Influence of air mass history on the columnar aerosol properties at Valencia, Spain

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 112,, doi: /2007jd008593, 2007 Influence of air mass history on the columnar aerosol properties at Valencia, Spain Víctor Estellés, 1 José A. Martínez-Lozano, 1 and María P. Utrillas 1 Received 27 February 2007; revised 4 May 2007; accepted 22 May 2007; published 15 August [1] The physical and radiative properties of atmospheric aerosols have been obtained in Valencia (latitude , longitude 0.418, 60 m a. s. l.), a city of the Spanish Mediterranean coast, by the inversion of direct solar irradiance and diffuse sky irradiance measurements made with a CIMEL CE318 system, from January 2002 to July The data acquired by the CE318 were used to determine the instantaneous values of the aerosol optical depth (AOD), the columnar water vapor content (w) and the Ångström wavelength exponent (a). The SKYRAD code was used to obtain the size distribution, the asymmetry parameter, the complex refractive index and the single scattering albedo of the aerosols. By dividing the source regions into sectors, and with the help of a simple model that quantified the relative influence of each sector in the final character of the air masses arriving over the measurement site, it was possible to classify and understand the dependence of these physical and radiative properties on the dominant air mass type. Citation: Estellés, V., J. A. Martínez-Lozano, and M. P. Utrillas (2007), Influence of air mass history on the columnar aerosol properties at Valencia, Spain, J. Geophys. Res., 112,, doi: /2007jd Introduction [2] Knowledge of aerosol physical and chemical characteristics is fundamental for understanding their effect on the Earth s climate both at regional and global levels [Charlson and Heintzenberg, 1995]. Therefore it is of great importance to be able to characterize these properties by means of complementary techniques which will allow as much detail of the climate mosaic to be described as possible. In this context, the Mediterranean is one of the most interesting regions of the world since its atmosphere is subject to very varied influences of both local and remote origin. The specific region of our study, on the western extreme of the Mediterranean, is subject to the influence of particles brought from such distant regions as the Sahara, industrial continental Europe, and oceanic regions. The characteristics of the aerosols coming from each of these regions are very different, and confer highly variable temporal properties to the atmospheric aerosols measured in any Mediterranean region. [3] In this work we have tried to characterize the transport of the aerosols effected by air masses in movement across several source regions. For this we have used a modified classification of the classical definition of air masses [Barry and Chorley, 1998] based on thermodynamic properties which was used to create a simple model that could be used to relate and classify these properties as a function of the air mass origin. 1 Solar Radiation Research Unit, Department of Earth Physics and Thermodynamics, Universitat de València, Valencia, Spain. Copyright 2007 by the American Geophysical Union /07/2007JD [4] An air mass is traditionally defined as a volume of air whose physical properties, especially the temperature and humidity, remain relatively constant across areas of hundreds to thousands of square kilometers [Barry and Chorley, 1998]. Such physically homogeneous air masses acquire their properties by interaction with the surfaces above which they are found (source region), whenever they must remain stationary for a certain length of time. The source regions should also be sufficiently extensive and homogeneous in order to impart their characteristics to these air masses. [5] Our study was carried out near to the Spanish Mediterranean coast where the Atlantic Ocean is one of the most important source regions both because of its size and its proximity. The desert region of North Africa constitutes another first-order source region with its own marked characteristics. There are other source regions which, being more distant are less important such as the Arctic region as well as closer low-magnitude source regions, that impose their own characteristics on the air masses moving over them, such as the European continent and the Mediterranean Sea. The latter can be considered a relatively small maritime area located between two continental regions. Because of its relatively small size, the Mediterranean does not easily develop clear defined air masses, in spite of leaving a characteristic footprint over the primary air masses. 2. Instrumentation and Measurement Site [6] To characterize the aerosol properties we have used the sun photometric technique which gives results that are representative of the whole atmospheric column. The sun photometric measurements were made with a CIMEL 1of12

2 CE318 photometer designed for the automatic measurement of direct solar irradiance and sky radiance. It measures in the 440, 670, 870, 940 and 1020 nm nominal wavelengths. The bandwidth or Full Width at Half Maximum (FWHM) is about 10 nm. The sensor head is equipped with a 1.2 Field of View (FOV) double collimator. The most important error in the calculation of the aerosol optical properties is the uncertainty of the calibration. Given that the CE318 is really a sky-sun photometer and makes measurements of two different parameters, direct solar radiation and diffuse sky radiation, it needs two types of calibration. Since it is an instrument that is not included in the AERONET network, it is necessary to comment briefly on the method used to calibrate the photometer in order to evaluate the measurement uncertainty. However, we address the reader to a previous work [Estellés et al., 2007] where further details are given for this instrument calibration method and history. [7] The calibration of the direct component can be performed using the Langley plot technique or by transference from other instruments. Here we used the latter. To perform calibration transference it is necessary to have a master instrument which possesses a recent calibration of sufficient quality, preferably made using the Langley plot in optimal conditions. The secondary instrument obtains its calibration by comparison with the master, by making simultaneous measurements in clear sky conditions and under some stability limitations. The total calibration uncertainty for our instrument gets typically %, in the order of AERONET field instruments [Holben et al., 1998]. Moreover, the total calibration