Aerosols are important climate forcers (Ramanathan and Carmichael, 2008; Myhre et al., 2013), but the magnitude of their forcing is highly uncertain and depends on altitude, position relative to clouds, the surface albedo and the optical properties of the aerosol as well as cloud indirect effects. While absorbing aerosols such as black carbon (BC) are likely to increase climate warming (Shindell and Faluvegi, 2009), scattering aerosols such as sulfate have a cooling effect (Myhre et al., 2013). In addition to atmospheric radiative forcing, deposition of absorbing aerosols on snow or ice reduces the albedo and can thus induce faster melting and efficient surface warming (Jacobson, 2004; Flanner et al., 2009). The highly reflective surfaces of snow and ice as well as strong feedback processes make the Arctic a region of particular interest for aerosol research (Quinn et al., 2008).

The Arctic aerosol consists of a varying mixture of sulfate and organic carbon (OC), as well as ammonium, nitrate, BC and mineral dust (Quinn et al., 2007; Brock et al., 2011). Aerosols in the Arctic feature a strong annual cycle with a late winter–spring peak (the so-called Arctic Haze) and a summer minimum. Increased transport during the cold season (Stohl, 2006) and increased removal by wet deposition during the warm season can explain this annual variation (Shaw, 1995; Law and Stohl, 2007) and also shape the aerosol size distribution (Tunved et al., 2013).

Models have for a long time struggled to capture the distribution of aerosols in the Arctic (Shindell et al., 2008; Koch et al., 2009). The concentrations of BC during the Arctic Haze season in particular were underestimated, in some cases by more than an order of magnitude (Shindell et al., 2008), whereas summer concentrations were sometimes overestimated. The simulated aerosol seasonality is strongly dependent on the model treatment of aerosol removal processes. For instance, changes in the calculation of aerosol microphysical properties, size distribution and removal can change simulated concentrations by more than an order of magnitude in remote regions such as the Arctic (Vignati et al., 2010) and the calculated Arctic BC mass concentrations are very sensitive to parameterizations of BC aging (conversion from hydrophobic to hydrophilic properties) and wet scavenging (Liu et al., 2011; Huang et al., 2010).

The seasonal decrease of aerosol concentrations from winter to summer in the Arctic is likely also due to the different efficiency of scavenging by different types of clouds. There is a transition from inefficient ice-phase cloud scavenging in winter to more efficient warm cloud scavenging in summer, and there is also the appearance of warm drizzling cloud in the late spring and summer boundary layer. Including these processes in one model clearly improved its performance both in terms of absolute concentrations as well as seasonality for sulfate and BC (Browse et al., 2012). This result is in agreement with the observation-based findings that scavenging efficiencies are increased in summer both for light-scattering (of which sulfate is an important component) as well as for light-absorbing (of which BC is an important component) aerosols (Garrett et al., 2010, 2011). Another modeling problem may be excessive convective transport and underestimation of the associated wet scavenging in convective clouds, which can lead to model overestimates of BC in the upper troposphere and lower stratosphere (Allen and Landuyt, 2014; Wang et al., 2014). Despite remaining difficulties, simulations of Arctic aerosols with many models have improved considerably in the last few years by updating the model treatment of some or all of the above-mentioned processes (Fisher et al., 2011; Breider et al., 2014; Sharma et al., 2013; Lund and Berntsen, 2012; Allen and Landuyt, 2014).

Remaining problems may also be due to missing emission sources or incorrect spatial or temporal distribution of emissions in the inventories used for the modeling. The main sources of BC are biomass burning and incomplete combustion of fossil fuels and biofuels (Bond et al., 2004). Sulfate aerosols are formed by sea spray or originate from natural sources such as oxidation of dimethyl sulfide (DMS) or volcanoes. It is also produced from oxidation of SO2 emitted when sulfur-containing fossil fuels are burned or by metal smelting. Studies based on observed surface concentrations repeatedly suggest that the main source regions for Arctic BC and sulfate are located in high-latitude Eurasia (e.g., Sharma et al., 2006; Eleftheriadis et al., 2009; Hirdman et al., 2010). Stohl et al. (2013) suggested that gas flaring in high-latitude Russia is an important source of BC that is missing from most inventories. In their simulations, BC emissions from gas flaring accounted for 42 % of the annual mean BC surface concentrations in the Arctic. However, they also noted the large uncertainty of the gas flaring emissions.

