Tropospheric aerosols in the Mediterranean: 2. Radiative effects through model simulations and measurements

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 108, NO. D10, 4317, doi: /2002jd002807, 2003 Tropospheric aerosols in the Mediterranean: 2. Radiative effects through model simulations and measurements D. Meloni, 1,2 A. di Sarra, 1 J. DeLuisi, 3 T. Di Iorio, 2 G. Fiocco, 2 W. Junkermann, 4 and G. Pace 2 Received 29 July 2002; revised 30 October 2002; accepted 2 December 2002; published 30 May [1] The radiative effects of tropospheric aerosols at the island of Lampedusa, in the Mediterranean, have been investigated by comparing measurements and results from a radiative transfer model. The model was used to reproduce the measured ultraviolet irradiance spectra ( nm) in three cases of cloud-free conditions in May 1999, allowing the estimation of the aerosol properties and of the direct radiative forcing. Observations show very different aerosol loading and distribution, connected to the different origins of the air masses: low aerosol optical depths are associated with air masses from North Atlantic/Europe (25 and 27 May), and larger particles up to 7 km altitude and larger optical depths are observed for mineral dust coming from the Saharan region (18 May). The detailed description of the atmospheric structure and composition was used to initialize the radiative transfer model. The estimated single-scattering albedo and asymmetry parameter at 500 nm for the desert dust are in the range and in the range , respectively. Radiative transfer calculations show that differences of the surface ultraviolet irradiance larger than 10% may arise from the lack of a detailed knowledge of the aerosol size distribution. Aerosol may also increase or reduce the absorption effectiveness of tropospheric ozone, depending on the characteristics of the particles. Estimates of the direct aerosol radiative forcing in the spectral range nm at the surface and at the top of the atmosphere (TOA) were also derived. At the surface, aerosols produce a decrease of the instantaneous downward irradiance with respect to an aerosol-free atmosphere by 70.8, 37.2, and 39.1 W m 2, for 18 May (the aerosol optical depth is at 415 nm), 25 (0.165) and 27 (0.224), respectively. The radiative forcing per unit optical depth at the surface is largest for aerosol of continental/ marine origin, transported from North. The forcing at the TOA is negative, thus producing a cooling, in the desert dust case, and close to zero or positive for aerosol originating from North. INDEX TERMS: 0305 Atmospheric Composition and Structure: Aerosols and particles (0345, 4801); 0360 Atmospheric Composition and Structure: Transmission and scattering of radiation; 3359 Meteorology and Atmospheric Dynamics: Radiative processes; 3360 Meteorology and Atmospheric Dynamics: Remote sensing; KEYWORDS: aerosol, radiative forcing, atmospheric ozone, dust, atmospheric transmission Citation: Meloni, D., A. di Sarra, J. DeLuisi, T. Di Iorio, G. Fiocco, W. Junkermann, and G. Pace, Tropospheric aerosols in the Mediterranean: 2. Radiative effects through model simulations and measurements, J. Geophys. Res., 108(D10), 4317, doi: /2002jd002807, Introduction [2] The role of ozone in modulating solar UV and, to a less extent, visible radiation is reasonably well known [e.g., Cutchis, 1974; Brühl and Crutzen, 1989; Madronich, 1992; McKenzie et al., 1995; Zerefos et al., 1995, and others]. 1 Ente per le Nuove Tecnologie, l Energia, e l Ambiente (ENEA), Climate Laboratory, S. Maria di Galeria, Italy. 2 Dipartimento di Fisica, Università La Sapienza, Roma, Italy. 3 National Oceanic and Atmospheric Administration, Boulder, Colorado, USA. 4 Forschungszentrum Karlsruhe, Institute for Meteorology and Climate Research, IMK-IFU, Garmisch-Partenkirchen, Germany. Copyright 2003 by the American Geophysical Union /03/2002JD [3] Nevertheless, optical properties of atmospheric constituents like aerosols are essential for understanding the impact on solar radiation reaching the ground. Aerosol particles have two main effects on the atmospheric radiative budget: a direct one, through absorption and scattering, and an indirect one, since they can act as condensation nuclei for cloud droplets, thereby influencing the cloud properties and their radiative characteristics and lifetime. The concentration, shape, size distribution, refractive index, and vertical distribution of the atmospheric aerosols are highly variable, both in time and space; some of these quantities (especially the microphysical properties) are generally not well known, and difficult to be measured. [4] Radiative transfer models represent an important tool for the investigation of the effects of atmospheric constitu- AAC 4-1

2 AAC 4-2 MELONI ET AL.: TROPOSPHERIC AEROSOLS IN THE MEDITERRANEAN, 2 ents on solar radiation. These models, however, require among other parameters, the knowledge of the aerosol optical characteristics. [5] Aerosol effects on the UV irradiance have been studied by several authors through field observations and radiative transfer model studies. [6] Liu et al. [1991] suggested that the observed reduction in ground visibility over nonurban areas in the eastern United States and Europe since the industrial revolution might have balanced most of the UVB increase due to stratospheric ozone depletion. [7] Studies have been also conducted at continental site in North America [DeLuisi, 1997; Kerr, 1997], and at a rural site, Ispra (Italy) [Meleti and Cappellani, 2000]. Krzyścin and Puchalski [1998] have analyzed the time series of daily erythemal UV dose at ground level at Belsk (Poland) in the period , and its relation with total ozone, global solar radiation, and aerosol optical depth (t) at 550 nm, concluding