Study of desert dust events over the southwestern Iberian Peninsula in year 2000: two case studies

. Strong desert dust events occurring in 2000 over the southwestern Atlantic coast of the Iberian Peninsula are detected and evaluated by means of the TOMS Aerosol Index (A.I.) at three different sites, Funchal (Madeira Island, Portu-gal), Lisboa (Portugal) and El Arenosillo (Huelva, Spain). At the El Arenosillo station, measurements from an AERONET Cimel sunphotometer allow a more precise retrieval of the spectral AOD and the derived alpha ˚Angstr¨om coefﬁcient. After using different threshold values of these parameters, we conclude that it is difﬁcult to establish reliable and robust criteria for an automatic estimation of the number of dust episodes and the total number of dusty days per year. As a result, additional information, such as airmass trajectories, were used to improve the estimation, from which reasonable results were obtained (although some manual editting was still needed). A detailed characterization of two selected desert dust episodes, a strong event in winter and another of less intensity in summer, was carried out using AOD derived from Brewer spectrometer measurements. Size distribution parameters and radiative properties, such as arefrac-tive index and the aerosol single scattering albedo derived from Cimel data, were analyzed in detail for one of these two case studies. Although speciﬁc to this dust episode, the retrieved range of values of these parameters clearly reﬂect the characteristics of desert aerosols. Back-trajectory analysis, synoptic weather maps and satellite images were also considered together, as supporting data to assess the aerosol desert characterization in this region of study.


Introduction
The detection and characterization of desert dust aerosols on a global scale and hence in different areas of the world is necessary to evaluate dust aerosol radiative forcing Miller and Tegen 1998). Its influence, along with other types of aerosols, is considered to be crucially important, even though its magnitude is still highly uncertain (Shine and Forster, 1999;Haywood and Boucher, 2000;Haywood et al., 2001). Desert dust events have been widely studied in recent years using a variety of measurements, including those from ground-based, aircraft and satellite sensors (Karyampudy et al., 1999;Husar et al., 2001;Díaz et al., 2000;Ginoux et al., 2001;Kinne et al., 2001). Such data are valuable for assessing various available desert dust models (Longtin et al., 1988;D'Almeida et al., 1991). It is also becoming increasingly clear from recent observations that these models must be adapted to account for the physical and chemical modification of the dust particles as they are transported over long distances from their source regions.
In this paper, we focus on desert dust outbreaks originating in North Africa (Sahara, Sahel) that reach the Iberian Peninsula. In Spain, the so-called red rain (corresponding to wet deposition of desert dust) has been used as an indication of the frequency and extent of these desert dust outbreaks (Avila et al., 1997). Recent studies concerning African dust outbreaks reaching the Iberian Peninsula are reported using back-trajectory air masses. These studies evaluate the influence on ground particle mass concentration measurements of Air Quality networks, resulting from desert dust outbreaks (Rodriguez et al., 2001;Querol et al., 2002a, b;Sanchez de la Campa, 2002). The data for PM10 (particle diameters less than 10 µm) concentration are of particular interest  because the recent European Union Directive 1999/30/CE on Air Quality established PM10 limit values, which are frequently exceeded in rural areas of Spain and southern Europe due to the occurrence of desert dust outbreaks (Artiñano, 2001;Querol et al., 2002a, b) thereby making attainment of this directive difficult.
Outbreaks of desert dust plumes, similar to those observed over the Iberian Peninsula, have been analysed over the Mediterranean Sea (Prospero, 1996;Guerzoni et al., 1997;Moulin et al., 1997a), The Canary Islands (Bergametti et al., 1989;Smirnov et al., 1998;Díaz et al., 2001) and the Atlantic Ocean (Prospero, 1999;Moulin et al., 1997b). Pinker Ann. Geophys., 24,[1493][1494][1495][1496][1497][1498][1499][1500][1501][1502][1503][1504][1505][1506][1507][1508][1509][1510]2006 www.ann-geophys.net/24/1493 /2006/ et al. (2001) has also analyzed a strong outbreak which occurred between the end of January and the beginning of February 2000 over the western sub-Sahel, which has many similarities with our first case study presented in this paper. The opportunity to follow African desert dust events that reach the Atlantic Ocean and the Atlantic coast of the Iberian Peninsula from multiple stations is not very typical. In this study, available aerosol measurements at three different sites of Funchal on the Madeira Island, Lisboa on the Atlantic coast of Portugal, and El Arenosillo (Huelva) in the southwest coastal area of Spain, allows a detailed analysis of two desert dust outbreaks observed during 2000.
