We used observational data from the NASA Atmospheric Tomography Mission (ATom). The ATom campaign took place from July 2016 to May 2018 and included measurements in four different seasons (six flights each). During the flights with the NASA DC-8 aircraft, numerous profiles were recorded (see Fig. ), which makes the dataset perfect for the evaluation of atmospheric model simulations. The data used in this evaluation was taken from the “Merged Atmospheric Chemistry, Trace Gases, and Aerosols, Version 2 dataset” and are based on the following instruments: the UC-Irvine Whole Air Sampler (WAS; ), the Trace Organic Gas Analyzer (TOGA; ), the California Institute of Technology Chemical Ionization Mass Spectrometer (CIT-CIMS; ), the NOAA Chemical Ionization Mass Spectrometer (NOAACIMS; ), the Georgia Tech Ionization Mass Spectrometer (GTCIMS; ), the NOAA Nitrogen Oxides and Ozone Instrument (NOyO3; ), and the NOAA Picarro G2401 spectrometer (NOAA-Picarro; ). The ATom observations were subdivided into data for different regions for this study (Fig. ; namely Northwest Pacific, Southwest Pacific, East Pacific, Southern Ocean, South Atlantic, North Atlantic, and North Canada/Alaska/Greenland) based on expected homogeneous properties of organic compounds in those regions. A climatological comparison between the simulated year (2010) and the data from the ATom campaign (2016–2018) will be performed in order to further evaluate whether the MOM mechanism is able to reproduce remote conditions. For that purpose we divided the altitude range into 12 bins (linearly between 0.5 and 12.5 km) and defined a maximum of 336 (7 regions × 4 seasons × 12 altitude bins) points of interest (POI). For each POI we compared the mean value of all observations in the specific region, altitude bin and season to the mean value of all accordingly simulated values in 2010 in the range of ±1 d around the flights conducted in that region.

We also used the database of aircraft measurements of trace gases produced by . It is based on observations from numerous aircraft campaigns that took place during the period 1990–2001 to create observation-based climatologies of chemical species relevant to tropospheric chemistry. Although these measurements cover only limited time periods and regions, they provide valuable information about the vertical distribution of the analysed trace gases. Note that the field campaigns used in this evaluation have been extended to also include observations after the year 2000, such as those from the TOPSE and TRACE-P campaigns.

The aircraft evaluation is completed by aircraft measurements for organic aerosols presented by . This dataset includes organic aerosol measurements from 17 aircraft campaigns during the period 2001–2009. Here, similar to , we consider the aircraft campaigns representative for the period and the regions, i.e. as an observation-based climatology. Therefore, we compared these data with the model results for the year 2010, yielding an improved evaluation of the vertical distribution of the organic aerosols in the troposphere.

Data from the National Oceanic and Atmospheric Administration (NOAA) Institute of Arctic and Alpine Research (INSTAAR) Global Monitoring Program were also used in this study. The network consists of 44 background stations from the NOAA Global Greenhouse Gases Reference Network (GGGRN) , where pairs of whole air samples are collected weekly and shipped to a central laboratory for analyses. For this study, we considered measurements of ethane, propane, iso-butane and n-butane from this network.

The Clean Air Markets Division of the U.S. Environmental Protection Agency (EPA) operates the Clean Air Status and Trends Network (CASTNET), with a total of 97 stations. Here, we include weekly filter pack data from 81 stations for SO42-, NO3-, NH4+, Na+, Mg2+, Ca2+, K+ and Cl- for the year 2010. The data can be downloaded at https://www.epa.gov/castnet (last access: 2 March 2022).

In order to evaluate the simulated regional background OA concentrations, we use monthly averaged PM2.5 OA measurements during the year 2010 from the Interagency Monitoring of Protected Visual Environments (IMPROVE) program (http://views.cira.colostate.edu/fed, last access: 21 September 2021). IMPROVE is a cooperative measurement effort in the United States designed to characterize current visibility and aerosol conditions in scenic areas (primarily national parks and forests). This network includes 198 monitoring sites that are representative of the regional haze conditions over North America. IMPROVE co-located samplers collect 24 h samples every 3 d.