The radiative effects of aerosols are not so much determined by the surface concentrations as by the column loadings as well as the altitude distribution of the aerosol (Samset et al., 2014; Samset and Myhre, 2011). Nevertheless, in the past, model results for the Arctic were evaluated mainly against surface measurements due to their availability over long time periods. However, surface concentrations are not representative of concentrations aloft, which are controlled, at least in part, by different source regions and different processes. It is therefore important to evaluate models not only against surface measurements but also using vertical profile information.

The purpose of this study is to explore the capabilities of a range of chemistry transport models (CTMs) and chemistry climate models (CCMs) widely used to simulate the Arctic aerosol concentrations. The models use a common emission inventory, which includes gas flaring emissions and provides monthly resolution of the domestic burning emissions. Differences between their modeled aerosol concentrations are therefore solely due to differences in the simulated transport, aerosol processing (e.g., sulfate formation, BC aging) and removal. We concentrate our investigations on BC and sulfate, for which we collected data from six surface stations and five aircraft campaigns in the Arctic.

Figure 2 shows the simulated BC and sulfate column mass loadings as a function of latitude for the time periods of the Arctic Haze (March) and the much cleaner summer (July) in the Arctic, for the models for which this information was available. For BC in March, most models show a maximum near 20∘ N, with some models extending this maximum to 40∘ N. This approximately covers the latitude range with the highest global emissions where the models agree at least within a factor of 2 in their simulated column loadings. In contrast, larger differences between the models are found in the Arctic, where column mass loadings vary by more than an order of magnitude. Similar results are also found for sulfate in March, for which most models also show a maximum around 20–40∘ N; however, compared to BC, the models show a less pronounced decrease towards higher latitudes and two models even simulate increasing sulfate burdens with latitude. The relatively good agreement between the models in the BC and sulfate source region latitudes is not surprising, given that they all use the same emission data set. In contrast, the differences between the atmospheric column loadings in the Arctic must mainly be due to differences in the aerosol processing and removal and hence aerosol lifetimes, and probably differences in atmospheric transport. Most models with relatively low BC column loadings in the Arctic also have low sulfate loadings there, indicating similarities in the simulated removal of these two types of aerosols. A notable exception, however, is HadGEM3, which has moderately low BC but the highest sulfate loadings in the Arctic.

In July, the BC column loadings show a double peak in the southern tropics and northern subtropics. The southern tropical peak is due to the migration of the inter-tropical convergence zone (ITCZ) into the Northern Hemisphere, which leads to less efficient wet removal and dry conditions favoring biomass burning in the southern tropics. On the other hand, BC concentrations near 10∘ N show a deep minimum, due to the efficient wet removal near the ITCZ. Most models show a third peak in BC loading near 60∘ N, which results from open vegetation fires in the boreal region. North of 60∘ N, the BC loadings decline rapidly towards the North Pole. The sulfate column loading distribution in July lacks the peaks in the southern tropics and the boreal region because biomass burning is not a strong source of sulfate. HadGEM3 stands out against the other models even more than in spring, as its polar sulfate loadings are more than a factor of 5 higher than those of all other models, which show a smooth decrease with latitude north of 40∘ N.

In the simulated surface BC and sulfate mass mixing ratios the same basic patterns are found as in the column loadings, but with enhanced gradients between source areas and remote regions (Fig. 3). When looking at individual models, there are, however, notable differences for sulfate. ECHAM6-HAM2 has the highest sulfate surface mass mixing ratios of all models, especially in the Northern Hemisphere subtropics and mid-latitudes. Combined with the rather “normal” column sulfate loadings of this model, this indicates that ECHAM6-HAM2 does not transport sulfate away from the surface as quickly as the other models. On the other hand, HadGEM3, which has by far the largest sulfate column loadings, has the smallest surface concentrations. This deficiency was due to the implementation of the Global Model of Aerosol Processes (GLOMAP; Mann et al., 2010), which in this HadGEM3 version resulted in too little removal of the sulfate precursor SO2 during the venting from the boundary layer to the free troposphere. The longer sulfate lifetime there explains the high column loadings.

