Absorption properties of Mediterranean aerosols obtained from multi-year ground-based and satellite remote sensing observations

Aerosol absorption properties are of high importance to assess aerosol impact on regional climate. This study presents an analysis of aerosol absorption products obtained over the Mediterranean Basin or land stations in the region from multi-year ground-based AERONET and satellite observations with a focus on the Absorbing Aerosol Optical Depth (AAOD), Single Scattering Albedo (SSA) and their spectral dependence. The AAOD and Absorption Angstr¨om Exponent (AAE) data set is composed of daily averaged AERONET level 2 data from a total of 22 Mediterranean stations having long time series, mainly under the inﬂuence of urban-industrial aerosols and/or soil dust. This data set covers the 17 yr period 1996–2012 with most data being from 2003–2011 10 ( ∼ 89 % of level-2 AAOD data). Since AERONET level-2 absorption products require a high aerosol load (AOD at 440 nm > 0.4), which is most often related to the presence of desert dust, we also consider level-1.5 SSA data, despite their higher uncertainty, and ﬁlter out data with an Angstr¨om exponent < 1.0 in order to study absorption by carbonaceous aerosols. The SSA data set includes both AERONET level-2 and satellite 15 level-3 products. Satellite-derived SSA data considered are monthly level 3 products mapped at the regional scale for the spring and summer seasons that exhibit the largest aerosol loads. The satellite SSA dataset includes the following products: (i) Multi-angle Imaging SpectroRadiometer (MISR) over 2000–2011, (ii) Ozone Monitoring Instrument (OMI) near-UV algorithm over 2004–2010, and (iii) MODerate resolution Imaging Spec- 20 troradiometer (MODIS) Deep-Blue algorithm over 2005–2011, derived only over land in dusty conditions. Sun-photometer observations show that values of AAOD at 440 nm vary between 0.024 ± 0.01 (resp. 0.040 ± 0.01) and 0.050 ± 0.01 (0.055 ± 0.01) for urban (dusty) sites. Analysis shows that the Mediterranean urban-industrial aerosols appear “moderately” absorbing with values of SSA close to ∼ 0.94–0.95 ± 0.04 (at 440 nm) 25 in most cases except over the large cities of Rome and Athens, where aerosol appears more absorbing (SSA ∼ 0.89–0.90 ± 0.04). The aerosol Absorption Angstr¨om Exponent (AAE, estimated using 440 and 870 nm) is found to be larger than 1 for most sites over 9269 the Mediterranean, a manifestation of mineral dust (iron) and/or brown carbon pro-ducing the observed absorption. AERONET level-2 sun-photometer data indicate the existence of a moderate East–West gradient, with higher values over the eastern basin (AAE East. = 1.39/AAE West. = 1.33) due to the inﬂuence of desert dust. The North–South AAE gradient is more pronounced, especially over the western basin. Our additional 5 analysis of AERONET level-1.5 data also shows that organic absorbing aerosols signiﬁcantly a ﬀ ect some Mediterranean sites. These results indicate that current climate models treating organics as nonabsorbing over the Mediterranean certainly underestimate the warming e ﬀ ect due to carbonaceous aerosols. A comparative analysis of the regional SSA variability has been attempted using satellite data. OMI and MODIS 10 data show an absorbing zone (SSA ∼ 0.90 at 470–500 nm) over Northeastern Africa that does not appear in the MISR retrievals. In contrast, MISR seems able to observe the East–West SSA gradient during summer, as also detected by AERONET. Also, the analysis of SSA provided by satellites indicates that the aerosol over

Optical Depth (AAOD), Single Scattering Albedo (SSA) and their spectral dependence. The AAOD and Absorption Angström Exponent (AAE) data set is composed of daily averaged AERONET level 2 data from a total of 22 Mediterranean stations having long time series, mainly under the influence of urban-industrial aerosols and/or soil dust. This data set covers the 17 yr period 1996-2012 with most data being from 2003-2011 10 (∼89 % of level-2 AAOD data). Since AERONET level-2 absorption products require a high aerosol load (AOD at 440 nm > 0.4), which is most often related to the presence of desert dust, we also consider level-1.5 SSA data, despite their higher uncertainty, and filter out data with an Angström exponent <1.0 in order to study absorption by carbonaceous aerosols. The SSA data set includes both AERONET level-2 and satellite 15 level-3 products. Satellite-derived SSA data considered are monthly level 3 products mapped at the regional scale for the spring and summer seasons that exhibit the largest aerosol loads. The satellite SSA dataset includes the following products: (i) Multi-angle Imaging SpectroRadiometer (MISR) over 2000-2011, (ii) Ozone Monitoring Instrument (OMI) near-UV algorithm over 2004-2010, and (iii) MODerate resolution Imaging Spec-20 troradiometer (MODIS) Deep-Blue algorithm over 2005-2011, derived only over land in dusty conditions. Sun-photometer observations show that values of AAOD at 440 nm vary between 0.024 ± 0.01 (resp. 0.040 ± 0.01) and 0.050 ± 0.01 (0.055 ± 0.01) for urban (dusty) sites. Analysis shows that the Mediterranean urban-industrial aerosols appear "moderately" absorbing with values of SSA close to ∼0.94-0.95 ± 0.04 (at 440 nm) 25 in most cases except over the large cities of Rome and Athens, where aerosol appears more absorbing (SSA ∼0.89-0.90 ± 0.04). The aerosol Absorption Angström Exponent (AAE, estimated using 440 and 870 nm) is found to be larger than 1 for most sites over 9269 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | the Mediterranean, a manifestation of mineral dust (iron) and/or brown carbon producing the observed absorption. AERONET level-2 sun-photometer data indicate the existence of a moderate East-West gradient, with higher values over the eastern basin (AAE East. = 1.39/AAE West. = 1.33) due to the influence of desert dust. The North-South AAE gradient is more pronounced, especially over the western basin. Our additional 5 analysis of AERONET level-1.5 data also shows that organic absorbing aerosols significantly affect some Mediterranean sites. These results indicate that current climate models treating organics as nonabsorbing over the Mediterranean certainly underestimate the warming effect due to carbonaceous aerosols. A comparative analysis of the regional SSA variability has been attempted using satellite data. OMI and MODIS