that the changes in the UV radiation due to aerosol optical depth variations are, in the mean, comparable (but opposite in sign) to those due to ozone changes. [8] Kylling et al. [1998] used a radiative transfer model to compare simulated spectra with and without aerosols for the conditions occurred during the Photochemical Activity and Solar Ultraviolet Radiation (PAUR) campaign that took place in the Aegean in They found that, all the other atmospheric parameters being unchanged, UVB irradiance is reduced, with respect to the aerosol-free case, by 5 35% depending on the aerosol optical depth and single-scattering albedo. [9] Erlick and Frederick [1998] have investigated the effects of continental and urban aerosols on atmospheric transmission in the UV and visible, also considering variations of the aerosol properties with humidity. Their results show that midlatitude continental and urban aerosols may reduce transmission by % at 310 nm, and by % at 550 nm. [10] The influence of regional aerosol on UVB radiation transmission for a continental site in North America was quantified by Wenny et al. [1998], who developed a new method for retrieving the aerosol single-scattering albedo and asymmetry factor from radiation and size distribution measurements using Mie code and a radiative transfer model. The aerosol single-scattering albedo was found to vary from 0.75 to 0.93 and the asymmetry factor ranged between 0.63 and 0.76 at 312 nm. [11] Among all the different types of atmospheric aerosols, desert dust is of primary importance, first of all, because the extension of its potential source areas covers about one-third of the terrestrial surface; a further aspect which renders the desert aerosol one of the most prominent aerosol types is the long-range atmospheric transport due to strong winds and convective processes [d Almeida et al., 1991]. Although desert dust has a big impact on UV radiation through scattering and absorption, only a few studies of this kind have been carried out up to now. Dust is a mixture of various minerals, whose optical constants depend on dust origin, mobilization properties, physical and chemical transformation during transport in the atmosphere, as well as on the aggregation status of the minerals. [12] Sokolik and Toon [1996] estimated that, although the key quantities which influence the direct radiative forcing are affected by a wide range of uncertainties, the forcing due to mineral aerosol near dust source areas is comparable to the forcings observed for clouds. [13] Radiative forcing of mineral dust is sensitive to size distribution and complex refractive index [Claquin et al., 1998], the imaginary part being one of the critical parameters in the calculation of the forcing. The techniques currently used to determine the refractive index of dust aerosol have relatively high uncertainties (about 30 50%) that affect the modeled aerosol optical properties [Sokolik et al., 1993]. Sokolik and Toon [1999] described a technique to model the radiative properties of mineral aerosols from UV to IR taking into account their composition. They simulated the daily mean net forcing by dust of various compositions, showing that a mixture of aggregates can cause a positive radiative forcing, while a mixture of individual minerals may produce an opposite effect. [14] During the Photochemical Activity and Ultraviolet Radiation modulating factors II (PAUR II) campaign, an articulated set of measurements of ultraviolet radiation, aerosol properties, and atmospheric chemical composition was collected in the Mediterranean [Zerefos et al., 2002]. Measurements of UV and visible irradiance, ozone, and aerosol were performed also at the island of Lampedusa (35.5 N, 12.6 E) in May June 1999 [di Sarra et al., 2001a]. Instruments were deployed at the Laboratory for Climate Observation of the Ente per le Nuove Tecnologie, l Energia, e l Ambiente (ENEA), 45 m a.s.l. Lampedusa is a small island south of Sicily in the Mediterranean Sea, approximately 100 km off the west coast of Tunisia. The island is relatively far from significant sources of pollution. In late spring-early summer transport of Saharan dust is frequent in the central Mediterranean [see e.g., Moulin et al., 1997]. Several dust events occurred during the PAUR II campaign, and large aerosol amounts were detected over Lampedusa in different occasions [di Sarra et al., 2001b]. [15] In previous papers the evident effect of the Saharan dust aerosols on the UV, erythemal, and visible irradiance has been shown [di Sarra et al., 2001a, 2002]. Optical depths at 415 nm as large as 0.9 were observed during the dust transport events. The Saharan dust produced reductions of the surface irradiance at 350 nm up to 30% at the solar zenith angle of 20, and up to 60% at 60 [di Sarra et al., 2002]. [16] In this work the ground-based UV and visible irradiances in cloud-free conditions are simulated with a radiative transfer model. The model is initialized with as much and as accurate as possible information, based on the measurements carried out during the campaign, on the atmospheric structure. Particular efforts have been devoted to reconstruct the aerosol distribution and properties. By initializing the model with a large set of observations, it is possible to tune a few unknown parameters to obtain the best agreement between model results and measurements, and isolate effects attributable to different factors. A companion paper by Di Iorio et al. [2003], hereinafter referred to as DI] reports analyses based on lidar, particle counters, and Sun photometer observations at Lampedusa during PAUR II, aimed at retrieving aerosol microphysical and optical properties. Some of these properties will be used in this study to initialize the radiative transfer model.