The aim of this work is to detect and determine optical and physical properties of desert dust aerosols over the southwestern coast of the Iberian Peninsula in terms of AOD and Angström coefficient, α (if available), together with complementary data, such as back-trajectory analysis, satellite images and weather maps. Available AOD data at the three different stations have allowed the study of two different desert dust events and a detailed analysis of size distribution and other microphysical and radiative parameters (thanks to data from a Cimel photometer at one site).
The structure of the paper is as follows: Sect. 2 gives details of the measurement sites and instruments. Section 3 briefly describes the methodology and support data. Section 4 summarizes the measurements and results. The first part focuses on the detection and identification of desert dust episodes and the latter parts are focused on a more detailed description of the two selected episodes -a strong winter episode and a somewhat weaker summer episode.

Measurement sites and instruments
The station at El Arenosillo (37.1 N, 6.7 W, 10 m ASL and #213) in Huelva (Spain), is part of Instituto Nacional de Técnica Aeroespacial (INTA). The stations at Lisboa (38.77 N,9.15 W,105 m ASL and #082) and Funchal (32.64 N,16.89 W, 58 m ASL, and #287) on Madeira Island belong to Instituto de Meteorologia of Portugal. (The station numbers are those designated by the WMO network to track ozone atmospheric content.) These three stations are very good sites for solar radiation measurements and aerosol studies due to their geographical location with maritime background aerosols. Lisboa also has a significant urban aerosol component. Continental sources can have an important effect at El Arenosillo and Lisboa, but desert dust outbreaks, as we will show later, seem to have a major influence at the three sites. Figure 1 shows the location of the three stations (black points) on two SeaWIFS images, where one of the two selected desert dust storm case studies is depicted.
In this work we focus particularly on the application of Brewer spectroradiometer measurements to derive AOD data. The Brewer instruments were designed for ozone monitoring (Kerr, 1981) using five wavelengths (306.6, 310.1, 313.5, 316.8, 320.1 nm, respectively; some differences exit between instruments and models) and require 3 min per wavelength scan. They also were designed to measure the spectral UV global irradiances from 286.5-363 nm (depending on the model, single or double monochromator) with a step scan of 0.5 nm and with a nominal HWHM (Half-Width at Half Maximum) of 0.62 nm. Direct and global irradiance measurements have different fore-optics in the Brewer system (a rotating prism and a cosine receptor, respectively), each of which required different calibration procedures. The FOV (Field Of View) of the instrument is controlled by an iris, which can be set to about 2 • (used regularly for the direct sun measurements) and 10 • . Five neutral density filters with increasing attenuation (up to about a total of 10 −5 ) can be inserted into the beam path to protect the photomultiplier (the detection system) from overexposure to sunlight. These filters increased the difficulty of the calibration procedure.
The three Brewer systems follow the requirements established by the Brewer WMO network and are calibrated periodically according to NIST standards against the Brewer Reference Travelling Standard #17 operated by the International Ozone Services (IOS). Details about the Brewer instrument characteristics, performance and maintenance may be found at (http://www.woudc.org/). UV radiation measurements, made by YES pyranometers (1-min resolution) at El Arenosillo and Lisboa, were used in this work as support data to detect the presence of clouds. Details of these instruments, measurement and calibration procedures are found elsewhere (Yankee user guide, 1996;Vilaplana et al., 2002;Vilaplana, 2004). These measurements were also important for other research studies related to the decrease in UV radiation during high turbidity episodes, such as those due to desert dust outbreaks.
An AERONET (AErosol Robotic Network, ) Cimel sun-photometer (in operation at El Arenosillo station since February 2000) allows the determination of AOD and additional aerosol properties. El Arenosillo's Cimel photometer is a polarized instrument model CE318-2 with 8 filters, where the filters at wavelengths of 440, 670, 870, and 1020 nm, respectively, are used for aerosol studies. AOD values at these 4 wavelengths and the derived Angström exponent, α, are obtained as primary aerosol parameters. With the addition of sky radiance measurements and inversion algorithm methodology, other aerosol properties are determined, such as volume size distributions, refractive index, single scattering albedo, etc. For details of this instrument, the measurements and derived aerosol parameters, see Holben et al. (1998,  and Dubovik et al. ( , 2002a

Methods and supporting data
Multi-year data sets (meteorological variables, radiation data, etc.) are monitored at the three sites using different instruments. These atmospheric variables can be used to detect and evaluate desert dust episodes. Surface AOD is certainly an important parameter for desert dust detection. Examination of the AOD series shows the relatively high frequency of desert dust outbreaks over the three sites. However, a reliable criterion needs to be established to clearly identify desert dust events. Indeed, for weak desert dust outbreaks, AOD is not sufficient and complementary information is generally necessary. Thus, back-trajectory analysis, synoptic weather maps, satellite images and other parameters, such as the aerosol index (A.I.), are used to evaluate and characterize the desert dust aerosols transported over the study region. Table 1 summarizes the type of data used for the two selected case studies.
