2.2.1.

SEVIRI/MSG IR spectra analysis

The detection of desert dust is based on the analysis of the radiance sampled by SEVIRI on board MSG-1 (Meteosat-8) meteorological satellite²⁸ with capability to observe the high temporal frequency and also in the night. IR observations in the atmospheric windows between 8 and $12\mu \mathrm{m}$ have been exploited considering the different spectral trend of desert dust optical properties with respect to those of clouds and in clear conditions (i.e., without the presence of dust or clouds). In particular, BTs are calculated from the SEVIRI level 1.5 data in the 8.7, 10.8, and $12.0\mu \mathrm{m}$ channels.²⁹ Then, two BTDs are derived as (${\mathrm{BT}}_{12.0}$ to ${\mathrm{BT}}_{10.8}$) and (${\mathrm{BT}}_{10.8}$ to ${\mathrm{BT}}_{8.7}$) enabling the enhancement of the presence of desert dust [Fig. 2(a)].

Fig. 2

(a) Brightness temperature differences plot discriminating the presence of desert dust (solid squares in magenta) and (b) composite false color RGB image (see also Sec. 3.1.1), relative to IR SEVIRI observation at 8.7, 10.8, and $12\mu \mathrm{m}$ on July 27, 2005, 01:00 UTC. Data refer to the event (i) in Table 1.

2.2.2.

UV–VIS aerosol optical properties and OMI/Aura observations

Desert AOP as single-scattering albedo (SSA) and AOD present spectral trends that can be profitably used to highlight the presence of desert dust.^18,19 Within the UV–VIS spectral range, SSA relative to this type of particles increases with wavelength. Accordingly, the absorbing AOD (AAOD) decreases while the extinction AOD (EAOD) typically decreases slowly with respect to the other types of atmospheric aerosols. For seven different types of aerosols, Fig. 3 shows the spectral trend of the SSA and the normalized EAOD, ${\mathrm{EAOD}}_{\mathrm{n}}$

Fig. 3

(a) Spectral trend of normalized extinction aerosol optical depth, Eq. (1) and (b) single-scattering albedo derived from Ref. 30 for three desert dust and for other aerosol types.

By considering the optical absorbing features [i.e., the SSA plot of Fig. 3(b)], the significant increasing spectral trend of the desert aerosols SSA with respect to the other aerosol types can also confirm the dust presence.

UV–VIS spectra registered by OMI sensor onboard the Finnish–Dutch EOS Aura spacecraft are usefully employed. A spectral analysis of the AOP (level 2G gridded data products, OMAERUVG.003) from OMI/Aura data products (15 orbits/day; $13\times 24{\mathrm{km}}^2$ spatial resolution at nadir) into $0.25\mathrm{deg}\times 0.25\mathrm{deg}$ global grids products have been carried out to recognize the presence of the dust over the study area for the selected periods (cf. Table 1).

2.2.3.

MODIS satellite-based PM using VIS–NIR observations

A Northern Italy synoptic view of ${\mathrm{PM}}_{2.5}$ and ${\mathrm{PM}}_{10}$ concentrations is reconstructed³³ from MODIS/Terra and Aqua aerosol level 2 data (C005 collections) at $10\times 10{\mathrm{km}}^2$ spatial resolution, meteorological mixing layer, relative humidity fields simulated by MM5³⁴ model at spatial resolution of $12\times 12{\mathrm{km}}^2$, and in situ ${\mathrm{PM}}_{2.5}$ and ${\mathrm{PM}}_{10}$ as described in Ref. 35.

MODIS-based PM data (Fig. 4) are here selected to compute maps of ${\mathrm{PM}}_{2.5}$ to ${\mathrm{PM}}_{10}$ concentrations ratio averaged over the period of the selected events (Table 1). This ratio assumes typical values around or higher than 0.75 in case of anthropogenic pollution and around or lower than 0.6 in highlighting the presence of coarse particles, such as desert dust aerosol.³⁷

Fig. 4

2008 yearly average of MODIS-based particulate matter concentrations at surface over Northern Italy: (a) ${\mathrm{PM}}_{2.5}$ map and (b) ${\mathrm{PM}}_{10}$ map. Reproduced with permission, courtesy of Alessandra Cacciari.³⁶

2.2.4.