In summary, we find that the Arctic is a region with particularly large relative differences between the models, both for the surface mass mixing ratios (with differences of more than an order of magnitude) as well as for the column loadings, and both for BC and sulfate. This result must be related to differences in aerosol removal and lifetimes in the different models. We also found that, especially for sulfate, there can be an anticorrelation between simulated surface concentrations and column loadings. Hence there is a strong motivation to evaluate the models' performance in the Arctic, based on measurements taken both at the surface and aloft.

We start our discussion of the annual cycles of aerosol concentrations with the example of BC at the Zeppelin station in Spitsbergen (Fig. 4). Monthly medians as well as the 25th and 75th percentiles are calculated for every month based on hourly data for the two years 2008 and 2009. Maximum median eBC concentrations of 46 and 53 ng m-3 occur in March and April, while summer median values are only 2 to 3 ng m-3. Some of the models reproduce this seasonality with high winter/spring values and much lower summer values quite well, although in most of these models BC reaches its highest values already in January. Only the CanAM4.2 model seems to capture the observed spring maximum. All models except WRF-Chem capture the fact that summer has the lowest values of the year. OsloCTM2, TM4-ECPL and NorESM have smaller annual variation than observed. HadGEM3, which we have seen to produce lower BC surface concentrations than the other models in Fig. 3, strongly underestimates the measured eBC concentrations throughout the year. The variability of the modeled values within a month (described by the height of the bars) shows clear differences between the models. For instance, CESM1-CAM5.2 simulates far fewer variable BC concentrations than CanAM4.2 and DEHM or the measurements.

The eBC mass concentrations at the three other sites in the western Arctic (Alert, Barrow, Pallas) are quite comparable to those at Zeppelin station, with monthly median values of about 20–80 ng m-3 in late winter/early spring and of less than 10 ng m-3 in summer/early fall (see Fig. 5). One exception is EC measured at Station Nord, which in summer is higher than eBC measured at the other sites. At Alert, where both eBC and EC data are available, EC values in summer are also somewhat higher than eBC values (although lower than the Station Nord EC values), probably due to systematic differences in measurement techniques.

At the Tiksi station, which is closer to the main source regions of Arctic BC in high-latitude Eurasia (Hirdman et al., 2010), higher monthly median eBC values were measured (more than 100 ng m-3 in winter/spring, about 20–40 ng m-3 in summer) and the annual mean (81 ng m-3) is 2.5 times higher than the average for the other stations (31 ng m-3). The seasonality of measured eBC is strongest at Alert where the summer concentrations are very low, but the winter/spring concentrations are similar to the other sites in the western Arctic. This result points to a deepening of the seasonal minimum with latitude. While the aerosol concentrations in the Arctic during late winter/early spring are comparable to remote regions further south, the concentrations in summer/early fall are lower because of the effective cleansing of the atmosphere (Garrett et al., 2010, 2011; Browse et al., 2012; Tunved et al., 2013) and less efficient transport from source regions (Stohl, 2006). The highest eBC concentrations were observed in January (Alert), February (Barrow), March (Pallas, Tiksi) or April (Zeppelin), with no clear dependence of the time of the maximum on latitude; however, the maximum occurred earlier at the two North American sites than at the other sites.

