Aerosol extinction in a remote continental region of the Iberian Peninsula during summer

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 111,, doi: /2005jd006610, 2006 Aerosol extinction in a remote continental region of the Iberian Peninsula during summer Thierry Elias, 1 Ana Maria Silva, 2 Nuno Belo, 2 Sergio Pereira, 2 Paola Formenti, 3 Günter Helas, 4 and Frank Wagner 2 Received 23 August 2005; revised 1 December 2005; accepted 15 February 2006; published 20 July [1] Summer in Évora ( N, W), Portugal, is described in terms of aerosol properties of extinction of the solar radiation. We create a data set composed of (1) cloudscreened half-day averaged values of aerosol optical thickness (AOT) measured at 7 wavelengths by both a CIMEL Sun/sky-photometer and a YES shadowband radiometer and (2) half day averaged values of aerosol scattering coefficient (ASC) measured at the surface level at two wavelengths by a TSI nephelometer. Spectral dependence of both AOT and ASC gives the column and the surface Ångström exponents, a C and a S, respectively. Measurements are acquired in both 2002 and 2003 summers. Back trajectories are computed. A statistical study of the data set provides thresholds in AOT and a C for a classification of the days. The classification is applied with success to the case study of the 2003 summer heat wave episode and is generalized to the whole data set. In 23% of the cases, the turbidity in Évora is very low, with AOT441 < 0.12 and AOT873 < The air mass origin is the North Atlantic Ocean at 700 and 970 hpa. In 31% of the cases, the turbidity is high. Increase of AOT is due to forest fire emissions, originating in the Iberian Peninsula, with 0.30 < AOT441 < 1.10 and a C > 1.2, and to desert dust plumes transported from North Africa within 72 to 120 hours at 700 hpa, with 0.10 < AOT873 < 1.10 and 0.1 < a C < 1.0. The vertical profile is highly variable, and several cases of aerosol mixing in the column are identified. The duration of the aerosol episode during the 2003 summer heat wave is 16 days, which is exceptionally long. Citation: Elias, T., A. M. Silva, N. Belo, S. Pereira, P. Formenti, G. Helas, and F. Wagner (2006), Aerosol extinction in a remote continental region of the Iberian Peninsula during summer, J. Geophys. Res., 111,, doi: /2005jd Introduction [2] The effect of aerosols on the Earth s radiative budget must be investigated in the issue of the current climate change related to the human activities [Intergovernmental Panel on Climate Change, 2001; Kaufman et al., 2002]. Aerosols increase the atmospheric scattering and absorption of the solar radiation, attenuating the direct solar radiation at the Earth s surface and modifying the radiation reflected back to space by the surface-atmosphere system. The aerosol radiative forcing expresses the aerosol effect on the Earth s radiative budget. The magnitude of the aerosol radiative forcing is directly proportional to the aerosol optical thickness (AOT), and AOT depends on both the aerosol concentration and nature (particle size, composition, shape). Since the aerosol concentration and type are both 1 Centre National de Recherche Meteorologique/Groupe de Météorologie à Moyenne Échelle, MétéoFrance, Toulouse, France. 2 Centro de Geofísica de Évora, Universidade de Évora, Évora, Portugal. 3 Laboratoire Interuniversitaire des Systèmes Atmosphériques, Faculté des Sciences et Technologie, Créteil, France. 4 Biogeochemistry Department, Max Planck Institute for Chemistry, Mainz, Germany. Copyright 2006 by the American Geophysical Union /06/2005JD highly heterogeneous in space and time, aerosol observation requires a high spatial and temporal resolution over the globe to compute the aerosol radiative forcing in the climate system. Satellite observation is the designated technology to meet such an objective but a high level of resolution in both time and space cannot be achieved with a unique satellite. Constellation of spaceborne instruments might fulfill the requirement in future. However the uncertainty in satelliteinferred values of AOT remains high over land [Chu et al., 2002; Remer et al., 2005], where the surface contribution and heterogeneity are high, but also over ocean [Chylek et al., 2003]. To investigate the aerosol properties with high precision, ground-based observation is required. [3] Ground-based radiometric observation of radiation in the solar direction provides AOT at several wavelengths with a high temporal resolution and an uncertainty lower than 0.02 [Holben et al., 1998; Schmid et al., 1998]. Network of ground-based sites are constituted to compensate the spatial resolution limited to one point, as for example the Aerosol Robotic Network (AERONET). The site locations must be chosen in order to represent the global variability of the aerosol properties, which depends on many factors as the distance from the site to the aerosol sources, the strength and nature of the aerosol sources, the synoptical conditions (air mass motions), the meteorology 1of20

2 (humidity, precipitation). Because of their absorbing properties, desert dust and biomass burning particles are among the most important aerosol types to consider for aerosol radiative forcing calculations. [4] This paper presents aerosol data from the new ground-based station of Évora ( N, W), Portugal, which started to operate in 2002 and became part of the AErosol RObotic NETwork (AERONET) of Sun/skyphotometers [Holben et al., 1998] in July The objective of the Centro de Geofísica de Évora (CGE), with starting aerosol measurements in Évora, are several: (1) to characterize properties of remote continental background aerosols; (2) to generate an aerosol climatology for the southwest Iberian Peninsula; and (3) to evaluate the scientific interest of measurements performed in Évora in resolving uncertainties on main aerosol types encountered over the globe, in particular the desert dust particles and the biomass burning emissions. [5] Évora is the capital of a rural region, Alentejo, covering a third of the country territory but with only inhabitants. Évora being far from any