Introduction
For more than 15 years, two MODIS (Moderate Resolution Imaging
Spectroradiometer) satellite sensors have been monitoring tropospheric aerosols at a
global scale on a daily basis. The retrieved aerosol optical properties have
been used in numerous air quality studies as well as studies related to the
effect of airborne particles on various climatic parameters (e.g., radiation,
clouds, precipitation). The 1∘ × 1∘ daily gridded level-3
dataset is primarily used for global as well as regional studies, while the
single-pixel level-2 data with a 10 km resolution (at nadir) are mostly used
for regional and local scale studies. Nevertheless, the use of the coarse-resolution MODIS data has predominated, even in regional studies. The reasons
for this could be the smaller file size, which makes their processing and
storage easier, or the fact that they are easily accessible through
user-friendly databases, which also allow for a very basic analysis like
e.g., NASA's GIOVANNI website (http://giovanni.gsfc.nasa.gov/giovanni/)
(Acker and Leptoukh, 2007).
The same holds for studies focusing on the Mediterranean Basin, an area
which is considered of particular sensitivity as far as air pollution and
climate change is concerned (Lelieveld et al., 2002; Giorgi, 2006). The
Mediterranean Basin is one of the regions with the highest aerosol optical
depths (AODs) in the world (Husar et al., 1997; Ichoku et al., 2005;
Papadimas et al., 2008), causing significant climate forcing especially in
summer, which is characterized by low cloudiness and high incoming solar
radiation levels (Papadimas et al., 2012; Alexandri et al., 2015). The
Mediterranean is also recognized as a crossroads between three continents
where aerosols of various types accumulate (Lelieveld et al., 2002). Marine
aerosols from the Mediterranean Sea and even the Atlantic Ocean combine with
aerosols from continental Europe (urban and rural), dust particles
transported from the Sahara and Middle East, as well as biomass
burning aerosols from occasional wild fires and agricultural burning
(Lelieveld et al., 2002). Specifically, as discussed in Hatzianastassiou
et al. (2009), the Eastern Mediterranean, the region under investigation here, is
located at a key point of this crossroads. There is a significant number
of ground- and satellite-based studies on the abundance and optical
properties of tropospheric aerosols in the area; however, these studies are
either focused on specific spots, or they used a coarse spatial and temporal
resolution.
The ground-based instrumentation used in studies focusing on the aerosol
load and optical properties over the Eastern Mediterranean includes active
and passive sensors such as lidars (e.g., Papayannis and Balis, 1998; Balis
et al., 2004; Papayannis et al., 2005, 2009; Amiridis et al., 2005, 2009;
Mamouri et al., 2013; Kokkalis et al., 2013; Nisantzi et al., 2015), Cimel
sun photometers (e.g., Israelevich et al., 2003; Kubilay et al., 2003;
Derimian et al., 2006; Kalivitis et al., 2007; Kelektsoglou and
Rapsomanikis, 2011; Nikitidou and Kazantzidis, 2013), Brewer
spectrophotometers (e.g., Kazadzis et al., 2007; Koukouli et al., 2010),
Multi-Filter Radiometers (e.g., Gerasopoulos et al., 2009, 2011; Kazadzis et
al., 2014), ceilometers (e.g., Tsaknakis et al., 2011), Microtops
sun photometers (e.g., El-Metwally and Alfaro, 2013), etc. However, these and
other studies not referenced here either refer to specific spots with the
majority of the ground stations being situated in large urban centers (e.g., Athens, Thessaloniki, Cairo) or to specific events (e.g., Sahara dust
intrusions, biomass burning events).
On the other hand, AOD and other aerosol optical properties have been
studied over the greater Eastern Mediterranean region based on data from
Meteosat (Moulin et al., 1998), SeaWIFS (Koren et al., 2003; Antoine and
Nobileau, 2006; Mélin et al., 2007; Nabat et al., 2013), TOMS (Alpert
and Ganor, 2001; Israelevich et al., 2002; Koukouli et al., 2006, 2010;
Hatzianastassiou et al., 2009; Israelevich et al.,
2012, Kaskaoutis et al., 2012a; Nabat et al., 2013; Gkikas et al., 2013,
2014; Varga et al., 2014), MODIS Terra and Aqua (Barnaba and Gobbi, 2004;
Papayannis et al., 2005; Kaskaoutis et al., 2007, 2008, 2010, 2011,
2012a, b, c, d; Kosmopoulos et al., 2008; Papadimas et al., 2008, 2009; Rudich
et al., 2008; Carmona and Alpert, 2009; Karnieli et al., 2009; Gkikas et
al., 2009, 2013; Hatzianastassiou et al., 2009; El-Metwally et al., 2010;
Koukouli et al., 2010; Kanakidou et al., 2011; Gerasopoulos et al., 2011;
de
Meij and Lelieveld, 2011; Marey et al., 2011; de Meij et al., 2012; Nabat et
al., 2012, 2013; Nikitidou and Kazantzidis, 2013; Athanassiou et al., 2013;
Benas et al., 2011, 2013; Sorek-Hamer et al., 2013; Kabatas et al., 2014;
Kourtidis et al., 2014; Mishra et al., 2014; Flaounas et al., 2015; Kloog et
al., 2015), OMI/AURA (Kaskaoutis et al., 2010; El-Metwally et al., 2010;
Marey et al., 2011; Kaskaoutis et al., 2012b, c, Gkikas et al., 2013, 2014;
Sorek-Hamer et al., 2013; Varga et al., 2014; Flaounas et al., 2015),
CALIOP/CALIPSO (Amiridis et al., 2009, 2013, Mamouri et al., 2009; Marey et
al., 2011; Kaskaoutis et al., 2012c; de Meij et al., 2012; Nabat et al.,
2012, 2013; Mamouri and Ansmann, 2015), MISR/Terra (Kanakidou et al., 2011;
Marey et al., 2011; de Meij and Lelieveld, 2011; de Meij et al., 2012; Nabat
et al., 2013; Kabatas et al., 2014; Abdelkader et al., 2015), as well as
NOAA/AVHRR, MERIS/ENVISAT, AATSR/ENVISAT, PARASOL/POLDER, MSG/SEVIRI and
Landsat satellite data (see Retalis and Sifakis, 2010; Nabat et al., 2013;
Benas et al., 2013; Sifakis et al., 2014). To our knowledge, these studies
comprise the majority of works focusing on tropospheric aerosols over the
Eastern Mediterranean by means of satellite remote sensing, published in
peer-reviewed journals the last ∼ 15 years. As shown in Fig. 4
of this work, the publication rate of satellite-based studies focusing on
the Eastern Mediterranean aerosols nearly doubled every 3 years during
the period 1997–2014, which is indicative of the increasing scientific
interest in the area.
In a very large fraction of the satellite-based studies referenced above,
the data used are either of coarse mode (usually 1∘, which is
∼ 100 km for the midlatitudes) or focus on specific spots for
validation purposes. In a few cases, high-resolution data were used in
spatiotemporal studies; however, either these studies are restricted over
surfaces covered by water, or examine a short period only. For example,
Moulin et al. (1998) investigated the dust AOD patterns over the oceanic
areas of the Mediterranean Basin at a resolution of 35 × 35 km2 for a
period of 11 years (1984–1994) using Meteosat observations. A 7-year
climatology (1998–2004) of total and dust AOD for the same regions at a
resolution of 0.16∘ × 0.16∘ was compiled by Antoine and
Nobileau (2006) using observations from SeaWIFS. Mélin et al. (2007) merged
SeaWIFS and MODIS data and presented high-resolution AOD patterns (2 × 2 km2)
for May 2003. As far as MODIS is concerned, only Barnaba and Gobbi (2004)
presented a high-resolution (0.1∘ × 0.1∘) spatiotemporal
analysis for a period of 1 year (2001) over sea only. In a recent paper,
Athanassiou et al. (2013) presented, in detail, a method for compiling a
0.5∘ resolution AOD gridded dataset using level-2 MODIS Terra data for
the greater region of Greece (2000–2008). However, the spatial resolution
they used (∼ 50 km) is not high enough to reveal local sources
(e.g., cities, islands, river banks). Overall, there has so
far not been any detailed high-resolution spatiotemporal study of the AOD over the
Eastern Mediterranean.
In this paper, the AOD550 spatiotemporal variability over the Eastern
Mediterranean (30–45∘ N, 17.5–37.5∘ E) is presented at
a high spatial resolution (0.1∘ × 0.1∘) based on MODIS Terra and
Aqua observations. Level-2 MODIS data are used for the compilation of a
0.1∘ gridded dataset, which is validated against ground-based
observations. In order to calculate the contribution of different aerosol
types to the total AOD, the MODIS data were analyzed together with other
satellite data, ERA-Interim and MACC reanalysis data and the Goddard
Chemistry Aerosol Radiation and Transport (GOCART) model, using an algorithm
optimized for the surface properties of the Eastern Mediterranean region.
The different datasets used in this research are presented in detail in
Sect. 2, while a detailed description of the method is given in Sect. 3.
Sect. 4 includes the results from the MODIS validation procedure, the annual
and seasonal variability of AOD550 over the region with a discussion on
the local aerosol sources and the differences between Terra and Aqua and
the annual and seasonal contribution of different aerosol types to the total
AOD550. Finally, in Sect. 5, the main conclusions of the paper are
presented along with a short discussion on how these results could
contribute to future studies in the area.