drift along our complete database was found to be +2.6, 0.4, 2.9, 5.8 and 0.4% for channels 440, 670, 870, 940 and 1020 nm, respectively. The coefficients were interpolated between every pair of consecutive calibrations for finding the calibration at any intermediate date. A new in situ calibration method (SKYIL) was also applied to this instrument [Campanelli et al., 2007]. Their results were very consistent with the transferred calibrations and let us monitor the evolution of the calibration with time. [8] For the calibration of the diffuse component it is necessary to have a known radiation source. In our case four different references have been used to calibrate the instrument: Li-Cor (#370), Optronic OL455 (# ) and Bentham SRS8 (#4884 and #7281). [9] The radiance calibration factor is obtained from: kðlþ ¼ E 0ðlÞ S 0 ðlþ being E 0 the radiance of the source producing the signal S 0 in the photometer for a given wavelength l. To find this radiance it is necessary to know the spectrum of the source E 0 (l) and the photometer s response r(l), determined by the transmissivity of the interferential filters t(l) and the sensitivity of the detector s(l). The expression that allows us to know E 0 is: E 0 ¼ R l2 l 1 R l2 EðlÞrðlÞdl l 1 rðlþdl ð1þ ð2þ where r(l) =t(l)s(l) and l 1, l 2 are two extreme values for which r(l) is null or negligible. During the radiance calibration sessions, the temperature coefficient for 870 and 1020 channels was also determined for this instrument [Gómez-Amo et al., 2006] and applied on the field data. [10] From January 2002 a CE318 photometer (serial number 176) has been maintained at the Burjassot measurement station. This is located in the Physics Faculty (latitude , longitude 0.418, 60 m above sea level), in the Burjassot campus of the University of Valencia. Burjassot is a town with 35,000 inhabitants located in the Valencian metropolitan area which has a total population of some 1,357,000 inhabitants. Given its proximity to the principal population nucleus, 5 km, the station falls under the direct influence of the urban and industrial pollution typical of the metropolitan area. Its closeness to the western coast of the Mediterranean Sea, 10 km, is also decisive for identifying the type of aerosols in this region. Agriculture is another important activity in the region, especially irrigated horticulture and citrus fruit, although some nonirrigated agriculture is also found further inland. [11] The data acquired by the CE318 up to July 2005 were used to determine the instantaneous values of the aerosol optical depth (AOD), the columnar water vapor content (w) and the Ångström wavelength exponent (a). The SKYRAD code (downloadable at OPENCLASTR website: was used to obtain the size distribution, the asymmetry parameter, the complex refractive index and the single scattering albedo, all using the methodology described by Estellés [2006] and Estellés et al. [2006] and briefly introduced later. The so retrieved instantaneous aerosol and water vapor parameters were hourly averaged for the whole database; the so created database was daily averaged; and finally the monthly and seasonal statistics were also obtained. These different temporal averaged data sets were previously analyzed by Estellés et al. [2007] for a climatological overview of the site characteristics. For the air mass dependence of the aerosol properties, the daily database was finally chosen. In this way, we retrieve the mean statistics for all the aerosol parameters and the air mass characterizing indices that are later crossed for the classification. As the air mass characteristics do not change much in few hours unless a front is involved, the daily averages are a compromise between accuracy and computation time. 3. Methodology 3.1. Air Mass Classification [12] Air masses can be classified, with respect to the source region, in terms of two basic parameters: temperature and surface type. The temperature allows differentiation between arctic (A), polar (P) and tropical (T) depending on the latitude where the air masses are originated. The artic masses are originated in the highest latitudes in very cold environments, the polar masses in high latitudes of Canada, northern Atlantic and northern Europe, and the tropical masses in warmer regions at lower latitudes, in the Atlantic ocean, Mediterranean sea or the African continent. On the other hand, the surface type of the region may be continental (c) or maritime (m) in case they have been 2of12

3 developed over continental or oceanic surfaces, respectively [Barry and Chorley, 1998]. [13] Two of the main sources of cold air in our hemisphere are the continental anticyclonic weather systems from northern Canada, of continental polar type (cp) and the Arctic Basin of continental Artic type (ca). These masses originate over huge extensions of snow, especially in winter, such that they are cold and dry and therefore very stable near the surface. The warm air masses originate in the subtropical high-pressure cells and the extensive continental zones, especially in summer. These tropical air sources (T) can be divided into maritime (mt), when they come from the Atlantic Ocean, and continental (ct), if their origin is the north of Africa. The mt air masses are characterized by high temperature and humidity, stability and low cloudiness, except when they are displaced toward higher latitudes, where they are cooled from below, especially in spring. Generally the Mediterranean, although a relatively warm and small sea, closed in between two important continental surfaces, can impose its own specific imprint on the polar air masses that stagnate above it. [14] Although the traditional definition of air mass does not take into account the properties of suspended particles, it is reasonable to assume, a priori, that the continued interaction between the surface of the source regions and the air masses will not only determine the thermodynamic characteristics of the air but will also determine the nature and concentration of its aerosols. In reality it is normal that air masses are not pure masses either because they have been mixed, homogeneously or heterogeneously, with air masses with other characteristics or because they have spent a certain amount of time moving across different source regions. [15] The paths followed by the air masses were given by their back trajectories at different levels, allowing a broad approximation of the regions with which the air masses had interacted. One of the most used models for calculating back trajectories is the Hybrid Single-Particle Lagrangian Integrated Trajectory Model (HYSPLIT) developed