Introduction
Numerous studies have identified the Mediterranean basin as one of the most prominent "Hot-Spots" in projected climate change assessments (Giorgi, 2006, Giorgi andLionello, 2008). General Circulation Model (GCM) or Regional Climate Model (RCM) climate simulations have demonstrated that the Mediterranean is characterized by its 20 vulnerability to changes in the water cycle and predict a substantial precipitation decrease and warming, especially during the summer season. By the end of 21st century, the average prediction of the models suggests a significant loss of freshwater over the Mediterranean basin: −40 % for the period 2070-2090 compared to 1950-1999(Sanchez-Gomez et al., 2009. Climate simulations underline that the drying of region, which causes a northward shift of the mid-latitude storm track (Giorgi and Lionello, 2008).
Until now, most global and regional future climate simulations have only investigated the impact of global warming on the Mediterranean climate without clearly considering the influence of "Mediterranean aerosols" (pollution particles, smoke and mineral dust) that can significantly modify the radiation budget (Markowicz et al., 2002;Formenti et al., 2002b;Mallet et al., 2006;Roger et al., 2006). Specifically, atmospheric aerosols decrease the amount of shortwave (SW) radiation reaching the sea and continental surfaces. The column amount of atmospheric aerosol, quantified by the Aerosol Optical Depth (AOD), is one of the main factors causing this decrease. This aerosol-induced perturbation of the surface radiation budget can impact the Sea Surface Temperature (SST) (Foltz and McPhaden, 2008;Yue et al., 2011) and surface moisture exchanges by modifying latent heat fluxes (Ramanathan et al., 2001a).
In addition, due to their optical properties and especially their ability to absorb solar radiation, aerosols can trap SW radiation within the atmospheric layer where they re- 15 side. This additional absorption contributes to the direct heating of the atmosphere and causes changes in the atmospheric heating rate profiles, dynamical processes and more generally, the hydrological cycle (Solmon et al., 2008;Lau et al., 2009;Mallet et al., 2009). For example, Solmon et al. (2008) simulate that a change in dust absorbing properties could modify precipitation over Western Africa. Ramanathan et al. 20 (2001b) also conclude that heating caused by carbonaceous absorbing aerosols exported from India reduces the low cloud fraction over the Indian Ocean during the dry monsoon season. Therefore, a rigorous quantification of the effect of aerosols on the Mediterranean radiation budget and climate is required, including estimates of both the atmospheric aerosol load and its ability to absorb radiation. 25 Numerous studies of aerosol properties over the Mediterranean have documented AOD for pollution particles, smoke and dust aerosols using in-situ observations (Horvath et al., 2002;Formenti et al., 2002;Gerasopoulos et al., 2003;Kubilay et al., 2003;Meloni et al., 2004Meloni et al., , 2006Fotiadi et al., 2006;Pace et al., 2006;Roger et al., 2006; 9271 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Mallet et al., 2006;Tafuro et al., 2007;Saha et al., 2008). In brief, these studies report AOD values in the range 0.1-0.5, 0.3-1.8 and 0.3-0.8 in the spectral range of 440 to 550 nm for pollution, dust and smoke particles, respectively. In addition to analyses of local sun-photometer observations, several studies have used long time series of satellite-derived AOD at regional scales from Meteosat (Moulin et al., 1998), SeaW-5 iFs (Antoine and Nobileau, 2006), the MODerate resolution Imaging Spectroradiometer (MODIS) (Barnaba and Gobbi, 2001;Papadimas et al., 2008Papadimas et al., , 2009Nabat et al., 2012), the combination of MODIS andTOMS (Hatzianastassiou et al., 2009), MSG/SEVIRI (Lionello et al., 2012), or even all products (Nabat et al., 2012).
In contrast to the considerable scientific literature related to AOD distribution, aerosol 10 absorbing properties over the Mediterranean have received much less attention, and are poorly documented, in spite of their great importance for direct radiative forcing and overall regional climate. To the best of our knowledge, the only long term in-situ observations of absorbing aerosols available in the Mediterranean background atmosphere are those reported from Crete Island in the eastern Mediterranean by Sciare et al.

