Modelled DMS(g) from the model simulations (CLIM11, SAT and CLIM11 + ave-Obs; the later simulation is defined in Sect. 4.2.3), extracted along the Polar 6 flight path coinciding with aircraft samples, are compared with the measurements in Fig. 5. This figure includes the simulated average DMS(g) vertical profiles along the flight path. The GEM-MACH model captures DMS(g) mixing ratios close to the measurements and shows a general decay of the DMS(g) mixing ratio with height, indicating the influence of local sources of DMS. Using the same observation data, Ghahremaninezhad et al. (2017) highlighted the role of local sources in the Canadian Arctic with respect to DMS emission during summer. The observation results indicated a decrease in the DMS(g) mixing ratios with altitude up to about 3 km, and the largest mixing ratios were found near the surface above the ice edge and open water, coincident with increased particle concentrations (Burkart et al., 2017; Croft et al., 2016a, b; Ghahremaninezhad et al., 2017). The dominant influence of local sources on DMS(g) observed in the Arctic marine boundary layer during summer is further supported by the source sensitivity tests discussed in Sect. 4.2.

The scatter plot in Fig. 6 shows the statistical comparison of the model simulations (SAT and CLIM11) with the observation results. Overall, observation and model results are of a similar magnitude. The simulation using SAT is in slightly better agreement with the measurement based on the root-mean-squared error and mean bias values of 27.6 and -4.7 compared with 29.5 and -6.6 for the simulation using CLIM11; SAT also shows a better correlation coefficient (as shown in Table 2).

Figure 7a compares the time series of modelled DMS(g) mixing ratio values from the two simulations (using SAT and CLIM11) following the Amundsen cruise track with the GC-SCD and CIMS measurements. The discrepancy between the two measurement datasets can be attributed to the different sampling locations/heights onboard the Amundsen (e.g. CIMS's inlet at 16 m a.s.l. at the bow vs. GC-SCD's inlet on the bridge at 30 m a.s.l.) and the different sampling/analysis methods. For example, the lower DMS(g) measured by GC-SCD compared with the CIMS measurement could on occasions be attributed to the vertical gradient in DMS(g). This is particularly the case when DMS(g) is mainly driven by local fluxes, e.g. the 19–21 July episode. More details are available in Mungall et al. (2016) and Ghahremaninezhad et al. (2017) respectively.

Relatively high DMS(g) mixing ratios were observed during the biologically productive period of July; two main high DMS(g) concentration episodes were observed during the 18–21 and 26–27 July periods while Amundsen was traversing Lancaster Sound. The modelled DMS(g) from both simulations generally tracked the observations well but both missed the two abovementioned high DMS(g) episodes, although the simulation with SAT captured the higher DMS episodes in early August well. Overall the model simulations under-predicted the measurements. A statistical evaluation against the CIMS measurements shows a model mean bias of -126.4 pptv (NMB of -57.7 %) for the simulation using CLIM11 and -95.4 pptv (NMB of -43.6 %) for the simulation using SAT. These large negative biases are mainly driven by the model under-prediction for the period when the icebreaker was travelling back and forth along Lancaster Sound. The better model results using SAT during the early part of August correspond to the icebreaker's path through the Nares Strait, where the SAT DMS(aq) concentration, which was up to 9 nmol L-1 (Fig. S1), is considerably higher than the CLIM11 DMS(aq) concentration (<3 nmol L-1) and in better agreement with the in situ DMS(aq) measurement in the area. Also shown in Fig. 7b–d are the time series of the observed (from the Amundsen's Automatic Voluntary Observing Ships System, AVOS, available onboard the ship at ∼23 m a.s.l.) and modelled surface wind speed, air temperature and sea temperature. These are the physical parameters affecting the sea–air flux. Overall the model is in good agreement with observations, given the model temporal and spatial resolution, which suggests that the main cause for the model to under-predict the high DMS(g) events likely lies in the model's representation of DMS sources.

In the next section, we look into several potential DMS source uncertainties which may contribute to the model under-prediction.

This sensitivity test (no-Ice) is conducted to examine the effect of neglecting sea-ice cover in the DMS(g) flux calculation. This is essentially an extreme case used to consider the potential contribution from melt ponds; this is done by assuming that the entire ice-covered portion of the Arctic ocean is covered by melt ponds, that the DMS(aq) concentrations in these melt ponds are the same as the open-water sea surface water DMS(aq) concentration and that the flux exchange is the same as over open water.

