Carbon monoxide (CO) and ethane (C2H6) trends from ground-based solar FTIR measurements at six European stations, comparison and sensitivity analysis with the EMEP model

Trends in the CO and C 2H6 partial columns ( ∼0– 15 km) have been estimated from four European groundbased solar FTIR (Fourier Transform InfraRed) stations for the 1996–2006 time period. The CO trends from the four stations Jungfraujoch, Zugspitze, Harestua and Kiruna have been estimated to−0.45± 0.16 % yr−1, −1.00 ± 0.24 % yr−1, −0.62±0.19 % yr−1 and−0.61±0.16 % yr−1, respectively. The corresponding trends for C 2H6 are−1.51± 0.23 % yr−1, −2.11±0.30 % yr−1, −1.09±0.25 % yr−1 and −1.14± 0.18 % yr−1. All trends are presented with their 2-σ confidence intervals. To find possible reasons for the CO trends, the global-scale EMEP MSC-W chemical transport model has been used in a series of sensitivity scenarios. It is shown that the trends are consistent with the combination of a 20 % decrease in the anthropogenic CO emissions seen in Europe and North America during the 1996– 2006 period and a 20 % increase in the anthropogenic CO emissions in East Asia, during the same time period. The possible impacts of CH 4 and biogenic volatile organic compounds (BVOCs) are also considered. The European and global-scale EMEP models have been evaluated against the Correspondence to: J. Mellqvist (johan.mellqvist@chalmers.se) measured CO and C 2H6 partial columns from Jungfraujoch, Zugspitze, Bremen, Harestua, Kiruna and NyÅlesund. The European model reproduces, on average the measurements at the different sites fairly well and within 10–22 % deviation for CO and 14–31 % deviation for C 2H6. Their seasonal amplitude is captured within 6–35 % and 9–124 % for CO and C2H6, respectively. However, 61–98 % of the CO and C2H6 partial columns in the European model are shown to arise from the boundary conditions, making the globalscale model a more suitable alternative when modeling these two species. In the evaluation of the global model the average partial columns for 2006 are shown to be within 1–9 % and 37–50 % of the measurements for CO and C 2H6, respectively. The global model sensitivity for assumptions made in this paper is also analyzed.

. In contrast, ethane (C 2 H 6 ) has shown a constant negative trend in both the 1980s and 1990s of roughly 1-3 % per year Mahieu et al., 1997).
These changes are important as tropospheric chemistry to a large extent is controlled by the hydroxyl (OH) radical, also often referred to as the detergent of the atmosphere. One of the main OH sinks is the reaction with CO (Crutzen et al., 1999). In this oxidation, the greenhouse gas carbon dioxide is formed but the reaction is also related to the formation or destruction of tropospheric ozone (O 3 ), depending on the NO x concentrations in the ambient air. C 2 H 6 is the second most common organic trace gas in the troposphere after methane (CH 4 ) and is, like CO, destroyed by the OH radical. C 2 H 6 has also shown to be a major route for the formation of peroxyacetyl nitrate (PAN) which acts as a NO x reservoir and thereby accelerates O 3 formation in the troposphere (Blake and Rowland, 1986;Finlayson-Pitts and Pitts, 2000).
Trends in the concentration of CO and C 2 H 6 as seen from a particular location can result from both changes in emissions and changes in chemical production and loss processes. Observations alone cannot usually distinguish between these factors. In principal, chemical transport models (CTMs) can account for all the major processes affecting CO and C 2 H 6 , but such models are also limited by both inherent deficiencies and not least by the quality of the emissions data upon which such models rely.
The purpose of this paper is to make use of one such CTM to explore the trends seen in ground-based solar FTIR (Fourier Transform InfraRed) measurements of CO from four European stations. The model used is the globalscale model of the European Monitoring and Evaluation Programme developed by the Meteorological Synthesizing Centre -West (EMEP-MSC-W). The FTIR datasets used in this paper are compiled within the NDACC network (Network for the Detection of Atmospheric Composition Change, http: //ndacc.org/) for which a wide range of atmospheric species is measured in the mid-infrared spectral region with highresolution spectrometers, generally since the early 1990s, and even the mid-1980s at Jungfraujoch. An important part of the current study is also to evaluate how well the EMEP model reproduces observed CO and to a lesser extent C 2 H 6 levels, both as an opportunity to evaluate this model using novel measurements, and to give confidence to the use of the model for such sensitivity analysis. It can be noted that many comparisons between CTMs and satellite and in situ measurements regarding CO have been performed but often with poor results, especially for the spring time maxima Shindell et al., 2006).

