Introduction
Numerous and various aerosols affect the Mediterranean basin ,
located at the crossroads of air masses carrying both natural (desertic
particles, sea salt, volcanic ashes, etc.) and anthropogenic (black carbon,
sulphate, etc.) particles. Because of their microphysical and optical
properties, these aerosols can have strong effects on the regional radiative
budget e.g., with ensuing impact on climate
and ecosystems of the Mediterranean
. Among these aerosols, the Saharan desert dust particles
represent an important contribution of aerosols for this region
. Indeed, dust particles coming from suspension,
saltation and creeping processes associated with wind erosion
can move from northern Africa to the Mediterranean Sea and Europe
. These dust outbreaks are mainly driven by the
synoptic meteorological conditions : they are more frequent in
the eastern basin in winter and spring, in the central basin in spring and in
the western basin in summer . The ChArMEx initiative
(Chemistry-Aerosol Mediterranean Experiment,
http://charmex.lsce.ipsl.fr) has been launched for a few years in the
framework of the MISTRALS (Mediterranean Integrated STudies at Regional And
Local Scales) programme in order to improve our knowledge of aerosols and
their impacts on climate in the Mediterranean. Thus, in early summer 2012,
the ChArMEx/TRAQA (TRansport and Air QuAlity) campaign focused on the
characterization of the polluted air masses over the Mediterranean basin
through the study of representative case studies. A particularly intense dust
event has been measured at the end of June with different observation means
(balloons, aircraft, surface and remote-sensing measurements) and
consequently represents a documented case to evaluate the aerosol schemes of
regional climate models. Indeed the analysis of study cases is made possible
by the use of a reanalysis as lateral boundary forcing which provides the
real chronology of these events.
The aim of the present work is consequently to evaluate the direct and
semi-direct effects of dust particles during summer 2012 both at the daily
time scale and at the summer scale. We consider here a modelling approach
with the following requirements. First of all, in order to simulate dust
outbreaks, models need prognostic dust schemes (emission, transport,
deposition) to uplift dust particles from arid areas and transport them in
the atmosphere. Many climate models indeed use only monthly aerosol
climatology e.g. that cannot correspond to this kind
of study. However, disregarding the chemistry-transport models (e.g. CHIMERE,
MOCAGE) that do not have aerosol–climate interactions, several aerosol
schemes already exist in different climate models (e.g. MACC, ECHAM-HAM,
IPSL), evaluated in different intercomparison exercises (e.g. AEROCOM,
, ACCMIP, ). With regards to dust aerosols,
most of the climate models can simulate the main patterns of dust emission
and transport , but large uncertainties remain in the
characterization of dust properties and the resulting impact on climate
notably because of differences in dust emission
parameterizations . Over the Euro-Mediterranean region, several
studies have considered the effects of aerosols on climate using simulations
with a prognostic scheme, both for anthropogenic aerosols
and dust particles .
Moreover, the role of the Mediterranean Sea is essential in climate
feedbacks , so that ocean–atmosphere coupled
regional models have recently been developed . The importance of this coupling in the
aerosol–climate interactions in the Mediterranean has even been recently
highlighted . However, up to now, aerosol–climate studies with
prognostic aerosol schemes have been achieved either with the COSMO
or with the RegCM model and have not included
an ocean–atmosphere coupling yet, even if an ocean–atmosphere coupling is
currently developed between RegCM and ROMS .
In addition, as the Mediterranean is also characterized by local winds,
complex coastlines and orography, high-resolution modelling is needed to correctly reproduce the atmospheric circulation .
From our knowledge, none of these regional models can have simultaneous ocean–atmosphere coupling and prognostic aerosol schemes. In the present
study, a new version of the coupled regional climate model system (RCSM) of
the CNRM, called CNRM-RCSM5, has been developed, including an aerosol
prognostic scheme derived from the GEMS/MACC project
in addition to the atmosphere, ocean and land-surface components. This new
model tool thus complies with all the criteria mentioned above and should be
able to help us to evaluate the direct and semi-direct effects of dust
aerosols at the daily time scale. The data brought by the TRAQA campaign
provide the opportunity to a first evaluation of the dust aerosol scheme
before assessing the radiative aerosol effects. Additionally, including the other
aerosol species allows a comparison of total aerosol optical depth (AOD) with
remote-sensing measurements. Thus the present work aims at studying the
radiative effects of dust aerosols in the Mediterranean area during summer
2012. The question of the difference between the use of climatological and
prognostic aerosols in this model will also be raised, notably to study the
consequences of this choice both on the daily and seasonal (for summer)
variability of different meteorological parameters (radiation, temperature,
cloud cover).
After a description of the aerosol scheme in Sect. 2 and its evaluation in
Sect. 3, the radiative effects of aerosols are studied in Sect. 4 before the concluding remarks in Sect. 5.
Methodology
The CNRM-RCSM5 model
Four different components are included in this regional climate model system:
the atmosphere with the regional climate model ALADIN-Climate
, the ocean with the regional model NEMOMED8
, the land-surface with the model ISBA and the
aerosols, simulated interactively within ALADIN-Climate (see details in 2.2).
ALADIN-Climate is a bi-spectral semi-implicit semi-Lagrangian regional model
with a 50 km horizontal resolution and 31 vertical levels in the present
work. The version 5.3 is used here bringing some improvements compared to the
previous version 5.2 used in . As in the version used in
, the long-wave (LW) radiation scheme is now based on the rapid
radiation transfer model (RRTM, ), while the short-wave (SW)
scheme initially developed by has a finer spectral resolution
(six bands). We also use here a spectral nudging method described in
, which enables us to keep large scales from the boundary
forcing and thus impose the true natural climate variability that is
essential to represent dust events notably. Here the wind vorticity and
divergence, the surface pressure, the temperature and the specific humidity
are nudged. The function used imposes a constant rate above 700 hPa and a
relaxation zone between 700 and 850 hPa, while the levels below 850 hPa are
free. The spatial wavelengths are similarly nudged beyond 400 km, with a
relaxation zone between 200 and 400 km. Thus this method gives the model
enough freedom to generate the aerosols at the surface while keeping the
large scale conditions that are essential to simulate the true chronology.
The ocean model NEMOMED8 and the land surface model ISBA are the same models
as used in . The ocean–atmosphere coupling is achieved by the OASIS3 coupler at
a 3 h frequency, which represents an improvement compared to CNRM-RCSM4 described in . Note finally that
contrary to CNRM-RCSM4, the coupling to the river routine scheme is not included in the present version of CNRM-RCSM5.
Stations of the AERONET network (black crosses, see the list
of the corresponding numbers in Fig. ). Red crosses indicate the
stations providing measurements of surface radiation and temperature (see the list in Table ).
The aerosol scheme in ALADIN-Climate
Until the version 5.2 of ALADIN-Climate aerosols were represented in this model through
monthly climatologies of aerosol optical depth for five aerosol types (desert dust,
sea salt, black carbon, organic matter and sulphate) distributed vertically according to
constant profiles. In the version 5.3 used here, a prognostic aerosol scheme has been
included, adapted from the GEMS/MACC aerosol scheme . It includes
the same five aerosol species that can be directly emitted from the surface for dust
and sea-salt particles or from external emission data sets for black carbon, organic
matter and sulphate precursors. The spatial domain of our simulations has consequently
been extended compared to the previous study of in order to include all
the sources generating aerosols that can be transported over the Mediterranean basin. As
far as dust particles are concerned , the following sources are notably
included in the domain: North African sources (Morocco, Algeria, Tunisia), the Hoggar
mountains, the Tibesti Mountains, the Bodélé depression, Libya, Egypt and sources
near the Red Sea (northeast Sudan, Djibouti). No aerosol is included in the lateral boundary forcing.
