Long-term visibility variation in Athens ( 1931 – 2013 ) : a proxy for local and regional atmospheric aerosol loads

This study explores the interdecadal variability and trends of surface horizontal visibility at the urban area of Athens from 1931 to 2013, using the historical archives of the National Observatory of Athens (NOA). A prominent deterioration of visibility in the city was detected, with the long-term linear trend amounting to −2.8 km decade−1 (p < 0.001), over the entire study period. This was not accompanied by any significant trend in relative humidity or precipitation over the same period. A slight recovery of visibility levels seems to be established in the recent decade (2004– 2013). It was found that very good visibility (> 20 km) occurred at a frequency of 34 % before the 1950s, while this percentage drops to just 2 % during the decade 2004–2013. The rapid impairment of the visual air quality in Athens around the 1950s points to the increased levels of air pollution on a local and/or regional scale, related to high urbanization rates and/or increased anthropogenic emissions on a global scale at that period. Visibility was found to be negatively/positively correlated with relative humidity/wind speed, the correlation being statistically valid at certain periods. Wind regime and mainly wind direction and corresponding air mass origin were found to highly control visibility levels in Athens. The comparison of visibility variation in Athens and at a non-urban reference site on Crete island revealed similar negative trends over the common period of observations. This suggests that apart local sources, visibility in Athens is highly determined by aerosol load of regional origin. AVHRR and MODIS satellite-derived aerosol optical depth (AOD) retrievals over Athens and surface measurements of PM10 confirmed the relation of visibility to aerosol load.


29
Visibility is defined as the greatest distance at which a black object of suitable dimensions (located on the 30 ground) can be seen and recognized, when observed against the horizon sky during daylight, (WMO 1992).

31
Visibility represents one of the dominant features of the climate and landscape of an area. Although it is highly 32 affected by atmospheric circulation and the prevailing meteorological conditions, under clear sky conditions it is 33 mainly determined by the loading of atmospheric aerosols (Davis, 1991;Lee, 1994

38
Aerosols in the atmosphere contribute to light extinction by scattering and absorbing, thus reducing visibility 39 (Appel et al., 1985;Chan et al., 1999;Elias et al., 2009;Singh and Dey, 2012). The impact of particulate matter 40 on visibility depends on its physical (e.g. particle size distribution) and chemical properties (Dayan and Levy, 41 2005). In particular, visibility is inversely related to light extinction coefficient, which is determined by scattering 42 and absorption of light by gases and particles, the latter (e.g. sulphate and carbon containing particles) being the 43 main contributor (Malm, 1999

44
Sulphate and carbon containing particlesplaya major role in light extinction, while the role of relative humidity 45 (RH) on visibility is also important (Larson and Cass, 1989;Malm, 1999), as when RH reaches saturation values, 46 visibility deteriorates due to fog formation and the hygroscopic growth of SO4 2-, NH4 + and NO3particles (Tang, 47 1996; Sing and Dey, 2012). At local and regional level, wind speed and direction are also very important factors, 48 as they determine the transport and origin of air pollution.

49
Although the use of visibility as a viable atmospheric variable has been disputed by many researchers due to the 50 numerous biases related to observational procedures (Davis, 1991), visibility statistics have been increasingly 51 used as a surrogate for aerosol load (Zhao et al., 2011), especially since visibility records span quite long-term 52 periods. Today, there is a large number of studies that use visibility observations to investigate the spatial and 53 temporal variation of the optical properties of the atmosphere, mainly in relation to pollutant emissions and 54 aerosol load. These studies refer to global, regional and local scales. On a global scale, a decrease of clear sky 55 visibility over land from 1973 to 2007 is reported by Wang et al. (2009). This is interpreted in terms of aerosol 56 concentrations and its impact on incident solar irradiance.A significant decrease of visibility is observed over 57 3 Asia, South America, Australia and Africa , while over Europe visibility increased after the 1980s, as 58 a result of air pollution mitigation measures. Vautard et al. (2009) found a significant decrease in the frequency of 59 low visibility days in Europe after the 1980s, which is spatially and temporally correlated with SO2 emissions.