industrial and urban sources of aerosols and at more than 150 km from the Atlantic coast, the background level of turbidity is low and must represent continental background conditions. In these conditions, the measurements should be sensitive to any departure of the aerosol load. The intensity of forest fires in summer 2003 in Portugal was very high [Barbosa et al., 2004], injecting in the atmosphere large quantity of particles emitted by the biomass combustion. The effect of such biomass burning products should be detected in Évora. Portugal being located at the most south western point of Europe, Évora is also expected to be a major site in surveying the intrusion in Europe of desert dust transported from Africa. Ansmann et al. [2003] report a North African dust plume reaching northern and central Europe through the Mediterranean Sea and the Iberian Peninsula in October Desert dust was also observed during the VELETA 2002 field campaign that took place in July 2002 in south Spain [Alados-Arboledas et al., 2003, 2004]. Dust can also cross the Iberian Peninsula along a west-east line after overpassing the North Atlantic Ocean. Évora can be a milestone completing the AERONET sites of Dakar ( N, W), Cape Verde ( N, W), Barbados ( N, W), Canary islands ( N, W), which survey the evolution of the desert dust during the transport over the North Atlantic Ocean [Chiapello et al., 1999]. Outbreaks of European pollution were also observed in Portugal, as in Sagres [Ansmann et al., 2002; Silva et al., 2002] during the field campaign ACE-2 in June July 1997 [Raes et al., 2000]. By covering part of the western region of the Iberian Peninsula, Évora also complements a dense Iberian network composed by the stations of Palencia ( N, W), Valencia ( N, W), Boecillo (41.54 N, 4.7 W), Granada (36.18 N, 3.6 W), and El Aresonillo ( N, W) [Cachorro et al., 2000; Martinez-Lozano et al., 2001; Pedrós et al., 2003; Estellés et al., 2004a]. [6] The Évora station provides with surface in situ measurements of the aerosol scattering properties by a three-wavelength nephelometer, and with column-integrated aerosol extinction measured by both a MFR-7 shadowband radiometer and a CIMEL CE318-2 Sun/sky-photometer. Aerosol scattering and extinction measurements are made at several wavelengths for providing information on the aerosol size and consequently on the aerosol type. As Évora experiences a high insulation rate during the summer months because of sparse cloudiness, aerosol measurements were made intensively during 2002 and 2003 summers. [7] This paper presents the first study of the main aerosol types detectable in Évora, in terms of magnitude and spectral dependence of both the column aerosol extinction and the surface aerosol scattering, and in terms of occurrence and duration of the aerosol episodes. The surface in situ aerosol scattering coefficients (ASC) are approximated to aerosol extinction by neglecting the aerosol absorption. The aerosol turbidity conditions are classified in function of the intensity of the event (AOT and ASC magnitudes) and the mean size of the particles (Ångström exponent). Air mass back trajectories are computed for verifying the coherence between the air mass origin and the identified aerosol type. Association of surface in situ and radiometric column measurements informs about the aerosol vertical distribution in Évora. Available data sets and analysis are similar to what is presented by Gerasopoulos et al. [2003]. The objective of the paper is twofold: (1) to propose a classification based on aerosol optical thickness and Ångström exponent thresholds to characterize all situations encountered in Évora and (2) to give a first picture of the aerosol properties in Évora, enclosed in the remote continental Iberian Peninsula. [8] The sampling site is described in section 2. Instrumentation is described in section 3, as well as the experimental procedures and results from the calibration survey. The data set of aerosol extinction properties measured by the three instruments is defined in section 4. Comparing measurements made by several instruments when they are made at common wavelengths ensures the data quality. Back trajectory computations are discussed in section 5. Aerosol episodes which occurred in Évora during both 2002 and 2003 summers are commented in section 6.3 with an emphasis on the case study of the 2003 summer heat wave in section 6.2, using the classification defined in section 6.1. Discussion and summary of results is delivered in section 7 and conclusions are given in section Site Description [9] Aerosol and basic meteorological observations (temperature, wind speed and direction, relative humidity) were performed at the observatory of the CGE at the University facilities, located within the town center of Évora ( N, W, 300 m above sea level), Portugal. Évora is the capital of the Alentejo, a rural region which occupies one third of the Portuguese territory and which is inhabited by less than 10% of the entire Portuguese population. Évora is the biggest town of the region with a population of about inhabitants. Local pollution is due only to private cars, as no big industries work in the area. The nearest big urban and industrialized area is Lisbon at 100 km away from Évora, with a population of approximately 1 million, which is surrounded by the industrial belt of Great Lisbon. A further significative industrialized area is Sines, situated on the Atlantic coast at 100 km south of Lisbon and at around 200 km from Évora. The Atlantic Ocean is at 150 km 2of20