Chemistry and sources of CO and C H 6
CO concentrations in the atmosphere are affected by emissions and chemical formation. Primary emissions of CO result from incomplete combustion of carbon-containing fuels. In developed countries the major anthropogenic source is related to emissions from the transport sector (Finlayson-Pitts and Pitts, 2000). Natural CO sources include oxidation from organics, primarily from atmospheric methane and biogenic hydrocarbons (BVOC), and from biological processes in soils and ocean. An overview of the estimated global CO sources is presented in Table 1. In Table 1 it can be seen that the major source for CO is the natural one and that the anthropogenic source globally only contributes to approximately 15 % of the total yearly emissions. Almost all of the CO emissions from fossil fuel combustion and 2/3 of those from biomass burning are located in the Northern Hemisphere (Holloway et al., 2000). The CO source from oxidation of biogenic hydrocarbons is roughly equally divided between the hemispheres and that from methane oxidation is slightly larger in the Northern Hemisphere. The major CO sink is the reaction with the OH radical, seen in Reaction (1), accounting for 90-95 % of the total CO loss . CO is also a key compound in the formation and destruction of tropospheric O 3 depending on the background concentrations of NO x (Crutzen et al., 1999).
CO + OH(+O 2 ) → HO 2 + CO 2 (R1) In the Northern Hemisphere, Reaction (R1) is the dominant sink for OH, and even in polluted European boundary layers, CO accounts for a significant fraction of the OH loss (Simpson et al., 1995). The OH radical is produced from photodissociation of O 3 (λ ≤0.32 µm) and reaction with water vapor according to Reactions (R2) and (R3), respectively. Since the formation of the OH radical is strongly dependent of the amount of sunlight, the CO concentration shows a strong intra-annual behavior with low values in the summer and high in the winter.
In Reaction (R3) only a small fraction (∼10 %) of the energetically excited oxygen atoms, O( 1 D), react with water molecules and produce OH, the others recombine with O 2 to form O 3 (Finlayson-Pitts and Pitts, 2000). The main C 2 H 6 source is of anthropogenic origin and includes production and transport of fossil fuels and use of biofuel. Another important source is biomass burning (Xiao et al. 2008, Table 1). Xiao et al. (2008) also estimate the anthropogenic and biomass burning emissions in Europe to be 2.1 and <0.1 Tg yr −1 , respectively. As much as 84 % of the C 2 H 6 sources are located in the Northern Hemisphere, with highest emissions in Asia followed by North America Table 1. Global sources of CO and C 2 H 6 (Tg yr −1 and % of total). CO data are from Holloway et al. (2000) and C 2 H 6 data are from Xiao et al. (2008  and Europe. The main sink, which causes up to 95 % of the C 2 H 6 removal, is the reaction with the OH radical in which the ethyl radical and water are formed, see Reaction (R4).
The ethyl radical is then transformed through a complicated pattern of oxidations and reductions and ultimately ends up as CO (Aikin et al., 1982).

FTIR measurements
In this paper, FTIR partial column data of CO and C 2 H 6 from six European stations are used from the time period of 1996-2006. Information regarding the FTIR stations can be found in Table 2. A column is defined as the integrated amount of a species from the measurement station to a certain altitude, usually expressed as the number of molecules per unit area (molecules cm −2 ). When retrieving data from solar FTIR measurements, a synthetic spectrum is calculated in a forward model in which the atmosphere is divided into discrete layers. To compute the synthetic spectrum, a priori vertical distributions, line parameters for the target and interfering gases, as well as temperature-and pressure profiles are needed. In the FTIR case line parameters from HI-TRAN are often used (Rothman et al., 2005). To account for the line broadening caused by the spectrometer, the instrument lineshape is included in the calculation. The synthetic spectrum is then compared with the measured one and the forward model parameters are adjusted in an iterative way until the modeled spectrum is sufficiently close to the measured one. This inverse process is an ill-posed problem that is solved in this paper with the Optimal Estimation Method (OEM) in which the final retrieved profile is a statistical weighted combination of an a priori profile of the target gas and the measurement (Rodgers, 2000). In order to use the OEM, uncertainties in the a priori profile and measurements have to be assumed/known. The partial column is obtained by multiplying the retrieved profile with a pressure profile and summing the result to the height of interest. The retrieval procedure is implemented in an algorithm called SFIT2  which is used for all the participating stations except Kiruna, which uses PROFFIT . These two algorithms have shown to be within 1 % agreement Duchatelet et al., 2010). The micro-windows used in the present retrievals are in the region from 2057 to 2159 cm −1 for CO and from 2976 to 2977 cm −1 for C 2 H 6 . For CO, the interfering species O 3 , H 2 O, CO 2 , N 2 O and OCS are fitted simultaneously, while CH 4 , O 3 and H 2 O are co-fitted in the C 2 H 6 window. The retrieval strategy used for CO and C 2 H 6 was developed within the UFTIR (http://www.nilu.no/uftir/) project and are further described by De Maziere (2005) and Vigouroux et al. (2008). Error budgets of ground-based solar FTIR CO and C 2 H 6 measurements has been reported by Zhao et al. (2002) for two Japanese stations and by Rinsland et al. (2000Rinsland et al. ( , 2007 for Jungfraujoch and Kitt Peak. The reported total errors for CO are between 5.2 % and 6.7 % and for C 2 H 6 between 8.4 % and 8.9 %. The three papers have used the same microwindows, interfering species and retrieval algorithm (SFIT2) as in this paper but uses site specific a priori, pressure and temperature profiles. We conclude that the errors budgets from the references are applicable for the other stations used in this paper due to their varying altitude and latitude. The instrumental line shape, which is site specific, also affects the error. Although, the instrumental line shapes at all stations are monitored with gas cells measurements on a regular basis and the error is therefore assumed to be small. These measurement are retrieved with the LINEFIT program developed by Hase et al. (1999).