Observations, reanalysis data and model simulations
MODIS Terra and Aqua satellite observations
The main data used in this work come from the level-2 MODIS Terra (MOD04_L2)
and MODIS Aqua (MYD04_L2) Collection 051 dataset and have been acquired
through NASA's Level 1 and Atmosphere Archive and Distribution System (LAADS)
(http://ladsweb.nascom.nasa.gov). MODIS Terra and Aqua have a daytime
Equator crossing time at 10:30 LT (morning) and 13:30 LT (noon),
respectively. MODIS instruments, with a viewing swath of 2330 km, measure
backscattered radiation at 36 spectral bands between 0.415 and
14.235 µm with a spatial resolution of 250, 500 and 1000 m,
providing a nearly global coverage on a daily basis. Aerosol optical
properties for the standard MODIS aerosol product are retrieved using two
different “Dark Target” (DT) algorithms. One is used over land surfaces
(Kaufman et al., 1997; Levy et al., 2007a, b, 2010; Remer et al., 2005), and
the other over oceanic regions (Tanré et al., 1997; Levy et al., 2003;
Remer et al., 2005). The “Deep Blue” algorithm (DB) (Hsu et al., 2004,
2006) has been used for retrievals over bright land surfaces (e.g., deserts)
where the DT algorithm fails. Only recently did updates to the algorithm allow
for extending the spatial coverage of the DB aerosol product over all land
areas (Hsu et al., 2013; Sayer et al., 2013, 2014). AErosol RObotic NETwork
(AERONET) Cimel sun photometric measurements have been extensively used for
the validation of the MODIS over-land and over-ocean products (e.g., Chu et
al., 2002; Remer et al., 2002, 2005; Levy et al., 2010; Shi et al., 2013).
In this work, AOD550 over both land and sea and the fine mode ratio
(FMR550) over sea from Collection 051 were used at a spatial resolution
of 10 km (at nadir). The uncertainty of the MODIS aerosol optical depth has
been estimated at ±(0.05 + 0.15 AOD) over land (Chu et al., 2002; Levy
et al., 2010) and ±(0.03 + 0.05 AOD) over ocean (Remer et al., 2002)
relative to the AERONET AOD. Specifically, for the DT data used in this work,
only high-quality retrievals are used over land. This means that the data
have a quality assurance confidence (QAC) flag equal to 3 (high confidence).
For retrievals over sea we use data with a QAC flag of 1 (marginally good),
2 (good) and 3 (see Levy et al., 2009 for details). The pre-launch
uncertainty of FMR550 is ±30 % over ocean (Remer et al.,
2005), while over land this parameter is by no means trustworthy and should
only be used in qualitative studies (e.g., see Georgoulias and Kourtidis,
2011). In cases where the DT algorithm does not provide products over land,
especially over bright arid and semi-arid regions of northern Africa,
AOD550 values from the DB algorithm are used in our work. The expected
uncertainty of the DB product used here is ±(0.05+0.2 AOD) relative
to the AERONET AOD (Hsu et al., 2006). The analyzed datasets cover the
period from March 2000 to December 2012 for Terra and from July 2002 to December 2012 for Aqua
MODIS, covering the region of the Eastern Mediterranean. The Collection 051
DB data for Terra are only available until December 2007 due to calibration
issues; nevertheless, these data are carefully used within our analysis to
obtain a complete image of the aerosol load over the region.
AERONET ground-based observations
For the evaluation of the MODIS AOD550, version 2.0 level 2.0 high-quality cloud-screened data from 13 AERONET Cimel network ground stations in
the region of the Eastern Mediterranean have been acquired
(http://aeronet.gsfc.nasa.gov). The stations were selected such that their
operation period covers at least 2 years and there are at least 100 common
days of co-localized AERONET and MODIS observations. AERONET Cimel
sun photometers measure solar radiation every 15 min within the spectral
range from 340 to 1020 nm (Holben et al., 2001). The spectral measurements
allow for the retrieval of columnar aerosol properties (see Holben et al.,
1998; Dubovik and King, 2000; Dubovik et al., 2000, 2002). The AERONET AOD
uncertainty is on the order of 0.01–0.02 (Eck et al., 1999), being larger at
shorter wavelengths. Here, we use quadratic fits on a log–log scale to
interpolate the AERONET data (AODs at 440, 500, 675 and 870 nm) to the MODIS
band-effective wavelength of 550 nm (Eck et al., 1999; Levy et al., 2010).
Therefore, we can directly compare the MODIS AOD550 retrievals against AERONET
observations. Simultaneous measurements of the Ångström Exponent
(AE) for the spectral range 440–870 nm (AE440–870) from the 13 AERONET
stations mentioned above were also utilized in this work in order to account
for days with dust dominance. The uncertainty of the AE is significantly
higher than the AOD uncertainty, especially under low AOD conditions. Li
et al. (2014) found that the uncertainty for a typical northern hemispheric
AERONET station (GSFC) is ∼ 0.6 during winter, when AODs are
significantly lower compared to summer (∼ 0.15).
LIVAS CALIOP/CALIPSO dust climatology
Dust aerosol optical depths at 532 nm (AOD532) from CALIOP/CALIPSO
(Cloud-Aerosol Lidar with Orthogonal Polarization instrument aboard
Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations
satellite) at a resolution of 1∘ × 1∘ are also used here for the
period 2007–2012. CALIPSO measures cloud and aerosol properties flying at a
705 km sun synchronous polar orbit with a 16-day repeat cycle and an
Equator crossing time close to that of the Aqua satellite (13:30 LT). The
dust product used here comes from a Saharan-dust-optimized retrieval scheme
that was developed within the framework of the LIVAS (Lidar Climatology of
Vertical Aerosol Structure for Space-Based LIDAR Simulation Studies) project
(Amiridis et al., 2015), and has been presented in detail in
Amiridis et al. (2013). In brief, the LIVAS dust product is optimized for Europe by applying
a lidar ratio of 58 sr instead of 40 sr to Level 2 dust-related backscatter
products. This correction results in an improvement of the AOD532
product. Comparison against spatially and temporally co-located AERONET
observations (Amiridis et al., 2013) returned an absolute bias of
∼ -0.03. The corresponding reported biases for the original
CALIPSO data are significantly higher (∼ -0.10). The bias is
even lower when compared against MODIS satellite-based observations. Other
improvements of this product are related to the use of a new methodology for
the calculation of pure dust extinction from dust mixtures and the
application of an averaging scheme that includes zero extinction values for
the non-dust aerosol types detected. Overall, this product (hereafter
denoted as LIVAS dust product) exhibits better agreement with observations
from MODIS and AERONET and simulations from the BSC-DREAM8b dust model over
northern Africa and Europe than the standard CALIPSO data; hence it is an ideal
tool for the evaluation of other satellite-based products.
Earth Probe TOMS and OMI satellite observations
For this work, ultraviolet (UV) Aerosol Index (AI) data (Herman et al., 1997) from the
Earth Probe TOMS (Total Ozone Mapping) spectrometer aboard Earth Probe for
the period January 2000–September 2004 at a resolution of 1∘ (latitude) × 1.25∘
(longitude) and the OMI (Ozone Monitoring Instrument) sensor aboard EOS AURA
for the period October 2004–December 2012 at a resolution of 1∘ × 1∘ were
acquired through the GIOVANNI web database
(http://giovanni.gsfc.nasa.gov/giovanni/). Earth Probe TOMS continued the
record of the first three TOMS instruments aboard Nimbus-7, Meteor-3 and
ADEOS flying in a sun synchronous orbit at an altitude of 740 km with an
instantaneous field of view size of 39 × 39 km2 at nadir. The
instrument had an ascending node Equator crossing time at 12:00 LT, covering
85 % of the globe on a daily basis from July 1996 until December 2005. The
satellite was originally set to a 500 km sun synchronous orbit, but was set
to its final orbit after the failure of ADEOS satellite in June 1997. OMI is a
UV–VIS (ultraviolet–visible) nadir solar backscatter spectrometer (Levelt et al., 2006) that
continues the long TOMS record. OMI flies in a sun synchronous polar orbit
at an altitude of 705 km with an ascending node Equator crossing time at
13:45 LT. Its 2600 km viewing swath allows for almost daily global
coverage, while the spatial resolution of the instrument is 13 × 24 km2 at nadir.
The AI (also known as Absorbing Aerosol Index), which is calculated by the
two instruments, constitutes a qualitative indicator of the presence of UV
absorbing aerosols in the atmosphere such as biomass burning and dust
(Torres et al., 1998). Positive AI values generally represent absorbing
aerosols, while small or negative values represent non-absorbing aerosols.
The Version 8 algorithm is applied to spectral measurements from both TOMS
and OMI sensor to produce a consistent long-term AI time series (Li et al.,
2009). AI is calculated from the difference in surface reflectivity derived
from the 331.2 and 360 nm measurements, exhibiting an uncertainty of ±30 %
(Torres et al., 2007).
ERA-Interim reanalysis data
Wind speed (ws) data at 10 m above surface from the ERA-Interim reanalysis
(Dee et al., 2011) are used for 09:00 and 12:00 UTC on a daily basis for the
period 2000–2012. We use 09:00 and 12:00 UTC data in order to be closer to
the Terra and Aqua overpass time in the area, respectively. The various
ERA-Interim reanalysis fields are produced by ECMWF's Integrated Forecast
System (IFS) assimilating satellite- and ground-based observations. The
system includes a 4-D variational analysis with a 12 h analysis window.