Retrieval of chl-a concentrations from MERIS data

To assess the effect of dust deposition on Lake Garda waters, 26 MERIS level 1B FR (300 m) images relative to years 2005 to 2007 (Table 1) were used. The images were processed through the ESA-BEAM toolbox, in order to remove radiometric disturbances as well as atmospheric and adjacency effects. In particular, the BEAM Smile correction was applied to all images.³⁸ The Smile corrected images were processed with the BEAM plug-in Improved Contrast between Ocean and Land (ICOL), to remove the adjacency effects of land on water.³⁹ To retrieve chl-a concentrations, the MERIS case 2 water BEAM processor (C2R)⁴⁰ was used as it was successfully adopted to produce MERIS-derived chl-a concentration in Lake Garda.^41,42 MERIS level-1 and C2R level-2 flags were used for excluding pixels with risk of clouds and glint. The trend of chl-a concentration was finally evaluated in a pelagic area (about $17.5{\mathrm{km}}^2$ in Fig. 1) as it is most likely the least affected by residual adjacency effects.

3.1.1.

SEVIRI/MSG dust tracking

As introduced in Sec. 2.2.1, BTDs analysis of the SEVIRI data has been applied to the selected events (Table 1) for tracking and evaluating the origin of the desert dust plume.

Sequences of RGB (red, green, blue) composite images ($\mathrm{R}={\mathrm{BT}}_{12.0}-{\mathrm{BT}}_{10.8}$; $\mathrm{G}={\mathrm{BT}}_{10.8}-{\mathrm{BT}}_{8.7}$; $\mathrm{B}={\mathrm{BT}}_{10.8}$) are produced using the SEVIRI/MSG BTD at spatial and temporal resolutions of 4 km and 15 min, respectively.

SEVIRI BTDs showed a significant transport of dust over the Mediterranean area and toward Italy for the dust events (i) and (ii) (cf. Table 1). In Fig. 2(b) desert dust plume over Mediterranean Sea toward the Italian coasts appears pink or magenta, while thick high-level clouds are red-brown, thin high-level clouds appear very dark, and clear sky pixels are blue.^43,44 As the plume approaches Italy and also considering a decrease of particle concentrations during its transport, the dust comes less distinguishable.

3.1.2.

Aerosol optical properties derived from OMI and MODIS-based PM_2.5 to PM₁₀ ratio

Figure 5 presents 3-days averaged maps of three spectral values of AOP relative to the first dust event (cf. Table 1 and Fig. 2). The averaged values of EAOD and SSA are shown at 354, 388, and 500 nm. As typical properties of desert dust aerosol, the increasing spectral trend of SSA values can be clearly recognized in the values presented by the pixels located between the two main islands of Italy (southwestern part of the country). Within the desert plume over the sea, SSA values averaged in the Mediterranean area (from 39.3°N–8.3°E to 37.3°N–13.6°E) turn to be $0.81\pm 0.02$, $0.84\pm 0.02$, and $0.91\pm 0.02$ at 354, 388, and 500 nm wavelengths, respectively.

Fig. 5

UV–VIS desert dust detection over Mediterranean Sea as retrieved from OMI/Aura satellite data processing. All maps show 3-days averaged values from July 27 to July 29, 2005 of (a–c) the extinction AOD (EAOD) and (d–f) single-scattering albedo (SSA) at the wavelengths of (a, d) 354, (b, e) 388, and (c, f) 500 nm.

