The regime of aerosol asymmetry parameter over 1 Europe , Mediterranean and Middle East based on 2 MODIS satellite data : evaluation against surface 3 AERONET measurements

16 Atmospheric particulates are a significant forcing agent for the radiative energy 17 budget of the Earth-atmosphere system. The particulates' interaction with radiation, 18 which defines their climate effect, is strongly dependent on their optical properties. In 19 the present work, we study one of the most important optical properties of aerosols, 20 the asymmetry parameter (gaer), over sea surfaces of the region comprising North 21 Africa, the Arabian peninsula, Europe, and the Mediterranean basin. These areas are 22 of great interest, because of the variety of aerosol types they host, both anthropogenic 23 and natural. Using satellite data from the collection 051 of MODIS (MODerate 24 resolution Imaging Spectroradiometer, Terra and Aqua), we investigate the spatio25 temporal characteristics of the asymmetry parameter. We generally find significant 26 spatial variability, with larger values over regions dominated by larger size particles, 27 e.g. outside the Atlantic coasts of north-western Africa, where desert-dust outflow 28 takes place. The gaer values tend to decrease with increasing wavelength, especially 29 over areas dominated by small particulates. The intra-annual variability is found to be 30 small in desert-dust areas, with maximum values during summer, while in all other 31 areas larger values are reported during the cold season and smaller during the warm. 32


Introduction
Atmospheric aerosol particles interact with radiation, mainly the short wave (SW or solar) part of the spectrum, modifying the energy budget of the Earth-atmosphere system.The aerosol effect is either direct, through the scattering and absorption of solar radiation, and thus reducing the incoming solar radiation flux at the surface, indirect, through the modification of cloud properties, or semi-direct, due to the absorption of solar radiation and consequent modification of the atmospheric temperature profile, convection, and cloud properties (e.g.Graβl, 1979;Hansen, 1997;Lohmann and Feichter, 2005).
The interaction of particles with the solar flux, which defines their climate role, strongly depends on their optical properties (Hatzianastassiou et al., 2004;2007), which cannot be covered globally by surface in situ measurements.Besides the aerosol optical depth (AOD), one of the most important optical properties of atmospheric particles, which is used in radiative transfer, climate, and general circulation models, is the asymmetry parameter (gaer).The asymmetry parameter describes the angular distribution of the scattered radiation and determines whether the particles scatter radiation preferentially to the front or back.The globally available satellite based AOD data are considered to a great extent as reliable and adequate, due to significant developments in surface and satellite measurements during the last two decades, and particularly the arrival of MODIS in 2000, which is regarded as one of the most reliable datasets (Bréon et al., 2011;Nabat et al., 2013).On the other hand, despite the important role of the asymmetry parameter, relevant global coverage data are measured only for the few last years, or are available in long-term aerosol climatologies such as Global Aerosol Data Set (GADS, Koepke et al. 1997) and Max Planck Aerosol Climatology (MAC, Kinne et al., 2013).Even so, asymmetry parameter data are usually examined for regions with limited geographical extent and temporal coverage (Di Iorio et al, 2003), without intercomparison between alternative data platforms.
The goal of the present work is the study of the spatiotemporal distribution of the aerosol asymmetry parameter, using the most recent data from MODIS (MODerate resolution Imaging Spectroradiometer, collection 051).Emphasis is given to the comparison between the provided MODIS data and respective reliable surface measurements of the global AERONET, in order to gain insight on the quality of the former.
For this study we focus on the region defined by latitudes 5°N to 70°N and longitudes 25°W to 60°E, including North Africa, the Arabian peninsula, Europe, and the greater Mediterranean basin (Fig. 1).This area is selected because it is of particular scientific interest due to the simultaneous presence of a variety of particles, both natural and anthropogenic (e.g.desert dust, marine, biomass burning, anthropogenic urban / industrial pollution) as shown in previous studies (Lelieveld et al., 2002;Smirnov et al., 2002;Sciare et al., 2003;Pace et al., 2006;Lyamani et al., 2006;Gerasopoulos et al.,2006;Engelstaedter et al., 2006;Satheesh et al., 2006;Kalivitis et al., 2007;Rahul et al., 2008;Kalapureddy et al., 2009;Alonso-Pérez et al., 2012;Zuluaga et al., 2012;Kischa et al., 2014) which makes this area ideal for aerosol studies.The presence of a variety of aerosols in the area is due to the fact that two of the largest deserts of the planet are partly included in our area of interest, i.e. the Arabian desert and the Sahara, while one finds also significant sources of anthropogenic pollution from urban and industrial centres, mainly in the European continent.Moreover, our area of interest and primarily its desert areas are characterised by a large aerosol load (large optical depth, Remer et al. 2008;Ginoux et al. 2012).In addition, significant regions in this area, more specifically the Mediterranean basin and North Africa, are considered climatically sensitive, since they are threatened by desertification (IPCC, 2007;2013).Finally, one more reason for the selection of study area is that the present study complements previous ones made by our team (e.g.Papadimas et al., 2008Papadimas et al., , 2012;;Hatzianastassiou et al., 2009) analysing other key aerosol optical properties, namely AOD, for the same region.This is the first study (to our knowledge) that focuses on asymmetry parameter over a geographically extended area, while at the same time compares satellite with ground-station data.

