Introduction
The Arctic is one of the fastest warming regions on Earth
. Early studies have shown that 20th century Arctic warming
was mostly a consequence of the increased concentration of well-mixed
greenhouse gases (e.g., CO2 and CH4), associated with the
effect of shorter-lived climate forcers, especially aerosols and ozone
. As a result, mitigating Arctic warming requires first
and foremost global reductions of carbon emissions. However, controlling
short-lived species could be a faster and more cost-effective way to limit
Arctic and global warming, while also improving air quality
e.g.,, since aerosols and ozone are also harmful air
pollutants.
Global climate and chemistry-transport models are key tools used to
understand the past and future roles of short-lived pollutants. However,
modeling aerosol and ozone pollution in the Arctic has proven very
challenging in the past. Studies by , ,
and have shown that most models at the time strongly
underestimated black carbon (BC) observed at the Arctic surface, and
overestimated it aloft. In addition, models often failed to reproduce the
observed seasonal cycle of surface aerosol pollution, which peaks in late
winter and early spring due to enhanced transport from the midlatitudes and
lower deposition efficiencies . Studies have since shown
that these model biases were likely caused by the limited horizontal
resolution , missing local emission sources
, and poorly known removal processes. Specifically,
, , , and
showed that Arctic BC could be improved by the use of more complex aerosol
wet removal schemes within models. However, implementing these schemes does
not fully resolve model disagreement with BC measurements
, and recent research
indicates that differences in wet scavenging efficiencies
are still the main cause of differences in Arctic BC burdens between models.
Concerning ozone, , , and
showed that most models exhibit strong biases in ozone
precursors such as nitrogen oxides (NOx), carbon monoxide
(CO), peroxyacetyl nitrate (PAN) and several oxygenated volatile
organic compounds (VOCs), and underestimate ozone in the middle and high
Arctic troposphere by ∼ 10 to 30 %. Similarly, results
from the model intercomparison indicate that models are
strongly biased in the Arctic for both ozone and its precursors. These biases
are attributed to uncertainties in emissions, pollution transport and
processing, overestimated stratosphere–troposphere exchange and uncertainties
related to the hydroxyl radical OH.
The main known causes of model error in the Arctic (except emissions) can in
theory be addressed by using regional models, for which global coverage can
be traded for increased process complexity and higher resolutions. Several
recent case studies have shown the validity of this approach, by using the
regional WRF-Chem model (Weather Research and Forecasting model, including
chemistry; ) in order to understand the effect of
local pollutant emissions from shipping at high latitudes
and the mechanisms of pollution transport
from the midlatitudes to the Arctic
. However, these case
studies were based on short, relatively local simulations, while Arctic
pollution transported from the midlatitudes can only be studied using long,
quasi-hemispheric simulations, which can resolve both remote and local
sources of Arctic pollution. Such a quasi-hemispheric WRF-Chem simulation was
performed for the first time and evaluated in the intercomparisons of
and . Unfortunately, in spite of its
good performance for local case studies, WRF-Chem performed poorly in terms
of aerosols , failing to reproduce observed aerosol
concentrations and their seasonal evolution in spring and summer 2008.
showed that WRF-Chem performs reasonably well for ozone, but
other research indicates that the version of WRF-Chem
used in can be strongly biased low for ozone over
snow-covered ground due to overestimated dry deposition and underestimated
photolysis rates. In this context, the main objectives of this study are to
improve WRF-Chem results for Arctic aerosols and ozone compared to the
previous large-scale model intercomparisons of and
, to identify potential areas of further improvements in the
WRF-Chem model, and to define a model setup that can be used in future work
to study aerosol and ozone pollution on continental scales in the Arctic,
defined in this study as the region north of 60∘ N.
The model setup and emissions are presented in Sect. .
Section presents how the WRF-Chem 3.5.1 model was
updated for this study. The effect of these updates on Arctic aerosols and
ozone is evaluated in Sect. , where results are also
validated against surface and airborne measurements in the Arctic.
Conclusions are presented in Sect. .
WRF-Chem
WRF-Chem is a regional meteorological, chemistry,
and aerosol model based on the mesoscale meteorological model WRF-ARW
(Advanced Research WRF; ). WRF-Chem is fully
integrated within WRF, and uses the same grid, time step, advection scheme,
and physics schemes as WRF. The developments presented in this study
(presented in Sect. ) are based on the version 3.5.1 of
the model (the current version in March 2017 is 3.8.1 and only includes two of the updates presented here, Sect. ). The
version used here also includes the additions to WRF-Chem 3.5.1 related to
the KF-CuP cumulus scheme and described in .
Model setup, domain, and simulation period
WRF-Chem 3.5.1 setup.
