CO , NO x and 13 CO 2 as tracers for fossil fuel CO 2 : results from a pilot study in Paris during winter 2010

Measurements of the mole fraction of the CO 2 and its isotopes were performed in Paris during the MEGAPOLI winter campaign (January–February 2010). Radiocarbon (CO2) measurements were used to identify the relative contributions of 77 % CO2 from fossil fuel consumption (CO 2ff from liquid and gas combustion) and 23 % from biospheric CO2 (CO2 from the use of biofuels and from human and plant respiration: CO2bio). These percentages correspond to average mole fractions of 26.4 ppm and 8.2 ppm for CO 2ff and CO2bio, respectively. TheCO2 analysis indicated that gas and liquid fuel contributed 70 % and 30 %, respectively, of the CO2 emission from fossil fuel use. Continuous measurements of CO and NOx and the ratios CO CO2ff and NOx CO2ff derived from radiocarbon measurements during four days make it possible to estimate the fossil fuel CO 2 contribution over the entire campaign. The ratiosCO CO2ff and NOx CO2ff are functions of air mass origin and exhibited daily ranges of 7.9 to 14.5 ppbppm−1 and 1.1 to 4.3 ppbppm −1, respectively. These ratios are consistent with different emission inventories given the uncertainties of the different approaches. By using both tracers to derive the fossil fuel CO 2, we observed similar diurnal cycles with two maxima during rush hour traffic.


Introduction
Worldwide, approximately 20 "megacities" have a population of more than 10 million inhabitants.According to the United Nations, half of the world's population now lives in cities, and in the near future, the number of megacities and their population density figures are expected to grow considerably.In the context of global warming and in the framework of the Kyoto Protocol, it is important to better characterise greenhouse gas (GHG) emissions from megacities, which constitute a significant emission source at the global scale.More than 70 % of global fossil fuel CO 2 emissions are concentrated in cities (Duren and Miller, 2012).
With 12 million inhabitants (approximately 18 % of the French population), Paris and its suburbs constitute the largest megacity in Europe.According to the regional French emission inventory provided by AirParif (http://www.airparif.asso.fr/),CO 2 emissions from Paris and its agglomeration (the Ile-de-France region or IdF) were 50 Mt in 2005, or 13 % of the total French anthropogenic CO 2 emissions, although the IdF region, with a surface area of 12 011 km 2 , composes only 1.8 % of the French territory.The annual CO 2 emission density within the IdF region increases from Published by Copernicus Publications on behalf of the European Geosciences Union.
approximately 5000 t CO 2 km −2 limits of the suburbs to approximately 70 000 t CO 2 km −2 at the centre of Paris, reflecting the change in population density.Based on the Kyoto Protocol, France is committed to reducing its greenhouse gas emissions by 8 % with respect to its 1990 emissions by 2012.A decrease of 6.0 % was observed in 2010 compared to the reference year 1990, primarily due to reductions in N 2 O and CO 2 emissions (CITEPA, 2012).In addition, France adopted the "climate and energy package", which aims to reduce greenhouse gas emissions by 20 % with respect to the emissions of 2005 by 2020.Currently, regional and local GHG emissions data are primarily derived from statistical inventories that are gathered using a bottom-up method that involves the compilation of activity and process data.Economic statistics are gathered from the local to the national level to produce gridded maps of GHG emissions (EDGAR: Olivier et al., 2001;Olivier and Berdowsky, 2001).Peylin et al. (2011) showed that such emission inventories can vary up to 40 % on the national scale even for developed countries like the Netherlands, where we can expect the best datasets available for use in compiling the inventory.Atmospheric measurements can provide an alternative to develop methods of emission verification, in association with bottom-up inventories.Atmospheric methods (top-down methods) rely on a combination of high-precision measurements of atmospheric mole fractions of GHGs such as CO 2 as well as on measurements of atmospheric tracers and atmospheric transport models.At global to regional scales, inverse models have been used successfully to quantify GHG emissions for CO 2 (Rayner and Law, 1999;Bousquet et al., 2000;Rödenbeck et al., 2003;Peylin et al., 2005), CH 4 (Rödenbeck, 2005;Bousquet et al., 2006) and N 2 O (Hirsch et al., 2006;Thompson et al., 2011).On the local and regional scales, continuous observations of GHGs can be combined with observations of radon-222, which is used as a tracer for atmospheric dilution (van der Laan et al., 2009;Yver et al., 2009;Hammer and Levin, 2009) to estimate surface emissions.
The use of atmospheric measurements of CO 2 to estimate surface emissions in urban areas is of growing interest.During the last several years, a few projects focusing on quantifying CO 2 fluxes from cities using top-down approaches have emerged in the USA and in France, including the InFLUX project in Indianapolis, USA (http://influx.psu.edu/), the Los Angeles Megacity project (Duren and Miller, 2012) and the CO 2 -Megaparis project in Paris (https://co2-megaparis.lsce.ipsl.fr).One challenging issue associated with the atmospheric top-down approach is the attribution of emissions to specific processes.The first step in this process is to distinguish between CO 2 emitted from anthropogenic activities and CO 2 emitted from biospheric activities.Even in urban areas, variations in the atmospheric CO 2 mole fraction contain signatures from both fossil fuel combustion and biogenic sources.These signatures can be decrypted by additional information that is provided by specific tracers.Only 14 CO 2 measurements allow for the direct quantification of the contribution of fossil fuel CO 2 to observed CO 2 mole fractions (Levin et al., 2003;Turnbull et al., 2006;Vogel et al., 2010) because CO 2 from fossil fuels does not contain 14 C. Levin and Karstens (2007) showed that fossil fuel emissions and biospheric fluxes are of similar orders of magnitude in Europe.Because precise 14 CO 2 measurements are difficult to perform, carbon monoxide is often used as a tracer for fossil fuel CO 2 emissions (CO 2 ff) in order to separate the biospheric and the fossil fuel components from the total measured CO 2 mole fraction.The CO / CO 2 ff ratio depends on the condition of combustion and on the fuel type, and can range from 17.5 ppb CO ppm −1 CO 2 for industrial sources and 14.5 ppb CO ppm −1 CO 2 for residential heating to 5.3 ppb CO ppm −1 CO 2 for traffic according to the national emission inventories by the Centre Interprofessionnel Technique d'Etude de la Pollution Atmospherique (CITEPA, http://www.citepa.org/)for the year 2010 in France.From 2000 to 2010, improvements in combustion efficiency due to regulations that are intended to improve air quality led to a significant reduction in CO emissions and also to changes in CO / CO 2 ff fossil fuel ratios, which decreased from 20 to 13 ppb CO ppm −1 CO 2 in France.This reduction was primarily caused by the introduction of catalytic converters for cars, which decreased the road traffic CO / CO 2 ff ratio from 21 to 5 ppb CO ppm −1 CO 2 .
The European project MEGAPOLI (Megacities: Emissions, urban, regional and Global Atmospheric POLlution and climate effects, and Integrated tools for assessment and mitigation, http://www.megapoli.info)aims to assess the impact of megacities on air pollution, to quantify the feedback regarding megacity air quality and to develop integrated tools for predicting air pollution in megacities.To address these three objectives, the MEGAPOLI project organised two intensive measurement campaigns in Paris: one in July 2009 and one in January-February 2010.During the campaigns, numerous atmospheric pollutants and parameters were monitored by different groups (Royer et al., 2011;Healy et al., 2012;Dolgorouky et al., 2012).The French Atmospheric Network for Greenhouse Gases Monitoring (RAMCES), which is part of the Laboratory for Climate and Environmental Sciences (LSCE), participated in the frame of the CO 2 -Megaparis project to the measurement campaign in the winter of 2010, which extended from 15 January to 19 February.During this campaign, we focused on continuous CO 2 and CO measurement in Paris and the Plateau of Saclay (20 km south-west of Paris, Sect.2).In addition, air was sampled in flasks for 14 CO 2 analysis (Sects.2.2.2 and 3.1).Radiocarbon and stable CO 2 isotope measurements were used to compute the respective contributions of fossil fuels and biospheric CO 2 to the overall CO 2 mole fraction in Paris (Sects.3.2 and 3.3).These direct estimations of CO 2 ff were used to evaluate the potential of alternative proxies to radiocarbon such as CO and NO x (Sect.3.4).