Sea-salt aerosols are generated by wind stress on ocean surface either because
of air bubbles bursting at the sea surface or from spume droplets directly torn off
the wave crests by the wind. have reviewed different approaches to
model these processes. The current formulation used in ALADIN-Climate is based on
the studies of and that provide surface mass fluxes
at 80 % relative humidity depending on 10 m wind, integrated for the three size bins
defined in the scheme: 0.03 to 0.5, 0.5 to 5 and 5 to 20 µm.
Note that the size distribution of emitted sea salt also depends on other factors, such as the sea surface
temperature , which are not taken into account in this current version.
Dust emission processes depend on several factors such as soil characteristics (chemical
composition, humidity, roughness) and surface wind speed. In the GEMS/MACC
scheme, the dust parameterization follows , who propose a
simplified formulation of dust emission based on the wind speed and
thresholds according to the fraction of bare soil and soil moisture. In
ALADIN-Climate, this function has been replaced by the
parameterization that takes into account more soil characteristics coming
from the ECOCLIMAP database , which provides information on the
erodible fraction and the sand and clay fractions, allowing a classification
of the soil textures. After the determination of an erosion threshold based
on the soil distribution, the soil moisture and the roughness caused by
nonerodible elements, the horizontal saltation flux is calculated
proportionally to the third power of the wind friction velocity. The vertical
flux is then inferred from this saltation flux, according to an empirical
relationship given by , which notably depends on the soil clay
content. The emitted dust size distribution is based on the work of
. More details about this dust emission parameterization can be
found in , who have used the same dust emission scheme in
RegCM4. Once emitted dust particles are integrated in the three dust size
bins of the scheme: 0.01 to 1.0, 1.0 to 2.5
and 2.5 to 20 µm.
The external emission data sets for the three other aerosol types come from ,
who have provided inventories at 0.5∘ resolution of different species
for climate models. These inventories include numerous sectors such as energy
production, industries, domestic activities, agriculture, transport and
fires. Organic and black carbon particles are separated between hydrophile
and hydrophobic particles. SO2 emitted particles can be transformed in
SO4, but 5 % of them are directly emitted as SO4 aerosols
. Volcanic sulfur emissions are also included, as well as
dimethylsulfide particles from oceans see.
All these aerosols gathered in 12 bins are then transported in the atmosphere
before possible dry or wet deposition. More details about transport and
deposition can be found in . Optical properties (single
scattering albedo and asymmetry factor) are fixed for each aerosol type, as
defined in . The complexity of this aerosol scheme is similar to
the one used in RegCM, but it does not include detailed chemical processes
that can be found in COSMO-ART . However, it enables our model
to keep a low cost of calculations so that multi-annual simulations could be
carried out for aerosol–climate studies. Note also that nitrate aerosols are
not considered in this model.
Simulations
Three simulations have been carried out with CNRM-RCSM5, driven by the
ERA-Interim reanalysis as initial and lateral boundary
forcing. First of all, the PROG simulation includes the whole aerosol
prognostic scheme described previously. Secondly, in order to estimate the
effect of aerosols on meteorological variables such as temperature and
radiation, a simulation without aerosols is needed: the NO simulation does
not include any aerosols. Thirdly, as the objective of this study is also to
discuss the choice of using climatological or prognostic aerosols, another
simulation, called PROG-M, uses monthly AOD provided by PROG so that PROG
and PROG-M share the same average aerosol content at the monthly scale.
Comparisons between these simulations will enable us to estimate the aerosol
effects on the radiative budget and regional climate and the implications of
using a prognostic aerosol scheme instead of monthly climatologies. While an
improvement on daily SW radiation variability is expected with the use of
prognostic aerosols, it is more difficult to answer a priori for other daily
parameters, 2 m temperature (T2m) and sea surface temperature (SST), and more generally for consequences on the
summer average. The three simulations cover the summer 2012 period from 1
June to 31 August. A 1-month spin-up period has been performed for each
simulation in order to have realist aerosol concentrations on 1 June.
Observation data
Stations used for the composite study. The
total number of days when observations are available and among them the number of dusty days have been indicated.
Short name
Station
Lat
Long
Available days
Dusty days
MUR
Murcia
37.8
-0.8
83
23
BAR
Barcelona
41.3
2.1
85
10
MAL
Palma de Mallorca
39.6
2.6
74
13
ALI
Alicante
38.3
-0.6
90
15
AJA
Ajaccio
41.6
8.5
88
7
CAR
Carpentras
44.1
5.1
84
4
MON
Montpellier
43.6
4.0
75
7
NIC
Nice
43.7
7.2
88
4
PER
Perpignan
42.7
2.9
80
6
FES
Fès
33.9
-5.0
61
36
LIO
Gulf of Lions (buoy)
42.1
4.6
83
9
AZU
Azur (buoy)
43.4
7.8
78
5
LAM
Lampedusa
35.5
12.6
89
24
SED
Sde Boker
30.9
34.8
92
5
For the evaluation of the aerosols and their direct radiative effects, different observed data sets are used in the present work.
Simulated AOD is compared to satellite data from the MODerate resolution Imaging
Spectroradiometer (MODIS, collection 5.1, standard and Deep Blue algorithms,
1∘ resolution; ), the Multiangle Imaging
SpectroRadiometer (MISR, Level3; ) and the SEVIRI
radiometer onboard the geostationary satellite Meteosat Second Generation. For the latter instrument, we use the algorithm of ,
which provides high-resolution AOD over both ocean and land surfaces.
Nowadays, this algorithm is being implemented on the production chain of the
ICARE thematic centre (http://www.icare.univ-lille1.fr) under the name of
AERUS-GEO Aerosol and surface albEdo Retrieval Using a directional
Splitting method; application to GEO data by, a daytime
averaged product.
Ground-based observations from 30 stations of the AErosol RObotic NETwork (AERONET, )
will also be considered (Fig. ). These sun-photometer observations provide high-quality data (Level 2.0),
which have been downloaded from the AERONET website (http://aeronet.gsfc.nasa.gov). All AOD data have
been calculated at 550 nm using the Ångstrom coefficient when necessary to make comparisons and evaluation easier.
The TRAQA campaign has also provided interesting observations for dust
aerosols, namely vertical profiles from lidar instruments in Barcelona and
San Giuliano (Corsica). The Barcelona lidar system is part of the
ACTRIS/EARLINET network (Aerosols, Clouds, and Trace gases Research
InfraStructure Network/European Aerosol Research Lidar Network,
). The extinction coefficient profiles were retrieved by
means of the two-component elastic lidar inversion algorithm constrained with
the AERONET sun-photometer-derived AOD . In San Giuliano
(42.28∘ N, 9.51∘ E), aerosol vertical profiles were
acquired with a 355 nm backscattering lidar. The aerosol extinction
coefficient profiles are estimated using the Klett's method and a fixed lidar
ratio from hourly averaged attenuated range-corrected lidar
signals. Additionally, an ATR-42 research flight operated by SAFIRE (Service des
avions français instrumentés pour la recherche en environnement) has also
been realized during the TRAQA campaign. This study uses the airborne data
from the Passive Cavity Aerosol Spectrometer Probe (PCASP), which measures
particles between 0.1 and 3.2 µm.
In addition, the Météo-France and AEMET networks have provided daily
radiation and 2 m temperature measurements (see Fig. and
Table ). Radiation measurements have been completed by the
stations of Sde Boker (SED, SolRad-Net network, AERONET website), Lampedusa
(LAM, coll. ENEA) and two Météo-France buoys located in the Gulf of
Lions (LIO) and near the French Riviera (AZU). Lampedusa and the two buoys
also provide SST measurements. All 14
stations providing surface radiation and temperature have been added in
Fig. (red crosses). It is worth mentioning that available data
are
provided by stations that are located for most of them in the western
Mediterranean. However, in summer most of the dust outbreaks occur in this
region because of frequent low-pressure systems over Morocco that favour the
dust export over the western Mediterranean .