Data
Before presenting the data used in this study, a short introduction of the parameter studied is given here for readers more or less unfamiliar with it.The asymmetry parameter (or factor) is defined by: where P is the phase function, which represents the angular distribution of the scattered energy as a function of the scattering angle Θ and it is defined for molecules, cloud particles, and aerosols.The phase function can be expressed using the Legendre polynomials l  (see Liou, 2002) and 1  in Eq. ( 1) stands for l=1.The asymmetry parameter is the first moment of the phase function and it is an important parameter in radiative transfer.For isotropic scattering, g equals zero, which is the case for Rayleigh molecular scattering.The asymmetry parameter increases as the diffraction peak of the phase function sharpens.For Lorenz-Mie type particles, namely for aerosols and cloud droplets, the asymmetry parameter takes positive values denoting a relative strength of forward scattering, with values increasing with particle size.It can also take negative values if the phase function peaks in backward directions (90-180°).The phase function and its simple expression, the asymmetry parameter, along with the extinction coefficient (or equivalently the optical depth) and the single scattering albedo, constitute the fundamental parameters that drive the transfer of diffuse intensity (Joseph et al., 1976) and are used in modelling.Hence, the importance of aerosol asymmetry parameter is easily understood for enabling computations of aerosol radiative properties and effects (e.g.forcings).
In this work, we use MODIS aerosol asymmetry parameter (gaer) data, which we compare with in-situ measurements at AERONET stations.We provide a detailed description of the utilised data in the following sections.

Satellite MODIS Terra and Aqua data
MODIS is an instrument (radiometer) placed on the polar-orbiting satellites of NASA (National Aeronautics and Space Administration) Terra and Aqua, 705 km from the Earth, in the framework of the Earth Observing System (EOS) programme.Terra was launched on 18 December 1999, while Aqua was launched on 4 May 2002.The two satellites are moving on opposite directions and their equatorial crossing times are at 10:30 (Terra) and 13:30 (Aqua).MODIS is recording data in 36 spectral channels between the visible and the thermal infrared (0.44 -15 μm), while its swath width is of the order of 2330 km, which results in almost full planetary coverage on a daily basis.
Aerosol properties are monitored in 7 spectral channels between 0.47 and 2.13 μm and final results are derived through algorithms developed for aerosol quantities both over land and ocean (Kaufman et al., 1997;Tanré et al., 1997;Ichoku et al., 2002;Remer et al., 2005).MODIS data are organised in "collections" and "levels".
Collections comprise data produced by similar versions of the inversion algorithms, with the recent collection "051" including also outputs from the "Deep Blue" algorithm.Levels are characterised by data of different quality analysis and spatial resolution.
In this study we use daily MODIS data for the asymmetry parameter (gaer) provided on an 1°x1° grid (namely 100x100 km), from Collection 051, Level 3.These data were measured at wavelengths 470, 660, and 860 nm, only over oceanic regions, since they were derived through the algorithm "Dark Target" over ocean.These wavelengths were selected in order to match as much as possible those of the available corresponding AERONET gaer product (see sect.2.2).The period of analysis stretches from 24-2-2000 to 22-9-2010 for MODIS-Terra and from 4-7-2002 to 18-9-2010 for MODIS-Aqua.
The MODIS C051 gaer data are a derived product of the MODIS algorithm over ocean.This MODIS algorithm (http://modis.gsfc.nasa.gov/data/atbd/atbd_mod02.pdf, Remer et al., 2006) retrieves as primary products the AOD at 550 nm, the Fine (Mode) Weighting (FW, also known as fraction of fine-mode aerosol type, FMF) and the Fine (f) and Coarse (c) modes used in the retrieval, along with the fitting error (ε) of the simulated spectral reflectance.The algorithm reports additional derived parameters, such as the effective radius (re) of the combined size distribution, the spectral total, fine and coarse AODs or the columnar aerosol mass concentration.
Among them, gaer is also derived and reported at seven (7) wavelengths, 470, 550, 660, 860, 1200, 1600 and 2120 nm.The derived parameters are calculated (Levy et al., 2013) from information contained within the look-up table (LUT) and/or other retrieved products.For example, knowing the resulting total AOD and FMF, and which aerosol types were selected (or assumed), one can go back to the LUT, and recover additional information about the retrieved aerosol, such as the gaer.Hence, it should be noted that the derived gaer product is dependent on the used aerosol models (modes), since the algorithm is based on a LUT approach, assuming that one fine and one coarse lognormal aerosol modes can be combined with appropriate weightings to represent the ambient aerosol properties over the target (spectral reflectance from the LUT is compared with MODIS-measured spectral reflectance to find the "best"least squaresfit, which is the solution to the inversion).In the C051 algorithm there are four fine modes and five coarse modes, for which the spectral (at the aforementioned 7 wavelengths) aerosol asymmetry parameter values are given in Remer et al (2006).
We also used Level 3 daily Ångström exponent data from MODIS-Aqua C051, and also spectral aerosol optical depth data from MODIS-Aqua C006 datasets, from which we computed C006 Ångström exponent.These data were used to assess the validity of gaer data and their temporal tendencies, as discussed in section 3.2.3.

Ground based AERONET data
AERONET (Aerosol Robotic NETwork) is a global network of stations focused on the study of aerosol properties.AERONET currently encompasses about 970 surface stations (number continuously evolving) equipped with sun photometers of type CIMEL Electronique 318 A (Holben et al., 1998), which take spectral radiation flux measurements.
The optical properties of aerosols are extracted through the application of inversion algorithms (Dubovik and King, 2000).Data are provided on three levels (1.0, 1.5, and 2).In the present work, we use the most reliable cloud-screened and quality assured Level 2 data.AERONET calculates the asymmetry parameter at wavelengths 440, 675, 870, and 1020 nm.We employ daily Level 2 asymmetry parameter data from 69 stations (Fig. 1), contained in our study area (N.Africa, Arabian peninsula, Europe).
We choose only coastal stations, in order to maximize the coexistence of satellite marine gaer data with surface data.Also, in order to compare corresponding data between the satellite and station platforms, we perform comparison only for the 440, 675 and 870 nm.
3 Satellite based results