The spatial resolution of the ERA-Interim data is ∼ 79 km, with
60 vertical levels from the surface up to 0.1 hPa, while the data can be
acquired at various resolutions (in this work 1∘ × 1∘) through
ECMWF's website (http://apps.ecmwf.int/datasets/data/interim-full-daily/).
Over ocean, the 10 m ERA-Interim wind speed exhibits a bias of less than
-0.5 m s-1 compared to quality-controlled in situ observations on a global
scale (Dee et al., 2011). Specifically, for the region of the Eastern
Mediterranean examined here, the 10 m ERA-Interim wind speed exhibits a bias
of -0.96 m s-1 (-16 %) compared to satellite-based observations from
QuikSCAT (Herrmann et al., 2011).
MACC reanalysis data
The daily MACC total and dust AOD550 data for the period 2003–2012 come
from the aerosol analysis and forecast system of ECMWF, which consists of a
forward model (Morcrette et al., 2009) and a data-assimilation module
(Benedetti et al., 2009). AOD550 measurements from the two MODIS
instruments aboard Terra and Aqua are assimilated by the MACC forecasting
system through a 4D-Var assimilation algorithm to produce the aerosol
analysis, leading to an improved AOD representation compared to observations
(see Benedetti et al., 2009; Mangold et al., 2011). Five aerosol species are
included within MACC, namely, mineral dust, sea salt, sulfates, black carbon
and organic matter. Three different sized bins are used for mineral dust and
sea salt particles, while the black carbon and organic material are
distributed to a hydrophilic and a hydrophobic mode. Dust and sea salt
emissions are given as a function of surface wind speed, while the emissions
of the other species are taken from inventories. The spatial resolution of
the MACC reanalysis data is ∼ 79 km with 60 vertical levels
from the surface up to 0.1 hPa and can be acquired from
http://apps.ecmwf.int/datasets/data/macc-reanalysis/ for the period
2003–2012. The MACC total and dust AODs have been evaluated against ground- and satellite-based observations (see Elguindi et al., 2010; Bellouin et
al., 2013; Inness et al., 2013; Cesnulyte et al., 2014; Cuevas et al., 2015),
showing that the MACC aerosol products generally capture the daily,
seasonal and interannual variability of aerosols well. As discussed in Bellouin
et al. (2013) the uncertainties of MACC total AOD550 (∼ 0.03)
and dust AOD550 (∼ 0.014) arise from uncertainties
in the MODIS retrievals, which are assimilated into the model and errors in
the forward modeling of total and component AODs.
GOCART data
Daily total and dust AOD550 data from the GOCART
chemistry–aerosol-transport model simulations (version 006) are used in this
study for the period 2000–2007. The GOCART model (see Chin et al., 2000,
2002, 2004, 2007; Ginoux et al., 2001, 2004) uses the assimilated
meteorological fields of the Goddard Earth Observing System Data
Assimilation System (GEOS DAS), which are generated by the Goddard Global
Modeling and Assimilation Office (GMAO). The data that are used were
acquired from an older version of NASA's GIOVANNI web database
(http://disc.sci.gsfc.nasa.gov/giovanni/) and come from a simulation
implemented at a spatial resolution of 2∘ (latitude) × 2.5∘
(longitude) with 30 vertical sigma layers (Chin et al.,
2009). The model includes physicochemical processes of major tropospheric
aerosol components (sulfates, dust, black carbon, organic carbon, sea salt)
and precursor gases (SO2 and dimethylsulfide), incorporating various
atmospheric processes. The total AOD550 from GOCART compared to
ground-based observations from the AERONET exhibits a relative mean bias
(mean(GOCART) / mean(AERONET)) of 1.120, 1.135 and 0.959 over Europe, northern Africa and for the whole globe, respectively.
Ancillary data
Apart from the main datasets presented above, three additional datasets were
used in order to support our findings. OMI/AURA daily gridded (Bucsela et
al., 2013) tropospheric NO2 columnar data (OMNO2d version 2.1) at a
spatial resolution of 0.25∘ × 0.25∘ were acquired from NASA's
GIOVANNI web database (http://giovanni.gsfc.nasa.gov/giovanni/) for the
period 2005–2012. The quality-checked data used in this work correspond to
sky conditions where the cloud fraction is less than 30 %. Planetary boundary
layer (PBL) SO2 daily gridded columnar data (OMSO2e version 1.1.7) were
also acquired from GIOVANNI for the same period. The OMSO2e gridded data
(0.25∘ × 0.25∘) used in this work are produced from best level-2
pixel data, screened for OMI row anomalies and other data quality flags. The
PBL SO2 column retrievals are produced with an algorithm based on
principal component analysis (PCA) of the OMI radiance data (Li et al.,
2013). Finally, monthly precipitation data from the 3B43 TRMM and Other
Sources Monthly Rainfall Product (version 7) at a spatial resolution of
0.25∘ × 0.25∘ for the period 2000–2012 were obtained from
GIOVANNI. This dataset is derived from 3-hourly precipitation retrievals
from the Precipitation Radar (PR), the TRMM Microwave Imager (TMI) and the
Visible and Infrared Scanner (VIRS) aboard the TRMM (Tropical Rainfall
Monitoring Mission) satellite merged with other satellite-based
precipitation data and the Global Precipitation Climatology Centre (GPCC)
rain gauge analysis (Huffman et al., 2007).
Results and discussion
Validation of MODIS gridded data using ground-based observations
As discussed in Sects. 2 and 3, the high-quality (QAC: 3) DT level-2
Collection 051 MODIS data used in this work were validated in detail against
data from 13 AERONET stations (see Fig. 1). The stations were selected to
make sure that their version 2.0 level 2.0 high-quality cloud-screened Cimel
sun photometric observations were covering at least 2 years and there were at
least 100 common days of AERONET and MODIS observations. The exact
geolocation of the AERONET stations is given in Table 1 (see also Fig. 1)
along with the period of available data, the hosting country, the type of
the station (e.g., urban/rural, coastal/continental) and the
corresponding mean overpass time of Terra and Aqua MODIS. First, we spatially and temporally
collocated the MODIS and AERONET observations by
temporally averaging AERONET measurements within ±30 min from the
MODIS overpass time (see Levy et al., 2010) and spatially averaging MODIS
measurements centered within a 25 × 25 km2 window around each station
(see Koukouli et al., 2010). The use of a collocation window equal to the
one used for the gridding procedure, practically allows us to validate the 0.1∘ × 0.1∘ MODIS gridded product at
the same time.
The regression lines between MODIS and AERONET AODs are shown in Fig. 3,
while details about the validation results can be found in Table 2. Overall,
the MODIS Terra DT Collection 051 data overestimate AOD550 by
11.59 % (normalized mean bias – NMB), with 63.28 % of the data falling within
the expected error (EE) envelope and 67.78 % within the pre-launch
expected error (plEE) envelope. The expected error envelope is defined as
AOD – | EE | ≤ AODMODIS≤ AOD + | EE |,
with EE being ±(0.05 + 0.15 AOD) (Levy et al., 2010) and
plEE being ±(0.05 + 0.20 AOD) (Kaufman et al., 1997). On the other
hand, the MODIS Aqua DT Collection 051 data overestimate AOD550 by
25.18 % (NMB), with 57.14 % of the data falling within the EE envelope
and 61.87 % within the plEE envelope. The percentage of the MODIS Terra
and Aqua data falling within the EE envelope is close to the 57 % given
in Remer et al. (2005) for the Eastern Mediterranean. The validation results can be found in Table S1
for each station separately. The results discussed in this paragraph are comparable to the ones appearing in
previous studies focusing on the Mediterranean region (see Papadimas et al.,
2009; Koukouli et al., 2010). In general, it is shown here that the MODIS
Terra Collection 051 data exhibit a better agreement with the ground-based
observations from AERONET than MODIS Aqua data do. Therefore, the statistics
appearing for MODIS Terra throughout the paper could be considered more
robust.
Results of the comparison of spatially (using a spatial window
around each station) and temporally (±30 min from the MODIS overpass
time) collocated MODIS Terra and Aqua (italics) Collection 051 level-2 and
AERONET sun photometric (quadratically interpolated) AOD550 observations
for the Eastern Mediterranean stations. The algorithms used for the
production of the validated MODIS data (DT and DB), the spatial window used
for the spatial collocation (25 × 25 km2 or 50 × 50 km2 window
around each station) with the AERONET data, the average MODIS and AERONET
AOD550 and the corresponding ±1σ values, the mean
difference between them, the normalized mean bias (NMB) and the corresponding
root mean square error (RMSE), the percentage of the collocation points that
fall within the expected error (EE) envelope and the pre-launch expected
error (plEE) envelope (expected uncertainty – EU envelope for DB data), the
correlation coefficient R, the slope (a) and the intercept of the regression
line (b) and the number of the collocation points are given in the table. L10
denotes the use of a collocation window of 50∘ × 50∘
as in Levy et al. (2010), while HQ denotes the use of high-quality data
only.
Alg.
Window
MODIS TERRA
AERONET
Mean diff.