Paris
Paris is the largest city in France; it has 2.2 million inhabitants (source: INSEE, January 2009) and a population density of 21 196 inhabitants per square kilometre, which is one of the highest densities in Europe.Paris and its suburbs include several industrial and transportation hot spots, although numerous industrial activities were removed from the Paris agglomeration in the 1960s.Two sites were instrumented for atmospheric measurements within the city centre during the MEGAPOLI campaign (see Fig. 1).The Laboratoire de l'Hygiène de la Ville de Paris (LHVP), located at 48 • 50 N, 2 • 21 E and at 62 m above sea level (a.s.l.), lies within the southern part of Paris (13th district).This laboratory is located on the edge of the Parc de Choisy, approximately 100 m from a main road.A large number of instruments were deployed at the LHVP station to monitor atmospheric compounds such as aerosols, volatile organic compounds (VOCs), ozone, nitrogen oxides (NO x = NO + NO 2 ) and meteorological parameters.A cavity ring-down spectrometer (CRDS) was installed for continuous atmospheric carbon dioxide (CO 2 ) and carbon monoxide (CO) measurement, and a flask sampling unit was also installed.All inlet lines were positioned on the roof of the LHVP, 15 m above the ground level (a.g.l.).
A second site was equipped for atmospheric measurements at the QualAir station in Jussieu (http://qualair.aero.jussieu.fr/), a campus of University Pierre et Marie Curie, 48 • 51 N, 2 • 21 E and 38 m a.s.l., which is located in the city centre of Paris.This station is approximately 2 km from the LHVP.This site was already equipped with several atmospheric monitoring instruments (including those for CO measurement) and was adapted to use larger instruments than those at the LHVP.An automatic flask sampler (Neubert et al., 2004) and a tunable diode laser (TDL) spectrometer, developed and operated by the Laboratory of Molecular Physics for Atmosphere and Astrophysics (LPMAA, Croizé et al., 2010), were installed for continuous CO 2 and δ 13 CO 2 measurements.The two inlet lines were located on the roof of the building, 22 m a.g.l.
We used the meteorological parameters of the QualAir station (wind speed and wind direction) as these figures are higher for this station because wind shielding is less important and measurements at this site are more representative than at the LHVP station.During the campaign, wind directions with approximately equal frequency from all four directions and a mean wind speed of 2.7 m s −1 have been observed.

The Plateau of Saclay
The Plateau of Saclay is a semi-urban flat area that is located approximately 20 km south-west of Paris.The plateau contains atmospheric measurement stations at Gif-sur-Yvette and SIRTA, which are 5 km apart (see Fig. 1).The sites are mainly surrounded by agricultural fields, forests and villages and can be described as suburban.
Gif-sur-Yvette station (Gif station), 48 • 42 N, 02 • 09 E and 160 m a.s.l., is part of the LSCE and belongs to the RAM-CES team.The closest villages and small towns are Saint-Aubin (673 inhabitants) and Gif-sur-Yvette (21 352 inhabitants), which are located 500 m north-west and 1 km south of the station, respectively (Yver et al., 2009).The station is equipped with a gas chromatograph (GC) system that performs continuous measurements of CO 2 , CH 4 , N 2 O, SF 6 , CO and H 2 (Pépin et al., 2001;Yver et al., 2009;Lopez et al., 2012).The analytical laboratory is also equipped with an isotope ratio mass spectrometer (IRMS) that is dedicated to the atmospheric measurement of CO 2 isotopes: δ 13 C and δ 18 O (WMO-GAW, 2005).The inlet lines are located on the roof of the laboratory building, 7 m a.g.l.
The Instrumental Site of Research by Atmospheric Remote Sensing observatory (SIRTA, Haeffelin et al., 2005), located at 48 • 43 N, 02 • 12 E and 156 m a.s.l., in Palaiseau, is approximately 5 km east of Gif-sur-Yvette station.During the winter campaign, NO x were monitored by the Interuniversity Laboratory of Atmospheric Systems (LISA).
During the 2010 winter campaign, two primary wind regimes were observed: winds from the north (0-45 • ) that blew 22 % of the time and transported polluted air masses from Paris, and winds from the west (225-270 • wind sector) that also blew 22 % of the time.The mean wind speed in both cases was 5.4 m s −1 .During the rest of time, the wind regimes were well distributed among the remaining sectors except for the eastern one (45-90 • ), which was underrepresented (with winds blowing from that direction less than 3 % of time).