As shown in Fig. 8a, by neglecting the sea-ice cover, there is an enhancement in the model-simulated DMS(g) that mainly occurs over two periods: the 15 July – when the icebreaker was sailing along the coast of Baffin Island; and the 22–25 July – when the icebreaker was traversing the eastern end of Lancaster Sound (Fig. 2). In both instances the Amundsen was near the melt pond areas or ice edges. The sea-ice fraction in GEM-MACH (based on analysis) shows an area of sea-ice cover over Baffin Bay on 15 July and an ice edge located at the eastern end of Lancaster Sound on 22 July 2014 (see Fig. S4). However, there is no enhancement in the modelled DMS along the Amundsen path during the 18–21 and 26–27 July periods, when high DMS episodes were observed with the additional DMS sources from sea-ice-covered surfaces, indicating that the melt pond sources did not contribute to the two high DMS(g) events observed onboard the Amundsen. Figure 9a shows the impact of neglecting sea-ice cover on the modelled July mean DMS(g). The highest increase in the DMS(g) mixing ratio is around the Chukchi Sea (up to 300 pptv).

This sensitivity test is inspired by the sensitivity study conducted by Mungall et al. (2016) and is based on the observation results of Ferland et al. (2011). The latter study reported that the productivity of Hudson Strait water is equal to that of the northern Baffin Bay, whereas the Hudson Bay and Foxe Basin water is about a quarter as productive as northern Baffin Bay (Ferland et al., 2011). We assumed a linear relationship between productivity and DMS(aq) formation, and set the DMS(aq) concentration values in Hudson Strait to be equal to the ship-based DMS(aq) measurement values in northern Baffin Bay (e.g. 20 nmol L-1), and the DMS(aq) in Hudson Bay and Foxe Basin to a quarter of that value. The DMS(aq) value (20 nmol L-1) is quite consistent with the SAT DMS(aq) in this region (see Fig. S1a for 1–3 July and Fig. 1b for 4–11 July; the lower SAT DMS(aq) in this region on 20 and 29 July may in part be due to the ice conditions or other limitations of satellite-based DMS(aq) measurements, e.g. potential issues near marginal ice or near the ice edge, sparse temporal coverage, and so on.).

Mungall et al. (2016) conducted this sensitivity test using the GEOS-Chem model at 2∘×2.5∘ resolution, and concluded that the Hudson Bay system contributed significantly to the high DMS(g) event observed onboard the Amundsen during the 18–19 July period. However, the results from our sensitivity test do not seem to support their findings. As seen from Fig. 8b, the enhancement in DMS(aq) in the Hudson Bay system (HB + HS) caused a very small change in modelled DMS(g) along the Amundsen path. One possible reason for the difference in model results between GEM-MACH and GEOS-Chem could be in model resolutions (e.g. GEOS-Chem model, 2∘×2.5∘ and GEM-MACH, 15 km).

The difference in the modelled July-averaged DMS(g) mixing ratios with and without the HS + HB DMS(aq) enhancement, shown in Fig. 9b, indicates that the impact of the Hudson Strait and Hudson Bay system is rather locally confined during the study period. The HS + HB DMS(aq) enhancement led to an increase in modelled DMS(g) of up to 300 pptv in the Hudson Strait area but the area of increase does not extend much beyond the area of DMS(aq) enhancement, indicating either a short DMS(g) lifetime and/or inefficient transport in this case.

For this experiment we further updated the DMS(aq) in Baffin Bay and Lancaster Sound using the in situ measurements of the surface seawater DMS(aq) concentration onboard the Amundsen cruise (Mungall et al., 2016). The sampling area is divided into three subregions, Lancaster Sound, northern Baffin Bay–Southern Nares Strait and central Baffin Bay, each with averaged DMS(aq) measurement values of 7.9, 11.0 and 4.5 nmol L-1 respectively for July and a value of 4.0 for August. These values were used to replace the CLIM11 DMS(aq) values in the respective regions. It is worth noting that these value are comparable to the SAT DMS(aq) concentrations shown in Fig. 1. By updating DMS(aq) for the sampling region, GEM-MACH predicts the relatively higher DMS(g) mixing ratios and captures the elevated DMS(g) event days (18–22 July). These results show the importance of the local source (e.g. the Lancaster Sound region) of DMS(g) emissions during July (Fig. 8c). Figure 10 compares the CIMS DMS(g) measured data onboard the Amundsen with GEM-MACH simulations, using CLIM11, SAT and CLIM11 + ave-Obs. The statistical evaluations in Table 2 and Fig. 10 indicate a significant improvement in the model–observation comparison with this update.

