Introduction
Aerosols are of concern for urban air quality, but also an
important part of the climate system. Aerosol particles are liquid or solid
particles that are suspended in the atmosphere for some time, and range from
about 0.02 µm to several tens of µm in diameter (d).
Aerosols enter the atmosphere through primary production (e.g., dust or sea
salt emissions) or by nucleation and condensation of gases in the atmosphere,
and have both natural and anthropogenic sources
e.g.,. Climate is impacted by aerosols both directly
by their influence on the radiation budget and indirectly by their influence
on cloud optical properties and precipitation e.g.,.
Accurate multi-scale modeling of aerosols is difficult due to the variety of
processes involved, and thus aerosol-related processes remain the largest
source of uncertainty in assessments of anthropogenic radiative forcing
. Consequently, achieving the best possible representation
of the complex processes related to aerosols in models is an important task.
The atmospheric aerosol burdens are controlled by the aerosol sources,
chemical processes and rates of removal from the atmosphere. Removal can be
through transformation of aerosols (e.g. coagulation to form larger
particles; volatilization) or their complete removal from the atmosphere
which occurs by dry removal (dry deposition and gravitational settling) at
the surface and through wet deposition, i.e. removal by precipitation, which
is the focus of this study. While dry removal occurs only at the Earth's
surface, wet deposition can remove aerosols efficiently from the whole
troposphere. Clouds can form when moist air is cooled below the saturation
point of water vapor e.g.,. Within saturated air,
aerosol particles can act as nuclei for the water vapor to condense upon. The
efficiency of aerosol particles in serving as cloud condensation nuclei (CCN)
depends on their size and chemical properties as well as on the ambient
conditions. At low temperatures, ice crystals may also form on ambient
particles, which then act as ice nuclei (IN) . The
critical level of relative humidity (RH) determining which aerosol particles
are activated as CCN is described by Köhler theory .
When a droplet evaporates completely, non-volatile material is returned back
to the atmosphere, but often as transformed CCN or IN with different
physicochemical properties compared to the original particles. On the other
hand, if the cloud water precipitates to the surface, CCN or IN contained in
the droplets or ice crystals are also removed from the atmosphere. Since each
drop of precipitation can account for millions of cloud droplets, nucleation
scavenging is the most important mechanism for wet removal
. Nucleation removal of aerosol particles within clouds is
thought to account for more than 50 % of the aerosol particle mass
removal from the atmosphere globally .
Aerosol particles can also be collected by falling precipitation
through impaction (below-cloud
scavenging). The rate at which removal by impaction happens is dependent on
the probability of a collision of a falling hydrometeor with an aerosol
particle and the efficiency of subsequent collection of the particle by the
hydrometeor.
This paper describes and tests a new scheme for aerosol wet removal
implemented into the Lagrangian particle dispersion model FLEXPART. It is
based on the mechanisms of nucleation removal within the cloud and impaction
removal below the cloud. Section of this paper provides a short
description of FLEXPART in general, and introduces the new wet removal
scheme. In Sect. , we describe how the new scheme was tested and
compared with observations, and Sect. describes the results of
these tests. Finally, in Sect. conclusions are drawn.
Model description
The Lagrangian particle dispersion model FLEXPART
computes the transport and turbulent diffusion of atmospheric tracers (e.g.,
gases or aerosols). The model calculates trajectories based on meteorological
input data and can be used from local to global scales. Computational
particles follow the flow of the atmosphere resolved in the meteorological
input data, with random motions describing parameterized turbulence
superimposed on the particles' trajectories. Furthermore, a stochastic
particle column redistribution scheme is used to describe convection
. The meteorological data are usually taken from
operational analysis or re-analysis products. The reference version of
FLEXPART can ingest data from European Centre for Medium-Range Weather
Forecasts (ECMWF) or the National Centers of Environmental Prediction (NCEP).
Other versions of FLEXPART use e.g. data from the Weather Research and
Forecasting (WRF) model or the Norwegian Earth System
Model (NorESM) . We base our following discussion on the
reference version 10.0 in its configuration for ECMWF data products.
The aerosol removal scheme in FLEXPART has remained
relatively unchanged since its incorporation in the late 1990s. Other,
similar Lagrangian models like NAME and HYSPLIT have had recent updates to
their aerosol removal . However, the overall
level of detail also in these models remains low compared to known theory
e.g.. One reason for this is the limiting factors that
constrain the possible ways of treating aerosol removal within the Lagrangian
model framework. A main consideration within this framework is that each
transported computational particle is independent of others. Extensions of
this concept to allow for non-linear chemistry exist
, also for FLEXPART , but
the reference version of FLEXPART is a purely linear transport model. Within
such a linear model, it is impossible to include aerosol processes which
depend on the aerosol concentration (e.g., coagulation or non-linear chemical
reactions). Furthermore, to facilitate consistency between forward and
backward runs of FLEXPART, parameterizations that depend on the age of the
aerosol (i.e. time after emission for primary aerosols) should be avoided as
well. This limits the level of sophistication that can be incorporated into
an aerosol removal scheme. Nevertheless, a realistic treatment of aerosols is
possible even with these limitations.
Each computational particle released in FLEXPART represents an aerosol
population with a lognormal size distribution. While gravitational settling
is calculated only for the mass mean diameter of this aerosol population and
applied as an additional vertical velocity component when particles are
advected, dry deposition for details about the dry deposition in
FLEXPART, see is calculated for several weighted bins of the
size distribution a particle represents. The particle mass is then reduced by
the dry deposition for the computational particle as a whole, thus not
changing its size distribution. This simplified treatment of aerosol size
distribution can be extended easily by simulating several different types of
computational particles, each with its own size distribution (or discrete
size, if this is preferred). Removal processes acting differently for the
different aerosol particle sizes will then also modify the overall size
distribution.
The calculation of wet removal in FLEXPART can be divided in two parts: one
regarding the definition of the location of clouds, cloud water and
precipitation, and the other regarding the parameterization of the physical
removal of aerosols and gases during precipitation events. Both parts have
been revised and results will be presented in this paper.
Clouds and precipitation in FLEXPART
For a particle residing in a column with precipitation, it must first be
determined whether it is located within the cloud, above the cloud, or below
the cloud, before its wet scavenging can be calculated. Above the cloud, no
scavenging occurs; within the cloud, nucleation scavenging is used; and below
the cloud, the impaction scavenging scheme is activated. A new option has
been implemented in FLEXPART, so that the cloud vertical extent can either be
derived from three-dimensional ECMWF fields of specific cloud liquid water
content (CLWC) and specific cloud ice water content (CIWC) or from the summed
quantity specific cloud total water content (CTWC = CLWC + CIWC).
CTWC can be calculated by FLEXPART's ECMWF pre-processor to save storage
space required for the FLEXPART input data. Details of how the cloud water is
computed by the ECMWF Integrated Forecast System (IFS) model can be found in
and the processing of these
data is described in . If no cloud water content data
are available in the FLEXPART input files, cloud vertical extent can be
diagnosed from the vertical distribution of RH as in previous versions of
FLEXPART . However, this is considered much less accurate.
Multiple layers of clouds may appear both in the RH based parameterization
and in the ECMWF CTWC data. Not all of these cloud layers may be
precipitating but, because of lack of detailed information, in FLEXPART we
assume that all levels of clouds contribute to surface precipitation. An
inspection of the ECMWF cloud fields suggests that this assumption is of
minor importance as cloud layers with significant gaps in between account for
fewer than 10 % of the large scale precipitation events.
