CO emission and export from Asia: an analysis combining complementary satellite measurements (MOPITT, SCIAMACHY and ACE-FTS) with global modeling

This study evaluates our current knowledge of pollution outﬂow from Asia using simulations from the LMDz-INCA global model combined with the available satellite observations of CO during the spring of 2005. CO is measured by several space-borne instruments using di ﬀ erent techniques, allowing complementary perspectives. The CO 5 observations from the MOPITT and SCIAMACHY nadir sounders, which provide vertically integrated information with excellent horizontal sampling, and from the ACE-FTS solar occultation instrument, which has limited spatial coverage but allows the retrieval of vertical proﬁles, are used. Combining observations from MOPITT (mainly sensitive to the free troposphere) and SCIAMACHY (sensitive to the full column) allow the 10 evaluation of the boundary layer CO. Underestimated emissions in eastern Asia, and possibly too low lofting in the model, lead to an underestimate of the export to the free troposphere, as highlighted by comparisons with MOPITT and ACE-FTS. Both instruments observe large trans-Paciﬁc transport extending from ∼ 20 ◦ N to ∼ 60 ◦ N, with high upper tropospheric CO observed by ACE-FTS above the eastern Paciﬁc (with values of 15 up to 300 ppbv around 50 ◦ N at 500 hPa and up to ∼ 200 ppbv around 30 ◦ N at 300 hPa). The low vertical and horizontal resolutions of the global model do not allow the simulation of the strong enhancements in the observed plumes. However, the transport patterns are well captured, and are mainly attributed to export from eastern Asia, with increasing contributions from South Asia and Indonesia towards the tropics. Additional 20 measurements of C 2 H 2 , C 2 H 6 and HCN by ACE-FTS provide further information on the plume history. C 2 H 2 and C 2 H 6 enhancements are well correlated with the CO plumes, indicating common sources and rapid trans-Paciﬁc transport. HCN observations show that the biomass burning contributes mainly at latitudes lower than ∼ 40 ◦ N.

observations from the MOPITT and SCIAMACHY nadir sounders, which provide vertically integrated information with excellent horizontal sampling, and from the ACE-FTS solar occultation instrument, which has limited spatial coverage but allows the retrieval of vertical profiles, are used. Combining observations from MOPITT (mainly sensitive to the free troposphere) and SCIAMACHY (sensitive to the full column) allow the 10 evaluation of the boundary layer CO. Underestimated emissions in eastern Asia, and possibly too low lofting in the model, lead to an underestimate of the export to the free troposphere, as highlighted by comparisons with MOPITT and ACE-FTS. Both instruments observe large trans-Pacific transport extending from ∼20 • N to ∼60 • N, with high upper tropospheric CO observed by ACE-FTS above the eastern Pacific (with values of 15 up to 300 ppbv around 50 • N at 500 hPa and up to ∼200 ppbv around 30 • N at 300 hPa). The low vertical and horizontal resolutions of the global model do not allow the simulation of the strong enhancements in the observed plumes. However, the transport patterns are well captured, and are mainly attributed to export from eastern Asia, with increasing contributions from South Asia and Indonesia towards the tropics. Additional