Option name
Selected option
Chemistry and aerosol options
Gas-phase chemistry
SAPRC-99
Aerosols
MOSAIC 8 bins
+ VBS-2 SOA formation and aqueous chemistry
Photolysis
Fast-J
Meteorological options
Planetary boundary layer
MYJ
Surface layer
Monin–Obukhov Janjić Eta scheme
Land surface
Unified Noah land surface model
Microphysics
Morrison
SW radiation
RRTMG
LW radiation
RRTMG
Cumulus parameterization
KF-CuP
The model setup is presented in Table . Briefly, the
gas-phase chemistry mechanism is SAPRC-99 (Statewide Air Pollution Research
Center, 1999 version; ). Photolysis rates used in the
gas-phase chemistry calculations are calculated by the Fast-J scheme
. Aerosols are represented by the MOSAIC (Model for
Simulating Aerosol Interactions and Chemistry; ) model,
with eight size bins between 39 nm and 10 µm. The
version of the SAPRC-99–MOSAIC-8bin mechanism used here includes bulk aqueous
chemistry, as well as secondary organic aerosol (SOA) formation represented
by the VBS-2 (volatility basis set with two volatility species;
) scheme, treating the partitioning of organic
aerosols between the volatile and the condensed phase using the “volatility
basis set” approach . In this study, VBS-2 only includes
SOA formation from the oxidation of anthropogenic and biogenic VOCs. SOA
formation from semi-volatile and intermediate-volatility organic compounds
(SVOCs and IVOCs) was not included due to its high computational cost and due to the
lack of accurate global SVOC and IVOC emission inventories.
The MYJ (Mellor–Yamada–Janjić) scheme is used to represent the planetary
boundary layer, with the associated Janjić Eta surface layer scheme
. The land surface is represented using the Noah LSM (unified
Noah land surface model; ). Radiative calculations are
performed using the RRTMG scheme (Rapid Radiative Transfer Model for Global
applications; ), which is coupled here with WRF-Chem-predicted ozone and aerosol optical properties. The recommended microphysical
scheme to be used with MOSAIC is the Morrison two-moment scheme
. The Morrison two-moment scheme calculates cloud
formation, cloud properties, and precipitation on the grid scale, as well as
aerosol activation in clouds, aqueous chemistry for activated aerosols, and
wet removal. Subgrid clouds are represented using the KF-CuP (Kain–Fritsch
+ Cumulus Potential) parameterization developed by .
KF-CuP is a convective parameterization based on the Kain–Fritsch
cumulus scheme and the cumulus potential
scheme. The version of KF-CuP implemented in WRF-Chem
also represents the effect of cumulus clouds on aerosols and
trace gases (additional details are given in Sect. ).
Initial and boundary conditions for meteorology, as well as sea-surface
temperatures (SSTs) and sea ice, are specified using
NCEP FNL (National Center for Environmental Prediction, final analysis);
boundary conditions, SSTs, and sea ice are updated every 6 h. In
addition, WRF-Chem winds, temperature and humidity are nudged to the FNL
analysis in the free troposphere (grid nudging) with the same 6 h
update time. Initial and boundary conditions for chemistry are taken from the
global model MOZART-4 (Model for Ozone and Related chemical Tracers;
), and also updated every 6 h. The simulation
domain (polar stereographic projection) is presented in
Fig. . It includes remote sources of pollution potentially
transported to the Arctic in less than 30 days , a transport
time larger than the mean ozone and aerosol lifetimes in the troposphere
(22 days and less than 10 days, respectively). Simulations are performed for
the period from 1 March to 1 August 2008, in order to include both a period
with active long-range pollution transport to the Arctic (March to early May)
and a period when pollution removal processes are more prevalent (late May to
July). The month of March is discarded as spin-up. In order to be
computationally feasible, simulations are run at a relatively low horizontal
resolution of 100 km × 100 km, which is,
however, 2 to 3 times finer than the typical resolutions used by most global
models investigating Arctic aerosol and ozone . Simulations are performed for the year 2008, when many
measurement datasets are available as part of the POLARCAT (Polar Study using
Aircraft, Remote Sensing, Surface Measurements and Models, Climate,
Chemistry, Aerosols and Transport; ) project, and to allow
comparison with the WRF-Chem simulation presented in and
.
WRF-Chem simulation domain and location of the measurements used in
this study. Stations measuring ozone are shown as red circles. Arctic aerosol
measurement sites (BC and sulfate) are shown as blue squares. ARCTAS spring
and summer flight tracks north of 70∘ N (as in
) are shown in green and pink, respectively.
Emissions
Anthropogenic emissions are from the ECLIPSEv5 dataset (Evaluating the
Climate and Air Quality Impacts of Short-Lived Pollutants;
), except shipping emissions from RCP8.5 (Representative
Concentration Pathway 8.5; ). The ECLIPSEv5 inventory includes
BC emissions from gas flaring in the Russian Arctic, which have been shown to
improve the representation of Arctic BC by . Fire emissions
are from FINNv1.5 (Fire INventory from NCAR, Version 1.5;
). Soil NO emissions are from the POLMIP
(POLARCAT Model Intercomparison Project) inventory .
Biogenic emissions from vegetation are calculated online by the MEGAN model
(Model of Emissions of Gases and Aerosols from Nature;
). Other emissions calculated online by the WRF-Chem
model include sea salt, mineral dust (both from the GOCART model;
), and lightning NOx emissions .