Meteorological information in FLEXPART is available only at the resolution of
the ECMWF input data. However, a grid cell with precipitation may, in
reality, also contain areas without precipitation, and this can reduce the
efficiency of aerosol wet scavenging substantially . The grid
surface precipitation intensity (It) is the sum of the advective
precipitation intensity Il and convective precipitation intensity
Ic from the meteorological input files. To scale this to sub-grid
precipitation intensity (I) the empirical relationship for the fraction of
a grid cell experiencing precipitation (F) is maintained from previous
versions of FLEXPART, described in . If a particle is found
to be in or below a cloud with precipitation, the scavenging coefficient
Λ is determined by either the in-cloud or below-cloud scheme
described in the following two sections.
In-cloud removal in FLEXPART
The nucleation scavenging in FLEXPART is activated only for particles
residing in the precipitating fraction of a grid cell F, see
, and only at altitudes where cloud water is present. For
consistency with I, the column cloud water is also scaled by the
precipitating fraction of the clouds, to get the sub-grid precipitating cloud
water (PCW):
PCW=CTWCFcc
Here, cc is the surface cloud cover and so F/cc is the fraction of
cloud water in the precipitating part of the cloud. If PCW > 0 in-cloud
scavenging is applied.
An important intermediate quantity to determine is the in-cloud removal rate
of aerosols due to the removal of cloud water by precipitation, which is
given by the cloud water washout ratio I/PCW. To obtain accurate
values for I/PCW, it is important that I and PCW are
consistent. Both values are derived from ECMWF data, however, I is derived
from accumulated precipitation values (i.e., precipitation accumulated during
one ECMWF data output interval, typically 1 or 3 h), whereas PCW is
an instantaneous quantity, and this can cause small inconsistencies.
Furthermore, I/PCW does not take into account the efficacy of
turbulent overturning and the replenishment rate of cloud water from
condensing water vapor. The aerosol scavenging coefficient Λ
(s-1) is now given as
Λ=FnucIPCWicr,
where Fnuc, the nucleation efficiency, is the fraction of the
aerosol within the cloud that is in the cloud water (see Fig. ).
While icr represents the cloud water replenishment rate, it
cannot be determined from the ECMWF output data. Therefore, the determination
of the constant icr was done on the basis of empirical testing in
FLEXPART and must be considered a tuning parameter.
Compared to the previous FLEXPART scheme described in ,
icr/PCW replaces the cloud water representation that
was calculated based on an empirical relationship with precipitation rate
(cl=2×10-7I0.36). The overall best results were obtained
for icr set to a value of 6.1 for the ECMWF cloud water
fields, which is used for all simulations in this paper. This resulted in a
somewhat slower in-cloud removal rate with the new compared to the old
parametrisation. Comparison of the two parametrisations also shows that using
icr/PCW gives overall weaker dependence on I,
compared to cl in the old removal scheme. For simulations where in-cloud
removal constitutes a large fraction of the removal, i.e. especially for
soluble accumulation mode aerosols, the empirical value of
icr has a large impact on overall removal rates.
In reality, Fnuc depends on many different variables such as
aerosol size, chemical composition, surrounding aerosols, temperature and
cloud phase and microphysical properties. However, a complete
parameterization of Fnuc is not possible in FLEXPART because of a
lack of information. What can be constrained within FLEXPART is that most
aerosols have very different nucleation efficiency for liquid, mixed-phase
and ice clouds. Therefore, we introduced as a new feature that for
determining the nucleation efficiency (Fnuc), we now distinguish
the efficiency of aerosols to serve as cloud condensation nuclei
(CCNeff) and ice nuclei (INeff). By contrast, in the
old scheme all aerosols had Fnuc≡0.9. For ice clouds,
Fnuc is set equal to INeff, for liquid water
clouds, Fnuc is set equal to CCNeff, and for
mixed-phase clouds, we use α, the fraction of the cloud water in ice
phase shown in Fig. as a black line see for details on
calculations of α, to interpolate between
Fnuc and CCNeff:
Fnuc=1-αCCNeff+αINeff.
The fraction of cloud water that is in the ice phase (α) if
CTWC is used (black line) and the fraction of aerosols that reside within
cloud droplets (Fnuc) (colored lines and dots) as a function of
temperature. For Fnuc, partitioning values for aerosol number
from (red line), from (magenta
dots) and from (valid for black carbon (BC) particles)
(blue line) are shown. For the BC partitioning, ice fraction was converted to
temperature using α.
There are no unique globally representative values for
CCNeff or INeff because they depend not
only on the aerosol particle itself, but vary also with aerosol
concentrations and cloud properties (e.g., updraft velocities). Some general
considerations can however be made. In a review of measurements conducted at
the high alpine station Jungfraujoch, showed that
Fnuc varies significantly with both aerosol size and cloud phase.
found that the fraction of particles with dp>0.1 µm activated in a cloud dropped from 56 % in liquid
summer clouds to 0.08 % in winter ice clouds. The lower ice phase values
are attributed to the Bergeron–Findeisen process
, by which relatively few ice crystals grow
at the expense of many more liquid droplets. When the droplets evaporate the
non-volatile aerosol content is released back to the atmosphere. This
temperature dependent effect is illustrated in Fig. , where the
partitioning between cloud water and surrounding air of total aerosol number
according to is shown (magenta dots). Also shown in
Fig. are the similar results of (red line)
and the BC partitioning (blue line) reported by . Hence it
is generally assumed that for most aerosol particles
CCNeff>INeff.
found that the average scavenged fractions in clouds
during spring in Cumbria, UK, were 0.77 for sulphate and 0.57 for soot in
clouds formed in continental air, and 0.62 and 0.44 respectively, for clouds
formed in marine air. The time and place for these measurements suggest that
these were mainly liquid phase clouds. In other studies
, it was found that larger aerosol
particles have a higher nucleation efficiency than smaller particles. Such
information can be used by FLEXPART users to prescribe appropriate
CCNeff and INeff values for different
aerosol particle types and sizes.
Below-cloud removal in FLEXPART
Raindrops and snow flakes fall at approximately terminal velocity through the
air and may scavenge aerosol particles as they collide
with them in the ambient air below the cloud base. This below-cloud
scavenging process depends both on the probability that the falling
hydrometeor collides with an aerosol particle (collision efficiency) and the
probability of attachment (coalescence efficiency). Both probabilities
together determine the collection efficiency. Collection efficiencies of both
snow and rain have a minimum for aerosol particle sizes near
0.1–0.2 µm in what is known as the Greenfield gap
. Notice that dry deposition is also least efficient
for such particles. For aerosol particles of these sizes, neither Brownian
diffusion nor impaction is efficient. Whilst Brownian diffusion is the
dominant process of attachment for sub-micron particles, inertial impaction
is the dominant process for larger aerosol sizes and becomes dominant above
∼ 1 µm, though there are large discrepancies between
theoretical predictions and observations e.g.,. The
collection efficiency is strongly dependent on the sizes of both the falling
hydrometeors (and their terminal velocity) and the aerosol particles. It also
depends on the precipitation type.
The below-cloud scavenging parameterization in FLEXPART differentiates
between rain and snow because especially for large aerosol particles a large
difference in scavenging efficiency is found between the two, where snow is
more efficient than rain . Of many possible
parameterizations for liquid precipitation, the one of was
chosen, for which all the required information is available in FLEXPART. The
parameterization takes into account rain intensity I (used to parameterize
droplet size) and the aerosol dry diameter and is based on field measurements
over 6 years in Hyytiälä, Finland. The scavenging coefficient
λ (s-1) for particles below a cloud is given by
log10λλ0=C*a+bdp-4+cdp-3+ddp-2+edp-1+fII00.5,
where C* is a scalar, dp=log10DpDp0, λ0=1 s-1, I0=1 mmh-1, and Dp0=1 m. Coefficients for
factors a–f are given in Table . While originally intended for
particles of size 0.01–0.51 µm, the parameterization by
is one of few parameterizations that takes into account
data for larger aerosol particles up to 10 µm diameter, and
should thus provide reasonable results also for these larger particles. For
rain C*=Crain and is a preset scalar variable that makes
modifications to the removal scheme possible. The suggested default value for
Crain is 1.