NMB %
RMSE
In EE %
In plEE %
R
a
b
Obs
MODIS AQUA
DT
25 km
0.223 ± 0.163
0.200 ± 0.123
0.023 ± 0.106
11.59
0.11
63.28
67.78
0.76
1.007
0.022
6697
DT
25 km
0.247 ± 0.173
0.197 ± 0.121
0.050 ± 0.109
25.18
0.12
57.14
61.87
0.78
1.113
0.027
6283
DT
50 km(L10)
0.204 ± 0.152
0.194 ± 0.124
0.010 ± 0.085
5.10
0.09
70.17
74.64
0.83
1.016
0.007
6054
DT
50 km(L10)
0.224 ± 0.155
0.194 ± 0.125
0.030 ± 0.088
15.34
0.09
66.76
70.45
0.82
1.018
0.026
5557
DB
25 km
0.226 ± 0.177
0.186 ± 0.128
0.040 ± 0.162
21.38
0.17
–
51.90
0.47
0.657
0.104
2580
DB
25 km
0.242 ± 0.217
0.182 ± 0.118
0.06 ± 0.196
33.03
0.20
–
55.30
0.44
0.815
0.094
5345
DBHQ
25 km
0.229 ± 0.158
0.186 ± 0.132
0.043 ± 0.141
22.82
0.15
–
52.41
0.54
0.651
0.108
498
DBHQ
25 km
0.260 ± 0.220
0.186 ± 0.138
0.074 ± 0.204
39.84
0.22
–
52.34
0.42
0.670
0.136
896
To be in line with the global validation of the DT Collection 051 product by
Levy et al. (2010), we also performed a validation with the specifications
used in their work. We used a 50 × 50 km2 window for the spatial
collocation of the MODIS and AERONET data, while only days with at least five
MODIS retrievals and two AERONET measurements were taken into account. The
increased size of the collocation window improves the results of the
validation. As shown in Table 2, MODIS Terra DT Collection 051 data
overestimate AOD550 by 5.10 % (NMB), with 70.17 % of the data
falling within the EE envelope and 74.64 % within the plEE envelope. For
MODIS Aqua, the NMB is 15.34 %, while the percentage of the measurements
falling within the EE and plEE envelope is 66.76 and 70.45 %,
respectively. These results for the Eastern Mediterranean are close to the
global ones presented in Levy et al. (2010).
As discussed in Sect. 3.1, data from the DB algorithm were used over bright
arid and semi-arid regions of northern Africa for the production of the
0.1∘ × 0.1∘ MODIS gridded dataset for grid cells with no DT data.
Therefore, in this work we also perform a validation of the DB Collection
051 product over the region of the Eastern Mediterranean. In the case of DB
data, we first make use of all the available DB observations without any
quality filtering over the 13 AERONET stations. A spatial window of 25–30 km
has been typically used in the past for the collocation of MODIS DB data
with the AERONET observations (see Shi et al., 2011; Ginoux et al.; 2012;
Sayer et al., 2013, 2014), which is in line with the 25 × 25 km2 window
used here. The MODIS Terra DB data overestimate AOD550 by 21.38 %
(NMB), with 51.90 % of the data falling within the expected uncertainty
(EU) envelope, assuming a DB expected uncertainty of ±0.05 ± 20 %AODAERONET
(Hsu et al., 2006). The MODIS Aqua DB Collection 051
data overestimate AOD550 by 33.03 % (NMB), with 55.30 % of the
data falling within the expected uncertainty envelope. We repeated the
validation procedure for DB data, only taking the highest quality
data into account. The sample of available measurements was diminished by a factor
of 5 in the case of MODIS Terra and 6 in the case of MODIS Aqua, but the
results were quite similar to the ones for the unfiltered data.
Therefore, the use of unfiltered DB data during the gridding procedure does
not insert any significant uncertainty. The DB results for the 13 AERONET
stations examined here do not have the same agreement with the DT results and
the ones presented in previous studies utilizing DB Collection 051 data for
other stations and larger regions (see Shi et al., 2011; Ginoux et al.,
2012). However, it has been reported that stations in the region (e.g., Sede
Boqer in Israel) are among the ones with the greatest discrepancies between
MODIS DB and AERONET measurements (Ginoux et al., 2012). Nevertheless, as
commented in Sect. 3.1, the DB data constitute only a small fraction of the
data used for the production of the MODIS gridded dataset (only ∼ 1 %
of the 30 000 grid cells covering the Eastern Mediterranean have
only DB retrievals), and therefore they do not affect its
quality significantly. Only areas in northern Africa are expected to be affected by the
use of DB data due to the extended lack of DT data there.
As discussed in Sect. 3.1, the gridding procedure was repeated four times
using a gridding window of 25, 50, 75 and 100 km using MODIS Terra
AOD550 data for the year 2004, showing that the 25 km window is optimal
for capturing local pollution sources. In order to see how the size of the
gridding window affects the agreement between MODIS and AERONET data, we also
proceed to a validation of MODIS DT data against AERONET measurements
using different spatial collocation windows (25, 50, 75 and 100 km) and two
quality criteria, a “strict” one: at least two AERONET measurements for each
MODIS–AERONET pair and a “stricter” one: at least five MODIS retrievals and
two AERONET measurements for each MODIS–AERONET pair, as in Levy et al. (2010).
The results for the DT MODIS Terra and Aqua data are presented in Table S2.
In general, it is shown that the increased size of the
spatial collocation window leads to an improvement of the bias between
satellite- and ground-based observations. This is probably due to the
inclusion of more observations into the calculations, which diminishes the
noise of the MODIS observations. In addition, as expected, the stricter
quality criteria lead to a better agreement between MODIS DT and AERONET
data. Not only taking the NMB into account, but also the regression lines and
the other metrics appearing in Table S2,
it is concluded that the 50 km window is the best choice for the validation procedure in line with Ichoku
et al. (2002). On the other hand, the 25 km validation results are close to
the 50 km ones (see Table S2), and at the same time the 25 km gridding window
allows for a more efficient detection of local aerosol sources as shown in
Sect. 3.1. Taking this into account, we suggest that the 25 km window used
for the production of the 0.1∘ × 0.1∘ gridded MODIS dataset is the
optimal selection for studying the spatial variability of aerosols,
preserving, at the same time, the representativeness of the real aerosol load
over each specific spot.
Aerosol spatial variability and hot spots
The AOD550 spatial variability over the greater Eastern Mediterranean
region for the period 2000–2012 as seen from the Terra MODIS 0.1∘ × 0.1∘
dataset is presented in Fig. 4. Several aerosol hot spots that
coincide with megacities (e.g., Cairo, Istanbul), large cities (e.g., Athens,
Ankara, Alexandria, Izmir, Thessaloniki) or even medium-sized cities (e.g., Larissa, Limassol), industrial zones (e.g., OSTIM Industrial Zone in Ankara,
Turkey), power plant complexes (e.g., Maritsa Iztok complex in the Stara
Zagora province in Bulgaria, Ptolemaida–Kozani power plants in western
Macedonia, Greece), river basins (e.g., Evros River basin at the borders
between Greece and Turkey), etc., can be detected on the map. Indicatively,
in Fig. 4 we give a list of 35 local particle pollution sources in the
region; however, careful inspection of this map and the seasonal maps
presented in Fig. 6 allows for the detection of many more aerosol sources.
The results from the analysis of Aqua MODIS data are pretty similar as shown
in Fig. S10. A significant number of the local aerosol
sources can also be detected on the OMI 2004–2012 tropospheric NO2 and
PBL SO2 maps given in Fig. 5a and b, which reveal the origin of
aerosols over these regions (e.g., traffic, industrial activities).
However, there are regions of high aerosol load that cannot be seen in Fig. 5a
and b and vice versa, which is indicative of the significant role of other
anthropogenic or natural processes that contribute to the local aerosol load
(e.g., fires, soil dust from agricultural activities or arid regions, Sahara
dust transport).
AOD550 patterns over the Eastern Mediterranean as seen by MODIS
Terra during the period March 2000–December 2012 (March 2000–December 2007 for regions of northern Africa covered by DB data only). The color scale corresponds to the
AOD550 levels, while the top x axis and the right y axis correspond
to the longitude (∘ E) and latitude (∘ N), respectively. The
position of 35 aerosol hot spots is marked on the map (numbers from 1 to 35),
while the names of the places and the countries where the hot spots are
located appear on the right of the map. In the same figure, the exponential
growth of the number of satellite-based articles focusing on aerosols over
the greater Eastern Mediterranean from 1997 to 2014 is shown (black line).
The black dots represent the number of articles published within 3-year
intervals. The bottom x axis and the left y axis correspond to the years and
the number of published articles, respectively. The exponential growth
corresponds to a near doubling of the publication rate every 3 years.
![]()
AOD550 levels, the corresponding ±1σ values and
the number of gridded values used for the calculations over the Eastern
Mediterranean (EMT), over the land-covered part (EML), over the oceanic part
and over the nine subregions of the Eastern Mediterranean appearing in Fig. 1
based on the MODIS Terra and Aqua (italics) observations.
