Sea salt aerosols (SSA) are produced via wave breaking in the open ocean (Lewis and Schwartz, 2004; De Leeuw et al., 2011, and references therein). Over polar regions, SSA can also be generated via sublimation of saline blowing snow (Simpson et al., 2007; Yang et al., 2008), wind-blown frost flower crystals (Rankin et al., 2000; Domine et al., 2004; Xu et al., 2013), and by leads in sea ice (Nilsson et al., 2001; May et al., 2016). SSA production from blowing snow events requires strong winds (>7 m s-1) and depends on the salinity of snow cover on sea ice (Yang et al., 2008). Frost flowers grow over new sea ice formed from open leads under low ambient temperatures (<-20 ∘C; Kaleschke et al., 2004). These two sea ice sources of SSA have been proposed to help explain polar observations of wintertime maxima in SSA mass concentrations (Wagenbach et al., 1998; Weller et al., 2008; Jourdain et al., 2008; Udisti et al., 2012; Huang and Jaeglé, 2017), the depletion of sulfate-to-sodium mass ratio in winter SSA relative to bulk sea water at sites in Antarctica (Wagenbach et al., 1998; Rankin et al., 2000; Jourdain et al., 2008; Hara et al., 2012) and some sites in the Arctic (Jacobi et al., 2012; Seguin et al., 2014) as well as the increase in Na+ deposition fluxes during glacial periods relative to interglacial periods (Wolff et al., 2006; Fischer et al., 2007; Abram et al., 2013).

In a previous study (Huang and Jaeglé, 2017), we used the GEOS-Chem chemical transport model to examine the relative roles of blowing snow and frost flowers as sources of polar SSA during winter. Our study was based on the blowing snow parameterization developed by Yang et al. (2008) and the frost flower parameterization of Xu et al. (2013). We compared our simulations to in situ observations of SSA mass concentrations at three surface sites in the Arctic and two sites in coastal Antarctica, showing that blowing snow appeared to be the dominant source of polar SSA during winter. Here, we further constrain the spatial and temporal distribution of polar sources of SSA by using observations of aerosol extinction coefficients from the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) instrument onboard the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) satellite together with the GEOS-Chem model.

One of the main uncertainties in estimating blowing snow emissions is the salinity of surface snow on sea ice. In our previous work (Huang and Jaeglé, 2017), we assumed constant surface snow salinity over Arctic (0.1 practical salinity unit, or psu) and Antarctic (0.03 psu) sea ice. In reality, surface snow salinity is highly variable in time and space. The sources of sea salt in snow over sea ice include the upward migration of brine from the sea ice surface, incorporation of frost flowers, and SSA deposition from the nearby open ocean (Domine et al., 2004). Initial sea ice formation is accompanied by upward salt transport, such that first-year sea ice (FYI) has a high salinity, reaching 20–100 psu at the sea ice surface (Weeks and Lee, 1958; Martin, 1979; Weeks and Ackley, 1986). Nakawo and Sinha (1981) found that sea ice salinity decreases rapidly within the first week of sea ice formation in fall, and then decreases more slowly between December and May in the Canadian Arctic. Worby et al. (1998) showed that ice with a thickness of less than 0.05 m displayed salinities of 9–28 psu, while ice thicker than 0.05 m had salinities of 4–8 psu, which decreased linearly with ice thickness. As snow accumulates on FYI throughout winter, the brine is wicked upward, resulting in brine-wetted snow (Barber et al., 1995). Snow salinity is highest in the first 10 cm above FYI, with values of 1–20 psu (Geldsetzer et al., 2009), and then decreases rapidly as the snow cover gets thicker, with low salinities on the surface of thick snow (Nandan et al., 2017). Older and thicker multiyear sea ice (MYI), is desalinated by flushing and gravity drainage during repeated summer melt cycles, such that the overlaying snow has very low salinity. Krnavek et al. (2012) reported that surface snow salinity sampled on MYI was 0.01 psu, compared to 0.1 psu for snow on thick FYI and 0.8 psu on recently frozen thin FYI in March near Barrow, Alaska.