Trainou station
Trainou station (47 • 57 N, 02 • 06 E, 131 m a.s.l.) is a site located 100 km south of Paris (see Fig. 1) that is mainly surrounded by agricultural fields and forests.The nearest city is Orléans, which is 15 km south-west of the site and has 116 000 inhabitants.Trainou can thus be considered a rural site, especially in comparison to the Paris area.Air inlet lines are located at 5, 50, 100 and 180 m a.g.l. on a television tower (TeleDiffusion de France).Continuous GC analysis for CO 2 , CH 4 , N 2 O, SF 6 , CO and H 2 mole fractions have been conducted since 2007 (Messager, 2007;Yver et al., 2010;Lopez et al., 2012).The analysed meteorological variables from the data assimilation system of the European Centre for Medium-Range Weather Forecasts (ECMWF) were used for Trainou station.During the campaign, the station experienced approximately the same wind regimes as Gif-sur-Yvette station, with 24 % of the wind coming from the northnorth-east (0-45 • ) and 25 % of the wind from the west-southwest (225-270 • ).The mean wind speeds were 4.6 m s −1 and 4.3 m s −1 for the two sectors, respectively.

In situ analysers
The cavity ring-down spectrometer (G1302 Picarro, Wastine et al., 2009) that was installed at the LHVP analyses CO 2 , CO and water vapour (H 2 O) mole fractions with a sampling frequency of 1 Hz.This instrument was calibrated for CO 2 and CO before and after the campaign with four calibration tanks based on the WMO-X2007 scale.During the campaign, two high-pressure cylinders were analysed for 30 min every 10 h.One cylinder was used as a working standard to correct for potential temporal drift by the analyser, and the second cylinder was used as a target gas to evaluate repeatability.During the measurement period, a systematic drift in the CO 2 raw data of 0.08 ppm per month was observed, whereas no significant variation in the CO measurements was detected.After drift corrections for the entire campaign, the target gas showed a repeatability level (1 sigma) of 0.03 ppm for the CO 2 measurements and 10.1 ppb for the CO measurements when 1 min averages were used.Because the ambient air was not dried prior to the CRDS analysis, a water vapour correction for CO 2 was used together with a constant correction (−15 ppb) for CO measurements, which was empirically determined during several weeks of testing in our laboratory in Gif-sur-Yvette (Kaiser et al., 2010).
At Jussieu, a tunable diode laser spectrometer has been installed and is operated by the LPMAA to continuously monitor the atmospheric total CO 2 mole fraction and δ 13 C for four consecutive days in February (Croizé et al., 2010).The analysis technique is based on the absorbance correlations between a sealed reference cell and the sample cell.More details regarding the instrumental set-up and calibra-tion strategy used are provided in Croizé et al. (2010).All of the reference and calibration cylinders used in this instrument were calibrated at the central laboratory of LSCE by GC and IRMS.
The two gas chromatographs (Agilent HP-6890 and HP-6890N) located at Gif-sur-Yvette (suburban site) and Trainou (rural site) were optimised for semi-continuous atmospheric measurements.The GC are equipped with two detectors to simultaneously analyse CO 2 and CH 4 with a flame ionisation detector (FID, Messager, 2007) and N 2 O and SF 6 with an electron capture detector (ECD, Lopez et al., 2012).In 2006 at Gif-sur-Yvette and in 2009 at Trainou, two Peak Performers (PP1) were coupled to the GC for additional analysis of CO and H 2 using a reduction gas detector (RGD, Yver et al., 2009).In addition to continuous measurements, the GC at Gif-sur-Yvette has been optimised for flask analysis and tank calibration.All of the GC measurements (continuous, flasks and tanks) were calibrated on the WMO-X2007 scale.
The isotope ratio mass spectrometer (Finnigan MAT 252) located at Gif-sur-Yvette measures the isotopic composition of atmospheric CO 2 (δ 13 C and δ 18 O in CO 2 ) from flasks (Schmidt et al., 2005).An automated sampling line was setup according to the "BGC-Airtrap" design as described in Werner et al. (2001) to cryogenically separate CO 2 (along with N 2 O) from the other air constituents.The "airtrap" is coupled to the IRMS inlet system and used to analyse the air sampled in the flasks.During the campaign, the IRMS was used for semi-continuous analysis of the atmospheric CO 2 isotopes at Gif-sur-Yvette during a 3 day period.Each analysis sequence consisted of injections of an ambient air standard followed by a target gas and up to 10 flasks, with another ambient air standard injected at the end.The carbon dioxide isotopic data are reported in the δ notation (Eq.1); R is the ratio of the heavy isotope to the light one.The δ 13 C values are referenced to the VPDB scale.
In addition to CO 2 and CO, NO x have been monitored at the LHVP and SIRTA stations.NO x were measured continuously using two chemiluminescent analysers (AC31M, Environment SA) that were operated by the LSCE at the LHVP and by the LISA at SIRTA.A summary of the instrumentation used for this study during the MEGAPOLI campaign is given in Table 1.