Additionally, the MACC reanalysis is also used in the present work as a
means of evaluating the CNRM-RCSM5 simulations. This reanalysis
includes data assimilation of AOD from the MODIS instrument.
Evaluation of the simulated aerosols
In this section, an evaluation of the simulated aerosols during summer 2012 is carried
out against different available observations and climatologies. Depending on the
parameter, several types of data sets are indeed required.
Total AOD: spatial evaluation
Mean aerosol optical depth at 550 nm in summer 2012
(JJA) simulated by CNRM-RCSM5 and MACC (top) and measured by three satellite instruments (MODIS, MISR and AERUS-GEO, bottom).
Spatial correlation coefficients between AOD of the
different data sets presented in Fig. .
Data sets
MODIS
MISR
AERUS-GEO
MACC
CNRM-RCSM5
0.64
0.77
0.65
0.74
MODIS
0.81
0.69
0.84
MISR
0.68
0.84
AERUS-GEO
0.61
The AOD spatial distribution is firstly evaluated against different satellite
products (MODIS, MISR and AERUS-GEO). The average total AOD in summer 2012
for each data set is shown in Fig. . The general spatial
pattern shows a good agreement between satellites and CNRM-RCSM5. The highest
values (up to 1.5) are indeed found over northern Africa and Arabian
peninsula while the Mediterranean Sea is affected by moderate AOD, ranging
from 0.15 to 0.3, from the north-east to the south-west.
In greater detail, some differences can be noted between the model and
satellite data. CNRM-RCSM5 AOD is closer to MISR over northern Africa, where
a large zone of AOD higher than 0.5 can be identified in both data sets, while
MODIS and especially AERUS-GEO show lower AOD. Similar conclusions can be
drawn for the Arabian peninsula. Dust export over the Atlantic Ocean is, on
the contrary, in very good agreement between the five products (AOD between
0.5 and 0.7). Over western and central Europe, MISR AOD is lower than MODIS,
AERUS-GEO and CNRM-RCSM5. Large differences in AOD are also present in
Eastern Europe and Russia, where MODIS shows higher AOD than the other
data sets. However, this region is in the limit of the domain seen by SEVIRI
(lower values in AERUS-GEO) and is also close to the border of the domain
used in CNRM-RCSM5, so that aerosols over this region may come from outside
the domain. Finally, AOD over the northern Atlantic Ocean is higher in
CNRM-RCSM5 than in satellite products, but the presence of numerous clouds in
this area limits the quality of the satellite data there.
In summary, Table presents the spatial correlations between these
four products. All the correlations are higher than 0.6, confirming the general agreement
and the ability of CNRM-RCSM5 to reproduce the main spatial patterns of AOD.
Total AOD: temporal evaluation
AOD (at 550 nm) temporal series between 1 June and 31 August 2012
simulated by CNRM-RCSM5 (red lines) and MACC (orange lines) and observed by AERONET sun photometers (black crosses), MODIS (blue points) , MISR (purple points) and AERUS-GEO (green points) at four
stations of the AERONET network: Oujda (a, number 10 in Fig. ), Mallorca (b, 2), Frioul (c, 8) and Lampedusa (d, 1).
Taylor diagram evaluating CNRM-RCSM5 (red), MACC (orange)
and satellite (MODIS in blue, MISR in purple and AERUS-GEO in green) data against 30 AERONET ground-based observations
in terms of daily AOD in summer 2012. Averages over the 30 stations for each data set are indicated
with crosses. The mean bias against AERONET is indicated in the caption between brackets (all 30 stations/9 stations located to the north of 33∘ N).
As far as the temporal dimension is concerned, an evaluation has been realized
against ground-based measurements from the AERONET network in the
Mediterranean area in terms of daily means. Indeed, AERONET measurements
benefit from a higher temporal resolution than data from moving satellites
and their accuracy is generally higher, about ±0.01
compared to ±0.05 for satellites . Figure
shows four temporal series across the Mediterranean basin
at Oujda (a, Morocco, number 10 in Fig. ), Mallorca
(b, Spain, 2), Frioul (c, France, 8) and Lampedusa (d, Italy, 1). All these
series show high daily variability because of frequent dust outbreaks in
this season. The spectral nudging technique used in CNRM-RCSM5 enables the
model to reproduce the true chronology of the synoptic meteorological
conditions as shown in , which is useful for driving dust
emission in the present work. As a result, the model is able to reproduce
the intensity and the chronology of most AOD peaks, such as those observed in
Oujda (18 June, 25 July) in Mallorca (19 June, 9 July, 10 August),
Frioul (28 June, 19 August) and Lampedusa (21 June, 13 August).
However, CNRM-RCSM5 overestimate a few dust events (e.g. 19 June in Frioul,
15 June in Lampedusa), but these differences remain in the minority.
Similar comparisons have been realized for 30 AERONET stations (see their locations
in Fig. ), the results are presented in a Taylor diagram (Fig. , adapted to daily time series from ). This
diagram represents three statistics: the correlation coefficient is the
azimuth angle, the radial distance from the origin is the standard deviation
normalized by observations, and the distance to the “REF” point on the x axis
is the root-mean-square error (RMSE). The average temporal correlation
coefficient for CNRM-RCSM5 is 0.70, while the ratio between simulated and
observed standard deviations is 1.01, revealing the ability of the aerosol
scheme to reproduce AOD daily variability. In addition, CNRM-RCSM5 has no
station with very low scores and has a low mean bias both when considering
all 30 stations (0.02) and only the stations to the south of 33∘ N
(0.03).
Additionally, the daily values for the satellite products have been added in Figs.
and as information for data users. It is indeed important to
note that in terms of daily variability, (1) MODIS and AERUS-GEO have a
higher temporal correlation with AERONET (0.73 and 0.76 respectively) than MISR
(0.15), probably because of a reduced number of available retrievals with
this instrument; (2) AERUS-GEO has the best scores among the satellite
products; (3) MODIS and AERUS-GEO have, however, respectively 5 and 3 stations
with RMSE higher than 1.25; and (4) all these products have a higher mean
bias than CNRM-RCSM5.
Contribution of aerosol species to AOD
Mean aerosol optical depth at 550 nm in summer 2012 (JJA) simulated
by CNRM-RCSM5 for the five aerosol types (sea salt, desert dust, organic matter, black carbon and sulphate).
Aerosol extinction coefficient simulated by
CNRM-RCSM5 (full black lines)
and observed by a ground-based lidar (dotted black lines) in Barcelona on 27 June at 12:00 UTC (left)
and in San Giuliano (Corsica) on 30 June 2012 at 12:00 UTC (right). The different coloured
lines represent the contribution of each aerosol type to the extinction coefficient.
Total AOD and components for the five aerosol types simulated by CNRM-RCSM5
and the MACC reanalysis in summer 2012 over Europe (continental area up to 30∘ E), the Mediterranean
Sea and northern Africa (continental area up to 25∘ N). Averages in summer from NAB13,
the climatology of , have also been indicated with the minimum and maximum summer
values (period 2003–2009). Total AOD from satellite data (MODIS, MISR, AERUS-GEO) is also given.
