Long-Term Ground-Based Measurements of Aerosol Optical Depth over Kuwait City

: We analyze ten years (2008–2017) of ground-based observations of the Aerosol Optical Depth (AOD) in the atmosphere of Kuwait City, in Middle East. The measurements were conducted with a CIMEL sun-sky photometer, at various wavelengths. The daily average AOD at 500 nm (AOD 500 ) is 0.45, while the mean Ångström coefﬁcient (AE), calculated from the pair of wavelengths 440 and 870 nm, is 0.61. The observed high AOD 500 values (0.75–2.91), are due to regional sand and dust storm events, which are affecting Kuwait with a mean annual frequency of almost 20 days/year. The long-term record analysis of AOD 500 and AE, shows a downward and upward tendency respectively, something which could be attributed to the continuous expansion and industrialization of the main city of Kuwait, in combination with the simultaneous increase of soil moisture over the area. By utilizing back trajectories of air masses for up to 4 days, we assessed the inﬂuence of various regions to the aerosol load over Kuwait. The high aerosol loads during spring, are attributed to the dominance of coarse particles from Saudi Arabia (AOD 500 0.56–0.74), a source area that contributes the 56% to the mean annual AOD 500 . Other dust sources affecting signiﬁcantly Kuwait originated from the regions of Iraq and Iran with contribution of 21%.


Introduction
It has been already identified and commonly accepted, that aerosols are holding a very important role to the Earth's climate (e.g., [1]). Despite the numerous scientific efforts during the last decades towards our better understanding of aerosol effects on climate, their large spatial and temporal variability, heterogeneity and spectral varying behavior still cause significant uncertainties at global scales [1]. At regional scales, aerosols can affect the climate via radiation forcing and interaction with cloud but also degrade air quality, visibility and public health. Especially, interest has been raised seasonal variability are presented in Section 3.1. In Section 3.2, we utilize cluster analysis in order to identify the major source regions of particles and the typical paths of their transport towards the northern edge of Eastern Arabia, in Kuwait. Finally, our conclusions are provided in Section 4.

CIMEL Sun-Sky Radiometer
The sun-sky radiometer is located on the roof of building 33 KH in the Physics Department of Kuwait University, in Khalidiyah campus (29.325 • N, 47.971 • E, elev.: 42 m, a.s.l.) and operates almost continuously since 2008 up to now. Khalidiyah is an urban area, approximately 4.8 km west from Kuwait City and 10 km away from the coastal line. Moreover, the site location appears to be around 25 km east of the desert areas of the state. The instrument is part of NASA's global sun photometric network AERONET [8], which includes more than 500 operational stations. The sun-sky photometer is capable of performing automatic measurements of the direct solar irradiance at the common wavelengths of 340, 380, 440, 500, 675, 870, 940, 1020 and 1640 nm and diffuse sky radiance at 440, 675, 870 and 1020 nm, respectively. These measurements are used afterwards to provide, both optical and microphysical aerosol optical properties in the atmospheric column [33,34]. Recently released Version 3 of AERONET direct product includes significant improvements for the direct sun measurements, related to the screening of clouds, the automated data quality assurance, the inclusion of data with higher air masses (up to 7, in contrast with v2 that is up 5) and additionally implements spectral temperature corrections based on laboratory measurements [35,36]. To benefit from the above innovations, in this study we used the level 1.5 version 3 AERONET data products, related to the columnar aerosol optical depth (AOD) and the Ångström exponent (AE), at various wavelengths.

HYSPLIT Model and Cluster Analysis
The climatological partitioning of air mass trajectories arriving over Kuwait is analyzed with HYSPLIT Lagrangian trajectory model [37]. The model is set up to provide 96 h (4 days) daily back-trajectories for a 10 year period (2008-2017) and for 6 arriving heights (0.5, 1.5, 2.5, 3.5, 5.5, 7.5 km), driven by GDAS (Global Data Assimilation System) reanalysis data, at a resolution of 1 • × 1 • . Cluster analysis of the 10-year back-trajectories is used to identify the distinctive directional clusters for the arriving air masses. In general, certain limitations on the use of coarse meteorological forcing need to be taken into account for a Lagrangian study (for example the inaccurate representation of convective motions). However, the use of a 10-year-long dataset in this study increases our confidence on the statistical robustness of our results, although systematical errors and limitations of this approach should always be considered when interpreting the results. For the cluster analysis, we used only the trajectories that approach the surface (i.e., include heights below 1.5 km along their paths). These trajectories imply some Planetary Boundary Layer (PBL) residence time for the corresponding air masses and thus are eligible of transporting pollutants towards Kuwait.

