Seasonal variability of surface and column carbon monoxide over megacity Paris, high altitude Jungfraujoch and Southern Hemispheric Wollongong stations

. Carbon monoxide (CO) is an atmospheric key species due to its toxicity and its impact on the atmospheric oxidizing capacity, both factors affecting air quality. The paper studies the altitude dependent seasonal variability of CO at the three different sites Paris, Jungfraujoch and Wollongong, with an emphasis on establishing a link between the CO vertical distribution and the nature of CO emission sources. The CO seasonal variability obtained from the total columns and from the 5 free tropospheric partial columns shows a maximum around March-April and a minimum around September-October in the Northern Hemisphere (Paris and Jungfraujoch). In the Southern Hemisphere (Wollongong) this seasonal variability is shifted by about 6 months. Satellite observations by IASI-MetOp and MOPITT instruments conﬁrm this seasonality. Ground-based FTIR is demonstrated to provide useful complementary information due to good sensitivity in the boundary layer. In 10 situ surface measurements of CO volume mixing ratios in Paris and at Jungfraujoch reveal a time-lag of the near surface seasonal variability of about 2 months with respect to the total column variability at the same sites. The chemical transport model GEOS-Chem is employed to interpret our observations. GEOS-Chem sensitivity runs allow identifying the emission sources inﬂuencing the seasonal cycle of CO . In Paris and on top of Jungfraujoch, the surface seasonality is mainly driven by anthro- 15 pogenic emissions, while the total column seasonality is also controlled by air masses transported by biomass and MOPITT (Measurements Of Pollution In The Troposphere (Drummond and Mand, 1996)). With respect to satellite measurements, ground-based FTIR instruments are more sensitive to the boundary layer 75 and can therefore provide complementary data which we compare with surface in situ measurements. Using custom GEOS-Chem model (Goddard Earth Observing System-chemical transport model (CTM), Bey et al. (2001)) simulations, we investigate the impact of local sources on the lower partial column and its variability as compared to the total column. is structured as follows. section 2, the different ground-based and instru- 80 ments will be the levels. This corresponds to a much thinner atmospheric layering than the effective vertical resolution indicated by the averaging kernels Figure 1. shows that the retrieval of CO essentially provides two independent measurement points of CO in the troposphere: the ﬁrst point delivers maximal information in the altitude range between 0 and 1000m and thus well the The second of the upper troposphere, with a maximum around 8-9km. The uncertainties in the CO column density and the proﬁle stem from a variety of sources. These sources been investigated detail by Té et al. (2012), following the procedure outlined by Rinsland et al. (2000). According to this evaluation, the random uncertainty is around only due to local emissions, but also due to other natural and anthropogenic contributions: biomass burning, long distance transport, chemical processes, i.e. oxidization of methane. The surface seasonal variability is directly inﬂuenced by the local emission due to human activities: fossil 400 fuel combustion, warming system, and industrial activities. In comparison the total column seasonal variability is additionally inﬂuenced by distant sources transported to the upper levels of the atmosphere. At Paris, the seasonality introduced by these distant sources outweighs the contribution of the local surface. The surface CO maximum in January-February corresponds to the winter season, where domestic heating is strong, where the PBL height is reduced and when oxidation by OH 405 high altitudes in Switzerland, as indicated by the difference between urban and mountain sites. Figure compares the four NABEL urban sites with an average altitude of 438.25m asl with the in situ surface CO obtained at Jungfraujoch with an altitude of 3578m asl. The low altitude sites located in urban areas show a similar seasonal variability as the surface CO at Paris, with a maximum around January and a 420


Introduction
Air is one of the most fundamental prerequisites for life and human beings inhale about 1500 litres of 20 air per day. This air contains, besides of the major gaseous components nitrogen, oxygen and argon, reactive trace gases and small particles that are of concern for human health. The survey and control of these trace components, which affect air quality, have thus become a field of major importance for environmental research and public health authorities, especially in large cities. The quality of air is a function of time and space, which depends on many parameters such as geographic location, 25 meteorological conditions, as well as sources and sinks of pollutants. It is thus strongly affected by natural and anthropogenic emissions.