Europe
CNRM-RCSM5
MACC
NAB13
MODIS
MISR
AERUS-GEO
Sea salt
0.01
0.02
0.00 [0.00–0.00]
–
–
–
Desert dust
0.04
0.06
0.05 [0.04–0.05]
–
–
–
Organic matter
0.04
0.02
0.02 [0.02–0.03]
–
–
–
Black carbon
0.01
0.01
0.01 [0.01–0.01]
–
–
–
Sulphate
0.08
0.10
0.10 [0.08–0.12]
–
–
–
Total
0.18
0.21
0.18 [0.16–0.20]
0.16
0.15
0.15
Mediterranean
Sea salt
0.01
0.02
0.01 [0.00–0.01]
–
–
–
Desert dust
0.11
0.10
0.12 [0.10–0.13]
–
–
–
Organic matter
0.03
0.02
0.01 [0.01–0.02]
–
–
–
Black carbon
0.01
0.01
0.01 [0.00–0.01]
–
–
–
Sulphate
0.07
0.09
0.08 [0.07–0.10]
–
–
–
Total
0.23
0.24
0.23 [0.19–0.25]
0.20
0.22
0.18
Africa
Sea salt
0.00
0.01
0.00 [0.00–0.00]
–
–
–
Desert dust
0.37
0.18
0.31 [0.25–0.33]
–
–
–
Organic matter
0.02
0.02
0.01 [0.01–0.02]
–
–
–
Black carbon
0.01
0.01
0.01 [0.01–0.01]
–
–
–
Sulphate
0.05
0.07
0.08 [0.06–0.09]
–
–
–
Total
0.45
0.29
0.41 [0.33–0.44]
0.33
0.32
0.21
Satellites and ground-based measurements do not provide the contribution of
the different aerosol types to AOD (the distinction between coarse and fine
modes is not sufficient), which is the reason why a comparison to the MACC
reanalysis and the AOD climatology from ,
named NAB13 thereafter, is presented in this section. Note that total AOD of
NAB13 corresponds to MODIS AOD by definition of this product and that the
total AOD of MACC has been added in Figs. , ,
and Table as information for data users.
Figure presents the mean AOD for summer 2012 for the five simulated aerosol
types. Dust aerosols prevail in the southern part of the domain because of sources in Sahara and
in the Arabian peninsula, while anthropogenic particles, especially sulphate and organic matter,
are responsible for local maxima in AOD in Europe. Sea-salt particles are essentially simulated
over the Atlantic Ocean, as well as the western Mediterranean Sea in lower quantities.
The different contributions to AOD for each aerosol type are given in
Table for CNRM-RCSM5, MACC and NAB13. NAB13 is based
on both model and satellite data, and MACC is based on model and data
assimilation. NAB13, which gives reliable estimations of
the different AOD components, is only available on the 2003–2009 period, so
that the average over this period with the minimum and maximum values have
been indicated. Averages have been calculated on the three domains defined in
: Europe, the Mediterranean Sea and northern Africa.
Over Europe, CNRM-RCSM5 is very close to NAB13 for total AOD (0.18 on average) and the
five aerosol types, even if the sharing between organic matter and sulphate aerosols is slightly
different. MACC simulate more dust and sulphate particles, but the three satellites' data have
lower AOD (between 0.15 and 0.16) so that CNRM-RCSM5 AOD is median. Over the Mediterranean Sea
a good agreement is shown among CNRM-RCSM5 (0.23 for total AOD), MACC (0.24) and NAB13 (0.23).
In addition, the proportion among the different aerosol types is similar in the three data sets.
However, as in Europe, satellite data have lower AOD (between 0.18 and 0.22).
More variability is noted with regards to AOD over northern Africa, notably because of the
dust component. CNRM-RCSM5 shows higher AOD (0.45) than NAB13 (0.41), MACC
(0.32) and the satellite data (between 0.21 and 0.33). However, interannual
variability is stronger in this region as shown by the larger amplitude in
NAB13 (0.33–0.44). Moreover, MACC does not assimilate AOD over the Sahara
because the standard algorithm of MODIS cannot retrieve AOD on bright
surface, so that an underestimation of dust aerosols in MACC had been
identified .
In summary, the evaluation of AOD for each aerosol type is complicated because of the heterogeneity
among the different data sets, but the contribution of aerosol types to AOD in CNRM-RCSM5 is
close to that in MACC and NAB13. It is worth mentioning that CNRM-RCSM5 does not include the
nitrate component. However, dust aerosols constitute the main focus of the following paragraphs.
Dust extinction vertical profile
CNRM-RCSM5 has shown its ability to reproduce AOD daily evolution correctly,
which is a parameter often evaluated in climate models. However, aerosol
direct and semi-direct forcing also depend on the profile and size
distribution of particles, rarely evaluated given the scarcity of
observations and affected by large uncertainties . Even if
total AOD is necessary to evaluate AOD against in situ or satellite
measurements that cannot separate the different aerosol types, more attention
is now given to the dust component which is the focus of this study. The TRAQA
campaign has well documented a dust outbreak over the Mediterranean Sea,
which is useful for this evaluation. However, a deeper evaluation of the other
aerosol components is outside of the scope of this paper.
The dust plume observed in the TRAQA campaign comes from the uplift of dust
particles in western Africa between 21 and 23 June. These particles have been
transported along the African coast up to southern Spain, driven by the
presence of a low pressure system over Morocco and a high pressure area over
the Azores. From 26 June, a low formed in the bay of Biscay generated a
south-westerly flow, bringing the dust plume over northern Spain. Successively
moving to the southeast, dust particles have also been transported over the
Mediterranean Sea. Figure presents the vertical distribution of
aerosols during the dust outbreak observed by lidars in the TRAQA campaign in
terms of extinction coefficient in Barcelona at 532 nm and in San Giuliano at
355 nm. Dust aerosols first reach Spain on 27 June, transported in the
mid-troposphere, as noted in the profile between 2000 and 5000 m with a
maximum extinction (0.18 km-1) at 3500 m. The two-component elastic
lidar inversion algorithm constrained with an AERONET AOD of 0.32 gave a
column-equivalent lidar ratio of 54 sr. This value is in the range of
50–70 sr established by of desert dust lidar ratio
observations by Raman lidar, which makes us confident of the result of the
lidar inversion. The altitude of these dust particles is quite similar in
CNRM-RCSM5 despite an underestimation of the intensity of the dust outbreak
and a slight overestimation in the higher layers. Under this dust layer, the
presence of sulphate aerosols is noted in the model, with an extinction
coefficient close to observations (0.03 km-1). In San Giuliano, where
the dust plume has arrived 3 days later, its altitude is also similar in
CNRM-RCSM5 and observations: between 2000 and 5000 m. As in Barcelona,
extinction is slightly overestimated in the high troposphere (above 6500 m).
In summary, the dust extinction simulated profiles have been evaluated against
these lidar profiles, showing the variability in the altitudes of dust aerosols. It should
also be mentioned that two profiles are not sufficient to conclude on the ability of the
model to estimate the dust vertical distribution. This kind of comparison would need to be
done for other places and situations; however, it is a difficult exercise because evaluating only the
aerosol vertical distribution requires finding cases where adequate observations are
available and where the model correctly simulates the transport of dust aerosols.
Dust vertical size distribution
Dust particle size distribution observed by the PCASP
instrument onboard ATR42 (flight 22) on 29 June at 08:00 UTC (dashed black lines),
the dust refractive index has been adjusted (1.53–0.002i). Full coloured lines
indicate the aerosol concentration for each aerosol bin of CNRM-RCSM5 (red is sulphate, blue is sea salt, orange is dust,
green is organic matter and purple is black carbon), while full black lines indicate the total concentration (kg m-3).