These SSA values and the corresponding spectral ${\mathrm{EAOD}}_{\mathrm{n}}$ values, calculated averaging over the same Mediterranean area the spectral EAOD (Fig. 5, first row) and using Eq. (1), are plotted in Fig. 6 for a comparison with the relative spectral AOP showed in Fig. 3. It can be noticed that there is very good agreement of the retrieved OMI values (blue square) with respect to the desert background aerosol model (red triangle).

Fig. 6

Comparison between aerosol models of Fig. 3 (rural/clean continental, urban, desert CV, and desert back) and OMI-retrieved spectral averaged AOP (squared symbols) over Mediterranean (Fig. 5) and over Lake Garda areas (Fig. 7), for the event (i), and for Lake Garda area for events (ii, iii, and iv); (a) spectral trend of ${\mathrm{EAOD}}_{\mathrm{n}}$ and (b) SSA values. Medit-event and LG-event indicate Mediterranean and Lake Garda areas, respectively.

Since the study area is Lake Garda, a zoom in Northern Italy of maps in Fig. 5 is showed in Fig. 7. In the last row of Fig. 7, the OMI SSA of Lake Garda presents averaged values of $0.96\pm 0.02$, $0.97\pm 0.02$, and $0.99\pm 0.02$ at 354, 388, and 500 nm, respectively. The mapped AOP spectral values confirm that the desert dust outbreak at the end of July 2005 reached also Northern Italy. These SSA values and the corresponding spectral ${\mathrm{EAOD}}_{\mathrm{n}}$ [Eq. (1)], calculated averaging over the Lake Garda area the spectral EAOD of Fig. 7, are also included in Fig. 6 (squared green symbols) to compare them with the modeled spectral AOP depending on aerosol types rural/clean continental, urban, desert CV, and desert background.

Fig. 7

Zoom over Northern Italy (Lake Garda area) of the same period of averaged maps presented in Fig 5. (a–c) the extinction AOD (EAOD) and (d–f) SSA at the wavelengths of (a, d) 354, (b, e) 388, and (c, f) 500 nm. Data refer to the dust event (i) of Table 1.

In this event, the spectral trend of retrieved OMI AOP (${\mathrm{EAOD}}_{\mathrm{n}}$ and SSA) confirms the presence of a significant component of dust in the study area. Figure 6 shows that the increasing spectral trend of OMI SSA and ${\mathrm{EAOD}}_{\mathrm{n}}$ is very similar to those ones of desert CV aerosol model (square red symbols, yellow filled). Nonetheless, the higher values of retrieved OMI SSA and the slight difference of retrieved ${\mathrm{EAOD}}_{\mathrm{n}}$ with respect to the SSA and ${\mathrm{EAOD}}_{\mathrm{n}}$ values, respectively, of desert CV aerosol type can be justified considering a proper mixing with a more scattering aerosol type as the rural. In particular, for these hydrophilic aerosol particles, high values of the relative humidity also contribute to increase the scattering features of the aerosol particles.⁴⁵ To provide an exhaustive picture of the desert dust case study, a selection of satellite-based information is presented in Fig. 8. In particular, ${\mathrm{NO}}_2$ tropospheric column amount is shown [Fig. 8(a)] as derived from daily OMI data processing⁴⁶ and averaged over the period of the event (i), July 23 to July 31, 2005. The values highlight a relatively small contribution of human activities to air pollution around the study area.

Fig. 8

On the top: average of period July 23 to July 31, 2005, dust event (i), of (a) OMI tropospheric ${\mathrm{NO}}_2$ concentrations and (b) MODIS-based ${\mathrm{PM}}_{2.5}$ to ${\mathrm{PM}}_{10}$ ratio values. On the bottom: averaged values for October 4 to October 10, 2007 period, dust event (ii), of (c) OMI tropospheric ${\mathrm{NO}}_2$ map and (d) MODIS-based ${\mathrm{PM}}_{2.5}$ to ${\mathrm{PM}}_{10}$ ratio map.