The co-operative programme for monitoring and evaluation of the long-range transmission of air pollutants in Europe (EMEP) is a science-based and policy-driven programme under the Convention on Long-range Transboundary Air Pollution for international co-operation to solve transboundary air pollution problems. We use data from 42 stations, although the number of stations providing sufficient data for a certain species ranges from 22 to 42 (see Table ). As Teflon filters are predominantly used, depending on the respective station, the concentrations of particulate nitrate can be systematically underestimated . The data were obtained from the EBAS database (http://ebas.nilu.no/, last access: 21 September 2021), and stations that did not provide full coverage of monthly mean values for the year 2010 were excluded from the analysis.

The Acid Deposition Monitoring Network in East Asia (EANET) regularly has monitored acid deposition since January 2001. A total of 13 countries currently participate in this effort, submitting data from a total of 54 sites. We use monthly averaged aerosol concentrations from 15 to 25 stations for the year 2010, with the data obtained from the Network Center for EANET, which are archived at https://monitoring.eanet.asia/document/public/index (last access: 2 March 2022). It is worth mentioning that EANET does not include OA observations. The simulated PM2.5 OA over East Asia is evaluated against collected short-term measurement data as summarized by .

MOPITT (Measurement of Pollution in the Troposphere) is a sensor onboard the NASA's Terra satellite. We use the MOPO3JM version 8 product of MOPITT, which provides monthly mean gridded column-integrated CO, vertical profiles of CO mixing ratios at 10 regularly spaced pressure levels from the surface up to 100 hPa and the corresponding averaging kernels . The MOPO3JM product is based on the joint near-infrared (NIR) and thermal-infrared (TIR) retrievals of CO. We apply the MOPITT averaging kernels to the EMAC CO profiles according to Eq. () to ensure the same level of smoothing as that in the MOPO3JM product. 1xrtv=Axtrue+(I-A)xa Here, xtrue is the EMAC profile, xa is the a priori profile, A is the averaging kernel matrix and I represents an identity matrix.

The MODerate resolution Imaging Spectroradiometer (MODIS) sensor is also located on the Terra satellite. Here, AOD550 data (i.e. aerosol optical depth at 550 nm) from the MODIS Level 3 (Col. 6.11) gridded product are used at a spatial resolution of 1∘×1∘. The data are available through the Atmosphere Archive & Distribution System (LAADS) (https://ladsweb.modaps.eosdis.nasa.gov/, last access: 21 September 2021). The Deep Blue algorithm was used here for the aerosol AOD.

Global observations of a suite of VOCs are retrieved from the thermal infrared measurements of the nadir-viewing IASI (Infrared Atmospheric Sounding Interferometer) instruments . The VOC dataset exploited here consists of total column densities of methanol, acetone, formic and acetic acids, and PAN, all retrieved on a near-global and daily basis from the IASI/MetOp-A observations with the aid of the ANNI (Artificial Neural Network for IASI) v3 retrieval framework. This neural-network-based retrieval approach does not rely on a priori information of the total column densities, and hence the products can be used directly for unbiased comparisons with model data (see ). The satellite products are filtered for measurements affected by clouds and poor retrieval performance. The uncertainties on the individual retrieved column densities can be large but are considerably reduced by averaging numerous observations in space and time, as has been done in this study by comparing annual averages on the model grid. A full description of the ANNI framework, the characterization of the VOC products, and comparisons of the satellite data with independent measurements can be found in , and references therein. These studies, thanks to comparisons with ground-based column density measurements of VOCs and column densities derived from aircraft profiles, indicate no large systematic biases in the IASI data and no dependence on the latitude, although an underestimation (locally up to 30 %) of the highest column densities over tropical source regions (e.g. the Amazon basin) has been identified for all the IASI VOC products (except PAN). The accuracy of the IASI measurements is therefore sufficient to provide a global evaluation of the EMAC performance, considering the large uncertainties that still affect the emissions and atmospheric modelling of the VOCs. For both IASI and the model, daily gridded averages were constructed at the spatial resolution of the model grid. To have similar temporal coverage over the year between model and satellite observations, the EMAC daily averages were masked when the corresponding IASI data were missing for the same day and location.

