The aerosol extinction coefficient is not retrieved in case of clouds below the reference level. Indeed, clouds will drastically decrease the signal-to-noise ratio or even prevent the laser beam to reach clear air. The retrieval of aerosol extinction coefficient requires avoiding any contamination of the profiles by low clouds. Low clouds can be detected in the lidar profiles because they are associated with a sharp increase in the return signal that is highly variable from one shot to the other. The cloudy profiles are automatically detected using a simple procedure commonly used in satellite image processing . We select a set of 3 consecutive profiles over a period of 24 h. The profiles are standardized. For each altitude and each set of 3 consecutive hourly profiles we compute the 3×3 neighborhood mean and standard deviation. Cloudy structures have an abnormal mean and standard deviation. We retain a threshold of 1.0 for the mean and 0.5 for the standard deviation, i.e. above such thresholds the profile is flagged as cloudy. Figure  is an example of cloud masking obtained between 24 and 27 May 2012. Gray areas correspond to the cloud mask. During this period, all the clouds between 2 and 6 km are detected (Fig. a) and the corresponding contaminated profiles are automatically discarded (Fig. b).

Such a procedure removes 52 % of the total number of the 14 370 hourly acquired profiles. On a monthly basis, the screening ranges between a minimum of 26 % of cloudy data in August 2012 to a maximum of 72 % in March 2013.

The aerosol extinction coefficient is retrieved from the range-corrected attenuated backscattering signal following the Klett's method and considering both molecules and aerosols. The calibration of the lidar profiles is given as the initial condition used for solving the differential form of the lidar equation for the range corrected signal S(R). The solution for the aerosol backscattering coefficient βa is βa(R)+βm(R)=S(R)exp⁡-2∫R0R[La(r)-Lm]βm(r)drS*(R0)-2∫R0RLa(r)S(r)T(r,R0)dr where βm is molecular backscattering coefficient, and T(r,R0)=exp⁡-2∫R0r[La(r′)-Lm(r′)]βm(r′)dr′. where La and Lm are the particulate and molecular lidar ratio, respectively. The extinction coefficient is given by αa(R)=βa(R)⋅La(R)

The initial condition S*(R0) (Eq. ) is taken in the far range R0 and above 7 km height where the aerosol contribution is negligible compared to the molecular one. S*(R0)=S(R0)βa(R0)+βm(R0)

The choice of an adequate aerosol lidar ratio (ratio of aerosol extinction to backscattering coefficient) is a key issue in the analysis of single wavelength backscattering lidar data. This parameter may vary between 27 Sr for maritime aerosol and 71 Sr for urban/industrial pollution at 550 nm . The lidar ratio is a function of the altitude as it depends on the vertical stratification of the atmosphere, and on the aerosol optical properties transported in the different layers, their hygroscopicity, and the profile of relative humidity. In case of a single wavelength backscattering lidar, we can't derive any information on the vertical variability of this parameter so the lidar ratio is first assumed to be constant as a function of the altitude. The AOD is used as a constrain for the lidar ratio selection in the extinction retrieval . Different values of La(R) are tested until the vertical integration of αa(R) (Eq. ) is equal to the sun photomoter AOD. When no AOD value is available (nighttime or clouds) we use the same La(R) as the previous retrieval. It means that during nighttime, the extinction profiles are retrieved using the La(R) given by the lastest inversion on the preceding day. Post-processing of daily mean AOD indicates that AOD above 1.2 corresponds to remaining cloud contamination and so the lidar extinction coefficient profiles are removed from the data set.

The fix lidar ratio retrieved when constraining the retrieval of the hourly extinction profiles with the satellite AOD ranges between 22 and 108 Sr with an average value of 54±23 Sr. This estimation is made over the 1836 h of simultaneous satellite and lidar observations during daytime. There is a slight increase in the lidar ratio as a function of the AOD. Considering AOD below 0.05 the average lidar ratio is 42 Sr while it is 62 Sr in high aerosol condition, i.e. when AOD is above 0.5.

