AMTAtmospheric Measurement TechniquesAMTAtmos. Meas. Tech.1867-8548Copernicus PublicationsGöttingen, Germany10.5194/amt-9-5535-2016Aerosol data assimilation in the chemical transport model MOCAGE during the TRAQA/ChArMEx campaign: aerosol optical depthSičBojanbojanovmejl@gmail.comEl AmraouiLaazizPiacentiniAndreaMarécalVirginieEmiliEmanueleCariolleDanielPratherMichaelhttps://orcid.org/0000-0002-9442-8109AttiéJean-LucCNRM-GAME, Météo-France – CNRS, UMR3589, Toulouse, FranceCECI, CERFACS – CNRS, UMR5318, Toulouse, FranceDepartment of Earth System Science, University of California, Irvine, USALaboratoire d'Aérologie, University of Toulouse – CNRS, UMR5560, Toulouse, FranceBojan Sič (bojanovmejl@gmail.com)22November20169115535555426February20167April201620September201622September2016This work is licensed under a Creative Commons Attribution 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by/3.0/This article is available from https://amt.copernicus.org/articles/9/5535/2016/amt-9-5535-2016.htmlThe full text article is available as a PDF file from https://amt.copernicus.org/articles/9/5535/2016/amt-9-5535-2016.pdf
In this study, we describe the development of the aerosol optical depth (AOD)
assimilation module in the chemistry transport model (CTM) MOCAGE
(Modèle de Chimie Atmosphérique à Grande Echelle).
Our goal is to assimilate the spatially averaged 2-D column AOD data from the
National Aeronautics and Space Administration (NASA) Moderate-resolution
Imaging Spectroradiometer (MODIS) instrument, and to estimate improvements in
a 3-D CTM assimilation run compared to a direct model run. Our assimilation
system uses 3-D-FGAT (first guess at appropriate time) as an assimilation
method and the total 3-D aerosol concentration as a control variable. In
order to have an extensive validation dataset, we carried out our experiment in the
northern summer of 2012 when the pre-ChArMEx (CHemistry and AeRosol
MEditerranean EXperiment) field campaign TRAQA (TRAnsport à longue distance et Qualité de l'Air dans le bassin méditerranéen)
took place in the western Mediterranean basin. The assimilated model run is
evaluated independently against a range of aerosol properties (2-D and 3-D)
measured by in situ instruments (the TRAQA size-resolved balloon and aircraft
measurements), the satellite Spinning Enhanced Visible and InfraRed Imager
(SEVIRI) instrument and ground-based instruments from the Aerosol Robotic
Network (AERONET) network. The evaluation demonstrates that the AOD
assimilation greatly improves aerosol representation in the model. For
example, the comparison of the direct and the assimilated model run with
AERONET data shows that the assimilation increased the correlation (from 0.74
to 0.88), and reduced the bias (from 0.050 to 0.006) and the root mean square
error in the AOD (from 0.12 to 0.07). When compared to the 3-D concentration
data obtained by the in situ aircraft and balloon measurements, the
assimilation consistently improves the model output. The best results as
expected occur when the shape of the vertical profile is correctly simulated
by the direct model. We also examine how the assimilation can influence the
modelled aerosol vertical distribution. The results show that a 2-D
continuous AOD assimilation can improve the 3-D vertical profile, as a result
of differential horizontal transport of aerosols in the model.
Introduction
In recent years, the role of aerosols in the climate system has been better
determined . As a consequence, efforts to accurately represent
aerosols in models also increased for
example.
The development of aerosol modelling enables us to better understand how
aerosols affect atmospheric chemistry, air quality, climate, aviation,
visibility, radiative budget and clouds. Still, the complexity of the
processes governing aerosol physics and chemistry has led to a large
diversity of parametrizations which with other uncertainties (e.g. dynamics,
emissions, initial conditions) produce large differences in the aerosol model
results .
At the same time, the number and quality of aerosol observations has also
increased through advances in sensing technology and techniques. The last
decade of aerosol research has brought on more accurate measurements of more
specific aerosol characteristics observed from local to global scales and
over long periods of time . In essence, we are in a golden
age of aerosol data, but still looking for approaches to merge all the
disparate measurements and synthesize that knowledge .
Observations are crucial in identifying aerosol properties and processes,
which in turn help build more accurate models. Additionally, with data assimilation
techniques, we are able to directly integrate observations in models in order
to improve modelled fields . Up to now, several
research groups made efforts to assimilate aerosols in the models. These
efforts are mainly focused on assimilating satellite data, usually aerosol
optical depth (AOD), since satellites provide continuous aerosol observations
on the global scale and yield a large number of individual observations, which
is desirable for assimilation systems. Many studies have used variational data
assimilation techniques. For example, a 3-D-VAR system for assimilating AOD
data was built by in the Naval Research Laboratory's (NRL)
Aerosol Analysis and Prediction System (NAAPS), by in the
Chinese Unified Atmospheric Chemistry Environment – Dust (CUACE/Dust)
system, and by in the National Centers For Environmental
Prediction's (NCEP) Weather Research and Forecasting-Chemistry/Goddard
Chemistry Aerosol Radiation and Transport/Gridpoint Statistical Interpolation
(WRF-Chem/GOCART/GSI) system. described the
assimilation of AOD in the European Centre for Medium-Range Weather Forecasts
(ECMWF) Integrated Forecasting System (IFS) using the 4-D-VAR method, while
used the same approach to assimilate satellite lidar
profiles in the RAMS/CFORS-4-D-VAR (RC4) model. Sequential assimilation
approaches are also documented: optimal interpolation used by, for example,
and in the Model of Atmospheric
Transport and Chemist (MATCH) and in the Polyphemus
system; or the ensemble Kalman filter used by, for example,
in the Model of Aerosol Species in the Global Atmosphere
(MASINGAR), in the SPRINTARS model,
in the WRF-Chem model, in the
Non-hydrostatic ICosahedral Atmospheric Model (NICAM) and
in the Ensemble Navy Aerosol Analysis Prediction System/Data Assimilation
Research Testbed (ENAAPS-DART) system.
Bin ranges of individual primary aerosol species present in MOCAGE.
Bin 1Bin 2Bin 3Bin 4Bin 5Bin 6Desert dust (µm)0.1–11–2.52.5–55–1010–3030–100Sea salt (µm)0.003–0.130.13–0.30.3–11–2.52.5–1010–20Black carbon (µm)0.0001–0.0010.001–0.0030.003–0.20.2–11–2.52.5–10Organic carbon (µm)0.0005–0.0030.003–0.10.1–0.30.3–11–2.52.5–10
In this study, we describe the development of the AOD assimilation module in
the chemistry transport model (CTM) MOCAGE e.g.Modèle de Chimie Atmosphérique à Grande Echelle. Our goal is to assimilate the regular daily
global mapping of the 2-D column AOD by the National Aeronautics and Space
Administration (NASA) Terra and Aqua satellites with the
3-D CTM modelling of the major tropospheric aerosols. We use the variational
3-D-FGAT (first guess at appropriate time) method , which
is implemented in the CTM using the PALM coupler Projet d'Assimilation par Logiciel Multi-Méthodes.
The assimilated fields can then be evaluated independently against a large
range of aerosol properties (2-D and 3-D) measured by other satellites and
in situ and ground-based instruments. We will estimate improvements in the
CTM modelling of all aerosol types achieved by the AOD assimilation compared
to a direct (unassimilated) model run. One focus will be on the potential to
improve air quality forecasting. Another will be how the continuous
multicycle AOD assimilation can influence the modelled aerosol vertical
distribution. To obtain an extensive validation dataset, we choose the 2012
summer field campaign of TRAQA (TRAnsport à longue distance et Qualité de l'Air dans le bassin méditerranéen) and centre our
modelling on the western Mediterranean basin. The TRAQA campaign was a
pre-ChArMEx (CHemistry and AeRosol MEditerranean EXperiment) experiment with
the objective to characterize the air quality in the western Mediterranean
basin . The validation data not only include the TRAQA
measurements (in situ aircraft and balloon measurements of size-resolved and
speciated aerosols), but also different remote-sensed AOD observations from
the ground (Aerosol Robotic Network, AERONET; ) and
satellite (Spinning Enhanced Visible and InfraRed Imager, SEVIRI; ).
This paper is organized as follows. In Sect. 2 we describe the MOCAGE CTM and
its aerosol module, in Sect. 3 the data assimilation system and in Sect. 4
and Sect. 5, respectively, the assimilated and the independent observational
data. Results from the assimilation model and its critical evaluation against
the direct forecasts of the CTM with independent data are presented in
Sect. 6. In Sect. 7 we discuss the overall performance of our assimilation
system, and in Sect. 8 our recommendations for future work.
Model description
MOCAGE is a global CTM developed in Météo-France. It serves as an
operational air quality model and simulates gases
and primary (directly emitted) aerosols
. It transports atmospheric species by a
semi-Lagrangian advection scheme . Turbulent diffusion
is implemented following , and convection is implemented following
. The dynamics within the CTM are forced by
meteorological analysis fields (pressure, winds, temperature, specific
humidity) from ARPEGE, the operational numerical weather prediction model of
Météo-France. MOCAGE has 47 vertical hybrid sigma-pressure levels
from the surface up to 5 hPa. The vertical resolution varies with
altitude, with a resolution of 40 m in the planetary boundary layer,
about 400 m in the free troposphere and about 700–800 m in
the upper troposphere and lower stratosphere. The version of the model used
in this study is described in , and has been evaluated with a
range of different remote-sensed and in situ measurements for cases and
regions relevant to this study.
The model can include nested domains over smaller regions. In this study, the
model is run in a two-domain configuration with a global grid of 2∘×2∘ and a smaller nested domain (MEDI02) with a grid of
0.2∘×0.2∘ over the Mediterranean basin and the Sahara. The MEDI02 domain, where we assimilate AOD data, has boundaries
20∘ W–40∘ E, 16–52∘ N. The lateral boundary
conditions for aerosols are provided by the global domain.
Aerosols in MOCAGE consist of externally mixed primary aerosol species.
Implemented species for this study are desert dust, sea salt, black carbon
(BC) and organic carbon (OC). The particle size distribution for each type is
divided into six size bins, characterized by the particle average diameter and
mass. The size ranges of bins for all considered aerosol species are shown in
Table . Each aerosol bin is then treated as a passive tracer:
aerosols are emitted, transported and removed from the atmosphere; however
there are no transformations or chemical reactions between aerosol types,
between size bins or with gases. The aerosol dry deposition scheme is
described in . The sedimentation is implemented as described
in . For the wet deposition, the model uses
for the implementation of the in-cloud scavenging, and
for the rain and snowfall below-cloud scavenging.
