During the 72 h intensive measurement period, information from different
models, platforms, and instrumentation was available. A detailed
characterization of the situation above the Mediterranean basin during the
campaign focusing on aerosol microphysical properties using the different
resources available is presented in Sect. 4.1, followed by the model
evaluation in Sect. 4.2.
Spatial–temporal characterization of aerosol microphysical
properties during ChArMEx/EMEP 2012
Ground-based column-integrated measurements
Column-integrated properties retrieved from the AERONET sun photometer are
presented in Fig. 2. Figure 2a and b shows the time series of the τ440nm and AE(440–880 nm) for the selected five stations during
the analyzed period, and mean values for each day and station are indicated
in Table 3.
According to these data, the lowest values of τ440nm were
measured at the Évora station during the whole period, with values below
0.18. The AE(440–880 nm) was close to 1, except in the early morning and
late evening, when it decreased down to ∼ 0.5. These values, together
with the columnar volume size distributions observed in Fig. 2c, indicate a
very low aerosol load, mostly related to aerosol from local sources, and no
impact of the northern African aerosol plume forecast to arrive at the
Iberian Peninsula. A decrease in the τ440nm value with time
was observed at the Granada station, with maximum values reaching up to 0.40
on 9 July around 16:00 UTC. During 10 and 11 July, τ440nm
values were between 0.10 and 0.20, except for the late afternoon of 10 July
from 17:00 UTC, when the aerosol load decreased and τ440nm
below 0.10 were observed. By contrast, values of the AE(440–870 nm)
increased from 0.3 on 9 July up to 0.7 on 11 July, with maximum values on the
late evening on 10 July (AE(440–870 nm) > 1). It is worth noting that
the AE(440–870 nm) was below 0.5 during the whole period except for the
late afternoon on 10 July, coinciding with the decrease in τ440nm, indicating a clear predominance of coarse particles
(e.g., Pérez et al., 2006a; Basart et al., 2009; Valenzuela et al.,
2014). The columnar volume size distributions for the different days agreed
with these data. Data from 9 July show a very large coarse mode and a small
contribution of fine particles. The contribution of fine particles was almost
constant during the 3 days, whereas the coarse mode was decreasing with time.
There was a predominance of the coarse mode during the whole period, with
maximum values of 0.13 µm3 µm-2 during the
first day. All these data are usually related to the presence of mineral dust
in the station and the temporal evolution of the analyzed properties clearly
suggests a decrease in the mineral dust event intensity throughout the
analyzed period and a possible mixing or aging of the mineral dust. At the
Barcelona station no AERONET data were available on 9 July. During 10 and
11 July, τ440nm values were relatively high and quite
constant (around 0.30) and the AE(440–870 nm) values were larger than 1.5,
indicating a strong contribution of fine aerosol particles. In the columnar
volume size distributions, similar values for the fine and coarse modes were
observed on 10 July, but larger values of the fine mode were obtained on
11 July. Therefore, it can be inferred from these data that the impact of the
northern African aerosol plume was almost negligible at this station.
RCS at 532 nm (1064 nm at Athens) in arbitrary units for the five
stations during the ChArMEx 2012 measurement campaign.
![]()
In Athens and Bucharest the aerosol plume presented very different
characteristics to those observed in the western region
(Table 3). In this region, large
τ440nm values (> 0.35) and large values of the
AE(440–870 nm) suggested a situation with high aerosol load mainly composed
of fine particles. At Athens both τ440nm and
AE(440–870 nm) values were very constant during the 3 analyzed days, except
for a slight decrease in the AE(440–870 nm) on 11 July (from ∼ 1.70
to ∼ 1.30). This is in agreement with the columnar volume size
distributions (Fig. 3c), where a slight increase in the coarse mode was
observed on 11 July when compared to 9 and 10 July. In the case of Bucharest,
τ440nm was almost constant on 9 and 10 July (around 0.37),
but increased on 11 July (over 0.60). The AE(440–870 nm) was almost
constant around 1.10 during the 3 days, indicating a balanced presence of
coarse and fine particles despite the increase in the aerosol load during
11 July. The columnar volume size distributions were very similar to those of
Athens on 9 and 10 July, but a larger presence of fine particles was observed
here on 11 July. According to these sun-photometer data, the aerosol plume
over this region was not composed of mineral dust particles, even though low
concentrations of mineral dust might have been advected over Athens on
11 July.
Volume concentration profiles of the total coarse mode and the fine
mode at Barcelona and Athens, and volume concentration profiles of fine,
coarse spherical, and coarse spheroid modes at Évora, Bucharest, and
Granada (from left to right) for different periods of 9, 10, and 11 July 2012
(from top to bottom).