Region
MODIS TERRA
No. of values
MODIS AQUA
No. of values
AOD550
AOD550
EMT
0.215 ± 0.187
61 496 654
0.217 ± 0.199
49 522 934
EML
0.219 ± 0.165
25 923 766
0.239 ± 0.189
21 008 713
EMO
0.213 ± 0.201
35 572 888
0.202 ± 0.205
28 514 221
NBL
0.183 ± 0.163
5 563 495
0.187 ± 0.162
3 853 688
SBL
0.197 ± 0.152
7 345 829
0.207 ± 0.152
5 272 449
ANL
0.223 ± 0.146
7 948 817
0.228 ± 0.148
5 539 261
NAL
0.282 ± 0.192
5 065 625
0.306 ± 0.238
6 343 315
BSO
0.198 ± 0.150
6 433 951
0.183 ± 0.134
5 262 438
NWO
0.209 ± 0.162
11 645 069
0.197 ± 0.154
9 231 630
SWO
0.226 ± 0.266
6 202 893
0.223 ± 0.310
4 925 665
NEO
0.214 ± 0.196
4 807 910
0.199 ± 0.166
3 896 554
SEO
0.221 ± 0.236
6 483 065
0.210 ± 0.239
5 197 934
(a) Tropospheric NO2 levels and (b) planetary
boundary layer SO2 levels (in 1015 molecules cm-2) over the
Eastern Mediterranean as seen from OMI/AURA (2005–2012),
(c) topography (GTOPO elevation data in meters above sea level) and
(d) annual precipitation levels (in mm yr-1) from 3B43 TRMM and
Other Sources Monthly Rainfall Product (2000–2012).
![]()
The topography (Fig. 5c) and precipitation (see Fig. 5d for annual
precipitation levels for the period 2000–2012 from TRMM) are also major
determinants of the local AOD550 levels. For example, regions with
mountain ranges in the Balkan Peninsula (e.g., Pindus mountain range in
Greece, Dinaric Alps that run through Albania and the former Yugoslav
republics, the Balkan mountain range in central Bulgaria) are characterized
by low AODs (see Fig. 4). In contrast, regions of low altitude are
generally characterized by higher AODs because the majority of anthropogenic
activities is usually concentrated there. In addition, low-altitude regions
surrounded by high mountains are characterized by higher AODs as aerosols
cannot be easily transported by the wind (e.g., the industrialized regions in
central Bulgaria, which are confined between the high Balkan and Rodopi
mountain ranges). As precipitation is the major removal mechanism of
pollutants in the atmosphere, regions with high AOD550 are, in many
cases, connected to low precipitation levels and vice versa (see Figs. 4 and
5d). It has to be highlighted that the AOD550 over these regions is
high primarily due to the emissions and the atmospheric processes forming
aerosol particles. The low removal rates from precipitation just preserve
the AOD550 levels high. A striking example is the region of Anatolia in
central Turkey, which is characterized by lower precipitation levels and
higher aerosol loads compared to the surrounding regions. Additionally, the low
precipitation levels are partly responsible for the high aerosol loads
appearing over northern Africa.
Overall, the mean AOD550 for the whole period of interest is estimated
at 0.215 ± 0.187 for Terra and 0.217 ± 0.199 for Aqua MODIS for
the Eastern Mediterranean region, which is ∼ 45 % higher
than the global average appearing in recent studies (e.g., Kourtidis et al.,
2015). Over land, higher mean AODs are generally recorded (0.219 ± 0.165
for Terra and 0.239 ± 0.189 for Aqua MODIS) than over the sea
(0.213 ± 0.201 for Terra and 0.202 ± 0.205 for Aqua MODIS). All
these values, along with the mean AODs for the nine subregions of interest
covering the Eastern Mediterranean, can be found in Table 3.
The AOD550 spatial variability on a seasonal basis from MODIS Terra and
Aqua is presented in Fig. 6 along with the difference between the two
products. The majority of the local aerosol sources over land are more
prominent in summer. The limited washout by precipitation (see also
Papadimas et al., 2008) and also the enhanced photochemical production of
secondary organic aerosols (Kanakidou et al., 2011 and references therein)
contribute to the high AODs appearing over local sources. In addition,
during summer, over the region, there is typically a significant transport
of aerosols (e.g., see Kanakidou et al., 2011 and references therein) and
gaseous pollutants like SO2 and NO2 (see Georgoulias et al.,
2009; Zyrichidou et al., 2009) and biomass burning aerosols from
central-eastern Europe. Over the sea, a profound maximum is observed in
spring, extending across the northern African coast and the neighboring oceanic
areas, which is due to the well-documented transport of significant amounts
of dust from the Sahara (see Barnaba and Gobbi, 2004, and the list of
references given in the Introduction). The seasonal variability of aerosols
and the relative role of different aerosol types and various processes are
discussed in more detail in Sect. 4.4.
Seasonal AOD550 patterns over the Eastern Mediterranean as
seen by MODIS Terra (left column) during the period March 2000–December 2012
(March 2000–December 2007 for regions of northern Africa covered by DB data only) and
MODIS Aqua (middle column) during the period July 2002–December 2012. The differences
between MODIS Terra and Aqua AOD550 on a seasonal basis appear in the
right column.
![]()
The difference between MODIS Terra and Aqua Collection 051 AOD550 over
the Eastern Mediterranean is -0.002 (-1.40 %) for winter, -0.009
(-3.27 %) for spring, -0.011 (-4.46 %) for summer and 0.008 (4.40 %) for
autumn. AOD550 levels from Terra MODIS are lower than those from Aqua
MODIS over land for all seasons. Over the sea, Terra MODIS AOD550
levels are lower than that of Aqua MODIS only in winter. The fact that Terra
MODIS measurements are systematically higher than that from Aqua over the
sea by ∼ 0.01 on an annual basis is in line with the findings
of previous global studies for Collection 5 (e.g., Remer et al., 2006, 2008).
Locally, one can see regions with positive and negative differences between
Terra and Aqua MODIS AOD550. The patterns of the Terra–Aqua difference
per season are presented in Fig. 6c, f, i and l, while the patterns of the
percent difference are given in Fig. S11. The largest part
of the Terra–Aqua MODIS differences over land and sea that are observed
here may be attributed to the known calibration and sensor degradation
issues of MODIS (for details see Levy et al., 2010, 2013; Lyapustin et al.,
2014; Georgoulias et al., 2016). A significant effort has been undertaken to
address these issues in the new (Collection 6) MODIS product (e.g., Levy et
al., 2013; Lyapustin et al., 2014; Georgoulias et al., 2016), and a
repetition of similar analysis with Collection 6 data in the future would
be a valuable contribution. Taking the aforementioned issues
and the retrieval uncertainty of MODIS into account, it becomes more than obvious that the
attribution of observed differences between Terra and Aqua to the diurnal
variability of aerosol load (e.g., over biomass burning regions) in the
region is a difficult task. It is shown in Fig. S12 that
the diurnal variability of AOD550 from AERONET ranges significantly
from station to station. The average hourly departure from the daily mean
for the total of the 13 stations ranges from ∼ -5 to
∼ 5 %. Specifically, for the MODIS Terra and Aqua overpass
times, the AERONET AOD550 difference ranges from ∼ -10
to ∼ 10 % (see Fig. S12b). The Terra–Aqua AOD550
difference is negative for the total of the 13 stations ranging from
∼ -25 to ∼ -5 %. It is shown in Fig. S12b
that the two differences exhibit a similar variability from station to
station, which indicates that part of the observed Terra–Aqua difference is
indeed due to the diurnal variability of aerosols. However, as mentioned
above, the diurnal variability of aerosols is a very delicate issue and
should be comprehensively addressed in a future study. The same stands for
other kind of variabilities, which could be connected to local and regional
anthropogenic activities like, e.g., the weekly cycle of aerosols (see
Georgoulias and Kourtidis, 2011; Georgoulias et al., 2015).
Contribution of different aerosol types to the total AOD<inline-formula><mml:math display="inline"><mml:msub><mml:mi/><mml:mn>550</mml:mn></mml:msub></mml:math></inline-formula>
Annual contribution
As mentioned above, in our work we attempt to estimate the contribution of
different aerosol types to the total AOD550 over the region of the
Eastern Mediterranean; this was calculated following the methodology presented in
Sect. 3.2. For the land-covered areas, based on MODIS Terra observations, we
estimate that 52 % (0.112 ± 0.087) of the total AOD550 is due
to anthropogenic aerosols, 32 % (0.074 ± 0.080) due to dust and
16 % (0.034 ± 0.026) due to fine-mode natural aerosols (see Fig. 7).
For the oceanic areas, 41 % (0.086 ± 0.085) of the total AOD550
is due to anthropogenic aerosols, 34 % (0.076 ± 0.185) due to dust
and 25 % (0.054 ± 0.018) due to marine aerosols (see Fig. 7). The
results based on observations from MODIS Aqua are similar. Over land, 50 %
(0.117 ± 0.093) of the total AOD550 is anthropogenic, 35 %
(0.090 ± 0.102) is due to dust and 15 % (0.035 ± 0.028) due to
fine-mode natural aerosols, while, over the sea, 40 % (0.079 ± 0.080)
of the total AOD550 is of anthropogenic origin, 33 %
(0.070 ± 0.181) is due to dust and 27 % (0.054 ± 0.018) due to
marine aerosols (see Fig. 7). These results, along with the relative
contributions and the annual τa, τd, τn and
τm levels for each one of the nine subregions of interest (see
Fig. 1), are given in Table 4.
Relative contribution of anthropogenic aerosols, dust and fine-mode
natural aerosols to the total AOD550 over the land-covered part of the
Eastern Mediterranean based on MODIS Terra (a) and MODIS
Aqua (c) observations and relative contribution of anthropogenic
aerosols, dust and marine aerosols to the total AOD550 over the oceanic
part of the Eastern Mediterranean based on MODIS Terra (b) and
MODIS Aqua (d) observations.
