Blowing snow over sea ice has been proposed as a significant source of sea salt aerosol (SSA) (Yang et al., 2008). In this study, using snow salinity data and blowing snow and aerosol particle measurements collected in the Weddell Sea sea ice zone (SIZ) during a winter cruise, we perform a comprehensive model–data comparison with the aim of validating proposed parameterizations. Additionally, we investigate possible physical mechanisms involved in SSA production from blowing snow. A global chemical transport model, p-TOMCAT, is used to examine the model sensitivity to key parameters involved, namely blowing-snow size distribution, snow salinity, sublimation function, surface wind speed, relative humidity, air temperature and ratio of SSA formed per snow particle. As proposed in the parameterizations of Yang et al. (2008), the SSA mass flux is proportional to the bulk sublimation flux of blowing snow and snow salinity. To convert the bulk sublimation flux to SSA size distribution requires (1) sublimation function for snow particles, (2) blowing-snow size distribution, (3) snow salinity and (4) ratio of SSA formed per snow particle.
The optimum model–cruise aerosol data agreement (in diameter range of 0.4–12
Over most of the Earth, primary sea salt aerosol (SSA) derives from wave breaking and bubble bursting at the open-ocean surface (e.g. de Leeuw et al., 2011). SSA is relevant to radiative forcing of climate because it can efficiently scatter solar radiation (O'Dowd et al., 1997; Murphy et al., 1998; Quinn et al., 2002). Moreover, SSA can serve as cloud condensation nuclei (CCN) (e.g. O'Dowd and Smith, 1993; O'Dowd et al., 1997, 1999) and even ice-nucleating particles (INPs) (Wise et al., 2012; DeMott et al., 2016) that influence global climate.
Observations of sulfate depletion relative to sodium in Antarctic aerosol
and snow samples first argued for a sea ice source of SSA (Wagenbach et al.,
1998; Rankin and Wolff, 2003; Jourdain et al., 2008; Legrand et al., 2017).
The depletion of sulfate is due to the effect of mirabilite
(
Saline crystals on sea ice, such as frost flowers (FFs) (e.g. Rankin et al., 2000, 2002; Kaleschke et al., 2004; Xu et al., 2016) with relatively high salinity and blowing snow (Yang et al., 2008) with relatively low salinity, were both suggested as potential sources of SSA. Evidence from laboratory chambers (Roscoe et al., 2011; Yang et al., 2017) and field measurements (Obbard et al., 2009; Hara et al., 2017) indicate that FFs are unlikely to be a major direct source. Global models with blowing snow as a SSA source can successfully reproduce winter SSA peaks at high latitudes (Levine et al., 2014; Huang and Jaeglé, 2017; Rhodes et al., 2017). In addition, chemistry transport model studies demonstrate that when this sea-ice-sourced SSA is treated as a source of bromine to the boundary layer, the polar springtime bromine explosion events as well as the associated ozone depletion events can be largely reproduced (Yang et al., 2010; Theys et al., 2011; Legrand et al., 2016; Zhao et al., 2016, 2017; Choi et al., 2018). However, the SSA production parameterizations implemented in models have not been fully validated against field data, and the possible physical mechanisms involved in the SSA formation are not completely clear.