Below-cloud scavenging coefficients as a function of aerosol size.
Shown are the new parameterizations of (blue lines) for
rain and (black line) for snow, and the old parameterization
of used in previous FLEXPART versions with the parameters
A=1e-5 and B=0.62 (green). Values are shown for four different
precipitation intensities: 0.1 (dotted lines), 1 (solid lines), 3 (dashed
lines) and 5 mmh-1 (stippled lines).
For snow scavenging, we use a parameterization reported by ,
which was also derived from Hyytiälä data, but during snowfall. It is
fitted with the same function as given by Eq. () but with
coefficients derived for snow and also given in Table . In this
study we have used a local temperature threshold of 0 ∘C is to
distinguish between rain and snow, but it is also possible to use rain and
snow precipitation intensity read directly into the model from ECMWF analysis
data. The Kyrö function is independent of precipitation intensity or type
of falling snow as is common for snow scavenging parameterizations see
e.g.,. The shape of the snow crystals is very
important for the scavenging efficiency, but cannot be derived from the ECMWF
data. This aspect is thus ignored, and the Kyrö function is averaged over
many different types of snow crystal shapes instead.
Parameters used in Eq. () for below cloud scavenging from
and .
a
b
c
d
e
f
C*
Io
λo
Laakso
274.36
332 839.6
226 656
58 005.9
6588.38
0.24498
Crain
1
1
Kyrö
22.7
0
0
1321
381
0
Csnow
1
1
Figure shows the below-cloud scavenging parameterizations for rain
and for snow for different precipitation rates and compares them with the old
parameterization used in FLEXPART, which was based on . The
aerosol removal rate is increased relative to previous versions of FLEXPART
for almost all precipitation rates. Aerosol chemical properties may also
influence the below-cloud scavenging coefficient. In FLEXPART, this influence
can – to some extent – be accounted for by setting the parameters
Crain and Csnow (C* in Eq. ), which
are scalars used to scale the collection efficiency for rain and snow, to
values different from 1. For example, with Crain=0
(Csnow=0), no below-cloud scavenging for rain (snow) would occur
in FLEXPART.
As parameterizations by both and are
based on bulk aerosol there may be differentiating factors for certain
aerosol types, though very little specific evidence of this exists
. Comparisons with other impaction scavenging
parameterizations see e.g., for rain show that the
scavenging values are on the middle to low side of
existing parameterizations and that differences between different
parameterizations cover at least one order of magnitude. Choosing values for
Crain and Csnow between 0.1 and 10 should cover this
uncertainty range.
Model simulations
Three different global model experiments were set up to test the new
scavenging parameterizations for different types of aerosols: BC, mineral
dust and sulphate. The main purpose of these experiments is to explore the
performance of simulations that cover a broad range of aerosol particle types
and sizes, evaluate simulated atmospheric concentrations against
observations, and calculate e-folding lifetimes.
Mineral dust
Mineral dust arguably constitutes the largest mass of aerosols in the
atmosphere. Dust particles span a wide range of sizes and can be found far
from their source . Small dust particles have been found to
mix somewhat with volatile aerosol components but particles larger than
0.5 µm are inert in the atmosphere . Mineral
dust is thus well suited to model with FLEXPART. Model experiments were set
up to examine the role of impaction and nucleation scavenging as well as dry
deposition and gravitational settling for different sizes of mineral dust.
Emission of mineral dust was calculated based on a module presented by
. In short, dust emission was initiated from bare land
when friction velocity exceeded a threshold value for initiation of
saltation, depending on soil properties and soil moisture content. The soil
fraction available for erosion was determined from land cover data
GLCNMO version 2, based on MODIS images. Vertical
fluxes of mineral dust were derived according to .
Particles were subsequently released in FLEXPART over a layer of 300 m
height, at a 0.5∘ resolution in 6-hourly time steps. We assumed an
aerosol particle size distribution in ten particle size bins, varying between
0.2 and 18.2 µm, as suggested by . FLEXPART
simulations were run in forward mode for the year 2010.
Radionuclide tracers attached to sulphate aerosols
An evaluation of modeled aerosol lifetimes was recently performed by
who made use of measurements of radioactive isotopes
released during the Fukushima Dai-Ichi nuclear power plant (FD-NPP) accident
in March 2011. The radionuclide cesium-137 (137Cs) was released in large
quantities during the accident and measurements suggested that they mainly
attached to the ambient accumulation-mode sulphate aerosols
. Another radionuclide, the noble gas xenon-133
(133Xe) was also released during the accident and can serve as a passive
transport tracer. Both radioactive isotopes were transported and measured
across the Northern Hemisphere for more than three months after their
release, providing a unique constraint on modeled aerosol lifetimes
.
We have used measurements of the aerosol-bound 137Cs and the noble gas
isotope 133Xe from March to June 2011 at 11 different measurement
stations of the Comprehensive Nuclear-Test-Ban Treaty Organization (CTBTO)
network see Fig. 1 of. All measured radionuclide
concentrations were corrected for their radioactive decay and converted to
activity per cubic meter for comparison with the model data. Detailed
descriptions of these measurements and how they can be used to determine
aerosol e-folding lifetimes were provided by
.
Over the 46 days of measurements (starting 14 days after the initial
emission) used to evaluate e-folding times of 137Cs (and, implicitly,
of the accumulation mode sulphate aerosol to which it attached),
found FLEXPART concentrations to decrease by three
orders of magnitude more than the measurements. The decrease started from an
initial overestimation of the 137Cs concentrations but later the
concentrations were underestimated at all but one CTBTO stations.
Consequently, a too short e-folding lifetime of 5.8 days was calculated for
FLEXPART as compared to 14.3 days derived from the measurements. In this
paper, we repeat the simulations of but with the new
removal scheme for aerosols.
Black carbon
FLEXPART has been used in several recent studies to model BC with a focus on
the Arctic . All these studies used a
FLEXPART version where the in-cloud scavenging efficiency of the reference
FLEXPART version had been reduced by one order of magnitude. This has
produced realistic concentrations for the Arctic. In this study, we tested
the new scheme against measurements at Arctic and mid-latitude stations to
assess how well BC concentrations are captured.
For BC, simulations were made both in forward and backward mode, and results
were compared to test model consistency. When run in backward mode, FLEXPART
output is a gridded emission sensitivity that can be coupled with emission
fluxes to obtain the concentrations at the release point. For all
simulations, concentrations obtained by forward and backward simulations by
FLEXPART differ only due to statistical noise.
Emissions used for BC simulations were ECLIPSE v4.0
available through the website http://eclipse.nilu.no. Added to these
are shipping emissions from AEROCOM and GFEDv3.1
emissions for forest and savannah fires , all
resolved monthly and on a 0.5∘×0.5∘ grid. European
measurements of aerosol absorption were collected from the Database for
Atmospheric Composition Research (EBAS) with the aim of using data from
stations with similar particle soot absorption photometer (PSAP) instruments.
The stations were selected to represent different environments, ranging from
locations close to pollution sources in Central Europe to remote locations in
the Arctic. We chose the sites Melpitz (MEL, 51.32∘ N
12.56∘ E) in Germany which is surrounded by strong BC sources,
Pallas (PAL, 67.80∘ N 27.16∘ E) in Finland and Southern
Great Plains (SGP, 36.50∘ N 98∘ W) in the US at
intermediate distances from the sources, and Zeppelin (ZEP,
78.93∘ N, 11.92∘ E), Barrow (BRW, 71.30∘ N,
156.76∘ W) and Alert (ALT, 82.50∘ N, 62.34∘ W) as
remote sites.