3.1 Aerosol optical depth andÅngström exponent α AOD is a key aerosol parameter, defining the level of turbidity in the atmosphere. Indeed, AOD is a good indicator of desert dust outbreaks which generate high atmospheric turbidity Eck et al., 1999;Dubovik et al., 2002a). In this work we focus particularly on the application of Brewer spectroradiometer measurements to derive AOD at El Arenosillo, Lisboa and Funchal sites. As is well known, the Brewer instruments are currently used for the ozone content retrieval (Kerr et al., 1981), but recently, research efforts have addressed the use of Brewer measurements for AOD retrieval (Carvalho and Henriquez, 2000;Gröbner et al., 2001, Kirchoff et al., 2001Kerr, 2002;Marenco et al., 2002;Cachorro et al., 2003;Kazadzis et al., 2005). Although not systematic and operational, AOD monitoring by Brewers system has been conducted recently at different sites (Kerr, 2002;Jaroslawsky et al., 2003;Cheymol and De Backer, 2003;Gröbner and Meleti, 2004;Kazadzis et al., 2005). The method we have followed for AOD determination using Brewer measurements is based on the work of Carvalho and Henriquez (2000) using the routine DS data. In addition, the values determined in this work have also been assessed with those calculated following the procedure used in Gröbner et al. (2001).
Errors associated with Brewer measurements are difficult and laborious to evaluate (Kerr 2002;Gröbner and Meleti, 2004;Arola and Koskela, 2004;Kazadzis et al., 2005), because ozone, AOD and global UV irradiance follow different calibration procedures and retrieval methods. Therefore, various methods have been developed for this determination, where the main problem is how to obtain direct irradiance for applying the Beer law (for details, see the recent work of Kazadzis et al., 2005). Absolute error values of 0.07 are estimated in Arola and Koskela (2004) and Kazadzis et al. (2005) for Brewer AOD values. Thus, taking 0.3 as a representative and realistic value of the AOD in the UV, we estimate a relative error of ∼25%.
With respect to the AOD derived by the Cimel sunphotometer, we used data provided by the AERONET procedure. An error of 0.02 is estimated for AOD sun-photometer data Eck et al., 1999). We also considered values of α (Ångström exponent), as it adds valuable information, together with AOD for desert dust detection. Here we use theÅngström formula AOD=βλ −α , where λ is the wavelength (in µm), α is theÅngström exponent, and β is related to the concentration of aerosols (β is also the AOD at λ=1 µm). TheÅngström exponent α is derived from a log-log fit of the AOD data using three wavelengths (440, 670 and 870 nm, respectively), according to the AERONET procedure Holben et al., 2001). This procedure uses quantitative information about the spectral curvature of the AOD, which, in turn, is related to the particle size distribution.. Values of α>1.0 are indicative of size distributions dominated by fine particles, while 0<α<0.8 tend to be Ann. Geophys., 24,[1493][1494][1495][1496][1497][1498][1499][1500][1501][1502][1503][1504][1505][1506][1507][1508][1509][1510]2006 www.ann-geophys.net/24/1493/2006/ dominated by large particles Holben et al., 2001;Dubovik et al., 2002a). AOD and α are the primary key parameters in defining the climatology of aerosols in a given region Holben et al., 2001;Dubovik et al., 2002a). Desert aerosols (Longtin et al., 1988;D'Almeida et al., 1991) are characterized by high AOD values and low α values, resulting from an abundance of large particles. Dubovik et al. (2002a) defines a range of climatological α values for desert aerosols at four representative sites; i.e.: (-0.1≤α≤0.7 at Cape Verde and (0≤α≤1.55 at Lanai). However, the observed wide range of values (including negative values) makes this parameter only a good indicator, but not a definite one. A more robust characterization of desert aerosols is obtained when size distribution and optical properties are combined. Radiance data, together with Cimel AOD data, allow the application of the inversion algorithm to retrieve these properties.
Although the extraction of physical aerosol parameters from satellite sensors is difficult (Stowe et al., 1997;Tanré et al, 1997), retrieval methods are being improved and evaluated. Usually, satellite retrieval results are generally validated using ground-based measurements, but in this study they are used as complementary information to ground-based data and for a qualitative visualization of desert dust events.