3 Figure 1. Seven geographical sectors used for the classification of the back trajectories ending at the Évora monitoring site (indicated by a dot) are drawn onto the map of western Mediterranean, including Portugal, Spain, France, and parts of Morocco and Algeria. westward of Évora and the Spanish border at 100 km eastward (map of Figure 1 gives an indication of Évora location). The localization far from particle sources furnishes excellent conditions to observe the local impact of long- and middle-range transport of particles. [10] Rainy seasons in Alentejo are October December and March April. The weather is particularly stable in summer with low air humidity and high air temperatures reaching beyond 40 C a few days per year. [11] Évora was the unique station of aerosol remote sensing in continental Portugal from January 2002 to December 2003 when a second station was set up in Cabo da Roca (38.78 N, 9.05 W), on the Atlantic coast at 30 km north of Lisbon. Up to that time, the closest AERONET station was El Aresonillo ( N, E), south of Spain. 3. Instrumentation and Measurement Processing [12] Aerosol measurements were performed in Évora with three ground-based instruments: a three wavelength integrating nephelometer, a multifilter rotating shadowband spectral radiometer, and a spectral Sun/sky-photometer. The instruments and the associated experimental procedures are described in sections 3.1, 3.2, and 3.3 respectively. Measuring periods in are summarized in Table 1, and acquisition wavelengths in Table 2. [13] Calibrations were performed regularly for ensuring a good quality of the measurements. Moreover each instrument was sent to the respective manufacturing company at least once during 2002 and MFR-7 and CE318-2 measurements performed in Évora during low turbidity conditions are used for Langley plots. CE318-2 calibration coefficients were also compared with calibration coefficients of other instruments collocated during the calibration experiment, which occurred during the first week of the VELETA 2002 campaign [Alados-Arboledas et al., 2003, 2004]. The calibration of the instruments is also reviewed in sections 3.1, 3.2 and Three Wavelength Integrating Nephelometer [14] A three wavelength integrating nephelometer (TSI- 3563, TSI Inc., St Paul, Minnesota, USA), starts acquisition in Évora in The optical chamber is set up in the university facilities while ambient air is pulled in by a vacuum pomp along an outdoors 10-m vertical tube. A PM10 filter (PMX inlet, SVEN LECKEL ING:, GmbH, Germany) selects particles smaller than 10 mm aerodynamical diameter. The flow rate was fixed at 30 l.min 1. The air Table 1. Running Period Times of the Three Instruments of the Évora Station During 2002 and 2003 a Instrument Winter Spring Summer Autumn Winter Spring Summer Autumn TSI-3563 yes yes yes yes yes yes yes yes MFR-7 yes yes yes yes yes CE318-2 yes yes yes a Yes means that measurements were performed during the indicated season. 3of20

4 Table 2. Acquisition Wavelengths of the Three Instruments of the Évora Station a Instrument TSI-3563 yes yes yes MFR-7 yes yes yes yes yes CE318-2 yes yes yes yes a Wavelengths are in nm. Yes indicates the acquisition wavelengths of each instrument. sample in the optical chamber is not heated, except because of the heat dissipation of the lamp, in order to keep as close as possible to the ambient conditions. Measurements provide the particle scattering and backscattering coefficient at 450, 550 and 700 nm (Table 2), continuously day and night. Measurements are stored as 2-min averages. [15] The operating principles of the TSI nephelometer as well as the methodology used to extract the aerosol scattering coefficients (ASC) is described by Anderson and Ogren [1998]. The collected air sample is lighted within the optical chamber of TSI Light scattered from 7 to 170 is measured and the aerosol scattering coefficient is derived after correction by the calculated molecular scattering and for the angular truncation and the non-lambertian errors [Anderson and Ogren, 1998]. The relation between scattered light and scattering coefficient depends on the instrument optics and calibration needs to be performed regularly to check the optics quality. Calibration was performed several times during the acquisition time period by measuring successively known light scattering by both 99.9% pure CO 2 gas and dry air. TSI-3563 has been calibrated three times: in July 1997 during the ACE-2 campaign in Sagres, Portugal; in August 2001 by TSI company in USA; and in December 2002 in University of Évora. The K 2 coefficients (Table 3) define the proportionality relationship between the aerosol scattering coefficient and the measured raw signal. The instrument optics remains stable since the coefficients vary of less than 2% between two successive calibrations, except at 700 nm from 2001 to 2002 when the coefficient varies of 10%. Moreover it was noted that the signal-tonoise ratio at 700 nm is too bad when the turbidity is small, consequently the scattering coefficient at 700 nm will not be considered Multifilter Rotating Shadowband Radiometer [16] A multifilter rotating shadowband radiometer (MFR-7 Yankee Env. System Inc., Turner Falls, MA, USA), measures downwelling global and diffuse irradiances in a half-space field of view, I glo and I dif respectively, in five narrowband channels (10 nm bandwidths) at 416, 500, 613, 670, and 867 nm (Table 2). Description is given by Harrison et al. [1994]. Measurements are acquired every 15 s and stored as 1-min averages. The acquisition started in February The data set presents a gap between February 2003 and July 2003 when the instrument was calibrated at the manufacturer company (Table 1). [17] The diffuse irradiance is subtracted to the global irradiance to provide the intensity I dir of the direct solar beam attenuated by the atmosphere. The atmospheric total optical thickness t tot is related to the intensity I dir by the Beer s law: I dir =K esd I 0 exp ( m.t tot ) where I 0 is the solar irradiance at the top of the atmosphere and K esd the Earth- Sun distance corrective factor. The optical air mass m is calculated using the formula of Kasten and Young [1989] in function of the acquisition time and the site coordinates, the atmospheric refraction and the effects of Earth s sphericity on the atmospheric path radiance. The Beer s law can be expressed in units of numerical counts, relating similarly the measurement V dir to the atmospheric property t tot through the calibration coefficient V 0 which has to be determined experimentally at each wavelength by the Langley linear regression technique [e.g., Harrison and Michalsky, 