![]()
(a) Anthropogenic aerosol (τa), (b) dust
(τd), (c) fine-mode natural aerosol (τn) and
(d) marine aerosol (τm) patterns over the Eastern
Mediterranean based on MODIS Terra observations during the period
March 2000–December 2012 (March 2000–December 2007 for regions of northern Africa covered by DB data
only).
![]()
Left column: latitudinal variability of anthropogenic aerosols
(τa), dust (τd), fine-mode natural aerosols (τn) and marine aerosols (τm) for four 5∘ longitudinal
bands (see embedded maps), covering the Eastern Mediterranean based on MODIS
Terra observations. Right column: latitudinal variability of dust extinction
coefficients at 532 nm in km-1 (color scale corresponds to the
extinction coefficients and left y axis to the atmospheric levels) and dust
aerosol optical depth at 532 nm (dotted line corresponding to the right
y axis) for the same four bands from LIVAS CALIOP/CALIPSO observations.
![]()
Relative contribution of anthropogenic aerosols, dust, fine-mode
natural and marine aerosols to the total AOD550 (bold) and the
corresponding τa, τd, τn and τm levels with their
±1σ values (in parentheses) over the Eastern Mediterranean
(EMT), over the land-covered part (EML), over the oceanic part and over the nine subregions of the Eastern Mediterranean appearing in Fig. 1 based on the
MODIS Terra and Aqua (italics) observations. The sum of the aerosol type AODs
per region does not necessarily correspond to the total AOD550 values
appearing in Table 3 as these results were for the total of the days with
aerosol retrievals, even for days when our aerosol type separation algorithm
was not applicable.
Region
Satellite
Contribution to MODIS TERRA/AQUA AOD550
Anthropogenic
Dust
Fine-mode natural
Marine
EML
TERRA
52 % (0.112 ± 0.087)
32 % (0.074 ± 0.080)
16 % (0.034 ± 0.026)
–
AQUA
50 % (0.117 ± 0.093)
35 % (0.090 ± 0.102)
15 % (0.035 ± 0.028)
–
EMO
TERRA
41 % (0.086 ± 0.085)
34 % (0.076 ± 0.185)
–
25 % (0.054 ± 0.018)
AQUA
40 % (0.079 ± 0.080)
33 % (0.070 ± 0.181)
–
27 % (0.054 ± 0.018)
NBL
TERRA
59 % (0.108 ± 0.101)
23 % (0.042 ± 0.046)
18 % (0.032 ± 0.030)
–
AQUA
59 % (0.110 ± 0.100)
24 % (0.045 ± 0.047)
17 % (0.033 ± 0.030)
–
SBL
TERRA
55 % (0.109 ± 0.088)
28 % (0.056 ± 0.058)
17 % (0.033 ± 0.026)
–
AQUA
55 % (0.113 ± 0.088)
29 % (0.060 ± 0.060)
16 % (0.034 ± 0.026)
–
ANL
TERRA
51 % (0.113 ± 0.075)
34 % (0.076 ± 0.068)
15 % (0.034 ± 0.023)
–
AQUA
50 % (0.114 ± 0.075)
35 % (0.079 ± 0.070)
15 % (0.034 ± 0.023)
–
NAL
TERRA
50 % (0.113 ± 0.083)
35 % (0.083 ± 0.085)
15 % (0.034 ± 0.025)
–
AQUA
48 % (0.118 ± 0.091)
38 % (0.099 ± 0.108)
14 % (0.035 ± 0.027)
–
BSO
TERRA
53 % (0.108 ± 0.103)
22 % (0.044 ± 0.101)
–
25 % (0.051 ± 0.016)
AQUA
51 % (0.094 ± 0.087)
22 % (0.042 ± 0.085)
–
27 % (0.051 ± 0.016)
NWO
TERRA
41 % (0.087 ± 0.090)
33 % (0.071 ± 0.142)
–
26 % (0.055 ± 0.020)
AQUA
40 % (0.079 ± 0.083)
32 % (0.066 ± 0.127)
–
28 % (0.055 ± 0.020)
SWO
TERRA
32 % (0.071 ± 0.070)
42 % (0.097 ± 0.257)
–
26 % (0.058 ± 0.018)
AQUA
32 % (0.093 ± 0.288)
41 % (0.072 ± 0.080)
–
27 % (0.059 ± 0.018)
NEO
TERRA
48 % (0.098 ± 0.094)
28 % (0.061 ± 0.144)
–
24 % (0.050 ± 0.016)
AQUA
46 % (0.086 ± 0.082)
28 % (0.057 ± 0.115)
–
26 % (0.050 ± 0.016)
SEO
TERRA
36 % (0.079 ± 0.070)
39 % (0.087 ± 0.224)
–
25 % (0.055 ± 0.016)
AQUA
36 % (0.075 ± 0.071)
38 % (0.080 ± 0.217)
–
26 % (0.055 ± 0.016)
For anthropogenic aerosols, the region with the highest relative
contribution is NBL (59 % for both Terra and Aqua MODIS), while the region
with the lowest relative contribution is SWO (32 % for both Terra and
Aqua MODIS) (see also Table 4). The spatial variability of τα
is presented in Fig. 8a for Terra MODIS and Fig. S13a
for Aqua MODIS, the patterns being similar in both cases. Over land, the
annual τa patterns are similar to the AOD550 patterns, the
highest values appearing over local particle pollution sources (cities,
industrial zones, etc.). Over the sea, τa is higher along the
coasts, while it drops significantly towards other directions. An
interesting feature here is that the oceanic region of Black Sea (BSO)
presents higher relative anthropogenic contributions than the rest of the
oceanic subregions but also than land areas with significant anthropogenic
sources (e.g., ANL and NAL). This is indicative of the transport of
atmospheric particles from central Europe and biomass burning aerosols
during the biomass burning seasons in April–May from Russia (across the
latitudinal zone 45–55∘ N) and July–August from southwestern
Russia and eastern Europe (Amiridis et al., 2010). These aerosols are
transported at much lower latitudes as shown in previous studies (e.g., Vrekoussis et al., 2005; Karnieli et al., 2009), reaching the Sahara
and the Middle East regions (Pozzer et al., 2015). The fact that τa drops gradually from the coasts is also seen in Fig. 9 where the
latitudinal variability of the optical depths of the different aerosol types
(τa, τd, τn and τm) is
presented for four bands that cover the whole Eastern Mediterranean. An
interesting feature is that τa increases nearby the shoreline
(particularly along the northern African coastal zone) before it gradually
decreases. Over land aerosols are located within the atmospheric boundary
layer, close to the emission sources, and hence their deposition and
removal from the atmosphere is more efficient than over the sea. The
particles, which are transported over the sea on the other hand, usually reach
greater heights, which prolongs their lifetime.
As shown in Fig. 9, the same feature is observed for dust. Indicatively,
τd and the relative contribution of dust to the total AOD550
on an annual basis over the oceanic regions of SWO and SEO are, in general,
higher than or comparable to the ones over NAL (see Table 4 for more details). In
Fig. 9, the MODIS-based τd latitudinal variability is presented
along with the latitudinal variability of dust AOD532 and extinction
coefficients of dust at 532 nm from LIVAS. As expected, in all cases τd decreases with distance from the large dust sources in the south and
southeast (Sahara and the Middle East), with local maxima over the
latitudinal zone from 35 to 40∘ N (especially for band 2 and band 3).
The latitudinal variability of τd is similar to the
latitudinal variability of dust AOD532 for all the four bands, despite
the fact that the MODIS-based data have a resolution 100 times higher
(0.1∘ vs. 1∘) and therefore are more sensitive to local
characteristics. Dust reaches heights up to ∼ 4–5 km in the
area; however, the largest fraction of dust mass is confined within the
first 2–3 km of the troposphere (see Fig. 9). The annual τd
patterns are shown in Fig. 8b for Terra MODIS (Fig. S13b for Aqua MODIS).
The main dust transport pathways over the oceanic areas of
the Eastern Mediterranean can be seen along with various local maxima over
land. The highest τd values over land appear over the regions of
NAL and ANL (see Table 4) and along the coasts. The high dust concentrations
appearing over these regions are not only due to the transport of dust from
the nearby deserts but also due to local dust sources. A recent study by
Liora et al. (2015) reports various local sources of wind-blown dust along
the coastal regions of Greece and Turkey, over the region of Anatolia in
Turkey, over the Greek islands, Crete and Cyprus, and regions close to the
coastal zone of Middle East. Their results are in good agreement with the
τd patterns presented in this work.
As shown in Fig. 7, fine-mode natural aerosols exhibit the lowest
contribution to the total AOD550 compared to the other aerosol types
over land. The spatial variability of τn is very low compared to
τa and τd as shown in Figs. 8c and 9. It is inferred
from the values appearing in Table 4 that τn increases slightly
as one moves from north to south; however, the relative contribution of fine-mode natural aerosols to the total AOD550 slightly decreases (i.e., 17.67 % over NBL and 14.97 % over NAL according to Terra MODIS
observations). The latitudinal variability and the percentages appearing in
Table 4 are in accordance with the relative contributions of biogenic aerosols
to the total AOD550 appearing over the Eastern Mediterranean in a
recent modeling study (Rea et al., 2015).
Similar to fine-mode natural aerosols over land, marine aerosols generally
have the lowest contribution to the total AOD550 compared to the other
aerosol types over the sea (see Fig. 7 and Table 4), except for BSO. The
variability of τm is very low compared to τa and τd.