Figure  presents the comparison between sun photometer, lidar and satellite AOD at 355 nm on a monthly basis. There is an excellent consistency for the months were all the measurements are available. A seasonal cycle is clearly observed with minima in winter and maxima in spring–summer. Local maxima are observed in July 2012 (0.4), July 2013 (0.44) and in March 2012 (0.28). However no clear maximum is observed in spring 2013.

The monthly mean aerosol extinction profiles (Fig. ) shows a weak variability. The extinction coefficient tends to increase above 4 km from April to June and in September which probably corresponds to the overpass of dust events. The extinction coefficient below 2 km is maximum from June to September which correponds in majority to the period of polluted photochemical or smoke aerosol events.

A TSI 3-wavelength integrating nephelometer was operated at the Ersa station (43.0∘ N, 9.35∘ E, 530 ma.s.l.) since May 2012. The nephelometer is located 37 km away from the lidar on a North–South axis. The pseudo-total backscattering (integration of the scattering function for scattering angles between 7and 170∘) is measured at 450, 550 and 700 nm. The error due to the truncation in the phase function is corrected following . The aerosol scattering coefficient is extrapolated to 355 nm from the measurements at 450 and 550 nm following the Ångström law.

Figure  shows the comparison between the daily mean total aerosol scattering coefficient derived from the integrating nephelometer and the mean aerosol extinction coefficient between 450 and 550 ma.s.l. derived from the lidar. The extinction coefficient is the sum of the scattering and absorption coefficient and so it has to be higher or egal to the scattering coefficient. Since the measurement sites are not collocated it is not possible to estimate the single scattering albedo, that is the ratio between scattering and extinction coefficient. However we have a similar trend showing low scattering, respectively extinction coefficient during winter period and a maximum value in July 2013. When considering the coincident data the correlation coefficient is 0.63 (N=302 data).

The variation in the aerosol concentration and optical properties between the different layer induces gradients in the attenuated range corrected lidar signal. In the lower part of the atmosphere where the signal to noise ratio is the higher, we classify at least two layers using the second derivative of the lidar signal . In the upper part of the atmosphere where the gradients and the signal to noise ratio are smaller the derivative of the signal doesn't lead a clear detection of the dust layer. As an indicator of the vertical development of the aerosol layer, we compute the AOD scale height, i.e. the altitude at which 1/e of the total AOD is below that point .

We detect a shallow marine boundary layer on most of the profiles. The average altitude is 380±260m. A weak seasonal cycle is observed with a summer maximum at 460 m in July and a winter minimum at 250 m in December. The daytime variability is also lower during winter months (November to March). We also detect a secondary maximum in the derivative of the lidar signal located between 700 m and 3.5 km and on average at 1.3±0.5km. This secondary gradient in the lidar signal can be due to the complex local atmospheric circulation because of the proximity of both the sea shore and the high topography of the island. The mean scale height is 4.1±1.6km. The scale height reaches 5 km during the dust outbreak in March 2013. Below the scale height the mean hourly aerosol extinction coefficient is 0.07±0.17km-1. It is 0.05±0.08km-1 in the marine boundary layer.

The two main phenomena leading to strong AODs over Corsica are dust outbreaks from Northern Africa and advection of pollution or biomass burning from Europe. The AOD at 550 nm and the Ångström coefficient are used to classify 3 types of aerosol conditions: maritime, dust and pollution aerosols. Maritime situations correspond to AOD below 0.3 whatever values of the Ångström coefficient. Pollution aerosol conditions are identified with AOD above 0.3 and Ångström coefficient above 1.2 while dust cases correspond to AOD above 0.3 and Ångström coefficient below 0.9 . From February 2012 to August 2013, only 28 out of 341 days of observation have an AOD above 0.3, i.e. 8 % of the observations. Among those 28 days, 43 % (12 days) are identified as dust and 32 % (9 days) as pollution events.

