Introduction
In the last decades, Arctic temperatures have increased approximately twice
as fast as the global average temperature, e.g. due to temperature and
ice–albedo feedbacks , changes in the Atlantic
Ocean thermohaline circulation , and the decline in
European anthropogenic SO2 emissions since 1980 .
This temperature increase has been leading to reductions in both Arctic sea
ice extent and thickness for the last few decades: for the period from
November 1978 (start of satellite records) to December 2012, the Northern
Hemisphere sea ice extent decreased by 3.8±0.3 % per decade
. This decrease is more pronounced in summer and autumn than
in winter and spring . Since global and thus Arctic
temperatures will further increase in the near future, the Arctic is expected
to become ice-free in late summer within the next several decades
.
Sea ice concentration (SIC) refers to the percentage of an area which is
covered with sea ice. Ocean areas with high SIC have a larger surface albedo
and reduced exchanges of heat, momentum, and gases between the ocean and the
atmosphere than areas with low SIC . With an open Arctic
Ocean, natural aerosol emissions will increase because more sea salt
particles and more dimethyl sulfide (DMS; a precursor for sulfate aerosol
particles) will be emitted . Under present-day conditions,
emissions from the ocean are already an important aerosol source in some
Arctic regions in summer: measuring aerosol particles with radii between 0.25
and 10 µm in Svalbard (a map of the Arctic can be found in the
Appendix; see Fig. ), identified
sea spray particles as the main source for Arctic summer aerosol particles.
In a modelling study, found that sea ice retreat might
increase the sea salt aerosol number emissions in summer by a factor of 2
to 3 by 2100.
Presently, the contribution of Arctic shipping to aerosol radiative forcings
within the Arctic is very small compared to other emissions .
However, sea ice retreat might cause an increase in shipping aerosol
emissions over the Arctic Ocean, since reduced summer sea ice enables ships
to cross the Arctic Ocean. Cargo ships could shorten their paths
, tourism could be expanded
, and the Arctic oil and gas production will likely be
intensified . Compared to other regions, the present-day
Arctic air is exceptionally pristine, and aerosol levels are very low. Hence,
increases in both natural and anthropogenic aerosol emissions might have a
strong effect on cloud properties and radiation. Furthermore, deposition of
black carbon (BC) on snow and ice lowers the surface albedo
and therefore has the potential to accelerate sea
ice retreat .
Aerosol particles influence clouds, e.g. by acting as cloud condensation
nuclei (CCN) or ice-nucleating particles (INPs). Freezing processes involving
INPs are called heterogeneous freezing; for a recent overview on
heterogeneous freezing modes, see . The ability of an
aerosol particle to act either as a CCN or an INP depends on its size and its
chemical composition . Hence, both aerosol concentration and
composition influence cloud properties substantially : at a
constant liquid water content (LWC), an increase in the number concentration
of CCN changes the cloud droplet number concentration (CDNC); it leads to
more but smaller droplets, which increases the total surface area of the
cloud. Since cloud droplets must reach a certain size before they form rain,
this process may delay the formation of precipitation .
However, an increase in aerosol concentrations could also lead to
enhanced precipitation due to the presence of INPs, which reduce the required
supercooling and/or supersaturation for ice initiation. An earlier freezing
of some cloud droplets, followed by the Wegener–Bergeron–Findeisen process,
may rapidly form cold precipitation . Aerosol–cloud
interactions can affect cloud properties and the onset and/or intensity of
precipitation further, as described in
for example. In Arctic mixed-phase clouds,
observations suggest that the number of precipitating ice particles decreases
by 1–2 orders of magnitude under polluted conditions when aerosol
concentrations are high .
However, clouds are not only affected by aerosol particles. Increasing
atmospheric temperature is expected to shift the melting and the freezing
levels – and thus also cloud ice – to higher altitudes. Additionally,
higher temperatures will increase evaporation from the surface and,
consequently, the available water vapour in the atmosphere. An open ocean
further amplifies the increase in water vapour. Analysing satellite data from
2000 to 2010, found a negative correlation between sea ice
extent and cloud cover over the Arctic Ocean, which was statistically
significant and especially pronounced between July and November. Recently,
showed with a coupled atmosphere–ocean model that enhanced
heat and moisture fluxes resulting from the reduction in sea ice cover are
indeed responsible for the simulated increases in cloud cover.
Both aerosol particles and clouds impact the Earth's radiation budget.
Whether an aerosol particle predominantly absorbs or scatters radiation
depends on its physical and chemical characteristics. Aerosol scattering of
shortwave (SW) radiation tends to cool the atmosphere, whereas absorption of
SW and longwave (LW) radiation tend to warm it . The sum of
scattering and absorption is called extinction. Since the aerosol extinction
(normalised by the aerosol mass) is generally largest when the size of the
particle is comparable to the size of the wavelength, the SW effect is more
important than the LW effect for the majority of atmospheric particles
. However, for large particles such as dust or sea salt, LW
effects can become relevant .
Similar to aerosol particles, clouds impact the Earth's radiation budget by
absorption of LW radiation (warming) and scattering of SW radiation
(cooling). To a smaller extent, LW radiation is also scattered and SW
radiation absorbed . The absorption and emission
of LW radiation is a function of the emissivity of the cloud (which depends
on microphysical cloud properties and the water path), the (height-dependent)
cloud temperature, and the surface temperature
. The scattering of SW
radiation is a function of the microphysical cloud properties, of the cloud
water path, of the solar zenith angle, and of the surface albedo
. Since aerosol particles
influence cloud microphysics, they also impact cloud radiative effects
(CREs). With a higher CCN concentration at constant LWC, more radiation is
scattered back to space and the cooling effect of clouds is enhanced. This is
the so-called “Twomey effect” , also referred
to as radiative forcing due to aerosol–cloud interactions
RFaci;. Furthermore, changes in cloud
lifetime e.g. delayed precipitation; “Albrecht
effect”; also affect the CREs. Together with
RFaci, these adjustments are referred to as the
effective radiative forcing due to aerosol–cloud interactions
ERFaci;.
Compared with the global mean, the SW radiative effect of Arctic clouds is
less important because of the large solar zenith angle and the high surface
albedo . Therefore, the LW absorption of clouds
becomes more important and can dominate the total CRE depending on the
specific time and location. Arctic clouds warm the planet in the annual
average and show a net cooling effect only in summer
.
How Arctic clouds and their radiative effects will change in the future is
still an open question. Generally, both the SW and the LW CRE are expected to
become stronger when more CCN are available . However,
compared to other temperature feedbacks, the contribution of changes in
Arctic clouds might be small .
suggested that the overall effect of enhanced aerosol concentrations is to
increase the net warming effect of Arctic clouds because LW radiation
dominates in the long polar winter. In contrast, a modelling study of
found that the increase in anthropogenic aerosol
emissions since pre-industrial times has led to larger changes in the annual
Arctic SW (-0.85 W m-2) than in the LW (0.55 W m-2) CRE at
the surface. However, their simulated LW radiation effect was approximately
1 order of magnitude smaller than suggested by the observation-based study
of . Whereas considered
measurements from a specific location (near Utqiaġvik, Alaska) and analysed
strong pollution events, simulated the effect over
the whole Arctic (defined as north of 71∘ N in their study) under
all conditions. Other explanations for the different results include model
uncertainties, especially regarding cloud cover and thin cloud frequency
. For the Arctic summer, showed
that an increase in the number of aerosol particles can either decrease or
increase the net CRE depending on the background aerosol concentration.
Therefore, the future increase in both natural and anthropogenic aerosol
emissions due to sea ice decline is expected to influence radiation both
directly and indirectly. The following studies investigated the impact of
future changes in either natural or anthropogenic aerosol emissions:
using the global aerosol–climate model CAM-Oslo and
using the global aerosol microphysics model GLOMAP
analysed the influence of enhanced natural aerosol emissions on Arctic clouds
in the future; we will discuss their findings in the comparison with our
results. The impact of Arctic shipping on black carbon deposition on snow and
ice by 2050 was studied by , who found only a small
contribution of BC from ships. used the chemical climate
model OsloCTM2 to study the impact of enhanced global and Arctic shipping in
2030. In their high-growth scenario, O3 had the largest impact on
radiative forcing in autumn (August to October).
In this study, we aim to quantify changes in future Arctic aerosol particles
from both natural and anthropogenic sources enabled by sea ice reductions.
Furthermore, we analyse changes in clouds and radiation, which are partly
caused by these changes in aerosol emissions. We use the state-of-the-art
global aerosol–climate model ECHAM6-HAM2, which allows us to study changes in
Arctic aerosols and their impact on climate.
Figure provides a simplified overview of how the increase in
Arctic temperature can affect radiation. The most important interactions
between atmospheric variables, aerosols, clouds, and surface properties are
included. The figure shows that the increase in temperature directly affects
sea ice, specific humidity, and aerosols. Changes in these variables can then
directly or indirectly impact clouds and radiation.
The model and the simulations, the boundary conditions, the emissions, and
the statistical method used are described in Sect. . In
the results and discussion section (Sect. ), we focus on the
months July to October, when both the decrease in SIC and the increase in
shipping through the Arctic Ocean will be most pronounced. In the conclusions
(Sect. ), our key findings are summarised.
Simplified sketch showing how different variables (may) vary as a
result of enhanced Arctic temperatures. Red dashed arrows denote expected
increases, blue dashed–dotted arrows expected decreases. Black solid arrows show
which components impact radiation. CDNC and ICNC stand for cloud droplet and
ice crystal number concentration, respectively. Note that an increase in
aerosol concentrations can either increase or decrease precipitation and thus
the total water content, as mentioned in Sect. .
Methodology
ECHAM6-HAM2
General information about ECHAM6-HAM2
ECHAM6-HAM2 is the combination of the general circulation model ECHAM6
with the two-moment cloud microphysics scheme by
and the aerosol model HAM2 .
ECHAM6 solves prognostic equations for vorticity, divergence, surface
pressure, and surface temperature and uses a flux-form semi-Lagrangian
transport scheme to advect water vapour, cloud liquid water, cloud ice, and
trace components.