The emission inventories for BC and OC are prepared as follows. The
anthropogenic component comes from for both domains
(Global and MEDI02). This inventory is defined monthly, and harmonized for
the year 2000 . Biomass burning (BB) sources of BC and OC
aerosols are introduced into the model with a daily frequency from the Global
Fire Assimilation System (GFAS) version 1.1 . The GFAS
assimilates the fire radiative power observed by MODIS, corrects the cloud
cover gaps, filters anthropogenic and volcanic activities and finally
calculates daily biomass burning aerosol emissions for BC and OC. We use
daily BB emissions for better synoptic forecasts, which is not possible with
the monthly mean emissions of .
Sea salt particles are emitted using the semi-empirical source function from
, which includes a particle size, wind speed, and sea
surface water temperature dependence. Desert dust aerosols are emitted by a
dynamical online scheme, which depends on the wind intensity and surface
characteristics. The scheme is based on . It covers
Africa, Arabia and the Middle East (13–36∘ N,
17∘ W–77∘ E), where input soil properties and
aerodynamical surface parameters have a resolution of 1∘×1∘, and north-eastern Asia
(35.5–47∘ N, 73–125∘ E) with the input parameter
resolution of 0.25∘×0.25∘.
Description of data assimilation system
The data assimilation system used in this study is MOCAGE-Valentina,
developed jointly by Météo-France and CERFACS (Centre Européen de
Recherche et de Formation Avancée en Calcul Scientifique). The
assimilation algorithm used is 3-D-FGAT 3-D first guess at
appropriate time;, which is a compromise between
the 3-D-Var and 4-D-Var methods. Observations are taken at their exact times
to the nearest minute; i.e. every measurement is compared with the background
at the time of measurement, as in 4-D-Var. The optimal analysis is estimated
only for a specified moment in the assimilation cycle, as in 3-D-Var, and not
for the whole trajectory, as in 4-D-Var. Thus, compared to 4-D-VAR, during
the assimilation the information given by observations is not propagated in
time, and consequently we do not need the linearized operator of the model
evolution and its adjoint .
The goal of the variational assimilation process is to minimize the cost
function, whose incremental form in 3-D-FGAT is
J(δx)=Jb(δx)+Jo(δx)=12δxTB-1δx+12∑i=0N(di-Hiδx)TRi-1(di-Hiδx),
where Jb is a part of the cost function related to the background;
Jo is a part of the cost function related to the observations;
δx=x-xb is the misfit between the background xb and
the state of the system x; di=yi-Hixb(ti) is the innovation
and represents the distance of the observation yi from the background
xb at time ti; Hi is the non-linear observation operator;
H is its linearized version (the tangent-linear); B is
the background error covariance matrix; and Ri is the observation
error covariance matrix at time ti. The matrices B and
Ri influence the weighting of the terms Jb and
Jo.
To find the optimal solution we minimize the cost function J by computing
its gradient:
∇J(δx)=B-1δx+∑i=0NHiTRi-1(di-Hiδx).
After estimating the analysis increment δxa, we add it to the
aerosol abundance at the beginning of the cycle. The model is then run over a
cycle length (1 h) to obtain the analysed trajectory. Its endpoint is used
as the initial background field for the next cycle.
Preconditioning
MOCAGE-Valentina uses the incremental form of 3-D-FGAT (Eq. ). In
order to minimize the cost function more efficiently and to improve
convergence, the increment δx is preconditioned by
v=B-12δx.
In this way the cost function becomes
J(x)=12vTv+12∑i=1N(di-HiB12v)TRi-1(di-HiB12v),
and its gradient is
∇J(δx)=v+(B12)T∑i=1NHiTRi-1(di-HiB12v).
In this formulation, there is no need for an explicit specification of the
inverse matrix B-1, and the preconditioning reduces the number
of iterations . In MOCAGE-Valentina, the cost function is
minimized using the limited-memory BFGS (Broyden–Fletcher–Goldfarb–Shanno)
method .
The minimization of the cost function with the preconditioned form gives an
increment of the analysis in the space of the variable v, which after the
minimization is converted back to model space:
δx=B12v.
More details on the assimilation algorithm are provided by
and .
The control variable
For aerosols, the modelled prognostic variables (i.e. the 3-D concentration
of aerosols of different composition) and the observations (i.e. the column
optical depth summed over all aerosols at visible wavelengths such as
550 nm AOD) are usually not the same
physical quantity. In the case of AOD observations, to define which variable
will be minimized, different choices are possible. A straightforward choice
is to use a prognostic variable as a control variable, as implemented by
. Following this approach in our system, the control variable
would correspond to a 4-D variable containing the 3-D fields of all 24
aerosol bins. The matrix B would have to include the variances and
covariances of all bins separately. This could be difficult to define, but
the analysis would be partitioned automatically into all bins by the system.
made the choice to use the 3-D total aerosol
concentration as the control variable. This makes the control variable
smaller, corresponding to a 3-D variable, where all bins are merged into a
single one. Considering the characteristics of our system, we decided to
follow the same approach as in . Compared to the
approach, the problem of minimization of the cost function is
better determined: the 2-D AOD observations constrain one 3-D variable as the
unknown, compared to 24 3-D variables as unknowns in the
approach. The matrix B does not have to be defined for all bins
separately, nor does it need to contain the inter-bin covariances.
Nevertheless, in order to linearize the observation operator, it is necessary
to decide how the analysis increment δxa will influence each
bin. The increment could be weighted by different quantities, like number or
mass concentration, or extinction coefficient. The real contribution of
different aerosol types in the increment is unknown, and we can only rely on
the model information. In this way, all repartition weights based on the
model stay biased in a similar way compared to the real repartition weights,
regardless of the choice of the repartition. Considering all possible choices
and the characteristics of our system, we decided to keep the
relative mass contributions of all bins constant (24 bins including 6 size bins of all
4 aerosol types). After the analysis increment is calculated, it is
repartitioned to the different bins in the model according to their
background fractions of the total aerosol mass.
The observation operator
The assimilation of AOD in MOCAGE-Valentina requires the development of an
observation operator (H), which transforms the control variable from the
model space, i.e. the total 3-D concentration, into the observation space,
i.e. AOD. The AOD (τ) is calculated by taking into account bin number
concentrations (nbin) of all bins (types included) at a certain
model level and the optical properties of individual species calculated by
the Mie code:
Hx=∑bin∑levCext(Dp,ñ,λ)Δzlevnbin=τ,
where Δzlev is the vertical thickness of the model level
“lev” (m), and Cext is the extinction cross-section
(m-1). To calculate Cext in the model we use Wiscombe's
Mie code scheme for spherical homogeneous particles
and aerosol refractive indices from the
Global Aerosol Data Set/Optical Properties of Aerosols and Clouds
GADS/OPAC, and . We
also take into account the hygroscopicity of sea salt aerosols.
The tangent-linear operator (H=∂H∂xi)
is a linearized version of the non-linear observation operator H around the
system state x, and it consists of partial derivatives of H with respect
to all input variables. It gives a first-order approximation (δτ)
of the difference between the unperturbed (Hx) and the perturbed results
(H(x+Δx)) of the non-linear operator H.
The tangent linear operator can be derived explicitly by the
finite-difference method, but it is a computationally expensive method.
Instead, the tangent-linear operator can be considered as a sequence of
linearized sub-operators of the non-linear observation operator and built
piece by piece by differentiating each line of code or loop separately. This
approach allows the parts of the code to be tested separately. In order to provide
a linearized trajectory around the model state x, the tangent linear
operator has to satisfy
limδx→0H(x+δx)-HxHδx=1.
As long as the perturbation δx is small enough that the perturbed
state x+δx stays close to the model state x, the test will give a
value close to 1.
The adjoint operator (HT) is the transpose of the
tangent-linear operator, and it satisfies
Hx,y=x,HTy,
where x,y represents the inner product of x
and y. Analogously to the tangent linear operator, the adjoint operator
can be considered as a sequence of operators. Each discrete operation in the
tangent linear operator has a corresponding operation in the adjoint
operator, but the order of execution is reversed.
The error covariance matrices
The background error covariance matrix is a key component of the data
assimilation system. It defines the background errors and the spatial
structure of the analysis. The background error covariance matrix
B is a matrix of size j×j, where j is the size of the
control variable. It can be represented as
B=ΞCΞT,
where Ξ is the diagonal matrix of the square root of the
variances, and C is the positive definite symmetric matrix of
correlations. In the case of the preconditioned cost function, the matrix is
formulated as
B12=ΞC12,
where C12 is the square root of the matrix
C. Because neither enough information is available to explicitly
estimate all correlation members, nor is there enough memory to store them,
the matrix B is modelled as an operator. To estimate the product
of the matrix B and a vector, MOCAGE-Valentina uses the
integration of a generalized diffusion-type equation in a reduced space
.
The observation error covariance defines the observation and
representativeness errors. In our study we neglect the observation error
correlations, which means that all non-diagonal members (covariances) in the
matrix R are zero. The matrix R is reduced to its
diagonal with the variances of measurements:
R=Dy=diag(σobs2).
The background and observation error variances, located along the diagonal of
B and R, influence the weight of the model and
observations in the cost function. In this study, we specified them as a
percentage of the first guess field for the background error variance and a
percentage of each measured AOD for the observation error variance. From the
published MODIS uncertainties we estimated the percentage for the errors of
the (“super-”)observations to be 12 %. Further, we used the χ2
diagnostics to estimate optimal values for the B error variances
. The χ2 test is an a posteriori
diagnostic, which allows us to check that the errors are properly specified. It
checks whether, for each assimilation window, it is true that
E2Jminp∼1,
where E is the expectation (statistical average), Jmin is the value
of the cost function at the minimum and p is the number of observations.
For this test, it is necessary to run the assimilation system for a prolonged
period of time and, in the case of sufficient number of observations, the
matrix B will not depend on its initial value any more. Because
this method is computationally expensive, a very rigorous optimization of the
errors is difficult to do. Therefore, we carried out several test runs to
determine optimal parameters for the matrices a posteriori. As the optimal
parameters, we estimated that the percentage for the errors of the model
should be twice as large as for the observations (24 and 12 %
respectively).
Missing secondary aerosols in MOCAGE are considered as a possible
underestimation of AOD in the model, and this is taken into account in the
covariance matrices error definitions. Therefore, the possibly smaller AOD
values of the background are compensated by a higher percentage for the error
of the model. In addition, it is considered that it is better to have overestimated
errors in the matrix B than vice versa .
Covariances of the background error matrix, which influence the spread of the
analysis to neighbouring grid boxes, are specified with constant correlation
lengths in the horizontal and the vertical. The constant and homogeneous
correlation lengths are modelled using a Gaussian function
in terms of distance in kilometres for the horizontal
lengths, and in terms of number of model levels for the vertical lengths
. The implemented horizontal correlation length is
45km. For the vertical correlation, with the column-integrated
observations, there is no explicit need for vertical correlation elements.
However, in the matrix B preconditioned cost function (Eq. )
it is advantageous not to have null vertical correlation lengths. Therefore,
we apply a vertical correlation of one model level. The type of the
correlation field influences the method by which the generalized
diffusion-type equation is solved. In the case of constant correlation
lengths in the limited-area domain, the equation is solved by the
finite-difference method.