The EMEP model
The European Monitoring and Evaluation Programme (EMEP, www.emep.int) started in 1977, a successful effort between almost all European countries to pool efforts in tackling the major environmental problem of the day, acid deposition. When the Convention on Long-range Transboundary Air Pollution (CLRTAP, www.unece.org/env/lrtap) was established in 1979, EMEP became an integrated part of the Convention, and has since played an important part in the development of emission reductions scenarios, for both the Convention (now comprising 51 Parties, including USA and Canada) and the European Commission.
The EMEP chemical transport model (CTM) (Simpson et al., 2003a is a development of the 3-D chemical transport model of Berge and Jakobsen (1998), extended with photo-oxidant chemistry (Simpson et al., 1995(Simpson et al., , 2003bAndersson-Skold and Simpson, 1999) and the EQSAM gas/aerosol partitioning model (Metzger et al., 2002). Traditionally, this model has been run in a domain centered over Europe, but also covering large parts of the North Atlantic. The European-scale model has a resolution of about 50 × 50 km 2 , true at 60 • N polar stereographic projection, and extending vertically from ground level to the tropopause and the lower stratosphere (100 hPa). The model has undergone substantial development in recent years, and is now applied on scales ranging from local (ca. 5 km grid size, Vieno et al., 2010) to global (with 1 degree resolution, Jonson et al., 2010). The model presented in this paper uses version rv3.5, see Simpson et al. (2010). Both natural and anthropogenic emissions are included. The anthropogenic emissions are provided by most European countries on the 50 × 50 km 2 grid, otherwise derived from global databases or expert estimates. Biogenic emissions of isoprene in Europe are based on Guenther et al. (1993) and Simpson et al. (1999), driven by landcover for the appropriate grid. Emissions from forestfires are available as 8-day averages from the GFED (Global Emission Fire Database) database of van der Werf (2006 (Källén, 1996;Lenschow and Tsyro, 2000). The performance of the EMEP model varies of course with the meteorological driver, but differences are modest for most pollutants, and affected mainly species subject to wet deposition or sensitive to near-surface dispersion, two factors that have little impact on column simulations of CO. Tarrason et al. (2008) discuss the differences in more detail, and Sect. 6 will present results from both European and global model versions for comparison.
When run at the European scale, initial and boundary values are required for the important long-lived pollutants, notably O 3 , CH 4 , CO and some hydrocarbons including C 2 H 6 . Concentrations of O 3 are most crucial to the model's photochemical calculations, and these are derived from climatological data of Logan et al. (1999), modified with a socalled Mace-Head correction to correct for observed monthly background O 3 changes (Simpson et al., 2003c). CO and C 2 H 6 are specified as simple functions of latitude, altitude and time-of-year. The values were chosen to loosely reproduce observations from a number of studies (e.g. Derwent et al., 1998;Ehhalt et al., 1991;Emmons et al., 2000;Warneck, 2000) and the equations for the boundary condition calculations for CO and C 2 H 6 are presented in Eqs. (1) and (2).
The CO and C 2 H 6 concentrations at ground-level (C 0 ) are calculated as a function of the average concentration (C mean ), the amplitude ( C) and the phase where d mm is the mid month number, d max is the day at which C 0 maximizes and n y is the number of days each year. Vertical (profile) concentrations are calculated according to Eq. (2), where h is the height and H z is the scale height with, 10 km used for C 2 H 6 and 25 km for CO.
The parameters used in Eqs. (1) and (2) can be found in Simpson et al. (2003c). The European EMEP model has been compared with observations and trends in several studies. Jonson et al. (2006) examined trends in ground-level ozone, Simpson et al. (2006) looked at deposition fluxes of sulphur and nitrogen, Fagerli et al. (2007) compared historical trends of sulphate, ammonium and elemental carbon against ice-core records in the Alps, and Fagerli and Aas (2008) examined trends in nitrogen compounds in air and precipitation from 1980-2003. In recent years the EMEP model has been extended to run at the hemispheric and global scales. Indeed, calculated results with an earlier hemispheric version of the model have been uploaded to the HTAP database (Hemispheric Transport of Air Pollution, www.htap.org). More than 30 models have uploaded model calculations for the year 2001 on this database, with model intercomparisons and comparison with measurements reported in several studies (HTAP, 2007;Fiore et al., 2009;Shindell et al., 2008;Sanderson et al., 2008;Reidmiller et al., 2009;Jonson et al., 2010). In general the EMEP model performs well compared to the other models in the database for surface species and depositions. In the free troposphere the EMEP model tended to underpredict ozone in the summer months compared to other models and to measurements (Jonson et al., 2006). For tropospheric CO, the EMEP results were at the high end of the ensemble of the hemispheric CO calculations. Since this HTAP study, changes have been made to the advection scheme (in particular with the inclusion of a convection scheme for dealing with e.g. sub-grid cumulus clouds), and ozone levels in the free troposphere are now higher, and in much better agreement with measurements (Jonson et al., 2010). These revisions have also led to substantially lower CO in the free troposphere, and results should now be close to the TF HTAP ensemble mean. (Within TF HTAP a new set of model calculations are planned for 2012. The EMEP model will take part in this model intercomparison). As the convective scheme is such a new addition to the EMEP model, and such schemes are more uncertain , we run here versions with and without the scheme.
Concerning emissions and chemical species, it should be noted that there is a large difference between CO and C 2 H 6 in the model system. CO is a well-defined pollutant with its own emission databases. Such CO emissions data are thought to be fairly reliable over Europe and North America at least, although with larger uncertainties in other parts of the globe. On the other hand, C 2 H 6 is a compound whose emissions are not explicitly mapped within the EMEP model. Instead, emissions of non-methane volatile organic compounds (NMVOC) are available, distributed in 10 socalled SNAP sectors, with e.g. SNAP4 representing production processes, SNAP5 extraction and distribution of fossilfuels, SNAP7 representing motor vehicles, etc. For each of these SNAP categories we have a default speciation profile, in which the percentage of C 2 H 6 is specified. This profile is largely based upon emissions data from the UK however (Passant, 2002), and its validity in other areas is questionable. As C 2 H 6 has very particular emission sources which will not be captured by such a simple percentage-contribution sys-tem, e.g. from gas-leakage, it is actually unlikely that the model's emission inventory for this compound is of sufficient quality to derive a meaningful trend analysis. Further, the model's C 2 H 6 is actually a surrogate compound, representing not just real C 2 H 6 but also some other low-reactivity compounds. For these reasons we will present the FTIR trends of CO and C 2 H 6 , but concentrate on the modeled trends of CO.