In HAM2, the aerosol components SO4 (sulfate), BC, organic carbon
(OC), sea salt, and mineral dust are considered . The size
distribution of the aerosol particles is described by four size ranges: the
nucleation mode (rm<5 nm; rm is the mode radius
of the aerosol particles), the Aitken mode (5 nm <rm<50 nm),
the accumulation mode (50 nm <rm<500 nm), and the coarse
mode (rm>500 nm). Only a soluble mode exists for the
nucleation mode, whereas a soluble or internally mixed mode and an insoluble mode
exist for the other three size modes. Therefore, seven aerosol modes are
considered in total, each described by a log-normal size distribution.
Coagulation and condensation can shift aerosol particles to larger modes
and/or from insoluble to internally mixed modes. Removal processes of aerosol
particles in ECHAM6-HAM2 comprise wet deposition, dry deposition, and
sedimentation. To link the simulated aerosol population with the CDNC and the
ice crystal number concentration (ICNC), parameterisations for cloud droplet
activation and ice nucleation are implemented
.
Regarding the sulfur chemistry, DMS is oxidised to SO2 (sulfur
dioxide), which can form sulfuric acid in the aqueous phase or in the gas
phase. Gas-phase sulfuric acid in the atmosphere can either nucleate, i.e.
form new small, soluble particles, or condense onto pre-existing aerosol
particles. Condensation can be limited by the available surface area of
aerosol particles, by the available gas-phase sulfuric acid, or by the
diffusion of the gas-phase sulfuric acid to the particle surface. If any
gas-phase sulfuric acid is left after condensation, the sulfuric acid
nucleates and forms new sulfate particles. Besides the available
concentration of sulfuric acid, nucleation depends on temperature and
relative humidity.
In the standard ECHAM6-HAM2 set-up, a minimum CDNC of 40 cm-3 is
implemented. This ensures that the global CDNC is not unrealistically low due
to missing aerosol species in the model such as nitrate or due to the
simplistic model description of organics (no explicit treatment of secondary
organic aerosols; neglect of marine organics). Without a lower threshold for
CDNC, the model might also underestimate the CDNC in the Arctic, where
organic aerosol particles are emitted from the sea surface microlayer
. However, since
the Arctic is a remote environment with low aerosol concentrations,
observations show that the value of 40 cm-3 is often undershot in this
region: between 15 July and 23 September, measured daily
median CCN concentrations between 15 and 50 cm-3 at a supersaturation
of 0.25 %. In July 2014, found a median CDNC of
10 cm-3 for low-altitude clouds (cloud top below 200 m) and of
101 cm-3 at higher altitudes. In October 2004,
conducted aircraft measurements in single-layer stratus clouds and found
averaged cloud droplet number concentrations of 43.6±30.5 cm-3.
Applying the standard CDNC threshold of 40 cm-3 would drastically
reduce the influence of changes in the CCN concentration and therefore impede
aerosol–cloud interactions. Thus, we decided to use 10 cm-3 as a lower
threshold for the CDNC everywhere and re-tuned this new model version. The
studies by and indicate that values
even below this lower threshold can occur. While these measurements are
representative for a specific point, our model represents average values over
a larger area (1.875∘×1.875∘), which should be less
variable than a point measurement. Nevertheless, we acknowledge that the
threshold of 10 cm-3 could still be too high under certain conditions.
In the Arctic, this threshold hit 11 % (weighted with liquid water content)
when averaged from July to October under present-day conditions. Without a lower threshold for CDNC, the
underestimation of the Arctic aerosol concentrations in our
model (see also Sect. 3) would
locally lead to unrealistically high precipitation formation rates and a
too-strong effect of increased aerosol emissions on clouds.
Aerosol emissions
Emissions of sea salt, dust, and oceanic DMS are calculated online and depend
on the 10 m horizontal wind speed (u10). Marine organic aerosol
emissions are not considered in this study. Sea salt emissions follow
with sea surface temperature (SST) corrections according to
. The correction is applied because SST affects sea salt
emissions by influencing bubble rising velocities, the gas exchange between
the bubbles and the water, the bubble bursting behaviour, and maybe also the
coverage of oceanic whitecaps . Dust emissions are
calculated as stated in , with some modifications based on
. The monthly mean DMS seawater concentrations are
prescribed according to , and the flux from the
ocean to the atmosphere is calculated following .
Changes in oceanic DMS concentrations are not straightforward to project:
taking primary production or SST as a proxy seems unjustified since Arctic
oceanic DMS concentrations also depend on taxonomic differences in
phytoplanktonic assemblages . Using a coupled
ocean–atmosphere model (with ECHAM5-HAM as atmospheric component), the study
by explicitly simulates DMS but only reports changes
between the time periods 2061–2090 and 1861–1890, which are not directly
comparable to the time periods we are interested in. Thus, we decided to
leave the oceanic DMS concentrations unchanged.
Besides dust, sea salt, and oceanic DMS, the emissions of all other aerosol
components or sulfate precursors are prescribed, mainly from the ACCMIP
emission inventory . For ship emissions, we used the
inventories by and , which are
described in the next paragraphs. Ship emissions are put into the second
lowest model layer (≈150 m). While OC and BC particles from ships
are exclusively emitted into the insoluble Aitken mode, the sulfate mass is
equally distributed between the accumulation and the coarse modes. It is
assumed that 2.5 % of SO2 from ships is emitted as primary
sulfate .
Our ship emissions are based on the inventories by and
, which include the species SO2, BC, and OC. The
shipping emissions for the year 2004 follow , who
combined the observational data sets COADS (Comprehensive Ocean-Atmosphere
Data Set) and AMVER (http://www.amver.com/, last access: 19 July 2018) considering ships above 100 gross tons. For the
global ship emissions in the year 2050, we use the ship
emission inventory and apply the same reduction in emission factors for 2050
as in the study by (80 % for SO2 and 20 % for
OC), which are based on the Amendments to MARPOL Annex VI adopted by the
International Maritime Organization in 2007.
For additional ship emissions in the Arctic in 2050, we take the ship
emissions by . They used the 2004 inventory by
as a “background” for calculating future Arctic ship
emissions in the year 2050 for transit shipping and for shipping that is
related to oil and gas production. Changes in ship emissions from the sectors
tourism, fishery, and local or national transport are not considered. For the
year 2004, no transit shipping was assumed, and the oil and gas shipping was
estimated based on oil tankers operating in the Arctic region. The expected
increase in these two sectors is related to SIC: less sea ice will facilitate
the passage through the Arctic ocean and expose new areas to oil and gas
production. assumed that emission factors of SO2
and OC will decrease due to regulations and improved technology but that
everything else (other aerosol emission factors; shipping
routes outside the Arctic) will remain constant.
We increased Arctic ship emissions by a factor of 10 to detect a significant
signal in aerosol particles. This is in agreement with the results of
, who studied the effect of ship emissions on tropical warm
clouds with ECHAM5-HAM. In the following, we show how realistic these 10-fold
emissions are in the context of other studies and recent findings.
Compared with other estimates of future Arctic transit shipping, the results
from lie between those from and
: the fuel consumption by is 1.4 to 2.4
times smaller than the values reported by . Depending on
the scenario, the estimated CO2 emissions by are
2 to 4.6 times higher in 2050 than the values reported by .
Recently, pointed out that both global and Arctic
ship emission inventories might underestimate BC ship emissions because
too-low BC emission factors were used. While the ship emission inventory by
used a BC emission factor of 0.35,
found – depending on the averaging method and the
area – factors between 0.79 and 0.92. These differences in BC emission
factors suggest that ≈2.5 times higher BC ship emissions might be
more appropriate for future transit and oil- and gas-related shipping than
the original estimate from . However, note that
also point out that small fishing vessels (<100
gross tonnage), which are not included in the analysis by ,
contribute substantially to ship emissions. Neglecting these emissions from
fishing activity likely leads to an underestimation of background ship
emissions. This is important because higher background emissions might lead
to a smaller impact of future transit and oil- and gas-related shipping (i.e.
smaller relative increase in total aerosol emissions). 10-fold ship emissions
(at least for BC) are achieved if we consider that
(i) transit shipping (which contributes most to the ship emissions by
over the pristine Arctic Ocean between July and October)
might be up to 4.6 times higher according to and (ii) the
BC emission factor used by is likely underestimated by a
factor of ≈2.5. Increasing the additional ship emissions (both
transit shipping and oil- and gas-related shipping) from
by a factor of 10 is an upper estimate and is probably too high to represent
conditions in 2050.
Heterogeneous freezing of mixed-phase clouds in ECHAM6-HAM2
In ECHAM6-HAM2, dust and BC particles (also those emitted by ships) can act
as INPs in the immersion mode when transferred to the internally mixed mode.
Heterogeneous freezing in ECHAM5-HAM is thoroughly described in the study of
. The only differences in ECHAM6-HAM2 are that (i) contact
freezing is limited to montmorillonite dust because contact freezing of BC is
controversial and that (ii) only particles in the accumulation and coarse
modes can induce freezing. The freezing rate is defined as the number of
cloud droplets that freeze per time and volume of air. Among other factors
such as temperature, the contact freezing rate depends on the volume-mean
droplet radius as well as the CDNC, while the immersion freezing rate depends
on the cloud water mixing ratio.
Calculation of aerosol radiative forcings and cloud radiative effects
Both aerosol radiative forcings and CREs are calculated online by calling the
radiation scheme once with and once without considering aerosol particles or
clouds; the difference between the two radiation calls is called radiative
forcing due to aerosol–radiation interactions (RFari)
for aerosols and CRE for clouds. While RFari is normally
used for the forcing by anthropogenic emissions being the only external
forcing to the system, a double radiation call with zero aerosols as the
reference provides the sum of the natural and anthropogenic radiative
forcing. For SW radiation, aerosol radiative forcings and CREs both depend on
the surface albedo. For example, an aerosol particle that scatters SW
radiation can either have a cooling or a warming effect depending on whether
the underlying surface has a lower or a higher surface albedo, respectively.
