The Generalized Aerosol Retrieval from Radiometer and Lidar Combined data algorithm (GARRLiC) and the LIdar-Radiometer Inversion Code (LIRIC) provide the opportunity to study the aerosol vertical distribution by combining ground-based lidar and sun-photometric measurements. Here, we utilize the capabilities of both algorithms for the characterization of Saharan dust and marine particles, along with their mixtures, in the south-eastern Mediterranean during the CHARacterization of Aerosol mixtures of Dust and Marine origin Experiment (CHARADMExp). Three case studies are presented, focusing on dust-dominated, marine-dominated and dust–marine mixing conditions. GARRLiC and LIRIC achieve a satisfactory characterization for the dust-dominated case in terms of particle microphysical properties and concentration profiles. The marine-dominated and the mixture cases are more challenging for both algorithms, although GARRLiC manages to provide more detailed microphysical retrievals compared to AERONET, while LIRIC effectively discriminates dust and marine particles in its concentration profile retrievals. The results are also compared with modelled dust and marine concentration profiles and surface in situ measurements.
The importance of studying the vertical distribution of aerosol plumes is prominent in regional and climate studies, since it can effectively change the radiative properties of the atmosphere and the presence of clouds (e.g. Pérez et al., 2006; Solomon et al., 2007). Ground-based monitoring of the aerosol vertical structure is effectively performed with the synergy of passive and active remote sensing instruments, in particular with multi-wavelength sun-photometers and lidars. The sun-photometer provides the columnar properties of the particles (e.g. Dubovik and King, 2000; Dubovik et al., 2006), whereas the lidar is capable of providing vertical profiles of the backscatter and extinction coefficients, along with vertical profiles of the particle microphysical properties, mainly for the fine mode (e.g. Müller et al., 2016). The combination of active and passive remote sensing has been tried so far mostly by using the sun-photometer-measured aerosol optical depth (AOD) as ancillary information for the lidar retrieval (e.g. Fernald et al., 1972; Ansmann et al., 2011, 2012). The GARRLiC (Generalized Aerosol Retrieval from Radiometer and Lidar Combined data algorithm; Lopatin et al., 2013) and LIRIC (LIdar-Radiometer Inversion Code; Chaikovsky et al., 2016) algorithms go a step further and use deeper synergies: the LIRIC approach derives the particle concentration profiles from the lidar measurements, using the columnar microphysical properties derived separately from the sun-photometer; GARRLiC advances the method even more, combining for the first time both sun-photometer and lidar measurements for the retrieval of the particle microphysical properties. As discussed in detail in Lopatin et al. (2013), combining the sun-photometer intensity measurements with the backscatter lidar information seems to result in better sensitivity to the particle shape and the ability to retrieve the refractive indices of fine and coarse particles separately as well as to extract the vertical distribution of the fine- and coarse-particle concentrations. Moreover, it can potentially provide higher accuracy for cases of low aerosol loadings, compared with the intensity-only retrieval.
GARRLiC and LIRIC have been developed in the framework of the Aerosols,
Clouds and Trace gases Research Infrastructure (ACTRIS,
GARRLiC and LIRIC input and output data are shown in Fig. 1, while short
descriptions are given herein: the LIRIC algorithm uses the particle
microphysical properties provided in the AERONET product as a priori
information in the inversion of the lidar measurements for retrieving the
aerosol volume concentration profiles. Using lidar measurements of elastic
backscatter at three wavelengths of 355, 532 and 1064 nm, LIRIC retrieves
the volume concentration profiles of fine and coarse particles. Moreover, the
cross-polarized lidar signal at 532 nm allows the decoupling of the coarse
mode into its spherical and non-spherical components. The error estimation of
the retrieved profiles is provided as well. Both LIRIC and GARRLiC suppress
unrealistic oscillations in the retrieved quantities (e.g. aerosol concentration) but otherwise do not constrain their absolute values. In this way the algorithms exclude
solutions that are mathematically possible but contain unrealistic
oscillations in the retrieved properties (see also Dubovik, 2004; Dubovik and
King, 2000). The GARRLiC algorithm synergistically combines the sun-photometer
sun and sky measurements at four wavelengths (at 440, 670, 870 and 1020 nm)
and up to 35 scattering angles, with the vertically resolved lidar
measurements of the elastic backscatter at three wavelengths (at 355, 532 and 1064 nm). The algorithm does not use the AERONET products, but it
instead calculates the size distribution, spherical particle fraction and
spectral complex refractive index, separately for fine and coarse particles.
