsewhere. The results are summarized in Table 4. No major change in BL is expected with respect to Zbinden et al. (2006) given that the small variations depend only on the time period update.

BL, MT and UT seasonal cycles
The BLC m (O 3 ), the MTC m (O 3 ), and the UTC m (O 3 ) range from 4.8 to 9.8, 12.1 to 23.8 and 2.2 to 9.9 DU, respectively (Fig. 12). Thus, using fully defined MOZAIC tropospheric columns and z Ld , UTC m (O 3 ) are enhanced by 2 DU and show smaller amplitudes than those in Zbinden et al. (2006). Interestingly, the UTC m (O 3 ) maximum is shifted from summer to late winter-early spring. This may be observed even Atmos. Chem. Phys., 13, 12363-12388, 2013 www.atmos-chem-phys.net/13/12363/2013/ Table 4. Pure tropospheric columns PTC(X, s) and related partial columns, UTC(X, s), MTC(X, s) and BLC(X, s), where X is O 3 and CO at season s over USeast, Germany and Japan in DU and in ×10 18 molecules cm −2 , respectively. These results are in bold. For O 3 , the additional small numbers/letters in parentheses refer to previous values on incomplete tropospheric columns (TOC) and partial columns as given in Zbinden et al. (2006). MTC and UTC (bold)  over Germany, where the z DT and z Ld variations are the lowest. In August, the use of z Ld over Japan needs to be regarded with caution due to the low sampling rate of the uppertropospheric layers. Besides the UTC m (O 3 ) shift, we also highlight the intense vertical expansion of MTC m (O 3 ). Thus, over Japan, the previous obvious spring MTC m (O 3 ) maximum has turned into a pronounced summer maximum because the O 3 -poor air mass due to the monsoon in the lower mid-troposphere is balanced by the high column height. The BLC m (CO), the MTC m (CO) and the UTC m (CO) range from 0.7 to 1.1, 0.9 to 1.4 and 0.1 to 0.6 × 10 18 molecules cm −2 , respectively. The BLC m (CO) of the three sites show an increase in amounts and amplitudes from west to east, with a late winter/early spring maximum and a summer minimum.
Even over Germany, where the seasonal z DT is very flat, in MTC m and UTC m , the cycles and amplitudes change when the monthly varying z Ld is used instead of a fixed 8 km altitude. This change becomes significant over USeast and Japan and is obviously linked either to the reservoir thickness or the O 3 amounts. The spring O 3 maximum in MT observed using a fixed 8 km altitude has turned into a broad April-August maximum using z Ld , and UT has become out of phase with an overall August minimum and spring maximum.

Comparison with spaceborne measurements
Our aim in this last section is not to validate satellite products but to demonstrate the benefits of such pure tropospheric climatology based on MOZAIC fully defined individual tropospheric columns. Thus, we compare our new O 3 and CO pure tropospheric columns with the publicly available remote sensing results from the Giovanni website (http://disc.sci.gsfc.nasa.gov/giovanni) on a monthly average basis. The seasonal cycle comparison between the spaceborne and MOZAIC data is performed using the TES level-3 version 2 (2.0 • × 4.0 • gridded data) and AIRS level-3 version 5 (1.0 • × 1.0 • gridded data). Our PTC(O 3 ) will be compared to the tropospheric O 3 columns from TES 9 (Beer et al., 2001, Worden et al., 2007 by selecting the periods -2007or 2007 will be compared to the CO columns from AIRS 10 (Susskind et al., 2003(Susskind et al., , 2010 by selecting the 2002-2009 period. We focus only on two sites because of their contrasting z DT seasonal cycle: Germany [47-52 • N, 6-10 • E] and Japan (Fig. 13) McMillan et al. (2011) for AIRS CO. Nevertheless, the simple comparison we provide here is interesting by itself because it results from two independent data sets, with their own limit and performance.