Cloud screening methodology
Desert dust episodes are frequently accompanied by clouds and rainfall. Radiometric systems are very sensitive to cloud coverage, with global radiation being a useful tool to detect the presence of clouds. AOD measurements are also useful for cloud detection, as demonstrated by the Cimel AOD data retrieval algorithm . The Brewer AOD data must also be examined for possible cloud contamination. Satellite images, such as those of Meteosat and SeaW-IFS, serve as complementary information in that regard. Examination of the Meteosat infrared images over the study region, we observe qualitatively that the February-March event is certainly affected by cloud coverage. For the July-August episode, no clouds appear in the IR images.
The Cimel measurements for El Arenosillo are processed using a cloud screening algorithm, following a criterion based on the variability of AOD during triads of measurements taken for each datum . If this variability is higher than a certain level, the measurement is discarded as cloud affected. Data at this level of screening (level 1.5 in the AERONET database) are those used here. No specific screening algorithm is applied to the Brewer AOD data, other than discarding those AOD data as determined by the Cimel criterion at the El Arenosillo station.
In general, we have used UV global radiation data from the YES UV-pyranometer to detect the presence of clouds and to discard any Brewer AOD data affected by them. The procedure consists of an evaluation of the time interval where the shape of the global radiation curve deviates from a cloudfree day radiation by more than 10%. The application of this screening criterion to the Brewer AOD data is consistent with the AERONET cloud screening technique used in the analysis of the Cimel AOD data for the February-March episode at El Arenosillo.
Unfortunately, no UV radiation data were available at Funchal. Although only qualitative and not very efficient, we inspected Meteosat infrared images as a means for discarding Brewer AOD data contaminated by clouds. During the first case study, the highest AOD days also coincided with the highest A.I. and clouds, making it difficult to describe the evolution of this episode in any quantitative manner.

Back-trajectories and synoptic weather maps
Back-trajectory analysis was also carried out as complementary information for dust events. Back-trajectories were computed with the HYSPLIT (HYbrid Single-Particle Lagrangian Integrated Trajectory) model, a program developed by NOAA (Draxler, 1996;Draxler andHess, 1997, 1998), to study the movement of a particle within an air mass. In this study, we used the FNL database, archived by NOAA (National Oceanic and Atmospheric Administration; www.arl.noaa.gov/ss/transport/archives.html). The output of the model is a set of points (one per hour) that permits the tracking of the positions of the air mass in the atmosphere in latitude, longitude and height, according to the vertical velocity field. In order to minimize the usual errors of atmospheric dynamic computations, only back-trajectories with a maximum duration of five days were considered (Stunder, 1996;Stohl, 1998;Derwent et al., 1998;Cape et al., 2000).
Daily back-trajectories for the year 2000 were computed for air masses arriving at 12:00 GMT at El Arenosillo (at Lisboa and Funchal, only those days around the two selected desert episodes were calculated) for final site altitudes of 3000, 2000 and 1000 m, respectively. The highest levels are particularly suited for desert aerosol assessment (Vergaz, 2001;Rodriguez et al., 2001;Querol et al., 2002a, b). From an inspection of these trajectories, we assign those arriving from North Africa as desert dust episodes, which allows an evaluation of the episodes and number of dusty days during 2000 (Vergaz, 2001;Vergaz et al., 2005). Additionally, synoptic weather maps were obtained from the NOAA FLN data set (http://www.arl.noaa.gov/ready/ amet.html) at the 700 hPa, 850 hPa and surface levels as complementary information for the two case studies.

TOMS Aerosol Index
A frequently used aerosol parameter to monitor desert dust outbreaks is the TOMS (Total Ozone Mapper Spectrometer on the Earth-Probe satellite) Aerosol Index (A.I.) Cakmur et al., 2001;McPeters et al., 1998). The TOMS A.I. is based on the backscattered radiance measured in the 331 and 360 nm channels and is indicative of the aerosol content over the study area. If the A.I. is large and positive, absorbing aerosols and high turbidity are present in the atmosphere; if it is negative, the aerosol is nonabsorbing (Hsu, 1999). Also, A.I. is insensitive to aerosols below 2 km. Moreover, weakly absorbing aerosols are difficult to detect. Nevertheless, we use A.I. as one of our indicators of desert dust outbreaks over the three sites.

Detection and evaluation of desert dust events in year 2000
The TOMS A.I. time series (Fig. 2a) from mid-1996 to the end of 2000 suggests an annual cycle, with a winter minimum which seems to have decreased slightly. We can also compute a mean A.I. and standard deviation for each of the three stations. The data set, however, is too short to establish any meaninful trend. Figure 2b shows A.I. values for 2000 only, where the strong events recorded for each station are easily detected. Figure 2c shows the number of occurrences  After some experimentation, we found that a threshold A.I. value equal to the mean plus twice the standard deviation (STD) is sufficient to detect most desert dust events.