MODIS-based ${\mathrm{PM}}_{2.5}$ and ${\mathrm{PM}}_{10}$ daily concentrations for the investigated periods (cf. Table 1) have been used to compute maps of the MODIS-based ${\mathrm{PM}}_{2.5}$ to ${\mathrm{PM}}_{10}$ ratio and then averaged over the dust and no-dust periods.

An example of this map averaged over the July 23 to July 31, 2005 time period comprising the first event (i) and referred to the Lake Garda area is reported in Fig. 8(b). The average value of the ${\mathrm{PM}}_{2.5}/{\mathrm{PM}}_{10}$ ratio over Lake Garda is $0.61\pm 0.05$, highlighting a significant presence of coarse particles and thus confirming the dust transport in the study area. This analysis applied to the entire satellite data for detecting the desert dust transport has been applied also to the second case study—event (ii)—listed in Table 1.

Similarly, OMI-derived AOP presented in Fig. 7 is computed as averaged over the October 4 to October 10, 2007 time window. In particular, the SSA values averaged over Lake Garda result equal to $0.92\pm 0.02$, $0.93\pm 0.02$, and $0.96\pm 0.02$ at 354, 388, and 500 nm, respectively [cf. Fig. 6(b), square yellow symbol], also providing a spectral increasing behavior for pixels where desert dust aerosol occurs. Corresponding ${\mathrm{EAOD}}_{\mathrm{n}}$ data are shown in the left plot of the same figure [Fig. 6(a)]; these values are compatible with a presence of local aerosol (rural and/or urban) mixed with a dust-transported component.

Similarly to what has been performed in the analysis of the first dust event, Fig. 8 presents satellite-based information used to provide an exhaustive picture of the second event. Thus, ${\mathrm{NO}}_2$ tropospheric column amount is shown [Fig. 8(c)] as derived from OMI observations over the October 4 to October 10, 2007 period. The geographical distribution of the ${\mathrm{NO}}_2$ columnar content reflects the well-known very high pollution due to anthropogenic activities of the area.

Together with this, the MODIS-based ${\mathrm{PM}}_{2.5}/{\mathrm{PM}}_{10}$ ratio [Fig. 8(d)] in the Lake Garda area provided values around 0.6 and 0.7, highlighting the presence of coarse particles and thus confirming the presence of dust transported in this area.

The two no-dust case studies, events (iii) and (iv) (cf. Table 1), are presented with the aim of providing a benchmark for the comparison with the dust events singled out, confirming the capability of the satellite data to reliably detect and monitor the desert dust transport from the sources toward the domain of interest.

The OMI AOP spectral data relative to Lake Garda area and related to the first half of June 2005—event (iii)—are represented in Fig. 6. In this case, the retrieved ${\mathrm{EAOD}}_{\mathrm{n}}$ and SSA data (square gray symbols) have been averaged between June 1 and June 14, 2005 for the three OMI aerosol wavelengths. The spectral trend displayed by these parameters is characteristic of a scattering aerosol polydispersion suspended in the troposphere with the spectral extinction (${\mathrm{EAOD}}_{\mathrm{n}}$) decreasing and typical of a mixing of urban (black circle) and mainly rural (green circle) aerosols components. In particular, the OMI-retrieved SSA values averaged over Lake Garda slightly decrease with wavelength with a value of $0.96\pm 0.02$ at the three OMI wavelengths and are very similar to SSA values of rural/clean continental aerosol model (values around 0.95 of green circles).

${\mathrm{NO}}_2$ vertical column [Fig. 9(a)] and MODIS-based ${\mathrm{PM}}_{2.5}/{\mathrm{PM}}_{10}$ ratio [Fig. 9(b)] highlight the presence of anthropogenic pollution in the analyzed period of June 2005. In particular, the ${\mathrm{PM}}_{2.5}/{\mathrm{PM}}_{10}$ ratio over Lake Garda was around 0.7 and 0.8.