Flask sampling strategy and radiocarbon measurements
In addition to continuous measurements, flask sampling was carried out at the LHVP and Jussieu stations.At the LHVP, a manual sampler was used to fill 43 pairs of 1 and 2 L glass flasks.Ambient air was dried using a magnesium perchlorate cartridge.We flushed the flasks for 10 min before pressurisation.The flasks were filled in the morning during the traffic rush hour on six days in February.At Jussieu, we filled 35 flasks of 2.5 L volume every 2 h for two complete days (9-10 and 14-15 February) and during one night (11-12 February).The samples were dried by passing through a water trap maintained in an ethanol bath at −60 • C as described by Neubert et al. (2004).The flasks are automatically flushed for 2 h before pressurisation.All flasks were analysed using a GC and IRMS to provide additional trace gas information and to independently determine the quality of the continuous CRD (CO 2 and CO) and TDL (CO 2 , 13 C) spectrometer measurements.
We selected 23 flasks (17 from LHVP and 6 sampled at Jussieu) for high-precision 14 CO 2 analysis.Flasks were chosen based on the atmospheric CO 2 mole fraction, and days with large temporal CO 2 gradients were favoured in order to maximise the 14 CO 2 signal with respect to measurement uncertainties (corresponding to ∼ ±1 ppm CO 2 ff).Compared to the whole campaign, the selected flasks represent well the typical diurnal gradients.The 14 CO 2 samples were prepared at the University of Colorado, Boulder, and measurements were performed via accelerator mass spectrometry (AMS) at the University of California, Irvine, following Turnbull et al. (2007).The radiocarbon values are expressed in notation (Stuiver and Polach, 1977).
As shown in several studies (Levin et al., 2003;Gamnitzer et al., 2006), the amount of 14 CO 2 in atmospheric samples can be used to derive the contribution of fossil fuel CO 2 to the measured atmospheric CO 2 mole fraction.The measured CO 2 mole fraction (CO 2 meas) is assumed to consist of a background CO 2 (CO 2 bg), a biospheric component (CO 2 bio) and a fossil fuel (CO 2 ff) component: (2) Each component has a characteristic 14 C value. 14Cff is equal to −1000 ‰ (free of 14 C), while 14 Cbg can be measured remotely (or assumed). 14Cbio is equivalent to 14 Cbg as 14 Cbio includes a correction for massdependent fractionation.Equation ( 3) is then derived as follows: CO 2 ff = CO 2 meas × 14 Cbg − 14 Cmeas 14 Cbg + 1000 .
In Eq. ( 3), other contributions to the 14 CO 2 budget such as the terrestrial disequilibrium isoflux due to heterotrophic respiration and 14 CO 2 production from nuclear power plants or the reprocessing sector have been neglected.Turnbull et al. (2006) showed that in winter, isoflux corresponds to ∼ 0.2 ppm CO 2 ff, which is negligible in comparison with the large CO 2 ff signal we observe in Paris (see Sect. 3.2).In France, the main production of 14 CO 2 by the nuclear sector is from La Hague (a nuclear waste treatment plant, located approximately 300 km west of Paris), which emits an amount of 14 C about 100 times higher than the French nuclear power plants.Using a diffusion model, no influence of air masses from La Hague during the four days of flasks sampling have been observed at Paris.

Comparison of continuous measurements with flasks sampling
The flask results were compared with the corresponding 2 min averages from the continuous analysers.At LHVP, the mean difference between 36 pairs of flasks and the CRDS analyser data was −0.03 ± 3.33 ppm for CO 2 and −1.9 ± 31.1 ppb for CO, indicating good agreement in both cases.
The comparison between the flasks and the TDL spectrometer for the CO 2 mole fraction and δ 13 C also indicates good agreement, with mean differences between the flasks and in situ measurements of 0.07±2.33ppm (N = 15) for CO 2 and 0.29 ± 0.20 ‰ (N = 16) for δ 13 C.The substantial standard deviation is likely related to the strong temporal variability in Paris, which often reaches several ppm for CO 2 and 10 ppb for CO in just a few minutes.The 0.29 ‰ mean difference observed in the δ 13 C analysis may have been caused by an underestimated correction of the TDL spectrometer measurements.The correction aims to account for the small temperature difference between the atmospheric air and the calibration gas after the atmospheric air has been passed through a cold trap to remove water vapour (Croizé et al., 2010).
The mean mole fraction differences for the CO 2 and CO measurements derived from continuous measurement versus flask sampling are consistent with the WMO-GAW recommendation for background measurements (WMO-GAW, 2009).The objective is to maintain a network intercompatibility level of ±0.1 ppm and ±2 ppb for CO 2 and CO, respectively.The δ 13 C figures do not meet the WMO-GAW recommended maximum difference of ±0.01 ‰.The substantial difference is due to instrumental accuracy.However, this difference is not likely to impact the conclusions of this study, which shows that large signals are associated with sampling close to sources in a polluted area.