In the case of a dominant mode (e.g. for pure dust cases), the algorithm is set
to retrieve the aerosol characteristics for one mode only. Although in
GARRLiC the microphysical properties are considered to be constant along the
column for each mode, the total values change along the column in the case of two
modes with different properties. The algorithm also calculates the volume
concentration profiles of fine and coarse particles. The concentrations are
considered constant below the lowest height of the lidar signals, which may
introduce errors in the retrieved profiles (e.g. Tsekeri et al., 2013). The
retrieval uncertainties of the microphysical parameters are provided as well,
following the approach described by Dubovik et al. (2000) and the profile
retrieval uncertainties are currently under development. GARRLiC and its
updates are available for download at
GARRLiC and LIRIC algorithm input and output parameters. For LIRIC, the output when using the cross-polarized signal at 532 nm is shown in the dashed box.
In the case of multimode aerosol mixtures and/or changes in microphysical properties with height due to particle hygroscopic growth, an inherent deficiency of both algorithms is the number of aerosol modes retrieved, with LIRIC retrieving three modes (fine particles, coarse spherical and coarse non-spherical particles) and GARRLiC two modes (fine and coarse particles). We need to highlight here that LIRIC retrieves three modes only for the volume concentration profiles, whereas otherwise it uses the AERONET products by providing, for example, a common spectral refractive index for all modes (Fig. 1). Both algorithms work well for individual aerosol components or mixtures of (mainly) fine (e.g. pollution) and (mainly) coarse (e.g. dust) particles, but they should not be able to fully characterize the mixture components in the case of more than one fine or coarse mode in the mixture, such as in smoke–pollution or dust–marine mixture cases. For the latter, LIRIC should provide an effective characterization for the volume concentration profiles, since it derives the coarse spherical (hydrated marine) particles and the non-spherical (dust) particles, but the characterization is not expected to be satisfactory for the particle microphysical properties.
In our study, we apply GARRLiC and LIRIC to cases of dust, marine and a dust–marine mixture during the CHARADMExp (CHARacterization of Aerosol mixtures of Dust and Marine origin Experiment) campaign in the south-eastern Mediterranean. This is the first time a detailed characterization of marine and marine mixtures with dust along the atmospheric column is performed for the area. So far, various studies have tried to characterize the aerosol radiative properties in the Mediterranean with satellite or ground-based AOD measurements (e.g. di Sarra et al., 2008; Kazadzis et al., 2009; Papadimas et al., 2012). Unfortunately, they fail to overcome their limitations, such as the non-realistic assumptions about the aerosol absorption properties and the lack of information on the real vertical aerosol structure (Mishra et al., 2014). The kind of characterization presented here is important for application in future satellite missions, not only for the Mediterranean but for large parts of the globe where dust and marine particles are present, such as in the Atlantic Ocean (e.g. Prospero, 1996).
The CHARADMExp campaign and the three cases (i.e. mainly dust, marine–pollution mixture and dust–marine–pollution mixture) are presented in Sect. 2. The methodology followed in our work is presented in Sect. 3, the GARRLiC and LIRIC results are shown in Sect. 4 and finally our conclusions are given in Sect. 5.
CHARADMExp was an experimental campaign of ESA, implemented by the National
Observatory of Athens (NOA), aimed to characterize dust and
marine particles along with their mixtures (
The Polly
The information close to the surface is very important for our study,
especially for the marine-particle characterization, since the marine
particles reside mostly below 1 km (e.g. Ho et al., 2015). Unfortunately,
this is also the lidar “overlap region”, with large uncertainty for the
lidar backscattered signal due to its partial collection from the telescope
(e.g. Wandinger and Ansmann, 2002). The Polly
The CIMEL CE318 sun-photometer is the instrument used in the AERONET
sun-photometer network, with more than 250 units worldwide. The technical
specifications of the instrument are given in detail by Holben et al. (1998).