Assimilated observations
The MODIS (Moderate-resolution Imaging Spectroradiometer) instruments observe
atmospheric aerosols on board Terra (since 2000) and Aqua (since 2002) from
complementary sun-synchronous orbits. The Terra overpass time is around 10:30
local solar time at the Equator in its descending (daytime) node, and the
Aqua overpass time is around 13:30 local solar time at the Equator in the
ascending node. We use MODIS Aerosol Optical Depth Collection C051 retrievals at
550 nm from Terra and Aqua, the ocean product retrieved with the
“best solution” and the reflectance-corrected land product. Their predicted
uncertainties are Δτ=±(0.03+0.05τ) over oceans and Δτ=±(0.05+0.15τ) over land . Over bright desert
areas, we use the “Deep Blue” MODIS product . For the
assimilation, we only considered the best quality data, with the highest
possible quality flag.
MODIS L2 resolution of 10km×10km is approximately
2 times finer than the model resolution of 0.2∘×0.2∘ over the control MEDI02 domain in which the assimilation is
performed. We have no way of treating two separate AOD values within a model
grid cell at about the same time, and so just average all observations in each
grid cell that occur on the same swath making so-called
super-observations;. Modelled and observed AODs are
then on the same 2-D grid, and the maximal number of observations per 1 h
slot over the whole domain is reduced in this way from ≈ 80 000 to
≈ 15 000. MODIS data from the Terra and Aqua platforms are
separated in time, except at high latitudes (not used here), and all AODs are
binned in the MEDI02 grid at 5 min intervals.
Independent observations for evaluationSEVIRI
SEVIRI (Spinning Enhanced Visible and InfraRed Imager) geostationary
observations over oceans are retrieved at 550 nm by
. The ICARE data centre operationally implemented this
algorithm and makes AOD data available on its website
(www.icare.univ-lille1.fr). This product was evaluated against other
satellite products and AERONET measurements by and
. The instrument makes an image of the whole Earth disk
every 15 min as seen from the equatorial geostationary orbit and at a longitude of 0∘. We sample the SEVIRI AOD every hour, and only use data
over water, since the retrieval over dark surfaces is usually more accurate.
The nadir horizontal resolution is 3 km, while over Europe it is
≈5km, and we average the SEVIRI AOD data that fall within
the same modelled grid box.
AERONET
AERONET (Aerosol Robotic Network) measures ground-based AOD from hundreds of
automated stations in the world with an accuracy of ±0.01 for a range of
wavelengths . We use all available L2 data from different
stations, and interpolate these data in logarithmic space to 550 nm (to
harmonize wavelengths between different stations and with the model) by using
available neighbouring wavelengths: 440, 500, 675 and 870 nm.
In situ observations from the TRAQA campaign
TRAQA (TRAnsport à longue distance et Qualité de l'Air dans le bassin méditerranéen) was a scientific project including a
measurement campaign intended as a pre-ChArMEx (CHemistry and AeRosol
MEditerranean EXperiment) experiment (http://charmex.lsce.ipsl.fr). It
took place over the north-western Mediterranean basin during the northern
summer of 2012. The main objectives of TRAQA were studies of transport, ageing
and mixing of the polluted air masses in and around the Mediterranean basin
and their impact on air quality. From 26 June to 11 July seven intensive
observation periods were conducted with the ATR-42 aircraft of
Météo-France; atmospheric balloons (sounding and drifting) and ground
instruments measured trace gases and aerosols. During the campaign, a desert
dust outbreak from Africa transported aerosols to the Mediterranean basin. It
was well observed around 29 June with several different instruments
.
PCASP
In our study, we use the data measured by the passive cavity aerosol
spectrometer probe (PCASP), which was on board the ATR-42 aircraft. The PCASP
measures the aerosol concentration and the aerosol size distribution with its
30 channels . PCASP measures particles with diameters
from 0.1 to 3µm, with channel ranges, calibration methods
and errors reported by . The reported Poisson counting error
is in the range 5–15 %. The high-frequency instrument data are averaged
into 1 min intervals for model comparison, with a horizontal resolution of
about 8 km at cruise speed.
LOAC
Finally, we also use data from LOAC (Light Optical Particle Counter)
instruments collected during TRAQA. LOAC is a light
aerosol optical counter measuring aerosol number concentration in 20 size
classes within a diameter range from 2 to 100µm in the
version of the instrument used in the TRAQA campaign. The uncertainty for
total concentration measurements is ±20 % when concentrations are
higher than 1cm-3 and up to about ±60 % when
concentrations are smaller than 10-2cm-3. LOAC uses the
technique of measuring aerosol at two scattering angles
. During TRAQA, LOAC was mounted on
meteorological sounding balloons, where its vertical resolution depends on
measurement frequency. The processed data that we use in our analysis have a
vertical resolution in the troposphere of about 0.3 to 0.4km,
which is similar to the model resolution.
Results
We run two MOCAGE configurations, one with and one without assimilation. The
assimilation is performed in the regional MEDI02 domain. The simulation
without assimilation is referred to as the direct model run and the
simulation with assimilation as the assimilation model run. The
simulated period for which we evaluate the model performance is for 19 days
from 25 June until 13 July 2012. The model is run after a spin-up period of
45 days, where the last 10 days for the assimilation model run are the
so-called assimilation spin-up. The assimilation cycles in the experiment
have a length of 1 h. The cost function is minimized until the convergence
criterion is reached, or when the maximum number of iterations of 200 is
reached. The analysis increment is added at the beginning of each
assimilation cycle.
Performance of the assimilation
Scatter plots of aerosol optical depths (where colours represent the
number of points) of assimilated MODIS observations and the 1 h
assimilation forecast (first guess) (a), and the assimilation
analysis (b) for each 1 h assimilation window (both runs start
from the same assimilated conditions of 1 h before). In each panel,
correlation (ρ), absolute bias (Δ), root mean square error
(RMSE) and standard deviation (σ) are noted. The assimilated data
correspond to the period of the TRAQA campaign from 25 June until 13 July
2012, and cover the MEDI02 domain.
The aerosol optical depth over Europe on 29 June 2012 at 12:00 UT,
(top left) simulated in MOCAGE by the model direct run; (top right)
simulated in MOCAGE by the MODIS assimilation model run; (bottom left)
observed by MODIS (Aqua + Terra) and used for assimilation in MOCAGE
(shown observations are collected during the whole day, and not only at
12:00); and (bottom right) observed by SEVIRI, which serves as an independent
dataset. The colours from white to red represent AOD from low to high
values.
In Fig. we evaluate the impact of the data assimilation on the
modelled fields (the 1 h assimilation forecast for the each 1 h
assimilation cycle and the corresponding assimilation analysis) by comparing
each of them directly with the observations which we assimilated, and this
should be considered as a “sanity” check of the system. The figure
represents the performance of the assimilation system and its ability to move
the modelled field closer to the observed values. The assimilated model can
more readily lower the overestimated values than elevate the underestimated
values. The statistics of the scatter plots against assimilated observations
show increased correlation (from 0.54 for the modelled fields to 0.82
for the assimilated fields), lowered root mean square error (from 0.27 to
0.18) and lowered standard deviation (from 0.27 to 0.17), demonstrating
what we expect from assimilation. Similar values of the root mean square
error and the standard deviation show the absence of important biases in the
system.
Hourly time series of aerosol optical depth at 550 nm of
SEVIRI data (black), the direct model (blue) and the assimilation model run
(red) over the western Mediterranean (0–10∘ E, 35–45∘ N)
for the period of the TRAQA campaign from 25 June until 13 July 2012. The
considered region is also marked in Fig. by the grey box.
Correlation, bias and root mean square error for both the direct model and
the assimilation model run as compared to SEVIRI data are given in Table 2.
Comparison with SEVIRI
The period of the TRAQA campaign is marked by two desert dust events coming
from Africa. In the western Mediterranean basin, where the campaign took
place, this produces elevated concentrations of desert dust aerosols from the
Sahara desert. Figure shows the desert dust event in the
Mediterranean basin on a particular day (29 June) seen by the direct model
run, the assimilation model run and the MODIS and SEVIRI instruments, and it
illustrates the impact of the assimilation on the modelled field. We look at
a longer time period in Fig. , which represents the time series of
the direct and assimilation model runs and the independent SEVIRI
observations over the western Mediterranean basin. Table 2 shows the
statistics corresponding to this figure. The two desert dust events in the
figure are highlighted with high values of AOD (>0.25). The first, stronger
dust outbreak increases AOD values during 4 to 5 days over the region (from
27 June to 1 July). Its extent is well simulated in the model direct run, but
its intensity is underestimated (Fig. ). The assimilation produced
fields that are closer to the SEVIRI observations. The second desert dust
event occurs at the end of the TRAQA period (from 8 to 11 July). It is weaker
than the first one and only localized in the part of the western
Mediterranean closer to the coast of Africa. The AOD values of the second
dust event, and also of the period between the two events, are underestimated
in the direct model run, although not as strongly as during the first event.
Data assimilation reduces the difference between the model and the
observations, improving all statistical parameters (Table 2). For example,
the correlation improves from 0.83 for the direct model to 0.96 for the
assimilated model run. Each of the two MODIS instruments overpasses each point
once during the daytime (one at approximately 10:30 and another at
≈ 13:30 in local solar time), and this provides sufficient
information to even improve the hourly AOD variation in the assimilated field
during different dates (for example, 9–11 July; Fig. ).
Correlation (ρ), absolute bias (Δ) and root mean
square error (RMSE) between SEVIRI observations and MOCAGE
direct/assimilation model run for the western Mediterranean during the TRAQA
campaign between 25 June and 13 July 2012. The mean number of SEVIRI
observations per hour is also given. The statistics correspond to
Fig. with the observations localized in the region
0–10∘ E, 35–45∘ N (marked in Fig. by the
grey box).
MOCAGE direct MOCAGE assimilation Nobs‾ (h-1)ρΔRMSEρΔRMSESEVIRI208750.830.140.170.960.080.09
Scatter plots of aerosol optical depths (where colours represent the
number of points) from the independent observation dataset (SEVIRI) and the
simulations: the direct model run (a) and the assimilation model
run (b). In each panel, correlation (ρ), absolute bias
(Δ), root mean square error (RMSE) and standard deviation (σ)
are noted. The included data correspond to the period of the TRAQA campaign
from 25 June until 13 July 2012, and cover the whole MEDI02 domain.
Figure compares AOD from the direct and the assimilation model
runs with SEVIRI observations over the whole control domain, and not only
over the western Mediterranean. The majority of the observed points
correspond to small AOD values where the direct model and the observations
agree well. For larger observed values, the scatter plots confirm that the
direct model run underestimated the AOD field, largely because the desert
dust outbreaks were underestimated in the model. As expected, the
assimilation reduced this disagreement, also displaying better statistics
compared to the direct model run (e.g. the correlation improved from 0.69 to
0.87).