Introduction
Asia is one of the most important pollution source regions. While anthropogenic emissions from North America and Europe appear to be stabilizing or decreasing, emissions from Asia have increased rapidly in the past decade, becoming the largest contribution (Richter et al., 2005;Akimoto, 2003). Asian energy consumption has more than 5 doubled since the 1980s due to population growth, rapid industrialization and increasing transportation, causing significant enhancement in the emissions over this region (Streets et al., 2000(Streets et al., , 2006Ohara et al., 2007). In addition to the degradation of regional air quality, it is now recognized that the long range transport of pollution from Asia, both from fossil fuel and biomass burning, has a strong influence on the chem-10 ical composition of the troposphere over the Pacific Ocean and downwind continents (e.g. Wild and Akimoto, 2001;Jaffe et al., 1999Jaffe et al., , 2004Heald et al., 2003;Bertschi et al., 2004;de Gouw et al., 2004;Hudman et al., 2004). This issue has been the focus of a number of recent aircraft campaigns (TRACE-P in 2001 , ITCT-2k2 in 2002(Parrish et al., 2004, INTEX-B in 2006). Satellite measurements have also 15 been a valuable tool for evaluating Asian emissions (Martin et al., 2003;Arellano et al., 2004;Pétron et al., 2004;Heald et al., 2004) and monitoring trans-Pacific transport (Heald et al., 2003(Heald et al., , 2006Allen et al., 2004;Edwards et al., 2004;Hudman et al., 2004). Current global model simulations of carbon monoxide (CO) show a significant underestimates throughout the Northern Hemisphere compared to satellite observations from 20 the MOPITT/Terra instrument suggesting a missing source year round (Shindell et al., 2006), which could be explained by too low emissions in eastern Asia in the current inventories.
In this study, we explore the information on the Asian pollution outflow provided by the available satellite observations of CO using comparisons with simulations from the by several satellite borne instruments. Nadir looking remote sensors allow the retrieval of global CO distributions with a good spatio-temporal resolution, either from measurements in the thermal infrared spectral region, with MOPITT on board the Terra satellite since 2000(Edwards et al., 2004Deeter et al., 2003), AIRS on board Aqua since 2002(McMillan et al., 2005, TES on board Aura since 2004 (Rinsland et al., 2006a, 10 Luo et al., 2007, and IASI on board MetOp since 2006 , Clerbaux et al., 2007a, or from measurements in the near-infrared with SCIAMACHY on board ENVISAT since 2002(Buchwitz et al., 2004, Frankenberg et al., 2005Gloudemans et al., 2005). However, the nadir geometry provides vertically integrated information, implying limited vertical resolution. CO retrievals in the thermal infrared spectral range 15 are characterized by a maximum sensitivity in the free troposphere with low sensitivity in the boundary layer and less than 2 independent pieces of information on the vertical profile, while the near-infrared allow sensitivity down to the Earth's surface but provides only total column measurements.
Instruments using solar occultation, such as the ACE-FTS on board SCISAT since 20 2003 (Bernath et al., 2005;Clerbaux et al., 2005Clerbaux et al., , 2007b, or emission in a limb viewing geometry, such as MIPAS/ENVISAT  or MLS/Aura (Pumphrey et al., 2007), allow the retrieval of CO profiles with an improved vertical resolution, but with a poorer horizontal coverage than nadir instruments. Of particular interest for this work investigating tropospheric pollution is ACE-FTS, which allows probing down to 25 the middle troposphere in cloud free conditions (Clerbaux et al., 2005(Clerbaux et al., , 2007b. Several recent studies have shown the value of limb and solar occultation data for tropospheric chemistry studies (Dufour et al., 2007;Folkins et al., 2006;Rinsland et al., 2006b;Coheur et al., 2007;von Clarmann et al., 2007). In particular, the CO measurements Here, we examine the complementary information provided by the solar occultation measurements from the ACE-FTS and the nadir observations from MOPITT and SCIA-MACHY for the analysis of the export of pollution from Asia.
This analysis is focused on spring 2005. Two reasons motivated the choice of this time period. First, spring corresponds to the maximum in Asian outflow due to active cyclonic activity and strong westerly winds. During this season, the Asian pollution can be transported across the Pacific in 5 to 10 days (e.g. Jaffe et al., 1999;Liang et al., 2004) with the strongest transport in the middle troposphere (e.g. Bey et al., 2001;Stohl et al., 2002). Secondly, the mid-latitudes of the Northern Hemisphere were well covered by ACE-FTS during this time period. 15 After a brief description of the global model and a first validation against MOZAIC aircraft measurements, the satellite observations from ACE-FTS, MOPITT and SCIA-MACHY are presented, as well as an intercomparison of the available CO measurements over Asia and the Pacific. Asian outflow and transport is then investigated using the different types of CO satellite measurements. The origin of the exported air masses Introduction EGU Aerosols) chemistry module Folberth et al., 2006). Here, the LMDz version 4 and INCA version 2 are used, in the nudged version of the model, driven by meteorological fields (winds and temperature) from the ECMWF reanalyses (ERA15). The simulations were performed with the standard horizontal resolution of 3.75 • in longitude ×2.5 • in latitude on 19 vertical levels extending from the surface to 5 3 hPa. Different convection schemes have been tested, and the best general agreement with observations was obtained using the Tiedke scheme. INCA describes the tropospheric oxidation of NMHC and CH 4 through 332 chemical reactions using 86 chemical species. In addition to the species commonly included in this chemical scheme, regional CO tracers have been included to track the origin of 10 CO. In particular, the emissions from five regions in Asia are transported independently in the model (Fig. 1). Comparisons between LMDz-INCA and surface CO observations from the NOAA Earth System Research Laboratory (ESRL) Global Monitoring Division (GMD) network (Novelli et al., 1992) for a time period extending from 1997 and 2001 has shown the ability of the model to simulated the interannual variability of CO (Szopa 15 et al., 2007). While meteorology explains most of the variability in the tropics, it is equally explained by biomass burning and meteorological conditions at high latitude stations.
In this study, the anthropogenic emissions from the EDGAR 3.2 inventory (Olivier and Berdowski, 2001)  EGU used. During the spring of 2005, fires were mainly burning in south-east Asia and, with a smaller contribution, in southern Siberia and Canada (Fig. 4). Model simulations were first evaluated against surface CO data from the GMD network (not shown). The agreement between the model and the surface data is very good for downwind measurement sites, although the surface CO is underestimated in 5 the model after mid-April at the high latitude stations. This could be due to an underestimate of the emissions from the Russian fires which increased at this time period, or to too high photochemical destruction at high latitudes in the model. However, the transport events are well captured across the Pacific Ocean, giving confidence in the model's ability to simulate long-range transport from Asia during this time period.
Evaluation of the vertical distributions of CO was undertaken using the vertical profiles from the MOZAIC aircraft data, measured during take off and landing of commercial aircraft (Nedelec et al., 2003). Figure 2 shows comparisons above Asia (Bangkok, Hong-Kong, Shanghai and Tokyo) and western North America (Vancouver and San Francisco) averaged over the available measurements during the spring of 2005. The 15 model tends to overestimate CO in the upper troposphere and lower stratosphere (UTLS), which could be due to too low vertical resolution in the model or to too large exchange between the troposphere and the stratosphere. The agreement between ∼500 hPa and 300 hPa is generally good, with a small tendency to underestimate CO (up to ∼9% in San Francisco). On the contrary, the model clearly underestimates 20 CO in the middle troposphere (by 10 to 20% between 850 and 500 hPa), although it remains within the measured variability, and at the surface (by less than 20% except above Bangkok, where CO is underestimated by 49% below 850 hPa). Part of this underestimate is explained by the relatively low horizontal resolution of the model: MOZAIC measures CO above airports, hence in highly polluted areas which are not 25 well resolved by global models. The variability in the model simulation is also much lower than in the observations since the model cannot capture the magnitude of strong transport events, due to the lack of vertical and horizontal resolution, but also to overestimated diffusion in the model ). The regional CO tracers, Introduction EGU also shown in Fig. 2, highlight the preponderant contribution from Asia, not only locally but also above western America in the middle and upper troposphere.