Geographical distributions
The spatial distribution of annual mean values of gaer is given in Fig. 2  particles.It should be reminded that the ability of atmospheric aerosol to absorb water affects the particle size (hygroscopic growth), as described by Köhler theory in the early 20 th century.It is also well known that relative humidity significantly affects aerosol optical properties (e.g.Pilinis et al., 1996;Kondratyev, 1999), namely AOD, single scattering albedo and gaer, by modifying the aerosol liquid water content, size and hence extinction coefficient and refractive indices.
In general, the largest gaer values (deep red colors) are observed off the coasts of West Africa (eastern tropical Atlantic Ocean) at all three wavelengths.High values are also found over the Red and Arabian Seas.These high values are due to strong dust outflows from the Saharan and Arabian deserts carrying out coarse aerosol particles (Prospero et al., 2002;Alonso-Pérez et al., 2012;Miller et al., 2008)  Gulf, compared to neighbouring areas (not shown here).These fine particles originate from the industrial activities in the Gulf countries related to oilfields or refineries (Goloub and Arino, 2000;Smirnov et al., 2002a,b;Dubovik et al., 2002).
The high gaer values over the northeastern tropical Atlantic Ocean as well as west of the Iberian coasts are possibly related with the presence of coarse sea salt particles.
On the other hand, the asymmetry parameter takes clearly smaller values over the Black Sea, where according to MODIS-Terra varies between 0.63 and 0.7 at 470 nm, 0.57 and 0.67 at 660 nm, and 0.55 and 0.66 at 860 nm, with the smallest values appearing over the Crimean peninsula (corresponding maximum Aqua values are smaller by 0.02).The small Black Sea gaer values can be associated with industrial but also biomass burning activities in nearby countries.A region of special interest is the Mediterranean basin since it hosts a large variety of aerosols like anthropogenic, desert dust or sea salt (e.g.Barnaba and Gobbi, 2004).The MODIS results over this region show relatively small gaer values, secondary to those of Black Sea, characterized by an increase from north to south, which is more evident at 660 and 860 nm.More specifically, based on MODIS-Terra, gaer over the Mediterranean takes values from 0.68 to 0.74 at 470 nm, while at 670 and 860 nm it ranges from 0.64 to 0.73 and 0.62 to 0.72, respectively.According to MODIS-Aqua the gaer values are slightly smaller again.The observed low values in the northern parts of the Mediterranean are probably associated with the presence of fine anthropogenic aerosols transported from adjacent urban and industrial areas in the north, especially in central Europe.In contrast, the higher gaer values in the southern Mediterranean, particularly near the North African coasts, can be explained by the proximity to the Sahara desert and the frequent transport of significant amounts of coarse dust (e.g.Kalivitis et al., 2006;Hatzianastassiou et al., 2009;Gkikas et al., 2009;2014).
The spatial distributions of climatological monthly mean gaer values from MODIS-Aqua at 470 nm reveal significant differences in the range and the patterns of the seasonal variability, depending on the area (Fig. 3).Thus, in tropical and sub-tropical areas of Atlantic Ocean (up to about 30°N), where dust is exported from Sahara, gaer keeps high values throughout the year, which reach or even exceed 0.74 locally.Over the regions of Arabian and Red Seas and the Gulf of Aden, which also experience desert dust transport, larger gaer values appear in the period from March to September, with a maximum in August (locally as high as 0.75-0.76).This seasonal behavior is in line with intra-annual changes of dust production over the Arabian peninsula indicated by MODIS Ångström Exponent (AE) and Deep Blue aerosol optical depth data (Ginoux et al., 2002), as well as over southwest Asia through in-situ data (Rashki et al., 2012), aerosol index from various platforms and MODIS Deep Blue AOD data (Rashki et al., 2014).Indeed, the production of dust there is relatively poor in winter, increases in March and April and becomes maximum in June and July (Prospero et al., 2002).Over the Arabian Sea, it is known that large amounts of desert dust are carried out during spring and early summer (Prospero et al, 2002;Savoie et al., 1987;Tindale and Pease, 1999;Satheesh et al., 1999).Nevertheless, according to MODIS, the seasonal variability of gaer remains relatively small there in line with a small seasonal variability in MODIS Deep Blue AE data (results of our analysis, not shown here).This can be explained by the presence of sea salt coarse particles throughout the year, with which dust particles co-exist.
A greater seasonal variability exists over the Persian Gulf, where gaer values are higher during spring and in particular in summer (up to 0.74 at 470 nm according to Aqua), and smaller in autumn and winter (area-minimum values smaller than 0.65).
This seasonal behavior can be explained taking into account the meteorological conditions over the greater area of the Gulf; mainly in spring and summer dry northwestern winds (Shamal) blow from northwest carrying desert dust from the arid areas of Iraq (Heishman 1999;Smirnov et al. 2002a,b;Kutiel and Furman, 2003).The transport of dust is gradually decreased in autumn and reaches its minimum in winter.
When the presence of desert dust is limited, a significant fraction of total aerosol load in the region consists of fine anthropogenic particles (Smirnov et al. 2002a,b), which can explain the observed relatively small gaer values in autumn and winter.
In the Mediterranean basin gaer exhibits a relatively small seasonal variation, with lower values tending to appear in summer, in line with the presence of fine anthropogenic or biomass burning aerosols in the area, transported from the Balkans or central Europe (Hatzianastassiou et al. et al., 2009).On the contrary, over the Black Sea, a clear seasonal cycle is apparent, with higher values in the cold period of the year and smaller in the warm one.More specifically, according to MODIS-Aqua, the values at 470 nm drop down to 0.61 in summer months whereas they reach 0.7 in January and December.This seasonality is in agreement with the summer biomass burning from agricultural activities and wildfires (Barnaba et al., 2011;Bovchaliuk et al., 2013), and the resulting abundance of fine particles.
It is also interesting to look at the geographical distribution of monthly gaer values in latitudes higher than 50°N, for which annual mean values were not given in Fig. 2 because of unavailability of data for all months.Off shore northern France (English Channel) and Germany the asymmetry parameter has small seasonally constant values (note that data do not exist for January and February).In these areas, the aerosol load consists mainly of anthropogenic polluted particles, which explains the small gaer values there.
In the Baltic Sea (values available from March to October) gaer shows a significant spatial and temporal variability.More specifically, it is small during summer whereas it increases, locally up to more than 0.7, in March and October.The smaller summer values can be explained by the presence of fine aerosols in the Baltic Sea originating from forest fires in Europe and Russia (Zdun et al., 2011).On the contrary, in autumn the local aerosol loading consists largely of coarse marine aerosols.It is also important to note that the Baltic Sea hosts significant amounts of anthropogenic industrial and urban aerosols throughout the year, but especially in summer (Zdun et al., 2011).
In the higher latitudes of Atlantic Ocean, where the presence of maritime aerosols is dominant, we note a remarkable month by month variation of asymmetry parameter, with low values in summer (values up to 0.59) against high values (up to 0.75-0.77) in spring (March, April) and autumn (October).This difference is possibly explained by the seasonal variability of aerosol size in the northern Atlantic.Apart from the presence of coarse sea salt throughout the year, in spring and summer small particles are formed through photochemical reactions of dimethylsulphide (DMS) emitted by phytoplankton decreasing the aerosol size.Moreover during summer fine anthropogenic aerosols are transported in the region from North America (Yu, 2003;Chubarova, 2009).These result in lower gaer values between May and August.
Based on MODIS-Terra, the patterns of spatial distribution are generally the same with Aqua, with slightly larger gaer values.At larger wavelengths (660, 860 nm) a decrease of gaer is observed, especially for its smallest values.Further details and an overall picture are given in section (3.2.1) which deals with climatological monthly mean values not at the pixel but at the regional level.