PSAPs measure the particle light absorption coefficient. Conversion of this
coefficient to equivalent BC (eBC) mass concentrations is not straightforward
and requires certain assumptions , leading to
site-specific uncertainties on the order of a factor of two. We have used
conversion factors of 6.50 m2g-1 for PAL and
5.50 m2g-1 for ZEP, where site-specific information was
available and 10 m2g-1 for MEL, ALT, BRW and SGP. For ALT and
BRW a gap with more than a month of missing data for 2007 was filled with
climatological values of all available data after year 2000. For PAL only
climatological observations were used.
Results
Wet scavenging event statistics
To explore how frequent in-cloud and below-cloud scavenging events are and
where they occur, we used a 3-month (December 2006 to February 2007) global
ECMWF data set (1∘×1∘ with 92 vertical layers) and
classified each grid cell as being either in a cloud-free column or, if
clouds exist in the column, in, below or above the cloud. The vertical extent
of each layer increases with altitude, which emphasises lower altitudes when
a raw count of events is done, so for a more realistic representation the
numbers presented here are weighted by the mass of each model layer (using a
standard atmosphere). Convective and large scale precipitation events were
differentiated using surface precipitation and for each event classified as
the larger of the two.
Cloud top heights and the frequency of scavenging events are shown in
Fig. , both using the ECMWF cloud water information (blue) and the
cloud parameterization based on relative humidity (red). Close to the
equator, the precipitating clouds from ECMWF have on average high cloud tops,
often extending all the way to the tropopause. For the period examined, more
than 96 % of the in-cloud removal events in the tropical band
(15∘ S–15∘ N) are convective. For the
15–60∘ latitude range the cloud tops are markedly lower and the
frequency of convective removal events drops markedly to 46 % which is a
result of both more stratiform clouds and fewer and lower convective clouds.
This can be seen in the left panel of Fig. as an extension of the
25–75 % percentile range, which indicates that there are both low
stratiform and high convective cloud tops. The fraction of large scale
in-cloud events in this area is 46 %. Poleward of 60∘, stratiform
precipitation dominates with 76 % of all events.
Left: the zonally averaged median cloud top heights of precipitating
clouds as a function of latitude, averaged over a 90-day period starting in
December 2006. Clouds are defined using either the FLEXPART relative
humidity-based parameterization (red line) or by CTWC data (blue line). The
shaded areas span the 25–75 percentiles. Right: number of potential removal
events globally, for in-cloud nucleation scavenging (solid lines),
below-cloud impaction scavenging (dashed lines) and the sum of the two
(dotted lines), for both the parameterized clouds (red lines) and when ECMWF
cloud water fields are used (blue lines). Note that the height scales are
different for the left and right panels.
Globally, in-cloud scavenging accounts for 85 % (91 % above
1000 m) of the aerosol wet removal events, of which 57 % occur in
convective clouds (for ECMWF clouds). The global fraction of in-cloud (solid
line), below-cloud (dashed) and total (dotted) removal events as a function
of altitude is shown in Fig. (right). For the ECMWF defined clouds
(blue) there are very few below-cloud scavenging events above 1000 m.
There is however a slight increase in the frequency of such events around
5000 m, which is due to multiple layers of clouds. In the instances
where precipitation was predominantly large scale (21 %), at altitudes
above 5000 m, in reality most clouds are likely non-precipitating
cirrus clouds, and the ECMWF precipitation is actually originating from lower
cloud layers. This could also be related to both convective and large scale
clouds residing in the same grid cell, but without information about the
three-dimensional distribution of hydrometeors, a correct diagnosis is not
possible and many of the high-altitude below-cloud scavenging events are
probably not real. However, in total this accounts for only 4 % of all
below-cloud scavenging events. Defining clouds on the basis of relative
humidity produces an almost 4 times higher occurrence (15 %) of such high
altitude (> 1000 m) below-cloud removal events, which is likely
unrealistic.
The water phase of clouds influences the removal efficiency for aerosols that
are inefficient IN but efficient CCN (or vice versa). The phase partitioning
is temperature dependent and varies with season, latitude and altitude. For
the 3 months examined, globally 16 % of the in-cloud removal events were
liquid only, 7 % were ice only, whereas the remaining 77 % were
defined as mixed-phase cloud removal events.
In previous versions of FLEXPART, clouds were parameterized using relative
humidity. As can be seen in Fig. , this leads to several
differences in the distribution of scavenging events from the ice and liquid
water based cloud distribution. For instance, the high frequency of clouds
extending all the way to the surface seems unrealistic, and often no clouds
could be found in a grid cell with precipitation (not shown). Altogether, in
the new scheme the cloud distribution is more consistent with the
precipitation data and thus it produces a more realistic distribution of
below-cloud and in-cloud scavenging events with 52 % of the events below
1000 m being below-cloud removal events.
Relative frequency of removal events for a pulse of dust emitted in
Central Europe on 14 April 2007. For illustration purposes, daily frequencies
were fitted with a polynomial.
Sensitivity analysis of lifetimes τF for dust particles of
different sizes (diameter d) and with different settings of the removal
parameters. Bold values: default parameter settings, italicised values: the
longest lifetime if only one deposition process is reduced and underlined
values: the changed parameter(s) relative to the default
settings. The last two rows, denoted by “ * ”, reports results
obtained with FLEXPART v9 with the old removal scheme, where below cloud
scavenging was parameterized as Λ=AIB, with A=1e-4 and
B=0.62.
Parameters
Lifetime (days)
Crain
Csnow
CCNeff
INeff
d=0.2 µm
d=2.2 µm
d=6.2 µm
d=10.2 µm
d=18.2 µm
1.00
1.00
0.15
0.02
31.8
11.6
1.8
0.8
0.3
1.00
1.00
0.07
0.02
39.1
12.5
1.9
0.8
0.3
1.00
1.00
0.01
0.02
51.6
13.7
1.9
0.8
0.3
1.00
1.00
0.15
0.01
32.7
11.6
1.8
0.8
0.3
1.00
1.00
0.15
0.00
33.4
11.7
1.8
0.8
0.3
1.00
0.50
0.15
0.02
35.8
12.6
1.8
0.8
0.3
1.00
0.10
0.15
0.02
40.5
16.3
1.8
0.8
0.3
0.50
1.00
0.15
0.02
34.1
12.5
1.9
0.9
0.3
0.10
1.00
0.15
0.02
36.4
13.6
2.0
0.9
0.3
0.10
0.10
0.01
0.00
141
31.9
2.2
0.9
0.3
A*
B*
0.9*
0.9*
5.4
3.7
1.7
0.8
0.2
A*
B*
0.09*
0.09*
19.2
9.4
1.8
0.8
0.3
While Fig. shows the global distribution of scavenging events, the
actual relative probability of in-cloud versus below-cloud scavenging events
versus dry removal events for a given particle depends on the distribution of
the aerosol. To illustrate this, we released a pulse of 1 million particles
representing dust of five different sizes (see Table ) at
10 m a.g.l. over Central Europe on 14 April 2007. Figure
shows the relative frequency of the different removal events for these
aerosol particles as a function of time after the release. For the purpose of
clearer illustration, we show a polynomial fit through the daily total number
of events of each removal type. Initially, below-cloud scavenging and dry
removal are the most frequent removal types. Exact numbers at the beginning
will vary depending on the location and time of the release. However, as
particles are transported to higher altitudes, the relative frequency of
in-cloud removal events increases, exceeding that of the other event types
from day 4. On day 7 after the emission pulse, the relative frequencies are
already similar to the global distribution of scavenging events in the
troposphere, where below-cloud scavenging accounts for only 15 % and dry
removal for only 3 % of the number of events. Notice that in terms of
aerosol mass removed, the importance of below-cloud scavenging and dry
removal will decrease even more quickly because the mass of particles
remaining in the lower troposphere will also decrease rapidly. This effect
has been discussed in . The time dependence of scavenging
is an important feature as most primary aerosols are emitted at or near the
surface. Figure also shows that, despite the global dominance of
in-cloud scavenging events, below-cloud scavenging or dry removal may be most
important, at least for aerosol types for which these removal mechanisms are
efficient. The dependence in the efficiency and nature of scavenging also
means that aerosol lifetimes are different for fresh and aged aerosols, as
discussed in .
