Aerosol Monitoring over Athens Using Satellite and Ground-Based Measurements

3.1. MERIS Data

MERIS data are currently available in 15 spectral bands, in the nominal spectral range 400–900 nm, in two modes; at full spatial resolution (260 m by 290 m), and at reduced resolution (1200 m), with a swath of 1150 km around nadir. MERIS provides a global coverage of the Earth every 3 days [21]. A series of MERIS Level 2 Full Resolution images acquired over the Prefecture of Attica including the Greater Athens Area (GAA), were pre-selected. Their selection was based on the criterion of identical orbit over the study area in order to avoid geometrical distortions that would affect the radiometry of the images. The images selected corresponded to six dates (Table 1) with minimum cloud cover and high acquisition quality. MERIS Level 2 images had already been corrected for gaseous absorption and Rayleigh contributions and represented “top-of-atmosphere reflectance values”. All images were acquired at approximately 11:30 a.m. local time. The pre-selected images were visually inspected in order to detect, and eventually delineate, thin clouds or dense dust cloud transportation (e.g., from Sahara). MERIS images were classified with respect to pollution levels on the basis of ground-based pollutant concentration measurements in order to designate a “reference image” that is, the image with the lowest pollution load, that was required to apply the DTA method.

3.2. MODIS Data

MODIS acquires daily imagery in 36 spectral bands from the visible to the thermal infrared (29 spectral bands with 1-km resolution, 5 spectral bands with 500 m resolution, and 2 with 250 m resolution, nadir pixel dimensions). The data used in this study are MODIS aerosol retrieval products based on distinguished algorithms over land and ocean to obtain AOD at 550 nm (AOD₅₅₀) and the proportion of AOD contributed by fine-mode aerosols (fine-mode fraction: FM), which are only created over cloud-free regions. Both AOD₅₅₀ and FM data correspond to Collection 5 (c005) data, where much of the high bias is removed [22]. However, in dust-aerosol regimes, retrieved AOD₅₅₀ will have greater error due to their non-spherical effects [23, 24]. The FM fractionwas initially given over ocean but it can now be obtained also over coastal areas. The retrieval algorithms are fully described by Remer et al. [25] and Levy et al. [22].

The MODIS study area included GAA centered at 37.5^°N latitude and 23.4^°E longitude with 10 km × 10 km (Level 2) spatial resolution. Since Athens is a coastal urban area the previous pixel is probably affected by the differences between land and sea. An additional uncertainty in both AOD and FM retrievals over coastal areas is caused by the land-ocean sub-pixel contamination. Even with improved formulation of FM, retrievals of both parameters still have large uncertainties because the assumption of the empirical surface reflectance ratios seems too simple to account for complexities of land-surface reflectance [17].

3.3. PM₁₀ and Gaseous Pollutant Measurements

The Ministry of Environment (MINENV) provided the necessary air-pollutant concentration data. MINENV operates an organized monitoring network covering GAA with eighteen stations on hourly basis. The examined data refer to hourly concentration measurements of NO₂, NO, SO₂, O₃, and PM₁₀ (Table 1). PM₁₀ concentrations were monitored at four air-pollution monitoring stations within GAA representative of different environments. “Aristotelous” station is established in the city center near a road with heavy traffic. High levels of primary air pollutants (e.g., NO₂, SO₂) are still measured at this station. “Zografou” station is located in the University of Athens campus in a suburban region with low traffic and far from inhabited areas. This location is at the east edge of the basin, at the foothills of mountain Hymettos. It should be noted that these two stations were used by Chaloulakou et al. [3] as representing urban and suburban environments from the PM monitoring point of view. “Agia Paraskevi” is a suburb of Athens located in the northeast edge of the basin. Last, “Likovrisi” station is situated in a north suburb with low population density and close to open areas near the Parnitha mountain. It should be noted that, despite the distance from the city center, air-pollution and PM concentrations at this station are significant due to the city circulation patterns, which carry the pollutants to the north.

PM₁₀ filter samples were collected using low-volume samplers. The sampling flow rate was 16.71 l min ⁻¹. The particles were collected on 47-mm Pallflex TX40 lm⁻¹ filters (Teflon-coated glass fiber filters), which were mounted in plastic filter holders. Each filter was inspected for its integrity prior to its use. Particle concentrations were determined gravimetrically using an electronic microbalance (Metter Toledo AT201), with 0.01-mg resolution. Air pollutants, such as nitrogen dioxide (NO₂), nitrogen oxide (NO), sulfur dioxide (SO₂), and tropospheric ozone (O₃), are concurrently measured by the MINENV network. NO₂ and NO, O₃, and SO₂ concentrations are determined using chemiluminescence, UV, and fluorescence techniques, respectively.