Satellite CO observations
3.1 Solar occultation observations from ACE-FTS ACE-FTS (Atmospheric Chemistry Experiment Fourier Transform Spectrometer) was 5 launched in August 2003 on board the Canadian ACE satellite (also known as SCISAT-1), and measures the infrared (IR) radiation in solar occultation mode (Bernath et al., 2005). Up to thirty occultation sequences are recorded daily, during fifteen sunrises and fifteen sunsets. The solar occultation technique allows the retrieval of a wide range of trace species due to the high signal to noise ratio of the measurements, and 10 the retrieval of vertical profiles with a good vertical resolution from the mid-troposphere to the thermosphere (Coheur et al., 2007). A global fit approach is used, described in Boone et al. (2005), in which all parameters are retrieved simultaneously using a general non-linear least-squares method. The early ACE-FTS CO retrievals are described and characterized in Clerbaux et al. (2005). In this analysis, we use the data from the 15 version 2.2 of the retrieval algorithm. A full validation of the CO retrievals is provided in Clerbaux et al. (2007b). The ACE-FTS vertical sampling is ∼2 km in the mid and upper troposphere, and the accuracy of the CO mixing ratios was assessed to range between 2% (retrieval error) and 15% (discrepancy with other observations). EGU ments). These signatures, attributed to pollution outflow from Asia, are further analyzed in Sect. 5. Additional measurement of HCN, also routinely provided by the ACE-FTS standard retrieval algorithm, and of C 2 H 2 and C 2 H 6 from the research version of the retrieval algorithm (Park et al., 2007), are also used to investigate the origin of the air masses 5 by the ACE-FTS.
3.2 Nadir observations from MOPITT/Terra MOPITT (Measurement Of Pollution In The Troposphere) was launched in December 1999 on board the polar orbiting Terra satellite. At nadir, the ground pixel size of the MOPITT measurement is 22 km×22 km, and the scan angle of 26.1 • across the satel-10 lite flight track allows a global coverage in ∼3 days. It measures the infrared radiation in a nadir viewing geometry using gas correlation radiometry. Here, we use the CO data from the Phase II version 3 of the retrieval algorithm, described in Deeter et al. (2003). The profiles are retrieved on 7 vertical levels, from the surface to 150 hPa using a nonlinear optimal estimation method. However, the vertical resolution achieved is limited 15 by the observation technique. Although studies have shown that the MOPITT profiles contain information on the CO vertical structure (Kar et al., 2004), this information is highly correlated. The number of independent pieces of information characterizing the profiles (commonly referred to as degrees of freedom for signal, DOFS) varies from ∼2 in the tropics to less than 1 at high latitudes, weighted towards the middle troposphere 20 (Heald et al., 2003;Deeter et al., 2004). The vertical resolution is estimated to be ∼6 km. Retrievals have been validated against aircraft measurements and shown to be biased high by ∼6% for the Phase II retrievals used here (Emmons et al., 2007). As the nighttime measurements have not been validated and appear to be biased relative to the daytime measurements (Heald et al., 2004), only daytime data are considered 25 here.
The total column of CO observed during the spring of 2005, averaged over the model 3.75 • ×2.5 • grid is mapped in Fig. 4 EGU again, strong CO enhancements are observed above eastern and south-eastern Asia and above the Pacific Ocean. The outflow from Asia is clearly identified above the Pacific Ocean, with a maximum between ∼20 • N to 60 • N. Note that looking at the individual retrieval vertical levels from MOPITT does not change the pattern of the export due to the lack of vertical resolution. The retrieval error is mostly driven by 5 the vertical sensitivity of MOPITT: DOFS strongly depends on surface temperature, with larger DOFS (and hence better vertical resolution) at higher surface temperatures (Deeter et al., 2004), so that the error rapidly increases towards high latitudes (lower signal measured, so larger contribution from the a priori information used during the retrieval).