Seasonal variability
In order to provide an easier assessment of the seasonal cycle of aerosol asymmetry parameter and its changes from one region to another, but also among the different wavelengths (470, 660 and 860 nm), the study region was divided in 6 smaller subregions (see Fig. 1).The average values of monthly mean climatological data of the pixels found within each sub-region's geographical limits have been computed and are given in Fig. 4, for every wavelength, both for Terra and Aqua.It appears that the seasonal cycle differs between the sub-regions, as it has been already shown in the geographical map distributions discussed in the previous section.
At 470 nm (Fig. 4i), the intra-annual variability of gaer is greater over the Black Sea, where it is as large as 0.06 according to MODIS-Terra and 0.05 according to MODIS-Aqua, the north-eastern Atlantic Ocean (0.04 and 0.05 for Terra and Aqua, respectively) and the seas of North Europe (0.05 for both Terra and Aqua).In these regions, there is a tendency for smaller values during summer.More specifically, in the Black Sea the smallest gaer value (0.64) is observed in June, over the seas of North Europe in July and over the north-eastern Atlantic Ocean in August.In these regions, the largest values appear in the cold period of the year.Reverse seasonality with a large seasonal amplitude is observed over the Persian Gulf, where the variability is as large as 0.08, according to both MODIS-Terra and Aqua.The seasonal cycle of gaer over the Middle East exhibits a smaller range of variability (0.02 for MODIS-Terra and 0.03 for Aqua) along with a reverse seasonal variation, with maximum values in summer and minimum in winter.In the other two sub-regions (Mediterranean and eastern Atlantic Ocean) the annual range of values is small ( 0.02).It is noteworthy that in the Mediterranean Sea, there is a weak tendency of appearance of double maxima in winter and spring.The spring maximum should be associated with the presence of desert dust particles, which are transported from Sahara, mainly in the eastern Mediterranean in this season (e.g.Fotiadi et al., 2006;Kalivitis et al., 2007;Papadimas et al. 2008, Gkikas et al. 2009;Hatzianastassiou et al., 2009;Gkikas et al., 2013).There is also a similar transport of Saharan dust in the central and western Mediterranean during summer and autumn (e.g.Gkikas et al., 2009;2013), but then the predominance is not so clear because of the co-existence of fine anthropogenic aerosols.Regardless of the annual cycle, smaller gaer values are clearly distinguished over the Black Sea and North Europe seas throughout the whole year.
At 660 nm, the gaer values are lower than at 470 nm, in particular over Black Sea, North Europe and North-East Atlantic, whereas the intra-annual variability (range of gaer values) increases up to 0.10 (Terra) and 0.08 (Aqua) over the Black Sea.This increase is mainly attributed to the reduction of summer values due to the strong appearance of fine aerosols in this season.Also, at 660 nm, there is a clearer double annual variation of gaer over the Mediterranean Sea than at 470 nm.At 860 nm the general picture is similar to that of 660 nm though a further increase of month by month variability is noticeable.
In general, our results indicate that over the regions characterized by a strong presence of desert dust particles (eastern Atlantic and the Middle East and Mediterranean Seas) the annual range of variability of gaer is smaller than in the other regions.An additional feature above regions with desert dust is the smaller decrease of gaer values with increasing wavelengths.This is attributed to the lower gaer spectral dependence of coarse compared to fine particles (e.g.Dubovik et al, 2002;J. Bi et al., 2011).
We should note that the MODIS-Terra and Aqua gaer seasonal cycles are about similar but with generally greater Terra than Aqua values.shown in units decade -1 for both Terra and Aqua over their common time period, namely 2002 -2010, only if the trend is statistically significant at the 95% confidence level.We also performed the same analysis for the 660 and 860 nm (not shown), with similar results to the 470 nm wavelength.