Analysis of PM₁₀ and gaseous pollutant data was initially performed for classifying the available images into highly, moderately, slightly polluted and pollution-free. According to the examined ground-based pollution information, the selected images acquired on 25/6/2003, 4/7/2003, 14/7/2003, 15/8/2003 and 31/8/2003 were considered as “pollution images” compared to that of 22/6/2003. In particular, among these dates the 4/7/2003 and the 31/8/2003 indicated rather high levels of pollution and AOD (i.e., PM₁₀ exceeded the upper limit of the mean daily value of 50 μgm⁻³), the 15/8/2003 shows light pollution, and the other two days present moderate pollution levels. Finally, the 22/6/2003 was identified as “reference image” (pollution-free) according to its low PM₁₀ values.

5.1. Satellite Images

Figures 1(a)–1(e) show true color Terra-MODIS images covering the whole Eastern Mediterranean and corresponding to thefive examined days. These images, due to their largerspatial coverage, are capable of detecting Saharan dust outbreaks above the Eastern Mediterranean basin. The MODIS pictures referring to 25/6/2003 (Figure 1(a)), 14/7/2003 (Figure 1(c)) and 15/8/2003 (Figure 1(d)) seem to be rather clear over the whole Greek territory, while the sea glint effect is quite intense. A clearly different pattern was obvious on 4/7/2003 (Figure 1(b)) when a thin dust plume was blown north of Libya affecting the Aegean Sea and continental Greece intensifying the aerosol load above Athens. However, even on this day the sea glint is obvious above the seas adjoining the Greek subcontinent. Figure 1(d), referring to 15/8/2003, represents the clearest day over Athens in accordance to the minimum air-pollutant and PM₁₀ concentrations (Table 1). In contrast, 31/8/2003 is a dusty day with a dust plume blowing northeastward of the Tunisian/Libyan coasts. The MODIS image clearly reveals the geographical extension of this dust plume moving northeast through the Central Mediterranean, Aegean Sea and continental Greece.

5.2. Back Trajectories

Analytical air-mass back trajectories for the five dates, obtained by the Hybrid Single-Particle Langrangian Integrated Trajectory (HYSPLIT) Model [30], were used to identify the pathway of air masses over Athens. The back trajectories were computed for 3 distinct arrival levels, namely 100 m, to give representative origin of air masses near the surface, 1500 m, which could serve as a representative height for the boundary layer, in which the majority of the aerosol particles was present, and 4000 m for the presence of Sahara dust layers to be captured. The 4-day back trajectory analysis, presented in Figures 2(a)–2(e), clearly verifies the two dusty days (Figures 2(b) and 2(e)), where the trajectories at both 4000 m and 1500 m showa Sahara origin. Particularly, they originated from areas in Northwestern Africa (Algeria and Morocco) under the influence of intense surface winds. In their trip over the Sahara desert they were both within the boundary layer as was also established via the HYSPLIT model. It should be noted that this direction is the main pathway for the dust plumes influencing both the Central [31, 32] and Eastern Mediterranean [33]. On the other three days the air masses influencing Athens came at all altitudes from the northern sector traversing Central Europe and the Balkan countries. It was established that these air masses usually carry anthropogenic or biomass-smoke aerosols over Greece [34]. From MODIS fire-mapping website http://maps.geog.umd.edu/firms/ it was found (not shown) that on 12–15 August a large number of fire events occurred in the Balkan counties probably influencing theair mass on 15/8/2003; therefore, an increase in α values was derived from MERIS (see Section 5.4).

5.3. DREAM Model Applications

The dusty days (4/7/2003 and 31/8/2003) were also successfully forecasted by the Dust Regional Atmospheric Modelling (DREAM) model [35], which has been extensively used for the identification of dust events in the Mediterranean [33, 36, 37]. The 0-hr forecast presented in both figures is the safest prediction and represents the real situation. Typical output results for 12 UTC (0-h forecast) are given in Figures 3(a) and 3(b) indicating the dust loading affecting the whole Greek territory on these days. In these figures the wind speed and direction at 3000 m are also shown, and they are consistent with air-mass trajectories reaching Athens (Figures 2(b) and 2(e)). From the wind pattern at 3000 m it can be revealed that an anticyclone was established in North Libya (Figure 3(a)) and North Algeria (Figure 3(b)) carrying dust particles over Greece. These anticyclones were also observed (not shown) on the preceding days to those selected. An important finding using the DREAM model is the dry deposition of dust particles occurring over Greece on both dusty days (not presented), thus leading to significant enhancement of PM₁₀ concentrations on the surface. This is very important to be mentioned since, in numerous cases, the air masses transport dust over Greece in the free troposphere without any signal in PM₁₀ measurements [38]. On the other three days, no dust load was found over Greece and for this reason the respective figures are not presented.