Method
The European-scale version of the EMEP model is first run over the time-period 1996-2006. These runs are intended to firstly evaluate the model's CO and (to a lesser extent) C 2 H 6 fields, and secondly to determine the extent to which the changes in emissions and chemistry over the European domain can account for changes observed in the FTIR network. The global-scale EMEP model is essentially the same as the European-scale model, but not so dependent on initial and external boundary conditions, and driven by a different meteorological driver. As well as accounting for intercontinental transport, the global model is also able to account for recirculation of those air masses from Europe which pass beyond the boundaries of the European model. A comparison of the global model against FTIR measurements and the European model is also performed. The global-scale model is then used in a series of sensitivity scenarios to investigate possible cause for the CO trends seen in the FTIR measurements. The model versions and sensitivity scenarios are summarized in Table 3.

Calculations of partial columns
Since both the European and global-scale EMEP model domains only extend up to 100 hPa, whereas the FTIR data extends throughout the whole atmosphere, partial columns from the FTIR measurements have to be derived for comparability. The partial columns consist of the tropospheric columns and those parts of the stratospheric columns that are below 100 hPa. These are simply calculated by subtracting the average part of the total columns above 100 hPa from the total columns. The averages are based on the retrieved FTIR profiles from Harestua and Jungfraujoch in the CO case and Harestua in the C 2 H 6 case. The CO average is 6.1 ± 2.2 × 10 16 molecules cm −2 and the C 2 H 6 average is 5.0 ± 2.2 × 10 14 molecules cm −2 , corresponding to ∼3-6 % and ∼2 % of the total columns for each species, respectively. The uncertainties are presented on a 2-σ level. FTIR partial columns, in this case above 12 km, have earlier been reported to be at most 10 % and 7 % of the total column for CO and C 2 H 6 , respectively (Zhao et al., 2002).
The EMEP model data was obtained from discrete constant layers for both the European and global-scale models. To adjust the model to the actual altitude of the FTIR measurement stations, a linear interpolation, based on the difference between the station altitude and average topographical altitude in the grid area used by the EMEP model, was applied. Fagerli et al. (2007) discussed this issue in more detail, but for this exercise, the simple height-based interpolation is assumed to be sufficient for the model evaluation. The trend sensitivity study is not affected by the model baseline, i.e. the interpolation, since it is based on relative changes.

Trends and tracers
The FTIR trends are estimated with a method developed by Angelbratt et al. (2011). The method is based on multiple linear regressions where other data such as surface pressure and total column of hydrogen fluoride (HF) are used to reduce the variability in the FTIR time series and thereby explain part of the atmospheric dynamics resulting, for example, from tropopause changes or from the presence of the polar vortex. The model also contains a function that captures the seasonal fluctuations and the linear trend. Trends are estimated from the EMEP model by using simple linear regression. When estimating confidence intervals for the trends both of the above-mentioned methods assume that the residuals (data minus trend model) are normal distributed, are free from auto-correlation and have equal variance around zero. The method of confidence interval on individual regression coefficients described in Montgomery et al. (2008) is used. Trends are not estimated for Bremen due to the short time series and Ny-Ålesund due to the lack of measurements during winter.
Tracers for CO and C 2 H 6 have been introduced into the EMEP model, in order to track concentrations originating from the boundary and initial conditions (BICs). Tracers are lost to OH using the same rates as for CO and C 2 H 6 , but the tracers do not influence the chemical simulations. Thus, at the start of the simulation the modeled concentrations of CO and C 2 H 6 are identical to their tracer counterparts. As the simulation proceeds, emissions and chemical production and loss affect the real species, but the tracers respond only to advection and chemical loss. As the European-scale model has a limited spatial domain it can be heavily influenced by BICs, so that the tracer concentrations frequently amount to a large percentage of the real concentrations. The global-scale model is mainly influenced at the start of the simulation (see Sect. 6).