Inter-annual variability and changes
In general, the estimated changes are relatively small.Terra produces widely statistically significant positive trends, showing that during the period of interest, the asymmetry parameter increased over the examined area, with very few exceptions.
The results from Aqua are statistically significant at considerably fewer cells, but also give a few points with decreasing gaer.Based on Terra data, the stronger increases are observed in the eastern and southern Black Sea, as well as over the Baltic and Barents Seas.According to MODIS-Aqua, negative trends are found over few Atlantic Ocean cells.Both Aqua and Terra report increases of gaer over the Persian Gulf, the Red Sea, South Black Sea, East Mediterranean, the coast of the Iberian Peninsula, and some coastal areas of West Africa.The differences encountered between the Terra and Aqua gaer trends may be attributed to the different time of passage of each satellite platform carrying the same MODIS instrument, given that everything else is the same.
Nevertheless, they may more probably be the result of calibration differences between the two MODIS sensors.It is known that there is a degradation of MODIS sensor (Levy et al., 2010;Lyapustin et al., 2014) impacting time series of MODIS products.
More specifically, it is also known that Terra suffers more than Aqua from optical sensor degradation.These calibration issues are known to affect MODIS AOD retrievals, producing an offset between Terra and Aqua, and they are also expected to affect aerosol asymmetry parameter, which is probably more sensitive to such calibration uncertainties than AOD.In this sense, the results of Fig. 5 shown here are not to be taken as truth but rather they are given as a diagnostic of a problematic situation with MODIS aerosol asymmetry parameter inter-annual changes.Such The overall gaer changes of Fig. 5 may hide smaller timescale variations of gaer, which are obtained by the time-series shown in Fig. 6. Results are given for the 7 subregions defined previously, at the three different wavelengths and for Terra and Aqua separately.A general pattern is the decrease of gaer values with increasing wavelength, in particular from 470 to 660 nm.The largest month to month and year to year variation is for Black Sea (Fig. 6i).Relatively large variability is also found in the sub-regions of NE Atlantic (6v), North Europe (6vi) and the Persian Gulf (6vii).On the contrary, small variability is noticed in the eastern Atlantic, where systematic dust outflows from Sahara take place leading to consistently high values of gaer.There are also some other interesting patterns, like the significant drop of gaer with wavelength in areas characterized by the presence of fine aerosols, namely the Black Sea, North Europe and the Persian Gulf (Figs,6i,vi,vii,respectively).The specific patterns of inter-annual changes of gaer are suggested by both Terra and Aqua, though a slight overestimation by Terra is again apparent in this figure.The obtained results of our analysis are meaningful and in accordance with the theory, underlining the ability of satellite observations to reasonably capture the gaer regime over the studied regions.

Possible uncertainties of MODIS aerosol asymmetry parameter
The MODIS aerosol asymmetry parameter is not a direct product of the MODIS retrieval algorithm, but it is rather a derived by-product.Since this parameter is dependent on aerosol modes used and relative weights, it is understood that there can be uncertainties associated with it.Therefore questions may arise about the validity of gaer and their spatial and temporal patterns presented in the previous sub-sections.
Given that, as already mentioned, it is an aerosol optical parameter that is valuable and highly required by radiative transfer and climate models, it is worth assessing it through comparison against another more common aerosol size parameter, namely the C051 MODIS Ångström exponent at the 550-865 nm wavelength pair (AE550-865) over ocean, which is an evaluated MODIS aerosol size product (Levy et al., 2010) that is extensively used in literature.similar patterns between C051 and C006.Small trends are found in both of them, in agreement with the small trends of asymmetry parameter reported in Fig. 5.We find that the sign of AE trends basically does not change from C051 to C006.This might be a signal that no changes of aerosol asymmetry parameter are expected in C006 and puts confidence on the C051 results given in the present study.

Evaluation against AERONET data
In order to evaluate the satellite-measured aerosol asymmetry parameter, we identified the AERONET stations inside our area of interest and finally utilised only the coastal ones, so that both satellite and surface data be available.The total number of these stations is 69, and their locations are shown in Fig. 1 (open and full circles).
Table 1 contains the comparison statistical metrics for all wavelengths (Pearson correlation coefficient, bias, root mean square error (RMSE), slope, intercept) of the comparison between surface daily mean data from AERONET and satellite data from MODIS-Terra and MODIS-Aqua, which correspond to the 1°x1° cell wherein each station is located.For this analysis, we use all cells and days with common data between Terra-AERONET and Aqua-AERONET.The mean differences are calculated as gaer(AERONET)-gaer(Aqua) and gaer(AERONET)-gaer(Terra).
In general, we may note that on an annual level, the MODIS-Terra and Aqua asymmetry parameter values at 470 nm are not in very good agreement with the respective data from AERONET at 440 nm, while the results at the largest wavelengths are more reassuring, though not being very satisfactory (increasing R and decreasing relative bias and RMSE values at 675/660 nm and 870 nm).At 870 nm (Table 1 and Fig. 9), correlation coefficients are found to be the largest and equal to 0.47 (AERONET-Terra) and 0.46 (AERONET-Aqua), while satellite data are slightly overestimated compared to the surface data (bias -0.035 or 5.54% and -0.015 or -