3 MELONI ET AL.: TROPOSPHERIC AEROSOLS IN THE MEDITERRANEAN, 2 AAC 4-3 [17] The objectives of this paper, through the comparison of modeled and observed irradiances are: (1) to study the dependence of UV and visible radiation at the surface on different aerosol loadings and optical properties, (2) to identify the role played by tropospheric ozone, also in the presence of large aerosol amounts, (3) to derive estimates of some important aerosol properties, such as the singlescattering albedo, that are of fundamental importance for the determination of the aerosol radiative effects, and (4) to estimate the local direct aerosol radiative forcing in the UV and visible. 2. Measurements [18] During the PAUR II campaign, many instruments were deployed at Lampedusa [di Sarra et al., 2001a]: among them, a lidar designed to measure tropospheric aerosol backscattering and depolarization, a double monochromator Brewer spectrometer for the measurements of total ozone and spectral ultraviolet irradiance, a MultiFilter Rotating Shadowband Radiometer (MFRSR) to derive aerosol optical depths at several wavelengths in the visible and near infrared, a Licor spectroradiometer to measure spectral global irradiance in the wavelength range nm. These four instruments were operative throughout the campaign. [19] The lidar developed at the University of Rome was designed to observe the tropospheric structure between 0.4 and 10 km during daytime. Aerosol backscattering coefficient and backscatter ratio profiles were derived with vertical resolution of 30 m averaging signals over 30 min. A detailed description of the instrument and of the measurements of the PAUR II campaign is given by di Sarra et al. [2001b]. [20] UV irradiance and total ozone measurements were performed by the Brewer MK III #123 in the spectral range nm at 0.5 nm steps. The estimated uncertainty on the measured irradiance is about 4 5%, and the uncertainty on total ozone is around 1% for cloud-free conditions [di Sarra et al., 2002]. [21] The MFRSR performs measurements of global and diffuse irradiances at seven spectral bands: the instrument is equipped with six separate detectors, each with a 10-nm wide (FWHM) filter, centered at 415, 500, 615, 671, 868, and 937 nm, respectively. A Si photodiode measures broadband radiation [Harrison et al., 1994]. [22] The Licor LI-1800 spectroradiometer measures global irradiance in the range nm with a 6-nm bandwidth at a wavelength interval selectable at 1, 2, 5, or 10 nm. A 2-nm step was chosen during the campaign, and the total scan time was about 27 s. The estimated accuracy is ±10 15% depending on wavelength. The Licor spectrometer signals show a diurnal oscillation, associated with temperature changes, with respect to both the Brewer and the MFRSR measurements. The amplitude of the diurnal oscillation (i.e., the difference of the measured irradiances at the same solar zenith angle, in the morning and afternoon, with respect to simultaneous measurements by other instruments) was derived by comparing the Licor measurements with the Brewer data convoluted with a 6-nm slit function, and by comparing the Licor measurements convoluted with a 10-nm bandwidth slit function, with the MFRSR observations. The amplitude of the daily oscillation of the Licor data is about 10% at a solar zenith angle of 40, and about 13% at 60. The daily oscillation was removed from the data, using as a reference the Brewer data. [23] During PAUR II the ultralight aircraft of the Fraunhofer Institute of Garmisch-Partenkirchen (Germany) performed 16 flights from the Lampedusa airport between 10 and 28 May, reaching a maximum altitude of 4500 m. A description of the instrumented aircraft is given by Junkermann [2001]. [24] Various instruments were installed aboard the ultralight aircraft: among them, an ozone UV-photometer, a Forward Scattering Spectrometer Probe (FSSP) for particles of diameter between 0.4 and 8 mm, a laser particle counter for particles of diameter larger than 0.3 mm, and a condensation nuclei counter, for particles of diameter between 0.01 and 3 mm. [25] Measured UV and visible spectra are reduced to the mean Sun-Earth distance and, together with O 3 total columns, interpolated at fixed solar zenith angles, i.e., on each day, at the same time. The interpolation of the spectra is made with a second-order polynomial, using the three closest in time-acquired spectra. Linear interpolation is used to derive the total ozone amounts at fixed solar zenith angles. [26] Previous studies have shown that very large changes of the aerosol distribution occurred during the campaign, the aerosol vertical distribution and amount depending on the origin of the air masses. When the air mass trajectory comes from Europe and/or North Atlantic, the aerosol optical depth is relatively low, and the particles are generally confined to the region below 3 km. When the air mass originates from Africa, much larger values of the optical depth are found, and the aerosol cloud may reach altitudes of 7 8 km. Also, the aerosol properties appear significantly different, since the two cases are characterized by low (for the Saharan case) and large (for the European case) values of the Ångström exponent [di Sarra et al., 2001b, 2002]. The spectral UV observations at the ground have shown that these different aerosol conditions largely affect the atmospheric transmittance [di Sarra et al., 2001a]. [27] To emphasize the role of the aerosol, cloud-free periods with different aerosol loading, when ground-based and airborne observations were available, were selected for the model computations. Three cases were identified: 18, 25, and 27 May, for different values of the solar zenith angle (60 on 18 May, 40 on 25 and 27 May). In Figure 2 of DI, 5-day isentropic trajectories ending at Lampedusa at approximately 2.5 km on 18, 25, and 27 May are depicted. To compare spectra measured in different aerosol conditions at 60 solar zenith angle, also 1 June, when only groundbased measurements were carried out, has been considered. However, on 1 June no detailed information on the aerosol microphysical properties was available, and the data for this day have not been used in the model analysis. As reported by DI on 25 and 27 May, the air masses pass over Europe before reaching Lampedusa: their composition is expected to be prevalently of anthropogenic particles and crustal aerosol from continental air masses, and marine aerosols. On 18 May and 1 June, the air masses spent the last several days over central Africa, and the particles are expected to be mainly composed of desert dust.