In this study, based on a comprehensive set of measurements for both blowing-snow particles and aerosol particles (Frey et al., 2019), made during a
winter cruise on board the icebreaker RV
The measurements used for the model validation were carried out during a
winter sea ice cruise in the Weddell Sea, Antarctica, aboard the German ice breaker RV
Our global chemistry transport model, p-TOMCAT, has a detailed process-based
SSA scheme (Levine et al., 2014). The following updates have been introduced
to this model in recent studies: more realistic model precipitation fields
(Legrand et al., 2016), a sea spray emission scheme following the work of
Jaeglé et al. (2011) and a modified surface snow salinity distribution
function (Rhodes et al., 2017). Both open-ocean-sourced and sea-ice-sourced
SSA (as dry NaCl) are tagged in 21 size bins with size ranging from 0.02 to 12
The meteorological forcing files for the model are 6-hourly reanalysis
ERA-Interim data from the European Centre for Medium-Range Weather
Forecasts (ECMWF). Monthly sea ice coverage and sea surface temperatures
(SSTs) are taken from the Hadley Centre Sea Ice and Sea Surface Temperature
(HadISST) dataset (Rayner et al., 2003). The model's horizontal resolution
is 2.8
The experiments carried out are summarized in Table 1. In the control run for sea-ice-sourced SSA (SI_Base_A) a constant water mass loss rate against time for the snow particle sublimation rate is assumed (see Sect. 3.3.1), and mode A (Fig. 2) is used to represent the blowing-snow-particle distribution function. There are three additional runs – SI_Classic, SI_Area and SI_Mass (included in the prefix of experimental names) – performed with the aim of investigating possible mechanisms involved in the SSA production (see Sect. 3.3.1). The control run for open-ocean sea spray is SI_Base_OO, following the scheme by Jaeglé et al. (2011).
Normalized 29 m SPC instrumental blowing-snow size distribution is
shown by the black line. Note that the dotted line is for small particles with a diameter
Model experiments for sea-ice-sourced SSA (with SI in prefix of each experiment) and sea spray fluxes (with OO in the prefix). Columns 2–10 show parameters applied to each experiment: sublimation function, shape of the blowing-snow size distribution, ratio of SSA formed per blowing-snow particle, snow age, salinity, threshold wind speed, RH (with respect to ice) and air temperature.
Apart from the global modelling investigations, an idealized theoretical
calculation of the SSA production flux is made to compare with the sea spray
flux under the same wind speed of 12 m s
According to the scheme proposed in Yang et al. (2008, 2010), the SSA flux
from blowing snow is proportional to the bulk sublimation flux
In order to demonstrate how to calculate the SSA flux from the bulk sublimation flux, here we simplify things by assuming (1) all snow particles have a
uniform salinity
Under the above assumption, the corresponding dry NaCl size,
At steady state, the SSA number production flux,
Obviously, how to derive
With Eq. (7), the bulk sublimation flux can be allocated into each snow
size bin.
If more than one SSA is formed per snow particle, and assuming they are all
equal in size, then at a ratio of
Figure 3 shows the equivalent dry NaCl diameter (
Equivalent dry SSA diameter (
As pointed out above, at steady state, the snow particle loss rate via the
sublimation process should be balanced by newly supplied/generated blowing-snow particles for each size bin to keep the snow particle size distribution
unchanged with time. In windy conditions, vertical mixing via eddy
turbulence is relatively fast; thus the timescale of mixing could be much
shorter than that for the sublimation process. For instance, for a droplet
with a size of tens of microns, to evaporate it completely may take a few
thousands of seconds (Mason, 1971), which is substantially longer than the
timescale of tens to hundreds of seconds in boundary layer turbulent mixing
(Caughey et al., 1979). Therefore, the newly generated small snow crystals
could be efficiently brought upwards, via rebound and splashing of snow
grains in the saltation layer (
As shown in Table 1, there are four sublimation functions applied to the
In SI_Classic runs (with SI_Classic in the
prefix), a simple relation function of
A third sublimation function of
A fourth function of
We hope that, by comparing the SSA size spectrum between model integrations and the observations, we may assess which of these functions could be most appropriate.
It has been found that suspended blowing-snow particles follow a
two-parameter gamma distribution function
The
Similar to the previous modelling study by Rhodes et al. (2017), a surface snow salinity distribution is applied (e.g. see Fig. 12 of Frey et al., 2019), which is based on surface snow samples (top 10 cm) collected in the Weddell Sea SIZ. In the Arctic, snow salinity values are trebled to reflect the likelihood that snow there is more saline than in the Antarctic due to reduced precipitation rate (Yang et al., 2008). Further, we make the rate of SSA emission from multi-year sea ice half that from first-year sea ice (Rhodes et al., 2017). We note that these assumptions will not affect the main conclusions of this study.