1994; Michalsky et al., 2001]. The Langley technique is applied to spectral morning and afternoon measurements made from January to September 2002 and in 2003, for optical air masses m between 2 and 6 and whenever stable low turbidity conditions prevailed in Évora: no cloud covering, low aerosol amounts and low wind speed. Knowing V 0 and m, the atmospheric spectral optical thickness t tot can be derived from V dir measured in any conditions. Molecular optical thickness calculated according to the Rayleigh theory [Hansen and Travis, 1974] is removed from the atmospheric optical thickness t tot. The atmospheric pressure is considered constant at 970 hpa. The gaseous absorption due to ozone is estimated by the atmospheric content provided by Earth-Probe Total Ozone Mapping Spectrometer [McPeters et al., 1998]. Eventually, AOT is obtained. The method is more extensively described by other authors [e.g., Schmid et al., 1998]. [18] Averaged values of V 0 as well as standard deviations are reported in Table 4a. Around fifty values of V 0 are obtained from measurements performed on morning and afternoon in Standard deviation is small, decreasing with the increasing wavelength. The averaged values are very close to the value derived from measurements on 22 March 2002 morning, when the aerosol optical thickness was exceptionally low, with AOT500 < 0.04 and AOT867 < 0.02, which is close to the uncertainty level. No differences are found between morning and afternoon values of V 0. The drift of the calibration coefficients in one year is smaller than 5%, decreasing to 1% with increasing wavelength. According to Michalsky et al. [2001], the uncertainty on AOT is smaller than [19] The YES Company also delivers spectral sensitivity coefficients necessary to convert raw counts into irradiances in Wm 2 nm 1. The experimentally estimated extraterrestrial solar irradiance I 0 is compared to values given by Neckel and Labs [1981, 1989] in Table 4b. The agreement is within 10%, which is excellent according to the numerous sources Table 3. Time Evolution of the TSI-3563 K 2 Calibration Coefficients at Three Wavelengths a Date 450 nm 550 nm 700 nm July E E E-03 August E E E-03 December E E E-03 a Calibration coefficients are in s/count. 4of20

5 Table 4a. Calibration Coefficients V 0 for MFR-7, Derived at 5 Wavelengths by the Langley Plot Technique a Date 416 nm 500 nm 613 nm 670 nm 867 nm 22 March /51/ /55/ /58/ /53/ /41/ /25/ /32/ /22/ /13/ /11/1.3 a Averaged value of V 0 in numerical counts, the number N of values as well as the corresponding standard deviation sd in % are given as V 0 /N/sd. of uncertainties. The agreement is particularly good at 416, 500 and 867 nm Multiwavelength Sun/Sky Photometer [20] A multiwavelength Sun/sky photometer (CE318-2, CIMEL Electronique, France) is set up on the observatory, next to the MFR-7. In 2002, the instrument runs from mid June to the beginning of October with a gap of three weeks in July when it was deployed in south of Spain during the field campaign VELETA [Alados-Arboledas et al., 2003]. The instrument was sent for calibration by AERONET/ PHOTONS in November Acquisition started again on July 2003, when the site became an AERONET station, and works since then. [21] Holben et al. [1998] provide a full description of the acquisition procedure by CE Direct solar radiance as well as sky radiance are measured at 441, 671, 873 and 1021 nm (Table 2) with a full field of view of 1.2. The acquisition is automatic and the measurement sequence depends on the local solar time and on the value of the optical air mass. Measurements are made at an approximate rate of 15 min with a higher frequency when the Sun is close to the horizon. An electronic test is made on the direct solar beam measurements to check the presence of thick clouds: in case the attenuation is large, measurements are not made. Uncertainty in AOT estimated by Holben et al. [1998] is included between 0.01 and Data are automatically downloaded to a computer in Since July 2003, data are automatically sent via an emitter antenna to NASA facilities and can be found on nasa.gov/. [22] In contrary to MFR-7, CE318-2 measures the atmospherically attenuated direct irradiance I dir by pointing to the Sun with a narrow field of view optics. However the procedure to derive the aerosol optical thickness from I dir is equivalent to the procedure used for the MFR-7. By combining the two instruments, aerosol optical thickness is obtained at 416, 441, 500, 613, 670, 671, 867, 873 and 1021 nm (Table 2). [23] Calibration experiments are performed during conditions of very low turbidity (AOT441 < 0.10 and AOT873 < 0.04), which occurred in Évora in 12 occasions in end of June and early July Averaged value of V 0 and associated standard deviations are given in Table 5. The averaged V 0 values obtained from measurements made in Évora are retrieved with a high confidence, as the standard deviation is smaller than 2% except at 441 nm where it is 4%. Moreover the instrument participated to the VELETA 2002 campaign, which started by a several-day-long calibration precampaign when several instruments were collocated at a high-altitude site where turbidity is generally low [Alados-Arboledas et al., 2004]. The VELETA 2002 calibration coefficient delivered after intercomparison with an instrument calibrated by AERONET, set as a reference, is reported in Table 5. The difference of the VELETA 2002 value with our average is very small, confirming the confidence in the coefficients used to derive the aerosol optical thickness for the 2002 data. For 2003 on, the calibration coefficients determined by AERONET are considered and no independent experiment campaigns are performed in Évora. The VELETA 2002 comparison of the aerosol optical thickness given by the CGE instrument with an instrument calibrated by AERONET gives a good agreement, within the AOT uncertainty [Estelles et al., 2004b]. 