4 AAC 4-4 MELONI ET AL.: TROPOSPHERIC AEROSOLS IN THE MEDITERRANEAN, 2 Figure 1. Backscatter ratio profiles obtained by lidar on 18, 25, and 27 May and 1 June (see text). [28] The irradiances were calculated at fixed solar zenith angles, within or close to those of the aircraft flights. On 25 May, the aircraft performed one flight from 1345 to 1432 UT and the model was run for a solar zenith angle of 40 in the afternoon (1357 UT). On 27 May, the aircraft flew between 1324 and 1446 UT and the calculation was made at 40 in the afternoon (1358 UT). On 18 May, clouds were present before 1500 UT; the aircraft flight was performed between 1325 and 1515 UT, and the calculations were made for 60 in the afternoon (1531 UT). [29] In Figure 1 the lidar-derived aerosol backscatter ratio (ratio between the signal due to aerosol and molecules, and the signal due to molecules only) profiles for the component of the signal with the same polarization of the linearly polarized laser beam, for 18, 25, and 27 May and 1 June, are shown. Significant differences in the aerosol amount and vertical extension appear, as expected, due to the different origin of the air masses. On 18 May and 1 June, for example, the strong convection over the Sahara desert lifted the aerosol particles to 7 8 km. Over North Atlantic and Europe such a strong convection is unlikely, and particles are mostly confined below 3 km. [30] In Table 1 the average values of the aerosol optical depths at 415 and 868 nm, measured by the MFRSR, and of total ozone, measured by the Brewer, are reported for the 4 days. Data for 25 May at 60 solar zenith angle and for 1 June at 40 are also reported. The averages are calculated over a 1-hour interval centered at the indicated solar zenith angle. Average values of the diffuse-to-direct radiation ratio (DDR) at 415 nm, measured by the MFRSR, are also reported. The standard deviation of the average for the aerosol optical depth and the DDR is indicated. It may be noted that on 18 May and 1 June, when low total ozone values are measured, the aerosol optical depth is large, while the opposite occurs on 25 and 27 May. A negatively correlated behavior between total ozone and aerosol optical depth has been found at Lampedusa during PAUR II and is attributed to the influence of tropospheric anticyclones that induce the transport of desert aerosols over Lampedusa, on the stratospheric vertical structure and advection [di Sarra et al., 2002]. The amount of diffuse radiation is significantly larger on 18 May and 1 June than on 27 and 25 May, respectively, indicating that the aerosol largely affects the transfer of visible radiation. [31] In Figure 2 the ultraviolet spectra measured at 40 solar zenith angle in the afternoon of 27 May and 1 June, and at 60 solar zenith angle on 18 and 25 May, are depicted. Significant differences in the measured irradiances appear. At short wavelengths, the effect of ozone absorption is dominating, and higher irradiances are observed on the days when the total ozone is low. At longer wavelengths the effect of the aerosol overcomes the effect of the ozone, and significantly higher irradiances are found on the days characterized by low aerosol optical depth. At 40 solar zenith angle the irradiance at 340 nm, where the absorption by ozone may be neglected, is by 14% larger on 27 May than on 1 June. Similarly, at 60 solar zenith angle the irradiance at 340 nm is 15% larger on 25 May than on 18 May. Figure 3 shows the spectra measured on the same days and solar zenith angles by the Licor spectrometer. The large aerosol effect is evident throughout the visible and near-ir spectral range. Differences also appear in correspondence with the water vapor absorption bands around 690, 726, and 936 nm; at 60 solar zenith angle the difference in the water vapor content of 18 and 25 May seems to compensate for the irradiance difference produced by the aerosol in the nm region. The peak at 764 nm is due to an O 2 absorption band. The influence of the aerosol on the ultraviolet radiation transfer is also evident from the values of the DDR at 415 nm observed on the different days, respectively, at 40 and 60 solar zenith angle. [32] In what follows we will describe the radiation transfer model used to reproduce the observed large changes of irradiance, the procedure to set the atmospheric parameters needed by the model, and the results of the comparison with the measurements. 3. Model [33] In this study the UVSPEC radiation transfer model [Mayer et al., 1997] is used to reproduce the measured spectra shown in Figures 2 and 3. The UVSPEC model allows the choice of one among three methods for the solution of the radiative transfer equation: the two stream [Meador and Weaver, 1979], the discrete ordinate method (DISORT) for plane-parallel atmosphere [Stamnes et al., 1988] and the pseudospherical approximation of DISORT (SDISORT) which accounts for the sphericity of the atmosphere [Dahlback and Stamnes, 1991]. We operated the model in six-stream mode for the calculation of the irradiance, and used the SDISORT solver. The model can simulate radiation quantities such as radiance, direct and Table 1. Solar Zenith Angle (SZA), Aerosol Optical Depth (t) at 415 and 868 nm, Ozone Amount (DU), and Diffuse-to-Direct Ratio (DDR) at 415 nm for the Cases Selected in This Study Day SZA t (415 nm) t (868 nm) Ozone (DU) DDR (415 nm) 18 May ± ± ± May ± ± ± May ± ± ± May ± ± ± June ± ± ± 0.05

5 MELONI ET AL.: TROPOSPHERIC AEROSOLS IN THE MEDITERRANEAN, 2 AAC 4-5 Figure 2. UV spectra at (a) 40 solar zenith angle for 27 May and 1 June, and (b) 60 solar zenith angle for 18 and 25 May. The ratio of the solid and dashed curves is shown in the upper graphs. diffuse irradiances, and actinic fluxes between 200 and 800 nm assuming an extraterrestrial solar spectrum and taking into account the atmospheric structure and composition, and the surface characteristics. [34] The atmosphere is divided into layers with constant gas concentration and aerosol optical properties. These properties may vary from layer to layer. Ozone is assumed to be the only absorbing gas. Thus large differences between the model and the observations are expected to occur in the spectral regions where absorption bands by other species (mainly water vapor and molecular oxygen) are present. [35] The model inputs are solar zenith angle, total ozone, ground albedo, vertical profiles of pressure, temperature, air density, and ozone; aerosols are characterized by singlescattering albedo, asymmetry factor, and optical depth at each layer. This study is limited to clear sky conditions, and clouds are not taken into account. [36] Estimates of the accuracy of UV radiative transfer models indicate that the input parameters play a large role. According to Weihs and Webb [1997], the main source of uncertainty lays in the limited accuracy of the input data, which are likely to produce spectral UV values with maximum errors between ±15 and ±26% at 305 nm, and between ±4 and ±15% at 380 nm. Errors are smallest for unpolluted sites and for longer wavelengths. Schwander et al. [1997] found that for a polluted region as central Europe the uncertainties in the bulk input parameters (O 3 and SO 2 amounts, ground reflectivity, ground pressure, temperature, humidity, and visibility) lead to uncertainties in spectral UV irradiances between 10 and 50%; limited accuracy of nonbulk parameters (vertical profiles of ozone, aerosol extinction, pressure, and temperature) adds another 5% in modeled UV irradiances. Smaller uncertainties are expected when detailed characteristics of the input parameters, including the vertical distribution, are available. [37] The high-resolution extraterrestrial spectrum is derived from the Atlas 3 measurements, shifted to air wavelengths, between 280 and nm, from Atlas 2 in the nm region, and from the Modtran 3.5 Figure 3. Spectra measured with the Licor spectrometer at (a) 40 solar zenith angle for 27 May and 1 June, and (b) 60 solar zenith angle for 18 and 25 May. The ratio of the solid and dashed curves is shown in the upper graphs.