Comparison with AERONET
Positions of AERONET stations used in this study for the period of
the TRAQA campaign in the northern summer of 2012. The grey box marks the region from
which we considered SEVIRI data used in Fig. and Table 2. The
colours represent elevation of the stations as marked on the colour bar.
Time series of aerosol optical depth at 550 nm of the
AERONET data (black line), the direct model (blue line) and the assimilation
model run (red line) for the period of the TRAQA campaign from 25 June until
13 July 2012. The presented AERONET data are from eight stations: Malaga
(ESP), Tabernas (ESP), Avignon (FRA), Ersa (FRA), Frioul (FRA), Lampedusa
(ITA), Limassol (CYP), Palma de Mallorca (ESP). The location of a particular
station is marked at the top of each panel. Correlation, bias and root mean
square error for both the direct model and the assimilation model run as
compared to the AERONET data are given in Table .
Correlation (ρ), absolute bias (Δ) and root mean
square error (RMSE) between AERONET observations and MOCAGE
direct/assimilation run for the period of the TRAQA campaign between 25 June
and 13 July 2012. Each of the stations in the table is identified by its site
name, latitude/longitude, station height, number of observations and above-mentioned statistical parameters. AERONET site locations are also presented
in Fig. .
MOCAGE direct MOCAGE assimilation Station (location)Alt. (m)NobsρΔRMSEρΔRMSEAubiere (FRA; 45.8∘ N, 3.1∘ E)4232250.5060.0410.1100.8590.0240.074Autilla (ESP; 42.0∘ N, 4.6∘ W)8736850.7690.0120.0850.8820.0080.051Avignon (FRA; 43.9∘ N, 4.9∘ E)328460.8510.0240.0870.8960.0050.055Barcelona (ESP; 41.4∘ N, 2.1∘ E)1253780.8020.1100.1690.9000.0380.084Burjassot (ESP; 39.5∘ N, 0.4∘ W)304880.6810.1320.1910.8150.0550.119Cabo da Roca (PT; 38.8∘ N, 9.5∘ W)140770.9650.1300.2630.9390.0320.120Calern OCA (FRA; 43.7∘ N, 6.9∘ E)12705090.7840.0130.0930.9050.0090.052Carpentras (FRA; 44.1∘ N, 5.1∘ E)1007380.7740.0230.0850.8760.0050.055Cerro Poyos (ESP; 37.1∘ N, 3.5∘ W)18301930.6320.0340.0610.6700.0550.074Davos (CH; 46.8∘ N, 9.8∘ E)15962100.5180.0640.0910.6770.0270.063Ersa (FRA; 43.0∘ N, 9.4∘ E)806750.7600.0430.1120.9460.0110.045Evora (PT; 38.6∘ N, 7.9∘ W)2938860.8260.0100.1280.9320.0190.061Frioul (FRA; 43.3∘ N, 5.3∘ E)406580.8710.0370.0960.9520.0140.044Granada (ESP; 37.2∘ N, 3.6∘ W)6808830.6770.0410.1290.9300.0030.057Huelva (ESP; 37.0∘ N, 6.6∘ W)2510020.7930.0100.1530.9360.0340.083Laegeren (CH; 47.5∘ N, 8.4∘ E)7352080.5860.0770.1280.6300.0370.103Lampedusa (ITA; 35.5∘ N, 12.6∘ E)4510580.5730.0840.1240.8450.0060.061Limassol (CYP; 34.7∘ N, 33.0∘ E)229780.3380.0800.1150.6400.0030.067Madrid (ESP; 40.5∘ N, 3.7∘ W)6809040.7310.0110.0970.8780.0040.058Malaga (ESP; 36.7∘ N, 4.5∘ W)407860.7020.1010.1730.9100.0470.088Messina (ITA; 38.2∘ N, 15.6∘ E)155730.5190.0680.1110.8350.0200.060Montsec (ESP; 42.1∘ N, 0.7∘ E)15745280.6620.0160.0780.8920.0090.044Nes Ziona (ISR; 31.9∘ N, 34.8∘ E)405930.2660.0530.1110.7880.0140.063OHP Observatoire (FRA; 43.9∘ N, 5.7∘ E)6806570.7420.0190.0890.8860.0000.053Oujda (MAR; 34.7∘ N, 1.9∘ W)6203300.4590.2020.2210.7560.0900.116Palencia (ESP; 42.0∘ N, 4.5∘ W)7506490.8590.0300.1040.9190.0020.051Palma de Mallorca (ESP; 39.6∘ N, 2.6∘ E)107970.7540.1290.1630.8880.0480.084Porquerolles (FRA; 43.0∘ N, 6.2∘ E)226370.8050.0050.0710.9230.0200.044Sagres (PT; 37.0∘ N, 8.9∘ W)264050.9010.0170.1970.9580.0230.088Sede Boker (ISR; 30.9∘ N, 34.8∘ E)4809500.2400.0090.0800.5520.0650.095San Giuliano (FRA; 42.3∘ N, 9.5∘ E)107680.6750.0840.1370.9080.0660.089Tabernas (ESP; 37.1∘ N, 2.4∘ W)5007400.7540.1290.1840.9270.0380.078Tizi Ouzou (DZA; 36.7∘ N, 4.1∘ E)1332410.6860.1950.2030.7640.0790.095Villefranche (FRA; 43.7∘ N, 7.3∘ E)1304800.7070.0640.1130.8730.0330.067Zaragoza (ESP; 41.6∘ N, 0.9∘ W)2509160.7220.0530.1000.8160.0400.073All sites298400.7400.0500.1150.8830.0060.070
We compare the model direct run and assimilation model run with the AOD data
from AERONET stations. In total, we consider measurements from 35 AERONET
stations, which are all in or around the Mediterranean basin. Their locations
are presented in Fig. . Time series plots for eight stations are
presented in Fig. , and the statistics for all stations in
Table . The stations in Fig. are chosen to
representatively cover the basin. The time series of the stations in the
western part of the Mediterranean basin and in Spain are marked by the strong
desert dust event, which was already discussed earlier. Stations in Spain
recorded the event before the stations in France, where it arrived a couple
of days later. The duration of the event is well simulated by both the direct
model run and the assimilation model run in all stations, but the intensity
is underestimated in the direct model run (Fig. ). However, the
assimilation model run matches the outbreak intensity well. The second,
smaller desert dust event at the end of the TRAQA period is only observed at
southern stations. Similarly, the assimilation model run corrects its
intensity underestimated by the direct model run. The stations in the east,
like in Lampedusa and Cyprus, were not influenced by these dust events. They
are mostly influenced by sea salt aerosols, and the data assimilation also has a positive impact
here. The assimilation model run, with only two MODIS
overpasses per day, also shows improved hourly variations of AOD in these
stations. These variations are not clearly visible in the model direct run,
but they are present in AERONET data with similar amplitudes as in the
assimilation model run. The statistics of all AERONET stations confirm the
overall positive effect of assimilating MODIS data (Table ).
Scatter plots of aerosol optical depths (where colours represent the
number of points) from the independent observation dataset (AERONET) and the
simulations: the direct model run (a) and the assimilation model
run (b). In each panel, correlation (ρ), absolute bias
(Δ), root mean square error (RMSE) and standard deviation (σ)
are noted. The included data correspond to the period of the TRAQA campaign
from 25 June until 13 July 2012, and covers all stations presented in
Fig. .
The AERONET findings confirm those obtained by the comparison with SEVIRI
observations. The scatter plot of all AERONET observations (Fig. )
reinforces the conclusion that the assimilation model run reduces the bias in
the AOD field of the direct model run and significantly improves the
statistical parameters.
In situ aircraft concentration measurements
To further assess the performance of the assimilation model run we evaluate
the impact of the AOD assimilation on aerosol properties other than AOD. To
do this, we compare the modelled aerosol number concentrations with the
aerosol concentrations measured in situ during the TRAQA period by the PCASP
instrument. During the campaign, flights with the ATR-42 were conducted
during 9 different days, carrying the PCASP instrument on board. The flights
passed over the whole western Mediterranean basin using Toulouse, Marseille
and Corsica airports, and covered different meteorological and air quality
conditions. Figure presents three representative time series from
these flights: flight A of 26 June 2012 from Corsica to Toulouse
(Fig. a), flight B of 27 June 2012 from Marseille to Corsica and
back to Toulouse (Fig. b) and flight C of 29 June 2012 from
Corsica to Toulouse (Fig. c). Fig. d show the
tracks of the flights and the availability of the MODIS observations for each
day of the flight. The data availability helps to understand the most recent
impact of the observations in the model, but one should carefully regard it
since it does not reflect the influence of the assimilated observations
during the days prior to the flights.
Aerosol number concentration (cm-3) from the PCASP
instrument on board the ATR aircraft (black line) compared with the direct
model run (blue line) and the assimilation model run (red line) for three
different flights: flight A on 26 June 2012 (a), flight B on 27 June
2012 (b) and flight C on 29 June 2012 (c). The altitude of
the aircraft is also given for all three flights. In addition, the maps of flight
tracks are presented (d) with the points of departure (D) and
arrival (A) for each flight, and the positions of the assimilated
observations during the day of flight (dark green). The colours of the tracks
represent the altitude of the aircraft during the flight, with values defined
in the colour bar. The aerosols are considered in the size range from 1 to
2.5µm.
Aerosol number concentration (cm-3) from the LOAC
instrument on board the meteorological sounding balloons. The presented flights
are performed: in the morning of 29 June 2012 (a), at noon of 29
June 2012 (b) and on 6 July 2012 (c). LOAC measurements
(black line) are compared with the direct model run (blue line) and the
assimilation model run (red line). For the third flight, we also present the
1 h forecast (first guess) started from the assimilated conditions of
1 h before (dashed red line). The aerosols are considered in the size
range from 2.5 to 100µm. In addition, for the third flight, we present
the 12 h back trajectories obtained by the National Oceanic and Atmospheric
Administration's HYSPLIT model (d). They show directions of the
origin of air parcels at the three different starting heights, 200, 1300 and
2500 m above sea level, for the starting location of 42.33∘ N, 5.08∘ E and the starting time of 6 July 2012
at 11:00 UTC.
During flight A (Fig. a), aerosol concentrations are rather low,
except for the anthropogenic pollution around Toulouse measured at the flight
end. The aircraft first visited the area of the Gulf of Genoa where, because
of no available AOD observations, the direct model run and the assimilation
model run show the same aerosol concentrations. The variability in that part
of the flight is well simulated, with slightly higher modelled aerosol
concentrations at these heights than what is measured. Later, on the way to
Toulouse, with more available satellite observations, the assimilation model
run lowers AOD and approaches the measured concentration values. With
situations of no observations or with sparse ones, data assimilation is not
able to have a major effect.