Improvements included in WRF-Chem 3.5.1
We identify several processes previously missing from the WRF-Chem model
version 3.5.1 and potentially important for the representation of Arctic
aerosols and ozone. This section presents the model updates included and
evaluated in this study. (1) We include the effect of cumulus on aerosols and
trace gases as represented in the KF-CuP cumulus scheme within WRF-Chem
(Sect. ). KF-CuP is used here, but it was included in
WRF-Chem 3.5.1 in and and released in
later WRF-Chem versions; it is here further coupled to other components of
the model and its impacts on Arctic aerosols and ozone are evaluated. Updates
developed specifically for this study include (2) the addition of
sedimentation aloft in the MOSAIC aerosol model
(Sect. ), (3) the inclusion of DMS emissions
and gas-phase chemistry in the SAPRC-99 gas-phase mechanism
(Sect. ), (4) the coupling of WRF snow to the dry
deposition scheme (Sect. ), (5) the inclusion of a
dependence of UV albedo on snow and ice cover in the Fast-J photolysis scheme
(Sect. ), and (6) the added heat sink from melting
sea ice in calculations of the surface energy budget in the Noah-LSM surface
model (Sect. ). The updates presented in this
section, except the KF-CuP scheme and the corrections to the Noah-LSM module,
are not yet included in the latest version of WRF-Chem (3.8.1).
The different simulations performed to evaluate these updates are presented
in Table . ALL_UPDATES is the reference simulation
with all updates implemented, and NO_UPDATES a simulation where all updates
presented in this section are turned off. We also perform simulations where
each update is removed, leaving all of the others switched on (e.g.,
NO_SEDIMENTATION). The NO_KFCUP_CHEM simulation does not disable the
KF-CuP cumulus scheme entirely, but rather only its impacts on trace gases and
aerosols (aerosol activation, aqueous chemistry, tracer transport, wet
removal). Due to limited computational resources, the updates related to
deposition and photolysis over snow are only evaluated separately (i.e.,
NO_SNOWDEP and NO_SNOWPHOT) for the months of March and April, when snow
cover is highest, but are evaluated together (i.e., NO_SNOWDEP_SNOWPHOT) for
the full study period (March–July).
Simulation list and description.
Simulation name
Description
ALL_UPDATES
All model updates included
NO_KFCUP_CHEM
No effect of cumulus on aerosols and trace gases in the KF-CuP scheme
NO_SEDIMENTATION
No aerosol sedimentation above the first model level
NO_DMS
No dimethyl sulfide (DMS) emissions or gas-phase chemistry
NO_SNOWDEP
No reduced dry deposition of gases over snow (March–April only)
NO_SNOWPHOT
No increased UV albedo over snow and ice (March–April only)
NO_SNOWDEP_SNOWPHOT
Combination of NO_SNOWDEP and NO_SNOWPHOT (March–July)
NO_NOAH_SEAICE
No heat sink from melting sea ice in the Noah LSM
NO_UPDATES
All updates above turned off
This section presents these previously missing processes in more detail,
their relevance to Arctic short-lived pollutants, and how they were taken
into account in the WRF-Chem 3.5.1 model. The effect of these changes on
Arctic aerosols and ozone are evaluated and discussed in
Sect. .
KF-CuP cumulus scheme and its effects on aerosols and trace gases
Aerosol–cloud and trace gas–cloud interactions in the MOSAIC aerosol model,
including wet removal and aqueous chemistry, were previously only represented
in WRF-Chem for grid-scale (resolved) clouds, but not for cumulus
(parameterized) clouds. recently included the KF-CuP cumulus
scheme in WRF-Chem 3.5.1, and modified it to take into account the effect of
cumuli on aerosols and trace gases in the model. Specifically, the KF-CuP
scheme within WRF-Chem represents the impacts of warm cumulus clouds on trace
gas and aerosol vertical transport, activation and resuspension of aerosols,
aqueous chemistry in clouds, wet removal of aerosol and trace gases, and
impacts of aerosol activation on cloud droplet concentrations. Based on
simulations in June 2007 in the southern United States,
showed that using KF-CuP could decrease column-integrated BC by up to
50 %, due to changes in wet removal, and increase SO42-
by up to 40 % in nonprecipitating conditions, due to aqueous
chemistry in clouds. However, neither the long-term or large-scale effect of using
KF-CuP nor its effect on ozone has yet been investigated. These
processes are very relevant for the Arctic, where most of the pollution is
known to originate from long-range transport , and
wet removal is the main process controlling aerosol transport to the Arctic
.
The version of KF-CuP used in this study includes secondary activation of
aerosols above the cloud base, which was not included in ,
and primary activation at cloud base. For primary activation, the model
calculates the maximum supersaturation using the
parameterization, with the cloud-base updraft speeds from KF-CuP, and the
simulated aerosol concentrations in the updrafts. Secondary activation
assumes a fixed maximum supersaturation of 0.1 %. Aerosol
activation is then calculated from the maximum supersaturations and the
critical supersaturations for each aerosol size bin. In addition, KF-CuP is
coupled here to the RRTMG radiation scheme, by passing the KF-CuP cloud
fraction, cloud water, cloud ice, and cloud droplet numbers to RRTMG,
following the approach of . The lightning NOx
emissions scheme of , previously coupled in WRF-Chem to other
cumulus schemes, is also coupled here with KF-CuP, by linking KF-CuP cloud
top heights, cloud fractions, and deep or shallow convection flags to the emission scheme. In this study, we
only evaluate the effect of KF-CuP on aerosols and trace gases. These effects
are evaluated by disabling in KF-CuP the effect of cumuli on tracer
transport, aerosol activation, aqueous chemistry,
and wet removal (NO_KFCUP_CHEM simulation). The effect of lightning
NOx emissions or of the coupling between cumuli and radiation are
not evaluated separately here, since they were already studied with other
cumulus schemes in and .