MODIS AOD Dataset
In order to cross check our climatological values and trends of aerosol optical properties, we used also the daily level-2 collection 6.1 MODIS/Terra AOD at 550 nm (AOD 550 ), for the time period 2008-2017. The Terra satellite is sun-synchronous and near polar-orbiting, with a circular orbit of 705 km above sea level, capable of scanning 36 spectral bands across a 2330 km wide swath. Daily overpass data for the specific area were extracted at a spatial resolution of 50 × 50 km 2 . Narrower area selection will create data gaps due to exact overpass location and previous studies have shown that such spatial resolution provide the optimal viewing for aerosol climatology (e.g., [38]). The overpass time is 10:20 UTC ± 45 min for Kuwait City.

ESA CCI Soil Moisture Dataset
The Soil Moisture CCI project is part of the ESA Program on Global Monitoring of Essential Climate Variables (ECV) and provides a multidecadal database of satellite observations [39][40][41]. For building this database, data products from multiple microwave instruments have been used. These data products were generated using different retrieving techniques and have different spatial and temporal resolutions. The data records are merged on a daily basis by applying a state of the art quality control [39]. For the present study, we have used the combined (passive and active sensors) product, averaged in the greater Kuwait area (28)(29)(30) • N, 46-48 • E) over the 2008-2016 period.

AOD Levels and Seasonal Variation
The aerosol optical properties used in this study, namely the AOD at 500 nm (AOD 500 ) and the Ångström exponent derived from the pair of wavelengths 440 and 870 nm (AE), cover the time period from January 2008 to December 2017 ( Figure 1). During this time span, measurements are available on >50% of the days. The observed time gaps are due to system malfunction and/or calibration and upgrade issues (~33%), while the remaining~17% can be attributed to the existence of clouds or rain. temporal resolutions. The data records are merged on a daily basis by applying a state of the art quality control [39]. For the present study, we have used the combined (passive and active sensors) product, averaged in the greater Kuwait area (28-30°N, 46-48°E) over the 2008-2016 period.