4. Definition of the Data Set 4.1. Half-Day Averaging and Cloud Screening [24] Since the nebulosity is lowest and the number of working instruments is highest (Table 1), the analysis is restricted to measurements acquired during both 2002 and 2003 summers, and more largely to the months of June, July, August and September. [25] Aerosol property measurements were acquired automatically, simultaneously and at the same location, by the three manufactured instruments. Particular processing is necessary for creating a consistent data set from the three measurement sets. While the instruments can provide a wide set of physical quantities, only properties related to aerosol extinction are considered here and not the angular scattering properties. TSI3563 provides the spectral scattering coefficients, which can be related to extinction if the aerosol single scattering albedo is assumed (it is assumed constant at 1.0 for the paper). The spectral aerosol optical thickness derived from measurements made by the two optical radiometers MFR-7 and CE318-2 is retained. Backscatter coefficient, sky radiance, flux measurements (all quantities Table 4b. Extraterrestrial Solar Irradiance I NL Given by Neckel and Labs [1981, 1989] and Differences With Estimations (as (I 0 -I NL )/I NL ) Derived From Both the MFR-7 Calibration Coefficients V 0 (Table 4a) and the MFR-7 Sensitivity Coefficients Delivered by the YES Company Date 416 nm 500 nm 613 nm 670 nm 867 nm Nekel and Labs, Wm 2 nm , % , % of20

6 Table 5. Calibration Coefficient V 0, Number of Measurements and Standard Deviations Given for CE318-2 at Four Wavelengths From Langley Plot Techniques Performed on Measurements Made in Évora in 2002 (as Table 4a) a Location 441 nm 671 nm 873 nm 1021 nm Évora 5752/12/ /12/ /12/ /12/2.2 VELETA a The VELETA 2002 value of V 0 is also given. related to angular distribution of solar radiation by particles) are disregarded in this study. The methodology to generate a consistent data set from three kinds of measurements is presented. [26] The effect of cloud contamination is treated differently for MFR-7 and CE318-2 data. Electronic cloud screening exists for the CE318-2, which however is not sufficient to get rid of all cloud contaminations. On another hand, the MFR-7 provides 1-min data whatever the cloud coverage. A further downstream cloud screening is necessary, which must be common for both MFR-7 and CE318-2 for sake of data consistency. Clouds are characterized by a high optical thickness of low spectral dependence. Moreover temporal variation of optical thickness is large if the cloud cover is heterogeneous: therefore we rejected measurements (the whole spectral series) for which jdaot1021/ Dtj> 0.5 h 1 (where Dt is in hours). This is similar to the principle of the Smirnov et al. [2000] method. This threshold has been tested on Évora measurements, and has been judged sufficient for detecting cloud effect in Évora. [27] Aerosol property measurements were acquired automatically, simultaneously and at the same location by the three instruments but the temporal resolutions of TSI3563, MFR-7 and CE318-2 are very different. In order to uniformize the temporal resolution of the data set, all measurements are averaged over mornings and afternoons, that is from sunrise to 1200 UT and from 1200 UT to sunset for the cloud-screened MFR-7 and CE318-2 data, and over and UT for TSI3563. Corresponding standard deviation is also calculated to express the temporal variability of AOT and ASC during the averaging period. A further test is applied on the standard deviation of AOT1021, which is limited to 75% of averaged AOT1021 for acceptance (and for AOT1021 > 0.05 as the measurement error is ). The averaged value is kept when more than 5 acquisitions in morning or in afternoon are available. After cloud screening, averaging and the test on standard deviation and measurement numbers, 137 values over around 90 days of CE318-2 operating remain for 2002 and 145 values for 2003 (over an equivalent number of operating days), which confirms the minor influence of clouds in summer. [28] The final data set is made of morning and afternoon averages of the scattering coefficient at two wavelengths, and morning and afternoon averages of AOT at 9 wavelengths. These data are used to calculate the surface Ångström exponent a S from the scattering coefficients at 450 and 550 nm, and the column Ångström exponent a C from the aerosol optical thickness measured by each optical radiometer, using a least square fit over all channels. [29] When measurements were made at common wavelengths by two instruments, correlation is studied to ensure a consistency of the data set acquired by the whole station. Comparison of CE318-2 columnar quantities with TSI-3563 surface in situ measurements requires considering independent characteristics, which are not measured (for example aerosol absorption, uniformity within the aerosol layer). Only qualitative conclusion can result of such a comparison. However, comparison of CE318-2 and MFR-7 measurements is straightforward since aerosol optical thickness is derived from both instruments. Sections 4.2 and 4.3 present these two instrumental intercomparisons Comparison MFR-7/CE318-2 [30] The aerosol optical depth measured simultaneously by MFR-7 and CE318-2 at and nm are intercompared in order to check the consistency of the data set. The scatterplot of the aerosol optical thickness measured at 671 nm (AOT671) by CE318-2 against AOT670 measured by MFR-7 is shown in Figure 2a for the 2002 (circles) and 2003 (triangles) summers. The equivalent scatterplot for the aerosol optical thickness measured at nm is shown in Figure 2b. [31] The range of values extending over one order of magnitude, the linear fit for each summer is calculated. Parameters of the fit are reported in Table 6. The agreement between both instruments is very good. At both wavelengths, the correlation coefficient is larger than The slopes are similar, and close to 1 for both years. The intercepts are close to 0.01 and smaller than the measurement uncertainty. The linear fit stays within the measurement uncertainty for the whole range of values. The averaged differences between AOT670 and AOT671 and between AOT867 and AOT873 are all smaller than 0.01 and the standard deviations are