In Table  we give the daily average aerosol properties according to the previous classification. Low AOD denomination corresponds to maritime situation for which the AOD at 550 nm is below 0.1. The average AOD at 550 nm is higher for the dust than for the pollution cases. However because of the high Ångström exponent for polluted fine particles, it is the opposite at 355 nm. Considering cases with a low AOD, the Ångström exponent is 0.52 indicating a main contribution of large sea salt particles. The lidar ratio at 355 nm for those cases is 49 Sr. The effective lidar ratio found for dust is 63 (±18) Sr. This value is higher than the one retrieved by but in the range of previously published values in the Mediterranean . However, in our case the lidar ratio is not vertically resolved and thus is biased by the contribution of marine and pollution aerosols in the marine boundary layer. In the case of the long range transport of pollution aerosols which occurs at a lower altitude as it is shown by the lower scale height, the lidar ratio of 69 Sr is more reliable and in a better agreement with previously published values . The high scale height for low AOD is due to the weak contribution of the clean boundary layer to the total AOD. The difference in the altitude of the transport for dust and pollution aerosols is highlighted by a higher scale height for dust than for pollution particles. In the next section, we focus on this difference by analyzing selected case studies.

Figure  shows the daily mean AOD over the western Mediterranean basin during the dust events which affected Cap Corse during 2012–2013 period. The major dust event in terms of AOD occurred in April–May 2013 (Fig. d) with a paroxysm on 30 April when the AOD reached 0.89 (daily mean at 355 nm). SEVIRI Ångström exponent is 0.27 indicating a large contribution of coarse particles to the AOD. Such corridor is a well known feature of the dust transport in the Mediterranean and forms when low-pressure systems enter the northwestern part of the basin in spring or autumn . The dust storm is associated with clouds that prevent from retrieving AOD over large parts of the plume and lead to a patchy structure when averaging during the whole dust event between 30 April to 4 May 2012. The second dust event in magnitude observed over the site occurred in June–July 2012 (Fig. b) with a SEVIRI AOD at 0.8 on 30 June 2012 . In this case the advection pattern is different. The saharan air layer advected over north-east Atlantic and western Spain during the dust event is recirculated to the northwestern part of the basin by westerly winds. A larger SEVIRI Ångström exponent of 0.6 is observed for this dust outbreak which can be linked to a smaller contribution of large particles to the AOD than in the case of spring event. A similar pattern is observed for the 19 June 2012 (Fig. a). Finally, the Fig. d shows the case of a dust storm in the western basin on July 2012 where the core of the dust plume doesn't reach the Cap Corse and the gulf of Genova so the AOD remains rather weak in this area.

The highest pollution event observed occurred on July 2013. Figure d shows the mean daily SEVIRI AOD between 13–20 July 2013. The SEVIRI AOD reached 0.53 (daily mean) on 14 July and increased to 0.6 on 15 July. The Ångström exponent is 1.33 so leading to an AOD of 0.95 at 355 nm on 14 July 2013. During this period, the pollution plume is observed all along the northern coast of the basin including the Balearic islands and Corsica (Fig. d) while a local maximum at 0.4 is observed in the gulf of Genova. A rather similar case was observed on 6 September 2012 (Fig. c). However the plume is limited to the gulf of Genova. The geographic position of the low pressure system in the northern basin affects the extent of the plume. In this case the low is located in the gulf of Genova, while in July 2013 it is located westward over Spain. The second major pollution event in term of AOD occurs on 26 March 2012. SEVIRI observations indicate a large plume crossing the northern part of the western basin (Fig. a). The corresponding daily mean AOD at 355 nm is 0.75 and the Ångström exponent is 1.25. This moderate value tends to indicate a possible contribution of large particles. In this case the low is southward over Sicily bringing more continental air from central Europe to the basin. The case of August 2012 (Fig. b) is ambiguous because despite an Ångström exponent of 1.3 the satellite pictures indicates a possible contribution of dust particles from northern Africa.

We further analyze the stratification of the aerosol layers and their possible origin and transport pathway. The air mass origins for the aforementioned specific events are analyzed using the Lagrangian particle dispersion model FLEXPART . The model is run in backward mode over 10 days. The meteorological forcing is given by ECMWF operational analysis at 0.5∘. The plume is computed using 1 000 000 particles released for a given range of altitudes and arriving in 1∘×1∘ box centered on the Cap Corse station. The ranges of altitudes are selected to highlight some particular features of the aloft aerosol transport. The residence time is computed for the whole duration of the simulation and integrated between the surface level and the top altitude of the release. The retroplume residence times (in seconds) gives the density of probability of the origin of the air mass that arrived at the station in the given altitude range from below altitudes.