Figure 9c shows the difference between July mean DMS(g) mixing ratios using CLIM11 + ave-Obs and CLIM11. The DMS(g) enhancement is largely limited to the locations with the updated CLIM11 DMS(aq) concentration values. The sensitivity tests result emphasizes the role of locally emitted DMS(g) in the atmosphere particularly in the marine boundary layer.

DMS(g) oxidation depends on the oxidants present and the temperature at which the reactions take place. Figure 11 shows the modelled DMS(g) average chemical lifetime in the atmosphere (Fig. 11a) and the contributions (%) from each of the three main reaction pathways to DMS oxidation – abstraction with OH (Fig. 11b), addition with OH (Fig. 11c) and abstraction with NO3 (Fig. 11d) – for July 2014.

The DMS(g) atmospheric chemical lifetime (or e-folding time) shown here is based on the decay of DMS(g) due to OH and NO3 radicals in the atmosphere, and it is mostly less than 1 d in the marine environment below the Arctic circle but much longer in the Arctic (Fig. 11a). In the lower Arctic (<70∘ N), the chemical lifetime of DMS ranges from less than 1 to 5 d, and in the high Arctic (>80∘ N), the DMS chemical lifetime is between 5 and 20 d; it is longest over the central Arctic ocean where the concentrations of atmospheric oxidants are lowest. Over the Canadian Arctic Archipelago this DMS lifetime has a large range, from as short as less than half a day (e.g. over Hudson Bay and Hudson Strait, Davis Strait, and southern Baffin Bay) to >10 d (e.g. Nares Strait and western Queen Elisabeth Islands). The relatively short lifetime over Hudson Bay and Hudson Strait is consistent with the results discussed earlier from sensitivity studies, where we found that the effect of Hudson Bay DMS is mostly confined locally. The DMS(g) lifetime for the central Arctic was predicted to be 2.5 d by Leck and Persson (1996). Sharma et al. (1999), using a 1-D photochemical box model, estimated the lifetime of DMS(g) due to OH and NO3 oxidation to be around 6–8 d in the central Arctic during August. The relatively shorter lifetime of DMS(g) in Sharma et al. (1999) compared with our study may be partly due to the slightly higher OH concentration simulated by their 1-D model (e.g. 0.01–0.02 pptv in their August case vs. 0.006–0.01 pptv in this study). In addition to OH and NO3 radicals, halogenated radicals may have an important role as an additional sink for DMS(g) in the Arctic atmosphere (von Glasow et al., 2004; Hoffmann et al., 2016). The multiphase chemistry of DMS(g) and the impacts of the halogens in the DMS(g) chemistry/lifetime/products in the Arctic atmosphere need to be considered in future GEM-MACH model studies.

The NO3 concentrations, and as a result the DMS(g) oxidation by NO3, decrease sharply above 70∘ N. Conversely, oxidation by the OH radical is more important north of 70∘ N during the bright month of July. Overall, the abstraction pathway with the NO3 radical (up to 95 %) below the 70∘ N and the addition pathway with the OH radical (up to 90 %) above 70∘ N are the dominant oxidation pathways for DMS(g) in the subarctic and Arctic respectively.

SO2 is one of the important products of DMS(g) oxidation in the atmosphere. SO2 concentrations were altered in the GEM-MACH model by including DMS(g) as a new biogenic source. The SO2 increment (July-averaged) due to DMS (or DMS-derived SO2, both absolute and relative percentage to total modelled SO2) are shown in Fig. 12. The absolute SO2 concentration difference in Fig. 12a is up to ∼600 pptv, and the relative contribution of DMS-derived SO2 to total SO2 is up to almost 100 % for some regions (Fig. 12b), which could be significant for the remote and clean Arctic environment during July. SO2 concentrations are increased in Hudson Bay and south Baffin Bay (around 100 pptv) by adding DMS(g) in the model. The absolute values of DMS-derived SO2 are small in these areas in comparison with other areas; however, due to the low background SO2 concentrations, DMS makes a significant contribution to SO2 in these areas, as shown in Fig. 12b. The relative SO2 increment plots (Fig. 12b) highlight the significant change of SO2 in the atmosphere by including DMS in the model. The SO2 concentrations are relatively low in the Arctic clean atmosphere during summer, and the relative increase in SO2 due to DMS(g) is more than 70 % over most of the Arctic Ocean.