Size distribution is also an essential physical parameter for aerosol–climate
studies, as optical properties depend on the particle size. Figure presents the size distribution observed during a sounding
realized by the ATR42 during the TRAQA campaign as well as the simulated
distribution. Note that the bin scheme used in CNRM-RCSM5 does not enable the
model to reproduce exactly the observed distribution, but the division in three bins for dust particles notably can still be evaluated. This sounding took
place in the Mediterranean Sea (43.05∘ N, 9.55∘ E) on 29 June,
when the dust plume has been transported over this area. In the lower layers,
a first maximum is observed in the smallest particles (around 0.1 µm), probably due to sulphate aerosols, as represented by
CNRM-RCSM5. The observed distribution shows that mass concentration is higher
for larger particles, especially between 2000 and 4000m, where dust aerosols
are located. This distribution is simulated by CNRM-RCSM5, notably between
2000 and 3000 m. Above 3000 m, coarse particles (larger than 2.0 µm) are underestimated. However, these particles have less impact on
extinction in SW radiation than submicronic particles, but they could play a
role in other processes (e.g. deposition).
These results finally show that the aerosol vertical and size distributions
simulated by CNRM-RCSM5 reproduce the main patterns seen in observations from the TRAQA
campaign, even if the simulated profile in Barcelona shows an underestimated extinction
peak between 3 and 5 km in altitude.
To summarize, we have shown in this section the strengths and the weaknesses of CNRM-RCSM5
to simulate the evolution of aerosols during summer 2012 in terms of spatial pattern and
daily variability, as well as the vertical profiles and size distribution of dust particles.
This model will be used in the following section to study the impact of dust outbreaks on
meteorological parameters (radiation, temperature) in summer 2012. In addition, an
intercomparison modelling study about this dust event observed in the TRAQA campaign
will be the subject of a parallel study led by Sara Basart.
Aerosol radiative effects
As seen previously in the AOD temporal series, the Mediterranean basin has
been affected by frequent dust outbreaks in summer 2012. This section aims at
assessing their impact on different meteorological parameters.
Direct radiative forcing (DRF)
Aerosol SW direct radiative forcing (DRF): (a) Average in
summer 2012 for PROG (colours) and the PROG-PROG-M difference (white lines, interval
is 5 Wm-2). (b) Standard deviation of daily DRF for PROG (colours). The white line
indicated the region where the ratio between the standard deviations of PROG and PROG-M is higher than 2.
Figure first shows the daily direct SW DRF
of aerosols in PROG. DRF is calculated online during the simulation, calling
twice the radiation code: with and without aerosols. A negative forcing of
aerosols at the surface is noted. It is stronger over regions under dust
influence – northern Africa, Arabian peninsula and the tropical Atlantic
Ocean – reaching -20 to -50 Wm-2, in line with . Over
Europe and the northern Atlantic, aerosol DRF ranges from -10 to -15 Wm-2
notably because of sulphate aerosols. Compared to estimations from literature
such as the studies of and , who have
found an average DRF of -30 and -26 Wm-2 respectively in Lampedusa, the
values given by CNRM-RCSM5 have the same order of magnitude even if they can
reach larger forcings. Also note that the Atlantic Ocean off Africa, under
the influence of dust export, shows the highest variability.
At the daily scale
Cloud cover (%, green bars
for PROG, curves for the other simulations), 2 m temperature (∘C, curves), cloud cover (%, green bars for PROG, curves for the other simulations), downward SSR (Wm-2, curves) and AOD
(green bars for PROG, blue line for PROG-M), from top to bottom, in Lampedusa (Italy) for
PROG (green), PROG-M (blue), NO (purple), ERA-Interim (black) and observations (dashed red).
As dust aerosols can interact with solar and thermal radiation, consequences
on meteorological parameters such as surface radiation and temperature might
be expected. In the present work, an effort has been made to gather
colocalized measurements of AOD, SW radiation and 2 m temperature or sea
surface temperature. The list of the 14 corresponding stations in the
Mediterranean basin used in this study is presented in Table .
Daily series of solar surface radiation (SSR), cloud cover and surface temperature are presented
in details for two stations representative of the Mediterranean basin, namely Lampedusa (LAM)
and the buoy in the Gulf of Lions (LIO). Lampedusa is located in an island close to
dust-emitting regions where clear-sky conditions are frequent in summer, while LIO is in
the northwestern Mediterranean, where more clouds are observed. Figures
and present respectively in LAM and in LIO the daily series of AOD,
downward SSR, cloud cover and surface temperature (2 m temperature and SST respectively),
observed and simulated by PROG, PROG-M and NO.
Same as Fig. but for the buoy in the Gulf of Lions (LIO) and SST instead of 2 m temperature.
First of all, NO is the only CNRM-RCSM5 simulation to have a high bias
against observed SSR (+18.0 Wm-2 in LAM, 31.2 Wm-2 in LIO)
compared to PROG-M (-6.0 Wm-2 in LAM, 13.6 Wm-2 in LIO) and PROG
(-3.5 Wm-2, 15.9 Wm-2 in LIO) due to the absence of aerosols in
NO. While the aerosol climatology is enough to reduce the bias in PROG-M,
PROG has the highest temporal correlation (0.87 against 0.81 for NO and 0.85
for PROG-M in LAM), and its standard deviation is the closest to observations
(a ratio of 0.88 against 0.74 both for NO and PROG-M in LAM). Indeed, PROG-M
and NO clearly miss some variations of SSR. When AOD is high (e.g. 21/06,
3–12/07, 29/07, 7/08 in LAM, 19/06, 27/07, 20/08 in LIO), PROG-M and NO
overestimate SSR, especially in case of low cloud cover. Inversely when AOD
is low (e.g. 24/06, 20/07, 10/08 in LAM, 5/06, 27/08 in LIO), PROG-M
underestimates SSR while NO benefits in this case from the absence of
aerosols. ERA-Interim has a monthly aerosol climatology similar to PROG-M
except that the aerosol climatology used in ERA-Interim is
probably less realistic and simulates radiation variations lower than
observed. As a result, the effect of aerosols on surface radiation has been
identified in both stations.
With regards to land surface temperature in LAM and SST in LIO, the three
CNRM-RCSM5 simulations have similar temporal correlations (between 0.72 and
0.73 for LAM, 0.98 for SST in LIO), while PROG-M and PROG
are on average cooler than NO because of the aerosol forcing. Even during dust outbreaks, it is not
possible to state that average temperature in PROG is closer to observations.
With regards to standard deviations, the daily variability is reduced in PROG
(0.89 in LAM against 0.92 for PROG-M and 0.95 for NO). The aerosol forcing
during dust events could indeed decrease the maximum daily temperature, while
the effect of dust particles on thermal surface radiation (TSR) could
increase night-time temperature and thus reduce T2m diurnal variability.
Evaluation of daily SSR simulated by NO, PROG-M, PROG and
ERA-Interim against 14 ground-based measurements located around the Mediterranean
basin, in terms of bias (Wm-2), temporal correlation coefficient and standard deviation (SD) ratio.
Bias
MUR
BAR
MAL
ALI
AJA
CAR
MON
NIC
PER
FES
LIO
AZU
LAM
SED
MOY
NO
31.0
2.8
54.3
39.0
18.0
22.3
35.9
37.6
34.7
48.2
31.2
35.1
18.0
5.6
29.6
PROG-M
7.6
-8.5
35.1
18.1
2.4
10.1
20.8
19.6
19.3
13.6
13.6
16.6
-6.0
-13.4
10.6
PROG
9.7
-7.5
36.0
21.2
5.1
11.5
24.0
22.9
21.0
16.5
15.9
19.1
-3.5
-11.7
12.9
ERA-Interim
12.8
4.6
53.7
25.4
-1.0
-4.3
17.0
10.1
27.7
34.7
10.2
7.2
-16.8
-12.9
12.0
Corr.