AOD Levels and Seasonal Variation
The aerosol optical properties used in this study, namely the AOD at 500 nm (AOD500) and the Ångström exponent derived from the pair of wavelengths 440 and 870 nm (AE), cover the time period from January 2008 to December 2017 ( Figure 1). During this time span, measurements are available on >50% of the days. The observed time gaps are due to system malfunction and/or calibration and upgrade issues (~33%), while the remaining ~17% can be attributed to the existence of clouds or rain.
During the aforementioned time period, various spikes in AOD500 can be observed (Figure 1), which are associated with local sources of particles or strong aerosol loads transported from far away sources. The same pattern is also clear for AE, with the majority of the observations to be around the median value of 0.57, indicating the aerosol coarse mode dominance in the atmospheric column over Kuwait City (Figure 2). The mean daily AOD500 found to be 0.45 ± 0.29, with 50% of the values lying in the range bin 0.15-0.58 and modal AOD interval of 0.34-0.40. The mean daily AE found to be 0.61 ± 0.39, with 50% of the values lying in the range 0.14-0.91 and with 0.12-0.20 representing the modal AE. The basic statistics (i.e., minimum, maximum, 1st quartile-Q1, 3rd quartile-Q3 and median) for all wavelengths are given in Table 1. Some extreme values of AOD500 that can be observed at the tail of the distribution in Figure 2a and reported also in Table 1 (max. AOD500 = 2.91), are due to regional severe dust storms (e.g., [42,43]). These events, account for most of ~65% of the day-to-day variability, observed for AOD500.  During the aforementioned time period, various spikes in AOD 500 can be observed (Figure 1), which are associated with local sources of particles or strong aerosol loads transported from far away sources. The same pattern is also clear for AE, with the majority of the observations to be around the median value of 0.57, indicating the aerosol coarse mode dominance in the atmospheric column over Kuwait City (Figure 2). The mean daily AOD 500 found to be 0.45 ± 0.29, with 50% of the values lying in the range bin 0.15-0.58 and modal AOD interval of 0.34-0.40. The mean daily AE found to Remote Sens. 2018, 10, 1807 5 of 16 be 0.61 ± 0.39, with 50% of the values lying in the range 0.14-0.91 and with 0.12-0.20 representing the modal AE. The basic statistics (i.e., minimum, maximum, 1st quartile-Q1, 3rd quartile-Q3 and median) for all wavelengths are given in Table 1. Some extreme values of AOD 500 that can be observed at the tail of the distribution in Figure 2a and reported also in Table 1 (max. AOD 500 = 2.91), are due to regional severe dust storms (e.g., [42,43]). These events, account for most of~65% of the day-to-day variability, observed for AOD 500 .  Sand and dust storms (SDS) seem to occur mainly during the transitional monsoon seasons of spring and fall, defined as when the daily AOD500 is one standard deviation above the mean value of its distribution (i.e., AOD500 > 0.75) and AE is less than 0.5. Spring (post-Northeast monsoon season) and fall (post-Southwest monsoon season) seasons seems to be mostly affected by SDS. For these events, mean AOD500 and AE values found to be 1.04 ± 0.32 and 0.13 ± 0.01, respectively. May seems to be the month in which SDS events can be observed with the highest frequency, while it seems almost 4-5 times less probable for such events to occur during the fall months ( Figure 3). During winter and summer months (monsoon periods), SDS appears with the least frequency, with some exceptional cases however present during February and June. For the entire time period of this study, Kuwait is found to be affected by SDS with a mean annual frequency of 19.4 days/year, ranging from 35 days/year (in 2008) to 13 days/year (in 2017), in agreement with recent studies performed with satellite images and in-situ measurements (i.e., dust traps) [44]. The lasting time period of these events, varies from 1 up to 5 consequent days, with the latter appearing during the months of May and June.
Regarding the annual variability of the total aerosol load, maximum AODs are found at spring with a peak in May and a monthly average value of 0.63, in coincidence with minimum AE of 0.28 (cf. Figure 4). The most intense loads at that month seems to have happened in 2012, while during  Sand and dust storms (SDS) seem to occur mainly during the transitional monsoon seasons of spring and fall, defined as when the daily AOD 500 is one standard deviation above the mean value of its distribution (i.e., AOD 500 > 0.75) and AE is less than 0.5. Spring (post-Northeast monsoon season) and fall (post-Southwest monsoon season) seasons seems to be mostly affected by SDS. For these events, mean AOD 500 and AE values found to be 1.04 ± 0.32 and 0.13 ± 0.01, respectively. May seems to be the month in which SDS events can be observed with the highest frequency, while it seems almost 4-5 times less probable for such events to occur during the fall months ( Figure 3). During winter and summer months (monsoon periods), SDS appears with the least frequency, with some exceptional cases however present during February and June. For the entire time period of this study, Kuwait is found to be affected by SDS with a mean annual frequency of 19.4 days/year, ranging from 35 days/year (in 2008) to 13 days/year (in 2017), in agreement with recent studies performed with satellite images and in-situ measurements (i.e., dust traps) [44]. The lasting time period of these events, varies from 1 up to 5 consequent days, with the latter appearing during the months of May and June.