Mineral dust
Since the below-cloud scavenging scheme has a strong size dependency, an
important goal for our mineral dust simulations was to investigate the
differences in lifetime for aerosol particles with a large range of different
sizes. Also, mineral dust particles are ineffective CCN
e.g., and, therefore, below-cloud scavenging is very
important for dust. To investigate the sensitivity of dust scavenging to
various components of the scavenging scheme, we performed simulations for a
range of parameter settings.
The resulting lifetimes (τF) are shown in Table . Lifetimes
were calculated as the times when the dust mass has decreased to 1/e of the
emitted mass. Values of τF are equivalent to e-folding times if the
removal rate is constant. While this is not the case – as shown in the
previous section, it allows a simplified lifetime calculation and is
sufficient for our purpose of investigating the systematic dependence of
lifetime on aerosol particle size and choice of scavenging parameters. It
also emphasizes the initial phase of removal when most of the emitted mass is
lost.
The accumulation mode particles of mineral dust are in the 0.2 µm
size bin, which is locatedclose to the minimum of both impaction efficiency
(Fig. ) and dry removal. Consequently, and especially since dust
particles are also inefficient CCNs, the 0.2 µm sized particles
have very long lifetimes. With the standard parameter settings in FLEXPART
for dust (Csnow=Crain=1; CCNeff=0.15;
INeff=0.02, highlighted in green in Table ), the
lifetime of accumulation mode-sized (0.2 µm) dust is almost
32 days. Even though dust particles are inefficient as CCN, wet removal
dominates the total removal for the two smaller reported size bins and
nucleation scavenging in liquid water clouds is the dominant removal process.
Only if CCNeff is decreased further by one order of magnitude,
its importance is diminished and the lifetime increases to > 50 days.
Compared to τF obtained from the old scavenging scheme, the
0.2 µm size bins have significantly increased τF. The
increase is in part due to fewer clouds extending all the way to the surface
(Fig. left), thus decreasing the low altitude removal most
important initially. However, most of the increase is due to the decreased
CCNeff and INeff. The version 9 simulation with
CCNeff=INeff=0.09 is equivalent to the
parameters used for BC v9 simulation in Sect. (Fig. ).
The loss of particles of size 2.2 µm is more strongly affected by
gravitational settling, but still dominated by wet removal. Impaction
scavenging is also about four times more efficient for aerosols of this size
than for 0.2 µm particles, and thus has a large impact on the
atmospheric lifetime. This is important especially close to the sources, when
the aerosols are predominantly in the lower troposphere where below-cloud
removal occurs most frequently. Consequently, the lifetime τF, 11.6
days, is substantially shorter than for the 0.2 µm particles.
There is also a strong sensitivity to the choice of the Csnow
value for scavenging due to ice, which is probably related to the strong size
dependence of the scheme.
For the even larger particles shown in Table , dry deposition
combined with relatively fast gravitational settling take over as the most
important removal mechanisms and thus very little effect is seen from
altering the wet removal parameters. For the 6.2 µm particles,
reducing all wet removal parameters by one order of magnitude, only increases
the simulated lifetime by 20 %, compared to the 350 % increase in
lifetime for the accumulation mode particles. For the 18.2 µm
particles, wet scavenging has virtually no impact on the lifetime, which is
entirely controlled by gravitational settling.
A multi-year study of mineral dust, using FLEXPART with the same removal as
here found very good correlation
between observations and model concentrations using a global network of
observations positioned at various distances from major source regions. While
the 32-day lifetime τF obtained for the 0.2 µm particles
seems long, the emission to column burden estimate of lifetime for the full
dust size distribution is only 4.3 days, which is on the low side of commonly
reported estimates e.g.,. Notice that the mass fraction
of dust aerosols with diameter < 1 µm is very low in our
emission scheme .
Radionuclide tracers representative of sulphate aerosols
The FLEXPART model set-up for simulating the aerosol-bound cesium transport
after the Fukushima accident was the same as in ,
except for the updates in the cloud and wet scavenging schemes described in
this paper. Furthermore, used only one aerosol size
mode, with d=0.4 µm. Here, a more realistic aerosol size
distribution was used, and compared to the measurements of 137Cs surface
activity by . For these simulations, the mass was emitted
in six different size bins (Table ) ranging from d=0.4 to
6.2 µm. The size bins with logarithmic mean diameters of [0.4,
0.65, 1, 2.2, 4, and 6.2] µm received 1, 2, 10, 40, 32, and
15 % of the emitted mass. The resulting relative aerosol surface size
distribution is shown in Fig. b at the time of the release (green)
and for an aged distribution after 40 days (cyan) together with the measured
137Cs aerosol surface activity size distribution (red) of
. It is worth noting that started
their measurements 47 days after the largest emission but probably sampled
mainly 137Cs from small later releases. The measured size distribution
of 137Cs is bimodal with peaks around d=1 and 0.02 µm. The
larger peak at 1 µm fits well the released size distribution in
FLEXPART. The peak of the aged size distribution is dominated by particles of
0.6 µm. While the initial release included a significant fraction
of particles with diameter larger than 1 µm (52 % by mass and
7 % by aerosol number), their fraction is reduced considerably by day 40
(3 % by mass and < 0.1 % by number). The smaller mode around
0.02 µm is not represented in the model but it accounts for only
5–6 % of the total mass.
For evaluating the modeled aerosol lifetimes in the same way as
, we calculate the ratio of the aerosol (137Cs)
to the passive tracer (133Xe) at each measurement station shown in
Fig. a. The ratios decrease with time due to removal of aerosols.
We further calculate the daily median ratios (median concentration for each
day over all stations), and fit an exponential decay model (grey lines in
Fig. c) to these daily ratios. The fit is done over days 15–65
after the start of emissions, for which sufficient measurement data exist
seefor details. This excludes the initial phase of
removal (as shown in Fig. ) and thus emphasizes the role of
in-cloud scavenging. We therefore use the e-folding time of the exponential
decay model as an estimate for the aerosol lifetime (τe).
(a) The concentration of 137Cs in the Northern
Hemisphere on day 15 after the initial release. The locations of the
observational sites used in this paper are marked with colored circles.
(b) Normalized initial (green) and aged (cyan) aerosol surface area
distribution of the aerosols used in the simulation. For comparison the
measured aerosol size distribution of is shown in red.
(c) Simulated 137Cs/133Xe concentration
ratios for the different stations as a function of time after the accident;
upside down black triangles represent median daily ratio values over all
stations. The circle colours used for the different stations correspond to
those used in panel a. The dark gray line shows the fit to the observed
concentrations see. The light gray line shows the
log-linear fit to FLEXPART version 10 model data, and the pink line the fit
to version 9 data. (d) Ratio of modeled to observed concentrations
as a function of latitude for the passive tracer 133Xe (blue circles)
and the aerosol-bound 137Cs (green circles). For reference also a 1:1
line is shown (dotted black) and a fit to the FLEXPART version 9 data is
shown in pink.