all smaller than [32] The minor disagreement between the two instruments is not due to the calibration coefficients since the behavior is closely similar over the two years and the calibration coefficients of both instruments were modified between both summers. [33] The scatterplot of the Ångström exponents obtained from the two data sets is shown in Figure 2c. The linear fit has a slope close to 1, confirming that the disagreement has no clear spectral dependency. This observation justifies the argument that the cause of the disagreement is instrumental and is not linked to the calibration. The correlation coefficient is 0.90 (Table 6) (Tables 7a and 7b), which is not as good as for the aerosol optical thickness. This is expected as a C is calculated from two measurements and combines two sources of uncertainty. Since the uncertainty of a C increases when the aerosol optical thickness decreases, cases of low turbidity were removed of the plot. No temporal drift is applied on the calibration coefficients, which could be a cause of the minor difference between CE318-2 and MFR-7. 6of20

7 However such inaccuracy would not appear on spectral data, as a similar drift would be applied at any wavelengths. The averaged difference in a C between both instruments is smaller than 0.1 and the standard deviation is smaller than 0.2. This comparison study shows that the agreement is excellent between CE318-2 and MFR7 measurements Comparison TSI-3563/CE318-2 [34] In Figure 3, the aerosol scattering coefficients measured at 450 nm (ASC450) by the TSI-3563 are related to the aerosol optical thickness measured at 441 nm by the CE318-2 and for the 2003 summer. Triangles represent morning averages and circles represent afternoon averages, and desert data are represented by large empty triangles or circles. A linear regression is applied separately on morning and afternoon data. 30 July and 2 August afternoons and 23 August morning were eliminated as they clearly show an aerosol event at surface level (section 6.2). The correlation is good, with a correlation factor of 0.8 for both linear fits. Assuming that the aerosol single scattering albedo is constant and equal to 1, and that the aerosol scattering coefficient is constant within the whole aerosol layer, the slopes give an indication on the aerosol layer height. The mean aerosol layer height is estimated to be 2 km for mornings and 2.8 km for afternoons. The difference between morning and afternoon values could be due to the ascendance of the boundary layer top along the day when the surface air is heated by the Sun and lifts up. These values are similar to those obtained by Bergin et al. [2000] at a continental site (of 3 km), and by Formenti et al. [2001] for a coastal site during summer (of 3.4 ± 0.3 km). Points corresponding to desert dust (DD) are situated at the top of the point cluster, for AOT larger than the fitted value for an equal scattering coefficient, showing that the aerosol scale height is larger for desert dust than for the other cases, which is also expected. [35] The range of data is not sufficient in 2002 to derive a linear fit but the averaged ratio AOT441 over ASC450 is calculated, giving a scale height similar to that of 2003, and with an equivalent difference between morning and afternoon values. These results show an acceptable consistency between AOT and ASC values. 7of20 5. Air Mass Back Trajectory Calculations [36] Daily 72-hour back trajectory of air masses ending over Évora at 1200 UT are calculated using the Hybrid Single-Particle Lagrangian Integrated Trajectory Model (HYSPLIT) [Draxler and Rolph, 2003] to interpret data in terms of the aerosol type, according to their origins. Two pressure levels are chosen: (1) 970 hpa, which is the surface level, to represent the air mass sampled by the nephelometer, and (2) 700 hpa (at about 3 km height) for the longrange transport of particles (as Saharan desert dust). On the basis of the most frequent synoptic situations affecting Figure 2. (a) Correlation of AOT measured by CE318-2 with AOT measured by MFR-7 at nm in both 2002 (circles) and 2003 (triangles) summers. Linear fits are also plotted for each year. Coefficients are y = 1.11x 0.010, correlation factor cf = 0.95 in y = 1.08x 0.010, cf = 0.99 in (b) Same as Figure 2a but at nm. y = 1.05x 0.012, cf = 0.98 in y = 1.08x 0.006, cf = 0.99 in (c) Same as Figure 2a but the Ångström exponent is plotted. y = 0.92x , cf = 0.89 in y = 1.05x , cf = 0.90 in 2003.

8 Table 6. Parameters of the Linear Correlation Between CE318-2 and MFR-7 Spectral Extinction Measurements AOT670 AOT873 a C Parameters Correlation coefficient Slope Intercept continental Portugal during the summer periods [Belo, 2004], back trajectories are classified into seven geographical sectors (Figure 1): (1) local (LO), which is representative of the air masses originating and staying for 72 hours in the rural surrounding of Évora, within a range of about 100 km radius; (2) Iberian Peninsula (IB), for air masses originating and staying over Spain and Portugal, except the local sector; (3) Europe (EU), for trajectories in northeast of Portugal, representative of air masses originating in the European continent (from Scandinavia to the northern coasts of the Mediterranean Sea), which stay most of the time over land; (4) Arctic (AR), in northern direction, for air masses passing over Great Britain and also for air masses under the unique influence of the high latitudes of the North Atlantic Ocean; (5) North Atlantic (NA), in western direction, for air masses coming from the mid latitudes of North Atlantic Ocean and from the northern American continent; (6) tropical Atlantic (TA), in southwestern direction, for air masses coming from the tropical latitudes of the North Atlantic Ocean and from Africa before overpassing the North Atlantic Ocean; and (7) Africa (AF), in southern and southeastern directions, to represent the influence of African continent as well as the influence of the southern part of the Mediterranean Sea. [37] Air mass origins at 970 and 700 hpa in 2002 and 2003 summers are analyzed