MODIS combined Dark Target and Deep Blue product averaged within 50 × 50 km 2 , for the under study time period. Throughout the year the MODIS data seems to deviate up to 15% with respect to the ground based observations. Minimum deviation appears in Autumn (~−1%), while in late winter and spring MODIS may overestimate AOD values by 30% on average. As a consequence, the main seasonal pattern is reproduced by MODIS, namely the spring maximum; however, spring levels are relatively enhanced.   Regarding the annual variability of the total aerosol load, maximum AODs are found at spring with a peak in May and a monthly average value of 0.63, in coincidence with minimum AE of 0.28 (cf. Figure 4). The most intense loads at that month seems to have happened in 2012, while during the next years (2012-2017), lower mean monthly aerosol loads are observed. Minimum AODs are found in the period of December to January (0.26-0.29), with moderate values of AE (~0.92), while, during the rest of the period, AODs vary around the average value. The maximum level of AE (0.96) is observed in December. The same annual pattern seems to be followed also by AOD 550 , obtained by MODIS combined Dark Target and Deep Blue product averaged within 50 × 50 km 2 , for the under study time period. Throughout the year the MODIS data seems to deviate up to 15% with respect to the ground based observations. Minimum deviation appears in Autumn (~−1%), while in late winter and spring MODIS may overestimate AOD values by 30% on average. As a consequence, the main seasonal pattern is reproduced by MODIS, namely the spring maximum; however, spring levels are relatively enhanced.
At Figure 5, monthly averaged values of the AERONET level 1.5 products are provided. Level 1.5 inversions data are preferred for climatological studies, even having larger uncertainty, because the quality control for level 2.0, filters out data with Solar Zenith Angles less than 50 • and AOD at 440 nm lower than 0.4, leaving non-significant amount of dataset for the calculations [42]. Single Scattering Albedo (SSA) spectral variation through the year, should be consider as an indication for the type of aerosols. Although SSA at 670 nm has always larger values, during winter months it is almost identical to the one at 440 nm, while it is significant lower at larger wavelength. In contrast, for the period from May to October (with the exception of August) this pattern is reversed. Moosmuller et al. [45] linked the iron concentration of mineral dust aerosols to significant lower SSA at 405 nm than 870, which indicates a dominance of these particles during these months. While aerosols associated with urban pollution clearly increase SSA with wavelength [46]. The mixed picture mainly recorded at August should be linked to polluted dust mixture. Size distribution monthly average plot demonstrates significant larger coarse mode around spring-summer months (April-August) which should be linked to higher dust loads. The fine mode has relatively higher concentrations during winter months, where the presence of dust particles is limited and the contribution of urban particles to the total aerosol load is increased.   At Figure 5, monthly averaged values of the AERONET level 1.5 products are provided. Level 1.5 inversions data are preferred for climatological studies, even having larger uncertainty, because the quality control for level 2.0, filters out data with Solar Zenith Angles less than 50° and AOD at 440 nm lower than 0.4, leaving non-significant amount of dataset for the calculations [42]. Single Scattering Albedo (SSA) spectral variation through the year, should be consider as an indication for the type of aerosols. Although SSA at 670 nm has always larger values, during winter months it is almost identical to the one at 440 nm, while it is significant lower at larger wavelength. In contrast, for the period from May to October (with the exception of August) this pattern is reversed. Moosmuller et al. [45] linked the iron concentration of mineral dust aerosols to significant lower SSA at 405 nm than 870, which indicates a dominance of these particles during these months. While aerosols associated with urban pollution clearly increase SSA with wavelength [46]. The mixed picture mainly recorded at August should be linked to polluted dust mixture. Size distribution monthly average plot demonstrates significant larger coarse mode around spring-summer months (April-August) which should be linked to higher dust loads. The fine mode has relatively higher concentrations during winter months, where the presence of dust particles is limited and the contribution of urban particles to the total aerosol load is increased.  Figure 6 shows the percentage deviation of the monthly mean values of AOD500 and AE (obtained by AERONET) and AOD500 (obtained by MODIS), by excluding the seasonal influence on the recorded dataset. More precisely, in Figure 6a, a negative trend of the order of 3.3 ± 0.7% per year is observed in terms of the aerosol load, as estimated from the AERONET data (black dashed line). Similar trend 2.9 ± 0.9% is also observed for the AOD550 MODIS dataset (green dashed line in Figure  6a). Both of them have passed statistical significance testing with null hypothesis of an approximate Student's distribution with 5% significance (p < 0.05). In turn, a positive change of the order of 4.8 ± 0.1%, is depicted in AE (Figure 6b), which indicates that over the years, the size range of the particles observed over the atmospheric column, have been shifted to lower values.  Figure 6 shows the percentage deviation of the monthly mean values of AOD 500 and AE (obtained by AERONET) and AOD 500 (obtained by MODIS), by excluding the seasonal influence on the recorded dataset. More precisely, in Figure 6a, a negative trend of the order of 3.3 ± 0.7% per year is observed in terms of the aerosol load, as estimated from the AERONET data (black dashed line). Similar trend 2.9 ± 0.9% is also observed for the AOD 550 MODIS dataset (green dashed line in Figure 6a). Both of them have passed statistical significance testing with null hypothesis of an approximate Student's distribution with 5% significance (p < 0.05). In turn, a positive change of the order of 4.8 ± 0.1%, is The aforementioned trends can be attributed to the urbanization -industrialization of the main city of Kuwait and to a decrease of dust emission from nearby deserts. The land cover maps presented in Figure 7, were generated using Landsat Image (2006; 15 m resolution) and Spot Image (2010; 5 m resolution). Image Classification and field verification were used to identify the classes showed in Figure 7. As presented here, the land usage of the main city has changed and the more pronounced differences between the years of 2006 and 2010, seems to be related to the degradation of the limited greenery in the built-up areas, as well as to the change of the bare land to greenery-grassland, outside the main built-up area. Moreover, during the last decade Kuwait underwent a rapid urbanization since the population almost doubled, reaching the 4.2 million with a density of 235.56 inhabitants/km 2 . For example, in 2007, due to the discovery of new oil reserves, Kuwait had the third highest net migration rate in the world at 16.01%, which led to a large and growing immigrant population (http://worldpopulationreview.com/countries/kuwait-population/). Since then, the state has a mean positive growth rate of ~6.5% (http://countrymeters.info/en/Kuwait), leading to an increase of anthropogenic activity and land usage changes, the two main factors considered to be responsible for the observed positive trend of the AE exponent.  