We selected for case studies only those events for which the A.I. exceeded the threshold at all three stations simultaneously. Only two such episodes were selected: a strong episode in early March (4-14 March) and another at the end of July/beginning of August. A third event was a common episode to Funchal and Lisboa on 26-28 February, but not to El Arenosillo, and thus was not selected.
The A.I. appears to be a useful parameter for detecting strong desert dust events, but not moderate or weak episodes. This will be discussed later in the context of evaluations of the El Arenosillo data. This is also confirmed by the poor correlation between A.I. and mean daily Cimel AOD data at 440 nm over the region (see Fig. 3), where only a qualitative correlation appears for AOD greater than 0.4 and A.I. greater than 1.
To detect and evaluate desert dust outbreaks over El Arenosillo during 2000, we have examined the Cimelderived AOD time series (Fig. 4). Several episodes of high turbidity can be observed, many of which could be considered the result of desert dust outbreaks. Although the large AOD values can be assumed to be a result of desert dust events when compared to normal atmospheric conditions, we need to define a threshold for AOD values which provides an unambiguous identification of desert dust episodes.  used for A.I., we compute a mean and standard deviation 0.19 and 0.14; respectively, to estimate a threshold of 0.3. However, since AOD can be large for non-dust conditions, it is necessary to include the wavelength information provided by theÅngström exponent, α, to remove the nondust cases.
Generally, values of α greater than 1 are associated with fine particles, while those less than 1 have a significant contribution from coarse (dust) particles. As can be seen in Fig. 4 a strong anti-correlation exists between AOD values and α during high turbidity episodes, i.e. when AOD increases, α decreases, with α approaching values near zero. The average α is 0.8, with a STD = 0.4. We adopt a threshold of 1.0 as the maximum value for α for a dust episode. With the AOD and α criteria, we have 14 events with a total of 27 days. We can also evaluate desert dust episodes by means of back-trajectory analysis, taking the levels there and assigning those with a clear origin over a geographical sector covering North Africa (for details, see Vergaz, 2001;Vergaz et al., 2005). For 2000 (not including January) at El Arenosillo, there were 21 desert events with a total of 60 days.
Comparing the event selection results between these two methods indicates that the threshold values using AOD and α are perhaps too restrictive. We notice that the days corresponding to the beginning and the end of the episode are not accounted for and only the "central" days are included in the automatic filtering of the Cimel data. Using a lower threshold AOD value of 0.26 and α ≤1, we find 18 events with a total of 39 days. These events indicated in Fig. 4, where single day events (3,5,8,11,13,15,16,17,18)   a bold number, while events 1, 2 and 9 (marked with a bold square) are the selected case studies. Although the number of events and days has increased and agrees better with the air mass trajectory method, the difference between the two methods remains high. As a consequence of the different results from the two methods described above, we decided to perform a joint examination of the Cimel AOD database and the air mass trajectories. We have taken all available data (not the filtered, average daily values) of the AOD-α database (with the latter threshold AOD and α values) and the corresponding daily back-trajectories, with the analyst deciding whether or not to include a particular day as belonging to a given dust episode. In this analysis, 18 events with a total of 72 days were obtained.
The coincidence between the number of episodes from the joint examination and automatic filtering of the Cimel data indicates that the thresholds chosen for the Cimel data were satisfactory. The large difference between the number of days may be explained by the uncertainties in selecting the beginning or ending of an episode, which were not taken account previously.
We conclude from the foregoing discussion that a truly automated algorithm for selecting dust events is not yet possible without some analyst intervention. Nevertheless, we have shown that through suitable selection of threshold values of aerosol parameters, such as A.I., AOD, and α, that reasonable event identification is possible. It is also clear that detailed analysis of data from several years is needed to refine the selection criteria. We note that the same variability in desert dust events occurs every year over the Iberian Peninsula (Rodriguez et al., 2001).

Case study 1: winter desert dust episode
As mentioned above, this case study, February -March, is characterized by great intensity and long duration. Figure 4 shows that this episode can be considered as composed of two desert dust events separated by four days with low AOD (below 0.2). However, due to the strong intensity and the great spatial extent of the associated dust plume, we analyse them together, as part of continuous meteorological conditions over the Saharan region. Furthermore, the episode is highly contaminated by clouds with rainy weather conditions, making its characterization by radiometric data complicated.