The e-folding lifetime estimate obtained by for the
previous version of FLEXPART was 5.8 days, indicating a too quick removal of
the aerosols compared to the measurement-derived τe value of 14.3 days.
However, there was only a slight underestimation of the atmospheric
concentrations, partly explained by an initial overestimation. The new
scavenging scheme produces a longer e-folding lifetime of 10.0 days
(Fig. c). The longer lifetime is mainly due to slower in-cloud
scavenging and a broader range of aerosol particle sizes emitted, which have
different removal efficiencies. Both the below-cloud scavenging as well as
the dry removal are size-dependent. This also explains the shift towards
smaller particle sizes from the initial distribution to the aged distribution
in Fig. b.
Sensitivity analysis of e-folding lifetimes τe for particles
of different sizes (diameter d) and with different settings of removal
parameters, for the Fukushima case study. The lifetime is also calculated for
the total size distribution (Distr., last column). In addition to the
lifetime, the relative bias (bias), calculated as the average of all the
daily mean concentrations simulated with FLEXPART divided by the observed
daily mean concentrations for all days after day 15, is also reported. Cases
where the simulated concentrations were too low to reliably estimate lifetime
or bias are denoted with LC*.
Removal coefficients
d=0.4 µm
d=0.65 µm
d=1.0 µm
d=2.2 µm
d=4.0 µm
d=6.2 µm
Distr.
#
Crain
Csnow
CCNeff
INeff
τe
bias
τe
bias
τe
bias
τe
bias
τe
bias
τe
bias
τe
bias
1
1.00
1.00
0.90
0.90
11.7
18.7
10.8
11
9.6
5
7.6
0.2
5.4
0.01
LC*
LC*
10.08
0.99
2
1.00
1.00
0.40
0.40
17.9
103
15.2
55.5
12.0
21.3
7.9
0.8
5.5
0.02
3.0
LC*
13.4
4.6
3
1.00
1.00
0.15
0.15
25.2
192
19.2
109
13.8
38
8.1
1.1
4.8
0.02
2.8
LC*
18.6
96
4
1.00
1.00
0.00
0.00
66.0
> 103
–
–
–
–
–
–
–
–
–
–
–
–
5
0.00
0.00
0.90
0.90
–
–
–
–
–
–
11.0
1.3
–
–
–
–
–
–
The e-folding times calculated individually for the different size bins are
reported in Table . Simulation #1 in the top row (green) show the
results with scavenging parameters set to values believed to be valid for
sulfate, which are also used in the simulation shown in Fig. . The
e-folding lifetimes range from 11.7 days for the 0.4 µm size
bin, to 5.4 days for the 4 µm bin. Even the smallest two aerosol
size bins have a shorter e-folding lifetime than what is derived from the
CTBTO measurements. For the largest size bin, concentrations after 15 days
were too low for a robust estimate of lifetime.
The second column in Table for each aerosol size bin reports the
ratio of modeled to observed concentrations averaged over the whole period,
assuming that all 137Cs was attached to aerosols of that size bin.
Assuming that 137Cs attached exclusively to particles smaller than
1 µm (first two size bins), which have the most realistic
lifetimes compared to the observation-derived lifetime, leads to a large
overestimate of the observed concentrations (ratios of 18.7 and 11). This
might to some extent be due to an overestimate of the emissions used here, by
. Indeed, other authors e.g. have
found smaller emissions, but the source term uncertainty of about a factor of
two cannot alone explain the overestimates by the smaller modes. Assuming
that all 137Cs attached to particles larger than 2.2 µm, on
the other hand, leads to underestimates of both the concentrations and the
lifetimes compared to the observations.
From the differences between the simulations for different aerosol sizes, it
is also possible to investigate the relative importance of different removal
mechanisms for the different aerosol sizes. Furthermore, several different
in-cloud parameters INeff and CCNeff were tested. In
simulations #2 and #3 in Table , INeff and
CCNeff were reduced to values of 0.4 and 0.15, respectively. In
simulations #4 and #5, in-cloud and below-cloud scavenging were separately
turned off completely. For these simulations, only one aerosol size was used.
Comparison of the lifetimes and ratio of these simulations with the original
137Cs simulation #1 (Table ) shows that for submicron
particles the governing removal process is in-cloud scavenging. For particles
in the range ∼ 0.05–0.8 µm, dry deposition is slow and
also the below cloud removal in FLEXPART is not very efficient, which leaves
in-cloud scavenging to control the lifetime. This is apparent from how
changes in removal efficiency influence the model values and lifetimes
differently for different aerosol sizes. When CCNeff and
INeff are reduced by 60 % to 0.4 in simulation #2, the
atmospheric burden is increased by a factor of 5 for 0.4 µm
particles. The lifetime however, only changes from 11.7 to 17.9 days, i.e. by
a factor of ∼ 1.6. For the four larger aerosol size bins much smaller
changes are found between #1 and #2 in concentration, lifetime and ratio,
due to the less dominant role of in-cloud scavenging for these particles.
The measurement data during the first 15 days after the start of the
emissions are insufficient to derive an aerosol lifetime. However, for the
model simulation #1, the intermittent e-folding time for the full size
distribution of 137Cs during the first 15 days is 6.1 days, compared to
the 10 days found over the 45-day period in Table . This is due to
the reduction of below-cloud scavenging and dry removal events (shown in
Fig. ) combined with a reduction of in-cloud scavenging as well,
as after 15 days a large and increasing fraction of the left-over aerosol
particles reside above the cloud tops. As particles with more efficient
removal are lost, the lifetime is more and more influenced by the
longer-lived particles over time and thus the model e-folding lifetime
estimate increase with time. This last effect applies in FLEXPART only when
the aerosol size distribution consists of more than one specific aerosol kind
(i.e. modal size or different removal parameters).
In Fig. d the mean model/observed concentration ratios at the
different stations are plotted against latitude. A prominent feature of
FLEXPART and indeed most models used by is a tendency
to overpredict concentrations at low latitudes and underpredict
concentrations at high latitudes. This tendency is also present with the new
removal scheme, where model / observation ratios decrease with latitude.
The green line shows a logarithmic fit to the station median data. The same
fit was done to the mean from a simulation using FLEXPART version 9 (pink).
This shows that the new model, while still having a systematic latitudinal
dependence, represents a clear improvement over the old version. One possible
explanation of the decreasing model / observation ratios with latitude
might be that in-cloud scavenging in ice clouds is too effective. However,
sensitivity simulations where only INeff was reduced (not shown)
revealed that this change had only a small effect in further reducing the
latitudinal bias. One of the possible causes of this is the high proportion
of mixed phase clouds (77 %) which reduces the impact of the latitudinal
dependence of the frequency of ice-phase clouds after that much time for an
emission pulse. Another possibility is that cloud phase is not well captured
by the ECMWF model, as in many other models . It may also
be relevant that the clouds have on average higher cloud tops near the
equator, so that temperature and thus the mixing state of clouds does not
have a strong enough latitudinal dependence in the Northern Hemisphere at the
time of this simulation (March–May).
Northern Hemisphere vertical distribution of BC for eight different
settings of the removal parameters. Top left panel shows the concentrations
for the reference settings for BC. The other panels show results of the
sensitivity simulations (see Table for details). Five vertical
layers were used and the horizontal resolution is 0.5∘. The white
line is the latitudinal column burden for each simulation and the dashed
black line repeats the results for the reference BC simulation, for
comparison purposes.