Aerosol sedimentation aloft in the MOSAIC module
In MOSAIC, as it is included in WRF-Chem (and up to the current version 3.8.1
in March 2017), aerosol sedimentation is only implemented in the lowest model
level and only takes into account the contribution of sedimentation to dry
deposition, but not its role in bringing particles from higher altitudes to
the surface. This is discussed but not corrected in . This
could be an issue in longer, large-scale simulations, since this could lead
to a build-up of large particles (e.g., dust), for which sedimentation is one
of the main sinks . In this study, a first-order explicit
sedimentation scheme is implemented above the first vertical level in MOSAIC,
using the same algorithm for calculating settling velocities as the one
already in use for sedimentation at the model surface. The effects of this
change are evaluated by performing a simulation without sedimentation aloft,
called NO_SEDIMENTATION; results are discussed in
Sect. .
DMS emissions and gas-phase chemistry for SAPRC-99–MOSAIC
The SAPRC-99–MOSAIC mechanism does not originally include dimethyl sulfide
(DMS) gas-phase chemistry in WRF-Chem 3.5.1, even though DMS is known to be
an important source of SO2 and sulfate in the Arctic during summer
. Here, a simplified representation of SO2 chemical
production from DMS is implemented in SAPRC-99, following the work of
and . We also use the “online” DMS
emission scheme in WRF-Chem, based on and
, as it was implemented in . For this
study, this scheme is refined by using monthly resolved maps of oceanic DMS
from the climatology of instead of a single oceanic DMS
concentration value as in .
The effects of these updates are evaluated by performing a simulation without
DMS chemistry or emissions, called NO_DMS; impacts on aerosols and ozone are
discussed in Sect. , but we show here in
Fig. how this update changes surface DMS and SO2
in June–July 2008. The modeled amounts and geographical distribution are
similar to previous studies e.g.,. DMS concentrations
are especially elevated at higher latitudes due to the high oceanic DMS
concentrations. As a result, DMS is also a major source of SO2, the
main precursor for sulfate aerosols (Sect. ),
over the open Arctic ocean: away from Arctic shipping lanes, DMS emissions
and gas-phase chemistry are responsible for 90 to 100 % of surface
SO2 in this region.
June–July average (a) DMS surface mixing ratios and
(b) SO2 surface mixing ratios due to the implementation of
DMS emissions and gas-phase chemistry in the model (ALL_UPDATES –
NO_DMS).
Coupling dry deposition of trace gases with predicted snow
Dry deposition of trace gases is known to be lower
in winter and over snow, due to the reduced stomatal uptake of gases by
plants, and due to the enhanced atmospheric stability over snow, i.e.,
increased surface and aerodynamic resistance to deposition. Reduced
deposition over seasonal snow cover was already taken into account for the
MOZART gas-phase chemistry mechanism in WRF-Chem's deposition scheme
, but not for other mechanisms (e.g., SAPRC-99, CBM-Z,
RACM). For these other mechanisms, the model only took into account reduced
deposition over permanently snow-covered surfaces (e.g., mountain tops) or
over sea ice. As a result, showed that WRF-Chem (run with
RACM chemistry) could underestimate observed ozone by more than
5 ppbv in wintertime conditions in the western United States.
In this study, we also correct WRF-Chem's dry deposition scheme for the
SAPRC-99 mechanism, by forcing wintertime conditions in the dry deposition
scheme (“Winter, snow on ground and near freezing” seasonal category in
WRF-Chem; ) when predicted snow height is above
10 cm, the threshold already in use in WRF-Chem for the MOZART gas-phase chemistry mechanism. Over the snow-covered surfaces that were
previously treated as vegetation-covered, this update reduces ozone
deposition velocities by as much as -0.25 cms-1 during April,
as shown in Fig. .
Change in ozone deposition velocity due to the implementation of
wintertime dry deposition over seasonal snow (April 2008 average,
ALL_UPDATES – NO_SNOWDEP).
UV albedo over snow and ice in the Fast-J photolysis scheme
In their study of high wintertime ozone pollution events in the western US,
also identified that the photolysis schemes implemented
in WRF-Chem 3.5.1 were only using one single value for broadband UV albedo at
the surface, 0.055, even though this value should be much higher over snow
or ice (up to 0.85). In order to correct this, changed
the broadband UV albedo to 0.85 in their simulations, the value measured at
the site of their study. This value cannot be used as such here, since it
corresponds to conditions of very high snow cover over bare ground, which are
not representative of our whole simulation region.
Here, the UV albedo in the Fast-J photolysis scheme is
calculated as an average (weighted by snow and ice cover) of the snow-free
(or ice-free) albedo and the snow-covered (or ice-covered) albedo. This value
is updated at each call of the photolysis scheme. Land-use-dependent
UV-albedo values over snow are taken from the satellite-derived dataset
presented in and are retrieved from a look-up table
(Table 2 in ), based on the WRF-Chem land use category
in each grid cell. The resulting UV-albedo values are much higher than the
base value of 0.055, up to 0.85 over 100 % sea ice or bare
snow cover. As a result, photolysis rates predicted by the Fast-J scheme are
also greatly increased over snow- and ice-covered regions in April, by +50
to +200 % for jO1D and jNO2
(Fig. ). The combined effect on surface ozone of this
change and of reduced dry deposition over snow are validated and discussed in
Sect. and .