Following the path of flight B (Fig. b) we see again rather clean
aerosol conditions. The modelled and assimilated curves differ, and the
assimilation has an effect on the shape of the time series curve, but it does
not improve the field compared to the measurements noticeably. The result
that the assimilation changes the model, but without a clear improvement,
could be due to different factors. This could happen if the simulated shape
of the aerosol vertical profile differs from the measured one. Since we
assimilate the column-integrated quantity, which does not contain the profile
shape information, the model and measurements at a certain height will not
match if the modelled profile shape is far from the real one, although the
AOD values could match well. A similar impact could arise if the modelled
mixture of aerosol types and the size distribution are different from the
real one. The difference in the sizes would be easily noticeable in the
aerosol number concentration, even if the modelled and observed AOD values
correspond well. The third factor that could contribute to the difference
between the modelled and the assimilated curve is a possible declining effect
with time of previous assimilation cycles on the assimilated curve.
During flight C (Fig. c), the aircraft flew directly through the
desert dust plume. The concentrations are elevated over a wide range of
heights. The assimilation model run significantly improves the aerosol number
concentration, by having it close to the measured ones for most of the flight
path within the plume. If satellite measurements are accurate, concentrations
at one height after an assimilation cycle can only closely correspond to measured
ones if the shape of the vertical profile is well simulated in the
direct model run. To further explore this, in the following subsection we
compare the modelled and the measured vertical profiles.
In situ balloon concentration measurements
During TRAQA, LOAC flew on three balloons, all launched from Martigues, near
Marseille (FRA). Two flights on 29 June 2012, and one on 6 July 2012, are
presented in Fig. . The first two flights flew through the desert
dust plume. The path of the second LOAC flight is near the path of the
aircraft flight C (Fig. c), which allows us to directly compare
the two measurements. The total horizontal motion of LOAC is fairly small
≤15km. Therefore, we will assume that LOAC measurements
represent the aerosol vertical profile above the launch place.
The first two flights (Fig. a and b) are launched at two
different times of the same day, in the morning and at noon, but they flew
through the same desert dust plume. In both cases, the assimilation model run
matches the measurements very closely. It simulates both the shape of
the profile and the aerosol number concentration well. The direct model run
simulates the shape of the vertical profile well, but it underestimates the
aerosol concentration in the plume that the assimilation corrects. For the
second flight, although the concentrations in the plume are clearly improved,
near the surface, the increase of aerosols led to a larger overestimation in
the model. Sometimes, the generalized multiplicative change of the aerosol
profile in the assimilation can produce unsatisfactory effects in some
layers.
LOAC measurements acquired during the balloon flight (Fig. b) are
co-located with the aircraft measurements (Fig. c). They match with the assimilation model run profile
well, which confirms the interpretation already discussed: the aerosol concentration after an assimilation
cycle at a certain height can only be correct if the profile shape is well
simulated in the direct model run.
The third LOAC flight (Fig. c) measured moderate aerosol
concentrations coinciding with an air pollution episode. The assimilation
model run matches well with the measured concentrations. The direct model run
underestimates the concentrations only in lower levels. However, when compared
with the assimilation model run, the assimilation changes the aerosol
vertical profile significantly: the concentrations are increased much more in
the lower levels, while in higher levels the change is less important. In
this case, the different shape of the profile in the direct model run and the
assimilation model run is a result of the continuous multi-day assimilation of
AOD over many assimilation cycles, and the mixing of the aerosols coming from
different levels and regions where they were already assimilated (or not) in
previous assimilation cycles. Different directions of the origin for air
parcel at different heights shown in Fig. d confirm this
assumption. This demonstrates that the continuous assimilation of good-quality AOD observations and/or model propagation of the increment can
correct a shape of the aerosol vertical profile, although a single AOD
assimilation cycle can only expand or shrink the profile shape (as the AOD
observations do not contain the information on the vertical). For the profile
in Fig. c, by comparing the forecast and the analysis of the same
assimilation window, we see that the single AOD assimilation cycle expands
the profile but does not change its shape, something that multiple cycles and
the model propagation of the increment do. Fig. d shows the
back trajectories of air parcels at different heights at the location
corresponding to Fig. c, and confirms that the aerosols on
different levels come from different directions and regions where they have
been already assimilated in previous cycles.
The profile evolution
The profile evolution in the continuous multi-day assimilation run is further
explored in Fig. . We follow the desert dust plume over the
course of 1 week from 25 June to 2 July 2012 from its sources in Africa
to its weakening and dissolution in the Mediterranean Sea
(Fig. c). The aerosols at the different layers are carried by the
winds at different velocities and directions, and, as particles, they undergo
different physical processes (e.g. dry and wet deposition, sedimentation).
Therefore, we cannot follow the dust plume by following an air parcel.
Instead, to track the plume we use a criterion based on the high values of
AOD. The plume, after its emission, heads west and passes over a couple of
other dust sources. Near the Canary Islands the plume turns north-east
towards the Mediterranean basin under the influence of a low-pressure system
centred near the British Isles. MODIS directly observes the plume each day
during the considered period (MODIS passes are marked by triangles in
Fig. b), and these observations in the assimilation model run
have a considerable impact on the plume. When comparing the direct model run
(Fig. a) and the assimilation model run (Fig. b), the
most obvious assimilation impact is the change of the intensity of the plume
in the first part of the trajectory. However, the profile shape also evolves
considerably: at different moments the plume maxima are shifting their peak
heights, and the different layers are changing their relative densities. The
change of the profile shape is visible on a large part of the plume
trajectory, which means that this can be an important feature of a multi-day
AOD assimilation.
There are different effects that the assimilation can have on the profile
shape. While following the plume based on the AOD values, different air
masses enter and exit our trajectory based on their different transport
velocities and directions. Thus, aerosols that are assimilated at different
places later can be gathered in one vertical profile. Then, the
continuous multicycle assimilation can additionally amplify or reduce these
differences in a profile. In Fig. a and b, the change of the
profile shape is noticeable on 28 and 29 June when the plume is near the
coast of Morocco. The assimilation increases aerosol mixing ratios in lower
layers, and decreases them in higher layers. We see a similar effect around
1 July in the northern part of the basin, where the profile shape changes due
to the altitude varying wind direction and velocity and thus the associated
pronounced air mass mixing.
The evolution of the aerosol vertical profile over the course of 1 week from 25 June to 2 July 2012 in the direct model run (a) and the
assimilation model run (b) and the difference between the direct and
the assimilated model run (c). The units are in mass mixing ratios,
with the values represented by colours identified in colour bars.
In (b) the passes of the MODIS over the plume are marked by the
triangles. The desert dust plume is followed from its sources in Africa to
its weakening and dissolution in the Mediterranean Sea (d).
Discussion
Results presented in this paper show that the AOD data assimilation is an
efficient technique to improve modelled aerosol fields. Assimilated fields
have better statistical performances than the direct model run in comparison
with the assimilated observations, and with independent AOD observations and
in situ measurements. The uncertainties in the direct model in the case of a
desert dust outbreak come primarily from the uncertainties in the desert dust
emissions. The dust emission into the atmosphere is a threshold process,
which is very sensitive to uncertainties in the wind field. The small changes
in wind can produce significant differences in the emitted quantities. The
AOD assimilation proved to be a technique capable of reducing the effects of
such uncertainties in the modelled aerosol fields.
In doing so, our assimilation system showed to be more efficient in lowering
overestimated AOD values in the model than increasing underestimated values.
This can be seen in Fig. , and it is directly related to how
matrices B and R are defined in the experiment. By
defining variances as the percentage of modelled and observed quantities, and
making this factor 2 times smaller for observations, we penalized the high
AOD values in the model. This directly affects the analyses, and is later
reflected in the forecasts.
To regulate this feature, one of the possible and the simplest (ad hoc)
approaches would be to limit the observation error in the matrix R
up to a fixed value, which would give more weight to observations in the case
of high observed AOD. This would have a partial effect, only influencing
observations above a certain AOD, and in the case of the substantial
underestimation in the model, it would have a limited impact. Another approach
would be to try to define the matrix B differently. Previous
studies show that a rigorously defined matrix B can slightly
improve the analysis quality . In
MOCAGE-Valentina, in the framework of the MACC (Monitoring Atmospheric
Composition and Climate) project, the influence of different matrices
B was assessed for the case of ozone assimilation
. One of the approaches was the percentage method used
in our experiment. The second approach was the monthly a posteriori
diagnostics computed from the data of a month before,
and it is adapted for operational purposes since the data from the past are
readily available. The third approach is to calculate diagnostics from an
ensemble of runs with perturbed emissions with homogeneous or heterogeneous
correlation length scales. The main conclusion is that all methods
significantly improve the modelled field, and that relative differences
between different methods are small compared to the rate of improvement by
assimilation. For aerosols, the transport processes are more important than
for ozone, but the work of could give an idea of what
to expect in the model from redefining the construction of the matrices
B and R. Moreover, it should be kept in mind that using
a dense observation field, like our MODIS super-observation field, limits the
effects of the spatial diffusion of the increment. This makes the covariances
of the matrix B less important than in the case of sparse
observations.
In our system, the lack of secondary aerosols is presumed to have an
influence on assimilation performance. This could lead to an underestimation
of the direct model AOD in the regions where the secondary aerosols have an
important influence. During the TRAQA period, the primary aerosols had a
dominant effect on the aerosol field, mainly because of the two desert dust
events that occurred during the campaign, and this was favourable for the
evaluation of our system. In the direct model, differences between the model
and the observations appear because of different model uncertainties,
including simplified and neglected processes. However, the differences do not
have a constant or cumulative nature; the model sometimes overestimates or
underestimates AOD. To take into account these uncertainties in the
assimilation process, we defined the variances in the matrix B in
such a way that it allows a margin for the background errors
: the percentage for the background errors is twice as
large as for the observation errors. The developments of an inorganic
secondary aerosol module in MOCAGE are carried in parallel with the
developments on the AOD assimilation system . This has a
beneficial effect in the model and consequently, is expected to improve the
analysis after its inclusion in the assimilation module. To take into account
the model uncertainties, there are also other possibilities. One would be to
add an additional term in the cost function where we would describe the
errors of the model evolution . This method can be used
in the 4-D-Var systems and, besides the implementation of the 4-D-Var method,
it demands additional computational resources in assimilation. In addition, it is
difficult to define the model error covariance matrix in it. Another
possibility would be to apply techniques of bias correction
e.g..
The impact of the AOD assimilation on the model found in our study is
coherent with findings of other studies
. Our approach is the most similar to
the approach used by . We choose the same control
variable, and the choice of the control variable is essential when the model
prognostic and the observed quantities differ as in the case of AOD. The
differences in our systems are the number of bins to which the increment is
repartitioned (11 bins for five species in and 24
bins for four species in our system) and the assimilation method (4-D-VAR
against 3-D-FGAT). derived the matrix B
using the NMC (National Meteorological Center) method .