Since the surface albedo decreases in our future simulations due to melting
of sea ice, changes in RFari and CRE can either be
caused by changes in aerosol or cloud properties or changes in surface albedo.
For clouds, we can distinguish the two causes by applying the cloud radiative
kernel method described in the study of , which is
independent of changes in surface albedo. With this method, we can
furthermore disentangle changes in LW CRE caused by changes in clouds from
those caused by surface temperature changes see also. A
higher surface temperature enhances the outgoing LW radiation from the
surface. Thus, more LW radiation can be absorbed by clouds and the LW CRE
increases. In addition, the cloud radiative kernel method allows for
diagnosis of how different cloud types low and free-tropospheric
clouds; and changes in different cloud properties (cloud
cover or amount, cloud optical thickness, and cloud top altitude) contribute to
the total changes in CREs. Note that with this method, grid boxes without
incoming radiation are set to missing values for both SW and LW CRE. While
this is not an issue for July, August, and September, most values between
85 and 90∘ N are missing in October. For the SW CRE, we set these
missing values to zero; for the LW CRE, September values instead of the mean
over September and October are shown for these grid boxes.
In our model, the reduction of snow albedo due to deposited BC is determined
through interpolations of a lookup table based on a single-layer application
of the SNICAR model . The BC concentration in the top
2 cm of snow is considered . The concentration depends
on the surface influx of snowfall as well as the influx of BC removed from
the atmosphere through dry deposition, wet deposition, and sedimentation.
Both BC scavenged by hydrometeors through in-cloud and
below-cloud wet deposition is assumed to reach the surface
within one time step (if hydrometeors do not evaporate in subsaturated regions
below clouds). Given that both the spatial and the temporal resolutions of
our model are low (1.875∘×1.875∘; 7.5 min), this
assumption seems justified. The concentration of BC in snow can be further
modified through scavenging by snowmelt and glacier runoff. Since the
scavenging ratios are low 0.2 for BC particles in the internally mixed
mode and 0.03 for those in the externally mixed mode;, the BC
concentration in snow increases after snowmelt. Lastly, while albedo
reductions of snow on land and on sea ice are considered, the impact of BC
deposition on bare sea ice is not. This is due to the different
characteristics of the sea ice albedo concerning its interaction with the
deposited BC, which would only lie on top of the ice instead of being
mixed in. However, as the spatial coverage of bare sea ice without any snow
cover is small in the model, the impact of omitting this darkening is
expected to be negligible.
Model simulations
A summary of the model simulations can be found in Table . ECHAM6-HAM2 is an atmosphere-only model, i.e.
SIC and SST need to be prescribed (see
Sect. ). To estimate the impact of future
Arctic warming and sea ice retreat on aerosol particles and clouds, we
conducted simulations under present-day (year 2004) and future (year 2050)
conditions. The following simulations were performed with a resolution of
T63L31 (corresponding to ≈1.875∘×1.875∘ with
31 vertical levels):
arctic_2004. Global greenhouse gas concentrations, SIC, SST, and prescribed
aerosol emissions (including ships) from the year 2004 are used.
arctic_2050_EM2004. The global greenhouse gas concentrations in the year 2050
follow IPCC's Representative Concentration Pathway RCP8.5 . To prescribe future
SIC and SST, we used results from an Earth System Model (ESM; see Sect. )
simulation. The same prescribed aerosol emissions are used as in 2004. Therefore, all anthropogenic aerosol
emissions between arctic_2004 and arctic_2050_EM2004 are identical.
arctic_2050. The same as arctic_2050_EM2004 but the prescribed aerosol emissions
are representative for 2050 (RCP8.5). The emission factors of SO2 and OC ship emissions are smaller
than in arctic_2050_EM2004 since regulations and technological improvements are taken into account.
Additional Arctic ship emissions are not accounted for.
arctic_2050_shipping. The same as arctic_2050 but with additional ship emissions in
the Arctic. These emissions are estimated from see Sect.
based on future transport and oil and gas extraction. Since these additional Arctic ship emissions induced no
significant changes in our test simulations (not shown), we increased the emissions by a factor of
10 (mass flux). By comparing arctic_2050 with arctic_2050_shipping, we can estimate
the impact of future Arctic ship emissions enabled by the smaller SIC.
Each simulation is run for 20 years with the same forcing for each year,
therefore yielding 20 ensemble members.
An overview of the different model simulations.
Model simulation
Greenhouse gas concentrations
SIC/SST
Ship emissions
Other anthropogenic aerosol emissions
arctic_2004
Year 2004
Year 2004 (AMIP)
Year 2004
Year 2004 (ACCMIP)
arctic_2050_EM2004
Year 2050 (RCP8.5)
Year 2050 (MPI-ESM RCP8.5)
Year 2004
Year 2004 (ACCMIP)
arctic_2050
Year 2050 (RCP8.5)
Year 2050 (MPI-ESM RCP8.5)
with emission factors for 2050
Year 2050 (ACCMIP RCP8.5)
arctic_2050_shipping
Year 2050 (RCP8.5)
Year 2050 (MPI-ESM RCP8.5)
with emission factors for 2050 and additional ship emissions by
Year 2050 (ACCMIP RCP8.5)
Boundary conditions
Both SIC and SST are prescribed in ECHAM6-HAM2. For future conditions, we
used model results from the Earth System Model MPI-ESM as input
simulation for the climate model intercomparison project phase 5
(CMIP5), RCP8.5;. We chose MPI-ESM because its atmospheric
component is ECHAM and the simulated future sea ice retreat is close to the
model median of CMIP5. An inconsistency in our simulations is that we did not
apply the SST and SIC mid-month correction to the MPI-ESM data as recommended
by , which is applied for the AMIP data that we used for
the year 2004 . Therefore, the seasonal variability in SIC
and SST is somewhat underestimated in 2050. However, compared to the large
differences in SIC and SST between 2004 and 2050, we do not expect this to
affect the main conclusions of our study.
As mentioned previously, future greenhouse gas emissions follow the RCP8.5
scenario, which shows a similar CO2 emission increase as the A2
scenario that assumed in their analysis. From 2004 to
2050, the global greenhouse gas volume mixing ratios change as follows: from
377 to 541 ppm for CO2, from 1.76 to 2.74 ppm for
CH4, from 319 to 367 ppb for N2O, from 256 to
107 ppt for CFC-11, and from 540 to 345 ppt for CFC-12 (CFCs are
chlorofluorocarbons). Also, most prescribed aerosol emissions (excluding DMS
terrestrial emissions, biogenic organic carbon emissions, and ship emissions)
follow RCP8.5, which decline in most industrial sectors from 2004 to 2050.
We refrained from averaging SIC and SST over several years (e.g. 2000–2010)
to avoid having spurious regions with intermediate SIC and SST. However, the
inter-annual variability in SIC is pronounced, and therefore we performed
test simulations using SIC and SST from (i) the years 2003 and 2004 from AMIP
and (ii) the first and the second ensemble members from the MPI-ESM CMIP5
simulation for the year 2050. Overall, the Arctic SIC in 2003 was somewhat
smaller than in 2004, and the SIC in the first ensemble member from MPI-ESM
was smaller than in the second ensemble member. We found that the basic
results and main conclusions do not depend on these differences in SIC but
looking at only 2 years or ensemble members for both
present-day and future might not be sufficient to confirm that all our
results are robust. In the following, we will always refer to the simulations
using SIC and SST from 2004 and future SIC and SST from the first ensemble
member of MPI-ESM.
To verify consistency between future shipping routes and sea ice extent, we
further compared the sea ice conditions used to calculate future ship
emissions with the sea ice conditions employed in our simulations
(Appendix ).
Statistical test
recently pointed out that the approach to accept
alternative hypotheses at any grid point where locally significant results
occur (which is commonly used in atmospheric sciences) leads to
overstatements of scientific results: with this so-called “naive stippling
approach”, a significance test is calculated for every grid point and all
grid boxes are stippled where the p value is smaller than 5 % (for a
significance level of α=0.05). This approach has two main limitations:
(1) assuming that the spatial correlation is zero, 5 % of the grid boxes
show on average stippling just by chance; (2) spatial autocorrelation – often large when analysing gridded
climate data – increases the false discovery rate (FDR) for the “naive
stippling approach”, i.e. the null hypothesis is often rejected although it
is true. As suggested by , we circumvent the problem by
controlling the FDR instead. The advantages of this approach are the
elimination of many spurious signals and the robustness concerning spatial
correlation. In this method, a threshold p value is calculated below which
the result is supposed to be signal, not noise. We assume that the spatial
correlation is moderate or large for the variables we are looking at.
Therefore, we set αFDR to 2⋅α
seefor explanation. For the individual grid points,
p values are calculated using the Wilcoxon–Mann–Whitney test instead of
the often-used Welch's test since the latter is only valid if the samples are
normally distributed (a condition which was sometimes not confirmed by the D'Agostino–Pearson test). The only exception where
we used the Welch's test is for testing the significance of the results from
the cloud radiative kernel method (see Appendix ):
we could not apply the Wilcoxon–Mann–Whitney test to the cloud radiative
kernel results because they are given as differences instead of absolute
values. Throughout this paper, the term “significant” is interchangeable
with “statistically significant”.
Results and discussion
First, the changes in natural aerosol populations, clouds, and their
radiative forcings and effects in a warming Arctic are assessed
(Sect. ).
Second, we determine the influence of additional Arctic shipping activity
related to transit shipping and petroleum activities on climate
(Sect. ).
Most figures show the mean over the 20 ensemble members for the reference
simulation on the left and differences between the perturbed ensemble mean
and the reference ensemble mean on the right. As mentioned previously, we
analyse the months July to October. Since the conditions change considerably
from July to October, averaging over these 4 months might hide significant
changes occurring in only 1 or 2 months. Therefore, we decided to average
the results from July to August (late summer) and from September to October
(early autumn). If the season is not specified in the text, results refer to
both late summer and early autumn. Most of the figures show results for early
autumn, except for changes in clouds and RFari associated with
enhanced Arctic shipping, which refer to late summer. When we compare our
results to other studies, we average over the same time and area as the
authors of the corresponding study did for a meaningful comparison.