Change in (a) jO1D and
(b) jNO2 photolysis rates at the surface due to the
implementation of an UV-albedo dependence on snow and ice cover in the Fast-J
scheme (April 2008 average, ALL_UPDATES – NO_SNOWPHOT).
Heat sink from melting sea ice in the Noah land surface model
In WRF version 3.5.1, the Noah land surface model did not take into account
the heat sink due to sea-ice melt (latent heat of ice melt) in the energy
budgets at the prescribed sea-ice surface. As a result, the surface model
could predict unrealistically high surface temperatures during the ice-melt
season. We corrected this issue by simply prescribing the skin temperature of
sea ice to 0 K when the model diagnoses surface melt. We have
shared this update with the WRF community, and it was included in WRF-Chem
after version 3.7.1. Implementing this correction can decrease 2 m
temperatures over sea ice by as much as 10 K during the melt
season. This is of concern since the temperature contrast between snow and
sea-ice-covered and snow- and sea-ice-free areas is one of the main factors
determining the location of the Arctic dome ,
whose northward retreat during summer isolates the Arctic surface from
pollution transported from the midlatitudes. As a result, erroneously small
latitudinal temperature contrasts could greatly increase long-range pollution
transport to the Arctic surface during summer. However, the exact magnitude
of this effect on Arctic aerosols and ozone has not been evaluated until now
(this is discussed and validated in Sect. ).
Effect of the model updates on aerosol and ozone concentrations in the Arctic
This section presents the effect of individual model updates on modeled
aerosols (Sect. ) and ozone
(Sect. ) in the Arctic. The new, updated
version of the model is also validated against airborne
(Sect. ) and surface
(Sect. and )
measurements of aerosols and ozone in the Arctic in 2008. Simulation
performance is evaluated in terms of root mean square error (RMSE), defined
as
1n∑i=1n(xmod,i-xobs,i)2,
where xmod and xobs are the modeled and
observed mass concentrations or volume mixing ratios, respectively.
Aerosols
Effect on zonal mean aerosol concentrations in the Arctic
Change in the April–July 2008 average zonal mean PM10 due
to (a) KF-CuP cumulus effect on aerosols and trace gases,
(b) aerosol sedimentation aloft, (c) the sea-ice-melt heat
sink in the Noah LSM, and (d) DMS emissions and chemistry. Note the
differences in scale between top and bottom panels.
The effect of the KFCUP_CHEM, SEDIMENTATION, NOAH_SEAICE, and DMS updates
on zonal mean total aerosol mass concentrations (which are equivalent to
zonal mean PM10 in WRF-Chem with MOSAIC) is presented in
Fig. . The effect of the updated trace gas
deposition and photolysis over snow and ice (SNOWDEP, SNOWPHOT) on
PM10 (not shown) is very low, less than 1 %.
Figure shows that aerosol sedimentation aloft
(SEDIMENTATION) and cumulus effects on aerosols and trace gases (KFCUP_CHEM)
have the largest impact on aerosols in the Arctic, -30 % at higher
altitudes. Sedimentation aloft is both a sink (particles transported below)
and a source (particles transported from above) of particles at lower
altitudes, which explains why it has little effect below 3 km. The
net effect of KF-CuP is to decrease aerosol mass; this indicates that the
effect is dominated by increased wet removal, as in , and is
not compensated for by increased sulfate formation in the aqueous phase
(cloud chemistry) or by increased vertical aerosol precursor and aerosol
transport (tracer convection).
The implementation of the sea-ice-melt heat sink in the Noah LSM strongly reduces
PM10 at the Arctic surface (< -20 %), and increases
aerosol concentrations aloft. In these simulations, local sources of
pollution at the sea-ice surface are negligible; because of this, aerosol
concentrations there are mostly due to downward mixing of aerosols and gases
from the free troposphere. The NOAH_SEAICE updates reduce surface
temperatures over sea ice during summer, increasing stability, decreasing
vertical mixing, and thus reducing this tropospheric source (sink) of surface
(free troposphere) pollution. DMS emissions and chemistry increase
PM10 by +2 to +4 %, due to increased SO42-
aerosols formed from SO2 in the marine boundary layer. However,
relative increases of PM10 from DMS remain rather low because of
the relative lack of open water for DMS emissions north of 60∘ N,
and the high background PM10 in these areas due to colocated
emissions of sea salt aerosols.
Validation against BC profiles from the ARCTAS aircraft campaign
ARCTAS (a) spring and (b) summer median SP2 rBC
(size range 90–1000 nm) profiles north of latitude
70∘ N (gray shading indicates 25th and 75th
percentiles) and WRF-Chem median BC (size range 80–1000 nm)
profiles interpolated along the same ARCTAS flights (red, ALL_UPDATES; blue,
NO_UPDATES; error bars indicate 25th and 75th percentiles).
In order to validate the modeled aerosol distribution, we compare in
Fig. results from the ALL_UPDATES and NO_UPDATES
simulations to vertical profiles of refractory BC (rBC) measured by a SP2
(single-particle soot photometer) during the ARCTAS (Arctic Research of the
Composition of the Troposphere from Aircraft and Satellites) campaigns in
April and July 2008 . As in ,
this comparison only includes observations and model results north of
latitude 70∘ N. The updated model is in much better agreement with
observations than the original NO_UPDATES simulation, especially in the
summer, where RMSE decreases by 13.8 ngm-3 in ALL_UPDATES.