The AOD observations have been obtained from the global AErosol RObotic NETwork (AERONET ). The cloud-screened quality-assured Level 2 AOD data used in this study were obtained from the website http://aeronet.gsfc.nasa.gov/cgi-bin/combined_data_access_new (last access: 21 September 2021), including the AOD daily averages.

Alkanes are VOC ubiquitously present in the atmosphere that are mainly emitted from anthropogenic activities. In Fig.  the scatterplots of model results and in-situ observations from the NOAA/INSTAR database are presented. The statistic for each plot is summarized in Table . The model seems to reproduce satisfactorily the mixing ratios of the C2–C4 alkanes with more than 98 %, 91 %, 84 % and 79 % of the simulated values lying within a factor of 2 of the observations for C2H6, C3H8, i-C4H10 and n-C4H10, respectively. Furthermore, the average ratios show that a slight overestimation is present for i-butane and n-butane, whereas the model underestimates the mixing ratios of ethane and propane. This is confirmed by the comparison with the database (Table ), which shows that both ethane and propane are underestimated by the model with respect to the vertical profile as well (column M/O‾). The comparison with ATom observations (Table ) also shows an underestimation (albeit smaller) of ethane and propane. However, for i-butane we see a slight underestimation in the evaluation using ATom observations in contrast to the NOAA/INSTAR database.

Figure  also shows a noticeable underestimation of the mixing ratios in the region between 30 and 0∘ N for propane and the butanes. This is clearly due to the underestimation of the oceanic emissions, as these stations (i.e. at Mahe Island, Seychelles (SEY); The American Samoa Observatory (SMO); Ascension Island, UK (ASC); and Easter Island, Chile (EIC)) are all strongly influenced by oceanic emissions. It must be pointed out that no oceanic emissions were included in the simulation for i-C4H10 and n-C4H10 but that these are necessary for a correct representation of butanes, as already shown previously . However, the underestimation in this region is not present in the comparison to the ATom observations.

In contrast to alkanes, the alkenes are mostly emitted by the biosphere. The comparison to the ATom campaign can only be based on a very limited number of observations, and thus it will not be covered in this discussion. As shown in Table  in the comparisons with the database, the numerical simulation of alkenes diverges more from the observations than the simulated alkanes do. Both C2H4 and C3H6 are underestimated compared to the observations (see the model mean against observed mean comparison in Table ). Nevertheless, while C2H4 is reasonably reproduced overall (the average of the ratios is equal to 1.06; see Table ), for C3H6 a strong underestimation is present (the average of the ratios is equal to 0.4; see Table ). This is shown in Fig. , where examples of the vertical distribution of C2H4 are shown. The issues in simulating C3H6 have been already pointed out by , where similar results were obtained with large underestimation of this tracer. As alkenes are mostly removed by reaction with OH, there are strong indications that these reactions are too fast; in addition, a substantial lack of emissions could be present, for instance from natural sources, as suggested by .

CH3OH is the most abundant oxygenated VOC in the Earth's atmosphere and is primarily emitted by terrestrial vegetation . Other sources include biomass burning, the oceans and secondary formation in the troposphere (see, e.g. ). The annual global distributions of CH3OH total column densities obtained from the IASI measurements and EMAC of the year 2010, as well as their differences, are presented in Fig. . Over land, the satellite and model spatial distributions of CH3OH are relatively consistent, with the main source regions observed within the tropics (mainly over the Amazon basin and central Africa), in Southeast Asia, and at Northern Hemisphere mid-latitudes and high latitudes. However, important differences in magnitude exist. Over tropical forests, EMAC indeed simulates annual CH3OH column densities up to 2 times larger than those retrieved from the IASI observations (3.0–3.5 × 1016 molec.cm-2). This can be partly ascribed to the high temperature bias that exists in the model over these areas , which induces an excess of biogenic VOC emissions from the MEGAN submodel due to its high temperature sensitivity . This is confirmed by a comparison of the simulated isoprene mixing ratios in the Amazon rainforest with observations from the AMAZE-08 campaign and from the ATTO tower . In comparison to the AMAZE-08 campaign, the simulated isoprene measurements are overestimated by more than a factor of 3 (6.1±1.2 nmolmol-1 simulated and 1.9±1.4 nmolmol-1 observed), while the overestimation factor for the ATTO tower is on average approximately 1.6, being higher in February–March and lower in October–November.