In Fig. 5, a plot of the Cimel AOD (440 nm), together with the α coefficient at El Arenosillo, documents the main features of the event. On 25 February the AOD is ∼0.2 exhibits large variability compared to normal conditions for low AOD values. TheÅngström α coefficient decreases from about 1.2 to 0.6, suggesting the presence of dust at the end of the day. During 26 February (see Fig. 1a) the dust is well established and AOD increases to 0.5 on 28 February (see SeaW-IFS image in Fig. 1b) and decreases to 0.4 on 29 February. AOD decreases to normal values in the following days. The α coefficient decreases during the AOD increase and vice versa, leading to a good anticorrelation during the episode 25-29 February. However, it must emphasized that AOD and α are highly variable during the entire period.   The second part of this episode began on 6 March, according to the Cimel data. The AOD reaches a maximum value of ∼ 1, while on 5 March it is only ∼ 0.2. There were no Cimel data available during the afternoon on 5 March and most of 6 March, because of clouds and rain (corroborated by UV radiation data, SeaWIFS and Meteosat images, not shown). Simultaneously, the α values decrease from about 1 to near 0, clearly indicating a desert dust event. The end of this event is not to define, because dust remains several days after the maximum, with AOD quite slowly. Based on the nearly normal AOD values, March 15 appears to be a reasonable date for the end of the episode. Figure 5b illustrates the Brewer AOD (320 nm) time series at Funchal and El Arenosillo from 3 March. Since Brewer data were not available in February, the first part of the event cannot be displayed. No Brewer data were available at Lisboa during the entire episode. Brewer AOD data and A.I. values in Fig. 5c are analyzed together for all three stations. At El Arenosillo, despite a limited number of measurements, the Brewer AOD data exhibit the same features as the Cimel AOD data. The highest A.I. (greater than 1 from 5-9 March) are observed in accordance with AOD.
At Funchal, the event shows a slightly different temporal behaviour than at the other two sites. Although the first part of the event also arrives on 25 February (24 February    During the summer event (July-August 2000), a better, more quantitative characterization was possible due to noncloudy conditions. Figure 6 documents the AOD and A.I. values for this event. The AOD (440 nm) at El Arenosillo (Fig. 6a) clearly shows that this episode is less intense compared to the winter episode. AOD begins to increase in the afternoon of 29 July, but does not exceed values greater than 0.2 (a typical value in summer). As α is stable at ∼ 1.3, it suggests that desert dust aerosols are not yet present.   On 30 July, the AOD values are still not large, ∼0.22-0.28, but α is decreasing, which suggests that the dust event is beginning to establish (also confirmed by back-trajectories, as discussed below). AOD values reach a maximum of 0.6 on 1 August and remain at similar values on 2 August. Values drop considerably during 3 August, but still exhibit dusty characteristics. On subsequent days (4-5 August), AOD again increases, but with larger values of α, which probably represents a mix of aerosol types. AOD and α show the expected anti-correlation during the event, with α decreasing from ∼1.3 on 29 July to ∼0.1 on 1 and 2 August, when AOD is highest. Using the Cimel AOD-α threshold criteria, the automatic dust event algorithm includes 31 July and 1 and 2August, but not 30 July and 3 August, which are included with analyst intervention.
Brewer AOD data (Fig. 6b) increases smoothly at El Arenosillo and Lisboa (with lower values at Lisboa) until the event is well established at both, between 31 July and  Inspection of back-trajectories at El Arenosillo (Fig. 7) shows that the 29 July (12:00 UTC) air mass trajectories at all three levels come from the Atlantic, but on 30 July they arrive from North Africa, suggesting that the event probably arrived during the night. The back-trajectory at 1000 m. is more complex compared with the other two levels (2000 and 3000 m respectively). The latter have a clear Africa origin, while the former indicates re-circulation, with the air mass coming from the Mediterranean area, but having its origin in North Africa. Back-trajectories on 2 August show a clear African origin, while on 3 August the air masses come from the West Atlantic, corroborating earlier conclusions from the Cimel-Brewer data about the end of the episode.
The event at Funchal exhibited some differences compared with the other sites, namely with respect to intensity and variability. It began on July 26, as indicated from backtrajectories (not shown) and was fully established on 1-4 August, with AOD values reaching a maximum of ∼ 0.8 on 2 August. The event persisted through 7 August, with decreasing, but relatively high AOD values and with α values consistent with desert dust. Interestingly, back-trajectories on 6 August (not shown) indicated a North Atlantic origin, suggesting that the dust plume may have undergone some recirculation over the ocean.