Taking into account all the information about the instrument and calibration
precision (Holben et al., 1998) the accuracy of the AOD measurements is
estimated to be of the order of
The GARRLiC-retrieved size distribution is evaluated against the surface
measurements of the Scanning Mobility Particle Spectrometer (SMPS). SMPS
provides the fine-particle number size distribution at
Moreover, we evaluate the particle concentration derived from GARRLiC and
LIRIC at the surface level with the surface in situ measurements of the
particular matter for particles with diameters less than 10
In order to compare the in situ measured size distribution and mass
concentration with the GARRLiC and LIRIC ambient retrievals, we need to take into
account the particle drying applied to surface measurements. The in situ
instruments dry the sampled air by adiabatic compression during the sampling
through their inlets and by the radiant heat from the lights inside the
instruments. The size and mass of the ambient particles thus change,
especially in the case of hygroscopic particles in humid conditions (e.g. Snider
and Petters, 2008). For the size distribution we evaluate this effect
qualitatively (see Sect. 4.2 and 4.3). For the PM
We derive
For the cases analysed herein, we consider a
The origin of the examined aerosol layers at Finokalia Station is
investigated with the use of source–receptor computations derived with
dispersion modelling tools. The corresponding emission sensitivity (i.e. the
residence time of the tracer particles inside the lowest tropospheric layers)
is calculated from backward Lagrangian simulations with the atmospheric
dispersion model FLEXPART-WRF (Brioude et al., 2013). The dispersion model is
coupled offline with the WRF_ARW atmospheric model (Skamarock et al.,
2008). The spatial resolution of WRF is 12
Desert dust emissions and transport are described with the BSC-DREAM8b model
(Nickovic et al., 2001; Pérez et al., 2006; Basart et al., 2012a). The
BSC-DREAM8b model is embedded into the Eta/NCEP atmospheric model and solves
the mass balance equation for dust, taking into account the different
processes of the dust cycle (i.e. dust emission, transport and deposition).
The updated version of the model includes a source function based on the
1 km USGS land use data, eight particle size bins (0.1–10
Sea salt emissions and transport are described with the atmospheric model
RAMS-ICLAMS (Solomos et al., 2011). The model is an enhanced version of RAMS
(Pielke et al., 1992; Cotton et al., 2003) and it includes a full description
of the sea salt life cycle in the atmosphere. The parameterization of sea salt
emission is based on the white-cap formation for the entrainment of sea salt
spray in the atmosphere (Monahan et al., 1986), also taking into account the
effects of RH on the size distribution of the particles (Zhang et al., 2005).
Sea salt flux close to the coastline is also calculated in the model
following the parameterizations of Leeuw et al. (2000) and Gong et
al. (2002). The dry and wet removal processes are treated with the
corresponding schemes described in Seinfeld and Pandis (1998). The simulated
sea salt mass is represented with a bimodal log-normal distribution. The first
(accumulated) mode has a mean diameter of 0.36
a) Range-corrected backscattered signal at 1064 nm in arbitrary
units (top) and volume depolarization ratio at 532 nm (bottom) from
Polly
In order to demonstrate the GARRLiC and LIRIC capabilities in characterizing
events with dust and marine particles, we analyse in detail three cases
acquired during CHARADMExp at Finokalia. The first case is a relatively
moderate dust episode with a low amount of marine and continental particles,
the second is a low-AOD marine and continental plume and the last is a
mixture of dust, marine and continental particles. Source–receptor
simulations are used to derive the particle origin and characterize the air
masses. Then, we compare the optical properties retrieved from GARRLiC,
LIRIC and collocated Klett retrievals (Klett, 1985). The GARRLiC and
LIRIC/AERONET fine-mode size distributions and PM
Backscatter and extinction coefficient retrievals at Finokalia,
Crete on 26 June 2014, at 04:00–06:00 UTC.