Ozone
First of all, the recent O 3 tropospheric climatology study from Ziemke et al. (2011) merits consideration. It combines OMI/MLS measurements over six years (2004)(2005)(2006)(2007)(2008)(2009)(2010). Interestingly, on two regions not sampled by ozonesondes, OMI/MLS results in Ziemke et al. (2011)  PTC m (CO) with AIRS CO total columns (right, in ×10 18 molecules cm −2 ) over Germany (top) and Japan (bottom). MOZAIC TC is the blue line and PTC the red line with box-and-whisker plots (as in Fig. 5). The solid green line, z DT , and the dotted green line, as defined in Fig. 4, (in kilometres, right green vertical axis) show the unvisited tropospheric layer, . The satellite results are plotted in black over Germany, and in pink, black and purple over Japan for Tokyo, Osaka and Nagoya, respectively. AIRS CO results are given twice a day for ascending node (13:00 LT, solid line) and descending node (07:00 LT, dotted line  (Ziemke et al., 2011, Fig. 5b) show a June maximum and remains high in July-August, whereas MOZAIC starts a sharp decline. Although OMI gives global distributions of tropospheric ozone, TES is the first spaceborne instrument to provide vertically resolved information on tropospheric O 3 , despite a low sensitivity below 900 hPa (Osterman et al., 2008). For these reasons, we now consider the seasonal cycles of the pure tropospheric O 3 derived from MOZAIC and the tropospheric results from TES (Fig. 13, left). Over Germany, the correlation is excellent (r = 0.93), with a TES positive bias of 9-14 DU. The O 3 seasonal cycles (Fig. 13, top left) are well phased with a summer maximum and a winter minimum. Interestingly, a larger winter TES positive bias, by 2-3 DU, is visible. Over Japan, a strong May maximum is observed on both cycles, and an additional secondary winter maximum is detected by TES (Fig. 13, bottom left). Consequently, the correlation drops to r = 0.60-0.76, with a TES positive bias of 9-18 DU. These results are consistent with, but all higher than, the 7 DU (2.8 DU) bias found by comparing TES tropospheric columns (with the averaging kernel applied) with 1425 sondes, as reported by Osterman et al. (2008). Herman and Osterman (2011) have reported that the "TES ozone profiles are positively biased (by less than 15 %) Atmos. Chem. Phys., 13, 12363-12388, 2013 www.atmos-chem-phys.net/13/12363/2013/ from the surface to the upper troposphere (from ∼ 1000 to 100 hPa) and negatively biased (by less than 20 %) from the upper troposphere to the lower stratosphere (from 100 to 30 hPa) when compared to the ozone-sonde data". Thus, besides the bias due to the instrument vertical resolution and/or the retrieval technique (not discussed here), the differences between MOZAIC and TES are reinforced because, firstly, the observation time influences the O 3 measured in the BL: TES' orbit is sun-synchronous with a 13:43 LT ascending node, comparable to OMI/MLS, while MOZAIC depends on commercial aircraft schedules. The 0-2 km partial column difference in the seasonal cycle using MOZAIC observations over Frankfurt selected at 02:00-06:00 and 11:00-18:00 LT is evaluated to be 2 DU maximum (not shown). Secondly, MOZAIC is almost insensitive to hydrometeorological conditions, while TES data processing requires cloud screening, thus introducing a probable bias due to specific meteorological conditions of sampling. Nonetheless, how can such enlarged winter in situ and spaceborne differences (> 3 DU) be explained at those two sites? In winter, the BL and MT have the lowest O 3 contributions (Fig. 12, top) and z DT is at a minimum. This suggests that the tropopause positioning makes the main contribution to the winter differences, probably because of the 6-7 km vertical resolution of TES measurements and that the tropospheric ozone column might contain some stratospheric information as reported by Osterman et al. (2008). We agree with Stajner et al. (2008), who emphasized the impact of tropopause location on tropospheric O 3 columns, inter-annual variability and trend estimates. Insufficient accuracy in the characterization of the tropopause altitude leads to strong stratospheric contamination in the pure tropospheric O 3 reservoir (Fig. 2) and thus may impact the assessment of trends in the UT.