Fig. 9

On the top: average of period June 1 to June 15, 2005, no-dust event (iii), of (a) OMI tropospheric ${\mathrm{NO}}_2$ concentrations and (b) MODIS-based ${\mathrm{PM}}_{2.5}$ to ${\mathrm{PM}}_{10}$ ratio values. On the bottom: averaged values for October 30 to November 30, 2006 period, no-dust event (iv), of (c) OMI tropospheric ${\mathrm{NO}}_2$ map and (d) MODIS-based ${\mathrm{PM}}_{2.5}$ to ${\mathrm{PM}}_{10}$ ratio map.

The month of November 2006—event (iv)—has been selected as the second case study for a no-dust analysis. Similarly to the previous cases, Fig. 6 shows the averaged values of the period for Lake Garda of ${\mathrm{EAOD}}_{\mathrm{n}}$ and SSA (black squares) at the three wavelengths. Again, the spectral trend of the parameters confirms the presence in the troposphere of a basically scattering aerosol polydispersion, with a significant presence of urban aerosol type (black circles in Fig. 6). The satellite-based observations in Fig. 9(c) show values of ${\mathrm{NO}}_2$ columnar amount, highlighting a significant contribution of human activities to air pollution. The MODIS-based ${\mathrm{PM}}_{2.5}/{\mathrm{PM}}_{10}$ ratio [Fig. 9(d)] provides averaged values higher than 0.8, indicating a main presence of fine particles in the study area.

The first five rows in Table 2 report the values of the whole data set of the atmospheric parameters retrieved from OMI/Aura and MODIS/Terra and Aqua satellite sensors. All data are averaged over each temporal period indicated in Table 1 and over the Lake Garda area.

Table 2

Atmospheric parameters derived from OMI/Aura (EAODn*, SSA, NO2, UV aerosol index) and MODIS/Terra and Aqua (PM2.5 to PM10 ratio) averaged over time period of the events indicated in Table 1 for the Lake Garda area. The uncertainties represent the spatial variability of averaged value (i.e., the RMS linked to the mean spatial value). The percentage of mixing ratio by number [Cm, see Eq. (2)] of the modeling aerosol type is also reported for each Lake Garda event.

A further indicator of the desert dust aerosol suspended in the troposphere is represented by the aerosol index (AI) that can be considered as an indicator of the presence of UV-absorbing aerosols (as mineral dust and smoke) against nonabsorbing aerosols (as sea salt and sulfate particles).⁴⁷ Accordingly, last four rows in Table 2 show OMI AI for each Lake Garda—event. Typically, AI values greater than 2 highlight the presence of dust. In the analyzed Lake Garda—events, the retrieved OMI AI values are between 0.8 and 1.0 also in the case of dust events (i) and (ii). Actually, these values prevent this parameter to be considered as a clear indicator of the presence of absorbing aerosol, like desert dust. This is mainly due to the small values of columnar EAOD. In fact, in the event (i) inside the dust plume over the Mediterranean area (Fig. 5) AI assumed values greater than 3.

To also provide a quantitative estimate of composition of the satellite-retrieved aerosol for each event, it can be assumed that the retrieved aerosol is composed by an external mixing³⁰ of the different aerosol types, desert CV, rural/clean continental, and urban, retrievable over the Lake Garda area. Thus, if $C_m$ [$0÷1$] is the mixing ratio by the number of the $m$’th aerosol type and ${\mathrm{EAOD}}_{\mathrm{n}}^{*}$ are the OMI-retrieved values reported in Table 2, the following system of equations can be considered

For each Lake Garda—event, the values of $C_m$ are also reported in Table 2 as solutions of Eq. (2). In the last row of the table, the percentage of the ${\mathrm{EAOD}}_{\mathrm{n}}^{*}$ is also reported as residual of the right member of Eq. (2). This is the fraction of EAOD retrieved, which cannot be fully explained by Eq. (2). Solution of this equation confirms a significant presence of desert dust for the events (i) and (ii) (89% and 75%, respectively), while for the events (iii) and (iv), the most important components are the rural/clean continental (61%) and urban (67%), respectively.