Each simulation consists of 20 ensemble members to account for the high
variability in Arctic climate. However, uncertainties associated with the
climate model used can of course not be captured with this approach. It is
well known that different global climate models deviate considerably, e.g.
when simulating aerosol–cloud interactions. Furthermore, models of different
resolutions generally have problems reproducing the structure of mixed-phase
clouds prevalent in the Arctic
, and the
future sea ice extent, as well as the prescribed aerosol emissions, is highly
uncertain . To gain a better understanding of the robustness
of our results, we compare them with other studies, both concerning relative
and absolute changes. In addition, we provide in the Supplement a comparison
of the simulation arctic_2004 with Arctic observations. While the
simulated ice water path (IWP) and the aerosol optical thickness (AOT; at
least in some Arctic regions) have a low bias, the surface concentrations of
BC and sulfate, the liquid water path (LWP), the cloud cover, and the SW, LW,
and net CREs at the surface and the TOA agree well with the observations. The
underestimation of AOT in our model is probably a combination of several
causes, including missing local aerosol sources in the model e.g. marine organics
or gas flaring emissions;, an
underestimation of aerosol transport from midlatitudes to the Arctic
, uncertainties in the optical properties and
emissions of aerosols e.g. for BC, see,
and the neglect of spume drops in the sea salt parameterisation by
. In general, it is very likely that our model underestimates
the total aerosol concentrations in the Arctic.
Changes due to warming and sea ice retreat
In the following, we analyse how a future temperature increase in the Arctic
affects natural aerosol particles, clouds, and radiation. For that,
simulation arctic_2050_EM2004 is compared with
arctic_2004. The Arctic sea ice area decreases from 6.1×106 to 3.4×106 km2 and from 5.7×106 to
2.3×106 km2 in late summer and early autumn, respectively.
To gain some insight into the importance of this retreat in sea ice, we
averaged some vertically integrated variables such as AOT or CDNC burden over
four different regions north of 60∘ N (see
Tables and for
late summer and early autumn, respectively): (i) the whole region north of
60∘ N; (ii) regions with open ocean in both 2004 and 2050
(SIC<0.5); (iii) regions with sea ice coverage in both 2004 and
2050 (SIC>0.5); and (iv) regions that are covered with sea ice in
2004 (SIC>0.5), but not anymore in 2050 (SIC<0.5). This
analysis is only qualitative since advection can hide significant changes
related to the sea ice retreat, the SIC values used for the calculations are
monthly means, and the threshold of 0.5 for SIC to differentiate open ocean
and sea ice is somewhat arbitrary.
Aerosol particles
Over the central Arctic Ocean, the decrease in SIC (Fig. )
enables emission fluxes of DMS and sea salt, which significantly increase
their burdens (Supplement Fig. S4; Tables ,
). As a second-order effect, significant
increases in u10 (Supplement Fig. S5) over the central Arctic Ocean in
early autumn increase sea salt and DMS emissions. In regions where the SIC
does not change, both changes in u10 (insignificant) and changes in SST
(Supplement Fig. S6) affect DMS and sea salt emissions, and thus their
burdens. For example, the decrease in the sea salt burden over the Bering
Strait is due to decreases in SST (caused by a model bias in the MPI-ESM sea
surface temperature compared to AMIP) and u10.
SIC in 2004 and 2050 for late summer (July–August) and for early
autumn (September–October).
Despite the pronounced increases in DMS burden, the sulfate burden shows no
large changes since it is dominated by other emissions (e.g. anthropogenic
SO2 emissions; not shown). Also the aerosol size distributions at
950 hPa (corresponding to ≈540 m;
Fig. a) and 800 hPa (corresponding to
≈1950 m; Fig. b) show only small,
non-significant changes from 2004 to 2050 (shown for early autumn; averaged
between 75 and 90∘ N). The number concentration slightly
increases in the nucleation mode in both seasons, which we attribute to the
enhanced DMS emissions. DMS is oxidised via SO2 to sulfuric acid,
which can form new particles. In late summer, the number concentration in the
Aitken mode increases to some extent. In early autumn, the number
concentration decreases at rap≈0.1 µm
(rap is the radius of the aerosol particles), which might be
caused by decreases in BC and OC burdens (not shown), but increases in the
coarse mode. The smaller BC and OC burdens can be explained by the increase
in precipitation, which leads to enhanced wet deposition (the BC and OC
emissions are identical between the two simulations). The increased number in
the coarse mode can be explained by the increase in sea salt emissions.
Aerosol number size distributions in 2004 (arctic_2004)
and 2050 (arctic_2050_EM2004); N stands for the number
concentration (assuming that air density ρair≈1 kg m3), rap for the radius of the aerosol
particles. The size distributions are shown for early autumn
(September–October) at 950 hPa (a) and 800 hPa (b),
averaged between 75 and 90∘N. The solid lines denote
ensemble means, the dotted lines the subtracted/added standard deviations.
Different colours (black, green) stand for different simulations (see
legend).
Absolute values for the year 2004 and differences between 2050 and
2004 (i.e. between simulations arctic_2050_EM2004 and
arctic_2004) for sea salt burden, DMS burden, AOT, LWP, IWP, cloud
cover (“CC”), in-cloud CDNC burden, and Tsurf for late summer
(July–August). The numbers are averaged over four regions between 60 and
90∘ N: (i) the whole region, (ii) grid boxes which are ocean in
both 2004 and 2050 (SIC<0.5; “Ocean”), (iii) grid boxes which are
covered by sea ice in both 2004 and 2050 (SIC>0.5; “Sea ice”),
and (iv) grid boxes which have sea ice in 2004 (SIC>0.5) but not in
2050 (SIC<0.5; “Transition”). Significant changes are marked
with a star. Note that the SST is prescribed, i.e. shows no inter-annual
variability.
Total region
Ocean
Sea ice
Transition
2004
2050–2004
2004
2050–2004
2004
2050–2004
2004
2050–2004
Sea salt (10-7 kg m-2)
1.2
0.18*
3.0
0.36*
0.18
0.12*
0.29
0.28*
DMS (10-7 kg m-2)
1.5
0.27*
3.2
0.39*
0.66
0.42*
0.92
0.73*
AOT (10-2)
3.6
0.26*
3.9
0.19
1.3
0.19*
1.6
0.19*
LWP (g m-2)
96
8.0*
108
7.3*
65
5.1*
67
7.9*
IWP (g m-2)
17
0.00
15
0.00
12
0.09
14
-0.06
CC (%)
77
0.08
81
0.85
88
-0.35
82
0.50
CDNC (1010 m-2)
6.0
0.47*
5.1
0.30*
1.9
0.22*
2.4
0.34*
Tsurf (K)
281
0.98*
278
1.6*
273
-0.36*
272
0.23*
As Table but for early autumn (September–October).
Total region
Ocean
Sea ice
Transition
2004
2050–2004
2004
2050–2004
2004
2050–2004
2004
2050–2004
Sea salt (10-7 kg m-2)
2.7
0.84*
6.6
2.0*
0.29
0.27*
0.45
0.53*
DMS (10-7 kg m-2)
0.62
0.12*
1.1
0.18*
0.32
0.24*
0.47
0.34*
AOT (10-2)
3.2
0.28*
3.5
0.32
1.3
0.03
1.4
0.21*
LWP (g m-2)
72
5.3*
92
2.0
24
14*
37
19*
IWP (g m-2)
21
0.59*
21
0.79*
12
0.17
14
0.76
CC (%)
87
0.05
89
-0.70*
92
1.3*
92
2.3*
CDNC (1010 m-2)
4.0
0.31*
4.3
0.30
0.96
0.28*
1.3
0.47*
Tsurf (K)
271
2.8*
277
1.8*
258
7.9*
264
7.4*
compared sea salt emissions for a nearly ice-free
summer (2100) with present-day conditions (2000) and found an increase in
mass emissions by a factor of ≈4 (present-day value
7.1 µg m-2 s-1); this is an average over JJA (June,
July, August) and 70 to 90∘ N. Note that we chose 2050 for our
simulations due to the availability of Arctic ship emissions for this year.
In the same region, found that sea salt emissions
increased by a factor of 10 (present-day value 6.9×10-3 µg m-2 s-1) in August when comparing a
hypothetically ice-free ocean with present-day conditions (2000). In our
simulations (70 to 90∘ N), sea salt emissions increase by a
factor of 1.8 and 1.7 in JJA and August by 2050, respectively, compared to
the present-day values of 1.52×10-3 and 2.42×10-3 µg m-2 s-1. The relative increase in emissions
is largest in the study by , where the absolute decrease in
SIC is largest, and is smallest in our study, where the absolute decrease in SIC
is smallest. Present-day emissions are a factor of ≈3 lower in our
simulations compared with , which results from the
differences in the two parameterisations with SST
corrections as shown in the study of .
The absolute present-day emissions reported by are at
least 3 orders of magnitude higher than in our simulations. This might
again be caused by the parameterisations used since differences in u10
and SST are too small to explain the large disagreement.
used a modification of the Mårtensson
parameterisation combined with the Monahan parameterisation for particles
>1.4 µm . However, neither
using the Mårtensson parameterisation nor us using the
Monahan parameterisation for particles rdry<4 µm (in
earlier simulations with ECHAM-HAM; not shown) found emissions as high as
. Therefore, we expect that differences in the number
fluxes of large particles (>4 µm), which contribute the most to mass
emissions , are responsible for the large discrepancy. When
we compare our simulated mass emissions in the Arctic (60 to
90∘ N) from July to October with the 11 CMIP5 models that provide
sea salt mass emission fluxes, our model shows the lowest sea salt emissions:
we arrive at a value of 5.9×10-3 µg m-2 s-1
under present-day conditions, while the CMIP5 models emit ≈4×10-2 µg m-2 s-1 (GISS-E2-H, GISS-E2-R, MIROC-ESM,
MIROC-ESM-CHEM), 7 to 9×10-2 µg m-2 s-1
(MIROC5, NorESM1-M, NorESM-ME), and 1 to 2×10-1 µg m-2 s-1 (GFDL-CM3, MIROC4h, MRI-CGCM3,
MRI-ESM1). Our simulated absolute increases in sea salt mass emissions are
therefore likely underestimated because our parameterisation does not account
for the contributions from spume drops and thus results in
small emission fluxes of large (i.e. supermicron) aerosol particles.