6 AAC 4-6 MELONI ET AL.: TROPOSPHERIC AEROSOLS IN THE MEDITERRANEAN, 2 model for wavelengths longer than nm [Kylling et al., 1998]. The spectral irradiance is calculated at 0.05 nm intervals. The high-resolution spectral irradiances are then convoluted with the Brewer and the Licor slit functions, assumed to be wavelength independent and approximated by triangles with 0.55 and 6 nm FWHM bandwidth, respectively. The instrumental slit functions were not accurately determined, and may somewhat differ from the modeled curves. [38] Two main modifications to the original model by Mayer et al. [1997] have been implemented: an increase of the atmospheric vertical resolution, to allow a more detailed description of the tropospheric structure, and a new parameterization of the aerosol to include in the model the observed aerosol properties. As discussed above, lidar observations show that during the PAUR II campaign large differences in the aerosol vertical distribution occurred, with maximum altitudes variable from 1 to 8 km [di Sarra et al., 2001b]. Thus we divided the atmosphere in 64 homogeneous layers from 0 to 120 km with a vertical resolution of 150 m in the lowest 2.1 km, 300 m between 2.1 and 8.4 km, 1 km up to 25 km, and growing thickness up to 120 km. [39] The original version of UVSPEC includes four prescribed aerosol types (urban, rural, maritime, and tropospheric) for the lowest 2 km of the atmosphere. In this study, the aerosol size distribution has been derived from the particle counters observations and from the optical depth measurements made during the campaign [DI]. The aerosol optical depth supplied to the model is derived from observations as well. Being the aerosol size distribution, as will be discussed below, derived from the measurements, no dependence on ground visibility and humidity was taken into account; no prescribed aerosol type and size distribution are assumed. The wavelength-dependent aerosol extinction and absorption coefficient, and asymmetry factor have been calculated applying the Mie theory to the retrieved size distribution. [40] Vertical profiles of pressure and temperature at 1200 UT are obtained from the NCEP analyses. The atmospheric vertical structure is assumed not to vary appreciably during the day. The temperature-dependent ozone absorption cross sections given by Burrows et al. [1999] are used. [41] The UV irradiance at a specific point is influenced by the albedo of the surface within a region of radius of about 20 km [Ruggaber et al., 1998], and the determination of the albedo to be used in the model should take into account all the different types of surface in the surroundings [Weihs and Webb, 1997]. Thus the albedo is calculated as the average of land and water albedo weighted for the respective surface area, corresponding to 0.08 in the UV. The albedo in the visible and near infrared, estimated in the same way, is Lambertian reflection by the surface is assumed. [42] Previous studies [see e.g., Pyne, 1972] indicate that the shortwave reflectivity of the sea surface depends on the Sun altitude, on the surface roughness, which is related to the wind speed, and on the atmospheric transmittance. While the effect of the surface roughness is negligible for solar zenith angles smaller than 60, and for low wind speed, as is the case of the selected days, the dependence on the atmospheric transmittance and Sun altitude can be significant. It must be considered that small changes of the albedo, when its value is small, produce minor effects on the surface irradiance. We have performed test calculations by using the solar zenith angle dependence of the ultraviolet albedo reported by Doda and Green [1980]; the effect on the surface UV irradiance is small, and a constant value was assumed for all days and solar zenith angles both in the UV and visible spectral ranges. [43] The Brewer instrument is located on the roof of the ENEA laboratory (45 m a.s.l.). The island of Lampedusa is flat (the maximum elevation is 120 m), and the shading effects due to obstacles above the horizon are negligible. Consequently, no obstacles are considered in the model calculations Atmospheric Parameters: Ozone [44] Total ozone was frequently measured by the Brewer spectrometer during the campaign. The observed values are linearly interpolated at the solar zenith angles where the model calculations are performed. No stratospheric ozone profiles were available at Lampedusa, while the lower tropospheric ozone concentration was measured by the UV photometer aboard the ultralight aircraft. Bhartia et al. [1985] have derived climatological midlatitude ozone profiles from solar backscatter ultraviolet (SBUV) and balloon observations, for total ozone values from 225 to 525 Dobson units (DU). The ozone profile at Lampedusa is assumed to correspond to the climatology matching the measured total ozone. However, since the climatological profiles by Bhartia et al. are not differentiated at altitudes below 4 km, measurements of the ozone UV-photometer are used in the lower atmospheric layers; the measured and the climatological profiles are assumed to join at 10 km altitude. Keeping fixed the ozone concentration at the measured values at low altitudes, the climatological part was scaled to fit the measured total ozone. The used ozone profiles, and the climatological profiles by Bhartia et al. [1985] for 18, 25, and 27 May are shown in Figure 4. The tropospheric ozone concentrations measured by the aircraft are larger than the climatological profile by 31, 80, and 78%, respectively, on 18, 25, and 27 May. It is noticeable that largest deviations from the climatology occur on 25 and 27 May, when the trajectories originate from Europe. Kourtidis et al. [2002] have shown that higher tropospheric ozone concentrations are associated in the Mediterranean to air masses originating in Europe; the increased ozone is accompanied by an enhanced concentration of NO x, indicating that photochemical production may play a significant role. As discussed by Brühl and Crutzen [1989], tropospheric ozone may play a significant role in the absorption of UV radiation due to the strong scattering and consequent enhanced photon path lengths in the troposphere. This role may be amplified when large tropospheric aerosol amounts, leading to a further increase of the scattering, are present Atmospheric Parameters: Aerosol [45] The radiation transfer model requires profiles of the aerosol extinction coefficient, single-scattering albedo (the fraction of the extinction due to scattering), and asymmetry factor (the average of the cosine of the scattering angle for scattered radiation). A detailed description of the methods to derive vertical profiles of the size distribution, on the basis of column optical depth and particle counters measurements is given by DI. The aerosol extinction coefficient profile at 532