The models capture the Arctic BC concentrations with variable success (Fig. 5). Most models capture the much higher concentrations in winter/spring than summer/fall, and some models can approximately reproduce the concentrations reached during the Arctic Haze season (see also Breider et al., 2014). However, as already seen for the Zeppelin station (Fig. 4) and the annual mean surface mass mixing ratios (Fig. 3), there is a large variability between individual models, with seasonal median values varying by about an order of magnitude both in spring and summer even when excluding the most extreme models (see also Table 2). Seasonal mean concentrations during January to March are underestimated by up to a factor of 27 for individual models and by more than a factor of 2 for the mean over all models, and only one model slightly overestimates the measured concentrations (Table 2). Nevertheless, this indicates clear progress since earlier studies (e.g., Shindell et al., 2008; Koch et al., 2009; AMAP, 2011), where it was reported that most models had a completely wrong seasonality and systematically underpredicted the Arctic Haze concentrations. For instance, in Shindell et al. (2008), none of their models came close to the measured concentrations at Barrow and Alert during winter and spring, with a model-mean underestimate of about 1 order of magnitude (their Fig. 7). It is also important to keep in mind that the eBC measurements are uncertain and could be biased high. However, EC and eBC values at Alert are very similar and we find a similar model underestimate of measured EC at Station Nord as well.

Our finding that Arctic BC concentrations in the spring tend to be underestimated by our models implies that these models would also underestimate radiative forcing by BC in the Arctic. This is particularly important because spring is the season when both aerosol concentrations are large and solar radiation is abundant. Furthermore, it is the season when feedback processes, e.g., via ice and snow melting, are most important (Quinn et al., 2008). The concentrations of BC in summer are much lower than in spring, so even with more abundant solar radiation, modeling problems in summer would have a relatively small effect on radiative forcing.

In contrast, five models overpredict the low concentrations in summer, the most extreme model by an order of magnitude (Table 2). Some models (e.g., HadGEM3) underpredict strongly throughout the year. For the sites in the western Arctic, the model deficiencies become worse with increasing latitude. For instance, at the northernmost site, Alert (82.5∘ N), all models underpredict for the full duration of the Arctic Haze season from January until April.

For Tiksi, the data comparison is less direct as measurement data from July 2009 to June 2010 were used. Nevertheless, it is clear that except for CanAM4.2 (which produces the highest modeled values at most sites) the models strongly underpredict for this site, especially in winter/spring. The most likely explanation for this is that the BC emissions in high-latitude Russia are underestimated in the ECLIPSE inventory. It is difficult to know where exactly the missing sources are located. However, we find that in the ECLIPSE inventory the BC emissions in Norilsk (88.2∘ E, 69.3∘ N; population 170 000) are zero. We do not suggest that Norilsk emissions are responsible for the strong underestimation of BC concentrations at Tiksi, but these discrepancies (and others for sulfur emissions discussed later) suggest that the high-latitude Russian pollutant emissions are underestimated and/or wrongly placed in the ECLIPSE inventory. Similar problems likely occur with most other global emission inventories. For instance, AMAP (2015) compared the ECLIPSE emission data set with 10 other inventories and found that the differences between the different inventories grow with latitude and are largest north of 70∘ N (i.e., high-latitude Eurasian emissions).

The seasonal cycle of sulfate at the monitoring stations is similar to that of eBC, with a clear maximum during the Arctic Haze season and a minimum in summer/early fall (Fig. 6). However, the seasonal cycle at the northernmost stations is less strong than for eBC, with about a factor of 5 difference between spring and summer, compared to a factor of 15 for eBC (Table 2). This is probably due to the influence of biogenic sources of sulfate in summer (Quinn et al., 2002) and/or a weaker seasonality in the emissions (e.g., smelter emissions of SO2 are probably relatively constant throughout the year).

The models have similar difficulties capturing the sulfate seasonality as they have for BC. Again, there is up to more than an order of magnitude difference between simulated seasonal median concentrations from different models, both in summer and in winter (Table 2). The model differences in summer are in fact even larger than for BC, probably related to different treatment of natural sources, especially dimethyl sulfide emissions from the Arctic Ocean. There is a tendency for models that strongly underestimate BC concentrations to also underestimate sulfate (e.g., the HadGEM3 model), but the correlation between the two simulated species from the different models is quite low, especially in summer. For instance, ECHAM6-HAM2 underestimates BC by factors of 26 and 1.6 in winter and summer, but underestimates sulfate only by about 13 % in winter and even overestimates sulfate by a factor of 3.8 in summer (see Table 2). As seen in Figs. 2 and 3, ECHAM6-HAM2 simulates relatively high surface concentrations of sulfate but low total column loadings, both at source and Arctic latitudes.