80
The present study explores the historical observations of visibility in Athens, which is the oldest time series of 81 visibility in Greece and, to our knowledge, one of the oldest, uninterrupted time series of visibility in the eastern 82 Mediterranean. The records are retrieved from the historical climatic archives of the National Observatory of 83 Athens (NOA) and span a period of more than 80 years . In the past, Carapiperis and Karapiperis 84 (1952) reported on the correlation between the visibility and the blue colour of the Attika sky, while 85 Kanellopoulou (1979) analysed visibility in Athens for the period 1931-1977 and reported a pronounced decrease 86 after the 1950s. Since then, there has been no other study to address changes in visibility, as well as the 87 4 factorsbehind these changesduring the last 40 years, when significant changes occurred in Athens in terms of 88 urban expansion, traffic load, 2004 Olympic Games construction and the economic recession (starting in 2008).

89
The inter-decadal variability and long-term trends of visibility in Athens are presented in the study. The role of 90 meteorology and aerosol load(of local and regional origin) on the variability and trends of visibility are 91 investigated and discussed, while the relationship between visibility and aerosol load is investigated, through the 92 analysis of satellite AOD retrievals over Athens, but also surface measurements of PM10 in Athens and Finokalia 93 station (Crete) over shorter periods.

105
Sea/land breezes appear along the NE -SW axis and play a dominant role in the accumulation of air pollutants 106 (Kalabokas et al., 1999a,b).

107
In order to compare our findings for Athens with a reference, remote site, the visibility records from the 108 Heraklion airport (HER) in Crete Island were used (Fig. 1). Heraklion is located about 330km south of Athens, 109 while its airport is 5km east of the city with no significant (or systematic) influence by the urban web.

111
Athens has a temperate climate with warm and dry summers andwet and mild winters, typical for eastern 112 Mediterranean. Table 1 presents monthly and annual normal values along with standard deviations of the daily   113 mean, maximum and minimum air temperature, precipitation amount and precipitation frequency (PF) (defined 114 5 as the number of days with total precipitation> 1mm, following WMO), relative humidity and wind speed in 115 Athens, based on the WMO reference period, 1971-2000. July and August are the warmest and driest months of 116 the year. The periods from May to September and from October to March represent the dry and wet periods of 117 the year respectively. Precipitation is sparse in summer (June-August), with the total amount averaging 20mm   118   and precipitation frequency averaging 3 days. Athens receives on average approximately 400 mm of rain per   119 year, corresponding to 43rainy days (Table 1).

120
During summer, the area is dominated by anticyclonic circulation that enhances air temperature and intensifies 121 urban heat island. Athens has been experiencing a significant warming since the mid 1970's, more pronounced in 122 summer, which is the additive result of regional warming and gradual intensification of the urban heat island 123 (Founda, 2011;Founda et al., 2015). Strong northeasterly winds in summer, known from antiquity as 'Etesians', 124 induce a relief on air temperature and air pollution levels in the city.

168
Regarding columnar aerosol load and using ground-based AOD measurements in Athens, Gerasopoulos et al.

169
(2011) showed that the greatest contribution (40%) to the annually averaged AOD, comes from regional sources   (Table 2). Classes are defined based on the greatest distance at 199 which a predefined object can be seen and recognized with the naked eye. The procedure requires that an 200 8 operator scans the horizon for predetermined objects. In the case of Athens, some historical buildings in the city, 201 but also certain objects of the surrounding landscape unaltered over the years, (e.g. objects on the mountains or 202 islands of the Saronic Gulf, Fig. 1

218
In an effort to explore the relationship between visibility and AOD over Athens, we used the Terra/Modis AOD 219 at 550nm, available since 2000. NASA's Terra satellite is sun synchronous and near polar-orbiting, with a 220 circular orbit of 705 km above sea level. MODIS is capable of scanning 36 spectral bands across a 2330 km 221 wideswath. MODIS aerosol products were used in order to analyze the temporal and spatial variability of 222 aerosols over the wide area of interest. In this study, we used daily level-2 collection 5.1 MODIS/Terra AOD at 223 550 nm. Daily overpass data for the specific area was extracted at a spatial resolution of 50x50 km 2 . Previous 224 studies have shown that such spatial resolution product ensures sufficient daily measurements without losing out 225 to the higher spatial resolution and hence provides a better opportunity of correctly viewing the atmospheric 226 aerosol load (Ichoku et al., 2002). The overpass time is 09:35 ± 45 min UT.

9
In addition, in order to further examine long-term satellite based AODseries in the area, we used the longest 228 satellite timeseries available from the Advanced Very High Resolution Radiometer (AVHRR). AOD retrievals 229 PATMOS-x AVHRR level-2b channel 1 (630nm) provide data over global oceans at high spatial resolution (0.  The separation of the time series into three sub-periodswas indicated by the fact that they represent periods of 277 changing trends. In the following, the much longer middle sub-period  was further separated into two 278 parts (1949-1975 and 1976-2003) as it corresponds to substantially different visibility conditions. Figure 5 279 illustrates the frequency of occurrence of different visibility ranges as described in Table 2 for different sub-280 periods.