However, these large aerosol particles have a comparatively low impact on
climate due to their low number concentrations. Since the total sea salt
number emissions of the parameterisation by are not
generally lower than in other parameterisations, we do not expect that our
simulated impact on CCN and radiation is completely different compared with
other sea salt parameterisations. To confirm this, we conducted an additional
simulation similar to arctic_2004, but with the old standard sea
salt parameterisation of ECHAM-HAM i.e. the parameterisation
by. This parameterisation results in considerably higher sea
salt mass emissions than the parameterisation by (9.8×10-2 µg m-2 s-1 averaged from 60 to
90∘ N and from July to October). Nevertheless, the resulting AOT and
CDNC are quite comparable: using the parameterisation by ,
the AOT is somewhat higher in the Arctic than with the parameterisation by
(0.039 compared to 0.034; averaged from July to
October), while the CDNC burden is slightly lower (4.8×1010 m-2 compared to 5.0×1010 m-2; in-cloud
values).
Clouds
Except for cloud cover, LWP, and IWP, the averages of cloud properties (such
as LWC or CDNCs) refer to in-cloud values, i.e. by averaging only over
periods and locations when and where clouds are present.
In general, the number of aerosol particles acting as CCN increases in the
future, which leads to enhanced CDNCs (Fig. d). The
increase in the number of CCN is not only caused by the increases in oceanic
aerosol emissions but also by changes in meteorology: the updrafts available
for activation increase in the boundary layer between 75 and
90∘ N in early autumn (Supplement Fig. S7), which supports the
formation of cloud droplets in this region. Averaged between 75 and
90∘ N, the CDNC burden increases by 10 % and 29 % in late
summer and early autumn, respectively. Relative changes are largest in
regions where sea ice melted (Tables ,
). Also LWC increases (see
Fig. b) because both the open ocean and higher air
temperatures increase the specific humidity. The increase in LWC can be
ascribed to both higher CDNCs and larger cloud droplets (not shown). Averaged
between 75 and 90∘ N, LWP increases by 10 % in late summer
and by 34 % in early autumn. Precipitation shows significant increases in
early autumn (Supplement Fig. S8). In late summer, changes are only
significant when averaged between 60 and 90∘ N and smaller
than in early autumn (+4 % compared to +9 %).
We also obtain increased CDNCs (which we attribute to increased CCN
concentrations) when averaging over all sky conditions. In contrast,
found small decreases in CCN concentrations (also averaged
over all sky conditions) over the Arctic Ocean. In their simulations, the
liquid clouds over the ocean suppressed new particle formation via aqueous
phase oxidation of SO2 (a process also considered in ECHAM6-HAM2).
Instead, particles grew to larger sizes and were efficiently scavenged by
drizzle. The different responses when compared to our simulations could, for
example, be caused by different oxidant concentrations (H2O2, O3) or
by the different handling of drizzle and precipitation:
derived drizzle rates from Arctic observations of cloud altitude and droplet
concentrations and scaled them by the low-cloud fraction. However,
cloud microphysical processes (e.g. diffusional growth, coagulation) are
explicitly calculated in our simulations and coupled with aerosol particles
via Köhler theory and freezing parameterisations. Drizzle is not considered
as a separate size class in our simulations; however, showed
that the impact of drizzle on the CDNC burden is rather small in the Arctic
in ECHAM5-HAM.
As expected, the higher temperatures in 2050 influence the occurrence of
cloud ice (both cirrus and mixed-phase) in our simulations by shifting the
isotherms and thus also cloud ice towards higher altitudes. Changes in ice
water content (IWC) (Fig. b) can be caused by
changes in the ICNC (Fig. d) and/or the effective
ice crystal radius (Fig. f). Both changes in the
ICNC and radius have a considerable influence at altitudes below 500 hPa,
whereas changes in radius dominate at higher altitudes. The increase in ICNC
near the surface is mainly caused by enhanced convection, which leads to
small but numerous ice crystals following the temperature-dependent empirical
parameterisation of .
Compared to the pronounced increases in LWP, changes in the IWP are small and
only significant over the whole Arctic region and over the ocean in early
autumn (slight increases; see Tables ,
). This can be explained by two opposing
effects: on the one hand, the total water path increases due to the higher
specific humidity. On the other hand, the temperature increase leads to a
higher fraction of liquid water to the total water path. In our simulations,
the first effect slightly dominates in early autumn. The absolute changes
might be underestimated since our model in general underestimates the ice
water content of clouds.
LWC and CDNC in 2004 in (a) and
(c) and differences between 2050
and 2004 (i.e. between simulations arctic_2050_EM2004 and
arctic_2004) in (b) and (d) (in-cloud values) in
early autumn (September–October). Hatched areas are significant at the 95 %
confidence level. The dashed lines show the 0 and the -35 ∘C
isotherms.
Especially in early autumn, significant changes in cloud cover occur (see
Fig. ). Cloud cover decreases where convective
precipitation is most enhanced (e.g. near Svalbard; see Supplement Fig. S9)
but increases where sea ice vanished, e.g. over the East Siberian Sea and
the Beaufort Sea (Fig. shows a map of the Arctic Ocean
where the regional seas are labelled). When averaged over the open ocean
area, cloud cover shows rather small but significant decreases in early
autumn, whereas it increases significantly and pronouncedly where sea ice
melted (Table ). The latter is consistent with
the findings from , who found increases in the October cloud
cover caused by sea ice reduction, which leads to an enhanced moisture flux
to the atmosphere. Also in our simulations, the surface fluxes increase
significantly over regions where sea ice melted (not shown).
IWC, ICNC, and effective ice crystal radius in 2004 in (a),
(c), and (e) and
differences between 2050 and 2004 (i.e. between simulations
arctic_2050_EM2004 and arctic_2004) in (b),
(d), and (f) (in-cloud values) in early autumn
(September–October). Hatched areas are significant at the 95 % confidence
level. The dashed lines show the 0 and the -35 ∘C isotherms. Note
that they are zonally and temporally averaged, and hence ice can exist at
altitudes below the 0 ∘C isotherm.
(a) Cloud cover in 2004 and (b) differences
between 2050 and 2004 (i.e. between simulations
arctic_2050_EM2004 and arctic_2004) in early autumn
(September–October). Hatched areas are significant at the 95 % confidence
level.
Aerosol radiative forcings
Unless otherwise stated, all aerosol radiative
forcings and cloud radiative effects refer to those at the top of the
atmosphere (TOA). As mentioned previously, RFari refers
to the instantaneous effect of all aerosols on radiation. In 2004, aerosol
particles have a negative RFari and thus cool the Arctic
under clear-sky conditions (i.e. absence of clouds; see
Fig. c), except over sea ice and Greenland,
where the surface albedo is high (see
Fig. a). If the presence of clouds is
considered, aerosol particles also warm the atmosphere over Alaska and
northeastern Siberia (late summer) and over the whole of northern Russia (early
autumn; shown in Fig. e). Part of this
warming might be caused by BC and dust aerosols above clouds (Supplement
Fig. S10): the clouds reflect more SW radiation than the snow- and ice-free
surface and part of the scattered SW radiation can also be absorbed by
aerosol particles, causing an increase in aerosol absorption compared to
clear-sky conditions see e.g.. Moreover, the scattering
of aerosol particles could become less important in the presence of clouds,
which increases the relative importance of aerosol absorption to extinction.
Averaged over the whole Arctic region, aerosol particles have a cooling
effect under clear-sky conditions in 2004 (-1.23 W m-2 for late
summer and -0.65 W m-2 for early autumn) but a warming effect if
clouds are considered (0.12 W m-2 for late summer and
0.09 W m-2 for early autumn). Note that changes at the surface are
of opposite sign, i.e. the aerosol particles cool the surface under all sky
conditions. The simulated AOT has a low bias in the Arctic, which can affect
these estimates of the aerosol radiative effect. Depending on whether the
aerosol absorption or the scattering is underestimated, the aerosol radiative
effect is either under- or overestimated. It is also possible that both
effects cancel each other. In our simulations, both the cooling and the
warming are more pronounced in late summer than in early autumn due to the
higher solar zenith angle in late summer. Increases in the DMS and sea salt
burdens increase the AOT in 2050 (significant changes from 1.6×10-2
to 1.8×10-2 in late summer and from 1.5×10-2 to
1.7×10-2 in early autumn; averaged over 75–90∘ N).
While the AOT does not significantly change over open ocean, it significantly
increases over regions where sea ice melted
(Tables , ). The
absorption aerosol optical thickness significantly decreases in early autumn
(1.16×10-3 to 1.05×10-3, averaged over
75–90∘ N), which can be explained by the decrease in BC burden.
In both late summer and early autumn, RFari shows
significant decreases under both clear-sky
(Fig. d; shown for early autumn) and all sky
(Fig. f) conditions, especially in regions
where the surface albedo decreased (compare
Fig. b). We cannot distinguish between the
RFari induced by surface albedo changes and that induced
by changes in aerosols, but we expect that the increase in natural aerosol
emissions decreases RFari since sea salt and sulfate
particles are nearly pure scatterers.
The radiative forcing due to BC deposition on snow decreases significantly
(see Supplement Fig. S11) because less snow-covered sea ice and less snow on
land exist. However, the radiative forcing due to deposited BC as well as its
absolute changes are small compared to other radiative forcings and CREs.
This is also displayed in Tables and , which
show the area-averaged absolute differences in radiation, radiative forcings,
and radiative effects north of 60∘ N and north of 75∘ N,
respectively.