Satellite AOD errors are defined for retrievals over water using a
multi-regression formula, and for retrievals over land using the percentage
approach with defining a minimal possible error. Their 4-D-Var analysis
results qualitatively showed a very similar impact of assimilation as in this
study.
We assimilated the data of MODIS, which has two overpasses per day during daytime.
Satellite data with higher temporal resolution exist. SEVIRI data with a
temporal resolution of 15 min were used as independent data to evaluate the
results. Assimilating such data could further improve the agreement between
observations and the assimilation model run, but considering the moderate
temporal variability of AOD fields, we would not expect a substantial
improvement. In addition, the SEVIRI AOD products are less accurate compared to
MODIS. The TRAQA period used in our experiment is in northern summer with a
good likelihood of having a cloud-free field, and two overpasses per day were
able to cover a significant part of the control domain each day. Possibly, a
higher temporal resolution of data for assimilation could have a stronger effect
on the model, especially during the winter-time.
Summary and conclusion
In this study we present the development and validation of the
MOCAGE-Valentina system for assimilating aerosol optical depth (AOD). Our
system assimilates aerosol optical depth with the 3-D-FGAT method, and uses
the total 3-D aerosol concentration as a control variable. We examine how 2-D
AOD observations in a continuous multicycle assimilation can improve the
model aerosol representation, including the vertical aerosol profile. We use
accurate in situ aircraft and balloon measurements plus other remotely sensed
data to provide independent validation of the impact of the assimilation. The
MODIS L2 data are assimilated with a model resolution of about 0.2∘
and a 1 h assimilation cycle over the region covering northern Africa, the
Mediterranean basin and southern Europe for the period of the TRAQA campaign in
the northern summer of 2012.
The assimilated model fields show greatly improved aerosol representation
compared to the independently observed datasets, including the 3-D
distributions. The comparison with SEVIRI and AERONET AOD observations, as
independent datasets, confirms the significant positive effect of the AOD
assimilation to the model. For example, the comparison with AERONET data
shows that the assimilation decreased the bias in AOD (from 0.050 to 0.006) and
increased the correlation (from 0.74 to 0.88).
The TRAQA campaign provided independent 3-D data on aerosol concentration.
The assimilation sometimes improved the modelled fields and sometimes had
little effect. The best results as expected occurred when the shape of the
vertical profile was correctly simulated by the direct (unassimilated) model.
The shape of the aerosol vertical profiles does not change during one
assimilation cycle because AOD observations do not contain any vertical
information, but the profile shape can change and be improved by the AOD
assimilation because different parts of the column can be carried by winds
from different directions. The AOD assimilation can also impact aerosol size
and type for the same reason, but this was not evident in this experiment.
The AOD assimilation proved to be a very efficient technique to improve the
model forecast of bulk aerosols and a powerful tool for producing reanalyses
or studying past events.
As of an outlook of further developments, the next steps will consist of
improving the system performance and broadening its capabilities. For
example, it could be advantageous to assimilate observations from different
instruments at the same time. The aerosol observations from space are
available from various instruments located on different satellites, which can
provide different spatial and temporal coverage and resolutions. Combining
complementary data from different instruments could improve the system
performance, but the possible inter-instrument biases would need to be
carefully considered.
In addition, the same or even bigger positive effect of the assimilation could be
expected in the case of other strong aerosol events, like biomass burning or
a volcanic ash plume, where the model emission uncertainties are often even
larger than in the case of desert dust plumes.
By assimilating AOD observations at several wavelengths, we can get
information on aerosol size. If aerosol absorption can be measured, then we
can discriminate between carbonaceous and other aerosols. Then, with this
information we could modify the size distribution in the model. To achieve
this in the system, it would be necessary to study the relationship and
sensitivity between the size and bin distribution in MOCAGE and the aerosol
Ångstrom exponent obtained from multi-wavelength measurements.
If we want to introduce direct information of the vertical profile from
observations into the model, we would need to assimilate another type of
observations. Lidar observations, ground- or space-based, are an obvious
choice. The control variable defined as the 3-D total concentration is also
well adapted for the assimilation of lidar profiles. This facilitates the
implementation of the lidar assimilation into the system, and in the longer
term also makes feasible a simultaneous assimilation of the AOD and lidar
profiles as possibly complimentary datasets.
Data availability
The MOCAGE, PCASP and LOAC TRAQA data are available on the ChArMEx database
at http://mistrals.sedoo.fr/ChArMEx/ with subsequent DOI: the MOCAGE
direct run data (10.14768/MISTRALS-CHARMEX.1034) , the MOCAGE
assimilation run data (10.14768/MISTRALS-CHARMEX.1449) , the PCASP
TRAQA data (10.6096/MISTRALS-ChArMEx.998) and the LOAC TRAQA data (10.6096/MISTRALS-ChArMEx.833) .
The user must register before having access to the data. The MODIS/Aqua and MODIS/Terra Atmosphere L2
Aerosol Products (MYD04-C51 and MOD04-C51) were acquired from the Level-1
& Atmosphere Archive and Distribution System (LAADS) Distributed Active Archive
Center (DAAC, http://ladsweb.nascom.nasa.gov) . The SEVIRI/MSG product (SEV_AER-OC)
was acquired from the ICARE data centre (http://www.icare.univ-lille1.fr) . The user
must register before having access to the data. The AERONET level 2 data were
acquired from the AERONET network website (http://aeronet.gsfc.nasa.gov/) .
Acknowledgements
This work has been funded by Centre National de Recherches
Météorologiques (CNRM-GAME) of Météo-France and Centre
National de la Recherche Scientifique (CNRS). The authors would like to thank
Jean-Luc Attié, the TRAQA principal investigator, and all TRAQA/ChArMEx collaborators for
producing and providing the data used in this study. TRAQA was funded by
ADEME/PRIMEQUAL and MISTRALS/ChArMEx programmes and Observatoire
Midi-Pyrénées. We acknowledge the AERONET principal investigators and their staff for
establishing and maintaining the sites used in this investigation. We also
acknowledge the Global Fire Emission Database project and
for the biomass burning and carbonaceous aerosol
emissions, the MODIS mission team and scientists for the production of the
MODIS data, as well as and the ICARE data centre for
developing and producing the SEVIRI-retrieved aerosol data that we used in
this study. Edited by: W. Lahoz
Reviewed by: three anonymous referees
ReferencesAndersson, E., Haseler, J., Undén, P., Courtier, P., Kelly, G.,
Vasiljevic, D., Brankovic, C., Gaffard, C., Hollingsworth, A., Jakob, C.,
Janssen, P., Klinker, E., Lanzinger, A., Miller, M., Rabier, F., Simmons, A.,
Strauss, B., Viterbo, P., Cardinali, C., and Thépaut, J.-N.: The ECMWF
implementation of three-dimensional variational assimilation (3D-Var). III:
Experimental results, Q. J. Roy. Meteor. Soc., 124, 1831–1860,
10.1002/qj.49712455004, 1998.
Attié, J.-L. et al.: Transport of
Pollution and Air Quality experiment over the Mediterranean basin
(TRAQA/ChArMEx campaign), in preparation, 2016.
Bechtold, P., Bazile, E., Guichard, F., Mascart, P., and Richard, E.: A
mass-flux convection scheme for regional and global models, Q. J. Roy.
Meteor. Soc., 127, 869–886, 2001.Benedetti, A., Morcrette, J.-J., Boucher, O., Dethof, A., Engelen, R. J.,
Fisher, M., Flentje, H., Huneeus, N., Jones, L., Kaiser, J. W., Kinne, S.,
Mangold, A., Razinger, M., Simmons, A. J., and Suttie, M.: Aerosol analysis
and forecast in the European centre for medium-range weather forecasts
integrated forecast system: 2. Data assimilation, J. Geophys. Res.-Atmos.,
114, D13205, 10.1029/2008JD011115, 2009.
Boucher, O., Randall, D., Artaxo, P., Bretherton, C., Feingold, G., Forster,
P., Kerminen, V.-M., Kondo, Y., Liao, H., Lohmann, U., Rasch, P., Satheesh,
S. K., Sherwood, S., Stevens, B., and Zhang, X. Y.: Clouds and aerosols, in:
Climate change 2013: The physical science basis: Working Group I
Contribution to the Fifth Assessment Report of the Intergovernmental Panel
on Climate Change, Cambridge University Press, Cambridge and New York,
571–657, 2013.
Breon, F.-M., Vermeulen, A., and Descloitres, J.: An evaluation of satellite
aerosol products against sunphotometer measurements, Remote Sens. Environ.,
115, 3102–3111, 2011.
Buis, S., Piacentini, A., and Déclat, D.: PALM: a computational
framework for assembling high-performance computing applications, Concurr.
Comp.-Pract. E., 18, 231–245, 2006.Cai, Y., Snider, J. R., and Wechsler, P.: Calibration of the passive cavity
aerosol spectrometer probe for airborne determination of the size
distribution, Atmos. Meas. Tech., 6, 2349–2358, 10.5194/amt-6-2349-2013,
2013.Collins, W. D., Rasch, P. J., Eaton, B. E., Khattatov, B. V., Lamarque,
J.-F., and Zender, C. S.: Simulating aerosols using a chemical transport
model with assimilation of satellite aerosol retrievals: Methodology for
INDOEX, J. Geophys. Res.-Atmos., 106, 7313–7336, 10.1029/2000JD900507,
2001.
Courtier, P., Thépaut, J.-N., and Hollingsworth, A.: A strategy for
operational implementation of 4D-Var, using an incremental approach, Q. J.
Roy. Meteor. Soc., 120, 1367–1387, 1994.Dai, T., Schutgens, N. A., Goto, D., Shi, G., and Nakajima, T.: Improvement
of aerosol optical properties modeling over Eastern Asia with MODIS AOD
assimilation in a global non-hydrostatic icosahedral aerosol transport model,
Environ. Pollut., 195, 319–329, 10.1016/j.envpol.2014.06.021, 2014.
Daley, R.: Atmospheric data analysis, No. 2, Atmospheric and Space Science
Series, Cambridge University Press, 1993.
De Leeuw, G., Kinne, S., Léon, J.-F., Pelon, J., Rosenfeld, D., Schaap,
M., Veefkind, P. J., Veihelmann, B., Winker, D. M., and von Hoyningen-Huene,
W.: Retrieval of aerosol properties, in: The remote sensing of tropospheric
composition from space, Springer, 259–313, 2011.
Dee, D. and Uppala, S.: Variational bias correction in ERA-Interim, ECMWF,
2008.