Table shows that the decreased model error is almost solely
due (-12.6 ngm-3) to the KFCUP_CHEM update. Other updates
have little effect, which is understandable since small BC-containing
particles have slow sedimentation velocities, are not directly affected by
DMS, and because the NOAH_SEAICE update has the largest effect at the
sea-ice surface, which was not sampled by the aircraft. The updates have
little effect on correlation coefficients, which rise from 0.07 to 0.08 in
spring, and decrease from 0.48 to 0.43 during summer, indicating that neither
the base model nor the updated version is able to reproduce well the
vertical variability of BC in the Arctic troposphere. In addition, the
updated model still overestimates observations in summer, which could be due
to overestimated emissions from, for example, biomass burning, or underestimated
removal. showed that increasing the horizontal resolution
from 100 to 40 km could reduce summertime BC simulated by
WRF-Chem by 25–30 %, by improving the representation of wet
removal.
RMSE of individual WRF-Chem simulations relative to Arctic
observations of aerosols and ozone. All sensitivity simulations are performed
by deactivating updates from the ALL_UPDATES simulation; as a result, any
increase in RMSE relative to ALL_UPDATES indicates that a given update
improved RMSE. For surface measurements, RMSEs are calculated at each station
and given as a network average.
Simulation name
ARCTAS spring BC
ARCTAS summer BC
Surface BC
Surface SO4
Surface O3
(ngm-3)
(ngm-3)
(ngm-3)
(ngm-3)
(ppbv)
ALL_UPDATES
13.5
11.6
14.2
261
7.56
NO_UPDATES
18.8
25.4
23.0
332
8.89
NO_SEDIMENTATION
13.6
11.7
14.5
270
7.56
NO_KFCUP_CHEM
18.7
24.2
17.6
285
7.97
NO_NOAH_SEAICE
13.4
11.6
16.8
309
7.54
NO_DMS
13.5
11.3
14.4
263
7.61
NO_SNOWDEP_SNOWPHOT
13.5
11.6
14.4
279
8.35
and
13.8
38.8
34.8
493
9.4
Validation against surface measurements of BC and SO42- in the Arctic
WRF-Chem simulation results are evaluated in
Fig. against surface equivalent BC (eBC) and
non-sea-salt sulfate measurements in the Arctic. The eBC is calculated based
on light absorption measurements by particle soot absorption photometers
(PSAPs) and converted to concentrations by assuming a value for
mass-absorption efficiency. As a result, the uncertainty in eBC measurement
is of at least a factor of 2 . SO42- is obtained
from filters and analyzed by ion chromatography. The contribution from sea
salt is removed to obtain a non-sea-salt sulfate concentration comparable
with WRF-Chem aerosol sulfate. Additional details about these measurements
are given in .
In terms of BC, the updated model run (ALL_UPDATES) agrees much better with
surface eBC measurements than the NO_UPDATES simulation, especially during
summer (decreasing RMSE by -8.8 ngm-3). Table
shows that this is mostly due to the implementation of the KFCUP_CHEM
(-3.4 ngm-3 of RMSE) and NOAH_SEAICE
(-2.6 ngm-3 of RMSE) updates, other updates having very little
effect (<0.3 ngm-3 change in RMSE). The average Pearson
correlation coefficient increases from 0.43 to 0.87, indicating that the
seasonal cycle of BC pollution is also improved in the model.
For sulfate, the updated model performs much better at Alert and Barrow
during summer, and slightly better at other stations, due to the competing
effects of increased sulfate from DMS and decreased sulfate from KFCUP_CHEM
and NOAH_SEAICE. Surprisingly, DMS has relatively little effect on the
SO42- RMSE on average (Table ). This is because
including DMS emissions and gas-phase chemistry improves RMSE at Pallas
(Finland), Alert (Canada), Nord (Greenland), and Barrow (Alaska) (-11 to
-27 ngm-3) but degrades RMSE at Zeppelin (Svalbard)
(+66 ngm-3), where the model already overestimates sulfate.
Another surprising result is the impact of dry deposition and UV-albedo
updates on sulfate (Table ). This effect is likely mediated
by changes in oxidants (OH and ozone, as discussed in
Sect. ) and their impacts on SO2
oxidation. These updates also improve the modeled seasonal cycle of sulfate,
increasing the average Pearson correlation coefficient from 0.28 to 0.73.
Both NO_UPDATES and ALL_UPDATES tend to be biased low in April (especially
at the most remote Arctic sites, Alert, Barrow, and Nord), which could be due
to underestimated long-range transport caused by the limited resolution
.
Monthly median BC (left) and SO42- (right) observations
at Arctic surface stations (gray shading indicates 25th and 75th
percentile) and corresponding WRF-Chem results (red, ALL_UPDATES; blue,
NO_UPDATES).
Ozone
Effect on surface ozone in the Arctic
The effect of the SNOWDEP_SNOWPHOT,
NOAH_SEAICE, KFCUP_CHEM, and DMS updates on surface O3
concentrations in the Arctic is shown in Fig. .