The models generally underpredict sulfate most strongly at the northernmost station (Alert), which is consistent with the BC results (compare Figs. 5 and 6). The CanAM4.2 model, which had some of the highest BC concentrations, also gives the highest sulfate values (Table 2). It is the only model that matches the high measured sulfate values at Alert and Station Nord in spring. The reason why CanAM4.2 captures the spring peak better might be that this model has a less efficient removal through wet deposition under stratiform conditions compared to the other models (Mahmood et al., 2015).

At Pallas, the lowest-latitude station in this comparison, most models severely underestimate sulfate throughout the year (Fig. 6), although they tend to overestimate BC in spring there. One likely reason for the sulfate underestimation is the proximity of the Pallas station to the Kola peninsula, where metal smelters are a strong source of sulfur. According to AMAP (2006), SO2 emissions in Nikel, Zapolyarnyy and Monchegorsk together were about 170 kt year-1 in the year 2002. In the ECLIPSE version 4a inventory used for this study the SO2 emissions in these areas are only about 33 kt year-1 in total for the year 2005. Similar deficiencies were in fact reported also for other emission inventories for this region (Prank et al., 2010). Strong underestimation of the SO2 emissions from metal smelting in the Kola peninsula is therefore a likely explanation for why almost all models underestimate sulfate at Pallas so strongly. Similar discrepancies were in fact found for SO2 emissions in Norilsk, prompting a regridding of the ECLIPSE emissions (now available version 5a) using better location information for the metal smelting industry.

Figure 7 summarizes all rBC data from the ARCTAS and ARCPAC campaigns in spring 2008. Median concentrations are shown as a function of latitude (binned into 10∘ intervals) both for lower (< 3 km) and higher (> 3 km) altitudes, and as a function of altitude both for the high Arctic (> 70∘ N) and lower latitudes. As the campaigns focused on the Arctic, data south of 60∘ N are scarce and limited to North America. The models were sampled in their grid box containing a measurement location and at the time of a measurement and were subsequently binned in the same way as the measurement data to allow a direct comparison. For the free-running climate models, the same procedure was used, albeit with the caveat that the simulated meteorological situation at the measurement time does not correspond to the real conditions.

For the low-altitude (< 3 km) bin, the highest median rBC values were measured (see the second from top row of panels in Fig. 7) at 35 and 55∘ N, with a substantial concentration drop towards higher latitudes. The mid-latitude maximum reflects the location of the BC sources in North America, where ARCTAS and ARCPAC were conducted. Above 3 km (top row of panels in Fig. 7), the highest median rBC concentrations were measured further north, at 60∘ N, and the concentrations drop less strongly towards the North Pole than at lower altitudes. This is due to quasi-isentropic lifting occurring together with northward transport (Stohl, 2006). All models, except CanAM4.2, systematically underestimate the measured values for both altitude bins and for all latitudes, and they also underestimate the measured rBC variability. However, most of the models simulate a decrease of the concentrations with latitude that is consistent with the measured latitude dependence.

When plotted as a function of altitude (two bottom panel rows in Fig. 7), the measured values peak in the 4–5 km altitude bin, both for sub-Arctic and Arctic latitudes. The models, except for CanAM4.2, underestimate the measured median values throughout the entire depth of the profile. Some of the models, mainly those driven by observed meteorology, capture the rBC maximum in the mid-troposphere in the Arctic. However, the lower-latitude 4–5 km maximum is hardly reproduced by any of the models. One likely reason for the modeling problems is the strong biomass burning activity during spring 2008, which influenced a substantial fraction of the measurement data (Warneke et al., 2010; Brock et al., 2011). Even though this should be reflected in the GFED emission data for 2008, it seems possible that the GFED emissions are underestimated. Furthermore, as some of the flights targeted biomass burning plumes specifically, the influence of the biomass burning may be enhanced in the measurement data compared to the models, especially if the models did not capture the plume transport well enough and thus potentially simulated the biomass burning plumes at other locations than observed. This sampling bias is particularly strong for the CCMs that are not driven by observed meteorological fields.