It is interesting to note that the Brewer AOD at 320 nm is lower than that of the Cimel AOD at 440 nm during the maximum of the summer event (31 July-2 August), in contrast to the previous days, when background conditions prevail. This behaviour in the UV is also consistent with the flat spectral dependence of desert aerosols in the visible. Figure 6 shows a clear correspondence between dustrelated AOD and positive A.I. values at each station. The exception appears to be at Funchal on 26 and 28 July, where A.I. values greater than 1 are observed, while the Brewer AOD data are less than 0.3. Also, we have inspected the daily TOMS A.I. contour maps of the dust plumes of over the extended region around the three sites from 28 July to 4 August. These maps show that the intensity of the dust episode is greatly diminished by the time it reaches the measurement sites, compared to other locations closer to the dust sources. It is likely that inhomogeneities in the dust plume, as it is transported out over the Atlantic and back toward the Iberian Peninsula, accounts for the variability observed from site to site.
The synoptic weather conditions over the study region on 29 August are shown in Fig. 8. These appear to be the most frequent type of weather situations for dust transport over Spain during the summer months. The surface pressure map shows a thermal low over the North African Sahara as a result of the intense surface heating. An upper level high induces a convective system which pumps dust to high altitudes (3000-5000 m) leading to a large dust charge. If the high moves eastward, the air masses move from south to north, covering a wide area of Mediterranean, including the Gulf of Cadiz and the Iberian Peninsula, transporting the dust-laden layers. Most summer events follow this pattern (Rodriguez et al., 2001  (d) Fig. 9. Volume particle size distribution functions at representative days before, during and after the desert dust event of July-August at El Arenosillo, obtained by inversion of AOD and sky radiance measurements by the CIMEL instrument (AERONET inversion protocol).

Aerosol particle size distributions and radiative parameters
Relevant aerosol properties for this study are the particle size distribution, the asymmetry parameter g, refractive index and single scattering albedo SSA, which are obtained by the inversion method developed by  and . The values and evolution of these aerosol parameters obtained from the CIMEL measurements at El Arenosillo provides valuable information about the dust aerosols over this region. The inversion method uses AOD values and sky radiance measurements at all scattering angles and assumes that the Mie theory applies for spherical particles. The method is only applicable for clear-sky (noncloud) conditions. A check for nonsphericity was necessary for these data because of the appearance of an unusually large accumulation mode in the retrieval when using the spherical model. This artifact of the algorithm was resolved by Dubovik et al. ( , 2002b, using a randomly oriented spheroids model. We have used this approach when assessing the particle size distribution retrieval of dust aerosols. It is clear that care must be taken with the error associated with the particle size distribution and radiative parameters when using the Dubovik et al. ( , 2002a inversion method. All available level 1.5 (as defined by AERONET protocols) radiance and AOD data were used to retrieve the volume size distribution function, dV/dlnr. Several representative days of the July-August desert dust episode are shown in Figs. 9a-d, where we can follow the evolution of aerosol particle size distributions in some detail. The particle size distributions always exhibit two modes: an accumulation mode, with particle radius below 0.6 µm, and a coarse mode, with particle radius between 0.6 and 15 µm. The coarse particle mode is always dominant during conditions where AOD is greater than 0.4. At the onset of the event, in the early morning of 29 July (Fig. 9a), when no dust is yet present, the maximum aerosol volume particle concentration (using the sphere model) is less than 0.02 µm 3 /µm 2 , with similar concentrations for both modes. For 30 July to 3 August the sphere model led to anomalous results for the accumulation mode, hence, we have used the spheroid model for the retrievals during this latter period.  On 30 and 31 July, the coarse number concentration increases to 0.06 and 0.15 µm 3 /µm 2 , respectively (Fig. 9b), typical of values characteristic of desert aerosol. Figure 9c shows the size distribution on 1 August, with a maximum coarse mode concentration of 0.25 µm 3 /µm 2 , two orders of magnitude higher than on 29 July. The values for 3 August (Fig. 9d) and 4 August (not shown) decreased to those found for 29 July. For these two days, retrievals from both the sphere and spheroid models do not differ significantly. In addition to the two-orders of magnitude change in the volume size distribution during the dust event, the modal coarse radius decreased to 2-3 µm, compared to 5-6 µm for typical background conditions. Results for desert aerosol type size distributions are discussed in Dubovik et al. (2002a; (2001) are also not directly comparable to our findings, as they were obtained close to the aerosol sources. Smirnov et al. (1998) reported data from the ACE2 experiment at Tenerife (Canary Islands) at two different altitudes (sea level and 2360 m at Izaña) during dust outbreaks. At Izaña, as an illustration, the AOD changes from 0.02 to 0.2, while the coarse mode size distribution concentration changed by two orders of magnitude and the modal radius changed from 0.8 to 2 µm. Thus, our volume concentration values of ∼ 0.2 µm 3 /µm 2 for the coarse mode seem to be consistent with those reported by others (Smirnov et al., 1998;Dubovik et al., 2002a, b).