Surface albedo, aerosol net radiative forcing (clear-sky), and
aerosol net radiative forcing (all sky) in 2004 in (a),
(c), and (e) and differences
between 2050 and 2004 (i.e. between simulations
arctic_2050_EM2004 and arctic_2004) in (b),
(d), and (f) in early autumn (September–October). Hatched areas
are significant at the 95 % confidence level.
Cloud radiative effects
Not only the aerosol radiative forcing but also CREs change significantly.
Using the radiative kernel (RK) method, we first assess how CREs change only
as a function of cloud properties (i.e. independent of changes in surface
albedo or surface temperature). In this case, both the SW and the LW CRE (RK)
become stronger in late summer (Tables , ),
for example by -2.2 W m-2 for SW and +0.88 W m-2 for LW
when averaged between 75 and 90∘ N. In early autumn, changes in
CREs (RK) are significant when averaged over latitudes between 75 and
90∘ N (but not over the whole Arctic; see Tables ,
), where the SW and LW CREs (RK) change by -0.36 and
-0.96 W m-2, respectively. These decreases in the SW CRE (RK) north
of 75∘ N in early autumn (see also Fig. c) can be
attributed to increases in the cloud optical thickness and low cloud cover
(cloud top altitudes below 680 hPa; not shown). In contrast, the negative
changes in LW CRE (RK) north of 75∘ N (see also
Fig. f) are due to decreases in the free-tropospheric cloud
cover (cloud top altitudes above 680 hPa; not shown).
If we use the standard method for calculating CREs, which considers also
impacts due to changes in surface albedo and surface temperature, changes in
both SW and LW CRE are much more pronounced over the central Arctic Ocean in
early autumn than with the radiative kernel method (Fig. b,
e). Similarly to RFari, the large changes in SW CRE are mainly
caused by the smaller surface albedo (i.e. larger changes in
Fig. b than in c). In contrast, increases in LW CRE
primarily result from increases in surface temperature (Supplement Fig. S6).
The significant decrease in LW CRE over the Bering Sea (which only occurs in
Fig. e and not in f) can also be explained by changes in
surface temperature (a decrease in this case). Decreases in surface albedo
are highly correlated with increases in surface temperature over the Arctic
Ocean because the surface temperature of ice (which can be much lower than
270 K in early autumn, e.g. due to the ice–albedo feedback) changes to
the temperature of sea water (minimum temperature of 271.38 K).
Furthermore, changes in cloud cover and thickness affect both SW and LW CRE.
Changes in SW and LW CRE thus mostly occur at the same locations. Since they
are of opposite sign and on the same order of magnitude, they cancel each other out to a
large degree (Tables , ). While regionally
significant decreases and increases occur in the net CRE in early autumn, it
shows no significant changes when averaged between 60–75 and
90∘ N.
In late summer, the net CRE decreases significantly from 2004 to 2050 (by
-10 W m-2, averaged between 75 and 90∘ N), i.e. the
cooling effect of clouds increases, even though changes in surface albedo are
smaller than in early autumn (-0.12 compared to -0.21; averaged between
75 and 90∘ N). This is because (i) the SW component dominates in
these months due to the higher zonal zenith angle and (ii) the surface
temperature over the central Arctic Ocean does not show pronounced increases
like in early autumn (Table ), therefore not
enhancing the LW CRE. The surface temperature even decreases in some regions
because melt ponds on ice can have temperatures higher than 271.38 K (but
below 273.16 K) in late summer, while the SST is 271.38 K in grid boxes
with 0<SIC<1 (equilibrium conditions, i.e. heat changes lead to
changes in SIC, not SST).
Compared with the results by , our changes in the SW CRE
are rather small: averaged between 70 and 90∘ N (JJA), the
radiative effect increases from -63.7 to -107.7 W m-2 (i.e.
change by -44 W m-2) and from -47.1 to -55.4 W m-2 (i.e.
change by -8.3 W m-2) in their and our simulations, respectively.
The larger relative change reported by is likely caused
by the larger decrease in SIC (and, thus, albedo): while still considerable
parts of the Arctic Ocean are covered by sea ice in our simulations in 2050
(especially in June and July), only a small amount of sea ice is left in the simulations
by in 2100. For the present day, the absolute estimates of
SW CRE by the two models are similarly close to the satellite-derived value
by the Clouds and the Earth's Radiant Energy System (CERES), which is
-56.8 W m-2 averaged over the same months and latitudes for the
period July 2005 to June 2015 .
SW and LW CRE in 2004 in (a) and (d) and
differences between 2050 and 2004 (i.e. between simulations
arctic_2050_EM2004 and arctic_2004) in (b),
(c), (e), and (f) in early autumn (September–October).
In (b) and (e), the changes in CREs were calculated online
from two radiation calls (once with, once without clouds). In (c)
and (f), the changes in CREs were calculated with the radiative
kernel (RK) method (see text for more details). Hatched areas are significant
at the 95 % confidence level.
Absolute values for the year 2004 and differences between 2050 and
2004 (i.e. arctic_2050_EM2004–arctic_2004) in radiation,
radiative forcings, and CREs (in W m-2) averaged over all latitudes
north of 60∘N in late summer (July–August) and early
autumn (September–October). The arctic_2050_EM2004–arctic_2004
simulation accounts for changes between 2050 and 2004 associated with a warmer climate,
which leads to a reduction in SIC and therefore increased natural aerosol
emissions. RK stands for radiative kernel method (see text for details). The
star (*) denotes changes that are significant at α=5 %.
Late summer (2004)
Late summer (2050–2004)
Early autumn (2004)
Early autumn (2050–2004)
Net SW radiation
233
3.4*
67
0.70*
Net LW radiation
-231
0.60
-202
-1.7*
RFari
12×10-2
-9.5×10-2*
9.4×10-2
-3.4×10-2*
BC deposition
13×10-2
0.02×10-2
1.9×10-2
0.49×10-2
SW CRE
-67
-4.0*
-26
-0.45*
LW CRE
18
-0.04
21
0.55*
SW CRE (RK)
-67
-2.0*
-26
-0.00
LW CRE (RK)
18
0.92*
21
-0.07
The same as Table but averaged over all latitudes north of
75∘ N.
Late summer (2004)
Late summer (2050–2004)
Early autumn (2004)
Early autumn (2050–2004)
Net SW radiation
201
12*
29
2.5*
Net LW radiation
-228
0.77*
-196
-4.4*
RFari
53×10-2
-17×10-2*
15×10-2
-4.1×10-2*
BC deposition
21×10-2
0.32×10-2
2.0×10-2
0.99×10-2
SW CRE
-45
-10*
-7.8
-2.2*
LW CRE
9.3
-0.06
13
2.0*
SW CRE (RK)
-45
-2.2*
-7.8
-0.36*
LW CRE (RK)
9.3
0.88*
13
-0.96*
Impact of additional ship emissions
Future sea ice retreat will enable ships to cross the Arctic Ocean, thus
likely leading to enhanced shipping activity in late summer and early autumn.
In this section, we will study the influence of these anthropogenic aerosol
emissions on aerosol populations, clouds, and their radiative
forcings and effects by comparing the simulation arctic_2050_shipping
with arctic_2050.
Aerosol particles
Due to the increase in Arctic ship emissions (10-fold increase in the ship
emissions by in 2050), the burdens of BC and sulfate are
significantly enhanced in late summer (not shown). In early autumn, rises in
ship-related aerosol burdens are more pronounced (except for sulfate, which shows smaller, but still significant local changes) and also significant for OC
(not shown). The maximum increases
in aerosol burdens (see Fig. b) occur at the same
locations as the ship emissions, but significant increases can spread over a
large part of the Arctic (see Fig. c), as shown for
the example of BC. The largest absolute changes in BC concentration occur
near the surface, although significant changes reach altitudes as high as
400 hPa in early autumn (Supplement Fig. S12d). While the changes in natural
aerosol emissions (2050 versus 2004) only have a minor influence on the
number size distribution (Fig. ), the impact of
increased ship emissions is considerably larger.
Figure shows the aerosol number size
distributions averaged between 75 and 90∘ N, at both 950 hPa
(corresponding to ≈540 m; Fig. a)
and 800 hPa (corresponding to ≈1950 m;
Fig. b) for early autumn. At 950 hPa, the
number of particles in the nucleation mode largely decreases in both seasons
(Fig. a). For the Aitken mode, a small
decrease and a distinct increase occur in late summer (not shown) and early
autumn, respectively. The number concentration in the accumulation mode
increases to some extent in both late summer and early autumn. At 800 hPa
(Fig. b), the effect of ship emissions on
the aerosol size distribution is smaller than at 950 hPa.
The additional aerosol particles emitted by ships provide additional surfaces
for the condensation of gaseous sulfuric acid. Thus, the vertically
integrated condensation rate of sulfate increases where the ship emissions
occur (not significant; Supplement Fig. S13b). The vertically integrated
nucleation rate of sulfate shows neither a clear decrease nor a clear
increase along the shipping paths (Supplement Fig. S13d); if the increase in
condensation suppressed nucleation (as
Fig. a suggests), we would expect a decrease
in the nucleation rate. However, the vertical cross section of aerosol
particles in the nucleation mode shows that the number concentration indeed
decreases significantly near the surface (Supplement Fig. S13f).
The number concentrations in the accumulation mode (and the Aitken mode in
early autumn) increase both by direct emissions and by shifting aerosol
particles to larger sizes due to coagulation and condensation. Since ship
emissions occur near the surface, the influence at 800 hPa is much smaller
than at 950 hPa.
Panel (a) shows the BC burden in 2050 without considering
enhanced Arctic ship emissions. Panel (b) shows the difference
between a simulation with additional Arctic ship emissions and a simulation
without these emissions in 2050 (difference between
arctic_2050_shipping and arctic_2050). Hatched areas are
significant at the 95 % confidence level. Panel (c) shows ship emissions of BC (10-fold higher transit- and petroleum-related emissions) in
2050 based on the emission inventory
by . All values are for early autumn (September–October).