Comparisons like those shown in Fig. 7 were also performed for the other aircraft campaigns. For the sake of brevity, we further aggregate the data and only show results for latitudes north of 70∘ N and for median values below and above 3 km altitude (Fig. 8). For spring 2008, the aggregate plots for BC (Fig. 8e–f) show even more clearly than Fig. 7 that all models except CanAM4.2 underestimate the measured rBC concentrations both at low and high altitudes. The spring 2009 PAMARCMiP campaign, however, shows a different picture (Fig. 8c–d). This campaign was influenced very little by biomass burning. The measured median rBC mass concentrations at low (high) altitudes were about a factor 2 (3) lower than for the spring 2008 campaigns. Most models also simulated lower median BC concentrations than a year earlier, but the modeled reductions were less pronounced than the measured ones and, thus, about half of the models underestimated and the other half overestimated the measured median values. The vertical gradient of measured BC was also different in 2008 and in 2009. While in spring 2008, the concentrations above 3 km were higher than those below, the opposite was true in spring 2009, likely because of the weaker biomass burning influence in 2009. This feature can be seen very clearly in the vertical profiles shown in Fig. 9 and it is not well captured by the models, most of which showed a relatively flat vertical BC distribution.

The concentrations measured by the ARCTAS summer campaign in 2008 are much lower than those measured in spring 2008 and 2009, both at low and high altitudes (Fig. 8g–h), which is in agreement with the seasonality seen at the surface stations. Some of the models underestimate and others overestimate the measured concentrations, with the majority of the models overestimating, especially below 3 km. The mean values, averaged over all models, are about 2 (3) times as high as the measurements for altitudes above (below) 3 km. Some of the models reproduce the measured rBC maximum at 6 km (Fig. 9).

The HIPPO campaign in fall 2009 (Fig. 8i–j) was conducted about 1 month after the seasonal minimum at most surface sites and measured very low rBC mass concentrations, which is consistent with the surface observations. Most of the models overestimate the measured concentrations throughout the entire vertical profile (Fig. 9).

The HIPPO campaign in January 2009 (Fig. 8a–b) measured strong altitude differences: moderately high rBC mass concentrations up to 3 km, but the lowest concentrations of all campaigns above. This feature is well captured by some of the models (Fig. 9). The lack of high concentrations aloft is likely related to the minimal influence of biomass burning at this time of the year.

Overall, the aircraft measurements confirm the BC seasonality measured at the surface stations. They also confirm that most models underestimate the concentrations in spring (at least for the year 2008) but many models overestimate the concentrations in summer and fall. It thus seems that models produce a too weak BC seasonality throughout the depth of the troposphere. However, for the year as a whole there is a tendency towards model overestimates, in contrast to the surface sites. Even stronger model overestimates downwind of Asia over the Pacific, especially in the upper troposphere, were recently reported by Samset et al. (2014), who suggested that the BC lifetime in the models is too long. However, a uniform reduction of BC lifetime in our models would lead to strong underestimates of the BC concentrations at the Arctic measurement stations. Even our Arctic aircraft comparisons only support at most a very moderate BC lifetime reduction. Of course, regional and/or vertical differences in the model lifetime biases or excessive convective uplift could explain the contrasting findings of our study and Samset et al. (2014).

For sulfate, measured median concentrations in the Arctic during spring 2008 were lower above 3 km than below 3 km (Fig. 10a–b). All models, except CanAM4.2, strongly underestimate the measured sulfate concentrations, some models by more than an order of magnitude. This is consistent with the findings from the surface station comparisons (Fig. 6, Table 2). The models also do not give a consistent picture of the vertical distribution of sulfate, with some models correctly simulating lower concentrations above 3 km than below but others giving the opposite result. The model underestimates for sulfate are likely not related to a sampling bias towards frequent encounters of biomass burning plumes, as biomass burning plumes are relatively poor in sulfate (e.g., Brock et al., 2011). Instead, the underestimation suggests other missing sulfur sources or a too quick removal of sulfate from the atmosphere. Indeed, the latter would be consistent with the suggestion of Kristiansen et al. (2012) that sulfate lifetimes in models are too short in spring.