The observation of CO 2 , CO and NO x in Paris, at the Plateau of Saclay and at Trainou station
In the Paris region, CO 2 is mainly emitted by anthropogenic sources through the combustion of fossil fuel and biomass (biofuel and wood), although natural sources such as soil respiration also contribute to a degree that remains highly uncertain.Carbon monoxide is often considered as a reliable combustion tracer.It results from fossil fuel and biomass burning and has an atmospheric lifetime of approximately 2 months (assuming an OH value of 10 6 molecules cm −3 , Atkinson et al., 2006).Nitrogen oxides are typical markers of traffic exhaust emissions and have a lifetime of approximately 9 h (Atkinson et al., 2006).However, due to its short lifetime, NO x cannot be transported over long distances and consequently can only be used as tracers for local emissions.thropogenic sources.Trainou yielded the lowest values and synoptic variability, which was expected given that we sampled at 180 m a.g.l. and the rural site likely has weaker local CO 2 and CO sources.
The wind regimes were classified based on back trajectories obtained using the Lagrangian particle dispersion model, FLEXPART (version 8.2), which is described by Stohl et al. (2007).In Fig. 2, the blue areas correspond to times when the air masses that influenced the sites originated from the western sector (180 to 360 • ).We refer to this as the "oceanic regime" using the MEGAPOLI project nomenclature.The green areas correspond to times when air masses came from eastern Europe (0 to 180 • ) and crossed Germany or Benelux before arriving at the LHVP, the Plateau of Saclay and Trainou station.We refer to this as the "continental regime".The data collected on days for which air mass back trajectories could not be clearly classified (uncoloured ones) are not considered in the analysis that follows.Figure 3 presents the CO source contributions at LHVP for typical continental (Fig. 3a) and oceanic (Fig. 3b) air masses.Table 2 summarises the average mole fractions of CO 2 , CO and NO x for the two different regimes.On average, the three sites show higher CO 2 and CO mole fractions during the continental regime.According to a Student's t test, the average mole fractions for the continental and oceanic regimes are statistically distinct at all three sites.Carbon  dioxide and carbon monoxide, with lifetimes of 100 yr and 2 months, respectively, can be transported over longer distances than NO x can.Thus, higher CO 2 and CO mole fractions are associated with the continental regime because the air masses are progressively charged with pollutants on their trajectory to Paris.When the main wind direction shifts from east (continental) to west (oceanic), a decrease in CO 2 and CO mole fractions is observed.Nitrogen oxides cannot be transported over long distances; thus, at the LHVP and the Plateau of Saclay, NO x mole fractions are similar for the oceanic and continental regimes.
Figure 4 shows the mean diurnal cycles of CO 2 , CO and NO x at the three stations for both weekdays (in red) and weekends (in black) for the entire measurement period.The mole fractions of trace gases and the mean amplitudes of the diurnal cycles are both smaller at weekends than on weekdays.The mean CO 2 , CO and NO x diurnal cycles show two peaks at LHVP, in the morning (07:00 to 09:00 UTC) and at the end of the afternoon (18:00 to 20:00 UTC).At the Plateau of Saclay, the afternoon peak is weak and is only visible for CO and NO x .The mean diurnal cycle of the traffic flow in the IdF region during weekdays for the year 2010 is presented along with the NO x diurnal cycles at the LHVP (Fig. 4).This cycle exhibits the same pattern as the NO x , CO 2 and CO diurnal cycle, with maxima in the morning and at the end of the afternoon, which suggests the influence of traffic emissions on NO x measurements.Traffic counts, provided by the Directorate of Ile-de-France roads (DiRIF), are recorded for the main roads of the IdF region.The diurnal cycles of CO 2 and CO at Trainou (180 m a.g.l.) present a single peak at noon on weekend and weekdays primarily reflecting the diurnal dynamics in the planetary boundary layer (PBL) as shown by Lopez et al. (2012) for radon-222 and nitrous oxide.
In Fig. 5, the diurnal cycles of CO 2 , CO and NO x that are recorded at the LHVP are plotted as functions of the air mass origin.The vertical lines are the standard deviation for each hour.As previously explained, the mean concentrations are generally higher when the air masses originate from the east, but the amplitudes of the morning peaks for the three species are similar for the two wind regimes.In contrast, the afternoon peaks in the oceanic wind regimes are more marked for CO 2 , CO and NO x than are those of the continental regime.In the oceanic wind regime, the LHVP is more  sensitive to local pollution, which increases between 16:00 and 18:00 UTC due to the traffic rush hour and the decrease in the PBL height.The signals decrease at the end of traffic rush hour as the atmosphere is refreshed by relatively clean air masses from the west.This second peak is less marked during the continental regime because the atmosphere is subsequently refreshed by the polluted air masses coming from east of Paris.
At the LHVP, non-methane hydrocarbons including benzene and toluene were monitored during the whole campaign using gas chromatography equipped with an ionisation flame detector (see Gros et al., 2011, for more details).Benzene and toluene have lifetimes of approximately 11 and 2 days, respectively (assuming an OH value of 10 6 molecules cm −3 , Atkinson et al., 2006), and are typically emitted by the same sources.The toluene to benzene ratio is therefore often used to determine the photochemical age of an air mass: the lower the ratio, the older the air mass (Roberts et al., 1984;Warneke et al., 2007).This ratio may therefore be used to specify whether the observed pollution originated from a local source or was subject to long-range transport.On 4 and 5 February (oceanic regime), the ratio of toluene to benzene was recorded as approximately 5, whereas on 9 and 10 February (continental regime), the ratio decreased to 2. This result indicates that during the oceanic regime, air masses arriving in Paris were primarily affected by local pollution, whereas during the continental regime, long-range pollution sources were an additional factor.

Calculating the fossil fuel CO 2 contribution using radiocarbon measurements
In this study, Eq. ( 3) (see Sect. 2.2.2) is applied to the flasks sampled at the LHVP and Jussieu station to quantify the contribution of fossil fuel CO 2 to the ambient air in Paris.
The background value of 14 Cbg is derived from 14 CO 2 measurement at Mace Head (Ireland, 53 • 19 N, 9 • 54 W, 25 m a.s.l.), which is considered a background site for Europe.In February 2010, the monthly mean 14 CO 2 and CO 2 values at Mace Head (wind selected for the oceanic sector, Bousquet et al., 1996) were 43.7 ‰ (I.Levin, personal communication, 2011) and 393.5 ppm, respectively.The analysis of the flasks yielded 14 Cmeas between −54.04 and +15.59 ‰ and a resulting CO 2 ff contribution of 10.9 to 41.9 ppm relative to the background air at Mace Head. Figure 6 presents an example of a typical measurement day, 4 February 2010, at the LHVP.The 1 min averaged CO 2 mole fractions obtained via continuous measurement are plotted in orange, and the CO 2 mole fractions obtained from the flask measurements are indicated with red crosses.The background CO 2 mole fraction at Mace Head during the maritime background conditions (Bousquet et al., 1996) for February 2010 is plotted in blue.
Combining the CO 2 and 14 CO 2 measurements from the individual flask samples permits us to determine the contribution of CO 2 ff and CO 2 bio to the observed CO 2 mole fraction following Eq.( 2): see Fig. 6.Fossil fuel CO 2 increases by 22.7 ppm from 07:15 to 08:15 UTC to reach a maximum of 41.9 ppm.Subsequently, we observe a small decrease until 12:00 UTC followed by a strong decrease (17.0 ppm) to 17.2 ppm at 13:00 UTC.The NO x cycle (not shown) follows the same pattern, suggesting that local traffic contributes to Fig. 6.A comparison between the continuous measurements for CO 2 (in orange) and the data obtained using flask sampling (red crosses) at the LHVP site are presented for 4 February 2010 on the left axis.We added the background level of CO 2 as measured at Mace Head (Ireland), which is represented by the blue horizontal dashed line.The figures for CO 2 ff and CO 2 bio, as derived from flask sampling, are represented on the right axis by the black and green curves, respectively.The left and right axes have the same CO 2 amplitude.
the CO 2 ff signal in the centre of Paris.The morning CO 2 ff increase is due to the morning rush hour, and the subsequent decrease indicates the reduction in traffic as well as the PBL development.The figures for CO 2 bio are lower (0-10 ppm) and exhibit less variation than those of CO 2 ff.The same diurnal cycles are observed for CO 2 ff and CO 2 bio for the four sampling days at the LHVP and the single sampling day at Jussieu.Table 3 summarises the daily averages for CO 2 ff and CO 2 bio for each sampling day.The fossil fuel CO 2 enhancement with respect to the background varies from 16.2 to 32.8 ppm, and the biospheric CO 2 contribution varies from 4.5 to 10.2 ppm.The differences in the CO 2 ff and CO 2 bio mole fractions at the LHVP and Jussieu sites as they were observed on 10 February are related to the sampling period and location.
The contributions of the two CO 2 sources (fossil fuel and biogenic) to the CO 2 meas variability show that the majority of this variability is due to the injection of CO 2 ff into the atmosphere.On average, 77 % of the observed signal is caused by CO 2 ff (Table 3), mainly from road traffic, heating and the residential sectors.However, despite Paris being a megacity, the biospheric contribution of CO 2 fluxes was significant (23 % on average), even in winter.
This biospheric contribution of approximately 23 % can be attributed not only to net plant and soil respiration but also to the use of biofuel and to human respiration, which cannot be identified using 14 CO 2 measurements because these sources have been fixed or metabolised recently and have isotopic signatures that are similar to that of the ambient atmosphere.In this paragraph we attribute a percentage to the different biospheric sources in Paris using emission inventories and literature values.According to an AirParif emission inventory, CO 2 emissions from wood burning used for residential heating contribute less than 1 % of the total emissions in Paris and likely do not contribute significantly to the observed CO 2 bio enhancement.By 2010, the biofuel ratios in gasoline and diesel increases to approximately 7 % in France, which suggests that road traffic sources contribute significantly to the CO 2 bio results.In France, most of the biofuel is produced from rape, wheat and sugar beats (C3 plants), having a δ 13 C signature of approximately −28 ‰.AirParif estimates that the contribution of CO 2 from traffic is 17 % of the total fossil fuel CO 2 emissions for January 2008 after the emissions associated with biofuel in gasoline and diesel have been removed.According to our mean results for CO 2 ff and CO 2 bio as presented in Table 3, this contribution corresponds to an average CO 2 ff of 5 ppm from traffic.Because gasoline and diesel contain up to 7 % biofuel, the CO 2 bio enhancement due to traffic sources is 0.4 ppm, or approximately 12 % of the CO 2 bio in Paris during the winter.A study by Turnbull et al. (2011) determined that the use of biofuels contributed up to 1 ppm of the positive CO 2 bio signal when gasoline contained approximately 8 % ethanol in California.
A study by Ciais et al. (2007) shows that CO 2 emissions from human respiration in densely populated cities can reach 20 % of fossil fuel CO 2 emissions.Koerner and Klopatek (2002) show that human respiration is 500 kg CO 2 per year per person on average.With a population of 2.2 million in the centre of Paris, human respiration in the centre may be responsible for CO 2 emissions of approximately 1100 kt yr −1 .By comparing this value to the reported fossil fuel CO 2 sions of 7218 kt for the year 2005 in Paris reported by Air-Parif, human respiration contributes 15 % of the CO 2 ff emissions; this percentage is similar to the value reported in Ciais et al. (2007).Human respiration should be responsible for approximately 4 ppm of the CO 2 enhancement, and may therefore account for approximately 50 % of the CO 2 bio calculated.
Here we can conclude that the biogenic sources can be attributed to 38 % of the net plant and soil respiration, 50 % to human respiration and 12 % to the use of biofuel.These values are only representative for our sampling sites at the LHVP and Jussieu in February 2010.