![]()
Aerosol vertical distribution
Figure 3 shows the time series of the lidar range-corrected signal (RCS) in
arbitrary units at 532 nm (at 1064 nm in Athens) for the 72 h period at
the different stations. From these plots, it is clearly observed that at
Barcelona and Évora the aerosol load was mainly confined within the
planetary boundary layer, and the time series reveal the evolution of the
planetary boundary layer height, even though at Barcelona some aerosol layers
are observed in the free troposphere. Therefore, it is expected that most of
the aerosol particles are of local origin. However, at the rest of the
stations a more complex vertical structure was observed and the presence of a
lofted aerosol layer reaching up to 6 km a.s.l. at some periods indicated
the advection of different aerosol types.
The aerosol microphysical properties profiles retrieved with LIRIC for
different periods at the different stations are shown in Fig. 4. That is, the
volume concentration profiles of the total coarse mode and the fine mode were
retrieved at Barcelona and Athens, whereas the volume concentration profiles
of fine, coarse spherical, and coarse spheroid modes were retrieved at
Évora, Bucharest, and Granada because of the availability of
depolarization information.
At Évora it was clearly observed that the aerosol was located below
1000 m a.s.l., within the planetary boundary layer, and concentrations were
very low, ranging from 25 to 46 µm3 cm-3. No advected
aerosol layers were observed for the analyzed period.
At Granada a clear predominance of coarse spheroid particles reaching
altitudes around 6000 m a.s.l. was observed on 9 July, related to the
mineral dust event. A small contribution of fine particles was also observed
during the 3 days. Values of the volume concentration (below
50 µm3 cm-3 for the total concentration) indicate a
medium intensity dust event, which was considerably decreasing with time.
Concentration values around 30 µm3 cm-3 on 9 July for
the coarse spheroid mode went down to values below
20 µm3 cm-3. The altitude of the mineral dust layers was
also decreasing from 6000 to 4000 m a.s.l. for the highest layers.
At the Barcelona site, an aerosol layer dominated by fine particles with a
slight presence of coarse particles was observed between 2000 and
4000 m a.s.l. on 11 July, these coarse particles being possibly related to
a faint presence of mineral dust. The 5-day backward trajectory analysis
performed with the HYSPLIT model (Draxler and Rolph, 2003) (not shown)
indicates that air masses arriving at this altitude came from the north of
Africa through the Iberian Peninsula. This information, together with
previous studies (e.g., Wang et al., 2014), suggests that the mineral dust
plume was moving from the north of Africa towards the northeast, being
detected at Granada and later on at Barcelona. However, the possibility of
these coarse particles being linked to the presence of biomass burning from
the eastern Iberian Peninsula (see Fig. 5) cannot be dismissed.
Depolarization information would be crucial here to discriminate the origin
of the aerosol particles arriving at this height above Barcelona and would
provide very valuable information for the aerosol typing at the station.
MODIS FIRMS image indicating the active fires during the five
previous days to the 11 July 2012. The red line correspond to the air-mass
5-day back-trajectory arriving over Bucharest at 3000 m a.s.l. on 11 July
2012.
![]()
At the Athens station the aerosol reached up to 5000 m a.s.l. and total
concentration values of up to 55 µm3 cm-3 in the free
troposphere. The coarse mode was located below 2000 m a.s.l., whereas a
predominance of fine particles was observed at higher altitudes. The top of
the aerosol layer was increasing with time from 3800 to almost
5000 m a.s.l. This temporal evolution of the microphysical properties is
coherent with the optical properties shown in Sicard et al. (2015) for the
same period. It is worth pointing out that on 11 July, coarse particles were
detected between 3000 and 4800 m a.s.l. at this station, probably related
to the arrival of mineral dust as indicated by the column-integrated values.
Backward trajectory analysis with HYSPLIT (not shown) revealed a change in
the trajectory of the air masses arriving at 3500 m a.s.l., coming from
northern Africa, which would explain the presence of mineral dust on 11 July.
However, according to the trajectories and the different characteristics, the
mineral dust observed at Athens corresponds to a different plume than the one
observed above Granada and faintly above Barcelona.