by NOAA [Draxler and Rolph, 2003]. This model combines a Langrangian approximation for resolving air mass transport with an Eulerian approximation for the diffusion of pollutants. [16] In order to establish possible errors originating from the use of this model, Harrys et al. [2005] studied the results for various configurations for three different locations over a whole year. They found five principal factors caused deviations in the back trajectories: (1) differences in the computational methodology (3 4%), (2) temporal interpolation (9 25%), (3) the vertical transport model (18 34%), (4) the meteorological database (30 40%), and (5) combined differences between the meteorological data and the vertical transport model (39 47%). These differences were calculated as the mean squared difference of the horizontal distances between points on the back trajectories. Such differences were obtained with a flight time of 96 hours. In our case, the flight time was extended to 120 hours. Evidently the errors accumulate and the greater time will lead to greater differences. Nevertheless, the majority of studies found in the bibliography use five day back trajectories as a compromise between accuracy and the need to reconstruct as completely as possible the average life cycle of aerosols in the atmosphere. [17] For each day three back trajectories were calculated simultaneously starting at different altitudes above the measurement station: (1) 500 m above sea level, well within the boundary layer where the greater part of the interactions affecting the aerosols occur, although these same interactions create the greatest uncertainty in the back trajectory; (2) 1500 m above sea level trying to represent the top of the nominal boundary layer; and finally (3) 3000 m above sea level, representing the free troposphere, where there are hardly any aerosols except in the case of intrusion by Saharan dust which are produced by movements at high altitudes above the boundary layer. HYSPLIT model also allows choosing between three models for the vertical velocity motion: isobaric, isentropic and model of vertical velocity (three-dimensional). Following Draxler [1996] and Stohl [1998] the three dimensional model is the preferable when an accurate vertical velocity data field (as the FNL database) is available. In any case, preliminary test runs showed no important differences in the obtained classification for most of days when using different models. Therefore HYSPLIT was set up with the model of vertical velocity for computing the vertical motion. [18] In some previous works back trajectories have been used as a tool for analyzing the origins of air masses at a specified moment and thus for understanding the origin of the particles that they carry [Grousset et al., 2003; Niemi et al., 2004; Estellés et al., 2004]. In a smaller number of cases the graphical back trajectories or synoptic maps are classified manually with the aim of understanding the role of transport processes in the local aerosol climatology [Birmili et al., 2001; Gerasopoulos et al., 2003; Slater and Dibb, 2004]. In our case we have looked at the problem with a different methodology. Since the HYSPLIT model allows the possibility to obtain the air mass trajectory (and some of the key meteorological parameters) quantitatively, a simple classification model was implemented that would permit the back trajectories to be described using simple basic indices, and to assign to them the primary and secondary characteristics of the air masses. By performing this classification automatically it would be possible to apply objective criteria and also to study the sensitivity of the classification to different input parameters of the HYSPLIT model. [19] Figure 1 shows the chosen sectors used for classifying the origins of the various back trajectories. These sectors were defined considering both the previous definitions as well as other factors related to the physics of aerosols. The sectors were as follows: [20] Sector EU defines continental polar air masses and includes the northern Mediterranean coast. Given its basically continental character it was expected that these air masses would transport mineral particles, originating in the soil, but its differential characteristic actually is the urban type aerosol load including smoke and soot such that the size distribution would possess a well developed accumulation mode and a relatively low single scattering albedo. [21] Sector AF is characterized by tropical continental air masses, with a significant load of mineral dust due to the Saharan desert and the low-pressure cells generated in summer by the intense sunshine, causing the incorporation of dust into high-altitude atmospheric layers from where 3of12

4 Figure 1. Principal sectors defined for the back trajectory classifications (divided by a solid line) and the secondary class O (in dashed line). EU, European; AF, African; TR, Tropical; PO, Polar; AR, Artic. they are transported to other latitudes. The air masses coming from this sector were expected to have high turbidity and a size distribution with a coarse mode dominating the accumulation mode. [22] In sector TR, the Atlantic Ocean was subdivided into three different regions. The first of these was the region of tropical maritime air masses located to the west of Africa. Because of the global easterly circulation below 30 the air masses were predicted to possess a certain mineral footprint due to aerosols transported from the desert toward the Americas before being partially deflected toward our region. The air was expected to be warm and moist. Turbidity was not expected to be high but nonetheless higher than for the other two Atlantic regions. [23] Sector PO was the region of the Atlantic located between average latitudes (between 30 and 60 ) which would have an intermediate character between the maritime air masses of the tropics and the Arctic. These were polar maritime air masses generated by the movement of continental air masses from north America perhaps with an occasional residue to anthropogenic aerosols. [24] Sector AR was defined by the back trajectories originating in Canada or in the Arctic basin, with associated air masses of type ma or mp, according to the exact place of origin. This sector also included air masses generated in the sea to the south of Iceland. Given the time taken to cross the ocean the continental memory would be very vague. It was expected that these would be the cleanest air masses. [25] To complement these definitions another regional class was defined (O type, drawn by a dashed line), characterized by those trajectories whose