On an annual basis, high τm values appear over the
Aegean Sea and the oceanic area between Crete and the northern African coast,
while slightly lower values appear along the coasts of the Eastern
Mediterranean (see Figs. 8d and 9). The τm patterns follow the
near-surface wind speed patterns in the region (see Fig. S14),
being in accordance with the τm, marine particulate
matter concentration or sea salt emission patterns appearing in other
studies (Im et al., 2012; Nabat et al., 2013; Rea et al., 2015; Liora et
al., 2015).
Seasonal contribution
The contribution of different aerosol types to the total AOD550 over the
Eastern Mediterranean varies from season to season. The relative contribution
of each aerosol type over EML and EMO for each season is shown in Fig. 10.
Over land, the relative contribution of τa, τd and
τn to the total AOD550 exhibits a low seasonal variability. The
relative contribution of anthropogenic aerosols to the total AOD550
ranges from 49 % in SON to 55 % in DJF based on Terra MODIS
observations and from 48 % in MAM and SON to 52 % in JJA based on
Aqua MODIS observations. In contrast, over the oceanic regions the relative
contribution of τa, τd and τm to the total
AOD550 exhibits a significant seasonal variability. The relative
contribution of anthropogenic aerosols to the total AOD550 ranges from
27 % / 27 % in DJF to 50 % / 47 % in JJA based on Terra/Aqua
MODIS observations. The percentages appearing here are in accordance with the
values appearing in Hatzianastassiou et al. (2009), where a different
satellite-based approach was followed. Indicatively, for the greater Athens
area, an average summertime anthropogenic contribution of ∼ 50 %
was found here based on Terra MODIS data, which is within the summer period
range of 47–61 % indicated in the study by Hatzianastassiou et
al. (2009). In addition, the corresponding values for the greater
Thessaloniki area, Crete, Cairo and Alexandria are 53, 38,
48 and 41 %, respectively, within the range of values (57–73,
36–52, 34–56 and 23–60 %) shown in Hatzianastassiou
et al. (2009). Only in the case of Ankara do our results suggest a lower
anthropogenic contribution (52 % vs. 71–84 %). Particularly for
Athens, Gerasopoulos et al. (2011), following a different approach
incorporating ground-based AOD observations and trajectory modeling, reached
similar results (annual contribution of ∼ 62 % from local and
regional sources and continental Europe, which is expected to be mostly of
anthropogenic origin). Similarly, for Crete, Bergamo et al. (2008), using a
different approach, also utilizing ground-based data, found an annual
anthropogenic contribution of ∼ 43 %.
Seasonal relative contribution of anthropogenic aerosols, dust and
fine-mode natural aerosols to the total AOD550 over the land-covered
part of the Eastern Mediterranean based on MODIS Terra (a, e, i, m)
and MODIS Aqua (c, g, k, o) observations and seasonal relative
contribution of anthropogenic aerosols, dust and marine aerosols to the total
AOD550 over the oceanic part of the Eastern Mediterranean based on MODIS
Terra (b, f, j, n) and MODIS Aqua (d, h, i, p)
observations.
![]()
The seasonal patterns of the anthropogenic aerosols (τa) over the
Eastern Mediterranean based on MODIS Terra observations are presented in
Fig. 11a, e, i and m, while the seasonal variability of τa over
the whole region, over the land-covered part and the oceanic part and over
the nine subregions of interest is presented in Fig. 12. The results based on
MODIS Aqua observations are similar and can be found in Figs. S15a, e, i and m and S16. Generally, the local hot spots are
detectable throughout the year; however, they are becoming much more
discernible in spring and especially in summer. As shown in Fig. 12a, τa nearly
doubles during the warm period of the year (spring–summer),
with the seasonal variability being stronger over the sea (Fig. 12c) than
over land (Fig. 12b). A clear peak is observed in summer, August being the
month with highest τa levels. As discussed in Sect. 4.3.1, the
summer peak is mostly the result of three basic reasons. The first one is the
deficiency of wet removal processes compared to the cold period. As shown in
Fig. S17, based on the TRMM satellite observations, August and July are the
months with the lowest precipitation levels over the land-covered part (a
drop of ∼ 75 % compared to winter months) and over the
oceanic part (a drop of ∼ 90 % compared to winter months)
of the Eastern Mediterranean, respectively. The second reason is the
enhancement of the photochemical production of secondary organic aerosols in
summer (Kanakidou et al., 2011), and the third reason is the transport of
pollution aerosols from central Europe and biomass burning aerosols from
southwestern Russia and eastern Europe during the biomass burning season in
July–August (Amiridis et al., 2010). The Etesians, which are persistent
northerly winds that prevail over the Eastern Mediterranean during summer,
bring dry and cool air masses and aerosols from the regions mentioned above
while blocking, at the same time, the transport of desert dust in the region
and dispersing local pollution in urban areas to levels typical for rural
areas (see Tyrlis and Lelieveld, 2013, and references therein). As seen in
Fig. 12a–l, a smaller but distinct, in most cases, τa peak appears
in April, mostly as a result of the transport of biomass burning aerosols
from Russia (across the latitudinal zone 45–55∘ N). This is in
line with the findings of Sciare et al. (2008) who detected traces of these
biomass burning aerosols over the island of Crete in southern Greece.
Seasonal (a, e, i, m) anthropogenic aerosol (τa),
(b, f, j, n) dust (τd), (c, g, k, o) fine-mode
natural aerosol (τn) and (d, h, i, p) marine aerosol
(τm) patterns over the Eastern Mediterranean based on MODIS Terra
observations during the period March 2000–December 2012 (March 2000–December 2007 for regions of
northern Africa covered by DB data only).
![]()
Seasonal variability of anthropogenic aerosols (τa),
dust (τd), fine-mode natural aerosols (τn) and marine
aerosols (τm) over the Eastern Mediterranean (EMT), over the
land-covered part (EML), over the oceanic part (EMO) and over the nine subregions
of the Eastern Mediterranean appearing in Fig. 1 based on MODIS Terra
observations. The error bars represent the ±1σ values
calculated from monthly gridded data.
![]()
As discussed above, the relative contribution of dust to the total
AOD550 over land exhibits a low seasonal variability ranging from 29 %
in DJF to 36 % in SON based on Terra MODIS observations and from
33 % in JJA to 38 % in SON based on Aqua MODIS observations (see
Fig. 10). Over the oceanic regions the relative contribution of dust to the
total AOD550 ranges significantly throughout a year from
26 % / 28 % in JJA to 42 % / 39 % in MAM based on Terra/Aqua
MODIS observations. The percentages appearing here are in accordance with model
and observational studies. For example, de Meij et al. (2012), using the
atmospheric chemistry general circulation model EMAC (ECHAM/MESSy Atmospheric
Chemistry), showed that dust contributes on an annual level ∼ 30 %
to the total AOD550 over stations located in the area of the Eastern
Mediterranean. Gerasopoulos et al. (2011) found a ∼ 23 % contribution of northern African dust to the total AOD over Athens using
ground-based AOD observations and trajectory modeling. Taking into account
that part of the ∼ 39 % local and regional sources appearing in
Gerasopoulos et al. (2011) is due to local dust sources, especially in
summer, it turns out that their results are in agreement with the
∼ 33 % relative contribution found in this work for the greater
Athens area based on Terra MODIS observations. The seasonal patterns of dust
(τd) over the region based on Terra MODIS observations are shown in
Fig. 11b, f, j and n, while the seasonal variability of τd over the
whole region, over land, over the sea and over the nine subregions of interest
is shown in Fig. 12. The corresponding results based on MODIS Aqua
observations are pretty similar and can be found in Figs. S15b, f, j and n and
S16.
As seen in Fig. 11f, in spring, mostly due to the strong Sahara dust events,
very high τd values appear over land regions in northern Africa,
Middle East, Anatolia and oceanic areas across the Eastern Mediterranean
(especially below 35∘ N). Dust loading over the sea exhibits two maxima,
one at the coastal zone of Libya and one across the coastal zone of Middle
East. The same two maxima, but with much lower τd values, appear in
summer (Fig. 11j) and autumn (Fig. 11n). Over land, the τd
patterns are similar in summer and autumn, the maximum values appearing over
the Anatolian Plateau and areas of northern Africa and Middle East. During
winter, dust maxima appear across the coastal zone of northern Africa, with
relatively low τd values across the coastal zone of Middle East
(Fig. 11b). In winter τd levels are as low over land compared to the
other seasons (Fig. 11b, f, j and n) as precipitation levels (see Fig. S17), and hence the wet scavenging of aerosols peaks. At the same
time, the local emissions of dust are low for regions away from the large
area sources in the south (Liora et al., 2015). In contrast, over the sea,
τd levels in winter are similar to or slightly higher for some areas
than those in summer and autumn (see Figs. 11 and 12) as this is the season
with the second highest frequency (after spring) of strong (∼ 21 %)
and extreme (∼ 26 %) desert dust episodes in the
region (see Gkikas et al., 2013, for details). February is by far the winter
month with the highest τd levels (see Fig. 12) in line with the
findings of Pey et al. (2013), who showed that the intensity of African dust
episodes over stations in Greece and Cyprus peaks in February. Dust exhibits
a strong peak in spring, April being the month with the highest τd levels in line with other studies (e.g., Israelevic et al., 2012;
Varga et al., 2014). The peak in April is a result of the high cyclonic
activity over northern Africa during this month as shown by
Flaounas et al. (2015). According to the same study, low-pressure systems are responsible
for ∼ 10–20 % of moderate and ∼ 40–50 % of
high and extreme Sahara dust transport events over the Eastern
Mediterranean. Northern Africa (Sharav) cyclones develop mainly in spring and
summer, while Mediterranean cyclones develop in winter and autumn. The
Mediterranean cyclones are more intense than Sharav cyclones. The region is
also affected by events bringing particles from dust source regions in the
eastern part of the Mediterranean Basin (Negev desert in Israel, Sinai in
Egypt, Anatolian Plateau in Turkey) and the Arabian deserts (Basart et al.,
2009; Pey et al., 2013; Abdelkader et al., 2015). Dust remains in the
atmosphere for a period of 1–4 days undergoing chemical aging before being
removed (see Abdelkader et al., 2015, and references therein). The seasonal
variability of τd is much stronger and the spring maxima much
more prominent over the sea (see Fig. 12). This is expected, as dust is only
occasionally transported over the sea during episodic events, while over
land, local sources also contribute to the dust burden, especially in summer
due to the dryness of soil. For example, over NBL, a broad spring-summer
peak is observed, June being the month with the highest τd
levels. As one moves south (SBL, ANL and NAL) the April peak becomes more
prominent.