At Bucharest, a similar volume concentration of fine and coarse particles was
observed on 9 and 10 July, reaching total volume concentration values around
35 µm3 cm-3. The observed coarse particles were
spherical according to LIRIC; therefore, the presence of mineral dust at this
region can be totally neglected. On 11 July a strong increase in the fine
mode volume concentration was observed between 2500 and 5000 m a.s.l., with
values reaching up to 55 µm3 cm-3, suggesting the
advection of an aerosol plume dominated by fine particles at this altitude.
Again, this is in agreement with the optical properties presented in Sicard
et al. (2015), where a larger spectral dependence (related to finer
particles) is observed at Bucharest station in the height range between 3 and
4 km a.s.l. As suggested in the study by Sicard et al. (2015), this large
spectral dependence of the backscatter coefficient could have originated in
the presence of fine particles related to the advection of smoke. The
combined information provided by backward trajectory analysis and MODIS FIRMS
comes to confirm the presence of active fires along the air mass paths
arriving at Bucharest on 11 July (Fig. 5).
(a) Five-day backward trajectories arriving over Granada on
9, 10, and 11 July 2012 at 12:00 UTC (from left to right) computed by the
HYSPLIT model. (b) Locations of the main industrial activity in the
north of Africa (brown stars) taken from Rodríguez et al. (2011) together
with the 5-day backward trajectories arriving at the Granada experimental
site on 9 July 2012 at 12:00 UTC.
![]()
The use of the depolarization information as input in LIRIC in the stations
of Bucharest, Évora, and Granada provided additional information that is
very valuable for aerosol typing. In the cases of Bucharest and Granada, this
information turned out to be very useful for the characterization of the
aerosol types and their distribution in the vertical coordinates. The
differences in the aerosol type were already evidenced in the columnar volume
size distributions retrieved by the AERONET code (Fig. 2), and here LIRIC
confirmed that these two stations presented really different situations. The
volume concentration profiles retrieved with LIRIC indicated a predominance
of the spheroid mode in Granada and a predominance of spherical particles in
Bucharest, highlighting very different aerosol composition in the coarse
mode. However, at stations such as Barcelona or Athens where lidar
depolarization was not measured, ancillary information, e.g., backward
trajectories or sun-photometer-derived optical properties, was needed to
discriminate whether the coarse mode was related to non-spherical particles,
usually associated with mineral dust, or with spherical particles, mostly
present in cases of anthropogenic pollution or aged smoke. Therefore, here we
have a clear example of the importance and the potential of the
depolarization measurements in the vertical characterization of the aerosol
particles and for aerosol typing.
Temporal evolution of the aerosol microphysical property
profiles
The continuous analysis of the aerosol microphysical properties profiles
during the 3 days provided very valuable information about the dynamics of
the aerosol layers and revealed LIRIC's potential to retrieve information
with a high temporal resolution. Because of the uninterrupted lidar
measurements at Granada from 12:00 UTC on 9 July 2012 to 00:00 UTC on
12 July and the frequent AERONET retrievals due to good weather conditions, a
more detailed analysis was performed at this station. A total of 60 different
LIRIC retrievals were performed based on 60 lidar data sets and 21 AERONET
inversion products. The retrieval of microphysical properties was performed
using 30 min averaged lidar data (in order to reduce noise on the lidar
profiles) and the closest in time AERONET retrieval, considering only those
data with time differences lower than 3 h.
Time series of the volume concentration profiles (in
µm3 cm-3) for the fine mode (upper part), coarse
spherical mode (middle part), and coarse spheroid mode (lower part) for days
9, 10, and 11 July 2012 (from left to right).
![]()
In addition, the Granada station was affected by a mineral dust event during
the whole period as already shown in previous sections. This fact is of
special interest since the retrieval of the mineral dust microphysical is not
so straightforward, and they are not so well characterized. Up to our
knowledge not many comprehensive studies on dust microphysical properties
vertical profiles have been performed (Tsekeri et al., 2013; Wagner et al.,
2013; Granados-Muñoz et al., 2014; Noh, 2014) because of the difficulty
of the retrievals due to different factors, e.g., the high temporal variation
and non-uniform distribution of dust aerosol concentration around the globe
(Sokolik and Toon, 1999; Formenti et al., 2011), mineral dust's highly
irregular shape, and the chemical and physical transformations dust suffers
during its transport (Sokolik and Toon, 1999; Chen and Penner, 2005; Formenti
et al., 2011).