7 MELONI ET AL.: TROPOSPHERIC AEROSOLS IN THE MEDITERRANEAN, 2 AAC 4-7 available, is based on the similarity between the lidar and the particle counters measurements [DI]. [47] The aerosol layer was divided into a number of sublayers, where the size distribution is allowed to change, but the refractive index is kept constant; the refractive index is also assumed to be wavelength independent. Singlescattering albedo and asymmetry factor are calculated by using the size distribution by DI, for different values of the refractive index. A sensitivity study was performed to infer a variability range for the real and imaginary parts. The method and the results of the comparison between measured and modeled UV irradiances are described in detail in the next section. To study the influence of the aerosol size distribution on the surface irradiance, Mie calculations were also performed using the size distributions for maritime polluted aerosol, desert background (or wintertime), and wind carrying dust (or summertime) particles reported by d Almeida et al. [1991]: the first aerosol type was used when air masses originated from Europe, while the two desert types were chosen for the case when the air mass originated from Africa. The parameters of these three modal lognormal size distributions are reported in Table 2. Different values of the aerosol refractive index were assumed, depending on the prevailing component expected to be present in the particles, i.e., water-soluble for the continental aerosol, and mineral for the dust case. Refractive indices as a function of wavelength are also derived from the work of d Almeida et al. [1991]. [48] The main error source in the calculation of the aerosol optical properties from Mie theory is the nonsphericity of the particles. When t is known, however, the particle shape has a negligible influence on the estimates of the aerosol radiative forcing [Díaz et al., 2000]. Figure 4. Climatological ozone profile (dashed line) and climatological profile corrected in the troposphere to match the airborne measurements (solid line) for (a) 18 May, (b) 25 May, and (c) 27 May. The profiles are normalized to the total ozone amount measured by the Brewer spectrophotometer. nm is derived by multiplying the lidar-derived backscattering coefficient by the extinction-to-backscattering ratio e. The value of e is estimated by the combination of lidar and photometer observations, and is used in the lidar retrieval algorithm [see DI]. The optical depth of the aerosol layer sounded by the lidar is estimated by subtracting the upper troposphere and stratospheric aerosol optical depth. These contributions are derived from the work of Russell et al. [1993], who reported values of 0.01 and 0.003, respectively. [46] The estimate of the extinction profile in the lowest atmospheric layers, where lidar measurements are not 4. Results and Discussion [49] As discussed above, the radiation transfer model was initialized with all the available measured parameters, allowing for a detailed comparison of the observed irradiances with the measured spectra. In what follows two types of analyses will be performed. The first one is aimed at deriving an estimate of the imaginary part of the aerosol refractive index (integrated over the column), by searching the values that produce the best agreement between model and measurements. The second analysis is a sensitivity Table 2. Parameters of Three Modal Size Distributions by d Almeida et al. [1991] Used in This Study in the Indicated Cases Component Mixing Ratio by Volume Radius, mm s Maritime Polluted (25 and 27May) Water-soluble Soot Sea-salt Desert Background (Wintertime) (18 May) Mineral Mineral Mineral Wind Carrying Dust (Summertime) (18 May) Mineral Mineral Mineral

8 AAC 4-8 MELONI ET AL.: TROPOSPHERIC AEROSOLS IN THE MEDITERRANEAN, 2 study aimed at understanding the influence of some key parameters, namely aerosol size distribution and refractive index, and ozone vertical profile, on the model results. The effect of tropospheric aerosols on UV absorption by ozone will be also investigated. The role of these parameters in the observed different aerosol conditions (low and high load, continental/marine, and desert aerosol) will be emphasized Brewer Spectra Low Aerosol Amount: 25 and 27 May 1999 [50] The conditions of 25 and 27 May are very similar, thus providing the opportunity to compare model and measurements under relatively well-defined conditions, without the complexity introduced by large aerosol loadings. [51] Figure 5a shows the ratio between the model results and the spectrum measured by the Brewer on 25 May at 40 solar zenith angle. The figure displays the model results for three different values of the imaginary part n 00 of the aerosol refractive index; the real part n 0 was fixed at 1.366, that is the column average of the estimates of n 0 obtained by DI. Curves showing 10 nm running averages of the ratio are also depicted. The fluctuations in the model-measurement ratio are mainly attributed to possible differences between the assumed and the true instrumental slit function, and to possible small-wavelength shifts of the spectrometer. The comparison is limited to wavelengths larger than 300 nm, due to the increase of the experimental uncertainties below this wavelength. We assume that the 5% deviation from unity in the ratio limits the values of the refractive index that may be considered plausible, ±5% being the estimated error of the Brewer irradiances. The upper curve is thus obtained using the smallest value of n 00 for which the running mean of the model-measurement ratio remains below In the same way, the lower curve is calculated with the largest value of n 00 for which the running mean remains above The central curve is obtained with a refractive index of i, value that gives the best agreement. This procedure produces a wide range of uncertainty for the imaginary part n 00 ( 0.012/ 0.16), indicating that the simulated irradiance is not strongly dependent on the aerosol absorption properties. [52] Following the same criteria adopted for 25 May, for n 0 = 1.402, we obtain for 27 May that the plausible range for n 00 is ( / 0.044), and the refractive index which yields the best agreement is i (see Figure 5b). These values correspond to aerosol of very different compositions: values of n 00 on the order of 10 2 /10 3 are typical for dust-like, water-soluble, and mineral aerosol components, while for soot particles