The aforementioned trends can be attributed to the urbanization -industrialization of the main city of Kuwait and to a decrease of dust emission from nearby deserts. The land cover maps presented in Figure 7, were generated using Landsat Image (2006; 15 m resolution) and Spot Image (2010; 5 m resolution). Image Classification and field verification were used to identify the classes showed in Figure 7. As presented here, the land usage of the main city has changed and the more pronounced differences between the years of 2006 and 2010, seems to be related to the degradation of the limited greenery in the built-up areas, as well as to the change of the bare land to greenery-grassland, outside the main built-up area. Moreover, during the last decade Kuwait underwent a rapid urbanization since the population almost doubled, reaching the 4.2 million with a density of 235.56 inhabitants/km 2 . For example, in 2007, due to the discovery of new oil reserves, Kuwait had the third highest net migration rate in the world at 16.01%, which led to a large and growing immigrant population (http://worldpopulationreview.com/countries/kuwait-population/). Since then, the state has a mean positive growth rate of~6.5% (http://countrymeters.info/en/Kuwait), leading to an increase of anthropogenic activity and land usage changes, the two main factors considered to be responsible for the observed positive trend of the AE exponent.
Soil-surface conditions (i.e., soil moisture) and surface wind speeds have generally been regarded as the main factors for determining the mobilization of sand and dust in dust-source regions (e.g., [47][48][49][50][51]). The low soil moisture level, seems to be responsible for relaxing the cohesive forces among the particles, leading thus to easiest lift and transportation of coarse dust particles in the atmosphere. The observed high aerosol load, especially during spring season, along with the low AE values, can be attributed to the low soil moisture values that are prevailing in the desert area but also to the observed trends.
For verifying this, monthly de-seasonalized variation of soil moisture in the greater Kuwait is presented in Figure 8, which show an increase trend over the more recent years. In order to make these data comparable an additional filter have been applied, that each month should have at least 25 days with more than 90% quality controlled data points in the area. Only one month has been filtered out with this procedure. In particular, an augmenting trend of 1.9% per year is recorded, which had passed a statistically significant Student's t test for confidence at 95%. This increase is a factor that weakens dust emissions in the greater area (as shown at Kingmüller et al. [47]), which is linked to the recorded decrease in AOD, especially in coarse mode. Although, soil moisture is only one of the factors affecting the aerosol load, in the dust aerosol dominated area of study, a significant change in the soil conditions could explain an important part of the variation. Especially the very large positive variations in soil moisture, ongoing from summer 2015, are correlated to some of the lowest AODs of the period at both AERONET and MODIS databases. Figure 7. As presented here, the land usage of the main city has changed and the more pronounced differences between the years of 2006 and 2010, seems to be related to the degradation of the limited greenery in the built-up areas, as well as to the change of the bare land to greenery-grassland, outside the main built-up area. Moreover, during the last decade Kuwait underwent a rapid urbanization since the population almost doubled, reaching the 4.2 million with a density of 235.56 inhabitants/km 2 . For example, in 2007, due to the discovery of new oil reserves, Kuwait had the third highest net migration rate in the world at 16.01%, which led to a large and growing immigrant population (http://worldpopulationreview.com/countries/kuwait-population/). Since then, the state has a mean positive growth rate of ~6.5% (http://countrymeters.info/en/Kuwait), leading to an increase of anthropogenic activity and land usage changes, the two main factors considered to be responsible for the observed positive trend of the AE exponent.  Soil-surface conditions (i.e., soil moisture) and surface wind speeds have generally been regarded as the main factors for determining the mobilization of sand and dust in dust-source regions (e.g., [47][48][49][50][51]). The low soil moisture level, seems to be responsible for relaxing the cohesive forces among the particles, leading thus to easiest lift and transportation of coarse dust particles in the atmosphere. The observed high aerosol load, especially during spring season, along with the low AE values, can be attributed to the low soil moisture values that are prevailing in the desert area but also to the observed trends.
For verifying this, monthly de-seasonalized variation of soil moisture in the greater Kuwait is presented in Figure 8, which show an increase trend over the more recent years. In order to make these data comparable an additional filter have been applied, that each month should have at least 25 days with more than 90% quality controlled data points in the area. Only one month has been filtered out with this procedure. In particular, an augmenting trend of 1.9% per year is recorded, which had passed a statistically significant Student's t test for confidence at 95%. This increase is a factor that weakens dust emissions in the greater area (as shown at Kingmüller et al. [47]), which is linked to the recorded decrease in AOD, especially in coarse mode. Although, soil moisture is only one of the factors affecting the aerosol load, in the dust aerosol dominated area of study, a significant change in the soil conditions could explain an important part of the variation. Especially the very large positive variations in soil moisture, ongoing from summer 2015, are correlated to some of the lowest AODs of the period at both AERONET and MODIS databases.