The dust outbreak analyzed here started over the Granada station on
7 July 2012 as indicated by sun-photometer data and the model forecast from
previous days (not shown). Thus, it was already well developed when the
intensive measurement period started. The 5-day backward trajectories
analysis performed with the HYSPLIT model indicated that the air masses
arriving at Granada on 9 and 11 July came from Africa, passing by the
northern African coast above 2500 m a.s.l. and from the North Atlantic
Ocean through the southwestern Iberian Peninsula below this altitude
(Fig. 6). On 10 July the air masses came from the central part of the Sahara
through the northern African coast for heights above 5000 m a.s.l., from
the Atlantic Ocean going along the coast of Africa between 2500 and
5000 m a.s.l., and from the North Atlantic Ocean, overpassing the
southwestern Iberian Peninsula below 2500 m a.s.l.
Figure 7 shows the time series of the volume concentration profiles retrieved
with LIRIC. It is clearly observed that the dust event was decreasing its
intensity along the whole study period, with the largest aerosol
concentrations for the coarse spheroid mode retrieved on 9 July
(∼ 35 µm3 cm-3) and the lowest concentrations on
11 July (∼ 15 µm3 cm-3), in agreement with AERONET
data. Maximum values of total volume concentration were around
60 µm3 cm-3 on 9 July. There was a strong predominance
of the coarse spheroid mode during the whole period, with maximum values on
9 July in the afternoon, reaching values of up to
55 µm3 cm-3. Some fine particles were also observed,
with larger volume concentrations during the first day
(∼ 10 µm3 cm-3). For this first day of
measurements, fine particles reached altitudes of around 6000 m a.s.l.,
whereas on 10 and 11 July larger volume concentration values were confined to
the lowermost region from surface up to 3 km a.s.l. The presence of this
fine mode in the upper layers might be related to the advection of
anthropogenic pollutants coming from Moroccan industrial activity in the
north of Africa mixed with the mineral dust as reported in previous studies
(Basart et al., 2009: Rodríguez et al., 2011; Valenzuela et al., 2012,
2014; Lyamani et al., 2015). Figure 6b reveals that air masses overpassed
northern African industrial areas before reaching Granada. However, it is
also well known that mineral dust emissions produce a submicronic size mode
(e.g., Gomes et al., 1990; Alfaro and Gomes, 2001). Depolarization lidar
observations over the Mediterranean have illustrated that irregularly shaped
fine dust particles significantly contribute to aerosol extinction over the
boundary layer during dust transport events (Mamouri and Ansmann, 2014). A
more detailed analysis with additional data (e.g., chemical components
measurements, single scattering albedo profiles) would be needed in order to
come to a quantitative attribution of soil dust and anthropogenic particles
to the fine mode.
Time series of the δ532nmp profiles
retrieved from the Granada lidar system at different time intervals during
the ChArMEx July 2012 intensive measurement period. The dark blue color
represents regions and time periods where no data were retrieved.
![]()
τ550nm from MODIS/Terra (top) and τ675nm daytime
mean from MSG-SEVIRI (bottom) on 9, 10, and 11 July.
![]()
The contribution of the fine mode in the lowermost part may be due mainly to
anthropogenic sources of local origin. From 11 July around 12:00 UTC up to
the end of the study period, an increase in the coarse spherical mode
concentration was observed. This increase in the coarse spherical mode was
associated with a decrease in the particle linear depolarization profiles
δ532nmp obtained from the lidar data according to
Bravo-Aranda et al. (2013) as shown in Fig. 8. On 9 July the values of
δ532nmp were around 0.30 in the layer between 3
and 5 km a.s.l. These values are representative of pure Saharan dust
(Freudenthaler et al., 2009). However, they decreased down to 0.25 during the
following days, indicating either a possible mixing of dust particles with
anthropogenic aerosols or aging processes affecting the mineral dust. During
10 July in the late afternoon and 11 July, a decrease in the fine mode
coinciding with an increase in the coarse spherical mode was observed. The
simultaneous decrease in the fine mode and increase in the coarse spherical
particles together with the decrease in δ532nmp
point to processes such as mineral dust aging and/or aggregation processes.
However, additional analysis would be necessary to confirm this hypothesis.
τ550nm forecast by the (a) BSC-DREAM8b,
(b) DREAM8-NMME, (c) NMMB/BSC-Dust, and
(d) COSMO-MUSCAT models for 9, 10, and 11 July 2012 at 12:00 UTC
over Europe and northern Africa. The yellow stars represent the location of
the stations where microphysical properties profiles are retrieved with
LIRIC.