Carbon monoxide
The AIRS seasonal cycles of total columns and our pure tropospheric seasonal cycles are correlated by r = 0.89 (r = 0.80) on the descending (ascending) node over Germany. They show a larger difference in winter and a similar response at the summer minimum (Fig. 13, right). As the CO maximum along the vertical is below 2 km and as the impact of tropopause height is negligible, the winter difference may be explained by the weak sensitivity of the satellite instrument in the BL. The winter maximum is clearly not well captured by AIRS. Diurnal variations are zero from July to January and maximum in April. Considering Japan (Fig. 13, bottom right), the correlations are r = 0.96-0.98 on the ascending node (13:00 LT) and r = 0.93-0.97 on descending node (07:00 LT). From AIRS, we observe differences on the diurnal CO cycle which are larger in winter-spring and almost zero in summer. They also show a west-to-east negative gradient between the individual Japanese airports. In summer, we note that the best agreement is over Germany and that there is excellent agreement between our pure tro-pospheric seasonal cycles and AIRS in that season (Fig. 13, top right). The differences may be due to less thermal contrast between air in the lower BL and in the surface layer. Besides the low AIRS sensitivity in the BL, an additional source of differences is the cloud screening, a mandatory constraint for AIRS but not required for MOZAIC inducing a sampling effect in the intercomparison.

Conclusions
A new and comprehensive O 3 and CO pure tropospheric climatology has been derived from MOZAIC over 1994-2009 including 40 000 profiles. Eleven sites were documented between 24 and 50 • N and from Los Angeles to Japan in order to give a quasi-global picture of the northern midlatitudes. The previous low-biased tropospheric O 3 climatology  has been improved by adding the complete estimate of the MOZAIC unvisited tropospheric remainder as well as CO. The outcome is a fully defined pure tropospheric climatology from the ground to the individual dynamical tropopause based solely on the MOZAIC individual profiles. The O 3 validation was performed using composite results derived from coincident MOZAIC profiles and WOUDC soundings close to the most documented MOZAIC sites (USeast, Germany, Japan). The pure O 3 tropospheric columns reproduce 96-98 % of the composite tropospheric columns after the addition of WOUDC partial columns to MOZAIC when necessary. Therefore, the pure tropospheric profiles (by season) and columns (by month) are the most robust and accurate vertically integrated results based on in situ measurements. As far as we know, this is the first pure tropospheric climatology and is significantly different from climatological profiles at all the sites, as seen in particular for ozone or by referring to Logan (1994), McPeters et al. (2007), Stajner et al. (2008), Ding et al. (2008) and Tilmes et al. (2012). The first outcome is that the seasonal cycles of the pure tropospheric columns are in the range of 23.7-43.2 DU for O 3 and 1.5-6.8 × 10 18 molecules cm −2 for CO. Due to the photochemistry and OH removal efficiency, the maxima of the seasonal cycles are not in phase: February-April for CO, May-July for O 3 . In terms of zonal variability, for O 3 we globally observe greater summer contents over the northern US, the lowest contents regardless of the months considered over Europe and the greatest spring and summer contents in the Middle East and Beijing. The summer Asian monsoon results in a sharp decrease over Japan, not observed over Beijing, and, surprisingly, in a weakened maximum over Uaemi. In addition, Los Angeles is almost in the range of Asian pollution, except in summer, when incoming air from the Pacific likely strongly interplays. We found that Los Angeles is at the lowest sampling rate acceptable to be included in a climatological study. For CO, the Beijing minimum exceeds the maxima of all other sites for all seasons. The smallest amplitudes and the lowest winter-spring CO columns are detected in the Middle East.