Sensitivity analysis for CO trends
To find possible causes for the estimated CO trends in the FTIR dataset, see Sect. 6.1.2, a series of sensitivity scenarios are tested in the global-scale model; the scenarios are summarized in Table 3 and are described in detail below. As a baseline-simulation we use the standard model, with convection, but without forest fires (Gc) and all the sensitivity scenarios are compared with this simulation. The scenarios have been chosen to loosely represent known emission changes or climate effects and the mass balance is schematically described in Eq. (3), where P represents the production terms and D the destruction of CO in Tg yr −1 . In this paper, P Anthropogenic is defined as the CO emissions, mainly from combustion in industrial processes and the transport and energy sector while forest fires, savannah and agricultural waste burning are included in the P Biomassburning term. The P CH4oxidation and P BVOCoxidation are defined as the formation of CO from the oxidation of these species with the OH radical. CO can also be formed in biological processes in soils and oceans and these comprise the P Biological term in Eq. (3). The main sink of CO is the reaction with the OH radical and thereby the destruction term D OH . Trend = P Anthropogenic + P Biomass burning In the first two scenarios a reduction of anthropogenic CO emissions by 20 % in Europe and North American are tested (GcEUR20 and GcNA20) and this corresponds to a decrease per year of ∼2 % during the 11-year period of 1996-2006 for which the FTIR measurements are taken. In  and the EDGAR database v. 4.0 (2009) the anthropogenic CO emissions in Europe have been shown to decrease by 1.7 % yr −1 to 4.5 % yr −1 and the North American emissions have been show to decrease by 2.1 % yr −1 to 3.7 % yr −1 , where the trends are based on data from the late 1990s and early 2000. The reported emission reductions for both continents indicate that our reduction of 2 % yr −1 is a reasonable value to use in the two sensitivity scenarios. Further, we have decided to use a reduction in the lower range for both of the scenarios so the change in the modeled CO partial columns is not overestimated. Unlike North America and Europe, many countries in Asia have increased their CO emissions during the last decades. In the third scenario (GcEA20) we investigate how a 20 % increase in the East Asian anthropogenic CO emissions is affecting the CO trends seen in Europe. We define East Asia as China and Japan, and for model simulations use the same area as defined by Fiore et al. (2009). According to Monks et al. (2009) In scenarios four and five, GcEurAll20 and GcNAAll20, the emissions from all anthropogenic sources (CO, NO x , NMVOC, SO x , and NH 3 ) are reduced by 20 % for Europe and North America (again using the Fiore et al. (2009) domain definitions). Since both CO and the other species have decreased, these scenarios illustrate more realistic situations where the chemical interactions between the species are included in the simulation. Scenario six corresponds to a 20 % increase for the same species as in four and five but for East Asia (GcEAAll20).
In scenario seven the global surface temperature as used in the BVOC calculations are increased by 0.2 • C (GcT0.2). According to Hansen et al. (2006) the global temperature has approximately increased by 0.2 • C over the study period presented in this paper and since the BVOC emissions are strongly temperature dependent, such emissions should also have increased.
In the last sensitivity scenario (GcCH 4 1.2) it is investigated how the partial columns of CO are affected by the methane (CH 4 ) concentrations. Both surface concentration and total column measurements of CH 4 have been shown to increase from 1996 to 1999 and thereafter flatten out towards a zero trend for the rest of the studied period (Dlugokencky et al., 2009;Angelbratt et al., 2011). The average increase over the first three years was 0.4 % yr −1 . A proper modeling assessment of these CH 4 changes would need to run over many years because of the lifetime of CH 4 (∼10 years) ). Here we make a first order approximation and calculate the effect of a single 1.2 % yr −1 emission change calculated over one year (2006).