The SO2 formed from DMS(g) oxidation will further undergo oxidation in the atmosphere by OH radical to form sulfuric acid, which can either nucleate to form new particles or condense on existing particles. In GEM-MACH, the nucleation and condensation of sulfuric acid are treated as two competing processes. The H2SO4–H2O nucleation rate is parameterized following Kulmala et al. (1998), and the condensation rate is parameterized based on the modified Fuchs–Sutugin equation (Fuchs and Sutugin, 1971). The combined nucleation–condensation equation is solved using an accelerated iterative scheme as described in Gong et al. (2003). The sulfate mass produced by nucleation is placed in the model's smallest size bin; the treatment of condensational growth of particles is handled by the same mechanism as described in Jacobson et al. (1994). As a result, the inclusion of DMS will induce changes in modelled aerosols in GEM-MACH, both in mass concentration and size distribution (Croft et al., 2016a).

Figure 13 shows the changes (both absolute and relative) in modelled July-averaged aerosol sulfate mass concentration due to DMS (or the production of biogenic sulfate) at the lowest model level. This difference is approximately in the range from 1 to 20 ng kg-1, or ∼1.50 to 25.0 ng m-3 (<10 %), in the high Arctic (<80∘ N) and is higher in the lower Arctic (e.g. up to 100 ng kg-1, or ∼125.0 ng m-3, in Baffin Bay). Moreover, the increase in sulfate mass is significant for the east and southwest of the domain with higher DMS(g) (e.g. the North Atlantic and off the coast of southern Alaska).

Figure 14 shows the relative mass change due to DMS chemistry in aerosol sulfate (July-average) for four different size ranges: 10–50, 50–100, 100–200 and 200–1000 nm. Examination of the different size ranges indicates that the most significant relative sulfate additions due to DMS reside in the smaller sizes (10–200 nm). It is interesting to note that the enhancement in sulfate due to DMS is more pronounced in the size range of 50 to 100 nm at higher latitudes (>70∘ N). This is in contrast to the enhancement at lower latitudes (<70∘ N) which is more evident in the size range from 200 to 1000 nm. This could be an indication for more favourable conditions for nucleation under the cleaner environment at high latitudes, whereas condensation onto existing aerosol is favoured for DMS-derived sulfuric acid at lower latitudes, as found in Leaitch et al. (2013).

Abbatt et al. (2019) showed the highest increase in particle number concentration to be between 15 and 50 nm at Alert, Nunavut, during July and August associated with new particle formation and growth from natural sources (see Fig. 7 of Abbatt et al., 2019). They estimated the contribution of natural sources to particles in the 30 to 50 nm size range to be around 20 % to 70 %. Figure 14 shows a 20 %–50 % and an approximate 50 % increase in sulfate particle mass between 10 and 50 nm and 50 and 100 nm respectively for July around Alert. In general, GEM-MACH suggests the enhancement of particles between 50 and 100 nm to be higher than particles between 10 and 50 nm for the high Arctic. This difference between Abbatt et al. (2019) and GEM-MACH results could be partly due to other missing natural sources (e.g. organics, see Croft et al., 2019; Burkart et al., 2017; Willis et al., 2016) in the model. Possible inadequacy in the model representation of particle nucleation processes and the over-prediction of the condensation sink may also contribute to the size discrepancy between the model and observations. For example, in the model, new particles formed via nucleation are added to the first model size bin (10–20 nm) at sizes considerably bigger than nucleating particles in the real world (e.g. Kulmala et al., 2006). In addition, the size discrepancy between the model and observations could be due in part to the role of MSA in the nucleation/growth of particles, which is not considered for this modelling study. However, recent field measurement studies have found that while MSA correlates with particle growth in the Arctic, it is not a major contributor to the mass of Aitken-mode particles or their growth (Willis et al., 2016; Tremblay et al., 2019). Furthermore, using the GEOS-Chem model, Croft et al. (2019) found that the condensation of MSA does not lead to sufficient particle growth; they were also not able to reproduce observed size distributions in the Arctic, even when particle growth by MSA condensation was included in the model. Further studies are needed to quantify the contribution of MSA in nucleation/growth of aerosol in the Arctic atmosphere.