maximum distance traveled during the five days of the back trajectory was less than 600 km from the measurement station. This class was generally indicative of synoptic conditions of weak pressure gradients and thus stationary air masses. They did not possess any clear relationship with the other five cases, since they usually loitered around the arrival point, passing consecutively through the different sectors. It was expected that these air masses would have a relatively high turbidity due to their length of stay over the Iberian Peninsula, North Africa or the Mediterranean. This turbidity would also be increased by the accumulation of particles produced locally as there would be no adequate renewal of the air. During these episodes the effect of the sea breeze was particularly noticeable which, during the night or land phase drew the polluted atmosphere toward the sea, while during the day or marine phase returned them to the land. These phenomena increase the ageing of the soot aerosols and the growth of sulphate hygroscopic aerosols. [26] It may seem strange that the Mediterranean Sea was not included as an independent sector, especially since it is indeed a key factor in the renewal of the air masses on the coast and, in general, in the climatology of any station located on its coast. This decision was justified by two reasons. First, the situation eastward of the general circulation makes improbable the existence of many cases of Mediterranean air masses which affect the station. Secondly the small size of the Sea and its position between the European and African continental platforms makes the classification of a pure Mediterranean back trajectory very complicated. Preliminary studies carried out in order to optimize the choice of the most important sectors, identified the Mediterranean air masses as halfway between the AF and EU types, in relation to the aerosol properties. [27] To make an approximation of the character of a given back trajectory, a percentage index was defined z i, where i corresponded to a given sector: EU, AF, TR, PO or AR. These indices were obtained by dividing the number of hours spent by the air mass moving across each sector i (t i ) by the total travel time (T), being in this case 120 hours. Both time factors are weighted by two factors that account for the life time of the aerosol and the layer height from the ground. The definitions of the indices take the next form: z i ð% where the temporal sums are given by: ~t i ¼ X3 X N k¼1 j¼1 ~T ¼ X5 Þ ¼ ~ t i 100 ð3þ ~T wh kj wtkj Dti ð4þ X 3 X N i¼1 k¼1 j¼1 being w(h kj ) the height weighting factor given by a first exponential function: wh kj Dt i h kj ¼ exp and w(t kj ) the temporal weighting factor given by a second exponential function: wt kj S H t kj ¼ exp S T ð5þ ð6þ ð7þ 4of12

5 Figure 2. Comparative evolution of (top) the AF air mass index with (bottom) the instantaneous (dots) and daily (solid line) aerosol optical depth at 500 nm, during the month of August [28] In the equations above, ~T i is the weighted sum of the time spent by the three back trajectories in sector i, being Dt i the time step (one hour). On turn, ~T is the sum in hours, of the total time for the three back trajectories. In the weighting functions, h kj and t kj are respectively the height above ground level and the remaining time for the arrival at our site, for each point j of each trajectory k. S H and S T are the scale factors related to the boundary layer and the mean life time of the aerosols, and their values were set to 2000 m and 120 hours respectively. The first weighting function tries to take into account that the concentration of aerosols is higher in the lower layers where the most interaction with the surface is produced. An exception is made for the AF sector, because of the summertime existence of mineral dust in high layers over North Africa. Because of this highaltitude layers, the height weighting factor is not applied within this sector. [29] The second weighting function has been introduced for accounting for the fact that when an air mass formed over a source region moved over a surface with different characteristics (for example a cp air mass originating in north America moves over the Atlantic Ocean) the initial identity of the air mass starts to be lost as particles from the origin sector are being removed and the particles from the transit region are being incorporated. Since the average life time of particles in the troposphere is between 5 and 10 days, during the time of calculation the back trajectories a significant part of the initial characteristics of the air masses will be lost. [30] We then have a set of five z percentage indices (given by equation (3)) that described the character of the air masses as a function of five basic classes. These indices are quantitatively interesting and useful for establishing correlations with the optical properties of the particles. It is particularly interesting to apply these indices to the identification of the primary and secondary classes of the air masses. In order to assign a definitive character to each group of back trajectories in practice we would say that an air mass was purely class i if the index z i possessed a value greater than 80%. Otherwise it was treated as mixed and composed of all those classes that had indices greater than 20%. In the following we only consider either pure air masses or mixed air masses composed of only two classes. The threshold percentages used to define pure or mixed classes were taken as a compromise value. A stricter definition of the pureness of an air mass would have led to a rapid decrease in the number of cases and thus less representative statistics. On the other hand a lower cutoff index would have increased the number of cases but given smoother results. In fact, in our region it would be difficult to admit only a single air mass type. Even an air mass originated in the Artic, before passing over our region will have interacted during at least one day with the continental surface of the Iberian Peninsula. In other frequent cases, African dust high layers will coincide and be mixed with low-level Mediterranean air masses. [31] On occasions an air mass may stagnate over the local area, thus acquiring a different character to the rest of the classes. In these cases the back trajectories are short and meandering and their indices do not have any definite characteristics. It was in order to identify these cases that we introduced the O class. [32] The air masses classified as O class were identified in effect by the maximum distance from the measurement station. If the maximum distance