285
Cumulatively, visibility exceeded 10 km at a frequency of approximately 80% during this period.

286
A progressive shift of frequency distribution towards lower visibility categories is observed in the next sub-287 periods. In particular, the frequency of very good visibility (20-50 km) drops to 13% and 6% for the periods 288 1949-1975 and 1976-2003   shown. It is noteworthy that RH at NOA does not exhibit any significant trend over the years (as already shown 309 in Fig. 4) and its monthly distribution remains almost unaltered in all sub-periods. As it results from Fig. 6 (a-d), 310 visibility exhibits a seasonal cycle in all sub-periods, with better visibility occurring in the warm and dry months 311 of the year. Although seasonality is observed in all sub-periods, the pattern is more evident and robust in the first 312 sub-period (Fig. 6a), with much higher visibility values (up to 40%) in the warm and dry months. The pattern of 12 visibility in this period is almost a mirror image of the pattern of RH and reflects the influence of RH on visibility 314 and the anti-correlation between these two variables. The lowest values of RH correspond to July and August 315 (mean value of RH ~35% at 14:00 LST) and this probably results in visibilityimprovement. Moreover, strong 316 northeasterly winds that prevail in eastern Greece during these monthsenhance ventilation and induce drier 317 conditions in the city, therefore improving visibility.

318
The distinct seasonal cyclein visibility of the first sub-period changed in the following sub-periods (Fig. 6, Figure 9 presents the running correlation coefficient (15-yrs window) between visibility and relative humidity at 373 NOA, over theperiod 1931-2013. As expected, the correlation coefficient between visibility and RH is negative, 374 indicating the anti-correlation between these two variables. High RH enhances water uptake by airborne particles, 375 leading to higher light scattering and thus, visibility impairment. Actually, when RH exceeds a threshold level 376 (e.g. > 70%), some inorganic salts, such as ammonium, sulfate and nitrate, undergo sudden phase transitions from 377 solid particles to solution droplets and become responsible for visibility impairment, as compared toother 378 particles that do not uptake water (Malm, 1999

394
Apart from wind speed, visibility was also found to be sensitive to wind direction. A distinct variability of 395 visibility with wind direction is observed in Fig. 10, for all sub-periods. Lower values of visibility are related to 396 southerly winds, as they bring either dust from Sahara or warmer and more humid air masses from the sea (see 397 also Figs 1, 2b). Southeasterly winds are, in general, weak winds (see Fig. 3), while southwesterly winds are 398 associated with sea breezes from the Saronic Gulf (Fig. 1). In general, sea breezes and calm wind conditions 399 favor the accumulation of pollutants and the formation of secondary aerosols and photochemical smog in Athens 400 15 (Colbeck et al., 2002), thus reducing visibility. A number of S/SW events are also associated with strong wind 401 speeds occurring during Sahara dust outbreaks, which enrich Athens atmosphere with dust particles that decrease 402 visibility (Figs 2, 3). As it results from Fig. 10, the highest visibility occurs under northwesterly winds and this is 403 robust for all sub-periods. An explanation for this, is that air masses originated from northwesterly directions are 404 much drier as they have lost water vapor after passing over the high mountainous basin of the Greek mainland 405 (e.g. Pindos mountain), while air pollution is also blocked within the boundary layer by the mountain chain.

418
Are the changes in visibility in Athens due to local factors or can they be considered representative of a more 419 extensive area? To answer this question, the Athens visibility record was compared with visibility at a reference, 420 non urban station. From the available stations in Greece disposing long-term visibility observations, we chose the 421 station at Heraklion airport (HER) in Crete Island (Fig. 1). Actually, both sites, NOA and HER, are exposed,most 422 of the year,to air masses of similar origin (from north and northeasterly directions) travelling over the Aegean 423 Sea, in contrast to other sites of the country that are strongly affected by the mountainous volumes of the Greek 424 mainland. Visibility observations at HER are available since the mid 1950s. Figure 11 Mylona (1996) it comes out that emissions by countries of N-NE sector (as defined in Fig. 2a)have the largest 452 contribution in total European emissions. Sulphur dioxide emissions increased by a factor of approximately 2.5 453 between 1950 and 1980 in these regions, which is analogous to the increase of total European emissions over the 454 same period.According to Mylona (1996), the contribution of emissions from the former USSR (but also Turkey)