Black carbon
It has been notoriously difficult to model BC accurately. For example, Arctic
seasonal variations and Arctic haze periods are not captured well in most
models . Some of this can be accredited to BC aerosol
undergoing stages of transformation after its release to the atmosphere from
a hydrophobic to a hydrophilic state e.g.,. The aerosol
ageing processes that would influence in-cloud scavenging are not readily
included in FLEXPART and the constant removal parameters cannot account for
this transformation. Therefore, several aerosol parameter combinations were
tested with FLEXPART both in backward and forward mode. There are
observations that urban BC is transformed very quickly into particles with
aged, hydrophylic characteristics . Therefore, a
representation resembling physical properties of aged BC (BC #1 in
Table ) was selected as our reference set-up for BC. Our
assumptions regarding the values of CCNeff and INeff
were based on the findings of that BC is much more
efficiently removed in liquid water clouds than in ice clouds.
showed that aerosol composed of mainly elemental carbon
had the highest fraction of non activated particles. A size distribution with
a modal mean diameter of 0.15 µm was assumed.
In addition to our simulations for our reference BC species, seven other
simulations were performed to test the sensitivity of model results at
different latitudes, altitudes and times of the year to changes in the
parameters describing the different removal mechanisms. For this, parameter
settings were varied within ranges thought to be suitable for BC.
Table summarizes the parameter choices for these simulations.
Column burdens and vertical distribution of the eight simulations are shown
in Fig. . The concentrations are FLEXPART output from five vertical
layers with upper borders of 100, 1000, 5000, 10 000 and 50 000 m.
The BC column burdens (shown with white lines in Fig. on the right
hand side y axis) are overall somewhat high when compared to other studies
e.g.,, with the exception of simulation #4,
which has strongly enhanced in-cloud removal. The dashed black line shown in
all the panels is the column burden of the reference simulation (#1).
All simulations produce a quite similar latitudinal distribution. The
strongest sources of BC are at mid latitudes and most of BC at high altitudes
is also found in this region for all simulations. Thus, the highest column
burdens are found near 35∘ N in all simulations. The two simulations
with reduced in-cloud scavenging (#2 and #8), have the highest column
burdens. While increasing Crain by a factor of 10 (simulation
#5) reduces the burden significantly, a similar, but an even stronger effect
can be achieved with a reduced aerosol size (simulation #3), as smaller
particles have higher dry deposition velocities. This shows that in the
absence of efficient wet removal, dry removal can be important as well.
Though it generally accounts for less than 10 % of total removal in our
simulations for particles with d<1 µm, in simulation #3 it
accounts for 48 % of the removal. Only simulations #5–#8, which have
phase dependent changes to removal parameters, produce burdens with a
noticeable different dependence on latitude when compared to simulation #1.
Annual average BC concentration in the lowest model layer
(0–100 m) for the reference simulation (top left) for the year 2007.
White circles mark the locations of the measurement stations used for model
comparisons. The other panels show the relative difference to this reference
version (in %) for the seven other simulations using parameter settings
from Table .
Aerosol specifications for the eight simulations done for BC. The
first four columns report the aerosol removal parameters used, the following
columns show the median concentration (ngm-3) at each station and
the last column reports the median of all modeled values. Italicised values:
the value for each station that is closest to the observed values (bottom
row), bold values: default coefficients, and underlined values: changed
parameters.
Coefficients
Annual median Concentration (ngm-3)
#
Crain
Csnow
CCNeff
INeff
MEL
SGP
PAL
BRW
ZEP
ALT
ALL
1
1.00
1.00
0.90
0.10
700.4
234.1
33.9
7.4
9.5
6.2
33.3
2
1.00
1.00
0.30
0.03
736.8
252.2
61.6
10.5
16.0
8.09
58.6
3*
1.00
1.00
0.30
0.03
713.2
245.2
45.4
8.5
9.4
7.5
45.4
4
1.00
1.00
9.00
1.00
428.8
113.6
4.8
< 0.01
< 0.01
< 0.01
1.0
5
10.0
1.00
0.90
0.10
615.6
194.1
20.8
3.4
4.4
3.3
22.0
6
1.00
0.00
0.90
0.10
690.85
232.6
36.0
8.9
10.3
6.9
40.3
7
1.00
1.00
0.60
0.60
673.2
219.8
30.6
6.0
8.9
6.3
31.5
8
1.00
1.00
0.45
0.10
727.2
244.2
37.8
8.5
10.2
7.5
41.3
Observed
366.9
211.6
36.35
17.8
11.8
19.8
19.8
* Aerosol diameter was reduced to 0.02 µm.
Annual average calculated BC concentrations in the surface layer
(0–100 m) in the northern hemisphere are shown in Fig.
for the reference simulation (top left) and as differences from this
reference for the other seven simulations. Overall, there are only small
differences between the various model runs in the major BC source regions,
where the concentrations are strongly influenced by local emissions. Further
away from the source regions, differences in removal have a stronger effect.
Simulation #4, with enhanced in-cloud scavenging in both liquid and ice
clouds, stands out with very low concentrations in the Arctic and other
remote regions. The remaining simulations have concentrations within
±50 %. It is worth noting that there are a few distinct geographical
features in Fig. . For example, turning off the below-cloud removal
by snow (simulation #6) only has a small effect that can be seen north of
60∘ N. In simulation #8, where liquid in-cloud removal is reduced,
modeled surface concentrations are increased in remote tropical areas.
Simulation #7, where the overall removal efficiency is maintained, but no
differentiation of cloud phase is made, illustrates the relative effect of
the cloud phase dependency of in-cloud removal.
The monthly measured (black) and modeled (blue; simulation #1 in
Table ) BC concentrations at six measurement stations are shown in
Fig. . The station locations are marked in Fig. and are
at different distances from major source areas. The aerosols measured at the
different stations thus have very different ages. For simulation #1, at
Melpitz the mean mass weighted FLEXPART aerosol age is 1.3 days, at Pallas it
is 3.8 days and at Zeppelin it is 7.7 days. The age is defined as the time it
takes for the aerosol to reach the station after its emission. The aerosol
age depends not only on the transport, but also on the removal between
emission and observation.
Modeled and observed monthly BC concentrations at six different
measurement stations for the reference BC simulation. The black boxes cover
the 25–75 % percentile range, the black horizontal line the median, and
the black whiskers the 10–90 % percentile range of the observations.
Modeled median values are plotted in blue with vertical lines showing the
25–75 % percentile range. The stippled blue line shows the model mean.
Also shown are the median values obtained from simulation #7.
Increased removal efficiency would, on average, reduce aged BC more than
fresh BC, resulting in a less aged aerosol population. Systematic differences
in model bias for stations close to and stations far away from source regions
can thus allow to separate errors in emissions versus errors in simulated
aerosol lifetimes. In Table the median modeled concentrations at
the six stations are reported for all the sensitivity simulations. Seven of
the eight simulations overestimate the concentrations at Melpitz by a factor
of almost 2, especially in summer (Fig. ). This suggests that local
emissions around Melpitz are too high, as changes in the removal
parametrisation have little effect on the concentrations (Table ).
Moving away from the source regions, stations Southern Great Plains and
Pallas have model concentrations close to the observed average for all the
simulations except for simulation #4 which underpredicts the concentration
at these two stations by a factor of 2.1 and 8, respectively. Annual mean BC
concentrations at the three Arctic stations Alert, Barrow and Zeppelin are
underpredicted by the model (mainly due to very low simulated summertime
concentrations, see Fig. ). This alone would indicate a too fast
removal and thus a too short BC lifetime. However, indicative of total global
removal rates, the column burden is, also for the Arctic, on the high side of
most current model estimates and therefore also
burden/emission estimates of the BC lifetime of 9.0 days is higher than in
many other models .
Observations at all stations except Southern Great Plains have a seasonal
cycle, with lowest concentrations during summer and higher concentrations
during winter. The Southern Great Plains station has a somewhat different
seasonality than the other stations, with a peak in autumn, and this is quite
well captured by the model. The four higher-latitude stations all show a
pronounced winter/spring peak, which is well reproduced by the model. .