The role of frost flowers as a direct source of SSA remains subject to debate. Some studies have shown that strong winds inhibit frost flower formation and bury existing frost flowers with snow (Perovich and Richeter-Menge, 1994; Rankin et al., 2000), while field experiments show that frost flowers are difficult to break (Domine et al., 2005; Alvarez-Avilez et al., 2008; Obbard et al., 2009). In addition, laboratory experiments show that evaporating frost flowers form a cohesive chunk of salt (Yang et al., 2017), which is unlikely to be a direct source of SSA, even when exposed to large wind speeds (Roscoe et al., 2011).

In this study, we use 3 years (2007–2009) of CALIOP aerosol extinction coefficients to constrain the spatial and temporal distribution of polar SSA emissions with the GEOS-Chem model. Satellite observations and GEOS-Chem simulations are described in Sect. 2. In Sect. 3, we evaluate the model's ability to reproduce observed aerosol extinction over the Arctic and Antarctic sea ice regions with and without sea ice sources of SSA. In Sect. 4, we develop an empirical parameterization of seasonally varying surface snow salinity on FYI. In Sect. 5, we conduct a case study of an Arctic blowing snow event and the resulting SSA observed by CALIOP.

While the inclusion of blowing snow leads to improved agreement with CALIOP, we hypothesize that the remaining discrepancies in the magnitude and seasonal cycle of aerosol extinction coefficients are due to our simplified assumption of a temporally uniform surface snow salinity over FYI.

As discussed in Sect. 1, the surface snow salinity is highest over thin FYI with little snow cover early in the cold season, declining in the ensuing months as a result of thickening sea ice and increasing snowpack depth (Barber et al., 1995; Krnavek et al., 2012; Weeks and Lee, 1958; Weeks and Ackley, 1986; Nakawo and Sinha, 1981). As no systematic observations of surface snow salinity are available over sea ice, our approach is to find the monthly salinity of snow on FYI required to minimize the discrepancy between CALIOP extinction and the GEOS-Chem simulation. The salinity of surface snow on MYI is the same as in the STD+Snow simulation.

By using a linear regression between the GEOS-Chem simulation and CALIOP monthly extinction over Arctic sea ice, we derive a surface snow salinity on FYI of 0.9 psu in September, decreasing to 0.09 psu in April (with values of 0.36, 0.26, 0.19, 0.16, 0.16, and 0.14 between October and March). The inferred snow salinities decrease with time and are generally consistent with observations near Alaska reported by Krnavek et al. (2012): 0.8 psu for snow over recently frozen thin FYI and 0.1 psu over thick FYI. For Antarctic FYI, we infer snow salinities of 0.05 psu in April, 0.02 in May–June, and 0.018 in July–September. For the rest of the year the salinity is 0.015 psu. These decreasing trends in salinity between fall and spring are consistent with expectations based on the seasonal evolution of FYI thickness, sea ice surface salinity, and deepening snow cover (Worby et al., 1998; Warren et al., 1999; Massom et al., 2001; Kwok and Cunningham, 2015).

We use our empirically derived monthly snow salinities to conduct an optimized blowing snow simulation (STD+Opt. Snow). Over Arctic FYI, the model bias changes from -7 % (STD+Snow) to +8 % (STD+Opt. Snow), and for MYI, the model bias of -17 % changes to -2 % (Fig. 2b and c). Over Antarctic FYI, the model overestimate decreases from +43 % (STD+Snow) to +4 % (STD+Opt. Snow). Similarly, the model bias over MYI decreases to -10 % (Fig. 2e and f). The STD+Opt. Snow simulation displays cold-season extinction profiles that are within 5–10 % of CALIOP observations over sea ice (Figs. 3d–e and 5d–e). The seasonal cycles are in better agreement with observations, especially in the Arctic for October–December, when the inferred FYI salinities (0.36–0.19 psu compared to 0.1 psu in the STD+Snow simulation) lead to a near doubling of aerosol extinction (Fig. 3g–h). Over Antarctica, the amplitude of the seasonal cycle of aerosol extinction over FYI is reduced, in better agreement with CALIOP observations (Fig. 5g–h).

