Mediterranean intense desert dust outbreaks and their vertical structure based on 1 remote sensing data 2 3

21 The main aim of the present study is to describe the vertical structure of the intense Mediterranean 22 dust outbreaks, based on the use of satellite and surface-based retrievals/measurements. Strong and 23 extreme desert dust (DD) episodes are identified at 1° x 1° spatial resolution, over the period Mar. 2000 24 – Feb. 2013, through the implementation of an updated objective and dynamic algorithm. According to 25 the algorithm, strong DD episodes occurring at a specific place correspond to cases in which the daily 26 aerosol optical depth at 550nm (AOD550nm) exceeds or equals the long-term mean AOD550nm (Mean) 27 plus two standard deviations (Std), but being smaller than Mean+4*Std. Extreme DD episodes 28 correspond to cases in which the daily AOD550nm value equals or exceeds Mean+4*Std. For the 29 identification of DD episodes, additional optical properties (Ångström exponent, fine fraction, effective 30 radius and Aerosol Index) derived by the MODIS-Terra & Aqua (also AOD retrievals), OMI-Aura and 31 EP-TOMS databases are used as inputs. According to the algorithm using MODIS-Terra data, over the 32 period Mar. 2000 – Feb. 2013, strong DD episodes occur more frequently (up to 9.9 episodes yr) over 33

for one year, while other studies focused on the vertical structure of dust outflows towards the Atlantic  MODIS, nevertheless its performance is better over sea (uncertainty equal to ± 0.03 ± 0.05 × AOD; 209 Remer et al., 2002) than over land (± 0.05 ± 0.15 × AOD; Levy et al., 2010). 210 The following daily MODIS-Terra and MODIS-Aqua Collection 051 (C051) level 3 satellite data 211 (MOD08_D3 and MYD08_D3 files) provided at 1° × 1° latitude-longitude spatial resolution are used: 212 (i) AOD550nm, (ii) Ångström exponent over land (α470−660nm), (iii) Ångström exponent over ocean 213 (α550−865nm), (iv) fine-mode fraction (FF) of AOD over land and ocean and (v) Effective radius over 214 ocean (reff). It must be mentioned that the size parameters (α, FF) over land are less reliable compared 215 to the corresponding ones over sea, since they are highly sensitive to spectral dependent factors such as 216 errors in the surface model or sensor calibration changes. Over sea, the accuracy of size parameters is 217 strongly dependent on wind conditions. 218 Similar data have been used by . However, in the present study we have 219 improved data quality by using the quality assurance-weighted (QA) level 3 data (http://modis-220 atmos.gsfc.nasa.gov/_docs/QA_Plan_2007_04_12.pdf) derived from the level 2 retrievals (10 km x 10 221 8 km spatial resolution). Each level 2 retrieval, is flagged with a bit value (from 0 to 3) corresponding to 222 confidence levels (No confidence: 0, Marginal: 1, Good: 2 and Very Good: 3). Based on this, the level 223 3 QA-weighted spatial means are obtained by the corresponding level 2 retrievals considering as 224 weight their confidence level (bit value). In addition, the day cloud fraction as well as the number of 225 level 2 counts, which are both relevant to the performance of the satellite algorithm, are also used in  scattering (e.g. sea-salt) and absorbing (e.g. desert dust, smoke) aerosols. The retrieval algorithm (fully 240 described by Torres et al., 1998; takes advantage of the low surface albedo in the UV 241 spectrum range, even in arid and semi-arid areas, making thus possible the estimation of the AOD over 242 highly reflecting desert surfaces, where the major dust sources are located. Since the late 70's, the 243 TOMS sensor onboard Nimbus-7 (1978Nimbus-7 ( -1993  providing almost daily global coverage thanks to its wide viewing swath (2600 km with 13 km x 24 km 249 nadir resolution). Apart from AI measurements, OMI aerosol products include also the total and 250 absorption AOD and the single scattering albedo at 388 and 500 nm (Torres et al., 2007). Both EP-251 TOMS and OMI-Aura retrievals are available via the Mirador ftp server (http://mirador.gsfc.nasa.gov/) 252 9 of the Goddard Earth Sciences Data and Information Services Center (GES DISC). OMI-Aura data, as 253 MODIS, are provided at 1° x 1° spatial resolution while the EP-TOMS retrievals have been regridded 254 from their raw spatial resolution (1° x 1.25°) in order to match the other two datasets (OMI, MODIS).    excluded. In this case, from the raw AOD retrievals we have masked out the "pure" desert dust grid 408 cells, which were identified based on the concurrent fulfillment of the defined criteria for dust 409 occurrence in the algorithm (for Ångström exponent, fine fraction, aerosol index and effective radius).