As reflected in Eq. (1), SSA size is proportional to salinity with a
power of
How snow age affects blowing snow and SSA production is not completely
clear, though it has generally been thought that aged snow will be more
resistant to wind mobilization (Li and Pomeroy, 1997; Box et al., 2004).
Snow age was initially introduced to the parameterization to counteract the
relatively high snow salinity used (Yang et al., 2008). At present, this
parameter amounts to a crude tuning tool with no clear physical meaning.
Snow age
Actually, the “snow” here refers to all ice crystals on the surface snowpack that can be uplifted by air movement. These include fresh fallen snow, diamond dust, wind-cropped frosts or even aged snow that has been re-mobilized by wind erosion. The mixing of fresh snow and old saline snow changes the salinity distribution, a process that has not been considered by the model so far. Due to a lack of data, we do not know how fast fresh fallen snow acquires salts. This process may be fast and efficient during windy conditions through direct physical contact with salt-rich crystals. With further data, we may have a better representation of this process.
Here in this study, we follow a recent study by Huang and Jaeglé (2017)
by setting a snow age
As pointed out by Mann et al. (2000), sublimated water from blowing-snow
particles will raise the RH (with respect to ice) within the blowing-snow
layer, which will have a negative effect on the further sublimation of
wind-blown-snow particles, especially from the near-surface layer. A model
without consideration of this negative feedback may likely overestimate
sublimation and SSA production. The p-TOMCAT model gets its RH field directly
from ECMWF ERA-Interim data. Therefore, it is likely that the model surface
RH is underestimated in the cases with blowing snow. Figure 1c indicates that
the lowest model grid box RHs (with respect to ice) (at an average height of
According to Li and Pomeroy (1997), the threshold wind speed for blowing
snow is air temperature and snow age dependent. According to the bulk
sublimation parameterization of Déry and Yau (1999, 2001), a minimum
threshold of
Due to the large perturbation in air temperature, the threshold wind speed
calculated varies significantly in association with the temperature
perturbation (orange line in Fig. 1b). To test model sensitivity to this
term, model runs with fixed threshold speeds of 7, 8 and 9 m s
In the original parameterizations (Yang et al., 2008), a unit ratio (
Ratios of aerosol number density between model runs and the
observations (for overlapping size range of 0.4–12
Figure 1a shows a comparison of observed total aerosol number density along
the cruise track and model output (size ranging from
For the full analysis, we have regrouped the cruise data into three surface
types: open ocean (before 17 June), marginal sea ice and packed sea ice,
using an air temperature of
Figure 4 shows the simulated aerosol size spectrum in each surface zone. It
can be seen that, over the open ocean (Fig. 4a), sea spray (OO, blue line)
dominates over sea-ice-sourced SSA (in three model runs: SI_Base_A, SI_Classic_AX10 and
SI_Classic_BX20). By looking at the time
series, we find that sea spray shows a significant positive correlation to
the observations with a correlation coefficient of
Size distribution of sea spray and sea-ice-soured SSA at three
defined surface zones:
Once the vessel enters densely packed sea ice (Fig. 4c), the simulated sea
spray contribution drops significantly to only
In marginal sea ice (Fig. 4b), our simulations suggest that both sea-ice-
and open-ocean-sourced SSA are making a contribution to the observations.