Separation of fossil fuel CO 2 sources using δ 13 C measurements
Several studies (Zondervan and Meijer, 1996;Meijer et al., 1996;Djuricin et al., 2010) have used 13 C / 12 C ratios in combination with 14 C measurements to quantify the fractions of emissions by different fossil fuel sources.Tans (1981) estimated that the release of CO 2 via anthropogenic combustion implies average worldwide δ 13 C emissions for natural  gas (δ 13 Cgas), liquid (δ 13 Cliq) and solid (δ 13 Csol) fuel of −41.0, −26.5 and −24.1 ‰, respectively.More recently, Andres et al. (2000) shown that δ 13 C signatures depend on the geographic origin of fossil fuels and can range from −19 to −35 ‰ for liquids and from −20 to −100 ‰ for natural gas.A study by Widory and Javoy (2003) determined δ 13 C from the main pollution source in Paris by directly measuring exhaust gases.The authors determined that the mean δ 13 C source ratio for natural gas, liquid fuel and coal is −39.1 ± 1.1, −28.9 ± 0.4 and −24.8 ± 0.4 ‰, respectively.Based on the atmospheric δ 13 C and CO 2 measurements, we estimate the signatures of the main δ 13 C sources, using a Keeling plot (Keeling, 1958(Keeling, , 1961;;Pataki et al., 2003) for the daily observations.Assuming mass conservation and following Eq.( 2) (see Sect. 2.2.2), we derive Eq. ( 4), where CO 2 s represents the sum of the CO 2 sources (i.e.CO 2 s = CO 2 ff + CO 2 bio).Combining Eqs. ( 2) and (4), we obtain Eq. ( 5).We compute the relationship between δ 13 Cmeas and the inverse mole fraction of the corresponding CO 2 using a least-squares method following Pataki et al. (2003).The intercept corresponds to the mean signature of the source: δ 13 Cs (i.e. the last term of Eq. 5).These values have been corrected for biospheric δ 13 C emission, assuming that δ 13 Cbio = −24.7 ‰ (Bakwin et al., 1998), resulting in δ 13 Cff.
δ 13 Cmeas×CO 2 meas=δ 13 Cbg×CO 2 bg+δ 13 Cs×CO 2 s (4) Based on the data from the flasks sampled during the entire measurement period, δ 13 Cff at LHVP and Jussieu was equal to −36.1 ± 2.7 ‰ and −36.2 ± 1.1 ‰, respectively, where the errors are given by the uncertainty in the intercept from Eq. ( 5).According to AirParif, the use of coal in the Paris region accounts for less than 1 % of the total anthropogenic CO 2 emission; thus, we will neglect this source.Using the δ 13 Cgas and δ 13 Cliq values measured by Widory and Javoy (2003) and the δ 13 Cff source values from the Keeling plots yielded percentages of 70 ± 6 % for natural gas and 30 ± 3 % for liquid fuel.The emission inventory for Paris for 2008 that was provided by AirParif (described in Sect.3.5) estimates the contribution of natural gas to be 37 % of the total CO 2 emissions.
We surmise that the higher ratio of gas consumption derived using the atmospheric method (70 ± 6 %) compared to the AirParif emission inventory (37 %) likely arose from the spatial separation of automotive and residential sources and the lack of complete mixing prior to the sampling at our sampling location.We conducted our sampling during only a few days in winter, whereas the fraction provided by AirParif takes into account data from the entire year.We expected a higher fraction of gas consumption during the winter due to enhanced heating use.Pataki et al. (2006) conducted δ 13 CO 2 continuous measurements over Salt Lake City (Utah, USA) during the early winter of 2004-2005.Using the same approach as presented in this section, they found a natural gas contribution between 30 and 40 % during morning rush hour and 60 and 70 % at pre-dawn.Newman et al. (2008) analysed δ 13 CO 2 in flasks for the years 2002-2003 in Pasadena (Los Angeles Basin, California, USA).They found a Keeling intercept of −29.9 ± 0.2 ‰, which is less depleted than our value (−36.1 ± 2.7 ‰).This shows a stronger annual influence of liquid fuel use in the total CO 2 emission in Pasadena, compared with the winter in Paris.Administration of the US Department of Energy indicates that that 62 % of the state of California's energy from fossil fuel is from petroleum products and 37 % from natural gas.