5.4. Comparison between Satellite and Ground-Based Measurements

MERIS AOD values retrieved with the proposed method (DTA) precisely over the locations of the MINENV stations were compared to PM₁₀ measurements at the respective stations. The data sets for all days were correlated through linear regression and the correlation coefficients for each one of the 14 MERIS spectral bands are presented in Table 2. The correlations between satellite-based AODs and ground-based PM₁₀ concentrations are higher than 70% (R² > 0.50) when using any of the spectral bands in the visibleregion. The highest correlation (R² = 0.83) was found when using the MERIS AOD at 560 nm for all days combined. On the other hand, the correlation coefficients drop significantly in the NIR band (R² < 0.3). The correlations between AOD and PM₁₀ for individual days varied from R² = 0.63 to R² = 0.88; they are strong enough, especially in the visible spectrum but not in the infrared (>770 nm), to justify the above error levels. The slope values in Table 2 are small since the X axis corresponds to AOD(×10³). From all MERIS channels, channel 5 (560 nm) is considered the best correlated, where the slope is the highest and the intercept the lowest. In this channel the strong correlation justifies the dependency of PM₁₀ on AOD and, in general, the above correlation is statistically significant at the 95% confidence level for the visible spectrum. The revealed linearity and the low scatter at 560 nm allow a first-order prediction of the PM₁₀ levels near the surface based on the AOD values. The strong positive correlation between the AOD-derived from satellites and ground-based PM₁₀ concentrations suggests that the majority of the aerosol particles are within the boundary layer; their influence on the aerosol load increases with the slope of their correlation. Papayannis et al. [33] also proved via lidar signals that even during the dust events, the majority of the aerosols are within the boundary layer, below 1500 m. The relation between PM₁₀ and AOD also indicates to what extent and in which cases and columnar data reflect surface characteristics.

Figure 4 presents mean AOD, PM₁₀ and Aerosol Index (AI) values, averaged over GAA for each individual day. In the same figure AOD values from MERIS correspond to 560 nm and constitute the average values for all available AODs above all MINENV stations, retrieved at a spatial grid of 260 m by 290 m; while the vertical bars express one standard deviation from the mean value. The AOD₅₅₀ from both Terra-MODIS and Aqua-MODIS have been retrieved from the pixel centered above Athens city center (37.5^°N 23.4^°E) with a spatial resolution of 10 km × 10 km. AI data using version “8” algorithm were obtained from the TOMS sensor onboard Earth-Probe platform. AI is available on a daily basis, at a 1^° × 1.25^° resolution. Finally, PM₁₀ concentrations constitute the mean values in the time interval between 11 : 00 and 12 : 00 a.m. (local time) thus best corresponding to MERIS (11 : 30 a.m.) and to Terra-MODIS (10 : 30 a.m.) passes. Such a comparison between “MERIS retrievals” and “MODIS products” may include some arbitrariness, mainly due to the different AOD-retrieval codes and different spatial resolutions. It is, however, noteworthy that all satellite sensors and ground-based measurements were in agreement regarding the aerosol-load variation over GAA between the different days examined. It was found that the spatially-averaged AOD values from MERIS are highly correlated with the respective AOD product from Terra-MODIS (R² = 0.85) or Aqua-MODIS (R² = 0.83). Furthermore, the MERIS spatially-averaged AOD at 560 nm was correlated with ground-based PM₁₀ concentration measurements with R² to be as high as 0.82. The strong correlations between PM₁₀ and columnar AOD are mainly affected by dust events on two of the days, which significantly enhanced dust deposition on the surface and PM₁₀. Similar strong correlation between PM₁₀, sun-photometer based AOD and TOMS AI was found in Crete during Sahara dust events [38].

From Figure 4, the 25th of June and the 14th of July can be characterized as days with moderate air pollution levels; the 15th of August as day with low pollution levels, and the 4th of July and particularly the 31st of August as days with high or severe air pollution. Mean MERIS AOD at 560 nm was found to range from 0.13 to 0.33 when days are examined individually (Figure 4). The lower MERIS-AOD values, compared to those obtained by MODIS, are attributed to the “relative” method used to derive AOD including the use of a “reference” day, which was assumed to be pollution-free. The examination of this “reference” day in view to PM₁₀ showed low but not negligible pollution in the Athens atmosphere (Table 1), while Terra and Aqua MODIS AOD₅₅₀ was 0.195 and 0.175, respectively. As previously shown from back trajectories and DREAM applications, on two days (4 July 2003 and 31 August 2003), a Sahara dust plume affected GAA. These dust events further degraded air quality in the Athens urban environment, thus affecting mainly PM₁₀ concentrations. Furthermore, on these days the TOMS AI values were high, thus indicating the presence of dust. In Mediterranean coastal urban areas, such as Athens, which are close to North African arid regions, the effects of dust outbreaks play a key role in PM₁₀ concentrations and further contribute to the urban air quality and human health. Many studies [36, 39–41] have established that dramatic PM₁₀ enhancements in Southern Europe are with Sahara dust outbreaks. A PM₁₀ concentration enhancement with simultaneous increase in columnar AOD is characteristic of convective dust transport over Greece [38]. Back trajectories at 4000 m and 1500 m on 4 July 2003 and 31 August 2003 verified this type of transport; so did DREAM simulations, which showed dry deposition of dust particles over Athens on these days.