497
To investigate the relationship between visibility and AOD changes, the two parameters are plotted together after 498 data binning. Visibility and AOD measurements have been used as follows: Visibility at 12:00 UT was used 499 according to the indices defined in Table 2

519
An additional analysis was conducted to verify the relationship between visibility and particulate pollution from 520 surface measurements, using a short dataset of PM10 in Athensas described in Section 2.5. Figure 14presents

540
The present work analyses, for the first time, the historical record of visibility at NOA (Athens) from 1931 to 541 2013 and interprets its temporal variability and trends in terms of relevant changes in atmospheric properties 542 (related to local or regional processes) and/or meteorological conditions. Since this is the longest record of 543 20 visibility observations in Greece and one of the oldest in the broader area of the eastern Mediterranean, the study 544 providesunique information on the atmospheric properties of the area in the past, when air pollution records are 545 missing. The study period was divided into sub-periods corresponding to different trends in the time series of 546 visibility, each sub-period being affected by different factors.

547
The impact of meteorological conditions on visibility was investigated in different ways. Visibility in Athens was 548 found to follow a seasonal cycle, with higher visibility corresponding to the warm and dry months of the year.

549
Seasonality is more distinctin the first sub-period of the time series (1931)(1932)(1933)(1934)(1935)(1936)(1937)(1938)(1939)(1940)(1941)(1942)(1943)(1944)(1945)(1946)(1947)(1948), while after the 1950s, the 550 seasonal cycle attenuates.Visibility was found to be negatively correlated with RH, the correlation being stronger 551 in the early part of the time series and attenuating over the years. On the contrary, a positive correlation between 552 visibility and wind speed was found, statistically significant during the late part of the time series, suggesting the 553 increasing role of winds on the cleanup of the atmosphere from air pollutants.Visibility was also found to be 554 sensitive to wind direction, reflecting the influence of air masses origin on visibility. Lower visibility levels are 555 constantly observed under southerly winds,corresponding to sea breeze circulation, but also to dust outbreaks.

556
The study demonstrated that visibility in Athens has undergone a prominent impairment since the early 1930s.

557
The overall trend of the annual visibility averages was found equal to -2.8 km decade -1 . The impressively higher 558 levels of visibility in Athens before the 1950s (also characterized by strong seasonality) reflect the transparency 559 of the atmosphere at that period, coherent with the poorer aerosol load from anthropogenic emissions (urban 560 and/or regional). The dramatic decrease of the visual air quality in the 1950s coincides with a number of events 561 (end of wars, rapid urbanization and rapid increase of anthropogenic emissions on local and regional scale) and 562 points to the prominent role of aerosol load in the atmosphere of Athens. Air pollution has gradually incurred a 563 severe visual pollution in the city, with visibility lower than4 kmcorresponding to more than half of the year 564 during the decade 2004-2013.

565
The comparison of theannual averages ofvisibility in Athens andat a reference, non urban site(HER) in Crete, 566 revealed similar and statistically significant negative trends at both sites, suggesting the major contribution of 567 long and regional range transport of natural and anthropogenic pollution sources in the GAA. An improvement of 568 visibility at HER around the 1990s was not associated with synchronous improvement of visibility in Athens, 569 where visibility deterioration continued until the early 2000s. Although negative trends of main gaseous air 570 pollutants are reported in Athens at that period (Kalabokas et al., 1999a), the direct effect of such pollutants on 571 light extinction is negligible compared to suspended particles and particularly to fine particles (<1μm). 572 21 A strong anticorrelation was found between visibility and PM10 levels in Athens, measured at two different 573 stations (urban and suburban) over the period 2008-2012 (Fig. 14). The relationship between AOD and visibility 574 in Athens was also examined in the study, using MODIS and AVHRR satellite data (Figs 12, 13), and confirmed 575 their negative correlation.

576
The analysis showed a recent stabilization (or even slight improvement) of visibility in Athens,consistent with the 577 observed decreasing trends of PM10in the city from 2004 to 2014 (Fig. 15). This could possibly be related to 578 reduced local anthropogenic emissions as a result of important transport infrastructures, but also of the economic 579 recession in Greece. Although this last argument is already supported by some recent research studies,the impact 580 of economic recession on local emissions seems to be more complicated and drawing conclusions remains 581 tentative.Besides, in the same period,regional atmospheric pollution presents a decreasing tendency (Fig.