15
(2008), highlighting the major role of long-range transported biomass burning aerosols on Black Carbon (BC) concentration levels. Consequently, long-term observations of absorption in the atmospheric column, and specifically an analysis of the role played by dust aerosols, are still missing. In addition, aerosol single scattering albedo (SSA) products now available from several satellite missions (Hsu et al., 2004;Torres et al., 20 2007;Kahn et al., 2010) have not yet been analysed over the Mediterranean. This motivates the present work; its main objective is to characterise aerosol absorption over the Mediterranean region using available surface and satellites remote sensing data. We focus our study on (i) the aerosol absorption optical depth (AAOD), which is the fraction of AOD due to absorption only, (ii) the aerosol single scattering albedo (SSA), which 25 is the ratio of aerosol scattering to total extinction (i.e. scattering + absorption), and (iii) the spectral dependence of these optical parameters, calculated in terms of Angström Exponent (AE) or Absorption Angström Exponent (AAE) in the case of optical depth or 9272 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | absorption optical depth, respectively: AE = log(AOD λ1 /AOD λ2 )/ log(λ 2 /λ 1 ) and AAE = log(AAOD λ1 /AAOD λ2 )/ log(λ 2 /λ 1 ) (1) Subsequent to the data description (ground-based and satellite remote sensing data set used in our work) in Sect. 2, the results are presented and discussed in two main parts. First (Part 3.1-3.2), we report AAOD and AAE AERONET level-2 and 1.5 multi-

5
year observations for different Mediterranean sites. In a second time (Part 3.3), we exploit AERONET but also satellite SSA products derived at the regional scale from the Ozone Monitoring Instrument (OMI) (Torres et al., 2007), MODIS (Deep Blue algorithm;Hsu et al., 2004Hsu et al., , 2006 and the Multi-angle Imaging SpectroRadiometer (MISR) (Kahn et al., 2010). Our analysis of satellite data is focused on the summer (June-July-August, 10 JJA) and spring (March-April-May, MAM) periods, corresponding to the presence of the main Mediterranean aerosols (pollution particles, smoke and mineral dust; Barnaba et al., 2004) and highest AOD over the basin (Nabat et al., 2012).
2 Remote sensing ground-based and satellite observations 2.1 Surface AERONET observations 15 AERONET (Aerosol Robotic Network; http://aeronet.gsfc.nasa.gov/) is a federated network of ground-based sun-photometers and the associated data inversion and archive system, that routinely performs direct sun observations every 15 mn, and both almucantar and principal plane sky radiance measurements, and retrieves and distributes global aerosol columnar properties (Holben et al., 1998). Along with AOD observations, 20 the AERONET aerosol retrieval algorithm (Dubovik and King, 2000) delivers the complete set of column-effective aerosol microphysical parameters, including volume size distribution, refractive index at four wavelengths (440, 670, 870 and 1020 nm) and fraction of spherical particles (Dubovik et al., 2006, also see description in Dubovik et al., 2011. In addition, using these microphysical parameters, the algorithm provides other 25 9273 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | column-effective aerosol optical properties of interest to the scientific community, such as wavelength dependent SSA, phase function, and asymmetry parameter, as well as integral parameters of bi-modal particle size distributions (concentration, mode radii and variances) (Dubovik et al., 2002). In the present study, the analysis is mostly focused on AAOD and SSA for AERONET level 2.0, cloud-screened and quality-assured 5 AOD (Smirnov et al., 2000) and level 2.0 inversion products (Dubovik et al., 2002). The accuracy of AERONET retrievals is evaluated and discussed by Dubovik et al. (2000Dubovik et al. ( , 2002: the uncertainty of retrieved AAOD is estimated at the level of ±0.01 at wavelengths 440 nm and greater, whereas SSA uncertainty is estimated to be ±0.03 for AOD > 0.2 and ±0.07 for AOD < 0.2 (Dubovik et al., 2000(Dubovik et al., , 2002. 10 The 22 AERONET Mediterranean sites considered in our study are selected based on both their location (within or close to the basin) and their long term operation (> 2.5 yr). They are mapped in Fig. 1 and listed in Table 1 along with information regarding the location, period of observation and a brief description of each site. The significant number of AERONET sites available over the Mediterranean and their dis-15 tribution allows us to investigate aerosol characteristics under diverse conditions: over remote, anthropogenic-polluted and dusty locations. Figure 2 plots averaged level-2 AE and AOD of the different stations. The 4 stations of Thessaloniki, Erdemli, Nes Ziona and Modena show a high AOD range (0.28-0.32) whereas all other have an average < 0.23. The high AE (∼ 1.55) at Modena and Thessaloniki indicate urban pollution con-20 trolled by submicron particles. On the opposite, the low average AE value (< 1.0) at Nes Ziona reveals the major influence of large particles, likely soil dust and possibly sea salt. Lampedusa, Blida and Sede Boker also show a low AE (∼ 1), characterising a major influence of large particles in conditions of lower AOD (∼ 0.2). With intermediate AE values in the range 1.11-1.27 the stations of Malaga, Granada, Oristano, Forth 25 Crete and Erdemli correspond to relatively southern stations (Fig. 1) where the influence of desert dust from the south is still very significant. Larger AE values in the range 1.31-1.37 correspond to stations in a latitude band 38-40.6 • N (Burjassot, Potenza, Lecce, Messina and Athens) where the impact of dust is less visible but still signifi-cant. The rest of the stations has an AE in the range 1.41-1.49 and is composed of more northern stations in the western basin (Barcelona, Rome, Avignon, Ersa, Toulon, and Villefranche). As a result, a very significant trend is found between AE and station latitude, whereas AOD rather correlates with the station longitude (Fig. 3). Figure 4 provides the temporal distribution of the number of days per month with 5 AERONET level-2 absorption products available at each site considered. Numbers are most often < 10 and this number is rarely exceeded at other seasons than summer. The period covered spans from 1996 to 2012 but a single station of Sede Boker provides data before 2000. Figure 5 further plots some statistics. Twelve stations have at least 110 days of level-2 absorption products. The number ranges between 18 and 365 days 10 (at Ersa and Erdemli, resp.) and the maximum number of days is always found in the summer season which is known to be the maximum dust season in the western and central Mediterranean (Moulin et al., 1998). 88.8 % of data are from the period 2003-2011. As AERONET level-2 derivations correspond to a high aerosol load (AOD at 440 nm > 0.4) that is generally associated to dust events, we also consider level-1.5 15 inversion products in order to investigate the possible role of smoke aerosols on the absorption. We further limit the influence of dust events on the results by considering only level-1.5 data with an Angström Exponent > 1.0. In that case, uncertainties on AAOD are estimated at about ±0.01. 20 The MISR sensor aboard the NASA-Earth Observing System (EOS) Terra satellite provides estimates of AOD and SSA using top-of-atmosphere radiance measurements in four spectral channels (446, 558, 672 and 867 nm,  (Table 3). The key advantage of the MISR measurement approach is the use of different geometric views, making it possible to retrieve aerosol properties while accounting self-consistently for surface reflectance (Martonchik et al., 2009) and avoiding sunglint.