reached was less than a certain limiting value (600 km in our case), we considered that the back trajectory showed an air mass that during five days remained stagnant in the region. Characterization as a function of the five basic classes would make no sense in such a case and the air mass would have converted to a regional air mass with properties specific to the region. [33] Figure 2 shows an example of the evolution of the indices z of sector AF during the month of August 2004 with an hourly resolution, together with the evolution of the AOD at 500 nm at hourly and daily resolution. In Figure 2 a series of three intrusions of AF air masses can be seen, interrupted by the entry of maritime air which cleaned the local air in passing. The correspondence between these intrusions suggested by the indices and the evolution of the turbidity was generally very good. In the following the indices were obtained at 1200 UT so that the aerosol properties could be related to the daily air mass data. [34] Before analyzing the relationship between the air masses and the aerosol properties it seemed interesting to try an independent analysis of the annual variability of the dominant air masses of the region by means of the indices z. Figure 3. Incidence of each type of air mass in the study zone during the period considered. EU, European; AF, African; TR, Tropical; PO, Polar; AR, Artic; O, Regional. 5of12

6 characterizing the aerosol burden in the atmosphere. To calculate it, Beer s Law is used, relating the direct flux incident at ground level (F) with the extraterrestrial flux (F 0 ): tl ð Þ ¼ 1 m 0 ln FðlÞ R 2 F 0 ðlþ ð8þ where m 0 is the relative optical mass (which is assumed to be equal for all atmospheric components), R the Sun-Earth distance in astronomical units (AU), and t(l) is the total optical depth, which can be broken down into different contributions in the form: tl ð Þ ¼ t a ðlþþt R ðlþþt 3 ðlþþt w ðlþþt 2 ðlþ ð9þ Figure 4. Evolution of the z index for African air masses over the length of the study period. Figure 3 shows the daily cases of the incidence of each pure and mixed air mass type. The air masses that were most frequently found were of AF class and mixed AR-PO type. The histogram also showed the minimal incidence of TR maritime air masses in the region. This fact could be explained by the dominant global circulation pattern. [35] Figure 4 shows the evolution of the weighted index for AF air masses over the length of the study period. Figure 4 shows the values of the index as a solid line representing a 30 day moving average in attempt to more clearly illustrate the temporal trend. The evolution over the year only appears significant in the case of AF air masses, for which the maxima occurred in the summer months and the minima in winter. This annual pattern is not systematically compensated by the other air masses, as they showed no such apparent annual pattern. [36] In summary, in this work we propose a classification of the air masses affecting our measurement site and subsequently we study the characteristics of the aerosols present in relation to their origin and the trajectory of said air masses. Although the methodology and results have been only applied to this site, in general the results are expected to be valid in a regional scale (Iberian Peninsula), but taking into account that the atmospheric aerosol properties at a given site and time are the sum of a local and a remote contribution, it can be considered that in general our results should be also valid and comparable in the area of south western Europe. Although the local term would be highly dependent on site (closeness to urban sources, for example) the remote term dependent on the air mass would be similar Optical and Microphysical Aerosol Properties [37] The atmospheric aerosols would be completely defined in the atmospheric column if a given set of properties were retrieved. These aerosol main properties are optical (aerosol optical depth, single scattering albedo and asymmetry parameter) and microphysical (size distribution, complex refractive index and shape). [38] The most simple and useful parameter from this set is the aerosol optical depth (AOD or t), broadly used for being t a (l) the aerosol optical depth, t R (l) the molecular scattering optical depth (or Rayleigh), t 3 (l) the contribution due to the optical depth due to ozone absorption, t W (l) that due to water vapor absorption, and t 2 (l) that due to nitrogen dioxide. Further details on the computation of these contributions are given by Estellés et al. [2007]. [39] To obtain the associated uncertainty in the measurement of the aerosol optical depth the error propagation method is used [Russell et al., 1993]: ð Þ 2 ¼ 1 eðv 0l Þ 2 eðm 0 Þ 2 þ t al þ 1 eðvþ 2 m 0 m 0 V et al þ et ð Rl V 0l m 0 Þ 2 þ et ð 3l Þ 2 þ et ð 2l Þ 2 þ et ð Wl Þ 2 ð10þ [40] In this equation, the term with greatest weight is due to the calibration error. The daily variation of the uncertainty is modulated by the optical mass, with a maximum at solar noon. The uncertainty calculated in this way is similar in size to that given by Schmid and Wehrli [1995], and of the same order as the nominal uncertainty of AERONET ( ) [Eck et al., 1999]. [41] The Ångström wavelength exponent (a) in the nm range was obtained from the linear fit of the aerosol optical depth spectrum: log t a ¼ a þ b log l ð11þ [42] The a exponent is related to the particle size; the smaller particles are characterized by a more pronounced spectral dependence and therefore a greater value of a than for the larger particles, whose optical depth is far less spectrally dependent. [43] The single scattering albedo, asymmetry factor, size distribution and refractive index must be determined by inversion algorithms by inverting the available measurements of sky radiance in the almucantar and principal planes. Different inversion procedures have been proposed in the last several years [Wendisch and von Hoyningen- Huene, 1994; Nakajima et al., 1996; von Hoyningen-Huene and Posse, 1997; Dubovik and King, 2000]. In our case, we have employed the SKYRAD.PACK version 4.2. There is no any paper describing version 4 yet, although there is work in progress (T. Nakajima, personal communication, 2007). The older versions are described by Nakajima et al. 6of12