During summer 2008 (Fig. 10c–d), the measured median sulfate concentrations were about a factor of 4–6 lower than in spring 2008, consistent with the seasonality measured at surface sites. Median concentrations above and below 3 km are very similar. The models have very large differences in their simulated sulfate concentrations, with some models overestimating and others underestimating the measured concentrations in summer. This is again consistent with the findings from the surface site comparison (Fig. 6, Table 2).

Contrary to the year-round station measurement programs, the aircraft campaigns sample the atmosphere only during limited time periods and their representativeness with regard to climatological means may be questioned. Furthermore, from the aircraft measurements we have seen that spring 2008 and 2009 had very different measured rBC concentrations, and modeling problems were larger for spring 2008, when there was intensive biomass burning influence in the Arctic. A valid question is therefore whether the surface measurements show the same differences between 2008 and 2009.

To investigate how consistent a picture the aircraft campaigns give vis-a-vis the station measurements, we compare all aircraft data from the lowest 3 km and lowest 1 km to the values obtained from the surface stations for the same months (Fig. 11). Selecting data only for even lower altitudes is problematic as the data coverage becomes very poor. In Fig. 11, we also show the station measurements obtained for the years 2008 and 2009 separately. For eBC, the measurements obtained for the same month at the different stations and during different years are (with a few exceptions such as Barrow in January 2008) quite comparable with each other. In particular, April 2008 did not show higher eBC values than April 2009. This is consistent with the finding that the biomass burning layers in 2008 did not extend to the surface (Brock et al., 2011). At Alert, the EC values are similar to the eBC values, whereas the Station Nord EC values in summer and fall are higher than eBC values at other stations. The aircraft rBC measurements for all campaigns show consistently lower values than the eBC or EC measurements at the ground, except for the HIPPO campaign in January 2009 where, however, the data coverage particularly below 1 km is poor. It is possible that the BC concentrations show a strong gradient in the lowest 1 km and that surface concentrations are indeed systematically higher than concentrations just aloft. However, an alternative explanation could be that the rBC measurements are biased low against the eBC or EC measurements, given the different measurement techniques used. A direct comparison of all three measurement techniques at the Alert station also suggests a low bias of rBC against eBC and EC concentrations (S. Sharma, personal communication, 2014). For sulfate (Fig. 12) the measurements show a much larger variability than for BC, both between stations and between the two different years. For instance, the 25th percentile of the sulfate concentrations at Alert in January 2009 is higher than the 75th percentile of the other stations and also of Alert in January 2008. On the other hand, the sulfate concentrations measured during the two available flight campaigns in spring and summer 2008 are not systematically different from those measured at the stations, although the median concentration in summer 2008 is somewhat lower than at the stations. This is consistent with the eBC or rBC differences.

In this section, we perform a correlation analysis of BC and sulfate. Such an analysis allows some insights into the mixing state of the Arctic aerosol. BC and sulfate largely originate from different sources (although some sulfate is co-emitted with BC by combustion processes). A poor correlation between BC and sulfate means that BC and sulfate either arrive at the measurement stations in distinct air masses or that at least the different aerosol types (even if the air masses mix) remain externally mixed and thus are affected to a different and varying extent by removal processes. On the other hand, a strong correlation implies that BC and sulfate arrive in air masses where contributions from their different emission sources are mixed and that, furthermore, the aerosol must also be internally mixed, as otherwise different removal efficiencies for BC and sulfate would lead to decorrelation between the two species. Such a correlation analysis has in fact recently also been performed with measurement data from Station Nord (Massling et al., 2015). In our case, we can furthermore compare measured and modeled correlations, allowing some insights into how models treat the mixing of different aerosol types compared to reality.