Satellite MISR observations
Here, we consider the monthly level-3 data "MIL3DAE" products version F15 0031 for the period 2000 to 2011 (Table 3). As mentioned below, we focus our analysis on the 5 spring (MAM) and summer (JJA) seasons, when the major aerosol loading occurs over the Mediterranean basin. It should be remembered that MISR SSA values are categorical rather than quantitative (two-to-four bins under good retrieval conditions -e.g. "absorbing" and "non-absorbing"), and the quality of the value depends on retrieval conditions, such as AOD (Kahn et al., 2010). As such, MISR-retrieved SSA is useful mainly 10 for making qualitative distinctions among adjacent air masses containing aerosols having different SSA values (e.g. Kahn et al., 2009). In that sense, the aim of this work is to propose a first analysis of the SSA regional patterns derived by MISR over the Mediterranean. We integrate the data separately over the marine pixels of the western and eastern (Adriatic included) basins (Fig. 6).

Satellite OMI observations
The Ozone Monitoring Instrument (OMI) operating since October 2004 onboard of the EOS Aura satellite is a spectrometer with high spectral resolution (Levelt et al., 2006). OMI has a swath width of 2600 km and offers nearly global daily coverage with a spatial resolution for the UV-2 and VIS (UV-1) channels ranging from 13 × 24(48) km 2 at nadir. 20 Here, we use data from the OMAERUV v003 product containing retrievals from the OMI near-UV algorithm (Torres et al., 2007). The near-UV method of aerosol characterization uses remote-sensing measurements in two near-UV channels to detect aerosol absorption. This algorithm derives a variety of aerosol radiative properties, such as an aerosol index (AI), AOD, AAOD (uncertainty of ±(0.05+30 %)) and SSA (uncertainty of 25 ±0.03) for clear-sky conditions. Here, we use SSA retrieved at 500 nm (Table 3) for the period 2004 to 2010 and we integrate over the same two Mediterranean sub-basins as with MISR (Fig. 6). In general, OMI near-UV aerosol AOD, AAOD and SSA retrievals 9276 are more reliable over land than over water surfaces. The near-UV retrieval method is particularly sensitive to carbonaceous and mineral dust aerosols. The main source of uncertainty in the application of the near-UV technique to OMI observations is sub-pixel cloud contamination due to the large size of the OMI footprint. OMI measurements in the near-UV also depend on the height of the aerosol layer above the ground (Tor-5 res et al., 1998). Other possible sources of error include surface albedo effects and assumptions about particle size distribution and refractive index.

Satellite MODIS (Deep Blue) observations
The MODIS sensor is a 36-channel spectrometer (0.412-14.2 µm) with daily global coverage. The MODIS retrieval algorithm (hereafter referred as the MODIS standard 10 algorithm) retrieves AOD over water and dark land surfaces, and fine-mode-fraction over water, at 10 km spatial resolution, aggregated to 1 • × 1 • for the global level 3 product. The algorithm has two components, one applied over ocean (Tanré et al., 1997) and the other, based on the dark target approach, over land (Kaufman et al., 1997).
In parallel with the aerosol retrievals classically conducted over the dark vegetation, 15 the Deep Blue aerosol algorithm was developed to infer aerosol properties over highly reflective surfaces, using the blue channels radiance measurements (Hsu et al., 2004). When the retrieval algorithm successfully identifies the aerosols in the scene as "Fine or Mixed Aerosols", the corresponding values of AOD andÅngström exponent (AOD spectral dependence) are reported. For "dust-dominant" cases, SSA is retrieved (un-