Desroziers, G., Berre, L., Chapnik, B., and Poli, P.: Diagnosis of
observation, background and analysis-error statistics in observation space,
Q. J. Roy. Meteor. Soc., 131, 3385–3396, 2005.
Dufour, A., Amodei, M., Ancellet, G., and Peuch, V.-H.: Observed and modelled
“chemical weather” during ESCOMPTE, Atmos. Res., 74, 161–189, 2005.El Amraoui, L. and Sič, B.: MOCAGE direct aerosol run data for the period of the
TRAQA campaign, ChArMEx database, available at: http://mistrals.sedoo.fr/ChArMEx/
(registration required, last access: November 2016), 10.14768/MISTRALS-CHARMEX.1034, 2013.El Amraoui, L. and Sič, B.: MOCAGE assimilation aerosol run data for the period
of the TRAQA campaign, ChArMEx database, available at: http://mistrals.sedoo.fr/ChArMEx/
(registration required, last access: November 2016), 10.14768/MISTRALS-CHARMEX.1449, 2016.Elbern, H., Schmidt, H., and Ebel, A.: Variational data assimilation for
tropospheric chemistry modeling, J. Geophys. Res.-Atmos., 102, 15967–15985,
10.1029/97JD01213, 1997.
Fisher, M. and Andersson, E.: Developments in 4D-Var and Kalman filtering,
European Centre for Medium-Range Weather Forecasts, Reading, UK, 2001.Fisher, M. and Lary, D. J.: Lagrangian four-dimensional variational data
assimilation of chemical species, Q. J. Roy. Meteor. Soc., 121, 1681–1704,
10.1002/qj.49712152709, 1995.Fuzzi, S., Andreae, M. O., Huebert, B. J., Kulmala, M., Bond, T. C., Boy, M.,
Doherty, S. J., Guenther, A., Kanakidou, M., Kawamura, K., Kerminen, V.-M.,
Lohmann, U., Russell, L. M., and Pöschl, U.: Critical assessment of the
current state of scientific knowledge, terminology, and research needs
concerning the role of organic aerosols in the atmosphere, climate, and
global change, Atmos. Chem. Phys., 6, 2017–2038,
10.5194/acp-6-2017-2006, 2006.
Giorgi, F. and Chameides, W. L.: Rainout lifetimes of highly soluble aerosols
and gases as inferred from simulations with a general circulation model, J.
Geophys. Res.-Atmos., 91, 14367–14376, 1986.Guth, J., Josse, B., Marécal, V., Joly, M., and Hamer, P.: First
implementation of secondary inorganic aerosols in the MOCAGE version R2.15.0
chemistry transport model, Geosci. Model Dev., 9, 137–160,
10.5194/gmd-9-137-2016, 2016.Hess, M., Koepke, P., and Schult, I.: Optical Properties of Aerosols and
Clouds: The Software Package OPAC, B. Am. Meteorol. Soc., 79, 831–844,
10.1175/1520-0477(1998)079<0831:OPOAAC>2.0.CO;2, 1998.Holben, B., Eck, T., Slutsker, I., Tanré, D., Buis, J., Setzer, A.,
Vermote, E., Reagan, J., Kaufman, Y., Nakajima, T., Lavenu, F., Jankowiak,
I., and Smirnov, A.: AERONET – A federated instrument network and data
archive for aerosol characterization, Remote Sens. Environ., 66, 1–16,
10.1016/S0034-4257(98)00031-5, 1998.Holben, B., Eck, T., Slutsker, I., Tanré, D., Buis, J., Setzer, A.,
Vermote, E., Reagan, J., Kaufman, Y., Nakajima, T., Lavenu, F., Jankowiak,
I., and Smirnov, A.: AERONET Aerosol Optical Depth Level 2.0 Data, AERONET,
Goddard Space Flight Center, available at: http://aeronet.gsfc.nasa.gov/
(last access: November 2016), 2012.
Hsu, N. C., Tsay, S.-C., King, M. D., and Herman, J. R.: Deep Blue
retrievals of Asian aerosol properties during ACE-Asia, IEEE T. Geosci.
Remote, 44, 3180–3195, 2006.Hsu, C., Remer, L., Mattoo, S., and Chu, A.: MODIS Atmosphere L2 Aerosol Product –
Collection 051. NASA MODIS Adaptive Processing System, Goddard Space Flight Center,
available at: http://ladsweb.nascom.nasa.gov (last access: November 2016), 2008.Jaeglé, L., Quinn, P. K., Bates, T. S., Alexander, B., and Lin, J.-T.:
Global distribution of sea salt aerosols: new constraints from in situ and
remote sensing observations, Atmos. Chem. Phys., 11, 3137–3157,
10.5194/acp-11-3137-2011, 2011.Jaumouillé, E., Massart, S., Piacentini, A., Cariolle, D., and Peuch,
V.-H.: Impact of a time-dependent background error covariance matrix on air
quality analysis, Geosci. Model Dev., 5, 1075–1090,
10.5194/gmd-5-1075-2012, 2012.
Josse, B., Simon, P., and Peuch, V.-H.: Radon global simulations with the
multiscale chemistry and transport model MOCAGE, Tellus B, 56, 339–356,
2004.Kahnert, M.: Variational data analysis of aerosol species in a regional
CTM: background error covariance constraint and aerosol optical observation
operators, Tellus B, 60, 5, 753–770,
10.1111/j.1600-0889.2008.00377.x, 2011.Kaiser, J. W., Heil, A., Andreae, M. O., Benedetti, A., Chubarova, N., Jones,
L., Morcrette, J.-J., Razinger, M., Schultz, M. G., Suttie, M., and van der
Werf, G. R.: Biomass burning emissions estimated with a global fire
assimilation system based on observed fire radiative power, Biogeosciences,
9, 527–554, 10.5194/bg-9-527-2012, 2012.Kanakidou, M., Seinfeld, J. H., Pandis, S. N., Barnes, I., Dentener, F. J.,
Facchini, M. C., Van Dingenen, R., Ervens, B., Nenes, A., Nielsen, C. J.,
Swietlicki, E., Putaud, J. P., Balkanski, Y., Fuzzi, S., Horth, J., Moortgat,
G. K., Winterhalter, R., Myhre, C. E. L., Tsigaridis, K., Vignati, E.,
Stephanou, E. G., and Wilson, J.: Organic aerosol and global climate
modelling: a review, Atmos. Chem. Phys., 5, 1053–1123,
10.5194/acp-5-1053-2005, 2005.Kinne, S., Schulz, M., Textor, C., Guibert, S., Balkanski, Y., Bauer, S. E.,
Berntsen, T., Berglen, T. F., Boucher, O., Chin, M., Collins, W., Dentener,
F., Diehl, T., Easter, R., Feichter, J., Fillmore, D., Ghan, S., Ginoux, P.,
Gong, S., Grini, A., Hendricks, J., Herzog, M., Horowitz, L., Isaksen, I.,
Iversen, T., Kirkevåg, A., Kloster, S., Koch, D., Kristjansson, J. E.,
Krol, M., Lauer, A., Lamarque, J. F., Lesins, G., Liu, X., Lohmann, U.,
Montanaro, V., Myhre, G., Penner, J., Pitari, G., Reddy, S., Seland, O.,
Stier, P., Takemura, T., and Tie, X.: An AeroCom initial assessment –
optical properties in aerosol component modules of global models, Atmos.
Chem. Phys., 6, 1815–1834, 10.5194/acp-6-1815-2006, 2006.Kirchstetter, T. W., Novakov, T., and Hobbs, P. V.: Evidence that the
spectral dependence of light absorption by aerosols is affected by organic
carbon, J. Geophys. Res.-Atmos., 109, D21208, 10.1029/2004JD004999,
2004.
Köpke, P., Hess, M., Schult, I., and Shettle, E.: Global aerosol data
set, Tech. Rep. MPI Report No 243, Max-Planck-Institut für Meteorologie,
Hamburg, Germany, 1997.Lamarque, J.-F., Bond, T. C., Eyring, V., Granier, C., Heil, A., Klimont, Z.,
Lee, D., Liousse, C., Mieville, A., Owen, B., Schultz, M. G., Shindell, D.,
Smith, S. J., Stehfest, E., Van Aardenne, J., Cooper, O. R., Kainuma, M.,
Mahowald, N., McConnell, J. R., Naik, V., Riahi, K., and van Vuuren, D. P.:
Historical (1850–2000) gridded anthropogenic and biomass burning emissions
of reactive gases and aerosols: methodology and application, Atmos. Chem.
Phys., 10, 7017–7039, 10.5194/acp-10-7017-2010, 2010.
Laurent, B., Marticorena, B., Bergametti, G., and Mei, F.: Modeling mineral
dust emissions from Chinese and Mongolian deserts, Global Planet. Change,
52, 121–141, 2006.Lee, L. A., Carslaw, K. S., Pringle, K. J., Mann, G. W., and Spracklen, D.
V.: Emulation of a complex global aerosol model to quantify sensitivity to
uncertain parameters, Atmos. Chem. Phys., 11, 12253–12273,
10.5194/acp-11-12253-2011, 2011.
Liu, D. C. and Nocedal, J.: On the limited memory BFGS method for large
scale optimization, Math. Program., 45, 503–528, 1989.Liu, Z., Liu, Q., Lin, H.-C., Schwartz, C. S., Lee, Y.-H., and Wang, T.:
Three-dimensional variational assimilation of MODIS aerosol optical depth:
Implementation and application to a dust storm over East Asia, J. Geophys.
Res.-Atmos., 116, D23206, 10.1029/2011JD016159, 2011.
Louis, J.-F.: A parametric model of vertical eddy fluxes in the atmosphere,
Bound.-Lay. Meteorol., 17, 187–202, 1979.Lurton, T., Renard, J.-B., Vignelles, D., Jeannot, M., Akiki, R., Mineau,
J.-L., and Tonnelier, T.: Light scattering at small angles by atmospheric
irregular particles: modelling and laboratory measurements, Atmos. Meas.
Tech., 7, 931–939, 10.5194/amt-7-931-2014, 2014.Mahowald, N., Luo, C., del Corral, J., and Zender, C. S.: Interannual
variability in atmospheric mineral aerosols from a 22-year model simulation
and observational data, J. Geophys. Res.-Atmos., 108, 4352,
10.1029/2002JD002821, 2003.
Martet, M., Peuch, V.-H., Laurent, B., Marticorena, B., and Bergametti, G.:
Evaluation of long-range transport and deposition of desert dust with the
CTM MOCAGE, Tellus B, 61, 449–463, 2009.Marticorena, B., Bergametti, G., Aumont, B., Callot, Y., N'doumé, C., and
Legrand, M.: Modeling the atmospheric dust cycle: 2. Simulation of Saharan
dust sources, J. Geophys. Res.-Atmos., 102, 4387–4404,
10.1029/96JD02964, 1997.
Massart, S., Piacentini, A., Cariolle, D., Amraoui, L. E., and Semane, N.:
Assessment of the quality of the ozone measurements from the Odin/SMR
instrument using data assimilation, Can. J. Phys., 85, 1209–1223, 2007.Massart, S., Clerbaux, C., Cariolle, D., Piacentini, A., Turquety, S., and
Hadji-Lazaro, J.: First steps towards the assimilation of IASI ozone data
into the MOCAGE-PALM system, Atmos. Chem. Phys., 9, 5073–5091,
10.5194/acp-9-5073-2009, 2009.