Uncertainty analysis
A number of other tests have been conducted with the global model, which explores the importance of some model assumptions. As noted in the introduction, the convection routine is new to the EMEP model and optional. Although convection is unquestionably important to atmospheric transport, the parameterization of this in models is also quite uncertain . For this reason we have run a model version (G, Table 3) without convection. A second test with the global model is the introduction of forest-fire emissions. Although such emissions should in principal be part of a default model run, we only have 8-day average emissions. Thus, when comparing with daily FTIR data, we have chosen to omit these data from the baseline-case, but in test Gcff forest-fire emissions are added for comparison. As discussed in Sect. 5.1 (Fagerli et al., 2007), model results are derived as an interpolation between discrete model layers, and some uncertainty is associated with this procedure. Two tests (Gchigh, Gc-low, Table 3) are conducted in which the model results are taken from model-layers higher than, or lower than, this default. The uncertainty in the derived partial columns, Sect. 5.1, is investigated in two tests, FTIR-high and FTIRlow. Finally, we have conducted one more, extreme, test of the BVOC emissions, namely setting all such emissions to zero (Gcnobvoc). This test is designed purely to explore the magnitude of CO associated with BVOC emissions. Such emissions are in fact one of the most uncertain inputs to the CTMs, at least over North America and Europe where other emissions are known with reasonable accuracy, and according to Granier et al. (2000), BVOC contribute up to 18 % of the global budget of CO. Simpson et al. (1999) estimated a factor 2-3 uncertainty for European isoprene emissions (and much worse for other BVOC), and Warneke et al. (2010) found factor of two differences between two BVOC inventories in the United States, with measurement-based data unable to distinguish which was best.     Table 4. Compared to the low-level sites, the Alpine stations have a lower average partial column and seasonal amplitude due to their high altitude since a large part of the partial column is located below the station. As noted in Sect. 4, the European-scale model uses climatological (monthly) BICs for CO and C 2 H 6 , and these values strongly determine the model's concentrations, and hence are responsible for much of the good agreement. The BIC influence is quantified by tracers and those are presented in Table 5, where it can be seen that columns at the modeled stations close to the model horizontal boundary are dominated by the BICs, while columns at the stations further away from the boundary include more information generated within the model. It can also be seen that more information is from the BICs for C 2 H 6 than for CO.
A clear difference in CO between the model and the measurements can be seen in the years 1998 and 2002/2003 for most of the stations and this is particularly visible at Jungfraujoch and Kiruna in 1998. For C 2 H 6 these differences are not so clear although they can be seen for example at Jungfraujoch and Kiruna in 1998. During these two time periods, large scale forest fires were present in North America and Russia and the CO contributions to the atmosphere were captured by the FTIR measurements (Yurganov et al., 2004(Yurganov et al., , 2005. As noted in Sect. 4, forest fire emissions were not available at consistent time-resolution over the 11 years of this study, and so were omitted from the Europeanscale calculations. Furthermore, and probably most importantly, the model domain does not include North America and non-European parts of Russia and is hence highly dependent on the lateral boundary conditions. Compared to the measurements, the model shows a slight phase shift for all the participating stations regarding CO and C 2 H 6 . For CO, the model tends to overestimate the seasonal amplitude and underestimate the average partial column. For C 2 H 6 , on the other hand, the model overestimates both the average partial column and seasonal amplitude and deviates from the FTIR data by as much as a factor of two. For both species, the model captures the inter-station variability quite well for both the average partial columns and the seasonal amplitudes. Although C 2 H 6 has much less importance for EMEP modeling purposes (usually aimed at boundary layer ozone, or acidification and health issues) than CO, these results suggest a need to modify the global boundary conditions for the European EMEP model, and to re-evaluate the emission inventories for this compound.