However, neither sea-ice-sourced SSA nor sea spray alone shows a strong positive
correlation with the observations. For example, the time series show only a
small positive coefficient
Although the meteorological fields, such as wind speed (Fig. 1b),
temperature (Fig. 1c) and moisture (Fig. 1d), taken from the ERA-Interim
database, in general agree well with the observations, discrepancies between
them can be large during specific time periods. On average, model surface
wind speeds are lower than the observations, especially during storm events;
this is because global models with a coarse spatial resolution suffer
significant spatial averaging and cannot reproduce gusty winds. For example,
a mean wind speed of
At air temperatures of
During 11–13 July, there are two large aerosol enhancement events which are
completely overlooked by the model. As shown in Fig. 1b, they correspond
to relatively low wind speeds (in both reality and model), lower than the
calculated threshold speed of 7 m s
Apart from wind, moisture is another key factor that influences both
sublimation and SSA production. As shown in Fig. 1d, model RHs are generally
lower than the observations, which is likely due to the model not
considering the negative feedback of sublimated water vapour to the near-surface blowing-snow layer, which will limit further water loss from
suspended snow particles. Obviously, models without considering this
feedback effect could result in the overestimation of the SSA flux in some
circumstances. We perform two model experiments with fixed surface
RH (with respect to ice) equal to 90 % in SI_Base_A_R1 and 95 % in SI_Base_A_R2 to investigate this issue. As reflected in Fig. 1a
(orange line, with RH (with respect to ice) equal to 95 %) and Table 2, these two models
results are much closer to the observations. For instance, the model–data
ratio of aerosol number density in the sea ice zone reduces from the control
run 2.76 to 1.8 in the SI_Base_A_R1 and 1.1 in the SI_Base_A_R2 (Table 2). As a result, the time series correlation
coefficients between the model and the observations increase from
The blowing-snow-particle size distribution function also affects the SSA size distribution. A smaller
Averaged SSA size distribution from the whole sea ice zone
(including both marginal and packed sea ice). Observations are shown in the
black line with box symbols. Panel
When a SSA production ratio greater than 1 is applied, the size of the dry
NaCl formed will be reduced (refer to Eq. 11). Thus, at
Global model studies show that the observed winter SSA mass peaks at most
polar sites can only be reproduced when the sea-ice-sourced SSAs are
implemented (Levine et al., 2014; Huang and Jaeglé, 2017; Rhodes et al.,
2017). Figure 6 shows an updated p-TOMCAT result of seasonal Na
concentrations at eight polar stations (based on a 3-year integration,
2013–2015), which reinforces the importance of sea-ice-sourced SSA in
reproducing the winter peaks of sodium observed, as sea spray (solid green
lines) simply cannot do alone. As shown in Fig. 6, the model run
SI_Classic_BX20 (solid yellow lines) gives a
slightly higher Na mass concentrations than the control run
SI_Base_A (red lines); this is due to the
reduction of SSA size, e.g. by a factor of 2.7 when
Monthly mean Na mass concentration at eight polar sites.
Observations are shown in black with diamond symbols, with the uncertainty bars
representing
The three model runs (SI_Base_A,
SI_Classic_AX10 and SI_Classic_BX20) give very similar mass loading (Fig. 6) and
number density at a size of
Zonal mean SSA total number concentration (particles cm
With detailed blowing-snow data, the shape parameter as well as the scale parameter can be well constrained, and then the larger differences in predicted SSA number density in submicron mode among these model runs (mainly between SI_Base_A and SI_Classic_AX10 or SI_Classic_BX20) can be used as indicators for validation, when aerosol data in ultra-fine mode becomes available from SIZ locations.
Overall, the control run SI_Base_A
overestimates SSA number density (as shown in Fig. 1a) and underestimates
mass concentration at sites such as Alert, Barrow and Neumayer (Fig. 6),
indicating that the current model setups and parameterizations applied need
further constraints and evaluation against data. Model runs with a fixed
RH
As discussed in Sect. 3.3.1, under the assumption that one snow particle
only forms one SSA after sublimation, the SI_Mass_A run shows the least correspondence to the cruise observations, by several
orders of magnitude (Fig. 5a). Thus, it is safe to rule out the physical
mechanism represented by the sublimation function implemented, which assumes
that the SSA should come from an unsorted sample of suspended wind-blown-snow particles in the blowing-snow layer that lose their water completely
without any replenishment from newly generated snow particles.
SI_Classic_A and SI_Classic_B runs agree better than SI_Area_A and SI_Mass_A runs but
still cannot generate enough submicron-size SSAs to match the observations.