7 Table 1. Mean Aerosol Properties and Their Respective Standard Deviations (Represented as s) for the Primary Air Mass Classes a t a500 s t a s a n s n k s k w 0 s w0 g s g N AF EU AR PO TR O a AOD at 500 nm (t a500 ), Ångström exponent (a), real part of the refractive index (n), imaginary part of the refractive index (k), aerosol single scattering albedo (w 0 ), asymmetry parameter (g), and the number of cases (N). [1983, 1996]. The difference between old and new versions is mainly related to the computation of the refractive index. On the other hand, the SKYRAD code assumes the particles to have a spherical shape. The sphericity of the particles is not completely true but it has been largely assumed in the literature because of the complexity of the problem. However, the particles can be adequately described by spheres in most cases, mainly for small sulphate particles and hygroscopic aerosols like small sulphate particles and maritime salt aerosols for high humidity, usually present in coastal environments as it is our case. For mineral dust or coarse marine salts imbedded in dry environments the shape is seldom spherical; the presence of coarse mineral dust creates artifacts when using models such as the code from [Dubovik and King, 2000]. However, these artifacts have not been observed when using the SKYRAD code in presence of mineral dust. [44] The SKYRAD method allows the size distribution, the phase function and the single scattering albedo of the aerosols to be obtained from measurements of the direct solar irradiance (F) and diffuse sky radiance (E). When measured at ground level these components are given by: FðlÞ ¼ F 0 ðlþexp½ m 0 tl ð ÞŠ ð12þ EðQ Þ ¼ Fm 0 ½wtPðQÞþqðQÞŠ ð13þ where F 0 the extraterrestrial solar irradiance, w the single scattering albedo, P(Q) the phase function and q(q) a multiple scattering term. [45] Nakajima et al. [1996] developed an optimized radiative transfer code for a plane-parallel atmosphere called the REDuced Multiple scattering program (REDM). This RTC in its different versions constitutes the nucleus of the SKYRAD family of inversion algorithms. In this case version 4.2 was chosen ( clastr/). [46] To work with the diffuse component, the ratio R(Q) is defined: RðQÞ ¼ EðQÞ Fm 0 ¼ wtpðq ÞþqðQÞ ¼ bðqþþqðqþ ð14þ where b(q) is the total scattering coefficient, that includes simple molecular and aerosol scattering. The idea of the method is to iteratively eliminate the multiple scattering term q(q) from the data R(Q) to uncover the coefficient b(q). In each of the steps the code obtains the size distribution V(r) by inversion of b(q) and t a (l). This distribution is used as input for the radiative transfer code for recalculating in turn R 0 (Q), which is compared with the experimental data to evaluate the average square difference e(r). The process is repeated until the deviation is less than 10%. [47] Hourly averages of all instantaneous values were calculated to avoid giving greater weight to those hours with a greater number of measurements. Subsequently, on the basis of the hourly values, the daily averages were calculated and recorded in a database which in turn was used for studying monthly, seasonal and annual statistics. The results of this temporal characterization are given by Estellés et al. [2007], results that have been used here for the sectorial classification. The spectral optical properties were derived with the SKYRAD algorithm for the four experimental channels (440, 670, 870 and 1020 nm). For the presentation of the diagrams (refractive index, asymmetry parameter and single scattering albedo) we have used the mean value obtained from 440 and 670 nm, as a representative value for the visible band. 4. Results 4.1. Aerosol Optical Depth [48] For the analysis of the properties dependence on air mass type, we have employed box diagrams. In these box plots, the means are represented by a solid dot. The divisory segments in the boxes are the medians. The top/bottom box limits represent the monthly means plus/minus the standard deviations. The boxes bars are related to the percentiles 5% (U5) and 95% (U95). Percentiles 1% (U1) and 99% (U99) are represented by the crosses. [49] The main aerosol parameter averages (along with their respective standard deviations) in numerical format are presented for the primary air mass classes in Table 1. The number of cases for each pure class is also presented. Only 6 back trajectories cases could be classified as pure TR; it makes a low statistic for this case, but it is reasonable as these air masses are seldom found in pure state in our site (58 cases for bicomponent states of TR class). The total number of pure classes employed for the statistic was then 350, from 620 cases that were identified as mono (pure) or bicomponent air masses. The other cases up to 695 (the entire aerosol filtered database for daily statistics) were due to three or four component air masses. The entire back trajectory database consisted of 1337 daily triplets. [50] Although our instrument has only 4 useful channels nominally centered at 440, 670, 870 and 1020 nm, the AOD at 500 nm has been preferred for presentation as it is a very usual wavelength used for remote sensing application, representative for the visible band. The classification is mostly equivalent for 440 or 670 nm, and is given by Estellés [2006]. For retrieving the AOD at 500 nm, an interpolation between these two experimental values was done, by using the fit to Ångström law. [51] The diagram box from Figure 5 shows the daily variation of the AOD at 500 nm in relation to the origin of the air masses. The differences are especially significant for 7of12

8 Figure 5. Dependence of the aerosol optical depth at 500 nm on the origin of the air mass classes. EU, European; AF, African; TR, Tropical; PO, Polar; AR, Artic; O, Regional. air masses classified as pure, but the results for mixed air masses are also shown since they too are illustrative of the directional dependence of the optical parameters. The days of greatest turbidity were found under the influence of the pure AF type air masses and also in the mixed air masses formed in part by air from the African continent (mixed with continental EU and maritime TR air masses). The maritime TR air masses also showed a slightly higher average turbidity, perhaps due to the remains of desert origin mineral dust added to the effect of the higher temperatures and humidity which favored the hygroscopic growth of sea salts and other aerosols generated by gas-particle conversion. [52] It is interesting to note that although in the case of the AF type air masses the average and the median coincided, in the case of the TR type air masses the median was much smaller than the average, indicating that the high-turbidity cases were exceptional phenomena. In fact, if we observe only the medians, the values for TR air masses were only slightly above those for the other air masses (PO and AR). The lowest turbidity values were