GARRLiC retrievals (pink) of size distribution
On 26 June the Polly
Considering that the atmospheric column is dominated by dust (as shown in the
coarse-mode-dominated AERONET size distribution), we performed the one-mode
GARRLiC inversion. For both GARRLiC and LIRIC we used the lidar measurements
at 04:00–06:00 UTC (red box in Fig. 3a) and the sun-photometer measurements
at 04:54 UTC. Our results show that GARRLiC and LIRIC backscatter and
extinction coefficient profiles at 355, 532 and 1064 nm agree quite well,
with their differences being 10–20 % with respect to GARRLiC values,
well within the LIRIC uncertainties (Fig. 4a and b). Larger differences are
seen below
A special feature seen in GARRLiC, LIRIC and Klett backscatter profiles is
the larger backscatter at 532 than 355 nm. This is not usual for dust
particles, but it has been reported before: Veselovskii et al. (2016) showed a similar spectral dependence for dust during the SaHAran Dust
Over western Africa (SHADOW) campaign, which they attributed to large dust-particle spectral variation of the imaginary part of the refractive index.
More specifically, they managed to reproduce this backscatter spectral
dependence with imaginary part values of
Figure 5 shows good agreement between GARRLiC and AERONET retrievals (the
latter is used in the LIRIC retrieval) within the GARRLiC retrieval
uncertainties. Differences are seen only for the real part of the refractive
index, which for GARRLiC is at
As in Fig. 4 but for backscatter and extinction coefficient retrievals at Finokalia, Crete on 15 July 2014, at 12:30–14:30 UTC.
GARRLiC retrievals for fine (blue) and coarse particles (pink) of
size distribution
The concentration profiles from GARRLiC and LIRIC are in excellent agreement
at heights
To summarize, the GARRLiC and LIRIC retrievals perform well for the
dust episode on 26 July, considering the consistency with the Klett
retrievals, the BSC-DREAM8b modelled mass concentration profiles and the surface
in situ measurements of the fine-mode size distribution, as well as the
expected increase of the dust absorption in the ultraviolet. The
discrepancies seen for the retrieval closer to the surface and the PM
On 15 July the lidar measurements at 12:30–14:30 UTC showed a low-AOD layer
of non-depolarizing particles extending up to 3 km (Fig. 7a). The lack of
depolarization indicates spherical (hydrated) marine particles, which is also
supported by our source–receptor analysis (Fig. 7b). Specifically,
FLEXPART-WRF simulations show that the particles above Finokalia Station mainly have
a marine origin along the whole atmospheric column, with a possible
contribution of continental aerosol from southern Italy. This scenario is
further supported by AERONET measurements at 13:24 UTC of low AOD of
The low AOD is unfavourable for the GARRLiC and AERONET microphysical property retrievals, especially for the spectral refractive index and SSA (Dubovik et al., 2000; Lopatin et al., 2013). The latter requires an AOD of at least 0.4 at 440 nm for satisfactory accuracy in the case of sun-photometer-only retrieval (Dubovik et al., 2000). The lidar information combined with the sun-photometer measurements in GARRLiC is expected to improve the retrieval for low-AOD cases (Lopatin et al., 2013). Although the AOD requirements have not been quantified yet for GARRLiC, an AOD of 0.3 at 440 nm is considered sufficient. As reported in Dubovik et al. (2002), the marine particles rarely exceed the AOD of 0.15 at 440 nm; thus we do not expect a highly accurate refractive index and SSA retrievals from GARRLiC or from AERONET/LIRIC for the marine particles. Furthermore, the marine case analysed here has a much lower AOD; thus we consider the refractive index and SSA retrievals to be only indicative in this case. In addition, as seen in Fig. 7a, most of the aerosol load is located below 1 km, where the lidar incomplete-overlap region is located, which challenges the combined lidar–sun-photometer retrieval even more.