Using the least-squares method the Ångström exponent α was derived from MERIS-AOD values covering the spectral range 412.5–885 nm on each individual day. These values, associated with their standard deviations caused by averaging them over GAA, are presented in Figure 5. In the same figure the FM values from both Terra- and Aqua-MODIS sensors are also plotted. Despite the limited number of days and difficulties encountered (i.e., different spatial resolution, significant typical error in α computation) the analysis gave similar results with respect to α and FM values. The only discrepancy was revealed on 15 August 2003 where the computed large mean α value was not in accordance with respective FM values. On the other days, high α values were associated with high FM values. The two dusty days (i.e., 4 July 2003, 31 August 2003) were characterized by low values in both α and FM. The linear correlations between α and FM values were significantly strong; their variance was 93% and 70% in the case of Terra- and Aqua-MODIS, respectively. This result clearly establishes the agreement between the two satellite sensors despite their different spatial resolution and the differences in product definition (i.e., α and FM). Terra-MODIS correlates better than Aqua-MODIS with MERIS retrievals due to proximity between their passes (approximately one hour lag). Unfortunately, there are no systematic measurements of fine particles (e.g., PM_2.5 or PM_1.0) in GAA in order to compare with α and FM values.

Figure 6 shows the correlations between daily mean PM₁₀ and air-pollutant concentrations for all the available data presented in Table 1. Particularly, the correlation between PM₁₀ and NO₂ (R² = 0.67) is statistically significant (P < .001) indicating that most of the aerosols that contribute to PM₁₀ have originated from vehicle exhausts, which are the main source for NO₂ production. NO₂ is a highly variable atmospheric constituent, which plays a key role in the complex ozone cycle, both in the stratosphere where it is naturally present, and the troposphere, where its concentration may be high due to pollution. The respective correlation (PM₁₀ versus NO) is positive but not statistically significant (R² = 0.39), while no correlation between PM₁₀ and SO₂ seems to exist. Strong associations between particle-mass concentrations and primary gaseous pollutants have been previously reported in Birmingham, suggesting the importance of traffic-related emissions [42]. Gaseous pollutants (except ozone) are mostly associated with local sources, including exhaust emissions, domestic heating and industrial activities. These strong associations suggest that a large fraction of particles are emitted locally. On the other hand, the tropospheric ozone, which is a photochemical pollutant, exhibits a decreasing trend when PM₁₀ concentration increases. Within the Athens Basin, primary air pollutants (such as NO) are mainly produced in downtown Athens (“Aristotelous” station), while the highest tropospheric O₃ concentrations take place at stations far from the city center, and mainly in the northern areas of the Basin (“Likovrisi” and “Agia Paraskevi” stations). This is attributed to the fact that tropospheric O₃ is a very reactive pollutant and, therefore, the highest concentrations of other pollutants in the center (like NO₂, NO, SO₂) contribute to its destruction.

In Figures 7(a) and 7(b) the diurnal variation of PM₁₀ at two stations within GAA is presented for the five days considered. The PM₁₀ values are the hourly average in the time intervals presented. The two stations selected are the urban “Aristotelous” (Figure 7(a)) and the suburban “Zografou” (Figure 7(b)), in order to establish the differences in PM₁₀ concentrations caused by local emissions. It should be noted that similar values at these stations have already been published by Chaloulakou et al. [3]. The main remarks from both figures is the inexistence of a clear diurnal pattern especially at “Aristotelous” station, while at “Zografou” the PM₁₀ concentrations exhibit an increasing trend from early morning to midday, remaining nearly constant in the rest of the day. It is also obvious that except of the local sources the dust deposition on the two dusty days significantly affected the PM₁₀ concentrations (see daily mean values in both figures). In both figures, the daily threshold value of 50 μgm⁻³ suggested by EU is also shown in order to reveal the considerably higher PM₁₀ values over the Athens center. A full year monitoring of PM₁₀ concentrations in Athens [3] gave a daily average of 75 μgm⁻³ during the sampling period (1 June 1999 to 31 May 2000) at “Aristotelous” station. This annual mean is considerably higher than the EU annual PM₁₀ standard of 40 μgm⁻³.