2.43%, respectively).
It is important to note that the agreement of satellite and surface data is better in spring and summer, for all studied wavelengths.Specifically, in case of MODIS-Aqua gaer, the correlation coefficients increase up to 0.35, 0.50 and 0.54 at 440/470 nm, 660/675 nm and 870 nm, respectively, while the bias decreases down to 0.0005 (0.07%), 0.003 (0.46%) and 0.007 (1.11%), respectively.
Moreover, we find that for all seasons gaer values at 870 nm and 660 nm, both from MODIS-Terra and MODIS-Aqua, are overestimated compared to gaer (AERONET) at the corresponding wavelengths (stronger overestimation at 870 nm and by Terra).Finally we note an underestimation of gaer at 470 nm from MODIS-Aqua, relative to the data by AERONET at 440 nm, while very small biases (<0.5 %) are found between Terra and AERONET at the same wavelengths.
In Fig. 9 we present a scatterplot comparison between MODIS and AERONET gaer data pairs.There is bias towards larger gaer values from both Aqua and Terra compared to AERONET, with Terra overpredicting more than Aqua.The root mean square error to the fit between MODIS and AERONET is very similar between Aqua and Terra.There are concerns on the application of ordinary least squares regression, arising from the assumption that as the assigned independent variable, AERONET values should be free from error.We cannot guarantee the validity of this assumption, so we recognize that the reported R and slope values from Fig. 9 and Table 1, if viewed as metrics of agreement between MODIS gaer and real g, may be subject to the effect of regression dilution and consequently biased low.This possible bias for R and slope could be neglected only if AERONET errors can also be considered negligible.
With the above caveat in mind, the applied least-squares fit line to the scatterplot comparison between matched MODIS-AERONET data pairs (Fig. 9) indicates that MODIS overestimates gaer more in the smaller than larger values, i.e. more for fine than coarse particles.
We present the frequency distributions of asymmetry parameter daily values (Fig. 10) on the days when data from all three databases (MODIS-Terra, MODIS-Aqua and AERONET) were provided.Fig. 10a corresponds to the whole area of interest, while Figs.10b and c correspond to two broad sub-regions with basic differences in the aerosol source, namely Europe with great anthropogenic sources, and Africa, Middle East and Arabian peninsula, with predominant natural sources and mainly desert dust.
There is an apparent skew in the MODIS-Terra and MODIS-Aqua gaer distributions, while the AERONET distributions are more symmetrical.Moreover, the satellite data distributions show larger values and smaller standard deviations compared to AERONET, with the Terra overestimation being more exaggerated.The disagreement is more pronounced in the sub-region of Europe, while in the sub-region of North Africa / Arabian peninsula, the distributions of satellite and surface data agree more thus confirming the finding of Fig. 9 based on the slope of applied linear regression fit.Values over Europe are generally smaller than over North Africa / Arabian peninsula (Fig. 3), which can be attributed to the presence of larger size particles of desert origin in the latter sub-region, in contrast to Europe, where due to industrial activity and frequent biomass burning the presence of smaller size particles is important.Therefore, the smaller gaer values (<0.6) in the frequency distributions of the whole area, are overwhelmingly contributed by the European sub-region, contrasting with larger values (0.7-0.75) being contributed by both sub-regions and even more by N. Africa/Arabian peninsula at larger gaer.
The overall comparison between satellite and surface gaer data performed in the scatterplot of Fig. 9 and Table 1 does not allow one to have an insight to how the comparison behaves spatially, namely how it differs from one region to another.This is addressed in Fig. 11, showing the comparison of satellite and surface data at the wavelength of 870 nm separately between MODIS-Terra -AERONET and MODIS-Aqua -AERONET.For this comparison, we selected AERONET stations for which there is satisfactory overlap between the time series from AERONET and the time series from MODIS, namely the number of common days between AERONET-Terra and AERONET-Aqua is larger than 100.This criterion is satisfied by 36 stations for AERONET-Terra and by 34 for AERONET-Aqua shown in Fig. 11.For each AERONET station we compute the Pearson correlation coefficient between the station data and the corresponding MODIS-Terra or Aqua data at 870 nm, for the 1°x1° cell containing the station.Moreover, there is the information if the trends between AERONET and either MODIS-Terra or Aqua have the same sign (blue color) or not (red color).
In the case of the gaer (AERONET)gaer (Terra) comparison, at 5 stations, (i.e. in 14% of total 36 stations), the correlation coefficient R is larger than 0.5 (largest R found is 0.64 at station "Bahrain"), while at 13 stations (36 %) and 26 stations (72%) R is larger than 0.4 and 0.3, respectively.With respect to the agreement on the sign of the trends, at 24 out of 36 stations (67%) there is a trend sign match and at 12 stations (33%) a mismatch.Nevertheless, it should be noted that no systematic spatial behaviour, i.e. homogeneous spatial patterns, is found concerning the performance of MODIS-Terra gaer against AERONET in terms of either the magnitude of correlation or the agreement of trends between the satellite and ground datasets.A similar picture emerges for the comparison gaer (AERONET)gaer (Aqua).In this case, there are again 5 stations (15% of total 34 stations) with R>0.5 (maximum value R=0.61 again at "Bahrain"), while at 13 stations (38%) and24 stations (71%) R is larger than 0.4 and 0.3, respectively.Also, we see that at 22 stations (65%) there is a trend sign match and at 12 (35%) there is a mismatch.