The impact of additional future ship emissions
(arctic_2050_shipping versus arctic_2050) on aerosol
number size distributions; N stands for the number concentration (assuming
that 1 kgair≈1 m3), rap for the radius
of the aerosol particles. The size distributions are shown for early autumn
(September–October) at 950 hPa (a) and 800 hPa (b),
averaged between 75 and 90∘ N. The solid lines denote ensemble
means, the dotted lines the subtracted/added standard deviations. Different
colours (black, green) stand for different simulations (see legend).
As Table but for arctic_2050 (absolute
values) and arctic_2050_shipping–arctic_2050 (differences)
averaged over all latitudes north of 60∘ N in late summer
(July–August) and early autumn (September–October).
The arctic_2050_shipping–arctic_2050 simulation considers the impact of an
increase in future Arctic ship emissions in 2050.
Late summer (2050)
Late summer (2050ship–2050)
Early autumn (2050)
Early autumn (2050ship–2050)
Net SW radiation
238
-3.0*
68
-0.46*
Net LW radiation
-231
-0.01
-204
0.32
RFari
11×10-2
0.79×10-2
4.1×10-2
1.1×10-2
BC deposition
12×10-2
-0.26×10-2
2.1×10-2
0.15×10-2
SW CRE
-69
-2.9*
-26
-0.46*
LW CRE
18
-0.04
21
0.35
SW CRE (RK)
-69
-3.4*
-26
-0.46*
LW CRE (RK)
18
0.20
21
0.26
The same as Table (impact of additional Arctic
shipping) averaged over all latitudes north of
75∘ N.
Late summer (2050)
Late summer (2050ship–2050)
Early autumn (2050)
Early autumn (2050ship–2050)
Net SW radiation
213
-3.9*
32
-0.45*
Net LW radiation
-227
-0.47
-200
-0.75
RFari
41×10-2
1.3×10-2
11×10-2
0.52×10-2
BC deposition
19×10-2
0.64×10-2
2.5×10-2
-0.02×10-2
SW CRE
-57
-3.7*
-9.9
-0.38*
LW CRE
9.1
-0.23
15
0.61
SW CRE (RK)
-57
-4.4*
-9.9
-0.35*
LW CRE (RK)
9.1
0.18
15
0.46
Clouds
Although ship emissions have overall a larger effect on aerosol burdens and size distributions in early autumn than in
late summer, significant aerosol-induced changes in clouds predominantly
occur in late summer. In the following, we will therefore only discuss
results for late summer. The CDNC increases (Fig. b; increase
in CDNC burden by 33 % averaged between 75 and
90∘N) and the effective radius decreases with additional
ship emissions (Fig. d), consistent with the
RFaci. Overall, the increase in CDNC dominates over the
decrease in cloud droplet radius, leading to an enhanced LWC
(Fig. f). We attribute this increase in LWC to a slower
collision–coalescence process (cloud adjustments).
Using satellite data, studied the effect of ship
tracks on both mixed-phase and liquid clouds. In the late summer of 2050, the
clouds that are impacted by ships in our simulations are mostly liquid.
Therefore, we restrict our comparison to the influence of ships on liquid
clouds. Consistent with the observations by , we also
found decreases in the effective radius and increases in cloud optical
thickness. The relative changes in effective radius are larger in their
observations (-20 % at cloud top altitude) than in our simulations
(-2 % to -4 % at altitudes below 500 hPa; averaged between 75 and
90∘ N), whereas changes in cloud optical thickness compare well
(+20 % in both studies, averaged between 75 and 90∘ N). The
LWP slightly decreases in their analysis (-1 %; in-cloud); in contrast,
it increases in our simulations (+17 %; all sky, averaged between 75
and 90∘ N). While our simulated precipitation shows no clear trend,
the results by suggest that ship emissions delay
precipitation by enhancing cloud lifetime. The different results could be
explained by the location of the ship tracks analysed by
: the majority of their samples lie between
45∘ S and 45∘ N, and only very few data points are from the
Arctic. Precipitation formation at high latitudes differs considerably from
that at low latitudes since, for example, convection is usually much more important
at low latitudes.
While liquid clouds are significantly impacted by ships in our simulations,
this is not the case for cloud ice, either in late summer or in early
autumn. Theoretically, ship emissions could influence heterogeneous freezing
in ECHAM6-HAM2 by several processes, for example,
the increase in BC emissions could lead to enhanced immersion freezing by
BC;
the increased SO2 emissions could shift some dust particles from the insoluble
to the internally mixed mode, which shifts contact freezing to immersion freezing, i.e.
to colder temperatures, as found by , for example;
decreases in the droplet radius would decrease the contact freezing
rate;
increases in the CDNC would increase the contact freezing rate.
The last two effects might partly cancel each other since a larger number
concentration of CCN is expected to simultaneously decrease the droplet
radius and increase the CDNC. However, the first two points also seem to be
irrelevant as ship emissions have no significant impact on cloud ice in our
simulations. To better understand why and gain some insights into the
importance of the different heterogeneous freezing processes, we calculated
the number of ice crystals that freeze via each of these processes
(Fig. a, c, e). Immersion freezing by dust is
the dominant freezing process in the Arctic in late summer
(Fig. c). However, contact freezing by dust is
more important near the surface since it can induce freezing at higher
temperatures than immersion freezing (Fig. a).
With additional ship emissions, the number of ice crystals formed by contact
freezing decreases near the surface and increases at higher altitudes
(Fig. b). Since the relative changes in CDNC are
larger than the relative changes in droplet radius (which would increase the
contact freezing rate), we suspect that contact freezing near the surface is
reduced by shifting more dust particles to the internally mixed modes. This
is consistent with the slight (non-significant) increase in immersion
freezing occurring near the surface (Fig. d).
Compared to dust, BC initiates freezing only in very few cloud droplets
(Fig. e) because its influence is mainly
restricted to high altitudes where temperatures are sufficiently low to
initiate freezing. However, BC particles from ships are emitted near the
surface. Therefore, the largest increases in BC concentrations also occur
near the surface (Supplement Fig. S12b). As a consequence, BC immersion
freezing is slightly enhanced near the surface
(Fig. f), but absolute changes are orders of
magnitude smaller than the decreases in contact freezing of dust. These
findings lead to the conclusions that (i) BC immersion freezing is largely
not affected because of the low altitude of ship emissions; (ii) even if it
were, it would hardly matter because dust is by far the dominant INP; and
(iii) SO2 emissions from ships lead to a slight shift from contact to
immersion freezing near the surface, thus rather leading to a non-significant
decrease in cloud ice at low altitudes.
Heterogeneous freezing is still an active field of research, and
contradictory evidence exists concerning the ability of combustion aerosols
to act as INPs . Laboratory results suggest that soot starts
initiating freezing at temperatures ≤-30 ∘C
Fig. 1-7. However, found an increase
in INP concentrations in ship tracks at higher temperatures. The increases
were small at temperatures around -20 ∘C, moderate at
-25 ∘C (≈+0.5 L-1; saturation ratio of 1.22), and
quite pronounced at -30 ∘C (≈+2 L-1; saturation
ratio of 1.32). The ship plumes were measured near the port of Gothenburg
(57.7∘ N, 11.8∘ E) in 2013 and 2014, and the meteorology in
general represented climate conditions of the late-autumn maritime north. If
ship exhaust (not necessarily the BC particles) can indeed induce freezing at
higher temperatures than in the laboratory-based BC parameterisation used in
our model, the impact on cloud ice could be larger than in our simulations,
especially in early autumn when temperatures are colder.
CDNC, effective cloud droplet radius, and LWC in late summer
(July–August; in-cloud values): (a), (c), and
(e) show the absolute values for
2050 (reference), (b), (d), and (f) the difference
between a simulation with enhanced ship emissions and the reference
simulation (difference between arctic_2050_shipping and
arctic_2050). Hatched areas are significant at the 95 %
confidence level. The dashed lines show the 0 and the -35 ∘C
isotherms.
Number of cloud droplets that freeze heterogeneously per time step
(Nfreez) in 2050: (a, b) contact freezing by dust,
(c, d) immersion freezing by dust, (e, f) immersion
freezing by black carbon in late summer (July–August). On the left side,
absolute values for 2050 (reference) are shown. On the right side, the
difference between a simulation with enhanced ship emissions and the
reference simulation is displayed (difference between
arctic_2050_shipping and arctic_2050). Note that the
scale is logarithmic and that the lowest bin had to be decreased to
10-10 to display statistically significant increases in immersion
freezing by BC. Hatched areas are significant at the 95 % confidence level.
The dashed lines show the 0 and the -35 ∘C isotherms.
Aerosol radiative forcings
The higher aerosol burdens due to ship emissions lead to enhanced AOTs
(significant increase from 1.4×10-2 to 2.0×10-2 in late
summer and insignificant increase from 1.4×10-2 to
1.5×10-2 in early autumn; averaged between 75 and 90∘ N).
Changes induced by additional ship emissions are on the same order of
magnitude as the changes caused by additional sea salt and DMS emissions from
2004 to 2050 (≈+0.2×10-2). In contrast to the changes in
aerosol absorption from 2004 to 2050 (no significant changes in late summer;
decrease in early autumn), ship emissions lead to pronounced and significant
increases in the aerosol absorption optical thickness (from 1.12×10-3 to 1.19×10-3 in late summer and from 0.83×10-3 to 1.00×10-3 in early autumn; averaged between 75 and
90∘ N). This is not surprising since OC and predominantly BC are
important absorbers of sunlight.
In late summer, the SW component clearly dominates changes in the net
RFari (e.g. +13 mW m-2 in SW compared to
+0.40 mW m-2 in LW under all sky conditions; averaged between 75 and
90∘ N). Under clear-sky conditions, the ship emissions induce a
pronounced cooling (i.e. RFari decreases; see
Fig. b). This cooling reverses to a
non-significant warming under all sky conditions
(Fig. d). Again, this shows that the scattering
of aerosol particles becomes less important when the scattering of clouds is
considered as well, and that the aerosol absorption can be enhanced in the
presence of clouds.
In early autumn, changes in the SW component still dominate changes in the
RFari in the region of shipping activity (e.g.