NO
0.72
0.76
0.66
0.62
0.87
0.89
0.71
0.67
0.76
0.39
0.87
0.86
0.81
0.84
0.75
PROG-M
0.76
0.77
0.65
0.67
0.89
0.87
0.70
0.69
0.77
0.49
0.88
0.87
0.85
0.89
0.76
PROG
0.77
0.79
0.69
0.74
0.89
0.91
0.75
0.69
0.78
0.53
0.89
0.90
0.87
0.90
0.79
ERA-Interim
0.79
0.81
0.88
0.81
0.88
0.88
0.77
0.68
0.75
0.37
0.90
0.76
0.87
0.88
0.79
SD
NO
0.79
1.20
0.84
1.16
1.11
0.96
0.97
0.81
0.96
0.93
0.92
1.00
0.74
1.15
0.97
PROG-M
0.79
1.10
0.82
1.11
1.10
0.86
0.92
0.81
0.93
0.94
0.91
1.04
0.74
0.99
0.93
PROG
0.95
1.16
1.01
1.20
1.17
0.94
0.98
0.88
0.99
1.01
1.01
1.12
0.88
1.07
1.03
ERA-Interim
0.58
0.72
0.69
0.78
0.78
0.77
0.69
0.72
0.63
0.53
0.61
0.90
0.67
0.92
0.71
In order to confirm these results in the other stations, the evaluation of
surface radiation and 2 m temperature for the three simulations and the
ERA-Interim reanalysis in the 14 stations is presented respectively in Tables and . As far as radiation is concerned, the bias
is reduced both in PROG and PROG-M, reaching a level close to ERA-Interim
(between 11 and 13 Wm-2). A net improvement is noted in temporal
correlation, since it is higher in PROG than in PROG-M and NO in every
station. Daily variability in SSR is also higher in PROG for most stations,
representing an improvement compared to observations except where this
variability was already overestimated (e.g. Ajaccio). It is worth mentioning
that in Sde Boker PROG gets closer to observations by reducing SSR
variability. A misrepresentation of cloud processes could also explain some
of the discrepancies with observations. The lack of cloud cover in CNRM-RCSM
shown in could explain the remaining bias. ERA-Interim, which
does not have the daily aerosol variations and consequently misses some peaks
in surface radiation, succeeds in getting a high average correlation
coefficient (0.79) probably because of a better representation of clouds.
Moreover, changes in water vapour column amount may also affect the SSR to a
lesser extent.
Evaluation of daily 2 m temperature simulated by NO, PROG-M, PROG
and ERA-Interim against 13 ground-based measurements located around the Mediterranean basin,
in terms of bias (∘C), temporal correlation coefficient and standard deviation (SD) ratio.
Bias
MUR
BAR
MAL
ALI
AJA
CAR
MON
NIC
PER
FES
LIO
AZU
LAM
MOY
NO
0.3
-1.6
1.2
-0.5
-1.5
0.9
-1.5
-0.0
-2.0
0.0
0.6
1.6
-0.4
-0.3
PROG-M
-0.6
-1.7
0.8
-0.7
-1.7
0.8
-1.7
-0.3
-2.2
-0.4
0.4
1.4
-0.8
-0.5
PROG
-0.8
-1.9
0.7
-0.8
-1.8
0.8
-1.7
-0.3
-2.2
-0.4
0.4
1.4
-0.8
-0.6
ERA-Interim
-2.7
-2.8
-1.2
-0.1
0.1
-2.8
-1.3
-1.4
-1.6
-0.9
0.4
0.6
-0.5
-1.1
Corr.
NO
0.76
0.87
0.91
0.76
0.88
0.92
0.77
0.79
0.87
0.91
0.97
0.82
0.96
0.86
PROG-M
0.77
0.89
0.92
0.77
0.88
0.92
0.77
0.81
0.88
0.92
0.97
0.81
0.96
0.86
PROG
0.76
0.88
0.92
0.75
0.88
0.92
0.77
0.80
0.89
0.92
0.97
0.81
0.96
0.86
ERA-Interim
0.88
0.98
0.88
0.75
0.86
0.92
0.89
0.90
0.89
0.96
0.93
0.81
0.90
0.89
SD
NO
1.36
1.09
1.25
1.44
1.45
1.16
0.90
1.42
1.37
0.96
1.14
1.08
0.97
1.20
PROG-M
1.31
1.10
1.26
1.38
1.45
1.15
0.87
1.41
1.37
0.96
1.10
1.05
0.95
1.18
PROG
1.27
1.04
1.20
1.34
1.42
1.12
0.87
1.36
1.35
0.97
1.08
1.03
0.93
1.15
ERA-Interim
1.04
0.76
0.92
1.36
1.05
1.03
0.93
0.98
0.82
0.88
0.95
0.99
1.00
0.98
As far as surface temperature is concerned, no change in correlation coefficient is noted. The PROG
simulation is cooler than NO and PROG-M, increasing the negative bias. Nevertheless the daily
variability is slightly reduced, getting closer to observed variability. In addition, it is
worth mentioning that ERA-Interim has the highest scores in terms of correlation and
variability (standard deviation), probably benefiting from the assimilation of surface temperature .
As a result, these comparisons show that the prognostic aerosol scheme used in PROG
enables the model to better reproduce the evolution of surface radiation, which cannot be
done properly with an aerosol climatology. Besides, no improvement has been shown in the
scores of land and sea surface temperature. However, aerosol maxima over the Mediterranean
could be associated to particular weather conditions which are responsible for effects on
radiation and temperature that are not due to aerosols. That is the reason why a composite
study to isolate the effect of dust aerosols is carried out in the following section.
Composite analysis
Methodology
This section aims at highlighting the simulated and observed differences
between days of high aerosol load and the set of all the days in terms of
several meteorological parameters (radiation, temperature, cloud cover, etc.).
For the 14 stations defined previously, the days of high AOD, called
thereafter “dusty” days as dust aerosols are mostly responsible for these AOD
maxima, have been selected over the 92 days of summer 2012
(June–July–August). A day is considered as a dusty day provided that observed
AOD is higher than 0.2 and that simulated dust AOD in PROG is higher than
0.2. Days when observations were not available have been removed.
Average differences for several parameters have then been calculated between the dusty days
and the set of all the days for the three simulations (NO, PROG-M and PROG) and observations. The
differences obtained for NO will enable us to estimate the meteorological effect due only to changes
in weather parameters (cloud cover, wind, etc.) without considering the aerosols; those for PROG-M can estimate the average
effect of having an aerosol climatology; and those for PROG can estimate the added value of prognostic aerosols. The
objective is to isolate the effects of aerosols from weather changes that are systematically
observed during dust outbreaks. This method is first presented for the station of Murcia,
whose results are representative of the whole Mediterranean basin, and then generalized to the 14 stations.
Case of Murcia
Composite study for Murcia: differences between dusty days and
the set of all the days in observations (OBS), NO, PROG-M and PROG for AOD, downward SSR
(Wm-2), cloud cover (%), downward TSR (Wm-2), 2 m temperature (∘C) and
soil temperature (Ts, ∘C). The contribution of the different effects, namely
weather, aerosol (mean) and aerosol (variability), have been added.
Parameter
OBS
NO
PROG-M
PROG
Weather
Aerosol (mean)
Aerosol (var)
AOD
0.15
0.00
-0.01
0.19
0.00
-0.01
0.20
SSR
-19
-6
-7
-22
-6
-1
-15
Cloud cover
–
1
2
1
1
1
-1
TSR
–
10
9
11
10
-1
2
T2m
1.2
1.6
1.5
1.3
1.6
-0.1
-0.2
Ts
–
1.5
1.4
1.2
1.5
-0.1
-0.2
In Murcia, 23 days have been identified as dusty days over the 83 days when
observations are available; results are presented in Table .