![]()
According to δ532nmp profiles, a mineral dust
layer was clearly located above 2500 m a.s.l. or even at higher altitudes
depending on the analyzed period (see Fig. 10). Below this altitude, values
were lower indicating a mixing of the mineral dust with anthropogenic
particles from local origin. In the case of LIRIC, these vertical structures
were not so clearly defined, and a more homogeneous structure was detected.
Values of the fine and coarse mode volume concentration presented very low
variations with height when compared to δ532nmp
profiles. This vertical homogeneity is related to the assumption of height
independence of properties such as the refractive index, size distribution of
the modes, or the sphericity, which according to the results presented in
previous studies (Wagner et al., 2013; Granados-Muñoz et al., 2014), is
an issue that needs to be carefully considered in the analysis of the results
retrieved with the LIRIC algorithm.
Despite the limitations in the use of LIRIC, the analysis presented here
shows that LIRIC can reliably provide microphysical property profiles with
high vertical and temporal resolution even in cases of mineral dust. The
LIRIC algorithm can be a useful tool to detect changes in the aerosol
composition possibly associated with processes affecting the mineral dust
particles such as aging or nucleation, even though additional information is
needed for more in-depth analysis.
Evaluation of the mineral dust models
In order to obtain a general overview of the dust horizontal extension,
Fig. 9 shows the standard aerosol optical depth product retrieved using the
dark-target approach from MODIS/Terra (Remer et al., 2005, and references
therein) and the AERUS-GEO from MSG/SEVIRI (Carrer et al., 2014) for the
three analyzed days (9–11 July 2012).
Satellite data showed the presence of an aerosol plume extending from the
northern African coast towards the east with a higher aerosol load, as τλ values from MODIS sensor indicate, mainly affecting the
southeast of the Iberian Peninsula and the south of Italy (Fig. 9). As
indicated by the data presented in the previous section, this plume
corresponds to the mineral dust event, whereas a different plume is observed
above the Balkans area. The pathways of the aerosol plumes suggested by
satellite data are in agreement with both the meteorological analyses of
ECMWF and HYSPLIT air mass trajectories based on GDAS analyzed meteorological
fields at 2 km a.g.l. presented in the study by Wang et al. (2014). The air
masses were moving from Spain and Portugal to the east, whereas in the
Balkans region they were moving southwards.
Dust mass concentration profiles obtained with LIRIC (dotted line)
and BSC-DREAM8b-v2, DREAM8-NMME, DREAMABOL, and NMMB/BSC-Dust for Granada
station every 3 h on 9, 10, and 11 July 2012.
![]()
τ550nm data simulated by BSC-DREAM8b, DREAM8-NMME,
NMMB/BSC-Dust, and COSMO-MUSCAT are shown in Fig. 10. In general, when
comparing to the satellite data in Fig. 9, the aerosol plume located above
the Balkans region is not captured by the models. This is not surprising,
since it is not composed of mineral dust particles, as indicated by our
aerosol volume concentration profiles, shown in the previous section, and
suggested in previous studies (e.g., Sicard et al., 2015). The different
models correctly forecast the dust plume leaving the north of Africa and
moving towards the east and the dust plume reaching Athens, as also indicated
by satellite data. However, the decrease in τ550nm values
with time observed with satellite data and in LIRIC profiles is not well
captured by any of the different models. Regarding the extension of the dust
event, in general it is better captured by BSC-DREAM8b and NMMB/BSC-Dust,
whereas COSMO-MUSCAT and DREAM8-NMME tend to overestimate the mineral dust
horizontal extension when compared to the satellite data.
Focusing on the five stations analyzed in this study, the models showed that
the Granada station was affected by the mineral dust outbreak during the
whole analyzed period, in agreement with the analyzed data. No presence of
mineral dust was forecast above Évora, as expected from the measurements,
except for COSMO-MUSCAT, which predicted fair low values of dust τ550nm above the station. BSC-DREAM8b, DREAM8-NMME, and
NMMB/BSC-Dust indicated no presence of dust above Barcelona, even though it
was located close to the edge according to BSC-DREAM8b. As in the case of
Évora, almost negligible values were forecast above the station by
COSMO-MUSCAT. This would be in agreement with the previous data except for
the possible dust layer observed on 11 July.