The second outcome is the seasonal pure tropospheric profile climatology. At all sites, the O 3 pure tropospheric profiles, in DU, never return to a positive vertical gradient above 2 km (except on the poorly sampled winter profiles of Los Angeles), unlike the monthly averaged MOZAIC O 3 profiles. The spring-summer/autumn-winter seasonal dichotomy on O 3 seasonal profiles is confirmed, as is the deep summer decrease due to the monsoon over Japan, Beijing and USsouth with different intensities. Comparing Uaemi with Eastmed in summer, a strong negative O 3 anomaly below 6 km appears. It could be linked to the impact of sea breeze or depleted maritime air inflow from regions under Indian monsoon conditions. Regarding CO, we observe a summer-autumn/winterspring dichotomy with a clear winter maximum in the BL. Note that in the BL, the CO winter maximum over Beijing is 2-7 times (2-6 times) greater than over Japan (Europe), whereas only 2 times greater above 5 km. Additionally, comparing CO for all the sites in winter at 5 km, the minimum is encountered over the Middle East and southwestern US sites, while north-eastern US sites and Europe show both comparable and higher amounts in contrast to the strong Asian maximum values. The vertical profile study shows a large O 3 and CO homogeneity among the European sites with a west-to-east gradient. We found the upper pure tropospheric layers to be very consistent among chemical species, namely with less O 3 and more CO than when tropospheric/stratospheric reservoirs were undifferentiated, even if monthly averaged profiles were limited to the monthly averaged tropopause altitude.
For the third outcome, we provide the seasonal cycles of the BL, MT and UT partial columns using a new monthly varying criterion, z Ld , as the MT ceiling because of their distinct predominant processes. Consequently, the tropopause of all individual profiles within the time series will always be above z Ld , and thus UT is more faithfully represented. The z Ld is detected at 6-12 km on average, with a winter minimum and summer maximum. This approach highlights a predominant increase (decrease) in the MT O 3 amount associated with a summer-(winter-) wide MT thickening (thinning). The UT O 3 seasonal cycle, maximized earlier than when a fixed ceiling was used, is now more comparable to the UT O 3 seasonal cycle defined using a tropopause reference . For CO, it is interesting to note that the clear MT spring maximum is shifted to a broad April-August maximum, in contrast to what happens in the UT.
The last outcome highlights the benefits of this pure tropospheric climatology based on in situ measurements by comparing the seasonal cycles of the MOZAIC tropospheric columns with those derived from satellites. We focused on two sites, Germany and Japan, because of their contrasting tropopause altitude and seasonal variability. We found consistent seasonal cycles from MOZAIC and satellite data, although with noteworthy differences: (1) for O 3 , TES tro-pospheric columns are correlated with MOZAIC by r = 0.6 (Japan) and r = 0.9 (Germany) and are greater than MOZAIC by 9-18 DU, and the largest biases occur in winter; (2) for CO, AIRS total columns provide a response as much as 1.0 × 10 18 molecules cm −2 lower than MOZAIC (r > 0.9). Besides the instrumental bias, the tropopause is a probable source of discrepancy because just below the monthly tropopause, the MOZAIC profile MP m (O 3 ) exceeds our fully defined and pure tropospheric profile. We noted a greater O 3 bias in winter, when the tropopause is low (2-3 DU over Germany, > 5 DU over Japan). Therefore, the O 3 difference is probably mostly explained by stratospheric air contamination but is also explained by clear-sky conditions or the satellite overpass time capturing the daily maximum. All three factors can enhance the satellite response. We point out the need for accurate discrimination between the stratospheric and tropospheric reservoirs, and indeed the monthly tropopause is an unappropriated criterion. This point is crucial, merits consideration and must be clarified within the framework of tropospheric trends and radiative transfer studies.
These fully defined and pure tropospheric products will be available in the MOZAIC/IAGOS database to the scientific community. Such climatologies will be regularly updated thanks to the ongoing MOZAIC programme, IAGOS, whose database is available at http://www.iagos.org/.