Nadir observations from SCIAMACHY/Envisat
SCIAMACHY (SCanning Imaging Absorption spectroMeter for Atmospheric CHartog-raphY) was launched in March 2002 on board the ENVISAT satellite, and allows the observation of a number of trace gases with a horizontal resolution of 30 km ×60-120 km. 15 We use the CO total columns retrieved with the Iterative Maximum Likelihood Method (IMLM) using measurement in the near infrared between 2324.5-2337.9 nm (Gloudemans et al., 2005). This spectral region allows detection down to the Earth's surface, with equal sensitivity to all vertical layers, unlike the thermal infrared region which has a lower sensitivity in the boundary layer. In this paper version 7.4 of the IMLM algorithm 20 is used in which the instrument noise is estimated per individual detector pixel using the dark signal measurements instead of assuming an average value over the whole spectral range as is done in IMLM v6.3 used in de Laat et al. (2006Laat et al. ( , 2007 and Gloudemans et al. (2006). In addition, this version includes some minor improvements on the calculation of the cross sections. The CO total columns retrieved with IMLM v7.4 are 25 similar to those of v6.3 presented in de Laat et al. (2007), but with a smaller spread in the single measurements. The retrieval error due to random instrument noise can be large, typically 10-100% depending on the variations in surface albedo and solar 1718 Introduction EGU zenith angle. Especially over sea the instrument-noise retrieval errors are large, due to the low sea surface reflectance in the near infrared. Therefore measurements over sea and measurements with instrument-noise errors >1.5×10 18 molec/cm 2 are excluded from the analysis. The data are filtered using the same selection criteria as in Gloudemans et al. (2006) and are averaged over the whole season on the model 3.75 • ×2.5 • 5 grid as recommended in de Laat et al. (2006), using a weighted average based on the instrument-noise retrieval error. Whereas de Laat et al. (2006Laat et al. ( , 2007 only include data with cloud cover <20%, the Gloudemans et al. (2006) selection criteria are used here, including also data over low clouds. Although over low clouds the lowest part of the CO total column is missing, no significant bias is found when comparing the averaged  Fig. 4. Large enhancements are observed above eastern Asia, reflecting the main source regions. 15 Large values are also observed at high latitudes, probably associated with CO from the northern mid-latitudes accumulating during the winter. The retrieval error is larger above Indonesia, and generally along the Coast, as well as at latitudes above 60 • N.

Intercomparison of CO satellite observations
In order to compare vertical profiles retrieved from different instruments, the specific 20 characteristics of each observing system needs to be accounted for. The methodology introduced by Rodgers and Connor (2003) as been used in order to smooth profiles with the higher vertical resolution using the characteristics of lower resolution profiles. These characteristics are represented by the averaging kernel matrix A=∂x/∂x, corresponding to the sensitivity of the retrievalx at each vertical level to the true profile x. 25 A is a critical quantity; its rows define the vertical resolution of the retrieval (full width at half maximum), and the trace (sum of diagonal elements) of the matrix defines the number of independent elements that can be retrieved, or the number of degrees of 1719 Introduction EGU freedom for the signal (DOFS). A smoothed profile is then obtained as follows: where x high is the high resolution profile, x a,low is the a priori profile used for the retrieval of the low resolution profile, and A low is the averaging kernel matrix characterizing the low resolution profiles. 5 The vertical resolution of the solar occultation measurements from ACE is much larger than that of the nadir measurements from MOPITT. We use Eq. (1) for the intercomparisons, applying the MOPITT retrieval characteristics to the higher resolution ACE-FTS profiles. ACE-FTS only provides retrievals in the middle and upper troposphere, so that only these levels, and corresponding MOPITT averaging kernels, are 10 used in the comparison. Figure 5 shows several examples of comparisons, before and after applying the smoothing (Eq. 1) using the MOPITT averaging kernels (also plotted in Fig. 5) to the ACE-FTS profiles. Table 1  First of all, the comparison between the ACE-FTS retrieval and the ACE-FTS 15 smoothed profiles highlights the advantage of such high resolution measurements which can capture the vertical structure of pollution plumes in the upper troposphere. ACE-FTS smoothed and MOPITT are globally in very good agreement considering their respective uncertainty. The relative difference between the two measurements is ∼18% for all levels, with MOPITT giving higher CO values than ACE-FTS. The best 20 agreement is achieved for MOPITT DOFS≥2 (i.e., more than 2 independent elements of information contained in the retrieved profile), in the tropical regions (see Fig. 5d) because the information in the upper troposphere is less correlated to CO in the lower and middle troposphere. On the contrary, when the DOFS decreases, the MOPITT retrieval levels are all highly correlated, with a maximum sensitivity in the middle tropo- 25 sphere. CO in the lower/middle troposphere then contributes significantly to the upper troposphere MOPITT retrievals. For these cases, the method used for the comparison becomes questionable. Since the ACE-FTS data do not cover the lower troposphere, ACPD 8,2008 Analysis of CO outflow from Asia using satellite data S. Turquety et al. EGU this contribution is not included, and this results in an artificially low smoothed ACE-FTS profile, explaining the observed bias. This artifact can be corrected using CO profiles in the lower troposphere to complete the ACE-FTS profiles at lower altitudes. We verified this using the MOZAIC in situ measurements. However, coincident measurements proved difficult to find and a statistical comparison could not be undertaken.
Since the instrument noise error of single SCIAMACHY CO total column measurements is typically 10-100%, a specific comparison of single CO observations could not be undertaken. The distributions averaged over the whole period without taking into account the exact spatial and temporal collocations of the SCIAMACHY and MOPITT observations are shown in Fig. 4. They show good general agreement in the observed 10 patterns, with CO enhancements over eastern Asia and Indonesia associated with anthropogenic and biomass burning emissions. Over high altitude regions such as the Himalayas and the Rocky Mountains, MOPITT and SCIAMACHY agree well. SCIA-MACHY observes larger CO over large source regions (southern and eastern Asia in particular). This is largely explained by the ability of SCIAMACHY to sound the full 15 vertical column, whereas MOPITT is mostly sensitive to the free troposphere. SCIA-MACHY is also larger at high latitudes, which could be explained by surface emissions from biomass burning in Siberia and Canada that are not captured by MOPITT (too low sensitivity), or by transport from Europe at low altitudes over this region. However, this may not be significant since both MOPITT and SCIAMACHY show larger retrieval error 20 at high latitudes. On the contrary, MOPITT CO total columns are larger than those of SCIAMACHY over western Siberia and India. This suggests a bias between the datasets.