n 00 is between 0.1 and 1, depending on wavelength. [53] The retrieved ranges of values for n 00 on 25 and 27 May correspond to moderately to highly absorbing particles. [54] In the radiative transfer model, we have fixed the aerosol optical depth at 415 nm to the measured value. The profile of the aerosol extinction coefficient (whose integral matches at 415 nm the measured optical depth) is also derived from the observations, as previously discussed. Changes of the refractive index imply variations of the extinction coefficient, and the number of particles is varied by means of an altitude independent scaling factor to obtain Figure 5. Ratio of the modeled and measured UV spectra for (a) 25 May and (b) 27 May at 40 solar zenith angle. The curves refer to three different values of the imaginary part of the aerosol refractive index. The thick solid curves are 10 nm running averages of the ratio. the measured optical depth at 415 nm. Because of these constraints, the model has intrinsically a poor sensitivity on the real part of the aerosol refractive index. Some dependence on the real part of the refractive index is induced by the wavelength changes of the optical properties that are not fixed in the UV. For instance, the aerosol optical depth in the UV may somewhat change, depending on both the real and imaginary parts of the refractive index (the size distribution is fixed); these changes are, however, small since the aerosol optical depth at 415 nm is not allowed to vary. To investigate the role played by the real part of the refractive index, we have fixed n 00 at the value giving the best agreement between model and measurements, and have varied n 0 for both days, in the range These values correspond to large changes in the aerosol composition, 1.3 applying to water, and 1.7 to soot particles. The resulting change of the model-measurement ratio is within 2/+4% for 25 May, and within 4/+5% at all wavelengths for 27 May (10-nm running average). The sensitivity test was repeated also for the upper and lower values of the imaginary part n 00, and similar results were found. Thus as expected, little can be said on the real part of the refractive index with this method. [55] A larger sensitivity on the imaginary part should emerge, since its variations imply changes of the amount of absorption by the particles. A limited sensitivity of this method in the case of small optical depth is somewhat

9 MELONI ET AL.: TROPOSPHERIC AEROSOLS IN THE MEDITERRANEAN, 2 AAC 4-9 Figure 6. Comparison of the model-measurement ratio obtained with the aerosol size distribution by DI, and with that by d Almeida et al. [1991] for (a) 25 May and (b) 27 May. The model simulations with the d Almeida et al. distribution have been performed using refractive index estimated in this study, and the refractive index for the water-soluble component by d Almeida et al. [1991]. expected, since the effects of the aerosol on the radiative transfer, and on the surface irradiance, are small when the optical depth is small. [56] DI have used aerosol optical depth and backscattering and particle counters profiles to derive estimates of vertically resolved aerosol refractive indices, on the same days (18, 25, and 27 May) of our analysis. A comparison between the obtained refractive indices requires some caution. It must be remarked that the aerosol refractive indices by DI are derived at 532 nm, and are estimated for the height region where aircraft and lidar measurements overlap. Our results apply to the whole vertical column, also including the lowest altitude, below the minimum observation that may be sounded by the lidar (about 500 m), and the region above the aircraft maximum flight altitude (about 4.5 km). The two analyses, however, are using basically the same aerosol size distributions (vertically averaged in our case), and the convergence of the results may indicate the consistency between the lidar backscattering and the spectral radiometric measurements. The values of n 00 retrieved on 25 May are somewhat larger than those found by DI. [57] The sensitivity of the simulated spectra on the aerosol size distribution was examined comparing the UV irradiance calculated using the aerosol extinction and absorption coefficients and the phase function from the three-modal size distribution from the work of d Almeida et al. [1991] for the maritime polluted aerosol type (see previous section), and the results obtained with the measured size distribution. The changes of the size distribution have been implemented without modifications to the extinction profile and aerosol optical depth at 415 nm. In both cases the refractive index is assumed to correspond to the best estimated values. [58] Figure 6 illustrates the comparison for 25 and 27 May. Using the size distribution by d Almeida et al. [1991] the model-measurement ratio remains within ±5%. If the water-soluble aerosol refractive index and the size distribution by d Almeida et al. [1991] are used together, differences of 7 9% (25 May) and 3 8% (27 May) on the 10-nm running average curve appear. The change of size distribution and refractive index implies a modification of the aerosol optical depth in the UV spectral range: the t at 350 nm is 2.7% higher on 25 May, and 7.3% lower on 27 May for the d Almeida et al. distribution than for the DI distribution. [59] The sensitivity on the ozone profile has been studied by calculating the surface irradiance for two different vertical distributions, both corresponding to the observed total ozone amount. The first profile, P1, is given by the climatology by Bhartia et al. [1985], and the second one, P2, is the combination of the tropospheric ozone measurements and the climatology (see Figure 4). Figure 7 shows the ratio R mc, between the calculated irradiance spectra obtained with profiles P2 and P1, respectively. The increase of tropospheric ozone associated with the change of the ozone vertical profile produces a reduction of R mc, i.e., a reduction of the surface irradiance by approximately 1% at 320 nm, 6% at 305 nm, and up to 10% at 295 nm, both on 25 and 27 May. The decrease of the irradiance is due to the efficient absorption by tropospheric ozone (the reader is reminded that the two ozone profiles correspond to the same total ozone value). This phenomenon is particularly important at small solar zenith angles, since for low Sun condition UV radiation is mostly scattered in the stratosphere, sub- Figure 7. Ratio of the UV irradiance calculated with the climatological ozone profile to that calculated with the climatological profile corrected to the measured tropospheric ozone for (a) 18 May, (b) 25 May, and (c) 27 May.