The effect of aerosol sedimentation aloft (SEDIMENTATION) on ozone is very
low and is not shown. The updates related to deposition and photolysis over
frozen surfaces have a strong effect on surface O3. Based on the
1-month-long simulations NO_SNOWDEP and NO_SNOWPHOT in April, we find that
this is mostly due to changes in dry deposition (+10 ppbv in April,
against +1 to +2 ppbv for photolysis). Ozone also decreases
slightly over sea ice with the SNOWDEP_SNOWPHOT update. This is likely due
to the UV flux increase from the SNOWPHOT update, since ozone formation in
the Arctic boundary layer is NOx-limited , and
ozone increases when the UV flux decreases in NOx-limited regions
. Ozone concentrations at the surface are strongly reduced by
the NOAH_SEAICE update (down to -10 ppbv), due to the increased
stability and lower influx of ozone precursors and ozone from the free
troposphere to the surface. The KFCUP_CHEM update also has a strong effect
on ozone (+2 to +5 ppbv), especially at lower latitudes where
convection occurs. This could be due to tracer transport by midlevel
convective clouds, bringing polluted air down to the surface
. Adding DMS leads to a modest decrease in surface ozone
over the open ocean (-2 ppbv at most), which is associated with a
decrease in NOx mixing ratios (-10 to -20 %), due to
an increased HNO3 sink (+5 to +20 %) from increased
N2O5 uptake on the additional sulfate aerosols (-20 to
-90 % N2O5 at the sea surface).
Change in the April to July average surface ozone due to
(a) KF-CuP cumulus effect on aerosol and trace gases,
(b) improved trace gas deposition over snow and improved UV albedo
for photolysis over snow and ice, (c) sea-ice-melt heat sink in
the Noah LSM, and (d) DMS emissions and gas-phase chemistry.
Validation against surface measurements of ozone in the midlatitudes and in the Arctic
WRF-Chem results from the ALL_UPDATES and NO_UPDATES simulations are
evaluated against surface ozone measurements from the EMEP (European
Monitoring and Evaluation Programme) European network and the CASTNET (Clean
Air Status and Trends Network) US network, in addition to ozone measurements
from the Barrow (Alaska) and Summit (Greenland) polar observatories of
NOAA-ESRL (National Oceanic and Atmospheric Administration, Earth System
Research Laboratory). The evaluation against Arctic stations (north of
60∘ N, 17 out of 228 stations) is shown in
Fig. b. When all updates are included, RMSE is
reduced for all seasons (-1.3 ppbv on average), even though the
ALL_UPDATES simulations sometimes overestimate ozone in spring. This
overestimation is clearly due to the fact that WRF-Chem has no treatment of
halogen chemistry in the model, which is responsible for ozone depletion
events in polar regions during spring
e.g.,. Table shows that
improvements in RMSE are mostly due to the SNOWDEP_SNOWPHOT update
(-0.8 ppbv RMSE) and to the KFCUP_CHEM update (-0.4 ppbv
RMSE). The average Pearson correlation coefficient also increases from 0.67
to 0.73 in the updated model, only due to the SNOWDEP_SNOWPHOT update. The
effect of the NOAH_SEAICE update is low, since only stations Nord in
northern Greenland and Barrow in Alaska are located in an area with
significant summer sea ice, where this change affecting surface mixing ratios
could play a role. Figure shows that these
updates also have a relatively strong effect over the whole measurement
network, including subarctic sites, indicating that these processes should
also be taken into account when studying ozone at lower latitudes with
WRF-Chem. At subarctic sites (south of 60∘ N), model updates
decrease RMSE by 13 % on average, and by more than 50 %
at nine surface sites. These improvements in the midlatitudes are mostly due
to the KFCUP_CHEM update, and to the SNOWDEP_SNOWPHOT update at sites where
seasonal snow is present. , using WRF-Chem 3.5.1, also
showed that reduced deposition and enhanced photolysis over snow could
contribute to high wintertime ozone in the United States, when other
favorable conditions were present, such as shallow boundary layers and high
emissions.
Comparison between daily averaged surface ozone measurements (black)
and WRF-Chem results (red, ALL_UPDATES; blue, NO_UPDATES) (a)
averaged over all stations within the domain (228 stations) and (b)
averaged over Arctic stations only (latitude > 60∘ N, 17 out of
228 stations).
Discussion about the differences with the quasi-hemispheric WRF-Chem simulation in Eckhardt et al. (2015) and
AMAP (2015)
The ALL_UPDATES simulation performs better than the WRF-Chem 3.5.1
simulation presented in and . Compared
to these earlier results, RMSE is improved in ALL_UPDATES by
0.3 ngm-3 for ARCTAS Spring rBC, by 27.2 ngm-3
for ARCTAS summer rBC, by 20.6 ngm-3 for surface BC, by
232 ngm-3 for surface SO42-, and by
1.84 ppbv for surface ozone (Table ). However, the
NO_UPDATES simulations also perform better than the simulation in
and , compared to most datasets
(RMSE higher by 5.0 ngm-3, lower by
13.4 ngm-3, lower by 11.8 ngm-3, lower by
161 ngm-3, and lower by 0.51 ppbv, respectively). This indicates
that the model updates presented here are only partly responsible for this
improved RMSE, and that differences in setup between the simulations also
play a large role.