Analysis of the Asian outflow
Pollution is lofted from the Asian boundary layer by three main processes: rapid trans-25 port by convection or slower transport by frontal lifting ahead of cold fronts and orographic forcing. It is then rapidly advected in the free troposphere by westerly winds. 8,2008 Analysis of CO outflow from Asia using satellite data S. Turquety et al.

EGU
We analyzed the ECMWF horizontal wind fields used to nudge the model simulation as well as the vertical transport in the model. Strong vertical winds over India, southern and eastern China and over the western Pacific allow rapid transport of pollution from these regions into the free troposphere. It can then be exported across the Pacific Ocean by westerly winds between ∼30 • N and 60 • N. During the time period from 20 5 March to 10 April, air masses are transported to the north-east Pacific, while during the time period from 20 April to 25 May, the transport is confined to lower latitudes, ∼30-40 • N. This is consistent with the CO observed from MOPITT and ACE during these time periods (Figs. 3 and 4).
In this section, we first use the CO data from MOPITT and SCIAMACHY for a general 10 evaluation of the model over Asia and the Pacific. Comparison between these data and the model CO are then used to evaluate the Asian emissions. CO observations from MOPITT and ACE-FTS are then compared to the modeled CO to further analyze the export from Asia, and its transport across the Pacific. We then study the origin of the pollution plumes observed by using ACE-FTS measurements of other tropospheric 15 species (C 2 H 2 , C 2 H 6 , HCN).

Complementary view from MOPITT and SCIAMACHY
In order to get a better insight into the complementary information from the two nadir instruments, we use the model as an intermediate.  al., 2007), comparisons of the model results to the MOZAIC aircraft data also showed an underestimate of CO in the free troposphere of ∼10-20%. Such an underestimate is a common problem in many global models, as revealed in the intercomparison undertaken by Shindell et al. (2006), who attribute this to a year round underestimate of emissions from Asia. 5 On the contrary, the model total CO tends to be systematically higher than SCIA-MACHY (by 24% on average), especially above eastern India and western Siberia. This underestimate compared to MOPITT and overestimate compared to SCIAMACHY is partly explained by a systematic bias between MOPITT and SCIAMACHY (SCIA-MACHY apparently low compared to MOPITT), as discussed in Sect. 3.4. It could 10 also indicate too large CO at low altitudes in the model since MOPITT lacks sensitivity in the boundary layer, and could highlight an underestimate of pollution lofting. The model CO remains lower than SCIAMACHY over the largest emission regions (eastern China or the Gange Valley in particular), which could be another indication of too low emissions over Asia. Further evaluation of the Asian emissions and transport with the 15 satellite observations is presented in the following sections.

Evaluation of Asian emissions
As already mentioned, SCIAMACHY and MOPITT CO retrievals are characterized by different vertical sensitivities, inherent to the different spectral regions used: SCIA-MACHY is sensitive to the full column, while MOPITT in mainly sensitive to the free 20 troposphere, with low sensitivity in the boundary layer. Clerbaux et al. (2008) show that enhancements associated with urban pollution can be detected by MOPITT under specific conditions. However, this sensitivity, and hence the contribution from the boundary layer to the MOPITT column, remains low compared to SCIAMACHY. The difference between SCIAMACHY and MOPITT total columns therefore provides infor-25 mation about boundary layer CO, closer to emission sources. Buchwitz et al. (2007) showed that this difference allows a clear identification of pollution enhancements associated with urban areas. 1723