Furthermore, dry deposition of methanol in tropical rainforests is likely under-represented by the approach used here . The potential contribution of an additional non-stomatal pathway favourable under humid conditions has been shown by for a less soluble species. On the other hand, EMAC underestimates the CH3OH column densities at mid-latitudes and high latitudes of the Northern Hemisphere, indicating that biogenic and/or biomass burning VOC emissions are too low during summertime in these regions. The southeastern US is the region in North America where most of the simulated CH3OH enhancement exists, driven by substantial biogenic emissions in summertime, while larger column densities are measured by IASI during the same season further northwards in the boreal regions. Over the oceans, EMAC underestimates the observed CH3OH column densities by ∼5 × 1015 molec.cm-2, which might result from an insufficient transport from the continental source regions and/or from missing secondary source(s) (see, e.g. ). This is confirmed by the comparison of model results with the aircraft observations of the ATom campaign and the database, where the methanol model mean is considerably lower than the observed one. This is most prominent over Pacific regions in the Northern Hemisphere at lower altitudes, as depicted in Fig. . However, with respect to the vertical profile CH3OH is not generally underestimated in comparison with the ATom campaign, as it is instead overestimated at higher altitudes and for lower absolute values.

Acetone is the one of the most abundant oxygenated VOCs in the Earth's atmosphere after CH3OH . Its sources include the terrestrial vegetation, the oxidation of hydrocarbon precursors of biogenic and anthropogenic origin, the oceans, and biomass burning (e.g. ). The acetone column densities obtained from the EMAC simulation and the IASI observations exhibit major discrepancies in terms of global distribution (Fig. ). The satellite instrument detects the largest acetone column densities (up to 2 × 1016 molec.cm-2) at mid-latitudes and high latitudes of the boreal hemisphere, ascribed to important emissions of biogenic precursors during summertime , but a moderate burden at low latitudes and over the tropical forests. It is worth noting that this pattern is in agreement with the global acetone measurements obtained with the ACE-FTS satellite limb sounder . Conversely, EMAC simulates strong hotspots of acetone within the tropics, especially over South America and Africa, with annually averaged column densities larger than 2 × 1016 molec.cm-2, but underestimates the acetone abundance by up to a factor of 2 in the Northern Hemisphere. Such a mismatch between satellite and model distributions points to major deficiencies in the current emission inventories of acetone and its precursors. For example, the extremely large acetone column densities simulated by EMAC over the Persian Gulf and the Indo-Gangetic Plain can be attributed to the highly elevated emissions of propane – an important precursor of acetone – in these regions. In vegetated regions, an underestimation of acetone dry deposition, accounting for 20 % of the total loss globally , likely contributes to the mismatch. According to measurements, significant amounts of acetone are removed during the night through non-stomatal uptake . However, over remote areas and the oceans, the simulated acetone abundance is consistent with the IASI measurements, with vertical column densities in the range of 0.5–1.0 × 1016 molec.cm-2. The model nevertheless simulates a slight overestimation over the oceans at low latitudes compared to IASI, especially in the outflows of the tropical hotspots.