10 AAC 4-10 MELONI ET AL.: TROPOSPHERIC AEROSOLS IN THE MEDITERRANEAN, 2 Figure 8. Ratio of the modeled to the measured UV spectra for 18 May at 60 solar zenith angle. The curves refer to three different values of the imaginary part of the aerosol index of refraction. The thick solid curves are 10 nm running averages of the ratio. tracting influence to the enhanced tropospheric ozone [Brühl and Crutzen, 1989] High Aerosol Amount: 18 May 1999 [60] The dependence of the model results on the aerosol properties has been studied also for the UV spectrum of 18 May, obtained at 60 solar zenith angle. Existing estimates of the imaginary part n 00 of the atmospheric dust refractive index show a wide spread, which produces large uncertainties in the determination of the aerosol optical characteristics [Sokolik et al., 1993]. Figure 8 shows the model-measurement ratio for n 00 equal to , , and , with the real part n 0 fixed at The three selected values of n 00 produce, respectively, the +5% deviation, the best agreement, and the 5% deviation between model and measurements. These values identify a relatively narrow interval, which is indicative of the strong dependence of the surface UV irradiance on the aerosol absorption in the case of 18 May. Similarly to the previous cases, the real part n 0 does not seem to significantly affect the modeled spectra. The refractive index which gives the best agreement is i. [61] The imaginary part of the refractive index retrieved by DI above 3600 m, that is in the region of largest aerosol concentration, falls within the obtained interval of values. The refractive index inferred by the comparison between modeled and measured irradiances constitutes an essential information for the determination of the aerosol optical properties, like the single-scattering albedo w 0 and the asymmetry factor g. Table 3 shows values of w 0 and g calculated at 500 nm with the Mie theory for three separate aerosol layers, whose altitude limits are identified from the characteristics of the lidar profile [see DI]: from the ground to 2400 m, from 2400 to 3600 m, and from 3600 to 4200 m. It is worth noting that 4200 m is not the top of the aerosol layer (which reached 7 km), but the maximum altitude reached by the ultralight aircraft. The values reported in Table 3 are consistent with those computed for mineral dust by other authors. d Almeida et al. [1991] give a variability range of w 0 between and and of g between and at 500 nm for the desert aerosol in July. Sokolik and Toon [1996] found w 0 = 0.85 and g = 0.75 while Díaz et al. [2001] obtained w 0 = 0.87 and g = [62] To emphasize the role played by a detailed knowledge of the size distribution in conditions of high aerosol load, the surface irradiance spectrum was calculated assuming the measured vertical distribution and optical depth at 415 nm, and the two size distributions given by d Almeida et al. [1991] for desert dust: the desert background (wintertime) and wind carrying dust (summertime), that differ in the content of large particles (radius > 1 mm), which is larger for the second distribution [see DI]. Figure 9 shows the model to measurement ratio for the two distributions by d Almeida et al. [1991], using (1) the retrieved refractive index of i, and (2) the mineral aerosol refractive index tabulated by d Almeida et al. [1991]. In case (1) the model to measurement ratio is about 10 13% (wintertime) and 2 4% (summertime) higher than in the reference case (retrieved size distribution and refractive index), and is about 1 7% (wintertime) and 9 12% (summertime) lower in case (2). Thus the description of the aerosol optical properties through standard parameters can lead to an inadequate determination of atmospheric absorption and scattering. [63] The DDR, which is measured by the MFRSR, is correlated with the aerosol optical depth [see e.g., Yu et al., 2000]. The value of DDR is independent of the instrumental calibration (in the MFRSR measurements) and of the extraterrestrial solar spectrum (in the model results), and is a useful parameter for a comparison. The DDR was computed at 415 nm, in correspondence to the MFRSR measurements, for the three selected cases; the results are reported in Table 4. The model somewhat overestimates the measured DDR in the low aerosol cases, while a good agreement exists on 18 May. The reader is reminded that these results are relative to different values of the solar zenith angle, and should not be directly compared. [64] Also for 18 May, the model was run with different ozone vertical distributions with, in particular, different amounts in the troposphere. The enhancement of the Table 3. Aerosol Single-Scattering Albedo w 0 and Asymmetry Factor g at 500 nm Obtained With the Size Distribution of 18 May for Different Imaginary and Real Parts of the Refractive Index a w 0 g km km km km km km i i i i i a Calculations were performed for the three indicated layers.