Atmospheric carbon monoxide (CO) is an important trace gas, due to its toxicity and its impact on the atmospheric oxidizing capacity and air quality. For example, Sam-Laï et al. (2003) have studied the seasonal phenomenon of CO poisoning mainly due to defect heating systems. Barnett 30 et al. (2006) have investigated the link between outdoor air pollution and cardiovascular hospital admissions. Levy (2015) has studied the effect of CO pollution on the neurodevelopment. The major sources of CO are fuel and energy related industries, heating, motor vehicle transport, biomass burning, and the secondary oxidation of methane and of volatile organic compounds (VOCs such as isoprene and terpene), which are emitted by plants. Due to the fast reaction (R1), carbon monoxide 35 is the major sink for the main atmospheric oxidation agent, the hydroxyl radical OH (Weinstock, 1969;Bakwin et al., 1994).
A global increase of atmospheric CO thus leads to a decrease in global OH, which in turn augments the concentration of other, potentially harmful atmospheric trace gases (Logan et al., 1981; 2 Instrument description 2.1 Instrumentation at Paris, France The Fourier transform spectrometer (FTS-Paris) is a model IFS 125 HR Michelson interferome-85 ter from Bruker Optics, cf. http://www.bruker.com. Its maximum optical path difference is up to 258 cm, which corresponds to a spectral resolution of 2.4 × 10 −3 cm -1 .The instrument is equipped with IR optical elements (CaF 2 entrance window and beamsplitter, InSb detector), suited for groundbased atmospheric observations (Té et al., 2010). Solar absorption measurements are achieved by coupling the FTS-Paris instrument to a sun-tracker (model A547 from Bruker Optics) installed on 90 the roof terrace. The solar disk is tracked with an accuracy of less than 1 arcmin. The spectra contain rovibrational signatures of many atmospheric constituents, including numerous atmospheric pollutants. The spectral range determined by the above choice of optical elements and detectors is limited to the range between 1 and 5.4 µm. It is further narrowed down using appropriate band pass filters in order to optimise the signal to noise ratio when focussing on specific target gases. For CO, the chosen 95 optical filter and the InSb detector allow to cover the spectral domain from 3.8 to 5.1 µm, which corresponds to a typical NDACC configuration. More instrumental details and different measurement configurations are specified elsewhere (Té et al., 2010(Té et al., , 2012. Continuous in situ measurements of the CO surface concentration are performed using a commercial CO11M analyser (Environnement SA). The operating principle of the CO analyser is based on 100 the CO infrared absorption at 4.67 µm, which is the same spectral band covered by the FTS-Paris.
Ambient atmospheric air is drawn from the building rooftop into the analyser via PTFE tubing using a diaphragm pump, which is limited to a gas flow of 80 litres per hour. The pumped air is analysed in a 20 cm length multi-path absorption cell with an absorption path length of 5.6 m, using a globar IR source and a photoconductive PbSe detector. The CO11M analyser has a sensitive range between absorption lines of each atmospheric species observed in the solar spectra are used to retrieve its abundance in the atmosphere by appropriate radiative transfer and inversion algorithms (Pougatchev and Rinsland, 1995;Zhao et al., 1997;Hase et al., 2006). We have used the PROFFIT algorithm developed by F. Hase to analyse the Paris data using HITRAN 2008 (Rothman et al., 2009) as spectral database. PROFFIT is a code especially adapted for the analysis of solar absorption spectra 170 from the ground and it has been widely applied and tested (Hase et al., 2004;Duchatelet et al., 2010;Schneider et al., 2010;Té et al., 2010;Viatte et al., 2011). For the retrieval of CO, we have selected two micro-windows. The 2110.4 − 2110.5 cm -1 micro-window is centred around the weak 13 CO R(3) line, which is more sensitive to CO at higher altitudes and the 2111.1 − 2112.1 cm -1 micro-window around the strongly saturated 12 CO P(8) line. The left and right wings of that line are 175 particularly sensitive to CO in the Planetary Boundary Layer (PBL). The retrieval uses a grid with 49 altitude levels. This corresponds to a much thinner atmospheric layering than the effective vertical resolution indicated by the averaging kernels (Rodgers, 1990). Figure 1. shows that the retrieval of CO essentially provides two independent measurement points of CO in the troposphere: the first point delivers maximal information in the altitude range between 0 and 1000 m and thus well 180 represents the PBL. The second one is representative of the upper troposphere, with a maximum around 8-9 km. The uncertainties in the CO column density and the profile stem from a variety of sources. These sources have been investigated in detail by Té et al. (2012), following the procedure outlined by Rinsland et al. (2000). According to this evaluation, the random uncertainty is around 2.5%.  Figure 1. CO averaging kernels for each altitude of the a priori profile (from 0.06 to 20 km) using both microwindows (Paris site).