In the eastern region, the station of Athens was affected by mineral dust
during the 3 days according to the DREAM8-NMME model and COSMO-MUSCAT, only
on 10 July according to NMMB/BSC-Dust, and on 10 and 11 July according to
BSC-DREAM8b. As indicated by the analysis in the previous section, mineral
dust was observed only on 11 July and the models seem to not completely
capture the event at Athens. However, in this case the situation is quite
more complex than at the western stations. Athens is located at the edge of
the mineral dust plume during the 3 analyzed days. Slight changes in the
horizontal distribution of the dust related to the model uncertainty and the
relatively coarse horizontal resolution may highly influence the results. In
the case of Bucharest, BSC-DREAM8b, DREAM8-NMME, and NNMB/BSC-DUST foresaw no
influence of the mineral dust. Conversely, COSMO-MUSCAT forecast mineral dust
during the 3 days, with larger loads on 10 and 11 July, overestimating the
extension of the mineral dust plumes as previously stated.
Due to the relatively coarse horizontal resolution of the model data
presented in Fig. 10 compared to the single-site measurements at the five
analyzed stations, it is worth evaluating in more detail the mineral dust
mass concentration profiles provided by the models at the specific locations
of our interest. To perform this evaluation, mineral dust mass concentration
profiles provided by the BSC-DREAM8b, NMMB/BSC-Dust, DREAM8-NMME, and
COSMO-MUSCAT models are evaluated against LIRIC results. The main focus is at
the Granada station since this site presents a larger number of mineral dust
profiles due to the characteristics of the mineral dust event and allows
evaluation of the temporal evolution of the dust microphysical properties.
Figure 11 shows the dust mass concentration profiles provided by the four
models and LIRIC every 3 h from 9 July at 15:00 to 11 July at 18:00. From
the profiles presented in Fig. 11, Cm, the integrated mass
concentration for each profile and the correlation coefficient, R, between
LIRIC and the different models are calculated and presented in Fig. 12.
Figure 13 shows the profiles of statistical parameters such as R obtained
for LIRIC and the model time series, RMSE, NMB, and NMSD, calculated as
described in Sect. 3 for every altitude level. These three figures need to be
analyzed and discussed as a whole in order to cover all aspects of the model
performance regarding the temporal and vertical coordinates. An independent
interpretation of each of the presented statistical parameters might be
misleading at some points and lead to erroneous conclusions.
According to Figs. 11, 12, and 13, BSC-DREAM8b shows a good temporal
correlation with LIRIC, providing larger values on 9 July than on 10 and
11 July, as observed in the experimental data. The correlation coefficient
R between BSC-DREAM8b and LIRIC time series is larger than 0.5 for most of
the altitudes (Fig. 13a). However, the model strongly underestimates the
aerosol load during the 3 studied days, as indicated by the NMB in Fig. 13c.
Positive and larger than 0.5 values of R and the small difference between
LIRIC and BSC-DREAM8b values of Cm during most of the analyzed
period in Fig. 12 indicate that BSC-DREAM8b provides a good estimation of the
mineral dust vertical distribution.
A relatively good performance of DREAM8-NMME is observed up to 10 July at
06:00 UTC, when τ440nm was larger than 0.2. During this
period the model captured quite well the maximum values and the aerosol load
as observed in Fig. 11 and indicated by the integrated mass concentration
values in Fig. 12, close to those obtained with LIRIC. Despite this good
performance during the first part of the analyzed period, NMB values in
Fig. 13c suggest an overall underestimation of the aerosol load below
5000 m a.s.l., where it is higher, and overestimation above
5000 m a.s.l., where concentration values are lower according to LIRIC.
From 3500 m a.s.l., good temporal correlation is observed between LIRIC and
DREAM8-NMME, but R goes close to 0 below this altitude (Fig. 13a).
Regarding the vertical distribution of the load, Cm values in
Fig. 12 present very small differences with LIRIC before 10 July at 06:00,
but this difference increased afterwards. Absolute values of R in Fig. 12
are usually larger than 0.5 and larger than those retrieved for the other
models, indicating good correlation. However, they oscillate from negative to
positive values, indicating a vertical shift in the location of the dust
layers during some of the analyzed periods.
(From top to bottom) Time series of the integrated mass
concentration values (above 2 km in altitude) retrieved from LIRIC and the
four evaluated model vertical profiles for the period between 15:00 UTC on
9 July 2012 and 18:00 UTC on 11 July 2012. Time series of the correlation
coefficient R, between LIRIC-derived mass concentration profiles, and each
one of the four evaluated models for the same period. Time series of the dust
center of mass, Cm, obtained from LIRIC and the model profiles.