Compared to the ATom observations the model overestimates the mixing ratios, especially at high altitudes, leading to a large value of M/O‾, while the simulated and observed mean are of a comparable magnitude. However, 74.3 % of the simulated values are still within a factor of 2 of the observations (see Table ). Figure  depicts the vertical column densities of observations and simulations over the northern part of North America and Greenland, a region where CH3COCH3 is largely underestimated by the model in comparison to IASI. In Northern Hemisphere summer (July–August; ATom-1) this strong underestimation can also be observed for the ATom observations. However, for other seasons this is not the case. Overall, simulation results show only a small underestimation for this region. Compared to the database, the model results present a much better agreement, with only a somewhat small bias (see Table ). This apparent agreement with the observations is due to the uneven distribution of the field campaign present in the observational dataset (see , their Fig. 1, and , their Fig. 1). In fact, most of the campaigns in which acetone was measured took place over the ocean, where the model is correct or slightly overestimates the total column densities observed by IASI. In contrast, only few campaigns in the Northern Hemisphere are present in the dataset (such as ABLE-3B and POLINAT-2) for which the model strongly underestimates the observed values. As a result of the sparsity of these observations, Table  presents a quite fair agreement between model and aircraft observations, an agreement which is, however, not corroborated by the more spatially complete comparison with the IASI total column densities.

Formic acid is the dominant organic acid in the troposphere and a product of the degradation of a large suite of VOC precursors, but its observed abundance is generally severely underestimated by state-of-the-art global models (e.g. ). Similarly, in our simulation the EMAC model largely underestimates the HCOOH total column densities derived from the IASI observations by up to a factor of 4, particularly in remote environments (see Fig. ). Although the two main tropical source regions identified by IASI – the Amazon basin and central Africa – are reproduced by EMAC, the magnitudes of the simulated HCOOH column densities are too low in comparison to the satellite measurements. It also has to be mentioned that the apparent agreement over Amazonia is possibly due to the high temperature bias in the region and the subsequent excess of simulated isoprene emissions during the dry season. Over the other source regions (e.g. Southeast Asia and the southeastern US), the model underestimation is more pronounced, particularly in the Northern Hemisphere. The large enhancement of HCOOH column densities observed with IASI over western Russia (∼8×1015 molec.cm-2), attributed to the August 2010 wildfires, is not reproduced by EMAC (<2×1015 molec.cm-2) and suggests that the biomass burning emissions of VOCs are underestimated.

The general underestimation of simulated versus observational data is also confirmed by the comparison with aircraft observations. The model strongly underestimates the observations by the ATom campaign, especially at low altitudes and for high absolute values (see Table , the observed mean is higher by more than a factor of 4 compared to the model mean). In general, there is very low agreement between simulated values and ATom observations, as only around 35 % of the simulated values are within a factor of 2 of the observations and the Pearson correlation coefficient is only 0.3. Although the underestimation of the model results compared to the database is less apparent, it must be noted that only a limited number of merged data (53; see Table ) is available for the comparison with aircraft observations. It must be stressed that showed the importance of in-cloud chemistry for this tracer, which is missing in this study. The lack of adequate in-cloud chemistry would therefore explain the almost ubiquitous underestimation of this tracer.

Acetic acid is the second most abundant carboxylic acid in the troposphere and, based on the IASI retrievals, presents a spatial pattern, regional seasonality and vertical abundance that resemble those of HCOOH . Like the latter, CH3COOH is produced from the oxidation of various tropospheric precursors but has emission factors from biomass burning that are 3 to 10 times larger than those of HCOOH . CH3COOH is more difficult to detect in the infrared IASI spectra, and its retrievals are subject to larger uncertainties, particularly over ocean (see ). Therefore, here we limit the comparison with EMAC to the continents, excluding measurements over desert areas that are altered by surface emissivity artefacts (Fig. ). From the comparison, conclusions similar to those of HCOOH can be drawn for CH3COOH: the two main tropical source regions are relatively well reproduced by EMAC (with a better agreement over Africa in this case), whereas the observed CH3COOH levels are underpredicted in the Northern Hemisphere by up to a factor of 4. The comparison with aircraft observations (see Table ) again confirms this strong underestimation over the ocean as well, as the only campaign in the dataset including CH3COOH measurements is the PEM-Tropics-A, which was performed over the Pacific ocean.

These results confirm that the VOC emissions and oxidation pathways leading to the formation of CH3COOH in the troposphere are still poorly understood and constrained (e.g. ). For example, acetaldehyde – a major CH3COOH precursor via its reaction with OH – is well known to be largely underestimated by global models . On the other hand, the large model versus observations differences in Southeast Asia also point to missing emissions from biomass burning. Finally, in-cloud chemistry could be important in the CH3COOH formation, analogous to formic acid , and this process could bring model results and observations to a closer agreement.