Origin of Air masses Over Kuwait
The high columnar AOD values observed in Kuwait can be attributed to: (i) the anthropogenic activity and the local sources linked to the aerosol load inside the boundary layer and (ii) apparently to the dominance of dust particles in the atmospheric column. The latter can be further categorized to dust particles that have been transported through high altitudes from transnational regions and to particles circulated within the boundary layer, originating from local arid and semi-arid sources.
In order to investigate any possible relation between the measured AOD levels and the origin of the air masses arriving over the station, we used HYSPLIT model to generate four day back trajectories for air-parcels arriving at 12:00 UTC over Kuwait at 0.5, 1.5, 2.5 3.5, 5.5 and 7.5 km a.s.l. The HYSPLIT runs were performed for every single day with available AERONET retrievals. These trajectories were further filtered to select only the trajectories approaching the surface and categorized in six major clusters. For each cluster, the corresponding centroid has been determined

Origin of Air masses Over Kuwait
The high columnar AOD values observed in Kuwait can be attributed to: (i) the anthropogenic activity and the local sources linked to the aerosol load inside the boundary layer and (ii) apparently to the dominance of dust particles in the atmospheric column. The latter can be further categorized to dust particles that have been transported through high altitudes from transnational regions and to particles circulated within the boundary layer, originating from local arid and semi-arid sources.
In order to investigate any possible relation between the measured AOD levels and the origin of the air masses arriving over the station, we used HYSPLIT model to generate four day back trajectories for air-parcels arriving at 12:00 UTC over Kuwait at 0.5, 1.5, 2.5 3.5, 5.5 and 7.5 km a.s.l. The HYSPLIT Remote Sens. 2018, 10, 1807 10 of 16 runs were performed for every single day with available AERONET retrievals. These trajectories were further filtered to select only the trajectories approaching the surface and categorized in six major clusters. For each cluster, the corresponding centroid has been determined and drawn. The centroids represent the mean transport path of the air masses arriving over Kuwait (Figure 9). Remote Sens. 2018, 10, x FOR PEER REVIEW 10 of 16 and drawn. The centroids represent the mean transport path of the air masses arriving over Kuwait (Figure 9). The northeast monsoon is characterized by a northeasterly low level flow resulting mainly from the combination of a high pressure system at Siberia with a deep Australian low. The onset of the summer southwest monsoon is identified by the northward shift of the Intertropical Convergence Zone (ITCZ) and the development of the cross-equatorial Somali Jet that enforces the low level southwest winds from the east coast of Africa towards India [52].
The weakening of the northeasterly winter monsoon during the transition spring season is accompanied by the development of low-level thermal heat lows and the establishment of cyclonic flow patterns over the Arabian Peninsula. This circulation type is represented by CL1 28% and CL2 19% in (Figure 9). CL1 represents short range trajectories from the central and east parts of Saudi Arabia while CL2 represents trajectories originating from the northwest parts of Saudi Arabia, Red Sea and Egypt. These pathways imply the mobilization of dust from Arabian Desert sources towards Kuwait and result in high mean AOD500 of 0.55 ±0.02 (Table 2) mostly during March-June ( Figure  10a). Moreover, during May, air masses arriving from the Mediterranean coast of Africa over Middle East (Lebanon, Syria, Iraq) (CL3), contribute also to high AOD (AOD500 = 0.47 ± 0.03). The maximum frequency of occurrence for CL1, CL2 and CL3 appears to be in the spring transition period (May) with values of 18.2%, 28.1% and 22.7%, respectively (Figure 10b). The northeast monsoon is characterized by a northeasterly low level flow resulting mainly from the combination of a high pressure system at Siberia with a deep Australian low. The onset of the summer southwest monsoon is identified by the northward shift of the Intertropical Convergence Zone (ITCZ) and the development of the cross-equatorial Somali Jet that enforces the low level southwest winds from the east coast of Africa towards India [52].
The weakening of the northeasterly winter monsoon during the transition spring season is accompanied by the development of low-level thermal heat lows and the establishment of cyclonic flow patterns over the Arabian Peninsula. This circulation type is represented by CL1 28% and CL2 19% in (Figure 9). CL1 represents short range trajectories from the central and east parts of Saudi Arabia while CL2 represents trajectories originating from the northwest parts of Saudi Arabia, Red Sea and Egypt. These pathways imply the mobilization of dust from Arabian Desert sources towards Kuwait and result in high mean AOD 500 of 0.55 ±0.02 (Table 2) mostly during March-June (Figure 10a). Moreover, during May, air masses arriving from the Mediterranean coast of Africa over Middle East (Lebanon, Syria, Iraq) (CL3), contribute also to high AOD (AOD 500 = 0.47 ± 0.03). The maximum frequency of occurrence for CL1, CL2 and CL3 appears to be in the spring transition period (May) with values of 18.2%, 28.1% and 22.7%, respectively (Figure 10b).  