Massart, S., Pajot, B., Piacentini, A., and Pannekoucke, O.: On the merits of
using a 3D-FGAT assimilation scheme with an outer loop for atmospheric
situations governed by transport, Mon. Weather Rev., 138, 4509–4522, 2010.
Massart, S., Piacentini, A., and Pannekoucke, O.: Importance of using
ensemble estimated background error covariances for the quality of
atmospheric ozone analyses, Q. J. Roy. Meteor. Soc., 138, 889–905, 2012.
Ménard, R., Cohn, S. E., Chang, L.-P., and Lyster, P. M.: Assimilation of
stratospheric chemical tracer observations using a Kalman filter. Part I:
Formulation, Mon. Weather Rev., 128, 2654–2671, 2000.
Nho-Kim, E.-Y., Michou, M., and Peuch, V.-H.: Parameterization of
size-dependent particle dry deposition velocities for global modeling, Atmos.
Environ., 38, 1933–1942, 2004.Niu, T., Gong, S. L., Zhu, G. F., Liu, H. L., Hu, X. Q., Zhou, C. H., and
Wang, Y. Q.: Data assimilation of dust aerosol observations for the
CUACE/dust forecasting system, Atmos. Chem. Phys., 8, 3473–3482,
10.5194/acp-8-3473-2008, 2008.Pagowski, M. and Grell, G. A.: Experiments with the assimilation of fine
aerosols using an ensemble Kalman filter, J. Geophys. Res.-Atmos., 117,
D21302, 10.1029/2012JD018333, 2012.
Pannekoucke, O. and Massart, S.: Estimation of the local diffusion tensor and
normalization for heterogeneous correlation modelling using a diffusion
equation, Q. J. Roy. Meteor. Soc., 134, 1425–1438, 2008.
Parrish, D. F. and Derber, J. C.: The National Meteorological Center's
spectral statistical-interpolation analysis system, Mon. Weather Rev., 120,
1747–1763, 1992.Piguet B. and Perrin T.: ATR Core in-situ measurements – TRAQA,
ChArMEx database, available at: http://mistrals.sedoo.fr/ChArMEx/
(registration required, last access: November 2016), 10.6096/MISTRALS-ChArMEx.998, 2013.Rasch, P. J., Collins, W. D., and Eaton, B. E.: Understanding the Indian
Ocean Experiment (INDOEX) aerosol distributions with an aerosol assimilation,
J. Geophys. Res.-Atmos., 106, 7337–7355, 10.1029/2000JD900508, 2001.Remer, L. A., Kaufman, Y., Tanré, D., Mattoo, S., Chu, D., Martins,
J. V., Li, R.-R., Ichoku, C., Levy, R., Kleidman, R., Eck, T., Vermote, D.,
and Holben, B.: The MODIS aerosol algorithm, products, and validation, J.
Atmos. Sci., 62, 947–973, 10.1175/JAS3385.1, 2005.Renard J.-B.: LOAC – TRAQA – Martigues, ChArMEx database, available at: http://mistrals.sedoo.fr/ChArMEx/
(registration required, last access: November 2016), 10.6096/MISTRALS-ChArMEx.833, 2012.Renard, J.-B., Dulac, F., Berthet, G., Lurton, T., Vignelles, D., Jégou,
F., Tonnelier, T., Thaury, C., Jeannot, M., Couté, B., Akiki, R., Mineau,
J.-L., Verdier, N., Mallet, M., Gensdarmes, F., Charpentier, P., Mesmin, S.,
Duverger, V., Dupont, J.-C., Elias, T., Crenn, V., Sciare, J., Giacomoni, J.,
Gobbi, M., Hamonou, E., Olafsson, H., Dagsson-Waldhauserova, P., Camy-Peyret,
C., Mazel, C., Décamps, T., Piringer, M., Surcin, J., and Daugeron, D.:
LOAC: a small aerosol optical counter/sizer for ground-based and balloon
measurements of the size distribution and nature of atmospheric particles –
Part 1: Principle of measurements and instrument evaluation, Atmos. Meas.
Tech. Discuss., 8, 1203–1259, 10.5194/amtd-8-1203-2015, 2015.Rubin, J. I., Reid, J. S., Hansen, J. A., Anderson, J. L., Collins, N., Hoar,
T. J., Hogan, T., Lynch, P., McLay, J., Reynolds, C. A., Sessions, W. R.,
Westphal, D. L., and Zhang, J.: Development of the Ensemble Navy Aerosol
Analysis Prediction System (ENAAPS) and its application of the Data
Assimilation Research Testbed (DART) in support of aerosol forecasting,
Atmos. Chem. Phys., 16, 3927–3951, 10.5194/acp-16-3927-2016, 2016.Schutgens, N. A. J., Miyoshi, T., Takemura, T., and Nakajima, T.: Applying an
ensemble Kalman filter to the assimilation of AERONET observations in a
global aerosol transport model, Atmos. Chem. Phys., 10, 2561–2576,
10.5194/acp-10-2561-2010, 2010.
Seinfeld, J. H. and Pandis, S. N.: Atmospheric chemistry and physics: from
air pollution to climate change, John Wiley & Sons, New York, USA, 1998.Sekiyama, T. T., Tanaka, T. Y., Shimizu, A., and Miyoshi, T.: Data
assimilation of CALIPSO aerosol observations, Atmos. Chem. Phys., 10, 39–49,
10.5194/acp-10-39-2010, 2010.Shindell, D. T., Lamarque, J.-F., Schulz, M., Flanner, M., Jiao, C., Chin,
M., Young, P. J., Lee, Y. H., Rotstayn, L., Mahowald, N., Milly, G.,
Faluvegi, G., Balkanski, Y., Collins, W. J., Conley, A. J., Dalsoren, S.,
Easter, R., Ghan, S., Horowitz, L., Liu, X., Myhre, G., Nagashima, T., Naik,
V., Rumbold, S. T., Skeie, R., Sudo, K., Szopa, S., Takemura, T.,
Voulgarakis, A., Yoon, J.-H., and Lo, F.: Radiative forcing in the ACCMIP
historical and future climate simulations, Atmos. Chem. Phys., 13,
2939–2974, 10.5194/acp-13-2939-2013, 2013.Sič, B., El Amraoui, L., Marécal, V., Josse, B., Arteta, J., Guth,
J., Joly, M., and Hamer, P. D.: Modelling of primary aerosols in the chemical
transport model MOCAGE: development and evaluation of aerosol physical
parameterizations, Geosci. Model Dev., 8, 381–408,
10.5194/gmd-8-381-2015, 2015.
Slinn, W.: Some approximations for the wet and dry removal of particles and
gases from the atmosphere, Water Air Soil Poll., 7, 513–543, 1977.
Strapp, J. W., Leaitch, W., and Liu, P.: Hydrated and dried
aerosol-size-distribution measurements from the particle measuring systems
FSSP-300 probe and the deiced PCASP-100X probe, J. Atmos. Ocean. Tech.,
9, 548–555, 1992.Talagrand, O.: A Posteriori Validation of Assimilation Algorithms, in: Data
Assimilation for the Earth System, edited by: Swinbank, R., Shutyaev, V., and
Lahoz, W., NATO Science Series, Springer Netherlands, 26, 85–95,
10.1007/978-94-010-0029-1_8, 2003.Textor, C., Schulz, M., Guibert, S., Kinne, S., Balkanski, Y., Bauer, S.,
Berntsen, T., Berglen, T., Boucher, O., Chin, M., Dentener, F., Diehl, T.,
Easter, R., Feichter, H., Fillmore, D., Ghan, S., Ginoux, P., Gong, S.,
Grini, A., Hendricks, J., Horowitz, L., Huang, P., Isaksen, I., Iversen, I.,
Kloster, S., Koch, D., Kirkevåg, A., Kristjansson, J. E., Krol, M.,
Lauer, A., Lamarque, J. F., Liu, X., Montanaro, V., Myhre, G., Penner, J.,
Pitari, G., Reddy, S., Seland, Ø., Stier, P., Takemura, T., and Tie, X.:
Analysis and quantification of the diversities of aerosol life cycles within
AeroCom, Atmos. Chem. Phys., 6, 1777–1813, 10.5194/acp-6-1777-2006,
2006.
Thieuleux, F., Moulin, C., Bréon, F. M., Maignan, F., Poitou, J., and
Tanré, D.: SEVIRI Level-2 Aerosol Products (AOT and AE) over Ocean (SEV_AER-OC-L2),
ICARE data centre, University of Lille, available at: http://www.icare.univ-lille1.fr
(registration required, last access: November 2016), 2005a.Thieuleux, F., Moulin, C., Bréon, F. M., Maignan, F., Poitou, J., and
Tanré, D.: Remote sensing of aerosols over the oceans using MSG/SEVIRI
imagery, Ann. Geophys., 23, 3561–3568, 10.5194/angeo-23-3561-2005,
2005b.Tombette, M., Mallet, V., and Sportisse, B.: PM10 data assimilation over
Europe with the optimal interpolation method, Atmospheric Chemistry and
Physics, 9, 57–70, 10.5194/acp-9-57-2009, 2009.Tréemolet, Y.: Accounting for an imperfect model in 4D-Var, Q. J. Roy.
Meteor. Soc., 132, 2483–2504, 10.1256/qj.05.224, 2006.Vignati, E., Karl, M., Krol, M., Wilson, J., Stier, P., and Cavalli, F.:
Sources of uncertainties in modelling black carbon at the global scale,
Atmos. Chem. Phys., 10, 2595–2611, 10.5194/acp-10-2595-2010, 2010.
Weaver, A. and Courtier, P.: Correlation modelling on the sphere using a
generalized diffusion equation, Q. J. Roy. Meteor. Soc., 127, 1815–1846,
2001.
Williamson, D. L. and Rasch, P. J.: Two-dimensional semi-Lagrangian
transport with shape-preserving interpolation, Mon. Weather Rev., 117,
102–129, 1989.
Wiscombe, W. J.: Mie scattering calculations: advances in technique and fast,
vector-speed computer codes, Atmospheric Analysis and Prediction Division,
National Center for Atmospheric Research, 1979.
Wiscombe, W. J.: Improved Mie scattering algorithms, Appl. Opt., 19,
1505–1509, 1980.Yumimoto, K., Uno, I., Sugimoto, N., Shimizu, A., Liu, Z., and Winker, D. M.:
Adjoint inversion modeling of Asian dust emission using lidar observations,
Atmos. Chem. Phys., 8, 2869–2884, 10.5194/acp-8-2869-2008, 2008.Zhang, J., Reid, J. S., Westphal, D. L., Baker, N. L., and Hyer, E. J.: A
system for operational aerosol optical depth data assimilation over global
oceans, J. Geophys. Res.-Atmos., 113, D10208, 10.1029/2007JD009065, 2008.