We also examined whether a single fixed value of salinity over FYI can lead to similar improvements in the agreement with CALIOP. The resulting fixed salinities are 0.11 psu over Arctic FYI and 0.018 psu over Antarctic FYI, leading to good overall agreement with CALIOP over the Antarctic (NMB of +5 % on FYI and -9 % on MYI) but with no significant improvement seen in the Arctic (NMB of -7 % on FYI and -18 % on MYI). We found that over the Arctic, a simulation using a single salinity of 0.11 psu (STD+Const. Snow, Fig. S8g–h) yields results similar to the STD+Snow simulation and cannot explain the high extinction values during fall and early winter. Over Antarctic sea ice, the performance of a simulation with 0.018 psu over FYI shows results similar to the STD+Opt. Snow simulation. Thus there is a stronger case for using a seasonally varying snow salinity over Arctic sea ice than over Antarctic sea ice. We speculate that this might be linked to relatively smaller seasonal variation in sea ice thickness and snow depth for Antarctic sea ice compared to the Arctic. In their snow climatology, Warren et al. (1999) report that the mean snow depth at an Arctic sea ice site increased from 8.7 cm in October to 28.9 cm in March. Satellite-based observations of Arctic FYI thickness show an increase from 0.95 m in October to 2.15 m in May (Kwok and Cunningham, 2015). In contrast, over Antarctic sea ice, the mean sea ice thickness and snow depth remained fairly constant during fall–winter (April: 0.48 m for ice thickness and 0.11 m for snow depth, August: 0.52 m for ice thickness and 0.11 m for snow depth) as described in Worby et al. (1998).

Both STD+Snow and STD+Opt. Snow simulations underestimate CALIOP aerosol extinction over the Canadian Arctic Archipelago (Fig. 3f). Combining the STD+Opt. Snow and frost flower emissions could help improve the agreement in that region, but it would also lead to substantial overestimates over FYI and MYI for the rest of the Arctic. One possibility is that snow-covered frost flowers in the Canadian Archipelago increase the local surface snow salinity (Domine et al., 2004). The recent study of Hara et al. (2017) in northwestern Greenland proposed that snowfall buries frost flowers and the associated slush layer on new FYI. The resulting brine migrates vertically, enriching the surface snow layer, which can be mobilized under strong winds. We estimate that increasing the salinity of snow over the Canadian Archipelago to a value of 3 psu would help reconcile the optimized blowing snow simulation with CALIOP observations. Direct measurements of snow salinity in this region would help confirm this estimate. Similarly, we find that increasing the snow salinity near the Ross Ice Shelf region, where frost flowers are expected to occur, would improve the agreement with CALIOP observations.

Figure 6 evaluates the performance of the STD+Opt. Snow simulation against independent observations of SSA mass concentrations in Barrow, Alaska (71.3∘ N, 156.6∘ W); Alert, Nunavut, Canada (82.5∘ N, 62.5∘ W); Zeppelin, Svalbard, Norway (78.9∘ N, 11.9∘ E); Neumayer, Antarctica (70.7∘ S, 8.3∘ W); and Dumont d'Urville, Antarctica (66.7∘ S, 140∘ E). Descriptions of the in situ observations are provided in Huang and Jaeglé (2017). At Barrow, the optimized simulation improves the agreement with observed SSA mass concentrations in November–May. In particular, the enhanced salinity in October–November brings the model closer to the observations. At Zeppelin and Dumont d'Urville, the model bias in the STD+Opt. Snow simulation is reduced relative to the STD+Snow simulation. However, the model bias worsens at Neumayer (STD+Snow: -25 %, STD+Opt. Snow: -48 %), and the model bias remains large at Alert (STD+Snow: -50 %, STD+Opt. Snow: -32 %). As Neumayer and Alert are close to the frost-flower-producing regions, compared to other polar sites, this underestimation may be related to an underestimation in the snow salinity in those regions.