Figure  presents the vertical profiles of the aerosol extinction coefficient obtained during the aforementioned events. Lidar data were not available on 19 June 2012. Figure a to e is for dust events and Fig. f to i is for pollution events. The main feature corresponding to the pollution cases profiles is the low level of altitude in the transport of aerosols. The 26 March 2012 (Fig. f) and 6 September 2012 (Fig. h) are very similar in terms of vertical distribution showing a thick layer between 1 and 3 km height. The backward simulations of the air masses arriving at the station between 500 and 2000 ma.s.l. (Fig. a and b) indicates a most probable transport from Eastern Europe. The aerosol extinction coefficient in the pollution layer reached 0.15 and 0.25 km-1 on 26 March and 6 September 2012, respectively. The value of extinction coefficient encountered during those events might be explained by the advection of air masses contaminated by biomass burning by-products originating from southeast Europe. Indeed for both events, Global Fire Assimilation System biomass burning emission inventory shows biomass burning emissions on the pathways of air masses arriving to the Cap Corse (not shown here).

The case of summer 2012 (Fig. g) and summer 2013 shows additional elevated layers. The extinction coefficient below 700 m increased from 0.09 to 0.19 km-1 from 20 to 25 August 2012 showing the build up of the pollution event during this period. Below 2 km height, the air mass was originating from the gulf of Genova and coastal areas of Italy. The aerosol layer observed between 2.5 and 4 km height was originating from Spain and Southern France (Fig. c) and passed over the Pyrenees mountain range. The backward plume shows also a possible northwestern African origin and thus possibly carrying dust that might explains the rather low value for the Ångström exponent.

During the month of July 2013 we observed a large increase in the AOD (Fig. ). This increase was also well observed in the aerosol scattering coefficient measured at the Ersa station (Fig. ). Below 1 km the aerosols are originating from the Pô valley (northern Italy) and we observed a large variability in the extinction coefficient ranging between 0.1 on the 15 July (i, black solid line) and almost 0.2 km-1 (g, red solid line) on the 19 July. A large aerosol layer was observed aloft between 2 and 4 km height (Fig. i). The air mass backtrajectory (Fig. d) indicated that the air mass was originating from northern Europe and underwent an ascent over the Alps before arriving above Cap Corse. An additional aerosol layers was also observed between 1 and 2 km height during the pollution event.

The dust events observed during summer 2012 (Fig. a and b) are rather similar in terms of vertical distribution. We observed a thick dust layer with a top located at 5 km height. The maximum extinction coefficient in the dust layer is 0.12 km-1. The bottom of the layer is located between 2 and 3 km. The air mass origin simulations indicate a transport of dust from North western Africa. In late June (Figs. a and e) the plume is first advected over the Atlantic and then transported to the east over the strait of Gibraltar. While the dust case in July (Fig. b) shows an advection of the plume east of the coast of Spain (Fig. f). The event observed in July 2013 (Fig. d) has a lower extinction coefficient at 0.06 km-1 but with a similar vertical structure than the previous cases. The advection pattern of the dust (not shown) was the same as for the case in June 2012. On August 2013 (Fig. e) the profile shows a dust layer up to 7 km. The extinction coefficient is 0.1 km-1 however such a high altitude event has an impact on the estimation of the extinction coefficient as the reference zone is affected by aerosols. The air mass backward simulations show the advection of the dust from the northwestern Sahara with an ascent over the Atlas mountain range (Fig. g).

A large dust layer arrived on the 30 April (Fig. c, black solid line) between 1 and 4 km height with extinction coefficient over 0.3 km-1. To highlight the variability in the aerosol vertical distribution during the dust event in April–May 2013 (Fig. c) we plotted the aerosol extinction on the 30 April and the 2 May. As the dust event passed over the site (Fig. c, red solid line), we observed an increase in the extinction coefficient below 2 km while the aloft layer vanished. Figure h highlights an advection pattern that is different to the summer cases. The dust is transported in a southerly flow across north-western Sahara and then over Tunisia before reaching the station. This kind of synoptic situation favors first the transport in altitude of air masses uplift in northwestern Sahara and then the low level transport of dust from closest sources in southern Tunisia.