Owing to its complex photochemical sources and its thermal instability, PAN (peroxyacetyl nitrate), the main tropospheric reservoir species of NOx, is a very challenging tracer to simulate. The comparison with the IASI data reveals that the model correctly reproduces the main spatial patterns of PAN, with the source regions and main outflows over the oceans being correct. However, the model constantly underestimates the observed PAN column densities over the globe. The satellite column densities are indeed between 2 and 4 times the simulated ones, with the most pronounced negative bias observed at Northern Hemisphere mid-latitudes and high latitudes. The same conclusion can be drawn from the comparison with aircraft observations (Tables  and ), for which the model clearly underestimates the observed mixing ratios consistently using the database from and the measurements of the ATom campaign. A closer inspection reveals that PAN is especially underestimated in the middle and upper troposphere. The model–observation discrepancies can mostly be attributed to the unsatisfactory representation of different VOCs. For example, we have seen that the model results do not agree with observations for acetone, which is an important precursor of the peroxyacetyl radical in the free troposphere . Furthermore, the strongest model underestimation appears to be exactly where CH3COCH3 the most underestimated, confirming that deficiencies in simulated precursor patterns are the main cause of the PAN biases. Additionally, sensitivity simulations suggested that the PAN formation in global models is more sensitive to the representation of VOCs than the one of NOx . Finally, an insufficient vertical transport of VOCs from the planetary boundary layer to the free troposphere in the model might also reduce the amount of PAN because PAN is more stable at the lower temperatures of the free troposphere.

Observed particulate sulfate concentrations are reproduced well by the model. The monthly mean concentration is matched closely for North America (EPA) and slightly underestimated for Europe (EMEP) and East Asia (EANET) (see Table ). For EPA and EMEP data, more than 80 % of the simulated monthly mean concentrations lie within a factor of 2 of the observations, as do more than 70 % of the concentrations for EANET. The standard deviation of observed monthly mean values is lower than the root-mean-square error (RMSE) for all monitoring networks, which is an indication for a good quality of the model results . In addition, RMSE is lower in the present study compared to a similar analysis by in which a longer time period of 4 years (2005–2008) was investigated. also report a relatively large spread (observation standard deviation equal to 5.3 µgm-3) of the measured sulfate concentrations in EANET, which is likely caused by a comparably low number of stations that cover a large region with strong spatial gradients.

The close overall agreement of average concentrations simulated by the model (M‾) with the observations (O‾) for EPA can partly be attributed to underestimated high concentrations in summer (composite of June, July and August, JJA), which are compensated by overestimated lower concentrations from autumn to spring (see Fig. ). Relative deviations are largest in winter and autumn, when observed concentrations are much lower than simulated concentrations. In Europe (EMEP), the highest concentrations are observed during the winter months, which is captured by the model.

The east–west gradient of observed annual mean concentrations in North America is represented well in the model, although the amplitude is underestimated by the model: lower observed values in the west are overestimated, while higher observed concentrations in the east are underestimated by the model (Fig. ). This feature also led to a good agreement between observed and simulated mean concentrations in Table . The north–south gradient over Europe is also captured by the model. Despite the less dense spatial distribution of monitoring stations providing data for particulate sulfate, one can observe a correspondence between simulated and observed annual mean concentrations, especially over Japan. One outlier in Japan with a very low observed annual mean concentration is located at a relatively high altitude (Happo, 1850 m), and thus the comparison to model data on ground level may not be appropriate for this station. The same holds for two EMEP stations, namely Jungfraujoch (Switzerland) at 3578 m, and Chopok (Slovakia) at 2008 m, where annual mean concentrations are largely overestimated, possibly due to the coarse resolution of the model, which cannot correctly reproduce the orography (and hence the altitude) of these stations. We therefore compared SO42- concentrations with observations from aircraft campaigns, as compiled by . The results are presented in Fig. . The simulated sulfate concentrations agree well with the aircraft observations in the free and lower troposphere, with an underestimation in a few cases (see Fig. ; during the ITCT-2K4, ADIENT or IMPEX campaigns) but always within the measured standard deviations, confirming that the vertical profile of sulfate is generally well reproduced in the lower troposphere.