found for AR and PO air masses. The cleanest air masses were mixed AR-PO corresponding to currents originating in Canada. [53] The O type air masses showed very high turbidity. In fact its median value had the absolute maximum of the spectrum of classes. Habitually these air masses stagnate because of low pressure gradients in anticyclonic conditions which facilitate the accumulation of varied anthropogenic pollutants in the Mediterranean basin which add to possible dust from the Peninsula or from Africa. [54] A similar classification of aerosol optical depth and Ångström exponent on the air mass nature has been recently performed for El Arenosillo site in Huelva (Spain) on the south western coast of Spain (C. Toledano et al., Air masses classification and analysis of aerosol types at El Arenosillo (Spain), submitted to Quarterly Journal of the Royal Meteorological Society, 2007, hereinafter referred to as Toledano et al., submitted manuscript, 2007) for a similar database (years ). In general, slightly higher values of the AOD are found on our site for each of the equivalent air mass types (differences of ) being the greatest differences for AF and TR cases Ångström Wavelength Exponent [55] The Ångström exponent (a) is shown against air mass type in Figure 6. The minimum values of the exponent were found for AF air masses and its mixes with types EU and TR, as well as with pure TR air masses which had the minimum mean and median value of the exponent. In the case of the AF cases this result was as expected since the mineral dust from this sector usually provides a large quantity of coarse aerosols. With respect to the aerosols of maritime TR origin, these air masses (a mix, as we have supposed above, of sea salts immersed in highly humid air with mineral dust) are the cause of the low value of the exponent in this sector, although it should not be forgotten that they are much less frequently found that the remainder of the classes, so the statistics for this class is not as rather complete. [56] The maximum value of the exponent, by contrast, was found for the EU continental air masses. The reason was that these air masses carry along large quantities of anthropogenic type aerosols, largely soot from the burning of fossil fuels, and which are found in the accumulation mode of the size distribution. Finally the O type air masses showed intermediate values of the exponent, comparable to the values for PO. This result could be probably a balance between the existence of particles from pollution and the coarse continental dust, or sea salts in warm and humid environments, found in the region of the Mediterranean Sea. [57] In comparison with Toledano et al. s (submitted manuscript, 2007) classification, our retrieved exponents are again somewhat higher in all cases. This systematic difference on AOD and Ångström exponent is probably related to the local contribution of anthropogenic aerosols from the urban environment from Valencia. For example, for African (ct) air masses we obtain an AOD at 440 and Ångström exponent of 0.38 and 1.01 in front of 0.30 and 0.69 for El Arenosillo. For the EU class (cp) our AOD at 440 nm and wavelength exponent are 0.21 and 1.59, against 0.20 and 1.29 values for El Arenosillo. Again, an extra Figure 6. Dependence of the Ångström wavelength exponent on the air mass. EU, European; AF, African; TR, Tropical; PO, Polar; AR, Artic; O, Regional. 8of12

9 Figure 7. Mean aerosol volume distributions for each primary air mass class. EU, European; AF, African; TR, Tropical; PO, Polar; AR, Artic. addition of fine anthropogenic aerosols increases the turbidity and the wavelength exponent. [58] When comparing to other results from southern Portugal (Sagres, a coastal site 200 km west of El Arenosillo station) we found higher differences [Silva et al., 2002]. For example, the estimated AOD at 500 nm and Ångström exponent in Sagres under marine influence were 0.09 and 0.22 (low turbidity and coarse aerosol, probably due to a clean air mass where the marine salts dominate) in front of our 0.12 and 1.22 averages for these parameters (class PO). Our results point out again the effect of fine particles that screen the purely marine air masses. For continental-polluted dominant air masses in Sagres, Silva et al. [2002] obtained for the same parameters a value of 0.30 and In our case, the averages were 0.17 and Once again, our results were dominated by finer aerosols, although in this case their AOD was very high. In any case, it must be taken into account that their database was limited to 40 days in summer 1997, so a proper statistic is not available. Equivalent conclusions can be obtained when comparing our results to those of Elias et al. [2006] in Évora (Portugal) a site 200 km north from Sagres and El Arenosillo and 150 km far from important anthropogenic sources Aerosol Volume Distributions [59] Figure 7 shows the volume distributions corresponding to each of the air mass types. To produce this graph the averages were taken of the daily volume distributions. The volume distributions were retrieved with the SKYRAD algorithm between radius 0.05 and 15 mm. [60] In Figure 7, the AF class shows the best developed coarse mode, related to mineral dust; a relative high accumulation mode is also obtained, probably in relation to fine dust. A similar accumulation mode is found for the EU class, in this case in relation to the anthropogenic sources found in continental Europe, coincident with the sulphate range mode from Kaufman et al. [1994], as reported by Nakajima et al. [1996]. On turn, the EU distribution shows a trimodal shape, with the weakest modes within the salt and coarse particle ranges. The maritime air masses (TR, PO and AR) show the lowest accumulation modes with little differences between them. However, their coarser modes have clear differences: the AR and PO classes have a relatively low salt mode, meanwhile the TR salt and coarse mode appear clearly developed. A reason for such behavior could be related to the higher temperature and humidity of these air masses, that helps the development of hydrated aerosols at the same time that dust particles are also carry by them. In any case, the poor statistic for this air mass class does not allow us to make definitive statements in relation to its behavior Refractive Index [61] Figure 8 shows the dependence of the complex refractive index on the air mass type. The real part of the refractive index had maximum values for all the classes that were either directly or indirectly influenced by the AF air masses (AF, AFEU, AFTR, TR and TRPO). The absolute maximum of the averages was found for the AF