EGU
We use these complementary views of the vertical column to qualitatively evaluate the surface emissions over Asia in the model. Therefore, we define a boundary layer residual equal to: where CT SCIAMACHY is the total column retrieval from SCIAMACHY, CT MOPITT is the 5 total column retrieved from MOPITT, and CT a is the contribution from the a priori to the MOPITT total column retrieval. The MOPITT and SCIAMACHY data measured during the same day and collocated within the model horizontal resolution are used. This residual is also calculated using the model results, with and without applying the MOPITT averaging kernel to the model profiles. The contribution from the 10 a priori information on the CO profile x a,MOPITT in the MOPITT total column retrieval, CT a =T (I − A MOPITT ) x a,MOPITT with T the column operator, is removed in order to only keep the part of the column actually observed by MOPITT in the free troposphere. For fully self consistent observations, the residual then corresponds to the additional information on the true state x provided by SCIAMACHY with respect to MOPITT: This is true for the model BLR but, since both instruments rely on different observation techniques and retrieval procedures, some inconsistency may result in a bias in the observed BLR. Indeed over remote regions the observed BLR shows a tendency to be lower than the model BLR. This bias is removed by assuming that the observed BLR 20 over low CO regions far from the sources (chosen to be above west-central Asia) is equal to the model BLR. The observed BLR is then corrected by addition of a constant amount ∆=6.4×10 17 molecules/cm 2 . Note that this assumes that the model is correct above west-central Asia.
The measured and simulated BLR are shown in Fig. 6 for values exceeding the 25 SCIAMACHY and MOPITT errors. Values above 60 • N are also removed since both instruments show larger uncertainty at high latitudes (Fig. 4) EGU regions are clearly identified in both the measured and the modeled residual. The observed BLR allows a better resolved distribution of the emissions, with large BLR enhancements above the largest cities (particularly clear over China). The model BLR is lower by ∼0.33 molecules/cm 2 over Eastern Asia, which corresponds to ∼37% on average over this region (see Fig. 1 for region limits). This comparison suggests that 5 emissions are underestimated over this region, particularly over large urban areas.
The model BLR is also too low above Indonesia, Siberia and south-western Canada, probably due to an underestimate in biomass burning emissions. On the contrary, the model BLR is too large over western Siberia. This could be due to too much transport from Europe at low altitude, as already discussed in Sect. 4.1.
Our model simulation is based on the EDGAR inventory, adjusted for CO using inversion results from Pétron et al. (2004) based on 2001 MOPITT observations. Recent revision of the Asian emission inventories (Streets et al., 2006;Ohara et al., 2007) and inverse modeling work (Kopacz et al., 2008) (allowing higher spatial resolution than previous inversions, including Pétron et al., 2004) also recommend an increase 15 of emissions in Eastern Asia, in eastern and central China in particular. Further work would be required in order to allow a quantitative re-evaluation of the emissions: a better understanding of the apparent bias between SCIAMACHY and MOPITT, as well as an inverse analysis accounting for the instrument-noise error of each measurement would be required, which is beyond the scope of this study.

Trans-Pacific transport of CO
The good spatial and temporal resolutions of the MOPITT data are particularly interesting for the analysis of the spatial patterns and temporal variability of the transport. MOPITT clearly highlights the importance of trans-Pacific transport of pollution in the free troposphere between ∼20 • N and ∼60 • N (Fig. 4) EGU the free troposphere in the model, which could be explained both by too low surface emissions and by too low lofting into the free troposphere, as already discussed in Sects. 4.1 and 4.2.
The ability of the model to simulate the transport events was examined by looking at variations in the total averaged CO columns during the spring over five boxes, plotted in 5 Fig. 7. The agreement over Eastern Asia is very good, but the model underestimates CO over the Pacific Ocean and this underestimate tends to increase as the season progresses. This suggests an underestimate, not only of the export from Asia, but maybe also of the net chemical production of CO, which could be due to too large photochemical destruction or too low NMHC concentrations in the model. However, the 10 main features are captured by the model and are mainly attributed to the contribution from Asian sources, which control the temporal variations of CO over the entire region. Figure 8 shows the contributions from emissions in the main source regions to the total CO in the model. The main exporting regions for March-April 2005 are eastern Asia between ∼30 • N and 60 • N (dominated by fossil fuel combustion), and India and In- 15 donesia at lower latitudes. The large CO enhancement over the North-Western part of the region (western Siberia) in the MOPITT data is attributed to transport from Europe. These regional tracers indicate that the underestimated CO in the model could be associated with too low export from Eastern Asia, but also from Europe, or to transport at too low altitude.