410
Then, from the remaining data (non-dust AOD retrievals), the mean, the associated standard deviation 411 14 as well as the defined thresholds of AOD are computed for the whole study period, for each pixel, as 412 also done in the primary methodology. Finally, also similarly to the way done in the primary 413 methodology, the DD episodes were classified into strong and extreme ones. The obtained results, i.e.

414
frequency of occurrence and intensity of DD episodes, based on the primary methodology and 415 METHOD-B are discussed in Section 4.2.

416
As explained, a similar methodology and data were used in the study by . The agreement between MODIS and AERONET also improves when the former AOD products are 503 more spatially homogeneous, i.e. when they are characterized by smaller AOD standard deviations at 504 the grid-level (from < 0.25 down to < 0.05, Fig. 2 ii-b). However, our results also indicate that apart 505 from increasing correlation coefficients (up to 0.7-0.8) with increasing level-2 counts and decreasing 506 standard deviations, the number of intense DD episodes is decreased dramatically (about 40-50 for 507 more than 50 counts and standard deviation smaller than 0.05).

508
In order to assess the performance of the satellite algorithm when operated with non-weighted generally are considered as inferior to surface-based similar products, which are taken as the reference.

582
In order to examine this degree of uncertainty and to verify the successful performance of the 583 algorithm, we also tested using it along with AERONET retrievals. This has been made for 7 584 Mediterranean AERONET stations, depicted with cyan circles in Figure 1, during the periods for which 585 ground retrievals are available ( Arenosillo, FORTH Crete). The intense DD episodes were identified following the methodology 589 described in section 3, but using only AOD at 870 nm, α440-870nm (lower/equal than 0.7) and reff (higher 590 than 0.6) as criteria, based upon their availability from AERONET. Subsequently, the algorithm was 591 also operated again using satellite (MODIS-Terra, OMI-Aura, EP-TOMS) input data for the periods 592 with available retrievals in each of the 7 AERONET stations.

593
In Figure 4, we present the overall scatterplots between satellite and ground AODs when intense parameter. The results given in Table 2 are satisfactory, since the percentages range from 87 to 99%, 613 and confirm the validity of the defined thresholds.

614
The scatterplots in Figs. 4 i-b and ii-b also reveal some weaknesses of the satellite-based algorithm.

615
More specifically, it is found that for few DD episodes identified by the satellite algorithm the

741
The analysis has been also repeated applying the alternative METHOD-B described in Section 3. as assumed by the VFM retrieval algorithm. In our study, more than 95% of the aerosol type records 776 were pure dust, for the collocated cases between the satellite algorithm and CALIPSO observations. In 777 addition, in the majority of the defined boxes, the percentage of dust from the overall observations is 778 higher than 70%, confirming furthermore the validity of the algorithm DD episodes' identification 779 procedure. This is an excellent proof of the successful identification of DD episodes by the satellite 780 algorithm, since CALIOP-CALIPSO is an independent and vertically resolved platform and database. dust layers were equal to 2.5±0.9 and 4.2±1.5 km, respectively.

840
As to the variation of vertical extension with longitude ( Fig. 8-i), it is revealed that the base height 841 of dust layers is decreased towards the eastern parts of the study region. In the western Mediterranean, 842 the mineral particles are mainly detected between 2 and 6 km while over the central and eastern 843 Mediterranean the corresponding altitudes are equal to 0.5 and 6 km, respectively. It is well known,   This is in accordance with our longitudinal projection ( Fig. 9 ii-b), where β532nm is high varying from 971 0.004 to 0.008 km -1 sr -1 at these altitude ranges.

972
In summer, the intensity of dust episodes is smoothly decreased at higher altitudes, where dust 973 layers of considerable β532nm values are also found. More specifically, the highest backscatter 974 coefficients (up to 0.008 km -1 sr -1 ) are recorded near to the surface but also moderate values (up to 975 0.006 km -1 sr -1 ) are observed between 2 and 5 km, particularly over the southern parts of the study 976 region ( Fig. 9 iii-b). Most of these intense DD episodes occur in the western Mediterranean, where the  Here, aiming at addressing how dust layers' geometrical characteristics influence the agreement 997 between columnar AOD satellite and ground PM10 measurements, specific desert dust outbreaks that   (Figures 10-12). Moreover, the corresponding aerosol subtype profiles, acquired from the CALIOP 1008 website (http://www-calipso.larc.nasa.gov/products/lidar/browse_images/production/), are provided in  ATBD-MOD-30 for Level-3 Global Gridded Atmosphere Products (08 D3, 08 E3, 08M3), online: 57