class. This average value (1.41 ± 0.06) was lower than pure desert aerosols as compared to values from literature. For example, for sites typically affected by desert dust like Bahrain in Persian Gulf, the real part of the imaginary index is 1.55 ± 0.03 [Dubovik et al., 2002]. Nevertheless it is to be expected that the effective refractive index of the column would not usually be so extreme, and the average would remain around intermediate values. However, the AF class showed a high scattering with values that frequently reached levels indicative of situations where mineral dust prevailed. On turn, the maritime air masses (AR, PO and TR) show a real index ( ) very similar to that of Silva et al. [2002] for the maritime episodes (1.39), probably related to hydrated salt particles. Finally, the O type air masses had values that were intermediate between the air masses with an AF influence and those dominated by cleaner maritime classes. In some of these cases there was a clear dominance of mineral aerosols. Figure 8. Dependence of the (a) real and (b) imaginary part of the refractive index on the air mass. EU, European; AF, African; TR, Tropical; PO, Polar; AR, Artic; O, Regional. 9of12

10 Figure 9. Dependence of the single scattering albedo on the air mass. EU, European; AF, African; TR, Tropical; PO, Polar; AR, Artic; O, Regional. [62] With respect to the imaginary part of the refractive index, its dependence on air mass class was less significant. The minimum value of the index was found for EU type air masses, indicating perhaps greater absorption by soot. However, in the maritime air masses where the dominant component should be nonabsorbent salt, we found similar values for the imaginary part of the refractive index. This anomalous behavior could be explained a priori by three reasons: (1) Anthropogenic aerosols were not only found in the polluted EU air masses but also in the maritime origin air masses AR and PO, perhaps because of the longer life span of small soot particles acquired in transit over industrial zones of North America; (2) the source of these absorbent aerosols, found in air masses originating from the west and northwest, may be found in the Iberian Peninsula; the industrial regions of Madrid and the Basque Country lie under the paths of these air currents; and (3) the inversion code was unable to obtain the imaginary part of the index with enough accuracy. This parameter always has a high degree of uncertainty, especially for low turbidity [Di Carmine et al., 2005], which was the case for Atlantic origin maritime air masses. Therefore it was not possible to draw any definitive conclusions with respect to the characterization of this parameter Single Scattering Albedo [63] The single scattering albedo (SSA or w 0 ) is closely linked to the imaginary part of the refractive index and shows the same dependences and uncertainties. Although no publication is yet available for the SKYRAD uncertainty estimation on SSA, a current comparison of SKYRAD values with AERONET results at Gosan (Korea) in the last UNEP/ABC/EAREX project, has given a RMS difference between both SSA products of about 0.05 (K. Aoki, personal communication, 2007). [64] The SSA classification can be seen in Figure 9. It is difficult to find a simple explanation for its dependence. It can be seen that the EU origin air masses frequently had lower values of SSA than those originating in neighboring sectors, indicating situations in which the quantity of absorbing particles would have increased [Masmoudi et al., 2003]. In this work as in previous studies by other authors in the Iberian Peninsula using different methods [Lyamani et al., 2004], minimum values of the SSA were found in the winter months. Although these minima have sometimes been explained in relation to the greater rate of emission of soot during the winter (from heating) there has been no rigorous study of the specific weight of these sources in the obtained value of the SSA. Given that in winter the turbidity is very low, it could be that the reason for this behavior lay in the limitations of the codes for producing precise results in clean sky conditions Asymmetry Parameter [65] The asymmetry parameter had a slight and smooth dependence on the origin sector. The air masses that were more or less influenced by AF air masses had an average asymmetry parameter between 0.67 and The EU type air masses had a slightly lower average of 0.65, indicating less forward scattering or, what amounts to the same thing, particles with a phase function indicative of smaller size. The maritime origin air masses had an intermediate value between the previous values of from 0.66 to Figure 10 shows this result graphically. 5. Conclusions [66] The source regions affecting a Spanish Mediterranean measurement site have been divided into sectors allowing us to perform a modified classification based on previous air mass classifications based on thermodynamic criteria. These sectors were used to develop a simple model for characterizing the air masses, based on their residence time in each sector. Beginning with a definition of the character of a specific air mass in the measurement region we were able to relate the properties of the aerosols with the dominant character of said air masses. [67] The results of this classification showed that the optical depth had a significant dependence on the origin of the air mass. The highest turbidity values were found when the dominant air mass originated in North Africa, while the minima were found for PO-AR type air masses. The Ångström exponent showed the opposite trend, with maxima for EU type air masses and minima for TR-AF classes. [68] These trends observed in the AOD and the Ångström exponent were consistent with the dependences found in the size distributions. The major source of small particles occurred when the air mass crossed the European continent or when they were sourced locally, pointing to the influence of anthropogenic particles related to urban and industrial pollution. Inversely, the major contribution of coarse particles occurred when the dominant air masses were AF or TR. [69] The real part of the refractive index showed as well that the highest values were obtained in air masses from the south; those originating in Europe or locally had lower values between the typical values for sea salts and anthropogenic sulphates. Values typical of pure mineral particles were not observed with excessive frequency. [70] The imaginary part of the refractive index and the single scattering albedo showed the predominance of less absorbent aerosols when these were carried from desert 10 of 12