Constructing a continuous time series of fossil fuel CO 2
Precise 14 CO 2 measurements are difficult to perform and are only possible on discrete samples.Therefore, there are no direct methods of obtaining a continuous record of fossil fuel CO 2 .To derive such a record for the entire campaign (defined here as FFCO 2 ), we used two potential proxies: CO (see Gamnitzer et al., 2006;Levin and Karstens, 2007) and NO x .To our knowledge, this is the first time that NO x are used as proxy to derive the CO 2 fossil fuel contribution.Both tracers are known to be typical proxies for fossil fuel combustion, but the emission ratios with respect to CO 2 ff vary as a function of the combustion source.The daily ratios of each selected proxy of CO 2 ff for the four days when we took 14 C measurements at the LHVP site were used to construct the FFCO 2 curve.To compute the ratios R CO and R NO x , defined here as CO CO 2 ff × 1000 and NO x CO 2 ff × 1000, we subtracted the background mole fraction from the ambient measurements.To ensure consistency, we chose to use the background CO mole fraction from the winter 2010 at Mace Head, which was equal to 121 ppb (Derwent et al., 2001, http://agage.eas.gatech.edu/Stations/macehead.htm).For NO x background, we used a value of zero because of its short lifetime (close to 9 h).The CO 2 ff terms do not take into account the contribution of CO 2 from biofuel use.To determine the FFCO 2 values, Eqs. ( 6) and ( 7) were used.
The emissions ratios for each of the four days of measurements at the LHVP are presented in Table 4 along with their respective coefficients of determination (r 2 ).We observe that the ratios are functions of the wind regime (Table 4), with higher values on 4 and 5 February (oceanic regime) than on 9 and 10 February (continental regime).An average ratio for each air mass regime is applied to the appropriate segments of the continuous measurements to compute the FFCO 2 for each regime.For the few days without clearly identified air mass origin, we use the mean value from the two synoptic situations (Table 4).
In the upper panel of Fig. 7, we present the two curves inferred from R CO and R NO x : FFCO 2 (CO) and FFCO 2 (NO x ), respectively.The estimated FFCO 2 is greater during the continental wind regime: up to 80 ppm at three different times.In contrast, during the oceanic regime, the FFCO 2 mole fractions never exceed 40 ppm.At the bottom of Fig. 7, the FFCO 2 (CO) and FFCO 2 (NO x ) mean diurnal cycles are plotted.These plots show the same pattern: a peak during the morning in the rush hour and a second peak at the end of afternoon.The FFCO 2 (CO) and FFCO 2 (NO x ) mean mole fractions are 20.6 and 18.7 ppm, respectively.Differences are

Oceanic regime
Hour CO 2 ff flasks Fig. 7. Top: time series for FFCO 2 during the MEGAPOLI campaign calibrated using CO and NO x as proxies.The purple rhombuses correspond to the CO 2 ff from the flasks sampling.Periods during which the air masses originate from the west are highlighted in blue, and periods during which the air masses originate from the east are highlighted in green.No colour is attributed on the days without a clear air mass origin (see Fig. 2).Bottom: mean diurnal cycles of FFCO 2 (CO) in black and FFCO 2 (NO x ) in red at the LHVP station from 15 January 2010 to 19 February 2010.observed during the evening and over night (from 17:00 until 06:00 UTC) because of the short lifetime of NO x relative to the lifetime of CO.
A study by Vogel et al. (2010) shows that R CO has a diurnal cycle because anthropogenic CO 2 (and CO) fluxes are subject to strong diurnal variation.During the winter, the authors found a peak-to-peak amplitude of 8 ppb ppm −1 , with the lowest ratio at 08:00 UTC and the highest at 18:00 UTC.We assume a constant ratio during the day in this study.