The GARRLiC and LIRIC retrievals used the lidar measurements at 12:30–14:30 UTC (red box in Fig. 7a) and the sun-photometer measurements at 13:24 UTC. Figure 8 shows the retrieved backscatter and extinction coefficients at 355, 532 and 1064 nm, and the corresponding retrievals from the Klett method. For the latter we use LRs of 50, 45 and 45 sr for 355, 532 and 1064, which closely reproduce the sun-photometer-measured AODs of 0.1, 0.05 and 0.02 at 340, 500 and 1020 nm. The agreement between GARRLiC and LIRIC is satisfactory within the LIRIC uncertainties (Fig. 8a and b). Above 550 m, this is also the case for GARRLiC and Klett backscatter coefficient retrievals, whereas for the extinction coefficients the differences are within 30 % for 355 nm and 10–40 % for 532 nm relative to GARRLiC values (Fig. 8c and d). In the marine boundary layer (below 550 m) the Klett NF backscatter and extinction coefficients at 532 nm show much larger values than the ones retrieved from GARRLiC and LIRIC. This very vividly highlights the importance of the NF measurements in properly retrieving the marine-particle properties with lidars.
GARRLiC retrieves both fine and coarse particles in this case, which we
consider to be mainly of continental and marine origin, respectively. The
fine-particle volume size distribution shows
Figure 10a shows the GARRLiC and LIRIC volume concentration profiles, which
agree well within the LIRIC retrieval uncertainties above 550 m, whereas
below the GARRLiC concentration for the coarse particles is larger. Assuming
that the marine particles are comprised only of coarse particles, we derive
the marine mass concentration profiles from GARRLiC and LIRIC as shown in
Fig. 10b. The mass concentration profiles are calculated from the coarse
volume concentration profiles using a sea salt density of 1.3 g cm
GARRLiC retrievals for fine (blue) and coarse particles (pink) of
size distribution
To summarize, GARRLiC retrieves more fine particles than AERONET and surface
in situ measurements. The fine-particle SSA and refractive index are
characteristic of continental particles. The corresponding coarse-mode
retrieval probably fails for SSA and the imaginary part of the refractive
index, which are very difficult to retrieve with low AODs, but the real
part of the refractive index properly assigns the refractive index of marine
particles. Both GARRLiC and LIRIC concentration profiles seem to agree well
with the PM
On 4 July a mixture of dust, marine and continental aerosols was observed at
Finokalia Station. Figure 11a shows an advected
depolarizing dust plume at 4–6 km and a less depolarizing plume extending
from the ground up to 2–3 km at 04:00–06:00 UTC, with volume depolarization ratios at 532 nm
of 0.1 and
Potential of GARRLiC to retrieve marine- (light blue) and dust-particle (orange) size distribution (left) and spectral real part of the refractive index (right). The retrieval refers to measurements at Finokalia, Crete on 4 July 2014, at 04:00–06:00 UTC. The black line shows the AERONET retrieval at 05:49 UTC.
GARRLiC retrieves these three layers (Fig. 12a), but it cannot characterize
them effectively in terms of their refractive indices, since it is able to
retrieve only one refractive index for each mode. For example, the coarse
mode of the dust–marine mixture contains dust particles with a real part of
refractive index of
Figure 14 shows the potential of GARRLiC to retrieve the marine and
dust components of the mixture by changing the definition of the two
modes retrieved: instead of fine and coarse mode, GARRLiC is set to
retrieve two modes that span the whole size range so that both contain coarse
particles. It derives a dust mode that contains only coarse particles
and a marine mode that contains both fine and coarse particles, larger than those of dust. Raptis et al. (2015) showed similar results for the
marine and dust size distributions using their multimodal analysis for a
different dust–marine mixture case during the CHARADMExp campaign. The
retrieved real part of the refractive index is
As in Fig. 4 but for backscatter and extinction coefficient retrievals at Finokalia, Crete on 4 July 2014, at 04:00–06:00 UTC.
LIRIC provides the dust and marine vertical distribution, since it
disentangles the coarse-particle volume concentration profile into its
spherical (marine) and non-spherical (dust) components (Fig. 12a, right).