It is also possible that the discrepancies between observed and modeled aerosol extinction coefficients are due to other factors in the blowing snow parameterization as implemented in GEOS-Chem. For example, our simulation does not include the negative feedback of water vapor sublimation (Mann et al., 2000); as blowing snow particles sublime in unsaturated air, they cause an increase in water vapor and thus the cooling of the surrounding air. Both effects lead to an increase in RH near saturation, reducing the sublimation rate. Another underlying assumption is that 5 SSA are produced for each snowflake that sublimes (N=5). We conducted a sensitivity simulation assuming one SSA per snowflake, shown as STD+Snow (N=1) in the Supplement (Figs. S7 and S8). This change does not affect the total emission of blowing snow SSA, but it decreases the fraction of SSA in the accumulation mode (see Huang and Jaeglé, 2017). As the extinction efficiency of accumulation mode SSA is larger than that of coarse mode SSA, the assumption of N=1 leads to a 30 %–50 % decrease in modeled extinction relative to the STD+Snow (N=5) simulation. Overall, this results in improved agreement with CALIOP observations over Antarctic sea ice, but the CALIOP aerosol extinction is underestimated over the Arctic. Increasing the surface snow salinity over Arctic FYI can address the model discrepancy in aerosol extinction coefficients, but it will lead to a factor of 1.5–2 overestimate in SSA mass concentrations.

Figure 7 shows a case study of a blowing snow SSA event, which occurred on 6 November 2008 over the Arctic. The STD+Snow and STD+Opt. Snow simulations display enhanced extinction coefficients (40–80 Mm-1) over the Barents Sea along the 60∘ E longitude line extending from the North Pole to 70∘ N (Fig. 7b and c). This feature is due to blowing snow SSA, and it is not seen in the STD simulation (Fig. 7a) or the STD+FF simulation (not shown).

The CALIPSO 00:58–01:12 UTC overpass on 6 November 2008 transected this feature, with CALIOP aerosol extinction coefficients of 50–150 Mm-1 between points A (78∘ N, 52.5∘ E) and B (82∘ N, 110∘ E), labeled in Fig. 7a–c. The cross section along the CALIOP overpass shows that the large aerosol extinction coefficients are confined between the surface and 1–2 km altitude (Fig. 7e). This overpass region is mostly covered by FYI (Fig. 7d). We sampled GEOS-Chem along the same cross section, finding a very good correspondence in the spatial extent of the feature observed by CALIOP and simulated by the STD+Snow and STD+Opt. Snow simulations (Fig. 7g–i). The optimized blowing snow model predicts higher aerosol extinction due to larger surface snow salinity on FYI in November and is in better agreement with the 0–2 km CALIOP mean aerosol extinction coefficients (Fig. 7i).

We use the FLEXPART particle dispersion model (Stohl et al., 1998, 2005; Seibert and Frank, 2004) with meteorological data from ERA-Interim with a horizontal resolution of 0.5∘ to track the origin of this feature. We release 100 000 particles between points A and B at 0.01–2 km and track them back in time over 2 days. The air along the A–B transect in Fig. 7 originates from the 120–140∘ E sector at 60–90∘ N near the surface (0–100 m), on 4 November 2008 (Fig. 8a). This region is covered by both MYI and FYI (Fig. 7d). The STD+Snow model predicts enhanced blowing snow emissions in this region on 4 November 2008 (Fig. 8b). The CD transect in Fig. 8b shows the CALIOP overpass at 01:11–01:24 UTC, 4 November 2008. This CALIOP transect displays elevated 532 nm attenuated perpendicular backscatter and depolarization ratios below 200 m (Fig. 8c and d), which are co-located with strong surface winds (>5 m s-1). The pattern of elevated backscatter, surface winds and depolarization ratios satisfies the requirements of the CALIOP blowing snow detection algorithm described in Palm et al. (2011, 2017), who defined blowing snow events as layers with high color ratios (>1), high depolarization ratios (>0.25), strong surface winds (>4 m s-1), and enhanced backscatter signals below 300 m (ranging between 2.5×10-2 and 0.2 km-1 sr-1). The FLEXPART-predicted source region and CALIOP blowing snow feature are co-located with enhanced blowing snow emissions in the GEOS-Chem simulation (Fig. 8b). This case study thus shows that CALIOP can detect not only the blowing snow event (Palm et al., 2011) but also the resulting SSA produced after sublimation.