The modelled size-resolved sulfate increments due to DMS are compared to the measurements of size-resolved biogenic sulfate onboard the Amundsen cruise during the 2014 NETCARE campaign. Size-segregated aerosol samples were collected and analysed for sulfur isotopes in order to apportion total aerosol sulfate into different origins: biogenic, anthropogenic and sea salt (Ghahremaninezhad et al., 2016). The modelled size-resolved aerosols were mapped onto the size ranges of the observation. Figure 15 shows the comparison for three smaller size ranges: <0.49, 0.49–0.95 and 0.95–1.5 µm. The aerodynamic diameters were converted to dry diameters using a dry diameter correction factor of 2.3 (Ming and Russell, 2001). Based on their analysis, Ghahremaninezhad et al. (2016) indicated that most of the biogenic sulfate resides are in the smallest size range (<0.49 µm; refer to Fig. 6c in Ghahremaninezhad et al., 2016). At higher latitudes (16 July and later), the model simulation in this study compares well with the observations and also demonstrates that a larger fraction of DMS-derived sulfate (or biogenic sulfate) is found in aerosols with sizes <0.49 µm.

During July 2014, Leaitch et al. (2016) found a strong influence of the particles in the size range from 20 to 100 nm in the cloud droplet number concentrations (CDNC) in liquid clouds in the eastern Arctic over the Resolute Bay area. Quinn et al. (2017) also emphasized the role of the non-sea-salt sulfate aerosol as the primary CCN between 70∘ S and 80∘ N. Furthermore, Ghahremaninezhad et al. (2016) found that biogenic sulfate particles are the dominant non-sea-salt sulfate particles during July in the Arctic atmosphere, and >63 % of non-sea-salt sulfate fine aerosol (<0.49 µm) were from a biogenic source (DMS). Our modelling results indicate the formation/growth of biogenic sulfate aerosol in the size range from 10 to 200 nm. These results suggest that the non-sea-salt sulfate aerosol in the summertime Arctic is dominated by fine and ultrafine biogenic particles, which may act as a CCN and/or influence CDNCs and play a climatic role.

We examined the impact of possible additional SO2 formation from the OH-addition pathway via the MSIA–OH reaction, which is the crossover point between the addition and abstraction pathways (von Glasow and Crutzen, 2004). In this oxidation sensitivity test we considered a 75 % yield of SO2 from OH addition following Chin et al. (1996). Figure 16 shows the relative percentage of the July-averaged SO2 with and without the 75 % yield of SO2 from the OH-addition pathway. In this figure the difference (up to >50 % over the central Arctic, and 30 %–50 % over the Canadian Archipelago) is more pronounced above 70∘ N, where the effect of the OH-addition pathway is significant. As OH addition dominates DMS oxidation in the Arctic environment (particularly the high Arctic) due to low temperature (as discussed above), this additional SO2 formation mechanism can be important here.

Figure 17 shows the relative (percentage) change in the model-averaged sulfate aerosol with and without the 75 % yield of SO2 from the OH-addition pathway for different size ranges. This pathway exerts an important influence on the fine sulfate aerosol between 10 to 200 nm (mostly 50–100 nm) in size above 70∘ N. As shown in Fig. 17, the increase in sulfate aerosol mass in the size range from 10 to 200 nm is more than 30 % for some regions due to adding this pathway. These results emphasize the potential importance of the crossover point between the addition and abstraction pathways above 70∘ N.

Leaitch et al. (2013) estimated the July and August averaged sulfate mass concentrations at Alert to be equal to 25 and 84 ng m-3 for the particles in the size ranges of 20–100 and 20–200 nm respectively (based on 1 year of observations and a 100 % conversion from the integrated volume concentration measured under clean conditions). However, assuming that sulfate comprises about 40 % of the submicron particles in clean air at Alert (Leaitch et al., 2018), the sulfate mass concentrations are estimated to be in the range of, at least, 10 ng m-3 for a particle size range from 20 to 100 nm, and 35 ng m-3 for a particle size range from 20 to 200 nm at Alert from natural sources. The GEM-MACH results show the increase in the July-averaged sulfate particle mass due to DMS to be slightly less than 10 ng m-3 without the 75 % yield of SO2 and more than 15 ng m-3 with the 75 % yield of SO2 in the size range from 10 to 200 nm at Alert.