Trends
All FTIR stations show significant negative CO trends on the 2-σ level, as presented in Table 6. For the 1996 to 2006 time Table 6. Linear trends estimated from the partial columns (below 100 hPa) of CO and C 2 H 6 from FTIR measurements. The trends are presented with their 2-σ confidence intervals and used the average value of 2001 as reference.  Significant negative C 2 H 6 trends are estimated for all FTIR stations, see Table 6. The trends are stronger compared to those of CO and vary from −2.25 ± 0.35 % yr −1 to −1.09 ± 0.25 % yr −1 . Again Harestua and Kiruna are in close agreement and Jungfraujoch and Zugspitze differ despite their proximity.
To the author's knowledge, very few trend estimations have been performed for C 2 H 6 . Except the negative trends of 1.20 ± 0.35 % yr −1 and 2.70 ± 0.30 % yr −1 given in  and Mahieu et al. (1997), the trends in this paper could be  Gardiner et al. (2008). No significant trends for CO and C 2 H 6 have been found in the European-scale EMEP model. Since the model boundary conditions do not contain any trend components and most of the information at each station comes from the boundary conditions, see Table 5, we conclude that the model domain is too small for detecting trends from species with a lifetime as long as two months. To model a more realistic alternative regarding CO trends, the global-scale model will be used in Sect. 6.2.2.

Evaluation of global-scale model
The comparison of the global and European-scale model with the FTIR measurements for 2006 is presented in Figs. 3 and 4 for CO and C 2 H 6 , respectively. The average values for each station and species are presented in Table 7.
The global-scale model reproduces the FTIR data slightly better than the European model for the partial columns of CO at all participating stations. The biggest difference is at the stations located at higher latitudes where the European model tends to overestimate the partial columns while the global model is in good agreement with the measurements. The overestimation by the European model at high latitudes is probably due to the high influence from the boundary conditions at these locations. Both the global and European model have problems reproducing the partial columns of C 2 H 6 . The global model significantly underestimates the partial columns at all stations and for some by as much as a factor of two. The European model also underestimates the partial column but not as much as the global model.

Sensitivity of CO trends
The outcome of the global-scale model sensitivity scenarios for CO is presented in Table 8 as percent change per year for each of the participating stations. The outcome is based on average values of all sensitivity and baseline scenarios where January and February data is removed due to the high influence from the initial conditions. The differences between the baseline scenario and the sensitivity scenarios is then scaled with the 11 year period (1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006). The European anthropogenic CO reduction by 20 % (GcEUR20) has the largest impact on the modeled partial columns of all scenarios and could by itself account for a negative trend of 0.47-0.83 % yr −1 . Also, a 20 % reduction in the North American anthropogenic CO emissions (GcNA20) causes negative trends in the modeled partial column from 0.20-0.22 % yr −1 . The increase in the East Asian anthropogenic CO emission by 20 % (GcEA20) gives a positive contribution to the modeled European partial column trends of 0.15-0.18 % yr −1 ; this region has a slightly smaller impact on the absolute European trends than North America. A global increase in the CH 4 column (GcCH 4 1.2) and an increase in the global temperature by 0.2 • C during the 11-year period give a positive contribution to the modeled trends of ∼0.15 % yr −1 for all stations. Although these two scenarios are rough estimations, this highlights the fact that CO seems to be sensitive to both the global temperature and the CH 4 concentration.
When adding the five sensitivity scenarios discussed above (shown as Gc in Table 8), Jungfraujoch, Harestua and Kiruna have modeled trends that are close to the measured trends, while the modeled trend at Zugspitze deviates by a factor of two from the measured trends. The modeled trends follow the measured trends with a smaller trend at Jungfraujoch and larger trends at Harestua and Kiruna. It can also be seen that the modeled trend at Zugspitze is larger than the trends for Jungfraujoch. To exclude the altitude difference between the stations as a reason for this behavior, the trends at the ground layer are also modeled; this is presented in Table 9. In the comparison it can be seen that the trend dif- ference is 0.07 % yr −1 at the ground while it is 0.10 % yr −1 when adjusting for the station levels. From this we conclude that the modeled trend difference between Jungfraujoch and Zugspitze is not due to the different altitudes of the stations but rather has to do with the origin of the air masses at each station. This fact might also be an additional explanation for the measured trend difference between Jungfraujoch and Zugspitze, presented in Sect. 6.1.2. The modeled trends at Bremen and Ny-Ålesund stands out a bit compared to the other stations. The small trend at Ny-Ålesund is not surprising since the station is located in the far north and is hence less affected by the European CO reduction. Bremen on the other hand has a central location and is affected by European CO reductions for all wind directions. The 20 % reduction and increase in emissions of all anthropogenic species (GcEURAll20, GcNAAll20, GcEAAll20) represents the change in OH chemistry that occurs due to a change in the NO x , NH 3 , NMVOC and SO x concentrations and thereby the assumed change in the CO partial columns. It was shown that the difference between the Gc20 and Gc20All scenarios hardly differs at all between the model calculations for the three regions. Since OH affects the CO concentrations both as (1) a sink and (2) a source through the oxidation of VOCs and CH 4 , it turns out that these two processes cancel out each other and that the modeled CO partial columns are almost insensitive to changes in NO x , NH 3 , NMVOC and SO x .
When considering the mass balance for CO in Eq.
(3) we have investigated the presence of trends in three sources, namely: anthropogenic CO emissions, oxidation of CH 4 and BVOC (through the GcT0.2 scenario). Possible trends in biomass burning and in the OH radical concentration due to factors other than the emission changes explored above have not been taken into account. Both the OH radical and biomass burning have large inter-annual fluctuations, this is for example shown by Yurganov et al. (2004Yurganov et al. ( , 2005 and Montzka et al. (2011), and these fluctuations contribute to the uncertainties in the estimated FTIR trends. To outline the effect on the CO trends of possible changes in the OH radical and biomass burning further studies are needed.