Figure 13 shows correlation plots between monthly mean sulfate and eBC for the measurements and the models sampled at the different stations. In the observations, sulfate and eBC correlations for Alert, Pallas and Zeppelin are statistically significant at the 99.9 % level (Table 3). The slopes of the regression lines shown in Fig. 13 are reported in Table 3. For the observations, they are very similar: 10.1, 8.4 and 8.9 ng[SO4] m-3 (ng[eBC] m-3)-1 for Alert, Pallas and Zeppelin, respectively. For Barrow, where the correlation is not significant because of two eBC-rich outlier data points, the slope is smaller (6.4 ng[SO4] m-3 (ng[eBC] m-3)-1). The strong correlation between sulfate and eBC and the similarity of the slopes suggests that the sources contributing to the measurements at the different stations are similar and that the removal of sulfate and eBC is highly correlated, which would be expected for internally mixed aged aerosol as is typical for the Arctic.

Most of the models, on the other hand, show much weaker correlation between sulfate and BC, and some of the models have no significant correlation at all. Exceptions are DEHM, CESM1-CAM5.2 and WRF-Chem, which show mainly significant correlations and slopes that are comparable at the different stations and that are also quite similar to the observed slopes. This suggests that, with the given emissions, it is possible to reproduce the observed correlations. The lack of correlation between sulfate and BC in the other models – in disagreement with the observations – therefore suggests that they treat the two species differently, probably having a too large fraction of the aerosol as externally mixed. Correlations could also be degraded by a too strong influence of biogenic (dimethyl sulfide) emissions from the oceans or factors influencing SO2 to sulfate conversion such as the level of oxidants in the models. This could lead to varying fractions of sulfur present as SO2, and maybe these fractions are more variable in the models than in reality.

Based on the ECLIPSE inventory that is available for BC and for SO2, we estimated ratios between those two substances under the assumption that all SO2 is converted to sulfate. The SO2 to BC emission ratio of anthropogenic emissions in the ECLIPSE inventory is 25 globally and 40 north of 50∘ N. For the GFED biomass burning emissions the emission ratio is only 1.7 globally and 2.5 north of 50∘ N, and for the sum of anthropogenic and biomass burning emissions, we obtain ratios of 19 globally and 25 north of 50∘ N. The mean observed slopes of the observations (9.1 ng[SO4] m-3 (ng[eBC] m-3)-1) and the slopes modeled by DEHM (5.4 ng[SO4] m-3 (ng[BC] m-3)-1), CESM1-CAM5.2 (9.9 ng[SO4] m-3 (ng[BC] m-3)-1) and WRF-Chem (8.5 ng[SO4] m-3 (ng[BC] m-3)-1) are much lower than the emission ratio of anthropogenic emissions in the ECLIPSE inventory and they are also lower than the emission ratio for mixed anthropogenic and biomass burning emissions. This suggests that biomass burning emissions are relatively more important in the Arctic than elsewhere, that there are missing BC sources, that sulfur emissions are overestimated (although this is not so likely, given the too low SO2 emissions in high-latitude Russia in the ECLIPSE version 4a inventory used here), and/or that there exists a mechanism that enriches aerosols in BC relative to sulfate in the Arctic atmosphere. The latter could be related to the hydrophobic nature of freshly emitted BC.

Based on our comprehensive study of measured and modeled BC and sulfate in the Arctic, we can draw the following conclusions.

The simulation of BC concentrations in the Arctic has improved compared to earlier studies (e.g., Shindell et al., 2008; Koch et al., 2009; AMAP, 2011). For instance, our model-mean underestimate of Arctic eBC at Barrow and Alert is about a factor of 2, compared to 1 order of magnitude reported in Shindell et al. (2008). Nevertheless, the aerosol seasonality at the surface is still too weak in most models. Concentrations of eBC and sulfate averaged over three surface sites in the western Arctic are underestimated in winter/spring in all but one model (model means for January–March underestimated by 59 and 37 % for BC and sulfate), whereas concentrations in summer are overestimated in the model mean (by 88 and 44 % for July–September), but with overestimates as well as underestimates present in individual models.