First of all, as expected, the difference in AOD between dusty days and the
set of all the days is clearly positive in the PROG simulation (0.19), very
low in PROG-M (-0.01) but not necessarily zero, because the number of dusty days
varies from 1 month to another (AOD is monthly constant in PROG-M), and
equal to zero in NO (no aerosols). This difference in AOD is similar in the
observations and PROG, confirming the ability of CNRM-RCSM5 to reproduce
aerosol daily variability and making the comparison for other
parameters possible. The higher AOD during dusty days leads to a decrease in downward
SSR. The difference with the set of all the days reaches -22 Wm-2
against only -6 and -7 Wm-2 for NO and PROG-M respectively, while
measurements in the station show a difference of -19 Wm-2. The
difference in NO (-6 Wm-2) can be considered as the ”weather effect”
that is due to the choice of the days (meteorological and astronomical
variations). The duration of sunshine indeed varies during summer and
reaches its maximum at the solstice (21 June), which can explain a part of
the radiation differences in NO, in addition to changes in cloud cover.
PROG-M, which has a monthly climatology of aerosols, is useful to identify
changes in atmospheric circulation and cloud cover due to a monthly
climatology of aerosols (-1 Wm-2). The difference between PROG-M and
PROG gives the contribution of the daily variability of aerosols that is
necessary to reproduce observed radiation measurements. Few changes among
the three simulations are observed in cloud cover and TSR.
Average wind (km h-1, coloured barbs) and geopotential (mgp, black lines)
at 850 hPa for the set of all the days (left) and the dusty days (right) defined in Murcia (purple cross).
Temperature is also affected by weather changes, as dusty days are 1.6 ∘C higher in
NO than the set of all the days. This is probably explained by the predominance of stronger
southern fluxes during dusty days that can transport aerosols from Sahara to the Mediterranean
basin. Figure indeed shows the average circulation at 850 hPa during dusty
days and the set of all the days, indicating a reinforcement of south-westerly winds in
southern Spain advecting warm air. However, this increase in temperature during dusty
days is lower in PROG than in PROG-M and NO, which is closer to observed variations of
temperature. This decrease of -0.2 ∘C between PROG-M and PROG is caused by dust aerosols
that have reduced incoming solar radiation. In other words, without prognostic aerosols the
warming simulated by CNRM-RCSM during dusty days is too strong compared to observations, which
is corrected in PROG. A similar impact is observed in soil temperature.
As a result, radiation and temperature in Murcia have been shown to be better reproduced
in the PROG simulation, showing the added value of a prognostic scheme compared to monthly
climatologies to reproduce local meteorological variations.
Generalization
Average AOD (a), downward SSR (b), cloud cover (c),
downward TSR (d), 2 m temperature (e) and soil temperature (f)
differences
between the dusty and the set of all the days in 14 stations (presented in Table )
in summer 2012 for the NO, PROG-M and PROG simulations, as well as observations (AERUS-GEO for AOD,
ground-based measurements for the other parameters). For Lampedusa and the buoys in the Gulf of
Lions and Azur, 2 m temperature has been replaced by SST.
A similar composite study has been carried out for other stations (defined in Table ) where
daily radiation and temperature data were available. Figure presents the results per
station for six parameters (AOD, solar and thermal surface radiation, cloud cover, 2 m and soil temperature)
for the NO, PROG-M and PROG simulations, as well as for observations when available, while the average
composites are given in Table .
Same as Table but for the average over the 14 stations defined in Table .
Parameter
OBS
NO
PROG-M
PROG
Weather
Aerosol (mean)
Aerosol (var)
AOD
0.22
0.00
0.00
0.21
0.00
0.00
0.21
SSR
-23
-2
-5
-23
-2
-3
-18
Cloud cover
–
-2
-1
-2
-2
1
-1
TSR
–
12
12
14
12
0
2
T2m
1.4
1.7
1.7
1.5
2.0
0.0
-0.2
Land soil temperature
–
1.7
1.6
1.3
1.7
-0.1
-0.3
SST
1.3
0.9
0.9
0.9
1.3
0.0
0.0
As in Murcia, the difference in AOD between dusty days and the set of all the days is for
every station similar in observations (0.22 on average) and the PROG simulation (0.21). The
difference in PROG-M comes only from the number of dusty days varying from 1 month to another.
As a consequence, measurements reveal that downward SSR is on average 23 Wm-2 lower during
dusty days, which is correctly reproduced by PROG (-23 Wm-2). A part of this decrease (-2 Wm-2)
is explained by weather changes as simulated by NO, while added an aerosol climatology does not bring
significant differences (-3 Wm-2). Additionally, the decrease of SSR in dusty days varies from one
station to another (ranging from -2 to -53 Wm-2). The amplitude of the increase in AOD on dusty
days and changes in weather conditions explain this variability. For example in Mallorca, an increase
of 6 % in cloud cover on dusty days amplifies the dimming due to aerosol loads.
With regards to downward TSR, an average increase of 14 Wm-2 is simulated by PROG on
dusty days, but it is mainly due to weather conditions as NO and PROG-M also show an increase
of 12 Wm-2. Dust aerosols would consequently only represent an increase of 2 Wm-2.
Unfortunately, few LW observations are available. The measurements in the Gulf of Lions and in
Lampedusa show a lower increase than the simulations.
More observations are available for T2m, revealing a general increase of
temperature on dusty days (on average 1.4 ∘C). As in Murcia, this
increase is probably due to warm advection caused by southerly to
south-westerly winds responsible of these dust outbreaks. NO indeed simulates
an average increase of 1.7 ∘C but reduced to 1.5 ∘C in
PROG, indicating the cooling due to dust aerosols, which makes the simulation
closer to observations. This improvement is noted in 10 out of the 13
stations considered in the study (Fig. ) – these 10 stations being
the 9 continental stations and the buoy Azur. The other stations either do
not show a cooling (Ajaccio) or this cooling is not in line with observations (buoy of the Gulf of Lions, Lampedusa).
For these two latter stations, sea surface temperature also increases on dusty days (up to 2.0 ∘C in
the Gulf of Lions in NO), while PROG-M and PROG both alleviate this increase by 0.1 ∘C. However,
this reduction cannot be confirmed by observations. Maybe the 3-month period is not long enough to
identify the daily effects of aerosols on SST. With regards to land soil temperature, a cooling of -0.3 ∘C
due to dust aerosols is simulated by PROG, in relationship with the cooling in T2m.
In fact, this composite analysis has shown that significant differences are observed between
dusty days and the set of all the days, which come both from weather changes (notably due to south-westerly winds
bringing warm air) and from the presence of dust aerosols that alleviate this warming by reducing incoming
solar radiation. These results underline the importance of the use of prognostic aerosols to represent
daily variations in weather parameters such as temperature and radiation.
Impact of daily aerosol variability on the summer average
The question that arises from the impact of aerosols shown on surface radiation
and temperature during dusty days is whether using an aerosol prognostic scheme
instead of a monthly climatology also has an impact on the summer average.
Average difference in summer 2012 between the PROG and PROG-M
simulations in terms of (a) SW surface direct radiative forcing (Wm-2), (b) LW
surface direct radiative forcing (Wm-2), (c) 2 m temperature (∘C)
and (d) sea surface temperature (∘C).
As far as DRF is concerned, average differences in summer 2012 between PROG
and PROG-M are presented in Fig. both for SW (a) and LW (b)
radiation. The intensity of the average aerosol forcing is slightly lower (3 Wm-2) in PROG-M than in PROG for the SW component, while very few
differences are observed for LW radiation. Moreover, the daily standard
deviation of SW DRF is higher in PROG than in PROG-M, particularly over
northern Africa and the Mediterranean Sea, where it is more than twice as
high
(Fig. b). Indeed, dust emission is not a continuous
phenomenon, because it is associated with episodes of strong wind over
northern Africa. Consequently, dust particles show high variability over the
Mediterranean basin that PROG-M cannot take into account, contrary to PROG.