Summary and Conclusions
Using satellite data from collection (051) of MODIS-Terra and Aqua data, we examine the spatiotemporal variations of the aerosol asymmetry parameter (gaer) over North Africa, the Arabian peninsula and Europe.To our knowledge, this is the first time that a satellite (MODIS) based dataset of gaer, assessed and evaluated (against AERONET data), is used for the study region.This is important, since such an evaluated satellite dataset is very useful for many applications, like radiative transfer and climate modelling as well as for remote sensing.The advantages of MODIS gaer data are that: (i) They ensure complete spatial coverage over sea surfaces surrounding Europe, Mediterranean and Middle East, which is essential for investigating and understanding physical processes related to aerosols.These processes are strongly dependent on the aerosol radiative and optical properties, gaer being one of the three key ones (the other two being aerosol optical thickness and single scattering albedo).Such a complete spatial coverage is especially required by radiative transfer and climate models.
(ii) They provide with spectral gaer values, at 7 wavelengths from 470 to 2130 nm, which are of essential importance for radiative transfer models.Such spectrally resolved aerosol optical properties can induce significant differences in model computations of aerosol radiative effects (Hatzianastassiou et al., 2007).
(iii) They provide a relatively long temporal coverage, i.e. 8 years, which is significant for examining seasonal and inter-annual cycles and changes of this aerosol optical property, especially combined with the complete spatial coverage.This is also important since it provides a reasonable statistical bed for attempting evaluations through comparison against other gaer data like the AERONET.
(iv) They constitute the first to know so far satellite based gaer dataset; until now, the utilized gaer data in modelling or other analyses were taken from in-situ measurements or aerosol models, which both have their own deficiencies, namely limited spatial coverage or pure theoretical basis.
According to the obtained results, generally, the largest values of the asymmetry parameter, indicating the strongest forward scattering of radiation by atmospheric aerosols, are found over areas with aerosol load being dominated by large size particles of desert dust (tropical Atlantic, Arabian and Red Seas),.On the contrary, smaller gaer values are seen where a significant fraction of aerosol load comes from small size particles of anthropogenic origin, e.g. over the Black Sea.The results are consistent with the theory and thus prove a good performance of the MODIS retrieval of aerosol asymmetry parameter.Depending on the area of interest, the seasonal cycle of the asymmetry parameter varies markedly.More specifically, in areas with abundance of desert dust particles, the range of intra-annual variation is small, with the largest values during summer, while in other areas the seasonality is reversed, with the largest values during the cold season and the smallest during the warm season.
The asymmetry parameter decreases with wavelength, especially when one examines its spatially minimum values, while this decrease is weaker for the larger gaer values, corresponding to the presence of coarser particles.
The seasonal fluctuation is more pronounced with increasing wavelength in the examined regions, which is attributed to the different spectral behaviour of the asymmetry parameter for small and large particles.With respect to the inter-annual variability of the asymmetry parameter, we did not discern very important either increasing or decreasing tendencies, with absolute changes smaller than 0.04 in any case.On the other hand, we found opposing tendencies for the two satellite datasets.
MODIS-Terra observes mostly increasing tendencies, while Aqua gives also a few regions with decreasing tendencies.Generally, the largest intra-annual and interannual variations are seen over the Black Sea, while the smallest over the tropical Atlantic.However, some strong trends (especially from Terra) may be due to calibration drift errors, which may be addressed in collection 006.Along these lines we performed some preliminary comparisons between 051 and 006 Ångström Exponent trends from Aqua, which ensured that AE and gaer are very closely anticorrelated.These preliminary results, show that 051 Aqua AE trends resemble very closely the 006 trends, supporting that the gaer trends from collection 051 (at least for Aqua) reported in this study are credible.
The 051 MODIS gaer data is not a retrieved but a derived MODIS parameter.Given that the retrieval is strongly dependent on the assumptions made, namely on the aerosol modes used, uncertainties can be associated with its use in radiative transfer modeling.In order to examine these uncertainties, the gaer data were compared with Our results offer an interesting way to assess the uncertainty induced by the use of such satellite gaer data in climate and radiative transfer models that compute aerosol radiative and climate effects.Based on an overall assessment of satellite MODIS gaer through detailed comparisons against ground AERONET data, it appears that in overall MODIS performs satisfactorily in terms of magnitude of gaer values.This is indicated by the computed biases, which are smaller than 5% with respect to MODIS values, with better performance at smaller wavelengths.The root mean squared errors vary within the range 5-10% again being smaller for smaller wavelengths.These results indicate an uncertainty of MODIS gaer data over the study region up to of 10% at maximum.Previous analyses and sensitivity studies for the same study region (Papadimas et al., 2012) have shown that such gaer uncertainties can induce modifications of aerosol direct radiative effects (DREs) which are equal to 30% at the top-of-atmosphere (TOA) and 1% in the atmosphere and 10% at the surface, at maximum.Therefore, the uncertainty associated with the use of MODIS gaer is larger as to any aerosol related physical process taking place at TOA, namely planetary cooling or warming and its magnitude, smaller for processes at the Earth's surface, e.g.surface cooling and very small for aerosol processes and feedbacks in the atmosphere, like the aerosol semi-direct effect and its implications.Results from the same previous analysis (Papadimas et al., 2012)