There are many differences in model setup between the simulation in
and and the ones presented here. The
most significant are (1) the change of the gas-phase chemistry scheme
(SAPRC-99 here and CBM-Z earlier, but both being coupled to MOSAIC-8 bin
aerosols including aqueous chemistry); (2) the different fire emission
inventories (daily FINNv1.5 emissions here, monthly GFEDv3.1, Global Fire
Emissions Database, emissions earlier); (3) the larger simulation domain used
here, extending down to latitudes 10–35∘ N
(Fig. ), instead of 28–45∘ N earlier; and (4) the
inclusion of lightning NOx emissions here. Although it is difficult
to attribute precisely the improvement to each of these changes, the change
in fire emissions likely had a strong effect on modeled BC, since we find
that GFEDv3.1 BC emissions north of 60∘ N used in earlier WRF-Chem
simulations were 1.5 and 3.9 times higher in June and July than FINNv1.5 BC
emissions used here, a point also discussed in . Another
likely driver of errors for aerosols is the relatively small simulation
domain used earlier. This could have made WRF-Chem results too dependent on
the lateral boundary conditions from the MOZART-4 global model, in which
aerosols are represented by a simpler bulk aerosol scheme. The change of gas-phase mechanism, the use of a lightning NOx emissions scheme, and
the larger simulation domain used here also likely had an impact on ozone
results in the Arctic.
Conclusions
In this study, we update the WRF-Chem 3.5.1 model (with SAPRC-99 gas-phase
chemistry and MOSAIC aerosols) and perform quasi-hemispheric simulations of
aerosols and ozone in the Arctic region. This allows us to draw the following
main conclusions and perspectives:
Improved aerosols and ozone simulated by WRF-Chem 3.5.1 in the Arctic. Updating the model greatly reduces model errors compared to previous
WRF-Chem evaluations in the Arctic e.g.,. Specific
simulations with and without each model update allow us to characterize
which process has the most effect on Arctic pollution distributions.
Simulated airborne and surface BC in the Arctic is particularly sensitive to
the effect of cumulus clouds on aerosols and trace gases (wet removal,
aerosol activation, tracer transport, and cloud chemistry, represented by the
KF-CuP scheme) and to the representation of skin temperatures over sea ice,
affecting stability, in the Noah land surface model. Implementing these two
updates, as well as DMS gas-phase chemistry, also improves the representation
of sulfate concentrations in the Arctic, although the simple DMS chemistry
scheme used here appears to overestimate sulfate production at one Arctic
site (Zeppelin). In our simulations, neglecting sedimentation aloft does not
have a significant impact on BC or sulfate concentrations and has relatively
little influence on total aerosol concentrations, except in the upper
troposphere. Model updates also improve simulated ozone, both in the Arctic
and in the midlatitudes. The corrections to skin temperatures in the Noah LSM
have a strong impact on ozone over sea ice (-5 to -10 ppbv),
while the implementation of KF-CuP increases ozone by +2 to
+5 ppbv both in the Arctic and at lower latitudes. The main source
of improvement over land appears to be the implementation of a snow- and
ice-dependent UV albedo for the Fast-J photolysis scheme, and the decrease in
deposition velocities over snow-covered ground (>+10 ppbv combined
effect in spring where seasonal snow is present, in the Arctic and in the
midlatitudes). However, implementing these processes can sometimes degrade
model performance in the Arctic spring, by increasing ozone levels already
overestimated because of the lack of halogen chemistry in the gas-phase
mechanism.
Identification of potential areas of further improvement in the WRF-Chem model. The main discrepancies between modeled and observed ozone in
the Arctic occur in spring at coastal Arctic sites (e.g., Barrow, Alert,
Nord), where ozone depletion by halogen chemistry occurs. In order to study
springtime Arctic ozone it seems critical to include these processes in
WRF-Chem, as discussed earlier, e.g., in . WRF-Chem
underestimates aerosol surface concentrations in spring, which could be due
to underestimation of long-range transport due to the limited horizontal
resolution. found little improvement in BC transport
when decreasing the resolution from 2 to 1∘, but recent research
indicates that BC transport to the Arctic could be enhanced
in WRF-Chem simulations at much finer resolutions (< 10 km).
Simulations with the updated model have significantly lower RMSE for airborne
BC during summer (-54 %), but the model still significantly
overestimates BC aloft in this season. This could also be due to the low
resolution or to underestimated removal processes , since the
model does not currently represent aerosol activation in ice clouds and only
includes a simplified treatment of secondary activation in deep convective
clouds. Emissions from boreal fires could also be an important source of
uncertainty during summer, and for this reason it is important to validate
the different fire emission inventories in the Arctic.
Definition of a model setup that can be used in future work to study aerosol and ozone pollution on continental scales in the Arctic. The
updated model setup presented in this paper improves simulation of BC,
sulfate, and ozone in the Arctic. The updated results now appear to be in
better agreement than most global models included in the recent
intercomparisons of and , although
further model intercomparisons are needed to confirm this. There are many
pressing issues concerning short-lived pollutants in the Arctic and their
climate impacts which require reliable model results on the hemispheric
scale. For example, the relative importance of the different pollution
sources to Arctic pollution is still uncertain (local vs. remote, fossil fuel
vs. biomass burning, natural vs. anthropogenic). In addition, the attribution
of recent trends in Arctic composition can be difficult if long-range
transport from different source regions is not correctly reproduced. Other
Arctic issues could also benefit from accurate large-scale regional
simulations, such as the impact of Arctic air pollution on ecosystems (i.e.,
through deposition) and a more precise quantification of the climate impacts
of cloud–aerosol interactions in the Arctic and of BC deposition on snow
.