20
Comparisons between the model and the ACE-FTS CO allow the analysis of the vertical structure of the transport in the middle and upper troposphere. The comparisons during the two observation time periods (Fig. 9) show good consistency in the latitudinal variations, i.e. lower CO in the tropics due to enhanced photochemical production of OH and larger values at high latitudes due to slower photochemistry. The 25 transport patterns are also consistent, with CO enhancements over eastern Asia and the Pacific at mid-latitudes. As already highlighted by the comparisons to MOZAIC and MOPITT, the model tends to underestimate CO in the middle troposphere, especially at the end of the season, which can be attributed to too low emissions and export and/or ACPD 8,2008 Analysis of CO outflow from Asia using satellite data S. Turquety et al. EGU too high chemical destruction. Part of the underestimate could also be explained by the relatively low vertical and horizontal resolution of the global model which does not allow the reproduction of sharp pollution plumes, as well as too strong diffusion. The strong overestimate at high latitudes in the upper troposphere and lower stratosphere (UTLS), which also appeared in the comparisons with MOZAIC (Sect. 2), reflects too 5 large exchange between the troposphere and the stratosphere in the model. The vertical distribution of CO is shown on cross sections for latitudinal bands over Asia and the Pacific Ocean (Figs. 10 and 11) and illustrates the vertical structure of transport. The model tends to be too low in the middle troposphere, and too high in the upper troposphere (above ∼300 hPa). However, the CO enhancements observed by ACE-FTS are captured by the model. Note that the individual ACE-FTS profiles were compared to latitudinal averages in order to get a broader picture of the long range transport, and also to allow for transport errors in the model (plume may not be located in the right grid box). Pollution is lofted into the free troposphere over eastern and southern Asia. It is then transported eastward in the free troposphere. For the 15 first period (20 March-10 April), large CO enhancements are observed by ACE-FTS in the upper troposphere above the Pacific Ocean at latitudes  • N, with CO typically up to ∼200 ppbv around 400 hPa and at slightly higher altitude (∼300 hPa) moving eastward. In the region [15][16][17][18][19][20][21][22][23][24][25][26][27][28][29][30] • N, ACE-FTS enhancements are found at lower altitude (∼500 hPa), consistent with the model showing subsidence over the Ocean. Later in 20 April and May (Fig. 11), large CO values are observed in the 15-30 • N latitude band, with two enhancements: one over eastern Asia, associated with rapid vertical transport to the upper troposphere and one further east over the Ocean. The two export branches are clearly seen in the model in this latitude band, and also around [30][31][32][33][34][35][36][37][38][39][40][41][42][43][44][45] • N. 25 In addition to a good vertical resolution, ACE provides information on a series of trace gases. Among the other species routinely retrieved from the ACE-FTS measurements are acetylene (C 2 H 2 ), ethane (C 2 H 6 ) and hydrogen cyanide (HCN EGU eraged profiles measured above the Asian outflow are shown in Fig. 12, together with the main contributions per regions to the total modeled CO. C 2 H 2 , C 2 H 6 and CO have common sources from combustion and are removed by reaction with OH. However, the mean lifetime of C 2 H 2 (∼ 2 weeks) is much shorter than that of CO and C 2 H 6 (∼2 months). The relationship between C 2 H 2 and CO can be 5 used to study the chemical ageing of air masses (e.g., Smyth et al., 1996;Xiao et al., 2007;Park et al., 2007). Air masses with C 2 H 2 /CO ratio greater than 3 ppt/ppb are considered to correspond to young pollution plumes, containing recent emissions, while ratios lower than 1 ppt/ppb suggest aged (older than one week) and well mixed plumes (Russo et al., 1999). CO, C 2 H 2 and C 2 H 6 mixing ratios from ACE-FTS, and the corresponding model profiles, are shown in Fig. 12 for four regions over Asia and the Pacific. Note that NMHC emissions were not increased in the simulation presented here, although several studies have pointed the need to more than double the emissions of NMHC from the EDGAR inventory in extratropical regions (Streets et al., 2003, Xiao et al., 2007. 15 The C 2 H 2 , C 2 H 6 and CO profile shapes are well correlated in the observations and in the model, reflecting common sources. Large C 2 H 2 /CO ratios are observed within the CO plumes, typically around 2 ppt/ppb above the ocean, but reaching more than 3 ppt/ppb for observations above the eastern Pacific in the mid-troposphere around 30 • N. This highlights the rapid export and transport across the Pacific Ocean. The 20 regional tracers of CO indicate that anthropogenic emissions from eastern Asia constitute the main contribution. Transport from Indonesia also has significant contribution in the upper troposphere at low latitudes, while transport from Europe and North America contribute significantly at higher latitudes in the low and middle troposphere. Transport from India is located in the middle troposphere and has a strong impact on the model 25 CO above Asia, especially at the end of the season, but also above the Pacific.

Indications for plume origins in the ACE-FTS data
For C 2 H 6 and CO, the model is of the correct order of magnitude but underestimates enhancements associated with transport events and overestimates mixing ratios above ∼300 hPa. The background C 2 H 2 is correct in the model, but greatly underestimated in EGU the transported plumes, resulting in C 2 H 2 /CO ratios too low by a factor of 2 compared to ACE-FTS. This clearly suggests too large diffusion in the model, as discussed for CO in the previous section. This could also be partly due to underestimated NMHC emissions in the model, which would result in an underestimate of the chemical production of CO. The ACE-FTS HCN can be used as a tracer for the identification of biomass burning 5 signatures (Li et al., 2003). Large HCN mixing ratios were observed between 15 • N and 30 • N in the upper troposphere. Enhanced HCN extends to the north-eastern Pacific at lower altitudes in March-April 2005 (400 hPa) but with a small magnitude compared to the large CO enhancements in this region. The level of the HCN maxima does not always coincide with the CO enhancements (Fig. 11); it is generally observed at 10 higher altitudes (which is consistent with in situ observations, de Gouw et al., 2004). This indicates that biomass burning is not the most important contribution, especially at high latitudes. However, the correlation increases at lower latitudes, where the biomass burning outflow from southern Asia and Indonesia is expected. Larger HCN mixing ratios are observed during the second time period, extending across the Pacific. The 15 vertical distributions are consistent with the simulated biomass burning signatures of CO (Fig. 12). However, the lower HCN mixing ratios in March-April than in April-May suggest an increased influence of the fire activity during the season that is not well represented in the model (no such increase in the biomass burning CO mixing ratios). Figure 12d shows profiles measured at mid-latitudes above the eastern Pacific at the 20 end of April 2005, where CO was clearly underestimated in the model (Fig. 9). Large HCN values highlight a strong influence from a biomass burning plume that is not captured in the model, with no clear enhancement of the measured C 2 H 2 /CO ratios indicating relatively aged air. This could reflect a strongly underestimated source, but also too large diffusion of a biomass burning plume or underestimated lofting of the