Nitrate is less well represented than sulfate, showing a general overestimation of observed concentrations with an average ratio of simulated to observed concentrations of at least 3.75. This comparably large ratio is dominated by a few stations with small concentrations, where the median ratios lie between 1.70 and 2.84. About half (EMEP) to two-thirds (EPA, EANET) of the simulated monthly mean concentrations exceed a factor of 2 with respect to the observations. For EANET and EMEP, the number of monitoring stations is substantially lower than for EPA, which may partly explain the larger spread in both the model results and observations (OSTD, MSTD). The RMSE is larger than the observed standard deviations for all considered networks, suggesting a less faithful representation of this species. This overestimation may be attributed to the usage of Teflon filters, as nitrate can evaporate from the filters under warm and dry conditions . and showed that particulate ammonium and nitrate partially evaporate from the filters at temperatures between 15 to 20 ∘C and can evaporate completely at temperatures above 20 ∘C. Therefore, this effect predominantly affects measurements during the warmer seasons when nitrate concentrations are close to the annual minimum. An indication for overestimation due to evaporation would then be a closer agreement between model and observation in the colder winter months. However, both in relative and absolute terms, the overestimation of monthly mean nitrate concentrations is more pronounced in winter than in summer for the three considered networks (Fig. ). All regions show the same seasonal cycle with respect to averaged monthly mean concentrations: a maximum in winter and a minimum in summer both for model results and observations. The spatial gradient of the annual mean concentrations is captured well for North America (EPA), although the concentrations towards the east coast are generally overestimated. The low number of monitoring stations for EANET and EMEP does not allow us to draw solid conclusions; however, we can report that the north–south gradient of nitrate over Europe is reproduced by the model, with the exception of an outlier at high altitude in Central Europe (Fig. ). The region with the highest simulated annual mean concentrations and strongest spatial gradients is East Asia; however, it is sparsely covered by monitoring stations reporting full coverage for the considered year 2010.

Considering the ratio of averaged monthly mean ammonium concentrations, the model overestimates the observed concentrations but to a lower degree than for nitrate (Table ). This feature is also apparent in the fraction of simulated monthly mean values within a factor of 2 of the observations: 54.7 % for the network in East Asia and more than 60 % for Europe and North America. The magnitude of overestimation is similar for the three networks, ranging from 40 % to 50 %. RMSE values are close to the standard deviation of observations in all networks, again indicating a good representation by the model. The largest concentrations occur during the winter months (Fig. ). While the degree of overestimation is similar for the different seasons in EMEP and EANET, this does not hold for EPA: summer concentrations are systematically underestimated, whereas autumn to spring concentrations are generally overestimated. The concentrations for winter and spring constitute most of the values that are outside the aforementioned factor of 2 threshold regarding the observations for all considered regions. Both the observed north–south and east–west gradients of annual mean concentrations are represented well for the EPA network (Fig. ), but the simulated gradient is less pronounced.

The north–south gradient of observations over Europe is also captured by the model, although the number of stations is again comparably low. The simulated spatial patterns in East Asia also arguably match the observations well.

This section provides an overview for species frequently found in dust and sea spray aerosol: sodium, calcium, magnesium, potassium and chloride. As the ocean is a large source of these water-soluble species, concentrations over the sea and coastal areas are large, as opposed to lower concentrations over the continents. Sodium, potassium, magnesium and calcium can also be found in desert dust, for instance in the Sahara, as reported by and . However, most of the monitoring locations except for those in coastal areas are sampling quite remote sites far from the source regions, and thus they cannot be expected to display the full range of concentrations. In the following section, we present the results for sodium in more detail, followed by a brief overview of the remaining species.