According to Dubovik et al. (2002a), retrieval of the refractive index is only reliable when AOD is greater than 0.4. This criterion was satisfied only for 31 July-2 August. Refractive index values are shown in Fig. 10 for both the sphere and spheroid retrieval models. Again, the sphere model values are generally valid for nondust conditions, while the spheroid model is better for dust. Similar features for both models are indicated in the real and imaginary parts of the refractive index, although with different values. The real part of the refractive index, n R , ranges from 1.43-1.53 and the imaginary part, n I , from 0.0013 to 0.006.
For reference, Dubovik et al. (2002a) give a range of values between 1.48-1.56 for n R and 0.0006-0.001 for n I (at 870 nm). Our values for n R are within the same interval, but our values for n I are greater. We comment on aerosol absorption later when discussing single scattering albedo (SSA) values.
With respect to the wavelength variation of n R , we also found for the sphere model (see Fig. 10c for 2 August) the same dependence (increase with wavelength) described by Dubovik et al. ( , 2002a. The values of 1.33 for n R at 440 nm, however, seem unrealistic. In contrast to the sphere model, the spheroid model shows an inverse wavelength behaviour, with values decreasing with wavelength and a relatively large value at 440 nm. The two models agree for the two longer wavelengths (870 and 1020 nm, respectively).
The values for g and SSA at 870 nm are shown in Fig. 10b for the sphere and spheroid model retrievals, with g ranging from 0.64 to 0.71 and SSA from 0.88 to 0.97 during the three days. No significant differences are seen between the different model retrievals. Also, a slight diurnal dependence can be observed in the results (Fig. 10c) for SSA and to a lesser extent for g. In Fig. 10d for 2 August, we see an increase in SSA with wavelength, which seems to be a characteristic of desert aerosol (Dubovik et al., 2002a). Both sphere and spheroid model results agree except for low SZA (thick dotted line in Fig. 10d). This result indicates that SSA is well retrieved and not affected by the problems relative to the accumulation mode of the size distribution and real refractive index mentioned above.
Because aerosol light absorption is a key parameter in estimating climate aerosol forcing, comparison of our data with those of other studies is of great interest. As shown by Hansen et al. (1997) a change in SSA from 0.95 to 0.85 can change the radiative forcing from negative to positive, depending on the underlying surface. Although our values are quite variable, they are within the range obtained by Dubovik et al. (2002a), but indicating slightly high absorption. Indeed, Dubovik et al. (2002a) discuss the high SSA (less absorption) values retrieved at representative desert sites with respect to earlier established values in the literature (e.g. D' Almeida et al., 1991;Hess et al., 1998;WMO, 1983). As another, more recent example, Haywood et al. (2003), report on SSA values obtained from data collected during the SHADE campaign and using model calculations based on measured size distributions and n I in the range 0.008-0.0015. For n I =0.008, the retrievals for SSA are not as reliable as those for n I =0.0015.
Satellite measurements also indicate less absorption than older literature for desert aerosols, as discussed by Kaufman et al. (2001) and Sinyuk et al. (2003). Because of this tendency observed in recent studies, it is clear that additional measurements are needed for the assessment of dust aerosol absorption properties. Our study certainly shows the variability of desert dust arriving over the southwestern part of the Iberian Peninsula as a result of the frequency and intensity of these events. We conclude that further continuous effort in monitoring and characterization is needed in this region. Such work is now being carried out by the AERONET-PHOTONS network.

Conclusions
Detection and evaluation of desert dust episodes in the southern part of the Iberian Peninsula during 2000 have been accomplished by means of three different types of measurements and analysis procedures, with remarkable agreement among them. Despite the known year-to-year variability, this evaluation gives some insight into the frequency of dust episodes and the number of days per year affected by desert dust over this region. It also provides information on the range of values of specific aerosol parameters such as AOD and α.
In general, the events at the end of winter are more intense, but less frequent and shorter in duration than those of summer which have a more extended and diluted dust plume, reinforced by low rainfall and re-circulations. While we have observed many one-day events with desert aerosol characteristics, it is often difficult to assign any of these days as belonging to a particular episode.
The detailed characterization of two dust case studies, using a combination of microphysical (particle size distribution and derived parameters) and optical properties, allowed the www.ann-geophys.net/24/1493/2006/ Ann. Geophys., 24, 1493-1510, 2006 determination of the range of these parameters in this coastal area of southwestern Spain.