Comparison of R CO and R NO x ratios with emission inventories
In this section, we compare the R CO and R NO x emission ratios measured in Paris with those derived from emission inventories.Four different emission inventories are presented with different spatial and temporal resolutions.The reference years for the inventories vary from 2005 to 2010, which makes it difficult to compare the ratios directly.For France as a whole, CITEPA has calculated the GHG emissions on a yearly basis since 1960.In addition, AirParif provides a regional inventory for the IdF with a spatial resolution of 1 km × 1 km.(Olivier et al., 2001;Olivier and Berdowsky, 2001).The spatial resolution of IER and Air-Parif allowed for us to extract the exact grid cells that corresponded to Paris, whereas for EDGAR 4.2, we chose the grid cell that included most of the Paris area.
The total ratios for each inventory are presented in Table 5 for different reference years and spatial and time resolutions.For France as a whole, CITEPA estimates a higher total R CO ratio than does EDGAR 4.2, whereas the total ratios are similar for R NO x .For Paris, EDGAR 4.2 shows much lower total R CO ratios (4.3) than does the IER inventory (14.6).The R NO x ratios vary substantially across the different emission inventories; they are 3.9 (EDGAR 4.2), 3.0 (IER) and 2.1 (AirParif).These differences are due to discrepancies in the spatial source distribution for Paris and to the different ratios used for the same source sectors in different inventories.These large differences between inventories suggest that more work needs to be done to provide more consistent estimates.
In Table 5, we also summarised the percentage contributions of the three main source sectors (road transport, residential and industry) as well as the R CO and R NO x ratios for the different sectors for the different inventories.For France as a whole, the R CO and R NO x ratios stayed fairly constant between 2005 and 2010 except for the road transport ratio.The R CO ratio for road transport decreased by a factor of 2 in five years.This decrease is primarily a function of the environmental policy applied to road vehicles since 1970 and is also an effect of the adoption and improvement of catalytic converters for vehicles since 1993, which has reduced NO x and particularly CO emissions.
We compared the ratios determined using our top-down atmospheric approach (Table 4) with the available inventories (Table 5).We focus first on the R CO ratio and then on the R NO x ratio.During oceanic regimes, our mean R CO ratio of 13 is similar to the ratios from IER and CITEPA (Table 5).These high ratios are representative of the residential sector.This result is consistent with the ratio of benzene to toluene that we found in Sect.3.1, which indicates that our samples are sensitive to local emissions during the oceanic regime.Our R CO ratio for the continental regime is 8.5 and lies between the IER and CITEPA estimates and those of EDGAR 4.2.According to the AirParif inventory, the north and north-eastern areas of Paris have relatively significant CO 2 emissions from industrial sources (approximately 20 % of anthropogenic CO 2 ) compared with southern Paris (less than 5 % of anthropogenic CO 2 ).This result is in agreement with the ratio that we found for the continental regime, which indicates that the contribution of the road transport and industry sectors, which have low R CO ratios, are stronger.Moreover, the ratio of benzene to toluene during this regime indicated that long-range transport had occurred, which would have permitted the emissions from Benelux and Germany to have a greater influence.These are industrialised areas that are characterised by relatively low R CO ratios relative to those of France (Gamnitzer et al., 2006).We noticed that EDGAR 4.2 found a much lower R CO in Paris than did other inventories and that this figure was also much lower than our estimate.This difference is most likely due to the low R CO ratio of 1.0 for the residential sector for the Paris grid cell relative to the high value of 15.1 for France.
Because of its short lifetime, the R NO x ratio is less sensitive to contributions from different sources.Nevertheless, the ratios we found are within the range of the inventory estimates.During the continental regime, NO x were destroyed during transport over the continent, which leads to lower R NO x values.In the oceanic regime, the high ratios reflect the contribution of the road transport sector near the station.The ratios derived with our atmospheric approach for the different regimes agree well with the emission inventories given the uncertainties of the different approaches.

Conclusions
This pilot study has demonstrated that measurement of CO 2 , CO and NO x mole fractions together with isotopic CO 2 measurements can help to better quantify the fossil fuels component in cities. Radiocarbon measurements using flask sampling allowed for us to directly derive the fossil fuel CO 2 mole fraction in Paris, which was approximately 20 ppm (on average), in February 2010.Of the total CO 2 enhancement in Paris during this period, 77 % can be attributed to anthropogenic sources (CO 2 ff) and 23 % to biospheric sources (CO 2 bio).Using values from the literature, we estimated that half of these biospheric emissions are due to human respiration and that the other half are due to soil respiration (38 %) and the use of biofuel in gasoline and diesel (12 %).
Making use of δ 13 C measurements, we were able to determine the contributions of natural gas and liquid fuel to anthropogenic CO 2 emissions, which amounted to 70 % and 30 %, respectively.This estimation was possible in Paris because there are only two sources of fossil fuel emissions; the use of coal can be neglected because it produces less than 1 % of the total CO 2 emission.This calculation would be even more powerful if we could reduce the errors generated by the isotopic source signature of natural gas, which varies according to the supply of natural gas (i.e.whether it originates in Siberia or the North Sea).During future campaigns, it would be helpful to measure the source signature of natural gas from the gas supply network on a regular basis.
Carbon monoxide and nitrogen oxides, which were monitored continuously, were successfully used as proxies to determine the anthropogenic CO 2 emissions over the entire campaign.These proxies have been calibrated against the CO CO 2 ff and NO x CO 2 ff ratios.We determined that the ratios change as a function of the air mass origin.Therefore, it was possible to use different ratios for the proxies depending on the air mass regime.The data for the CO CO 2 ff and NO x CO 2 ff ratios are consistent with the information from the inventories although there are sizeable variations across the inventories themselves.According to these inventories, during the oceanic regime, we are more sensitive to the emissions from the residential sector, where the CO CO 2 ff ratio is high.During the continental regime, that ratio decreases, showing increasing influence of the road transport and industrial sectors.

Fig. 1 .
Fig. 1.Location of the different sampling sites.

Figure 2 Fig. 2 .
Fig. 2.The CO 2 , CO and NO x mole fractions at the LHVP, Plateau of Saclay (poS) and Trainou 180 m a.g.l.(TR3).Periods during which the air masses originate from the west are highlighted in blue, and periods during which the air masses originate from the east are highlighted in green.No colour is used for the days without a clear air mass origin.
Figure 3a and Fig. 3b indicate which grid cells contribute most to our signals.They are also obtained with the FLEXPART model (version 8.2, Stohl et al., 2007) coupled with the EMEP emission inventory (European Monitoring and Evaluation Programme, http://www.emep.int/)for the reference year 2005.

Fig. 3 .
Fig. 3. Maps showing the typical CO source contributions observed for the continental (a) and oceanic (b) wind regime during the measurement campaign at LHVP.

Fig. 4 .
Fig. 4. Mean diurnal cycles of CO 2 , CO and NO x at the LHVP, Plateau of Saclay and Trainou sites for weekend and weekdays with the associated standard deviations.In the NO x plot for the LHVP, the mean diurnal cycle of vehicle flow during the weekdays for the year 2010 in Ile-de-France is added in orange (DiRIF).

Fig. 5 .
Fig.5.The CO 2 , CO and NO x diurnal cycles at the LHVP as a function of the air mass regime.The green curves correspond to the continental regime, whereas the blue curves correspond to the oceanic regime.The error bars represent the standard deviation of the mean.

Table 1 .
Technologies, models, locations and accuracy of instruments for the analysed species with the respective measurement dates.

Table 3 .
CO 2 ff and CO 2 bio emissions with their respective contributions derived at the LHVP and at Jussieu.

Table 4 .
CO  CO 2ff (R CO ) and NO x CO 2 ff (R NO x ) ratios with their respective coefficients of determination (r 2 ) and the number of points used for each sampled day and the corresponding values used for each regime (3 last lines).

Table 5 .
CO 2 contribution estimates by different inventories for the three main emission sectors.From the emission values, we computed the total ratios R CO and R NO x and also the ratios by sectors.