The only daily variations of DRF in PROG-M are due to cloud cover variations,
as the aerosol effect can be partially masked by the presence of clouds.
As a consequence, the aerosol effect on surface temperature is on average slightly
different in PROG-M compared to PROG (Fig. c). The general cooling, due
to the presence of aerosols that scatter and absorb incident solar radiation,
preventing it from reaching the surface, is either reinforced (e.g. in the
south-western Mediterranean) or alleviated (e.g. in eastern Europe) when using an
aerosol interactive scheme instead of a monthly climatology. A similar difference
between PROG and PROG-M is found for SST (Fig. d). These changes
are probably due to the interactions between aerosols and weather conditions. As
seen previously in the composite study, the fact that high dust loads often occur
in southern fluxes could modify their impact on weather and climate. Moreover,
when using an aerosol climatology, the variability of the atmospheric aerosol content
is weaker, and the extreme values of AOD are not represented in the model.
Over the Mediterranean, while frequent AOD peaks are observed in the south-west due
to frequent dust outbreaks, the latter less often reach the Gulf of Lions and hence there are less
frequently AOD peaks there. The AOD standard deviation in PROG is, for example, 0.22 for
the Strait of Gibraltar and only 0.14 for the Gulf of Lions. In addition, there are
more days in the Strait of Gibraltar (32) where AOD is much higher (difference higher
than 0.1) in PROG than in PROG-M and in the Gulf of Lions (15), despite common
averages. Consequently, the aerosol effect can be more important in the Strait of
Gibraltar than in the Gulf of Lions, which must explain a cooler SST in the Strait
of Gibraltar. In addition, the days when AOD is high in the Gulf of Lions are
often cloudy, which alleviate the effect of aerosols. Indeed, dust outbreaks
over the northern basins are more frequent under southerly winds
that also favour humidity advection and cloud cover.
In summary, the choice of using an aerosol prognostic scheme instead of a monthly
climatology has not only an impact on daily weather and climate variability but also
on the summer average. This second impact has never been shown before over the Mediterranean to our knowledge.
Discussion
This study has shown the radiative effects of dust aerosols in summer 2012 over
the Mediterranean, but some points need to be discussed.
First, the choice to focus on a particular summer has been motivated by
the fact that summer 2012 was particularly affected by dust outbreaks. Thus,
a high number of dusty days were noted, providing an interesting case
to estimate the radiative effects of dust aerosols. However, one can wonder if the
results would change during a summer with few dust outbreaks, notably with regards to
the impact of the choice of prognostic aerosols. As a matter of fact, the composite
study and the analysis of the utility of prognostic aerosols should be redone for a
longer period to better understand the interactions between dust aerosols and regional
climate, even if finding adequate observations may represent an obstacle. It would be
also interesting to consider the effects of dust aerosols during the other seasons.
In addition, the choice of using the spectral nudging method may have influenced
the results, as it can be seen as a limitation of the effect of aerosols on the atmosphere.
Indeed, this relaxation towards the ERA-Interim inside the regional domain could, for
example,
prevent aerosols from modifying temperature and humidity profiles above 700 hPa and thus
have stronger semi-direct effects. This point is particularly interesting with
regards to the impact of the choice of prognostic aerosols instead of monthly AOD means.
Nevertheless, the spectral nudging method is essential to represent the real chronology of
dust events, making the comparison to observations possible. With regards to the
uncertainties of the model outputs, they will be more deeply evaluated in a multi-model
exercise currently carried out in the framework of the TRAQA/ChArMEx campaign.
Finally, the low complexity of the aerosol scheme used in the present work could
constitute another limitation. In particular, the low number of bins for dust
aerosols (only three), the absence of detailed processes representing the
formation of secondary aerosols, the choice of a bulk approach for aerosol
modelling and the absence of internal mixing are limitations to the present
work. Future developments on this aerosol scheme will be carried out to
improve the representation of aerosols in the model. For example, the
implementation of the Ångstrom exponent will make the definition of dusty days
for the composite study more robust. However, some of the simplifications
remain necessary to keep a low numerical cost in order to be able to carry
out easily multi-annual climate simulations with a coupling between the
different components of the regional climate system (atmosphere, aerosols,
land surface and ocean). Moreover, this scheme does not take into account the
second indirect effect of aerosols because of the huge uncertainties in their
parameterizations .
Conclusions
A prognostic aerosol scheme has recently been added in the regional climate model
ALADIN-Climate, enabling for the first time a regional coupled system
model (CNRM-RCSM5) including the atmosphere, prognostic aerosols, land surface
and the ocean components over the Mediterranean region. Simulations have been
carried out in summer 2012, first to evaluate the aerosols produced by the model
and then to estimate the radiative effects of dust outbreaks over the Mediterranean region.
CNRM-RCSM5 has shown its ability to reproduce the spatial and temporal variability
of AOD over the Mediterranean region in summer 2012. The general spatial patterns,
notably the locations of regions with high AOD, are in agreement with satellite data,
while the distribution in the main different aerosol types is close to the MACC reanalysis
and the independent climatology from . Daily variability is also correctly
simulated by the model, since the evaluation against 30 stations from the AERONET network
shows a mean bias of 0.02, an average correlation coefficient of 0.70 and an average
ratio of standard deviations of 1.01 as good as satellite data. In addition, the TRAQA
campaign has provided lidar and airborne measurements of a strong dust outbreak that
occurred at the end of June 2012. The aerosol vertical distributions observed in Barcelona
and in Corsica show that the model is able to reproduce the altitude of maximum extinction,
even when a slight overestimation has been noted in the upper troposphere. With regards to dust
size distribution, the three-bin scheme used in ALADIN-Climate simulates higher mass concentrations
for the largest particles, as well as a second maxima for submicronic particles, as observed
during the TRAQA campaign.
The simulated aerosol surface SW DRF is negative, ranging from -10 Wm-2 in Europe
to -50 Wm-2 in Africa, in line with previous studies. However, here the aerosol DRF is
shown to have much variability when using a prognostic aerosol scheme instead of a monthly
climatology. As a consequence, thanks to the prognostic aerosol scheme, downward SSR is
better reproduced compared to ground-based measurements from several stations across the
Mediterranean, both on days of high AOD (lower SSR) and low AOD (higher SSR), as correlation
and standard deviation are improved. The forcing due to the dust outbreaks also causes extra
cooling in surface temperature, but it is insufficient to improve significantly the correlation.
However, the average difference between a simulation using a prognostic aerosol scheme and an
aerosol climatology shows a cooling of 0.1 to 0.2 ∘C both in T2m and SST close to the
dust sources, notably in the south-western Mediterranean. Dynamics can also change in the
two simulations and thus modify surface temperature.
A composite study has been realized in 14 stations across the Mediterranean to
identify more precisely the differences between dusty days and the set of all the days.
During dusty days, SSR is shown to be reduced on average by 28 Wm-2 mostly because
of the dimming of aerosols (-17 Wm-2) but also because of weather conditions (-10 Wm-2).
In parallel, dust outbreaks that are responsible for dusty days also bring warm air, which explains
why T2m is observed 1.6 ∘C higher on dusty days. This warming is too strong (2.0 ∘C)
when considering only an aerosol climatology. The prognostic scheme reduces this average warming
of 0.2 ∘C, getting closer to observations.
Finally, this study has shown the improvement brought by a prognostic aerosol scheme compared
to a monthly climatology in terms of radiation and temperature during a summer. This methodology
could be applied on multi-annual simulations to evaluate the impact of prognostic aerosols at the
climate scale. Differences could be expected not only in terms of variability but also in average
climate as suggested by the differences shown in average SST in summer 2012 in the present work.