Figure 5
Figure 5 displays the geographical distribution of the slope of inter-annual trend of gaer over the study region, as computed from the application of the Mann-Kendall test to time series of deseasonalized monthly anomalies of gaer at 470 nm.Results are calibration issues are expected to be addressed, at least partly, in the new Collection 006 products.Nevertheless, a preliminary comparison between MODIS Aqua C051 and C006 Ångström exponent (AE), which is another common aerosol parameter strongly dependent on size, using data for the 550-865 pair of wavelengths spanning the period 2002-2010, does not reveal significant modifications in geographical patterns of AE inter-annual changes.This puts some confidence on the C051 gaer results given in the present study.The results of this analysis are presented in detail in the next sub-section (3.2.3).
Figure7a, displays the geographical distribution of long-term average AE for the whole studyperiod, i.e. 2002-2010.In this figure, the northernmost areas are blank with respect to gaer (Fig.7a) because there are no data during winter and a long-term average would be biased.The main geographical patterns in Fig.7aare in line with those of asymmetry parameter (Fig.2).For example, note the high AE values in the Black Sea (between about 1.3 and 1.8, yellowish-reddish colors), indicative of fine aerosols, the relatively high values in the Mediterranean Sea (between about 0.7 and 1.2, greenish-yellowish colors) and the low values (0.1-0.4,deep bluish colors) off the western African coasts corresponding to exported Saharan dust.Over the same areas, gaer takes inverse low and high values, for example smaller than 0.65 over the Black Sea and larger than 0.7-0.75 off the western African coasts (Figs2ii-b and 2iii-b), indicating the predominance of fine and coarse aerosols respectively, in accordance with AE.The consistency between gaer and AE data is shown by the strong anti-correlation between the MODIS AE550-865 and gaer data at 660 and 860 nm, shown in Figures7b and 7c, respectively.It should be noted that correlation coefficients are computed from any available data pairs, i.e. available data for both gaer and AE550-865 at a given pixel and day.Note that there are no blank areas in Figs7b and 7c, in contrast to Fig.7a.There are both AE and gaer data for all seasons except winter and therefore, correlation coefficients can be calculated for these regions.Strong negative correlation coefficients, larger than 0.7 and 0.8 in Figs7b and 7c, respectively, relate inversely high gaer values with low AE ones and vice-versa, over the same areas.In both cases (Figs7b and 7c), the correlation is slightly higher over sea areas characterized by the presence of fine aerosols (e.g.Black Sea or Persian Gulf) and lower over seas undergoing frequent transport of coarse dust particles (e.g.southern Mediterranean Sea, Arabian Sea or Atlantic Ocean off the western African coasts).The overall computed correlation coefficient between gaer and AE is equal to -0.95 over the Black Sea, -0.89 over the Mediterranean Sea, -0.87 and -0.94 over the Arabian Sea and Persian Gulf, respectively and -0.89 off the western African coasts (values given for AE550-865 and gaer data at 860 nm).These results indicate that the spatial patterns of MODIS C051 gaer product are reasonable as compared to the C051 Ångström exponent data.This shows that the use of gaer in modeling studies can be considered as reasonably reliable with regards to the consideration of fine and coarse aerosols over the examined study area, with slightly more confidence over areas characterized by the presence of fine particles, such as the Black Sea or Persian Gulf.Since questions may also arise about possible uncertainties regarding the long-term variability of MODIS C051 aerosol size products, due to the calibration issues discussed in the previous section, the corresponding MODIS C006 AE product is displayed in Fig.8a.Figs.8a and 7a are similar in the main geographical patterns of the two collections' AE product.The similarity between C051 and C006 AE data is also depicted in the computed correlation coefficients (Fig.8b), exceeding 0.8, and biases (in absolute and relative percentage terms, Figs8c and 8d, respectively).For the Mediterranean Sea, the Arabian Sea and Persian Gulf, biases are smaller than 0.1 or 10% in most areas and 0.2 or 20% almost everywhere.Relative biases larger than 30% are only observed over the open Atlantic Ocean).The overall computed correlation coefficient for the entire study region is 0.88 (0.86, 0.89, 0.95 and 0.84 for Mediterranean, Arabian, Persian and Atlantic sea surfaces off the western African coasts).The corresponding overall relative percent bias is equal to 15.6% (9.1, 6.7, 6.1 and 15.7 for the same sub-areas as above).Our results indicate that the uncertainty related to the use of C051 AE data is small, especially over the Mediterranean Sea, the Arabian Sea, the Persian Gulf and the Atlantic Ocean areas not far from the European, African and Asian coastlines.Our AE results are in line with those ofLevy et al.    (2013, Fig. 15) which refer, however, only to year 2008 (ours are for 2002-2010).In addition, a comparison is attempted in Figs 8e and 8f between the computed trends of C051 and C006 AE data over the common period 2002-2010, in order to assess whether changes are detected, which could be an indication of possible changes in corresponding asymmetry parameter trends.Figures 8e and 8f show the computed deseasonalized trends of slope values for both C051 and C006 AE.The results reveal proved that the exact magnitude of MODIS gaer DRE uncertainty can be estimated by simple linear equations relating DREs and gaer, separately given for TOA, atmosphere and surface.The results of the present analysis are useful since they assess for the first time the performance of satellite based products of aerosol asymmetry parameter over broad regions of special climatic interest.The obtained results are relatively satisfactory given the difficulties encountered by satellite retrieval algorithms due to the different assumptions they made.Nevertheless, our results and identified weaknesses remind that users should be aware of the gaer uncertainties and their consequences.The identified weaknesses may provide an opportunity to improve such satellite retrievals of aerosol asymmetry parameter in forthcoming data products like those of MODIS C006.The increased temporal coverage of gaer data, combined with the continued operation of MODIS, is expected to make possible the building of the first real satellite climatology of this important aerosol optical property.

Figure 1 .
Figure 1.The study region (5°N-70°N, 25°W-60°E) and the location of 69 AERONET stations used for validation of MODIS satellite aerosol asymmetry parameter (gaer) data.Solid red circles denote stations located in Europe and hollow red circles are stations in Africa, Middle East and the Arabian peninsula.Also shown are seven sub-regions selected for studying the seasonal variation of gaer.

Figure 9 .Figure 10 .Figure 11 .
Figure 9. Scatterplot comparison between gaer values at 870 nm from MODIS Terra (black color) and Aqua (red color) and corresponding values from AERONET stations at 870 nm (blue squares, Fig. 1).The 95% prediction bands as well as the mean bias (AERONET minus MODIS) and root mean squared error are given.

Table 1 .
Correlation coefficients (R), mean bias, root mean squared error (RMSE) and the slope and intercept values of applied linear regression fits between MODIS and AERONET gaer data.The statistical parameters are given separately for the pairs