Aerosol absorption optical depth
AAOD (level 2) has been derived from AERONET observations at different wavelengths. We first analyse AAOD at 440 nm obtained at the 22 selected AERONET sites ( Fig. 1, Table 2). Over "pure" urban-industrialized regions (Table 2), AAOD falls between 5 0.027±0.01 and 0.05±0.01 with the maximum (0.047±0.010) observed at Rome (Table  2). In most cases, AAOD is around 0.03±0.01 for other urban-polluted sites. Long-term analysis of the AERONET records at the Modena and Rome sites, having around 10 yr of observations, does not reveal a clear AAOD tendency excepted over Rome, where a decrease is observed (−0.0005 yr −1 ) between 2001 and 2011 ( Fig. 7). For periods of 10 3-5 yr between 2005 and 2008, Toulon and Thessaloniki display a possible decrease in AAOD (not shown), though the period of observations is too short to derive an AAOD tendency with confidence. Mediterranean urban AAODs obtained at 440 nm are in the same magnitude range as those observed over other urban sites such as Creteil/Paris (∼0.015 ± 0.01) or Mex-15 ico (∼0.05 ± 0.01), but are generally higher than AAOD obtained at GSFC (AAOD < 0.010) or over other sites along the US Atlantic coast (AAOD ∼ 0.01 for the same wavelength) as reported by Russell et al. (2010). Compared to AAOD obtained during SAFARI (Southern Africa Regional Science Initiative) for smoke aerosols in southern Africa (AAOD ∼ 0.2 at 440 nm, see Russell et al., 2010) or in the Japan-Korea region 20 during ACE-Asia for mixed particles (AAOD of about 0.1 at 440 nm), AAOD derived over the Mediterranean display lower values.
AERONET observations (  (2010), who showed a negative trend in AOD of −0.022 per decade using the ten-year (2000-2009) Data Assimilation quality MISR/MODIS combined aerosol product over ocean. Other Mediterranean sites under the influence of dust aerosols display significant AAOD ∼ 0.050. It should also be noted that similar 5 values are found for AAOD observed over Puerto Rico during the PRIDE dust experiment (AAOD ∼ 0.05-0.06 at 440 nm) and over dusty sites in Bahrain/Persian Gulf, Solar Village/Saudi Arabia or Cape Verde (Russell et al., 2010). A linear correlation between AAOD and latitude ( Fig. 3) is found significant (at the 0.01 level).

Aerosol level-2 product
As reported by Russell et al. (2010), the spectral dependence of AAOD, as defined in Eq.
(1), can provide useful information on the contribution different aerosol types make to the shortwave absorption: It follows that an AAE of 1.0 corresponds to λ −1 dependence of absorption. Due to its relatively constant refractive index, the absorption spectrum of fossil fuel Black Carbon (BC) aerosols is expected to exhibit AAE of about 1.0 (Sun et al., 2007). Laboratory studies and field measurements taken in urban areas support this statement (e.g. Bond In addition to the chemical composition of aerosols, Gyawali et al. (2009) have shown recently that the AAE of BC cores (with diameter > 10 nm) that are coated by scattering shells may deviate from the typically assumed AAE of 1 relationship. Lack and Cappa (2010) have shown that BC cores coated by scattering shells can produce AAE values 10 up to 1.6. This finding clearly complicates the attribution of observed AAE larger than 1 to BrC vs. mixing state. It should be also remembered that AAE is highly dependent of the wavelengths used in the calculation (Russel et al., 2010, Bahadur et al., 2012. The mean and associated standard deviations of level 2 AAE obtained over AERONET sites are reported in Table 2 (AAE presented in this study is always calcu-15 lated between 440 and 880 nm). Our results indicate that average AAE values obtained over the Mediterranean fall between 0.99 and 2.16. For the urban polluted sites (Lecce, Toulon, Rome, Thessaloniki, Messina, Barcelona, Burjassot and Athens), AAE is larger than 1, with a maximum of 1.5 in Rome (see Table 2). Such AAE values obtained over urban Mediterranean sites are similar to those obtained for other urban sites by Russell . We note that AAE obtained at urban sites in the Eastern Mediterranean basin are generally higher than those reported by Russell et al. (2010).
As shown in Table 2  One important result concerns the AAE regional gradient which appears over the Mediterranean. Analysis of AERONET level 2 data (excluding AERONET sites affected by the presence of dust, such as Blida and Lampedusa) display lower mean AAE values over the Western basin (Table 2 and Fig. 8b). Calculated AAE over Spain (including Burjassot, Granada and Barcelona) and France (Toulon, Villefranche sur Mer and Ersa) 15 are between 0.99 and 1.22 (Table 2) and clearly increase over Italy, Greece (Thessaloniki and Athens) and Crete Island, with AAE values comprised between 1.30 and 1.70. Average AAE for Eastern and Western AERONET sites are about 1.39 and 1.33, respectively (Fig. 8b). This AAE regional gradient indicates that mineral dust and/or organic absorbing particles might make a larger contribution to the total aerosol load 20 over the Eastern Mediterranean. As the AERONET level-2 product retains only highquality cases having AOD higher than 0.4 at 440 nm for estimating AAOD, this gradient supports a larger influence of mineral dust over the Eastern region. Figure 9 plots the particle size distribution retrieved from level-2 AERONET sky measurements (Dubovik et al., 2000) and averaged over the eastern and western stations. Larger concentra-25 tions are generally found in the eastern basin but the difference appears only significant for the coarse mode. If we fit those distributions by a bimodal lognormal distribution, the coarse mode is found to have a slightly larger relative contribution in the eastern basin (Table 4). In addition, Fig. 10 reports AAE for northern and southern AERONET sites 9281 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | classified for the western (Fig. 10a) and eastern (Fig. 10b) basins. In this case, the regional latitudinal gradient is clearly more pronounced over the western basin, where the mean AAE is about 1.77 and 1.22 for the northern and southern AERONET sites, respectively, due to the desert dust influence. This gradient is also well observed over the eastern basin with a moderate difference in AAE between the southern (mean AAE 5 of 1.47) and northern (mean AAE of 1.34) AERONET sites, respectively.