In this work, we used the GEOS-Chem chemical transport model to assess the ability of the CALIOP lidar onboard the CALIPSO satellite to provide constraints on sea ice sources of SSA. We find that mean CALIOP aerosol extinction coefficients below 2 km altitude reach 10–15 Mm-1 over sea-ice-covered regions during the 6 month polar cold season. The enhanced extinction is located below 2 km, with the largest values (20–35 Mm-1) occurring near the surface. We find that a standard GEOS-Chem simulation without sea ice sources of SSA underestimates CALIOP extinction by 50 %–70 % over Arctic and Antarctic sea ice. A simulation with frost flower SSA emissions is unable to explain the spatial and temporal distribution of CALIOP aerosol extinction. Adding a blowing snow SSA source results in improved agreement over the Arctic (NMB =-7 % for FYI and NMB =+17 % over MYI) but yields a 43 % overestimation of CALIOP extinction coefficients over Antarctic sea ice. Additionally, the simulation including blowing snow SSA tends to underestimate CALIOP observations during fall to early winter (October–December over the Arctic and April–June over the Antarctic).

We hypothesize that our assumption of constant surface snow salinity on FYI (0.1 psu over the Arctic and 0.03 psu over the Antarctic) in the blowing snow simulation could explain the remaining discrepancies between observed and modeled extinction. Given the paucity of snow salinity observations, we infer the monthly surface snow salinity on FYI required to minimize the discrepancy between CALIOP and the GEOS-Chem simulation. The resulting snow salinities decrease progressively between the beginning and end of the cold season (from 0.9 to 0.09 psu for Arctic FYI; 0.05 to 0.018 psu over Antarctic FYI). This decrease is consistent with the seasonally increasing sea ice thickness and accumulating snow depth. The optimized blowing snow model using the monthly varying snow salinities shows improved agreement with CALIOP observations and in situ observations of SSA mass concentrations at five surface sites. However, the optimized blowing snow model tends to underestimate the aerosol extinction over the Canadian Arctic Archipelago and off the Ross Ice Shelf. Both regions are predicted to favor frost flower growth, which could locally increase the salinity of snow when frost flowers are buried under snow. We find that increasing the Canadian Arctic Archipelago FYI snow salinity to 3 psu would help reconcile our simulation with CALIOP and in situ observations. Our work, however, cannot rule out other alternative factors contributing to the discrepancy between modeled and observed aerosol extinction, such as the impact of the negative feedback of water vapor sublimation and our assumption about the number of particles produced per snowflake. Systematic observations of surface snow salinity over multiple sea ice locations and times would help further constrain snow salinities in the Arctic and Antarctic. Furthermore, more extensive observations of sea salt aerosol size distributions during blowing snow events could help further refine and constrain these assumptions.

We conduct a case study of a blowing snow SSA event over the Arctic, which was detected by CALIOP on 6 November 2008 and predicted by our blowing snow simulation. Using FLEXPART, we find that the observed aerosol extinction layer originated 2 days earlier over sea ice below 100 m altitude. We demonstrate that CALIOP detects this blowing snow event below 200 m altitude with enhanced extinction and a large depolarization ratio, co-located with surface high winds.

Our work suggests that blowing snow emissions are the dominant source of SSA over sea ice covered regions during cold months. As SSA can act as a source of halogens, the inclusion of blowing snow in chemical transport models is important in understanding springtime bromine explosions and the resulting ozone and mercury depletion events (Schroeder et al., 1998; Simpson et al., 2007; Steffen et al., 2008; Gilman et al., 2010; Yang et al., 2010). Furthermore, these sea-ice sources of SSA can act as ice nuclei for cloud formation and may increase the downward longwave radiative forcing (Xu et al., 2013). Arctic sea ice has been rapidly changing over the past 30 years, with decreasing sea ice extent and thickness (e.g., Kwok and Rothrock, 2009), a shift towards less MYI and more FYI (Fowler et al., 2004; Maslanik et al., 2007), and a thinning of snow depth in spring (Renner et al., 2014; Blanchard-Wrigglesworth et al., 2015). All these factors are likely to have resulted in an increase in snow salinity on sea ice, hence increasing SSA emissions from blowing snow. In the Southern Hemisphere, sea ice extent is FYI-dominant, and its annual and seasonal trends are more complex, varying in space. Sea ice extent has been increasing over the Ross Sea, but it has been decreasing over the Amundsen–Bellingshausen Sea over the past decades (Turner et al., 2009; Parkinson and Cavalieri, 2012; Stammerjohn et al., 2012). Consequently, this may have resulted in a shift in the spatial pattern of blowing snow SSA emissions, with increased influence over the Ross Sea and reduced influence over the Amundsen–Bellingshausen Sea.