![]()
NMMB/BSC-Dust shows a better performance on 9 July, with
τ440nm values around 0.3, especially in the layer between
2500 and 6000 m a.s.l. The difference between LIRIC and the
model-integrated mass concentration is also lower during 9 July. However, in
general the model tends to underestimate the aerosol load below
4.5 km a.s.l. (Fig. 13c). Overestimation of the mass concentration is
observed above this altitude though. NMMB/BSC-Dust correctly follows the
aerosol load decrease with time as indicated by positive correlation values
in Fig. 13a, but it presents a lower temporal correlation compared to the
other models (except for COSMO-MUSCAT). Values of Cm in Fig. 12 are
close to those of LIRIC, indicating that it correctly forecast the location
of the aerosol load. Nonetheless, low values of R indicate that the
vertical distribution of the aerosol layers needs to be improved. For this
model it is worth pointing out the unrealistic increasing maximum at
5000 m a.s.l. at 15:00 and 18:00 on 10 July (Fig. 11). However, this
maximum is very similar to the one provided by LIRIC between 06:00 and
12:00 UTC. Therefore, it could be due to a time shift of the model when
compared to the LIRIC values. To check this hypothesis, the correlation
between LIRIC and the models considering a 3 h delay is calculated
(Supplement Fig. S5). Correlation between LIRIC and NMMB/BSC-Dust for
simultaneous data is on average below 0.5 (Fig. 13a), indicating that the
model does not reproduce very well the temporal evolution of the dust
profiles. This correlation slightly increases between 3500 and
4500 m a.s.l. when considering a 3 h delay between LIRIC and the model,
but decreases at the other altitudes. Therefore, it does not appear to be a
systematic delay between the model and LIRIC profiles. However, in the future
it will be beneficial for the modeling community to gather a more extended
database of continuous lidar measurements with similar characteristics to the
one presented here in order to further explore and improve the possible
existence of delays between the model forecast and experimental data.
Vertical profiles of the correlation coefficient between LIRIC and
the model time series for every altitude level, the root mean square error
RMSE, the normalized mean bias NMB, and the normalized mean standard
deviation NMSD.
![]()
COSMO-MUSCAT shows an increase in the mineral dust load during the analyzed
period, with an increasing maximum approximately located between 4 and 5 km.
This behavior is totally opposite to the one observed in LIRIC profiles that
shows a decrease in the volume concentration with time, as indicated by the
negative values of R in Fig. 13a. According to the integrated mass
concentration values in Fig. 12, COSMO-MUSCAT underestimates the dust load
during the first half of the analyzed period, whereas an overestimation of
the dust load occurs in the second half. These two opposite behaviors seem to
cancel and, as a result, NMB values in Fig. 13c are closer to zero below
4 km than for the other models, leading to erroneous conclusions. The
locations of Cm and R values in Fig. 12 indicate a good
performance of the model regarding vertical distribution on 9 and 11 July and
the afternoon of 10 July. Again, negative R values indicate a vertical
shift in the location of the maximum concentration values during some
periods, as also observed in Fig. 11.
The four models have been shown to have advantages and disadvantages, but a
clear superior performance of any of the four has not been observed. As a
general result, the four models tend to underestimate LIRIC values during the
whole period, except for COSMO-MUSCAT, which clearly overestimates the dust
mass concentration from the afternoon of 10 July onwards. DREAM8-NMME and
NMMB/BSC-Dust show a better performance, both regarding the dust load and the
temporal evolution of the event when the aerosol load observed with the
ground-based instrumentation is higher. The temporal evolution of the event
is mostly followed by the BSC models (namely the BSC-DREAM8b, DREAM8-NMME,
and NMMB/BSC-Dust models) as indicated by the positive correlation with LIRIC
time series, whereas COSMO-MUSCAT shows and opposite behavior (Fig. 13a).
BSC-DREAM8b shows the minimum values of the RMSE below 4 km, where most of
the aerosol load is located, and maximum values are obtained for DREAM8-NMME.
However, no statistically significant difference between the models is
clearly observed. BSC-DREAM8b, DREAM8-NMME, and COSMO-MUSCAT are not able to
capture the high temporal variability observed with LIRIC, as indicated by
the large absolute values of NMSD in Fig. 13d. They range between -0.5 and
-1 below 6 km a.s.l. for COSMO-MUSCAT and BSC-DREAM8b and between -1 at
the lower altitudes and 2 at the upper levels for DREAM8-NMME. NMMB/BSC-Dust
shows a good performance in this case, with values close to 0 from 3 km
upwards.