+8 mW m-2 in SW compared to +3 mW m-2 in LW under all sky
conditions; averaged between 75 and 90∘ N). Under clear-sky
conditions, the ship emissions lead to locally significant decreases in
RFari (see Supplement Fig. S14b). Under all sky
conditions, changes in net RFari are not significant
(Table ).
In early autumn, the BC deposition on snow leads to a small but significant
warming over part of the Arctic Ocean (see Fig. f).
Although these changes are pronounced in relative terms, they are more than
1 order of magnitude lower in absolute terms compared to the enhanced
cooling by clouds, which is discussed in the next section: averaged between
60 and 90∘ N, the radiative forcing of deposited BC
insignificantly increases by 1.5×10-3 W m-2 in early
autumn, while the SW CRE is significantly enhanced by -2.9 W m-2 in
late summer (Table ).
Based on the future Arctic ship emissions by ,
estimated how short-lived atmospheric pollutants might
change by 2030. Meteorology, sea ice extent, and emissions not related to
ships were not changed between 2004 and 2030 in their simulations. Therefore,
we compare our simulated changes which are only due to shipping (change from
arctic_2050 to arctic_2050_ship) with their results. In
their high-emission scenario, BC and OC annual ship emissions increase in the
Arctic by 2030 (BC by a factor of ≈5 and OC by a factor of
≈2), whereas SO2 emissions slightly decrease (by
≈4 %). In our simulations, annual Arctic BC, OC, and SO2
ship emissions increase by factors of 11, 10, and 7, respectively. Averaged
between 60 and 90∘ N and over August, September, and October,
find that the radiative forcing of aerosols
increases overall: +5 mW m-2 for sulfate, +5 to +6 mW m-2 for
BC, and nearly no changes for OC. The sum of these values is larger than the
value that we find (+5.7 mW m-2 averaged over the same time period
and area) although our increases in ship emissions are higher. It is possible
that the radiative forcing of all aerosols is more positive in the study by
because of the different SO2 emissions: in our
simulations, the SO2 emissions increase, which leads to cooling. In
contrast, the SO2 emissions in the study by
slightly decrease, which leads to a small positive forcing. Furthermore, the
effect of clouds on RFari might differ between the
simulations by and our simulations. The changes induced
by deposited BC are ≈1 mW m-2 in both the study by
and in our simulations. While the increase in BC
emissions is much larger in our simulations, less snow is available in 2050
compared to 2004.
Aerosol radiative forcing in late summer (July–August) 2050:
(a, b) under clear-sky and (c, d) under all sky conditions.
On the left side, absolute values for 2050 (reference) are shown. On the
right side, the difference between a simulation with enhanced ship emissions
and the reference simulation is displayed (difference between
arctic_2050_shipping and arctic_2050). Hatched areas are
significant at the 95 % confidence level.
The impact of additional ship emissions in the Arctic on
(b) in-cloud CDNC burden, (d) SW CRE, and (f)
radiative forcing due to BC deposition on snow. In (a),
(c), and (e), the reference without additional ship
emissions is shown (arctic_2050). Hatched areas are significant at
the 95 % confidence level. Panels (a) to (d) show results for late
summer (July–August), (e) and (f) for early autumn
(September–October). Note that the scale in (e) and (f) is
logarithmic.
Different contributions to the changes in SW (left) and LW (right)
CREs in late summer (July–August) caused by enhanced shipping: contribution
from changes in (a, b) cloud cover, (c, d) cloud top
altitude, and (e, f) cloud optical thickness. In (g) and
(h), the residual is shown. Hatched areas are significant at the
95 % confidence level.
Cloud radiative effects
In late summer, aerosol particles from ships lead to more but smaller cloud
droplets and an enhanced LWC (ERFaci), which increases
the reflection of solar radiation. Thus, we see an enhanced cooling effect of
clouds in most areas where the CDNC burden increases
(Fig. b, d), i.e. the SW CRE becomes significantly
more negative (≈-3.7 W m-2, averaged between 75 and
90∘ N). Changes in the LW CRE are smaller in terms of absolute
amount, not consistently spatially correlated with ship emissions, and not
significant (not shown). We additionally analysed the different contributions
to the changes in CREs from cloud cover, cloud top altitude, and cloud
thickness (see Fig. ). The residuals in
Fig. g and h show what can be attributed to neither
cloud cover, nor cloud top altitude, nor cloud thickness; it should ideally
be zero. While the changes in CRE caused by changes in cloud cover and cloud
top altitude are not significant (Fig. a–d), the
increase in cloud optical thickness leads to significant decreases and
increases in the SW and LW CRE, respectively (Fig. e,
f). Averaged between 75 and 90∘ N, the increased optical thickness
changes the SW CRE by -4.6 W m-2 and LW CRE by 0.52 W m-2
in late summer (significant). When we partition the contributions from low
and free-tropospheric clouds (defined as clouds with a cloud top altitude
below or above the altitude of 680 hPa), we find that 74 % of the changes
in SW cloud optical thickness occur in low clouds. This is not surprising
considering that the ship emissions occur near the surface.
studied the influence of model resolution on ship-induced
aerosol–cloud interactions and CREs (marine stratocumuli). They found that
the changes in SW CRE were overestimated by a factor of 2.6 with the coarser
model resolution (Δx=50 km, Δt=180 s) compared with the
higher model resolution (Δx=1 km, Δt=20 s). In case
this finding is generally applicable to numerical models, it could imply that
the SW CRE is also overestimated in our simulations.
In the study by , aerosol–cloud interactions lead to much
smaller changes in radiative forcing (-2 mW m-2; averaged between
60 and 90∘ N and over August, September, and October) than in our
simulations (-0.85 W m-2; averaged over the same period and space).
This is expected because our changes in future Arctic aerosol ship emissions
are considerably larger than in . Furthermore, it should
be noted that calculate RFaci using
an empirical relationship that estimates CDNC from aerosol concentrations. In
our case, CCN are calculated based on Köhler theory and we consider fast
adjustments, i.e. report ERFaci instead.
To summarise, ship emissions lead to a locally significant but very weak
positive radiative forcing over the central Arctic Ocean in early autumn
caused by absorption of deposited BC on snow. In contrast, the direct impact
of aerosol particles on the net radiation (RFari) is not
significant. The changes in CREs are significant and show that aerosol
particles enhance the cooling effect of clouds in late summer. When we
partition CRE into its different components, we find no significant radiative
changes induced by changing cloud top altitude or cloud cover, but the cloud
optical thickness increases and is responsible for the significant net
cooling. Since the cooling induced by aerosol–cloud interactions exceeds the
warming of deposited BC by at least 1 order of magnitude, ship emissions of
aerosols and their precursor gases overall induce a cooling in our
simulations.
Summary and conclusions
The main goal of this work was to analyse aerosol–cloud, aerosol–radiation,
and cloud–radiation interactions in a warming Arctic when sea ice extent
diminishes in late summer and early autumn. Simulations with ECHAM6-HAM2 were
conducted for the years 2004 and 2050. We also estimated the impact of
enhanced future Arctic shipping activity on climate.
Our results suggest that the future decrease in summer Arctic SIC will
significantly increase sea salt and DMS burdens in the Arctic due to enhanced
emissions. Both changes in aerosols and meteorology will lead to enhanced
CDNCs. Furthermore, not only the number concentration but also the size of
cloud droplets will generally increase because of higher specific humidities
leading to thicker clouds. In late summer, the net CRE at the TOA will become
more negative mainly because of the decrease in surface albedo associated
with melting of sea ice. Also, RFari will decrease in
late summer and early autumn mainly as a consequence of sea ice melting. The
decrease in both net CRE and RFari might delay Arctic
warming to some extent.
The simulated LWP, cloud cover, CREs, and surface concentrations of BC and
sulfate under present-day conditions compare well with Arctic observations.
However, our model has a low bias in AOT and cloud ice, which could impact
the simulated absolute changes in the radiative forcings and the CREs.
Missing aerosol sources such as nitrate the lack of an explicit treatment by SOA most likely contribute the simulated underestimation of AOT.
In future work, nitrate as well as a state-of-the-art SOA scheme will
therefore be incorporated into our model. Furthermore, inter-model
differences in sea salt emissions are large , and so are
the differences between our results and other modelling studies that
investigated changes in natural aerosols with declining sea ice. This
highlights that the results from this study – as from any climate model
study projecting the future – are uncertain.
Arctic ship emissions related to transport and oil and gas extraction have a
negligible impact on clouds and radiation in our simulations. Only when we
increase the ship emissions of by a factor of 10 is the
signal-to-noise ratio sufficiently large to detect ship-induced changes.
Considering that our model probably underestimates the background aerosol
concentrations in the Arctic, the simulated impact of the (10-fold) ship
emissions could be overestimated. With 10-fold ship emissions, the AOT
significantly increases by the same order of magnitude as natural AOT changes
from 2004 to 2050. RFari shows only minor, insignificant
changes in the presence of clouds, though. An increase in BC deposition on
snow leads to a very small local warming in early autumn. Meanwhile,
ERFaci induces a cooling in late summer. The magnitude
of changes in ERFaci are considerably larger than those
induced by the deposition of BC on snow, implying that ship emissions might
overall induce a cooling. In our simulations, only liquid clouds show
significant changes with increased ship emissions, while cloud ice is
unaffected. Considering the large uncertainty of heterogeneous freezing
processes, this result needs to be regarded with caution.
Compared to other changes (such as the decrease in surface albedo or the
increase in natural aerosol emissions), ship emissions of aerosols and their
precursor gases seem to have a small effect on climate considering that we
scaled the emissions up by 1 order of magnitude. However, even though this
study suggests that Arctic ship emissions of aerosols and their precursor
gases might have a negligible or slightly beneficial impact on climate, they
will also increase air pollution and might disturb local flora and fauna.
Furthermore, this study does not account for ship-induced changes in
greenhouse gases (e.g. O3), which are also important forcers
. More studies are required to confirm or refute
the findings of this work as well as to explore further ship-related
environmental impacts.