Assuming a very low contribution from dust and marine particles in the fine
mode we acquire the marine and dust concentration profiles from the
spherical and non-spherical coarse-particle concentration profiles,
respectively. Figure 12b shows that LIRIC marine and dust mass concentration
profiles have larger values than the BSC-DREAM8b dust and the RAMS-ICLAMS sea
salt models, respectively. In order to acquire the mass concentration
profiles, LIRIC dust and marine volume profiles are multiplied with the
density values of 2.6 g cm
Figure 15 shows the backscatter and extinction coefficients retrieved with GARRLiC, LIRIC and Klett methods. GARRLiC and LIRIC agree well within the LIRIC uncertainties (Fig. 15a and b). The agreement with Klett retrievals is satisfactory for the backscatter coefficients at 532 and 1064 nm above 550 m, within their uncertainties, with differences of 60–130 % seen for the 355 nm retrieval (Fig. 15c). As for the marine case in Sect. 4.2, the NF backscatter coefficient at 532 nm shows much larger values. The same holds for the NF extinction coefficient at 532 nm (Fig. 15d). The Klett extinction coefficients at 1–3 km are up to 60 and 50 % lower than GARRLiC at 355 and 532 nm.
Overall, this is a challenging case for both the GARRLiC and LIRIC algorithms. We
can claim that GARRLiC shows some potential in providing a successful dust
and marine microphysical property characterization when more information
(e.g. cross-polarized lidar signals) is included in the retrieval. Moreover,
the LIRIC capability of providing the vertical distribution of dust and
marine particles is mostly successful when the results are compared with our
source–receptor simulations and the surface in situ PM
GARRLiC and LIRIC algorithms provide the great innovation of retrieving the vertical distribution of aerosol microphysics utilizing the synergy of the elastic backscatter lidar and sun-photometer techniques. This way, the algorithms show the potential to effectively characterize the vertical distribution of fine, coarse spherical and coarse non-spherical particle concentrations in the case of LIRIC, and the concentration profiles of fine and coarse particles, along with their column-averaged size, shape and spectral refractive index, in the case of GARRLiC.
In this study we used both algorithms to characterize three cases of dust and
marine presence during the ESA-CHARADMExp experimental campaign. For the
first case GARRLiC achieves a successful retrieval of the dust vertical
distribution and microphysical characterization, which agrees well with AERONET
and climatological values for dust within the respective uncertainties. Both
LIRIC and GARRLiC concentration profiles are found to be consistent with the
BSC DREAM8b dust vertical structure, showing up to 100 % larger
values than the surface in situ PM
The difficulties posed in retrieving the concentration profiles and the microphysical properties of dust and marine-particle mixtures in the atmospheric column have to do with the low AOD of the marine plumes, the insufficient lidar information in the overlap region and the number of modes from the retrievals. For GARRLiC, the retrieval of multiple modes would be possibly feasible in the future with the incorporation of polarimetric measurements from the sun-photometer and/or the cross-polarized and Raman signals from the lidar. Moreover, we could try to increase the near-surface information from the lidar by performing the signal gluing technique between the FF and NF measurements and/or by using additional information available from in situ observations. We aim to continue investigating the GARRLiC and LIRIC potential for aerosol characterization and follow related improvements in the framework of the ACTRIS-2 project and the experimental campaigns that are dedicated to that objective.
The data used in this work are publicly available
in the Zenodo public data repository,
The authors declare that they have no conflict of interest.
This article is part of the special issue “EARLINET, the European Aerosol Research Lidar Network”. It is not associated with a conference.
The research leading to these results has received funding from the European Union's Horizon 2020 Research and Innovation Programme ACTRIS-2 (grant agreement no. 654109). The work has been developed under the auspices of the ESA-ESTEC project “Characterization of Aerosol mixtures of Dust And Marine origin” contract no. IPL-PSO/FF/lf/14.489. The work was also supported by the European Research Council under the European Community's Horizon 2020 research and innovation framework programme/ERC grant agreement 725698 (D-TECT). The publication was supported by the European Union's Horizon 2020 Research and Innovation programme under grant agreement no. 602014, project ECARS (East European Centre for Atmospheric Remote Sensing). The authors acknowledge support through the Stavros Niarchos Foundation. BSC-DREAM8b simulations were performed on the Mare Nostrum supercomputer hosted by the Barcelona Supercomputing Center-Centro Nacional de Supercomputación (BSC). Edited by: Richard Ferrare Reviewed by: three anonymous referees