The north sector clusters CL4, CL5 and CL6 are associated with the northerly flow throughout the year and most profound during the summer monsoon period (May-August) and early fall transition season (September) as seen by their frequency of occurrence in (Figure 10b). The mean AOD500 for CL4 is 0.32 ± 0.02, for CL5 0.25 ± 0.02 and for CL6 0.40 ± 0.02 (Table 2). CL4 and CL6 have similar centroids but are separated because they represent different scale transports. CL6 represents short range transport from north sector (mainly Iraq and Iran) while CL4 represents longer range transport from NE sector including Turkey and Eastern Mediterranean. In addition, the pathway of CL4 is more profound during summer months due to the seasonal persistent north winds over the Aegean Sea (Etesian winds). These clusters represent the north Shamal winds that mobilize dust from the nearby arid regions of Iraq and Iran towards Kuwait, with the shorter CL6 cluster being more typical of the channeling of Shamal flow towards the Gulf. The air masses originating from the northeastern Mediterranean and Turkey (CL4) (Figure 9) are also related with the transport of anthropogenic particles from the aforementioned region, linked with the lower AOD500 values (0.32 ± 0.02) ( Table 2). The lowest aerosol load (0.25 ± 0.02) is observed for air masses arriving from CL5, which accounts for 2% of the back-trajectories (Table 2) and maximum frequencies during January (38.4%) and December (30.8%) (Figure 10b). Overall, it seems that short trajectories (i.e., CL1, CL2, CL3 and CL6) are related with relatively high AOD values while the long trajectories (i.e., CL4 and CL5) are related with lower AOD observation, with respect to the overall statistically mean value.  The north sector clusters CL4, CL5 and CL6 are associated with the northerly flow throughout the year and most profound during the summer monsoon period (May-August) and early fall transition season (September) as seen by their frequency of occurrence in (Figure 10b). The mean AOD 500 for CL4 is 0.32 ± 0.02, for CL5 0.25 ± 0.02 and for CL6 0.40 ± 0.02 (Table 2). CL4 and CL6 have similar centroids but are separated because they represent different scale transports. CL6 represents short range transport from north sector (mainly Iraq and Iran) while CL4 represents longer range transport from NE sector including Turkey and Eastern Mediterranean. In addition, the pathway of CL4 is more profound during summer months due to the seasonal persistent north winds over the Aegean Sea (Etesian winds). These clusters represent the north Shamal winds that mobilize dust from the nearby arid regions of Iraq and Iran towards Kuwait, with the shorter CL6 cluster being more typical of the channeling of Shamal flow towards the Gulf. The air masses originating from the northeastern Mediterranean and Turkey (CL4) (Figure 9) are also related with the transport of anthropogenic particles from the aforementioned region, linked with the lower AOD 500 values (0.32 ± 0.02) ( Table 2). The lowest aerosol load (0.25 ± 0.02) is observed for air masses arriving from CL5, which accounts for 2% of the back-trajectories (Table 2) and maximum frequencies during January (38.4%) and December (30.8%) (Figure 10b). Overall, it seems that short trajectories (i.e., CL1, CL2, CL3 and CL6) are related with relatively high AOD values while the long trajectories (i.e., CL4 and CL5) are related with lower AOD observation, with respect to the overall statistically mean value.
The presented air mass cluster analysis over the study area can be further linked with a potential aerosol source type (i.e., dust, natural and anthropogenic, etc.). This approach has been already applied for other regions (e.g., [53,54]) and proved to be a great tool for estimating the contribution of each distinct cluster and consequently of each source type, to the mean AOD 500 value. Considering both the percentage appearance of each cluster and the respective mean AOD 500 (Table 2), from the direction and the characteristics of the air mass trajectories, we found that the greatest contribution comes from regional and local dust sources. This is shown in Figure 11, where dust particles from Saudi Arabia, accounts for 33% of the average AOD during the whole period of study. It is worth to mention here that including only the trajectories with PBL residence time in our analysis increases the accuracy of the presented source-receptor apportionment. Another important contribution (23%) is dust from western part of Arabian Peninsula transported over short distances, while regional dust sources located in Iran and Iraq contributes 21%. Since it is not to possible to discriminate the natural form anthropogenic contribution linked with CL3 and CL4, these clusters can be merged into one category representing the mixture of dust with anthropogenic ("Polluted dust"). Therefore, air masses of long-range transport from the Mediterranean coast of the African continent along with Turkey and Middle East, contribute another 22%. Finally, only the 1% of the statistically mean AOD 500 can be attributed to remote, clean areas. Saudi Arabia, accounts for 33% of the average AOD during the whole period of study. It is worth to mention here that including only the trajectories with PBL residence time in our analysis increases the accuracy of the presented source-receptor apportionment. Another important contribution (23%) is dust from western part of Arabian Peninsula transported over short distances, while regional dust sources located in Iran and Iraq contributes 21%. Since it is not to possible to discriminate the natural form anthropogenic contribution linked with CL3 and CL4, these clusters can be merged into one category representing the mixture of dust with anthropogenic ("Polluted dust"). Therefore, air masses of long-range transport from the Mediterranean coast of the African continent along with Turkey and Middle East, contribute another 22%. Finally, only the 1% of the statistically mean AOD500 can be attributed to remote, clean areas. Figure 11. Contribution of the discrete source areas to the statistically mean value of AOD500.