Aerosol level 1.5 product
As mentioned below, AERONET data quality criteria do not support the study of the brown carbon influence on absorbing properties in most cases, as smoke plumes are generally characterized by moderate AOD (0.2 < AOD < 0.5). However and as already 10 mentioned, smoke aerosols that contain high concentrations of organics are characterized by large AAE (for SAFARI African biomass smoke, AAE ∼ 1.45, calculated between 325 and 1000 nm; Russell et al., 2010). In order to investigate the possible role of smoke aerosols, we have conducted complementary analyses using AERONET level-1.5 retrievals with data screened for Angström Exponent > 1.0 to avoid mineral inating from the region surrounding the Black Sea and transported over Crete (Sciare et al., 2003(Sciare et al., , 2008. This influence of wood burning shows a maximum during summertime, consistent with the maxima in AAOD observed in that season over Moldova and Northern Greece, which are on the path of air masses originating from the Black Sea region and transported over the Eastern Mediterranean. Finally, over Barcelona,even 5 if the mean value is near unity, indicating that BC aerosol is the main contributor to absorption, a large range of AAE is also observed, with maxima around 1.6. Further analysis is needed to link the calculated AAE with the chemical composition of aerosols over different AERONET sites when such data are available. To conclude, such additional analysis of AERONET level 1.5 data reveal that, in addition to dust particles, 10 organic aerosols also contribute to shortwave absorption over the Mediterranean. This result demonstrates that current regional climate models treating OC as nonabsorbing over Mediterranean underestimate the total warming effect of carbonaceous particles by neglecting part of atmospheric heating.

AERONET level 2 observations
As mentioned previously, SSA (level-2) has been determined from AERONET inversions at different wavelengths. Long-term SSA observations (more than 5 yr, i.e. Modena and Rome) display average daily values between 0.70 ± 0.04 and 0.97 ± 0.04 (440 nm) with no significant trends. Over urban sites, Table 2 indicates that in most For "dusty" sites, average SSA is in the range of 0.90-0.92 ± 0.04 at 440 nm (see Ta- 20 ble 2) indicating that dust particles are moderately absorbing over the Mediterranean. For this aerosol type, comparisons between AERONET SSA and in-situ observations are more limited, as most of estimates were obtained with remote-sensing techniques (Di Biagio et al., 2009;Meloni et al., 2004Meloni et al., , 2008. In contrast to such "moderate" dust SSAs, several studies reported larger absorbing efficiencies over the Mediterranean. ing at 2000 and 4000 m coming from Africa. In these cases, the presence of polluted absorbing particles at lower atmospheric levels explains the low SSA values. More recently, in the case of dust aerosols transported over Barcelona, Sicard et al. (2012) indicate surprisingly low dust SSA (∼ 0.70 at 440 nm) due to the mixing of dust with smoke-polluted particles.