statistically. Results are presented in Figure 4. The main influence in the air mass origin over Évora is the North Atlantic Ocean, i.e., AR, NA and TA sectors, with accumulated 80% of occurrence at 700 hpa and 65 80% at surface level. The second important air mass influence is from both local region and the Iberian Peninsula, with 13 21% at 970 hpa and 4 10% at 700 hpa. Following, comes the influence from Africa with 7 9% at 700 hpa and 1 9% at 970 hpa. It must be noted that African air masses can reach Portugal directly in southerly winds and also after crossing the tropical Atlantic Ocean, which can take more than 72 hours and be consequently classified in TA. This means that the influence of African air masses may exceed 10% in Évora. Figure 5 gives examples of back trajectories of air masses loaded with desert dust (see Table 8) arriving in Évora at 700 hpa, over 72 (Figure 5a) and 120 hours (Figure 5b). Desert dust reaches Évora in (1) 72 hours with northerly winds from Morocco, as on 30 July to 2 August 2003 and 5 August 2003, and in (2) 120 and more hours through the Atlantic Ocean and in particular passing in the vicinity of Madeira Islands or Canary Islands, as on 3, 4, and 15 August [38] The European influence is smaller than 7% at 970 hpa and smaller than 4% at 700 hpa. No air mass at 700 hpa at Évora originates in Europe in IB and LO sectors have twice more influence at 970 hpa than at 700 hpa in both summers of 2002 and Large differences are noted between 2002 and 2003 for the air mass origins at 970 hpa, in what concerns the IB, EU, AF and TA sectors, and at 700 hpa in the frequency of the IB, EU and AR sectors 6. Aerosol Extinction in Évora 6.1. Classification of the Aerosols [39] Aerosol optical thickness measured by CE318-2 is studied on a statistical basis and histograms are presented in Figures 6a 6e. Averaged and cloud-screened values are used. Histograms of AOT441 are showed in Figures 6a and 6b, histograms of AOT871 in Figures 6c and 6d and the histogram of a C in Figure 6e, for 2002 (solid line) and 2003 (dashes). The width of the classes in AOT is 0.1 in Figures 6a and 6c and is reduced to 0.02 in Figures 6b and 6d for a better resolution. [40] Several features are observed which allow drawing criteria for a classification of the aerosol turbidity situations encountered in Évora. The most frequent values of AOT are smaller than 0.30 at 441 nm (frequency up to 75%) and smaller than 0.10 at 871 nm (up to 65%), for both 2002 and 2003 summers (Figures 6a and 6c). These two values will be used as thresholds to distinguish the predominant aerosol type in Évora. Figure 6d with a class width of 0.02 confirms the 871 nm threshold at around It is 0.08 with the condition of a class occurrence larger than 10% in 2002 and 2003 (horizontal line in Figure 6d), but it is 0.12 in 2002 and 0.10 in 2003 if the occurrence condition is lowered to 6%. The higher-resolution histogram of Figure 6b shows that the 441 nm threshold is 0.30 in 2003 and 0.34 in 2002 if the condition on occurrence is 3%. It is fixed at 0.30 for the paper. Table 7a. Criteria for Distinction of the Aerosol Events During Summer in Évora, in Terms of Spectral Aerosol Extinction in the Atmospheric Column a Event Type AOT441 AOT873 a C Clean (Cl) <0.12 <0.04 NC Maritime (Ma) NC <0.10 <1.0 Continental (Co) NC >1.0 Forest fire emissions (FF) >0.30 NC >1.0 Desert dust (DD) NC >0.10 <1.0 a NC indicates no conditions on this quantity. Table 7b. As Table 7a but Criteria Based on the Spectral Aerosol Scattering Coefficient at Surface Level Event Type ASC450, Mm 1 ASC550, Mm 1 a S Clean (Cl) <40 <30 NC Maritime (Ma) NC <55 <1.0 Continental (Co) NC >1.0 Forest fire emissions (FF) >100 NC >1.0 Desert dust (DD) NC >55 <1.0 8of20

9 Figure 3. Correlation between the aerosol scattering coefficient at 450 nm (ASC450) measured at the surface level and the aerosol optical thickness at 441 nm (AOT441) measured for the whole atmospheric column during the 2003 summer. Morning and afternoon averages are distinguished as well as measurements made during the desert events (DD) on 29 July to 6 August, August, September, and September. Linear fits are plotted for point clusters of both afternoon (dashed line) and morning (solid line): y = x , cf = 0.82 morning; y = x , cf = 0.79 afternoon. [41] Figures 6b and 6d show that AOT441 is generally larger than 0.04 and AOT873 larger than An interesting feature is observed in the data set presented in Figure 6b. A deficit of values of AOT441 between 0.10 and 0.12 is observed in both summers. It is very strong in 2002 when the class occurrence decreases from 10 to 2% from the class to the class and increases back to 7% to the class. Moreover, the occurrence of the class is 2 to 5 times smaller than the occurrence of any other class between 0.04 and This feature leads to the determination of the threshold to distinguish the clean background situation for AOT441 < The threshold at 871 nm for the same situation will be defined at a value of 0.04 in order to ensure that the class is characterized by a low turbidity level. Figure 6e shows that most values of a C are included between 1.0 and (to a frequency of around 70%). A plateau is observed for a C between 0.4 and 1.0 at a frequency value of 5 10% for 0.1-width classes. Data are rare for a C < 0.4 and a C > 2.0. [42] The standard deviation over all data of AOT is high, showing a high variability of the atmospheric situations: <a C > = 1.2 ± 0.4 for 2002 (± standard deviation) and <a C > = 1.4 ± 0.4 for 2003, <AOT441> = 0.23 ± 0.18 for both summers and <AOT873> = 0.11 ± 0.14 for 2002 and <AOT873> = 0.10 ± 0.08 for [43] Two quantities can be sufficient to summarize multispectral measurements of aerosol extinction: the aerosol turbidity, which is represented by the aerosol optical thickness at one wavelength and the spectral dependence, which is represented by the Ångström exponent. Consequently, two criteria only are necessary to classify the turbid conditions. Since the wavelength of maximum sensitivity of the aerosol extinction is correlated to the particle size, smaller wavelength is used for