Uncertainty analysis
As discussed in Sect. 5.4, we have conducted a number of tests designed to quantify some of the uncertainties in the calculations, concerning convection, forest-fires and BVOC emissions as well as uncertainties related to the derivation of partial columns from the FTIR measurements (FTIR high and low) and the interpolation between the model layers (Gc high and low). In Fig. 5 the results of these tests are illustrated for Harestua, the results are similar for the other stations participating in this study. It can be seen that the model with the forest fire module (Gcff) in general reproduces the FTIR measurements best compared to the other two versions (G and Gc) where G is close to Gcff while Gc is underestimating the measurements. This illustrates that the convection module decreases the estimated partial columns while the forest fire module increases the partial columns and that the two modules together almost cancel out each other. It is also seen that the derivation of the partial columns from the FTIR measurements is almost insensitive to the seasonal variation of the 100 hPa level. This is not unexpected since a very small fraction of the partial column is located around this level. The model-FTIR comparison is more sensitive to the interpolation between the layers and the BVOC emissions and when removing all North American BVOC emissions a large underestimation of the partial columns of CO is seen.

Conclusions
In this paper we have shown negative linear trends from partial columns of CO and C 2 H 6 measured with the groundbased solar FTIR technique at four European stations, Jungfraujoch, Zugspitze, Harestua and Kiruna, from the time period of 1996-2006. The negative CO trends vary from 0.45 ± 0.16 % yr −1 at Jungfraujoch to 1.00 ± 0.24 % yr −1 at Zugspitze and to ∼0.60 ± 0.20 % yr −1 at both Harestua and Kiruna. A similar pattern can be seen in the negative C 2 H 6 trends but here the trends are stronger and cover a range from 1.09 ± 0.25 % yr −1 to 2.11 ± 0.30 % yr −1 . All the estimated trends are significant on a 2-σ level, indicating that a decrease in the two species has occurred in Europe during the 1996-2006 period. To outline possible reasons for the mea-   sured negative CO trends the global-scale EMEP model was used in a series of sensitivity scenarios. It was found that the reduction in the European anthropogenic CO emissions during the 1996-2006 period could, to a large extent, explain the negative trends measured at the FTIR stations. Also, the decrease in North American and increase in East Asian anthropogenic CO emissions affected the measured CO partial columns in Europe but to a smaller extent than the European emission reduction. This paper should be considered as a first attempt to explain the CO trends seen in the FTIR measurements. Meteorology for the global-scale EMEP model was only available for 2006, the analysis can be improved in future if meteorology becomes available for the whole time period (1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006). Furthermore, of great interest is the effect of the variations in the OH radical and biomass burning on the trends in CO. This has been outside the scope of this article and a more detailed analysis, on a statistical basis, is needed to quantify the exact reasons for the measured trends.