Column data from Jungfraujoch
The Jungfraujoch data set corresponds to an update of the CO time series described in (Dils et al., 2011). It covers here the January 2009 to December 2013 time period and includes 1733 individual spectra recorded on 539 different days. Mean signal-to-noise ratio (S/N) is 2930, with the 2nd percentile still above 1000. We used the SFIT-2 (v3.91) algorithm (Rinsland et al., 1998)   and Wollongong (bottom). Gray lines present CO seasonal variability at each station fitted with sine functions.

Column data from Wollongong
The analysis of the Wollongong NDACC data follows very closely the method described above with a number of enhancements (not required in the CO analysis), and for the CO retrieval gives the same result. For the Wollongong data, HITRAN 2008 was adopted (Rothman et al, 2009), the mean of the 1980-2020 WACCM version 4 run used as the a priori CO profile (and a 4 km Gaussian interlayer correlation), with the a priori covariance matrix set to 1 standard deviation of the WACCM 220 profiles. A measurement signal to noise ratio of 200 was assumed. This gave a mean DOFS of 2.7.
The version 4 WACCM profiles were also used for the a priori profiles of all actively fitted inferring ). The error analysis used a NDACC community Python tool to estimate errors assuming a solar zenith angle of 50.2 o , representing the mean zenith angle for all Wollongong spectra. The resulting CO total column random errors were calculated to be 2.2%.

Data from the satellite instruments
The IASI-MetOp is a Fourier transform spectrometer with a medium spectral resolution of 0.5 cm -1 and a radiometric noise of about 0.2 K at 280 K using nadir viewing and working in the thermal infrared (TIR) range extending from 645 to 2760 cm -1 with no gaps. The CO products (L2) from the IASI sounder on the MetOp satellite are downloaded from the ETHER database, cf. http: 230 //www.pole-ether.fr, for the period from 1 January 2009 to 31 December 2013. The total column data were generated from the IASI radiance spectra in the 4.7 µm spectral range and from IASI L2 meteorological data (surface and vertical profile of temperature, humidity vertical profile and cloud cover) (August et al., 2012), using the Fast Optimal Retrievals on Layers for IASI (FORLI) code (Hurtmans et al., 2006). The CO total columns were compared to other CO satellite data (George 235 et al., 2009), from which a relative uncertainty between 4% and 10% could be estimated. The total columns are calculated from the ground altitude to 60 km height. For this paper, we have also additional vertical volume mixing ratio (VMR) profile and partial columns in the PBL and in the troposphere layers around Île-de-France; as well as the partial columns above 4 km height around the Jungfraujoch site.