Top panel: age of FLEXPART BC aerosol for simulation #1 at Zeppelin
every 6 h (blue) and smoothed with a 48 h running mean (red) for the
year 2007. The black line shows the annual mean age of 7.7 days. Bottom
panel: observed (black) BC concentrations at Zeppelin and the simulated
FLEXPART concentrations for simulation #1 and FLEXPART version 9 (blue and
red respectively). All data are smoothed with a 48 h running
mean.
In Fig. (bottom panel) a comparison between the observations and
model simulation #1 and a simulation using FLEXPART v9 is shown as a 48 h
moving average. With a Pearson's squared correlation coefficient of r2=0.44, simulation #1 captures nearly half of the variability of the
observations with generally higher concentrations during December–May, and
large peaks in the observations in January and December. There are noticeable
differences between the two simulations, but not all of them are due to wet
removal as FLEXPART v10 includes also other changes than the removal. Also,
the concentration simulated using v9 is a point estimate from a backward
simulation and simulation #1 a (1∘×1∘) grid average
from a forward simulation, so they are not directly comparable. Of most
significance however is the higher concentrations in the spring months
April–May, where simulation #1 capture the observed high levels of BC and
the v9 does not. On average for the year, v10 concentrations are about twice
as high as the v9 data with annual median (9.5 and 6.8 µg),
median (9.5 and 6.8 µg) and mean (47.6 and 21.1 µg)
values for the two respectively.
FLEXPART aerosol age at Zeppelin was also used to examine the role of the
removal processes in the variability. showed observed
concentrations of aerosol submicron mass had a strong dependence on
trajectory accumulated precipitation. Shown in the top panel in
Fig. is the mean model age corresponding to 6-hourly observations.
Also a smoothed 48 h fit is shown in red. Depending on the season,
the youngest BC aerosols are found in combination with either the highest
observed concentrations (winter) or very low concentrations (summer) and have
two different explanations. The high aerosol episodes with low age in winter
(e.g., peak on 15 January) are related to fast transport from the Yamal
Peninsula and the Kara Sea. In this area there are large emissions from the
gas and oil industry , and if these emissions are
transported quickly and nearly without removal to Zeppelin, concentrations
there increase strongly. In summer, there is a persistent background of
relatively low aerosol concentrations. Occasionally, this background is
reduced further by scavenging events occurring close to the station. This
removal only leaves BC from the small local sources of BC on Svalbard,
leading to both a low age and low concentrations of the simulated BC.
Summary and conclusions
This paper has presented the new FLEXPART aerosol wet scavenging scheme.
Firstly, a more realistic distribution of clouds was achieved by
incorporating three-dimensional cloud information from ECMWF. Secondly, a new
parameterization of wet removal within and below clouds was introduced,
considering also the water phase of the clouds and the precipitation type.
Reading of cloud liquid and ice water data from stored ECMWF data leads to
fewer inconsistencies with the ECMWF precipitation data than using the old
relative humidity-based cloud scheme, and is an important improvement of
FLEXPART. Using the ECMWF cloud water data, we diagnosed the frequency of
different types of removal events, and we found a dominance of in-cloud
scavenging events (91 % of all events) above 1000 m. At lower
altitudes than 1000 m, below-cloud scavenging events are slightly
more important (52 % of all events) than in-cloud scavenging events.
We performed model simulations for three different types of aerosols (mineral
dust, 137Cs attached to sulfate, and BC), to test different aspects of
the removal scheme. For each of these aerosol types, we performed sensitivity
simulations to explore the size dependence of the aerosol removal, to
determine atmospheric e-folding times, and to investigate the water phase
dependency of the aerosol removal scheme. We also compared simulation results
to observations of 137Cs and BC.
For both mineral dust and 137Cs simulations, the aerosol lifetime had a
maximum in the accumulation mode of 31.8 and 11.7 days, respectively. For the
BC particles, which are also in the accumulation mode, an e-folding
lifetime of 16 days was found. These lifetimes are long compared to lifetimes
typically reported in the literature. However, this can be explained by
differences in the definition of lifetime see discussion
in. For instance, estimating the lifetime by dividing the
aerosol burden with its emission rate – a common definition of lifetime used
by global aerosol modelers – results in a BC lifetime of 9 days. This is
quite comparable to lifetime values reported for BC in the literature, though
perhaps still somewhat longer than in most models
.
In our scheme, the lifetime of accumulation mode particles is controlled
mainly by in-cloud removal, as dry removal and below-cloud scavenging are
inefficient for these aerosol particle sizes. Therefore, the longer
e-folding lifetime is due mainly to transport of aerosols above clouds,
where they cannot be scavenged. This process is less important for the
burden/emission lifetime estimate, which depends mostly on the particles'
removal rate in the first few days after the emission.
Simulations for the accumulation mode particles with FLEXPART are highly
sensitive to the choice of CCNeff and INeff values,
which describe the particles' efficiency to serve as cloud condensation and
ice nuclei. Overall, it was found that the sum of CCNeff+INeff is more important for the removal efficiency than the
individual choice of values for CCNeff or INeff. For
the three aerosol types, we recommend the following values: Regarding
insoluble aerosols, found good agreement between
modeled and observed concentrations when using CCNeff=0.15 nd
INeff=0.02 for mineral dust. For BC, CCNeff=0.9
and INeff=0.1 gave the overall best results, and these values
are also comparable with what was found by . Soluble aerosol
(137Cs) concentrations compared best with CCNeff=0.9 and
INeff=0.9. The latter value is somewhat higher than
INeff values suggested by measurements of
e.g..
On the other hand, for particles larger than 1 µm, both
below-cloud scavenging and dry removal have a strong impact on the lifetime.
Consequently, these larger aerosols all have much shorter e-folding times.
CCN (and IN) efficiency has also been shown to increase with aerosol particle
size, thus contributing to the shorter lifetimes of particles larger than
1 µm. However, as their lifetime is mostly controlled by
gravitational settling and below-cloud scavenging, the choice of values for
INeff and CCNeff is not particularly critical for
super-micronic particles.
There are large uncertainties tied to the efficiency of impaction scavenging.
Nevertheless, the chosen schemes of and
capture the overall size dependence predicted by
impaction theory. For BC, for which the removal by snow is generally more
efficient (especially at low precipitation intensities), taking into account
the precipitation water phase leads to relatively stronger removal of BC at
high latitudes and so enhances the underestimation of BC concentrations at
the Arctic stations. This effect is however small compared to the aerosol
size dependence of below-cloud scavenging.
Despite all efforts to explore and correct this issue in FLEXPART, there is
still a tendency to underpredict BC measurements in the Arctic. Similarly, in
simulations of 137Cs from the Fukushima accident there is a latitudinal
gradient in model bias, with underprediction of observations at high
latitudes. For BC, assuming a larger efficiency of the particles to serve as
CCN than as IN reduced the Arctic underprediction and also produced a
seasonal cycle of BC concentrations that is closer to the observed one,
compared to simulations assuming equal CCN and IN efficiency. For 137Cs,
however, only a small improvement in the latitudinal gradient of model bias
was found. A reason for this may be that in the large fraction of clouds
defined as mixed phase (77 %), the Bergeron–Findeisen effect, as
represented in Fig. , may not be sufficiently strong.
Though there are limitations to the level of sophistication possible for
aerosol removal in linear Lagrangian models, the wet removal scheme
introduced in FLEXPART is capable of distinguishing and treating most aspects
of wet removal for aerosols with many different characteristics. Our results
show that the new scheme produces aerosol lifetimes and concentrations that
are realistic when compared with observations.