Sodium is represented well by the model, as averaged monthly mean concentrations from model and observations agree in general, particularly for EPA (Table ). The mean ratio M/O‾, however, shows an average overestimation by a factor of at least 2. The observational standard deviation (OSTD) for all regions is smaller than the RMSE, the latter being reduced compared to the results from a previous study by . For EMEP, more than three out of five simulated monthly mean values lie within a factor of 2 of the observations and more than 53 % and 46 % for EMEP and EANET, respectively, which is a substantial improvement compared to . The latter two networks again offer only a small number of stations. Considering the average of monthly mean concentrations separated by season, one can observe an underestimation of summer concentrations for all networks (Fig. ).

As mentioned before, the regional distribution of stations does not cover those regions with high concentrations. However, when including the information of all networks at once, one can observe that continental stations exhibit low annual mean concentrations, while high concentrations can for instance be observed at stations close to the coast (i.e. Japan, Florida, Denmark, Sweden), indicating that spatial features of the global sodium distribution are qualitatively captured (Fig. ).

Potassium, magnesium and calcium ions share spatial features with sodium in the model, exhibiting elevated values over the ocean, as well as over the Sahara and Gobi desert, and reduced concentrations over other continental areas.

The number of stations in EANET is quite low for all sea spray species, which, on the one hand, allows for a few outliers to dominate mean values and errors. On the other hand, EPA offers a large number of stations, yet the correspondence between model and observations is also quite low for sea spray and dust species. Average monthly mean concentrations for magnesium are represented well in the model, which is caused by a large overestimation of concentrations in winter months balancing an underestimation of summer concentrations (not shown).

Chloride concentrations are widely overestimated with respect to all three networks. The exceptionally high ratio M/O‾ for EMEP is caused by two outliers with M/O>1000 due to very low observed concentrations. Without these factors, M/O‾ is better, with a value of 13.08.

Figure  shows the comparison of model-calculated OA concentrations with measurements from the EMEP observational network over Europe, the IMPROVE network over rural North American locations, and short-term measurement data collected over East Asia as summarized by . As we only used measurements taken in the year 2010, a low number of observations are present for East Asia, and the comparison must be taken cautiously for this region. The comparison statistics are presented in Table . The model captures the monthly average concentrations of OA relatively well over these highly populated regions of the Northern Hemisphere. This is rather encouraging given the expected uncertainties of the emission inventory and the complex chemistry involved in simulating the secondary organic aerosol formation. It is worth emphasizing that the model considers the formation of SOA solely from the homogeneous gas-phase photochemical oxidation of its precursors. Therefore, the omission of other SOA formation pathways (e.g. from aqueous-phase and heterogeneous reactions) can add to the model bias. This is mostly evident during winter (e.g. over Europe), when the relative importance of these processes on SOA formation increases due to the lower photochemical activity and the limited conversion of gas-phase organic precursors to SOA. In addition, recent studies have provided strong evidence that the uptake of water-soluble gas-phase oxidation products (even small carbonyls like formaldehyde and acetic acid) can be the main driver of SOA pollution during haze events over East Asia . Given the coarse resolution of the model and its inability to simulate these SOA formation pathways, the observed total OA concentrations are expected to be systematically underestimated by the model over East and South Asia. Nevertheless, model results are in general in reasonable agreement with observations, with the exception of the strong underestimation over Beijing and Shijiazhuang, which brings the M/O‾ to very low value (i.e. 0.51). As pointed out by and , the emission inventories for semivolatile and intermediate-volatility organic compounds are insufficient and lead to the majority of the model biases in simulating OA in these regions that are influenced significantly by anthropogenic emissions.

Over Europe, the model also underestimates OA, with M/O‾ of 0.83 (see Table ). The model performs worst during winter. have identified the lack of biomass burning emissions as the main source of this discrepancy over Europe. and have more recently shown evidence that biomass burning also contributes significantly to SOA formation during winter following its oxidation by the NO3 radical in the dark. Over North America, the model overestimates OA over rural areas, with a M/O‾ equal to 1.73. However, as shown in Fig. , part of this discrepancy is explained by the low values of OA over the US national parks.

Despite the agreement of the surface OA concentrations between model results and observations, the OA are strongly underestimated in the free troposphere. In Fig. , the model is compared with 12 field campaign measurements . Our results show similar agreement to those obtained by , with the simulated values being below the observed values.