240
The MOPITT data were downloaded from the NASA website, cf. https://eosweb.larc.nasa.gov/ datapool. We are using the available data of the version 6 retrievals of CO vertical profiles and total columns, for the period from the beginning of 2009 to the end of 2013. The MOPITT retrieval history can be found at the link https://www2.acd.ucar.edu/mopitt/products. Since version 5 of the MOPITT retrieval algorithm, TIR (4.7 µm) radiances are combined with the near IR (2.3 µm) daily radiances 245 to improve the sensitivity to lower tropospheric CO over land. The retrieved vertical VMR profile is reported on 10 pressure levels (at the surface and every hundred hPa between 900 and 100 hPa). The retrieved CO total columns are obtained by integrating the retrieved VMR profile. In this paper, we are using the level 2 TIR/NIR products.  (Schultz et al., 2007) for the base year 2000. However, these global inventories may be overwritten by regional emission inventories such as over Europe, where the anthropogenic emissions of CO, NO x , SO x , NH 3 , propene, acetaldehyde, methyl ethyl ketone and 270 higher C3 alkanes are provided by the European Monitoring and Evaluation Programme (EMEP; http://www.ceip.at) regional inventory for the year 2010 (Benedictow et al., 2010). All these global and regional inventories are scaled to the years of interest according to the method described by van Donkelaar et al. (2008). Anthropogenic sources of ethane and propane are derived from an offline simulation (Xiao et al., 2008). The global biomass burning emissions are provided by the Global Fire scheme adopted at each station, then daily averaged and finally smoothed by convolution with the FTIR averaging kernels (AVKs) according to the formalism of Rodgers and Connor (2003). The regridding method used here is a mass conservative interpolation that preserves the CO total mass simulated above the altitude of the station (the CO mass below is ignored). The AVKs employed for  All three panels show clearly the seasonal variability of CO. We have used a sine function Eq. (1) to characterize this seasonal variability.
where y represents the abundance of CO (in total or partial columns or volume mixing ratio); y 0 is the mean value (offset); A and w are respectively the amplitude and the half-period of the seasonal cycle (assumed to be sinusoidal); t and t c the date and the phase shift in days. Table 1 summarizes the fitted w and A obtained at the three sites. is slightly, but not significantly higher, probably due to the lack of data before 2011. This seasonal variability is also observed by Rinsland et al. (2007) at Kitt Peak, which is the US National Solar Observatory at 2.09 km altitude located in the Northern Hemisphere, by Barret et al. (2003) at the 330 Jungfraujoch, and by Zhao et al. (2002) for Northern Japan. Our observations also agree with a recent 11 years climatology on purely tropospheric CO columns at Northern hemispheric sites (Zbinden et al., 2013), where observed maxima fall within the period from February to April. In the Southern Hemisphere, we observe an expected shift of 6 months as compared to the Northern Hemisphere, with a maximum in October and a minimum in April. We also note that the relative amplitude of 335 the seasonal variation is slightly higher at Wollongong (16% as compared to 14% at Paris), but still within error bars. Interestingly, the relative amplitude is lowest at Jungfraujoch, where the impact of the local surface emissions is small.
The seasonal variability of CO is also observed by the satellite IASI-MetOp and MOPITT instruments, cf. Fig. 3   for Jungfraujoch, and −20% for Wollongong. These deviations are consistent with previous inverse modeling studies (Kopacz et al., 2010;Hooghiemstra et al., 2012)     surements are consistent with the in situ data, even if they are much less affected by local pollution peaks. By comparing Figs. 2 and 5, we notice that the seasonal variability of the total column is 390 shifted by about 2 months as compared to the variability at the surface. In order to study the free tropospheric columns, we have recalculated the partial columns of CO between 2 and about 12 km over Paris, obtained by the ground-based FTS-Paris and the satellite IASI-MetOp instruments. Figure 5 (top panel) compares these free tropospheric partial columns with the output from GEOS-Chem.