Conclusions
The statistical values of the Aerosol Optical Depth (AOD) over Kuwait, were analyzed and presented in this work, as revealed from ten years of observations (2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017). This is the first long term, ground-based dataset presented for the area. The main conclusions of the study can be summarized as follows:

•
Mean daily AOD500 over Kuwait is 0.45 ± 0.29 and the corresponding value of Ångström coefficient (AE 44/870), is 0.61 ± 0.39. These values are indicative of the strong sources of particulate matter in the area, which contribute to the degradation of air quality at the regional scale.

•
Recorded high AOD500 values (0.74-2.91), are due to regional sand and dust storm events, which are affecting Kuwait with a mean annual frequency of almost 20 days/year. Ground-based retrieved AOD is compared against MODIS retrievals obtained over the area for the same time period. On an annual basis, MODIS found to have a mean overestimation of AOD up to 15%.

•
The long-term analysis of AOD500 illustrated a negative trend percentage, of 3.3 ± 0.7% per year. The corresponding trend analysis of AE observations shows a positive change of 4.8 ± 0.1%. Figure 11. Contribution of the discrete source areas to the statistically mean value of AOD 500 .

Conclusions
The statistical values of the Aerosol Optical Depth (AOD) over Kuwait, were analyzed and presented in this work, as revealed from ten years of observations (2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017). This is the first long term, ground-based dataset presented for the area. The main conclusions of the study can be summarized as follows: • Mean daily AOD 500 over Kuwait is 0.45 ± 0.29 and the corresponding value of Ångström coefficient (AE 44/870), is 0.61 ± 0.39. These values are indicative of the strong sources of particulate matter in the area, which contribute to the degradation of air quality at the regional scale.

•
Recorded high AOD 500 values (0.74-2.91), are due to regional sand and dust storm events, which are affecting Kuwait with a mean annual frequency of almost 20 days/year.  Ground-based retrieved AOD is compared against MODIS retrievals obtained over the area for the same time period. On an annual basis, MODIS found to have a mean overestimation of AOD up to 15%.

•
The long-term analysis of AOD 500 illustrated a negative trend percentage, of 3.3 ± 0.7% per year. The corresponding trend analysis of AE observations shows a positive change of 4.8 ± 0.1%.

•
The aforementioned trends and annual cycles of AOD 500 and AE, are strongly related to the soil moisture conditions over the desert area of Kuwait and land cover change and increased anthropogenic activity of the city during the last decade. • The following general points are concluded from the trajectory cluster analysis: (i) The greatest contribution (almost 80%) to the annually averaged AOD, comes from regional and local dust sources (56% from Saudi Arabia and 21% from Iran-Iraq). (ii) Natural and anthropogenic particles are transported from the northeastern Mediterranean and African coast mixed with dust from local regions, accounting for 22% of the mean annual aerosol load.
This analysis intended to provide data on the climatological aspects of columnar aerosol loadings over an arid environment in the northern part of Middle East, based on ground measurements. Such data sets are essential for the validation of satellite products obtained from passive sensors, which can then be used for long term climatic studies. However, for the detailed identification and aerosol characterization, range resolved multi-wavelength and multi-depolarization lidar measurements are needed in the area. With such information, it will then be feasible to estimate the aerosol optical and microphysical properties of each discrete atmospheric layer, assessing and discriminating the natural from the anthropogenic contribution on the total atmospheric load.