The location of Cm, which is an indicator of the vertical
distribution of the dust mass concentration, is similar in the case of LIRIC
and the models (Fig. 12). Despite the models being capable of reproducing the
temporal evolution of Cm, in general they tended to locate the dust
load at higher altitudes, as indicated by the larger values of Cm
obtained. Discrepancies are especially relevant in the case of DREAM8-NMME
after 10 July in the afternoon. During this event, the BSC-DREAM8b model
presented the lowest differences with LIRIC regarding Cm height.
COSMO-MUSCAT and NMMB/BSC-Dust presented the lower discrepancies on 11 July.
These results are comparable to those in the study by Binietoglou et
al. (2015).
Even though they forecast the Cm fairly well, the analyzed models
provided much smoother profiles than the ones retrieved with LIRIC, with
usually a single-broad maximum located at different altitudes depending on
the model. This result is not surprising due to the coarser vertical
resolution of the models compared to lidar profiles, which can provide more
detailed information about the vertical structures of mineral dust. The
vertical correlation between the models, shown in Fig. 12b, oscillates
between positive and negative values, indicating a shift in the location of
the maximum peaks in those cases when it is negative. R values range
between 0.01 and 0.85 in absolute value. The correlation obtained in the
present analysis is lower than the ones presented in Binietoglou et
al. (2015), where most of the data presented determination coefficient
(R2) values above 0.5. This is related to the fact that in the study by
Binietoglou et al. (2015) selected mineral dust events with higher aerosol
loads (τ440 > 0.15) were presented, whereas in this study the
continuous evolution of the dust event was analyzed with τ440 ranging
between 0.07 and 0.40. Therefore, according to the present study models seem
to show a better performance in cases of higher aerosol load.
Model profiles were also obtained at the stations of Athens, Barcelona,
Bucharest, and Évora in order to evaluate their performance at stations
where there is a slight or no presence of mineral dust. At Athens (Fig. S1 in
the Supplement) almost negligible mass concentration values were forecast by
the different models, with the exception of DREAM8-NMME. This model indicated
the presence of mineral dust in mass concentrations up to
100 µg m-3, reaching 4000 m a.s.l. on 10 July and up to
65 µg m-3 on 11 July, which is not in agreement with LIRIC
results. In spite of the disagreement, it is worth pointing out that the dust
layer observed at Athens between 3000 and 5000 m a.s.l. on 11 July
according to LIRIC data was correctly forecast by the different models. At
the Barcelona station (Fig. S2), DREAM8-NMME was not in agreement with the
experimental results since it forecast dust mass concentrations of up to
100 µg m-3 and located below 2000 m a.s.l. At Bucharest
(Fig. S3), large dust concentrations were forecast between 3000 and
7000 m a.s.l. by BSC-DREAM8b, DREAM8-NMME, and NMMB/BSC-Dust on 9 July. On
10 and 11 July the dust load forecast by the models was much lower, even
though it reached up to 50 µg m-3. This is not in agreement
with our experimental results since only coarse spherical and fine particles
and no mineral dust should be forecast here. Finally, at the Évora
station (Fig. S4), DREAM8-NMME forecast dust mass concentration lower than
10 µg m-3 below 2000 m a.s.l. COSMO-MUSCAT forecast similar
concentrations above 2000 m a.s.l. These mass concentration values are
almost negligible and therefore good agreement can be considered. In general,
good results were provided by the different models at the five stations.
However, DREAM8-NMME seems to be overestimating the dust mass concentrations
at those stations affected by aerosol types different to mineral dust.
An in-depth analysis of the causes of the discrepancies between the models
and LIRIC is beyond the scope of this study, especially taking into account
that they showed a similar performance here, with none of them proving to be
more accurate than the others. In general we observed that the BSC models
showed a similar behavior between them. Differences were clearly observed
when they were compared to COSMO-MUSCAT, based on a different philosophy.
However, none of them showed a statistically significant better performance.
Differences between the obtained results lie in the different approaches used
in the different models, the different meteorological fields used, dust
sources, horizontal and vertical transport schemes, different resolutions,
etc., as already pointed out in Binietoglou et al. (2015). Robust conclusions
in this respect cannot be drawn from this study and would require wider
databases with higher temporal and spatial coverage in order to cover the
different aspects of the model calculations, and more dedicated studies.
Nonetheless, the comparison presented here provided valuable results since it
addresses the points of discrepancy and proves LIRIC's potential as a tool
for future model evaluations. Information inferred from the results obtained
here could be used for the planning of future validation strategies and
campaign management.