The relative contribution of fine-mode natural aerosols to the total
AOD550 over land exhibits a very low seasonal variability, ranging from
15 % in MAM and SON to 16 % in DJF and JJA based on Terra MODIS
observations and from 14 % in DJF and SON to 15 % in MAM and JJA based
on Aqua MODIS observations (see Fig. 10). The seasonal variability is also
very low, the highest values appearing in spring and summer (Fig. 12).
Despite the generally low contribution of fine-mode natural aerosols to the
total AOD550 over the Eastern Mediterranean, τn levels are
similar to τd levels during winter months over specific regions
(NBL and SBL). The low seasonal variability can also be seen in Fig. 11c,
g, k and o where the patterns of fine-mode natural aerosols (τn)
are presented.
The seasonal relative contribution of marine aerosols to the total
AOD550 over the oceanic regions of the Eastern Mediterranean is shown
in Fig. 10. τm ranges from 20 % in MAM to 35 % in DJF based
on Terra MODIS observations and from 21 % in MAM to 36 % in DJF based
on Aqua MODIS observations (see Fig. 10). Like in the case of fine-mode
natural aerosols, the seasonal variability is very low, but here the highest
values appear in winter (Fig. 12). Due to the linear relation of τm and
near-surface wind speed within our algorithm (see Fig. 2), the
τm seasonal variability and patterns follow the wind speed ones
(see Figs. 11d, h, l and p and S14). Marine aerosol concentrations are lower
close to the coastlines, while the highest concentrations (see Liora et al.,
2015) and τm values within the Eastern Mediterranean appear over
the Aegean Sea (see Fig. 11). Overall, the τm patterns are in
accordance with the τm, marine particulate matter concentration and
sea salt emission patterns from previous studies (Im et al., 2012; Nabat et
al., 2013; Rea et al., 2015; Liora et al., 2015).
Summary and conclusions
In this work, satellite data from MODIS Terra (March 2000–December 2012) and Aqua
(July 2002–December 2012) were analyzed separately in order to examine the spatial
and temporal variability of aerosols over the Eastern Mediterranean. A high-resolution (0.1∘ × 0.1∘) MODIS gridded dataset was compiled using
a method that could be used in future regional studies. A number of tests
were implemented, and the dataset was validated in detail using
sun photometric observations from 13 AERONET stations. According to the
validation, the statistics appearing for MODIS Terra throughout the paper
could be considered more robust, while areas in northern Africa are expected
to be affected by the extended use of DB data, which do not exhibit a very
good matching with the ground-based observations. It is shown that the
gridding method we use offers the best compromise for studying the spatial
variability of aerosols on a regional or local scale, preserving, at the same
time, the representativeness of the real aerosol load over each specific
spot.
Based on MODIS observations, the average AOD550 levels over the region
of the Eastern Mediterranean are ∼ 0.22 ± 0.19, which is
∼ 45 % higher than the global mean. A number of aerosol hot
spots that coincide with megacities, large- and even medium-sized cities,
industrial zones, power plant complexes, river basins, etc., can be detected
on the AOD maps. A number of local aerosol sources can also be seen on
satellite retrieved tropospheric NO2 and planetary boundary layer
SO2 maps from OMI/AURA. This is indicative of the strong presence of
anthropogenic aerosols over these regions. Topography and precipitation also
play an important role. Generally, regions with mountain ranges are
characterized by low AODs, while regions of low altitude are characterized by
higher AODs. Regions with high AOD550 are, in many cases, connected to
low precipitation levels and vice versa. Precipitation is the major washout
mechanism of atmospheric pollutants. Low removal rates from precipitation
contribute in preserving the high AOD550 levels that are a result of
emissions and other atmospheric processes.
The AOD550 patterns over the Eastern Mediterranean exhibit a
significant seasonal variability, which is mostly driven by precipitation,
photochemical production of secondary organic aerosols, transport of
pollution and biomass burning aerosols from central and eastern Europe and
transport of dust from the Sahara and the Middle East. Differences
between the MODIS Terra and Aqua Collection 051 AOD550 over the Eastern
Mediterranean are generally small (∼ -8 % over land and
∼ 5 % over the sea). The comparison of the Terra–Aqua
differences with diurnal variabilities from the AERONET stations showed that
only a part of the observed differences is due to the diurnal variability of
aerosols.
The MODIS data were combined with data from other satellites (Earth Probe
TOMS, OMI/AURA), reanalysis projects (ERA-Interim,
MACC) and a chemistry–aerosol-transport model (GOCART) to calculate the
contribution of different types of aerosols to the total AOD550. The
algorithm used was optimized for the Eastern Mediterranean through a number
of tests and comparison with LIVAS CALIOP/CALIPSO dust retrievals and
AERONET ground-based observations. A different approach is used for land and
sea as there is not any reliable satellite-retrieved quantity to separate
the contribution of fine- and coarse-mode aerosols over land surfaces.
Overall, for the land areas, based on MODIS Terra observations, 52 %
(0.112 ± 0.087) of the total AOD550 is due to anthropogenic
aerosols, 32 % (0.074 ± 0.080) due to dust and 16 % (0.034 ± 0.026)
due to fine-mode natural aerosols (see Fig. 7). For the oceanic
areas, 41 % (0.086 ± 0.085) of the total AOD550 is due to
anthropogenic aerosols, 34 % (0.076 ± 0.185) due to dust and 25 %
(0.054 ± 0.018) due to marine aerosols. The results based on
observations from MODIS Aqua are similar to the MODIS Terra ones and are in
accordance with previous studies.
Over land, the τa maxima are detected over local particle
pollution sources (cities, industrial zones, etc.). Over the sea, τa is higher along the coasts, being significantly lower at greater
distance. Very high τd values appear over land regions in northern Africa, the Middle East, Anatolia and oceanic areas across the Eastern
Mediterranean, especially for latitudes below 35∘ N. Over the sea, dust
loading exhibits two maxima, one at the coastal zone of Libya and one across
the coastal zone of the Middle East. τd decreases with distance
from the large dust sources in the south and southeast. Generally, dust
reaches heights of up to ∼ 4–5 km in the area, the largest
fraction of dust mass being confined within the first 2–3 km of the
troposphere. The spatial variability of τn and τm is
very low compared to τa and τd, following the total
AOD550 patterns and the near-surface wind speed patterns,
respectively.
Over land, the relative contribution of anthropogenic aerosols, dust and
fine-mode natural aerosols to the total AOD550 exhibits a low seasonal
variability, while over the sea the relative contribution of anthropogenic
aerosols, dust and marine aerosols shows a significant seasonal variability.
τa nearly doubles during the warm period of the year
(spring–summer), August and April being the months with the highest τa levels.
The summer peak is mostly the result of low precipitation
levels, enhancement of the photochemical production of secondary organic
aerosols and the transport of pollution aerosols from central Europe and biomass
burning aerosols from southwestern Russia and eastern Europe during the
biomass burning season in July–August. The spring maximum in April is mostly
the result of transport of biomass burning aerosols from Russia in line with
previous studies. Dust exhibits a strong peak in spring (April), especially
over the southern regions. April is the month with the highest τd
levels as a result of the high cyclonic activity over northern Africa. The
seasonal variability of dust is much stronger, and the spring maxima are much
more prominent over the sea as dust is only occasionally transported there
during episodic events, while over land, local sources contribute to the
dust burden, especially in summer due to the soil dryness. The seasonal
variability of fine-mode natural aerosols is very low, the highest values
appearing in spring and summer. Marine aerosols also present a very low
seasonal variability, the highest values appearing in winter due to the high
near-surface wind speeds.
Overall, it is suggested that the AOD550, τa, τd,
τn and τm high-resolution gridded dataset that was
compiled in this work could be used in a number of future atmospheric and
biological studies focusing on the region of the Eastern Mediterranean (e.g., satellite- and ground-based studies on aerosol–cloud–radiation interactions,
experimental and field campaign studies on aerosols and clouds and research
on the impact of aerosols on human health and nature). It is also
acknowledged that a future update of the results presented here using more
recent releases of MODIS aerosol data (e.g., Collection 6) and aerosol
reanalysis datasets (e.g., NASA's Modern-Era Retrospective analysis for
Research and Applications Aerosol Reanalysis) would be a useful
contribution.