SSA spectral dependence
The SSA wavelength dependence obtained for urban and dusty AERONET sites is reported in Fig. 12a. We clearly note two opposite behaviours, an increase and a decrease of SSA with wavelengths, associated with dust and urban aerosol, respectively, and similar to that reported by Russell et al. (2010). Here, the SSA spectra for AERONET locations dominated by desert dust increase from ∼ 0.90−0.92 ± 0.04 (at 440 nm) to ∼ 0.95−0.96 ± 0.04 (at 1020 nm). In contrast, SSA is clearly decreasing for locations dominated by fine pollution aerosols, from ∼ 0.92−0.93 ± 0.04 (at 440 nm) to ∼ 0.91−0.92 ± 0.04 (at 1020 nm). Such a decrease seems less marked than those reported by Russell et al. (2010) for other anthropogenic sites. Figure 12b presents the average SSA spectra obtained for the western and eastern Mediterranean AERONET sites. The mean value calculated at 440 nm for the eastern Mediterranean sites is slightly lower (∼0.92 ± 0.04) than that over the western part (∼0.94 ± 0.04). This result is consistent with AAE observations, suggesting a more pronounced contribution of dust particles over the Eastern basin. As shown in Fig. 12b, 20 this finding is based only on the shortest wavelengths provided by AERONET (440 nm). Indeed, the difference in SSA becomes negligible for longer wavelengths (670, 870 and 1020 nm, Fig. 12b). In that sense, new AERONET photometer retrievals at 340 nm should be very helpful, but unfortunately are not yet available. Also, as shown in Fig. 12b, the SSA spectra for western AERONET sites decrease with increasing wave-25 lengths, from 0.94 ± 0.04 (440 nm) to 0.92 ± 0.04 (1020 nm), unlike the Eastern sites, where SSA increases between 440 and 670 nm, which is a signature of dust particles (Dubovik et al., 2002). 9285 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Concerning the OMI sensor, we observe a clear North-South gradient with lower values over the Southern Mediterranean and higher SSA (∼ 1 at 550 nm) over a large part of the European continent. Note that a region of lower SSA (∼ 0.85-0.90 at 550 nm) occurs, especially over Morocco and the Algerian coast. This result is consistent with recent findings presented by Valenzuela et al. (2012), who showed that particles have marked absorbing properties during dust events, corresponding to air masses transported from North Morocco and northwest Algeria. This "absorbing" zone ( Fig. 13) is also identified by MODIS DB, but with a larger regional extent compared to OMI. Indeed, MODIS DB indicates SSA between 0.90 and 0.95 over Northern Africa, South 15 of Turkey and Spain. Compared to OMI and MISR, Northern Africa appears "more absorbing" based on the MODIS DB product, but it should be remembered that the retrieval wavelengths are different among the three sensors (especially between MODIS and MISR), which could explain in part the lower values obtained from MODIS DB (∆SSA is around 0.01 between 470 and 550 nm, Dubovik et al., 2002). 20 As mentioned below, when using MODIS DB data, a large part of Algeria (25-30 • N, 0-10 • E), Tunisia and western Libya are characterized by SSA ∼ 0.90. This regional pattern is consistent with some identified dust sources reported by Laurent et al. Concerning MISR, we observe near unity SSAs (555 nm) during JJA over a large part of the African continent with some lower values (∼ 0.95) derived over Morocco, North Algeria and Tunisia, but the contrast is less marked than for the OMI and MODIS observations. Over Africa, MISR SSA is clearly higher (+0.1) than OMI. As reported by Kahn et al. (2010), MISR tends to obtain SSAs at or near unity, especially when 5 the AOD is too low to produce good SSA constraints. Kahn et al. (2010) indicate that information about particle properties in the MISR data decreases significantly when the total column visible AOD is below 0.15 or 0.2 (at 550 nm), and in such cases, should not be used for quantitative analysis according to the MISR Data Quality Statement distributed with the aerosol products. Aerosol type information content also diminishes 10 as AOD decreases below certain values for all remote sensing techniques (this effect is not as well documented for MODIS and OMI). Over the Mediterranean Sea, MISR derives lower SSA (∼ 0.92-0.96) compared to land surfaces, associated with a northsouth gradient. One can clearly note specific regions with low SSA (∼ 0.85-0.90) over the Aegean, Adriatic and Black Seas, possibly due to the presence of anthropogenic 15 absorbing particles.

SSA satellite observations
For the spring period (Fig. 14), Southern Europe (Italy, Greece, Turkey and Spain) appears to be less absorbing than during summer, with ∆SSA (spring minus summer SSA) of +0.025. The increase could be due to a lower concentration of polluted absorbing aerosols during spring compared to summer. In contrast, a large part of Northern 20 Africa appears as more absorbing (−0.01 < ∆SSA < −0.02) during spring compared to summer, with a maximum over the eastern part (Libya). This may due to higher emissions of dust during spring. This result is consistent with MODIS DB data, which shows a decrease of about −0.02 of SSA over Northern Africa during spring. MISR produces different results over Southern continental Europe, with higher SSA during summer 25 compared to spring (∆SSA ∼ +0.01). Similar increases in SSA during summer are also detected over the Mediterranean coastal zones of Algeria and Morocco. Over the Sea (especially for the Oriental basin), we obtain consistent results, with lower SSA during 9287 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | the summer, certainly due to the accumulation of absorbing polluted particles at this time.
Finally, at regional scales, MISR displays a nearly constant SSA (∼ 0.97-0.98) over the Mediterranean Sea during summer and spring as well as OMI (Fig. 15). Unlike OMI, the SSA geographical gradient observed using AERONET data is better captured 5 by MISR, amounting to lower SSA over the eastern basin (∼ 0.96-0.97) compared to the western one (∼ 0.98-0.99), especially during summer, when polluted particles are present. The uncertainty associated with MISR SSA (see Sect. 2.2 above) retrievals does not allow distinguishing different absorbing properties for the two basins during spring (Fig. 15). Due to the uncertainty associated with the OMI SSA product, no con-10 clusion about the differences in absorbing properties between the Eastern and Western basins can be drawn from OMI data (Fig. 15).