SI_Base_A and SI_Base_B are, instead, much closer to the observations with
mode-data ratios ranging within 0.8–2.8 (Table 2). As
discussed previously, SI_Base runs claim a particle
sublimation function of
There is a possibility that more than one SSA could be formed from one
saline snow particle. If this is the case, then the discrepancies between
SI_Classic_A (or SI_Classic_B) and the observations could be reduced. For
example, when a SSA production ratio of 10 per snow particle is applied to
SI_Classic_A (denoted as SI_Classic_AX10), or a ratio of 20 to SI_Classic_B (denoted as SI_Classic_BX20), then a result similar to the control run
(SI_Base_A) in a particle size of
Cruise data show that the blowing-snow-particle number densities decrease
significantly, e.g. by more than an order of magnitude from the near surface
(
Model experiments with the above two mechanisms implemented (e.g.
SI_Base_A and SI_Classic_AX10) can produce roughly the same number density at
a size range of
To highlight the above-mentioned SSA production mechanisms and make a direct
comparison with the sea spray flux, a theoretical calculation is performed with
results shown in Fig. 8. The bulk sublimation flux is calculated under
polar weather conditions of wind speed equal to 12 m s
At sub- to micron-size mode, SI_Classic_Aa
shows a comparable flux to sea spray calculated at the same wind speed
following the Jaeglé et al. (2011) scheme with a SST
Apart from a nearly 10 times increase in the number density, compared to
Classic_Aa, Classic_AaX10 also shows a shift
in the SSA size spectrum towards smaller bins with a roughly halved NaCl
size according to Eq. (11), indicating that more smaller SSAs formed, as
shown in Fig. 8d. Figure 8e shows that SI_Base_Aa and SI_Classic_AaX10
have the largest submicron SSA accumulation fraction, accounting for
The assumption that one blowing-snow particle only forms one SSA after sublimation means, at steady state, that the SSA number production rate should be the same as the snow particle loss rate and the replenishment rate of newly formed snow particles. For that reason, Eq. (2) can be used to describe the blowing-snow-particle production flux (in vertical dimension) due to the sublimation effect (Fig. 8b). However, our cruise data will not allow us to validate this flux and derive any robust conclusion.
The Weddell Sea cruise data give us a unique opportunity to constrain some
key parameters involved in SSA production, validate parameterizations and
investigate possible microphysical processes involved. Unfortunately, due
to a lack of data at smaller particle sizes, e.g.
All data used in plotting are stored in the UK Polar Data Centre, Natural Environment Research Council, UK Research and
Innovation (
EWW, MMF, AEJ and PSA designed the field experiment; MMF carried out the field measurements. XY designed and performed model experiments and interpreted the model output and the microphysical mechanisms proposed. RHR contributed to model development, SJN and IMB to CLASP, and KN to SPC instruments. XY prepared the manuscript with contributions from all co-authors.
The authors declare that they have no conflict of interest.
We thank the Alfred Wegener Institute for Polar and Marine Research, which made our participation in the cruise possible. We gratefully acknowledge financial support from the Natural Environment Research (NERC) in the UK through the project BLOWSEA. Rachael H. Rhodes was supported by a European Commission Horizon 2020 Marie Sklodowska-Curie Individual Fellowship. Eric W. Wolff is supported by a Royal Society Professorship. Xin Yang thanks Wuhu Feng and Martyn Chipperfield for support in using ERA-Interim data. Thanks also go to Stephen P. Palm for useful discussions.
This research has been supported by the Natural Environment Research (NERC) in the UK (grant nos. NE/J023051/1 and NE/J020303/1), the European Commission Horizon 2020 Marie Sklodowska-Curie Individual Fellowship (SEADOG (grant no. 658120)) and the Royal Society Professorship (grant nos. RP120096 and RP/R/180003).
This paper was edited by Christopher Hoyle and reviewed by two anonymous referees.