detecting events dominated by accumulation mode particles and longer wavelength is used for events dominated by coarse mode particles. Threshold values on AOT and on Ångström exponent are determined for the qualitative distinction of the aerosol events that occurred in Évora. Their values are fixed from the general knowledge currently available on aerosol properties and from the observations made in Évora. Values are reported in Table 7a. Five categories of aerosol turbidity are defined for Évora. The value a C = 1.0 is used to decline the turbid situations into two categories. The situation is labeled forest fire emissions when AOT441 > 0.30 and a C > 1.0. It is classified as desert dust when AOT873 > 0.10 and a C < 1.0. The low turbidity level is also divided into continental for 0.12 < AOT441 < 0.30 and a C > 1.0 and into maritime for AOT873 < 0.10 and a C < 1.0. The fifth category is the clean situation, already referred to, for AOT441 < 0.12 and AOT873 < The classification of surface measurements is made according to the AOT classification. Thresholds on ASC450, ASC550 and a S are defined from the AOT thresholds by assuming an aerosol layer between 2 and 3 km. Values are given in Table 7b. [44] The aerosol turbidity classification in function of spectral aerosol optical measurements is illustrated in Figures 7a and 7b where CE318-2 measurements made in 2003 are plotted as AOT441 in function of AOT873. The thresholds given in Table 7a are represented by solid lines that delimit 5 zones. The use of two wavelengths for making the distinction between forest fire emissions and desert dust particles, 441 and 873 nm respectively, is justified by the observation of the maximum of AOT441 belonging to the polluted class, while the maximum of AOT873 belongs to the desert dust class. Clean with a C < 1.0 is not observed Figure 4. Frequency distribution of the sector origin of the 72-hour back trajectories, ending at the position of Évora, and calculated at the pressure levels of 970 hpa and 700 hpa, for the summer (June to September) seasons of 2002 and of20

10 Figure 5. Back trajectories of air masses loaded with desert dust arriving in Évora at 1200 UT and at 700 hpa on (a) 30 July and 1, 2, and 5 August over 72 hours and on (b) 3, 4, and 15 August 2003 over 120 hours. (maritime cases are encountered for 0.04 < AOT873 < 0.09). Values of a C included between 1.0 and 1.2 are never observed during continental situations. It is interesting to note the large gap in AOT873 of a width of around 0.10 between desert dust cases and maritime cases (for a C < 1.0). In contrary, the cluster of points is continuous from clean situation to high turbidity level. [45] The classification is applied on measurements made in Évora and relation with air mass origins and aerosol measurements at surface level is examined. First a case study is presented in section 6.2 and second, the classification is generalized to all situations encountered over both summers, and each class of situation is commented in dedicated paragraphs in section Case Study: The Heat Wave of August 2003 [46] The case study is the episode of increased turbidity associated to the high temperature increase experienced in all Europe in July August 2003, which is associated by far to the worst forest fire season that Portugal faced in the last 23 years [Barbosa et al., 2004]. This event witnesses all kinds of situations, which occur in Évora during both 2002 and 2003 summers. Cloud presence is minor as the number of measurements per morning or afternoon exceeds 22 over 18 consecutive days, except on 5 occasions. Values of aerosol optical thickness, column Ångström Exponent, scattering coefficient and surface Ångström exponent, as well as the air mass origins, observed during the case study, are reported in Table 8 and classified within the four aerosol classes of clean, continental, desert dust and forest fire emissions. Values of morning and afternoon averages of aerosol optical thickness at 441 nm and at 873 nm are showed in Figure 8a in function of the number of the day in 2003 (nd). The selected period starts on 25 July 2003 morning (nd = 205) and stops on 19 August 2003 afternoon (nd = 230). Values of morning and afternoon averages of the aerosol scattering coefficients at 450 and 550 nm are plotted in Figure 8b for the same time period. The Ångström exponents a C and a S are showed in Figure 8c for the same period. [47] The aerosol optical thickness is as low as 0.10 at 441 nm and 0.03 at 873 nm on 25 July and 19 August, and varies over more than an order of magnitude in between. AOT441 exceeds 0.30 from 30 July morning (nd = 210) to 14 August afternoon (nd = 225), meanwhile AOT871 is larger than The turbidity is caused by two kinds of aerosols as the maximum of aerosol optical thickness at 441 nm during the turbid episode occurs on 12 August (nd = 223) with AOT441 = 0.81, while the maximum of AOT873 occurs on 2 August (nd = 213) with AOT873 = The Ångström exponent is different on these two dates, fine particles being more numerous on 12 than on 2 August. Figure 8a also shows that the difference between AOT441 and AOT873 is larger during the second half of the turbid time, demonstrating that fine particles are relatively more numerous after 6 August 2003 (nd = 217) than before this day. [48] Five subperiods are observed, according to the values of AOT and the spectral dependence. First the low turbidity level extends from 25 July morning (nd = 205) to 28 July morning (nd = 209), with AOT873 < 0.04 (first and second rows in Table 8). The second step consists in a strong influence of desert dust particles with AOT873 > 0.17, and a C < 1.0 (except on 30 July afternoon when a C = 1.2), which extends from 30 July morning (nd = 210) to 5 August afternoon (nd = 216). The third step occurs from 6 August morning (nd = 217) to 14 August morning (nd = 225) (third row in Table 8). Main influences are emissions from forest fires, causing AOT441 to be larger than 0.47 with a C larger than 1.4 (except on 6 August morning when a C = 1.1) (fourth row in Table 8). The fourth step extends 10 of 20