Conclusions
In this study, the export of CO from Asia was analyzed with the global chemistry and transport model LMDz-INCA based on the complementary pictures provided by the satellite measurements from MOPITT, SCIAMACHY and ACE-FTS. In addition to the full NMHC-CH 4 -O 3 chemistry, CO regional tracers were included in LMDz-INCA in or-5 der to track the signature from the main source regions. The surface emissions from the EDGAR 3.2 inventory have been updated for Asia by scaling emissions above this region to the inverse modeling results by Pétron et al. (2004) (which are based on MO-PITT CO for 2001). Evaluation of the model CO simulation against MOZAIC aircraft observations showed an underestimate of CO by ∼10-20% through the troposphere, 10 especially below ∼500 hPa and above ∼300 hPa. Further analysis of the Asian outflow was undertaken based on the satellite observations. MOPITT and SCIAMACHY both use the nadir viewing geometry and provide global CO distributions with good spatial coverage (typically global every 3 days). However, the nadir geometry implies vertically integrated information. Between 1 and 2 pieces of 15 independent information are available in the MOPITT retrievals, with maximum sensitivity in the free troposphere. SCIAMACHY only allows the retrieval of total CO columns, averaged temporally and/or spatially in order to decrease the random instrument-noise error, but equally sensitive throughout the troposphere. The ACE-FTS solar occultation observations offer valuable information on the vertical structure of the transport in the 20 middle and upper troposphere, in spite of the limited horizontal coverage. An intercomparison of the CO measurements provided by the three instruments over Asia and the Pacific Ocean highlights their complementary nature and shows the good consistency between the datasets, with an agreement better than 20% between MOPITT and ACE-FTS in the upper troposphere. Although SCIAMACHY observes larger CO above 25 the largest source regions, consistent with its better sensitivity to the lower altitudes, it tends to be systematically lower than MOPITT above remote regions, except over very high altitude regions, such as the Himalayas and the Rocky Mountains, where the EGU agreement is good. This reveals a bias between the two datasets, which needs further investigation. Accounting for this bias, the different sensitivities of MOPITT and SCIAMACHY allow the analysis of CO at low altitudes. Boundary layer residuals (BLR) were obtained by subtracting MOPITT columns (with the contribution from the a priori removed) from 5 the SCIAMACHY total CO. This BLR provides information on the CO source regions, with, in particular, large enhancements above the main urban areas. The comparison with the corresponding model BLR highlights an underestimate in Asian emissions in the model, in particular above eastern and central China, even after adjustment of the emissions. Recent inventories (Streets et al., 2006) and inverse modeling work 10 (Kopacz et al., 2008) also show that these source regions are underestimated in current inventories, and highlight the rapidly increasing fuel consumption in China (Ohara et al., 2007). Uno et al. (2007) show that model using updated Asian emissions still underestimate sources with respect to the GOME NO 2 observations, particularly over China. A more thorough analysis of the possible trends in the MOPITT CO observa- 15 tions would be required to better understand the recent evolution of the emissions. Our analysis also suggests an underestimate in emissions over south-east Asia, as well as Siberia and Canada during 2005, probably associated with too low biomass burning emissions.
The transport of pollution from Asia above the Pacific is clearly observed by MO- EGU also explained by underestimated export from CO source regions. In addition to a good vertical resolution, the solar occultation observation mode of the ACE-FTS allows for the detection of a wide variety of chemical species thanks to a high signal to noise ratio. Measurements of acetylene (C 2 H 2 ), ethane (C 2 H 6 ) and hydrogen cyanide (HCN) were used in order to further examine the origin of the observed plumes. 5 C 2 H 2 and C 2 H 6 are well correlated with the CO profiles, highlighting common sources and rapid trans-Pacific transport in the upper troposphere. The comparisons between the modeled and observed C 2 H 6 show the same characteristics as CO (too high in the UTLS, too low in the transport plumes), but C 2 H 2 is too low throughout the troposphere, consistent with too low export and too large diffusion in the model.

10
HCN is commonly used as a biomass burning tracer. Comparisons between the HCN observations and the model biomass burning signature show good agreement in the transport patterns, with strong enhancements in the 15 • N-30 • N latitude band, where the model regional CO tracers show a large signature from southern Asia and Indonesia, but reveals a possible underestimate in fire activity at the end of the season, 15 and/or too large diffusion of the fire plume in the model. Observed HCN and CO profiles are not well correlated for measurements at higher latitudes, indicating that biomass burning is not the dominant source. This is in agreement with the predominance of the eastern Asian contribution to the CO over this region in the model, revealed by the regional CO tracers. 20 We have shown that combining information from MOPITT, SCIAMACHY and ACE-FTS provides a more complete view of the Asian outflow, from surface emissions to transport in the middle and upper troposphere over the Pacific. Assimilation of the observations in the model would be needed to allow more quantitative analyses accounting for the specific properties of each dataset, including, in particular, their un-25 certainties and vertical resolution. Further analysis of ACE-FTS measurements will be undertaken to analyze the chemical composition of the air masses during export and their impact ozone production over the Pacific and further downwind. . Boone, C. D., Nassar, R., Walker, K. A., Rochon, Y., McLeod, S. D., Rinsland, C. P., and 5 Bernath, P. F.: Retrievals for the atmospheric chemistry experiment Fourier-transform spectrometer, Appl. Optics, 44, 7218-7218, 2005. Buchwitz, M., de Beek, R., Bramstedt, K., Noël, S., Bovensmann, H., and Burrows, J. P.: Global carbon monoxide as retrieved from SCIAMACHY by WFM-DOAS, Atmos. Chem. Phys., 4, 1945Phys., 4, -1960Phys., 4, , 2004,  Chemical isolation in the Asian monsoon anticyclone observed in Atmospheric Chemistry ACPD 8,2008 Analysis of CO outflow from Asia using satellite data S. Turquety et al.