The seasonal variability in the free troposphere obtained by the three different kinds of data is also 395 shifted by two months as compared to the surface seasonal variation. As the average lifetime of CO is estimated to be about two to three months, the seasonal variation of the CO in the atmosphere is not only due to local emissions, but also due to other natural and anthropogenic contributions: biomass burning, long distance transport, chemical processes, i.e. oxidization of methane. The surface seasonal variability is directly influenced by the local emission due to human activities: fossil There is also a temporal shift in seasonal cycles between surface and high altitudes in Switzerland, as indicated by the difference between urban and mountain sites. Figure 6   of CO is mainly driven by biomass burning sources modulated by the OH sink, (Buchholz et al., 2016). Similar to Paris, the surface CO discrepancy between model and measurement of −33% is slightly increased as compared to the value of −20% for the total columns. In order to study the influence of the different categories of CO and NMVOC emissions on the CO total column and its seasonality at the three sites, another three GEOS-Chem simulations were performed. These relied on the same setup as for the standard run (standard chemistry, horizontal resolution, time period. . . ), but in each of these runs we turned off either the biogenic, the anthro- the same results as compared to the CO total columns for the three studied sites.

475
This paper investigates the seasonal variability of CO total columns at three NDACC and/or TC-CON sites: Paris and Jungfraujoch in the Northern Hemisphere and Wollongong in the Southern Hemisphere. The variability of CO above the PBL has a seasonal maximum in March-April and a minimum in September-October in the Northern Hemisphere. This seasonal cycle is shifted by 6 months in the Southern Hemisphere. For both Northern-hemispheric sites, the seasonal variability of 480 the CO total columns seems to be mainly driven by anthropogenic emissions. On the contrary, the Southern-hemispheric site Wollongong is mainly influenced by the biomass burning contribution.
We have compared the ground-based FTIR data to satellite measurements from IASI-MetOp and MOPITT and to GEOS-Chem model outputs, which all of them confirm the observed CO seasonal variability. The GEOS-Chem model also shows that the CO seasonality at Paris and Jungfraujoch 485 is mainly controlled by anthropogenic emissions. This is different to Wollongong, where it is due 19 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2015-884, 2016 Manuscript under review for journal Atmos. Chem. Phys. Published: 8 March 2016 c Author(s) 2016. CC-BY 3.0 License. to biomass burning. For sites that are strongly affected by local anthropogenic emissions, we have observed a temporal shift in the seasonal patterns at the surface and in the higher atmospheric layers. This is likely because zonal mixing occurs on a shorter (1 -2 weeks) timescale as compared to complete vertical tropospheric mixing (1 -2 months). The observed time-lag between upper altitude and 490 surface CO is about 2 months in Paris and at the Jungfraujoch. The 2 months' shift is also confirmed by the GEOS-Chem model. In Wollongong, where low local anthropogenic emissions prevail and which is largely impacted by biomass burning, such a time shift is neither observed nor modelled.
Acknowledgements. We are grateful to Université Pierre et Marie Curie and Région Île-de-France for their financial contributions and to Institut Pierre-Simon Laplace for support and facilities. We thank the National

495
Center for Atmospheric Research MOPITT science team and NASA for producing and archiving the MOPITT CO product. Thanks are also due to the Swiss National Air Pollution Monitoring Network (NABEL) for delivering ground data around Switzerland. The University of Liège contribution to the present work has primarily been supported by the F.R.S. -FNRS, the Fédération Wallonie-Bruxelles and MeteoSwiss (GAW-CH program).
We thank the International Foundation High Altitude Research Stations Jungfraujoch and Gornergrat (HFSJG, 500 Bern). We are grateful to all colleagues who contributed to the acquisition of the FTIR data. The NDACC datasets used here are publicly available from the network database (ftp://ftp.cpc.ncep.noaa.gov/ndacc/station).
The Australian Research Council has provided financial support over the years for the NDACC site at Wollongong, most recently as part of project DP110101948. We also acknowledge the important contribution to the measurement program at Wollongong made by researchers other than those listed as co-authors here, in- [  Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2015-884, 2016 Manuscript under review for journal Atmos. Chem. Phys. Published: 8 March 2016 c Author(s) 2016. CC-BY 3.0 License. ground-based infrared spectroscopic measurements of tropospheric carbon monoxide and ethane, J. Geophys. Res., 103, 28 197, doi:10.1029/98JD02515, 1998.