Conclusions
A multi-year climatology of column-effective aerosol absorption properties obtained over the Mediterranean from AERONET ground-based and satellite (MISR, OMI, 15 MODIS Deep Blue) remote sensing observations is presented. The focus of this study was on characterizing Aerosol Absorption Optical Depth (AAOD) and Single Scattering Albedo (SSA), and their spectral dependence. The AAOD data set is composed of daily averaged AERONET level 2 data from 22 stations mainly under the influence of urbanindustrial and/or soil dust aerosols. The data sets span 1996-2012, but most data are 20 from the 2003-2011 period. The SSA data set includes both AERONET and satellite products. Since AERONET level 2 absorption products are limited to high aerosol load (AOD at 440 nm > 0.4), which are most often related to the presence of desert dust, we also considered level 1.5 AAOD data despite their higher uncertainty. Satellite-derived SSA data are monthly level 3 products mapped at the regional scale, and we focus 25 on the spring and summer seasons of maximum aerosol load in the Mediterranean. Satellite products include (i) the 0.5 • × 0.5 • product from Terra/MISR over 2000-2011, (ii) the 48 km × 48 km product from OMI near-UV algorithm over 2004-2010, and (iii) the 1 • × 1 • product from Terra/MODIS Deep-Blue algorithm over 2005-2011, derived only over land in dusty conditions. AAOD values obtained from sun-photometer observations over the Mediterranean at 440 nm ranged from 0.024 ± 0.010 to 0.050 ± 0.010 for urban sites (with maxima 5 observed over Rome), whereas for dusty sites, AAOD varied from 0.040 ± 0.010 to 0.055±0.010. The analysis of the corresponding SSA values showed that aerosol over Mediterranean urban-industrial locations appeared "moderately" absorbing, with SSA at 440 nm close to ∼ 0.94−0.95 ± 0.04, although in some locations the aerosol was as absorbing as heavily polluted sites such as Mexico City. 10 The spectral dependence of absorbing properties was also studied, using the Absorbing Angström Exponent (AAE) estimated between 440 and 870 nm. For most Mediterranean sites, AAE is larger than 1, indicating strong shortwave absorption that can be associated with the presence of Brown Carbon (BrC) and/or mineral dust (having high iron content) aerosols. Sun-photometer level-2 data analysis indicates 15 a moderate AAE regional gradient, with higher values obtained over the eastern basin (AAE East. = 1.39/AAE West. = 1.33), mainly due to the presence of mineral dust particles. In parallel, the North-South AAE gradient is more pronounced, especially over the western basin, with AAE North and AAE South of about 1.22 and 1.77, respectively.
The complementary analysis using level 1.5 AERONET retrievals with data screened 20 to avoid mineral dust aerosols (i.e., AOD > 0.2 and Angström Exponent > 1.0) show that some Mediterranean sites are affected by organic absorbing aerosols (mean AAE ∼ 1.15), especially Rome, Lecce, Burjassot and Athens. The effect is found to be lower for the Barcelona and Forth Crete AERONET sites (mean AAE of about 1.0). This result highlights that current regional climate models that treat OC as purely scattering over 25 the Mediterranean underestimate the total warming effect of carbonaceous aerosols and neglect part of the atmospheric heating due to particles. In addition, we explored the possibility of using satellite products for characterising the spatial distribution of aerosol absorption over the Mediterranean. Our results sug-9289 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | gest that columnar values of aerosol SSA obtained from OMI, MISR and MODIS DB reveal significant differences. OMI and MODIS detect an absorbing zone over Northeast Africa with SSA ∼ 0.90 (at 470-500 nm), which is not detected by MISR. In a contrast, MISR seems able to capture the East-West SSA gradient during summer, as observed from AERONET; however the satellite retrievals provide only qualitative constraints on 5 SSA. In addition, the analysis of satellite SSA indicates that the Mediterranean aerosol appears more absorbing during summer compared to spring, which could be due to a larger proportion of sunlight-absorbing particles, such as desert dust and smoke.
Acknowledgements. This work was undertaken is the framework of the MISTRALS/ChArMEx programme. We acknowledge the AERONET and PHOTONS sun-photometer networks and 10 the PIs of the 22 selected stations and their staff for their work to produce the dataset used in this study. We acknowledge the OMI PIs for providing satellite observations.

The publication of this article is financed by CNRS-INSU.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Barnaba, F. and Gobbi, G. P.: Aerosol seasonal variability over the Mediterranean region and relative impact of maritime, continental and Saharan dust particles over the basin from MODIS data in the year 2001, Atmos. Chem. Phys., 4, 2367-2391, doi:10.5194/acp-4-2367-2004, 2004. Bond, T. C. and Bergstrom, R. W.: Light absorption by carbonaceous particles: an investigative 5 review, Aerosol Sci. Technol., 40, 27-47, 2006. Chazette, P. and Liousse, C.: A case study of optical and chemical ground apportionment for urban aerosols in Thessaloniki, Atmos. Environ., 35, 2497-2506, 2001surements of Mediterranean aerosol radiative forcing and influence of the single scattering  Impact of dust aerosols on the radiative budget, surface heat fluxes, heating rate profiles and convective activity over West Africa during March 2006, Atmos. Chem. Phys., 9, 7143-7160, doi:10.5194/acp-9-7143-2009ing mediterranean aerosols lead to a large reduction in the solar radiation at the surface, Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Sanchez-Gomez, E., Somot, S., and Mariotti, A.: Future changes in the Mediterranean water budget projected by an ensemble of regional climate models, Geophys. Res. Lett., 36, L21401, doi:10.1029/2009GL040120, 2009 Characterization of carbonaceous aerosols during the MINOS campaign in Crete,

9305
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |   Particle Radius (µm) Fig. 9. AERONET level-2 aerosol volume size distribution averaged over the 8 eastern and 14 western sites (see Table 1). Inversion is performed in 22 logarithmically equidistant size bins between 0.05 and 15 µm in radius. Adjusted parameters are given in Table 4.

9311
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |