UV-visible observations of atmospheric O 4 absorptions using direct moonlight and zenith-scattered sunlight for clear-sky and cloudy sky conditions

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 107, NO. D20, 4424, doi:10.1029/2001jd001026, 2002 UV-visible observations of atmospheric O 4 absorptions using direct moonlight and zenith-scattered sunlight for clear-sky and cloudy sky conditions T. Wagner, C. von Friedeburg, M. Wenig, C. Otten, and U. Platt Institut für Umweltphysik, University of Heidelberg, Heidelberg, Germany Received 5 July 2001; revised 23 February 2002; accepted 8 March 2002; published 22 October 2002. [1] Atmospheric observations of the O 4 absorption bands at 360.5, 380.2, 477.3, 532.2, 577.2 and 630.0 nm are presented for different atmospheric conditions (clear and cloudy skies) and viewing geometries (direct and zenith-scattered light observations). From the observations of direct moonlight it was possible to derive absolute O 4 absorption cross sections for atmospheric conditions. We found that the relative shape of the observed absorption bands is similar to those of the O 4 spectrum measured by Greenblatt et al. [1990] in the laboratory. However, in general (except for the absorption band at 380.2 nm), the O 4 absorption cross sections derived in this study are larger by several percent compared to those of the other (mainly laboratory) observations. Using the observations of zenith-scattered light, we investigated the radiative transport through the atmosphere. Our observations under cloudy sky conditions confirmed that the light path enhancement due to multiple Mie scattering on cloud droplets is independent of wavelength. From the observations under clear-sky conditions we studied the influence of Mie scattering on aerosol. It was not possible to describe the selected clear-sky measurements by taking into account only Rayleigh scattering. We found that the comparison of the O 4 measurements with model results for different sets of assumed aerosol extinctions provides a new, very sensitive tool to derive aerosol parameters from zenith sky ground-based measurements. INDEX TERMS: 0394 Atmospheric Composition and Structure: Instruments and techniques; 0305 Atmospheric Composition and Structure: Aerosols and particles (0345, 4801); 0320 Atmospheric Composition and Structure: Cloud physics and chemistry; KEYWORDS: radiative transport, UV-vis spectroscopy, DOAS, O 4 absorption, oxygen dimer Citation: Wagner, T., C. von Friedeburg, M. Wenig, C. Otten, and U. Platt, UV-visible observations of atmospheric O 4 absorptions using direct moonlight and zenith-scattered sunlight for clear-sky and cloudy sky conditions, J. Geophys. Res., 107(D20), 4424, doi:10.1029/2001jd001026, 2002. 1. Introduction [2] The absorption of oxygen molecules in the UV-visnear-IR spectral range corresponds to different types of chemical bonds and transitions. Besides the discrete structured ro-vibrational bands of the electronic transition of the O 2 molecule and the structured bands of the bound van der Waal s molecule O 4, oxygen shows also broad unstructured absorptions due to the collision induced oxygen dimer (O 2 ) 2 [see, e.g., Greenblatt et al., 1990; Solomon et al., 1998, and references therein]. These broadband absorptions were first described by Janssen [1885, 1886] who also showed that the intensity of these bands varies with the square of the oxygen pressure. Since then several studies investigated the (O 2 ) 2 absorptions and it was shown that under atmospheric conditions these bands contain no fine structure [see Solomon et al., 1998; Naus and Ubachs, 1999, and references therein]. Greenblatt et al. [1990] concluded from the weak Copyright 2002 by the American Geophysical Union. 0148-0227/02/2001JD001026$09.00 temperature dependence of the (O 2 ) 2 absorption that (under atmospheric conditions) they are most probably related to an oxygen collision complex rather than to a bound dimer. Nevertheless, within this study we will use the term O 4 for this (O 2 ) 2 collision complex since it was conventionally used in previous studies. [3] The O 4 absorption bands in the UV-vis spectral regions analyzed in this study (360.5 630.0 nm) belong to transitions between dimers of ground state oxygen molecules and electronically excited states of both of the involved oxygen molecules of O 4 (see Table 1). At larger wavelengths, transitions with only one excited oxygen molecule also occur, and under atmospheric conditions, N 2 can also serve as a collision partner [see, e.g., Solomon et al., 1998, and references therein]. [4] Atmospheric absorptions of O 4 are important in different fields of atmospheric radiative transfer modeling: First, the atmospheric absorption of solar radiation by O 4 is expected to be responsible for about 1 2% of the total atmospheric absorption of the solar radiation [Pfeilsticker et al., 1997; Solomon et al., 1998; Mlawer et al., 1998; Zender, AAC 3-1

AAC 3-2 WAGNER ET AL.: UV-VISIBLE OBSERVATIONS OF ATMOSPHERIC O 4 ABSORPTIONS Table 1. Wavelengths and Transition Assignments of the O 4 Absorption Bands in the UV-Visible Spectral Region Upper State of Transition Wavelength, nm (From 3 g + 3 g Ground State) 343.4 a 1 + g + 1 + g (n =2) 360.5 1 + g + 1 + g (n =1) 380.2 1 + g + 1 + g (n =0) 446.7 a 1 + g + 1 + g (n =1) 477.3 1 + g + 1 g (n =0) 532.2 1 g + 1 g (n =2) 577.2 1 g + 1 g (n =1) 630.0 1 g + 1 g (n =0) a These wavelengths were not investigated in this study. 1999]. Second, since the atmospheric column of oxygen varies only slightly (depending on pressure), observations of the atmospheric absorptions of O 2 and in particular of O 4 can serve as an indicator for variations of the radiative transport through the atmosphere. O 4 and O 2 observations by ground-based, and satellite instruments have been used, for example, to detect and characterize clouds and aerosols [Kuze and Chance, 1994; Van Roozendael et al., 1994; Erle et al. 1995; Wagner et al., 1998a, 1998b; Pfeilsticker, 1999; Pfeilsticker et al., 1998, 1999; Veitel et al., 1998; Koelemeijer and Stammes, 1999; Wagner, 1999]. Although most commonly the observation of the atmospheric O 2 absorption (e.g., the oxygen-a band) is used for cloud retrieval algorithms [see, e.g., Kuze and Chance, 1994; Koelemeijer and Stammes, 1999] they are subject to several general shortcomings: First, when the highly fine structured O 2 absorption is measured by low resolving instruments (as often used for atmospheric absorption spectroscopy [Van de Hulst, 1957; Platt et al., 1979, 1997; Noxon et al., 1979; Solomon et al., 1987; Platt, 1994]), saturation effects can make the measured O 2 absorption insensitive to changes of the column density along the light path [Veitel et al., 1998; Heidinger and Stephens, 2000; Funk, 2001]. Second, the atmospheric light paths for the observation of scattered radiation depend strongly on the absorption strength of atmospheric species. In particular, the shape of the measured (low resolution) O 2 absorption band can become significantly different from that of direct light observations [Platt et al., 1997; Marquard et al., 2000; Wagner et al., 2000]. Third, inelastic scattering processes (Raman scattering) on air molecules can cause a so-called filling-in of the narrow O 2 absorption lines [Grainger and Ring, 1962; Fish and Jones, 1995; Sioris and Evans, 2000]. [5] All these shortcomings can be almost completely avoided by studying the atmospheric O 4 absorptions, which are relatively broad (FWHM in the range of a few nm [Greenblatt et al., 1990]) and show no spectral fine structure. Thus they can be spectrally resolved by a typical DOAS instrument [Platt, 1994]. In addition, the O 4 absorption bands cover the whole UV and visible spectral range from about 340 to 630 nm. Thus O 4 absorption measurements are very well suited for the investigation of the radiative transport through the atmosphere. [6] O 4 absorption cross sections in the UV and visible spectral range were measured under laboratory conditions in many studies (see, for example, Salow and Steiner [1936], Herman [1939], Dianov-Klokov [1964], Greenblatt et al. [1990], Newnham and Ballard [1998], and Naus and Ubachs [1999]; an excellent overview is given by Solomon et al. [1998]). Most of these measurements were performed at high pressures (>1000 mbar) and temperatures different from those of the troposphere where most of the atmospheric O 4 absorption takes place. Several studies also presented measurements of the O 4 absorption in the atmosphere. Perner and Platt [1980] reported O 4 absorption cross sections (for the bands at 343.4, 360.5, 380.2, 446.7, 477.3, 577.2 nm) for surface near atmospheric conditions derived by long path absorption DOAS measurements using an artificial light source. Volkamer [1996] derived O 4 absorption cross sections (for the bands at 343.4, 360.5, 380.2, 477.3, 532.2, 577.2 and 630.0 nm) under similar conditions using a multi-path reflection cell. O 4 absorption cross sections for the total atmospheric column were derived by Mlaver et al. [1998] for the bands in the near-infrared (1140 and 1260 nm) using solar radiation. From the observation of direct moonlight, Solomon et al. [1998] showed that the O 4 absorption of the atmospheric column (at 630.0 nm) is consistent with laboratory measurements by Greenblatt et al. [1990]. Osterkamp et al. [1998] determined O 4 absorption cross sections using atmospheric balloon soundings under varying atmospheric conditions (7 500 mbar, 203 260 K) for the absorption bands at 432, 477.3, and 577.2 nm [see also Pfeilsticker et al., 2001]. [7] Here we provide a set of O 4 absorption cross sections derived for the total atmospheric column (for Arctic winter conditions) including six O 4 bands from 360.5 to 630.0 nm. These O 4 absorption cross sections were obtained from direct moonlight observations, for which the atmospheric radiative transport can be modeled with high accuracy (because of the well defined geometry of the absorption path). Preliminary results of these observations were already presented by Wagner et al. [1996]. Our data set complements the observations from Mlawer et al. [1998] and Solomon et al. [1998] derived for a similar observation geometry. In this study (sections 2.2 and 2.3) we apply this O 4 cross sections to the interpretation of zenith-scattered light measurements performed with the same instruments for both clear-sky and cloudy sky conditions. From the observations under a heavy cloud cover it was possible to characterize the influence of Mie scattering; from the observations under clear sky the influence of atmospheric aerosols on the radiative transfer was investigated. We find that O 4 observations of zenithscattered light provide a new, very sensitive method for the investigation of atmospheric aerosols. 2. Instrumental Setup and Data Evaluation 2.1. Instruments [8] Two grating spectrometers were used to measure spectra of direct moonlight and zenith-scattered sunlight in the UV (318 384 nm) and visible (375 nm to 688 nm) spectral ranges. The spectral resolution (FWHM) ranges from 0.3 0.5 nm (4.5 7 pixels) for the UV instrument and from 1.1 3.1 nm (3.5 10 pixels) for the instrument of the visible spectral range. The spectral resolution of the visible spectrometer is <2 nm at wavelengths <600 nm. The spectral resolution of our instruments is thus much smaller than the widths of the O 4 absorption bands, which ranges from about 4 nm (FWHM) in the UV to about 14 nm in the red spectral region [Greenblatt et al., 1990]. Thus our

WAGNER ET AL.: UV-VISIBLE OBSERVATIONS OF ATMOSPHERIC O 4 ABSORPTIONS AAC 3-3 Table 2. Reference Spectra Used in the DOAS Analysis of the Different O 4 Absorption Bands a O 4 Band, nm Wavelength Range, nm Pixel Range Reference Spectra b Degree of Polynomial 360.5 347.1 375.3 420 820 O 4,O 3,NO 2,Ring 3 380.2 365.4 388.7 680 1010 O 4,NO 2,Ring 3 477.3 431.3 526.1 550 850 O 4,O 3,NO 2,H 2 O, Ring 3 532.2 488.7 556.7 450 670 O 4,O 3,NO 2,H 2 O, Ring 3 577.2 538.4 616.4 250 510 O 4,O 3,NO 2,H 2 O, Ring 3 630.0 586.8 654.2 120 350 O 4,O 3,O 2,NO 2,H 2 O, Ring 3 360.5, 380.2 c 347.1 388.7 420 1010 O 4,O 3,NO 2,2 Ring 5 477.3, 532.2, 577.2, 630.0 c 415.1 682.9 20 900 O 4,O 3,O 2,NO 2,2 H 2 O, Ring 7 a The results of the single band evaluations were used for the Langley plots. The multiband evaluations were used to determine the O 4 absorption spectra for atmospheric conditions (Figure 1). b For the analysis of the moonlight measurements no Ring spectra were included in the spectral analysis. c For these DOAS algorithms the (logarithm of the) Fraunhofer spectrum was included in the fitting procedure, and the reference spectra were fitted to the logarithm of the measured spectrum instead of to the logarithm of the ratio formed by the measured spectrum and the Fraunhofer spectrum. instruments are well suited to resolve the atmospheric O 4 absorption bands over the whole spectral range from 360.5 to 630.0 nm investigated in this study. More detailed information on our instrumental set-up is given by Stutz [1991], Fiedler et al. [1993], Otten et al. [1998], and Wagner et al. [1998a]. 2.2. Spectral Evaluation [9] The measured spectra were analyzed using the DOAS technique [Platt, 1994]. To remove the strong Fraunhofer structures the spectra were divided by a spectrum taken at low zenith angle J (this spectrum is in the following called Fraunhofer spectrum). Several reference spectra (see below) of the trace gases which show structured absorptions in the respective wavelength regions (including the O 4 cross section) as well as a polynomial (of degree up to 7, see Table 2) are fitted to the logarithm of this ratio spectrum using a non linear least squares algorithm [Stutz and Platt, 1996]. The polynomial accounts for broadband absorption structures as well as for atmospheric Mie and Rayleigh scattering. For the correction of the filling in of the Fraunhofer lines in the spectra of scattered sunlight (the so-called Ring effect [Grainger and Ring, 1962; Bussemer, 1993; Fish and Jones, 1995]) one or two Ring spectra are also included into the fitting routine. The first Ring spectrum was calculated assuming that Rayleigh scattering was the dominant atmospheric scattering process, the second Ring spectrum assuming that Mie scattering was the dominant atmospheric scattering process. Compared to the first Ring spectrum the amplitude of the second Ring spectrum increases towards smaller wavelengths. This reflects the strong difference of the wavelength dependence of Raman scattering (and Rayleigh scattering) compared to that of Mie scattering [Wagner, 1999]. Using two Ring spectra in the DOAS analysis of scattered light spectra minimizes the errors of the fitting results, especially when large wavelength ranges are analyzed. In this study we analyze the atmospheric O 4 absorptions in relatively small wavelength ranges around the different individual bands (360.5, 380.2, 477.3, 577.2, and 630.0 nm) as well as over larger wavelength ranges including several bands (see Figure 1). The selected trace gas reference spectra and the parameters of the different DOAS analysis are listed in Table 2. The following trace gas reference spectra were used: (1) an O 4 absorption cross section [Greenblatt et al., 1990] for 296 K, which was interpolated to the instrumental dispersion, (2) up to two H 2 O absorption spectra (for two column densities of 1 and 20 10 22 molec/cm 2 ) derived from the HITRAN database [Rothman, 1992] and convoluted to the spectral resolution of our instruments, (3) an O 3 absorption cross section for 223 K [Burrows et al., 1999] convoluted to the spectral resolution of our instruments, and (4) a NO 2 absorption cross section for 238 K [Harder et al., 1997] convoluted to the spectral resolution of our instruments. The wavelength calibration was performed by fitting the measured spectra to a highly resolved solar spectrum [Kurucz et al., 1984], which was convoluted to the spectral resolution of our instruments. Examples of the DOAS O 4 analysis including several bands are shown in Figure 1. For measurements with strongly enhanced O 4 absorptions (e.g., under heavy clouds, see section 2.2) it was possible to compare the wavelength calibration of the O 4 absorption cross sections to the O 4 absorptions in the atmosphere. We found slight shifts of the positions of the different bands (between +0.8 nm and 0.34 nm, see Table 3) and we used this new wavelength calibration for the O 4 analysis in our study. However, the influence of this new calibration on the derived O 4 absorptions was negligible (less than 0.5%). [10] Because the Fraunhofer spectrum also contains atmospheric absorption structures of O 4 (and other trace gases) the result of the DOAS analysis represents the difference of the slant column density (SCD, trace gas concentration integrated along the light path) of the measured spectrum observed at a zenith angle J and that of the Fraunhofer spectrum (SCD Fraunh ) measured at zenith angle J F. SCD diff ðjþ ¼ SCDðJÞ SCD Fraunh ðj F Þ ð1þ To derive the absolute SCD of the measured spectrum SCD Fraunh has to be independently determined and added to the SCD diff derived by the DOAS analysis. [11] The ratio of SCD and the vertical column density (VCD, vertically integrated trace gas concentration) defines the air mass factor (AMF), which can be calculated by radiative transfer models [Noxon et al., 1979; Solomon et al., 1987; Dahlback and Stamnes, 1991; Perliski and Solomon, 1993; Sarkissian et al., 1995; Rozanov et al., 1997; Marquard et al., 2000]: CD ¼ SCD diff ðjþþscd Fraunh ðj F Þ AMFðJÞ ð2þ

AAC 3-4 WAGNER ET AL.: UV-VISIBLE OBSERVATIONS OF ATMOSPHERIC O 4 ABSORPTIONS Figure 1. Atmospheric O 4 absorption spectra for different viewing geometries and different atmospheric conditions. Displayed is the O 4 absorption cross section of Greenblatt et al. [1990] scaled to O 4 absorption found in the logarithm of the ratio of the measured spectrum and Fraunhofer spectrum (thin line). In all three panels, simultaneous measurements in the UV- and the vis-spectral range (made by two different instruments) are combined into one spectrum. In Figure 1a, O 4 observations of direct moonlight are shown. Observations of zenith-scattered sunlight for cloudy sky (Figure 1b) and clear-sky conditions (Figure 1c) are displayed below. Because of the strong wavelength dependence of Rayleigh scattering for the observations under clear-sky conditions it was not possible to perform the spectral evaluation simultaneously over a large wavelength range (containing several O 4 absorption bands as for Figures 1a and 1b); thus results for the single band evaluations are combined into one spectrum (the fitting parameters are summarized in Table 2).

WAGNER ET AL.: UV-VISIBLE OBSERVATIONS OF ATMOSPHERIC O 4 ABSORPTIONS AAC 3-5 Table 3. Modification of the Wavelength Calibration of the O 4 Absorption Cross Section of Greenblatt et al. [1990] With Respect to the O 4 Absorptions of the Atmosphere a O 4 Absorption Band Spectral Shift Necessary to Match the Atmospheric O 4 Absorptions, nm 360.5 nm +0.56 ± 0.13 380.2 nm +0.80 ± 0.18 477.3 nm +0.05 ± 0.14 532.2 nm 0.15 ± 0.25 577.2 nm 0.36 ± 0.10 630.0 nm 0.15 ± 0.14 a The shifts of the different bands were determined with respect to the strongly enhanced O 4 absorptions under heavy clouds (see Figure 1b and section 2.2). The best shift was determined for the minimum of the residual structure. The range of uncertainty of the shifts was (somewhat arbitrarily) set to the values for which the residual structures increased up to more than 20% compared to the minimum value. For trace gases that show no diurnal variation (such as O 4 ), SCD Fraunh can be determined by the method of Langley plots. The derived differential slant column densities for different zenith angles are plotted as a function of the respective AMF: SCD diff ðjþ ¼ VCD AMFðJÞ SCD Fraunh ðj F Þ ð3þ The ordinate of the fitted line represents the SCD Fraunh ;the slope represents the VCD. For ground-based observations of absorbers like O 4 the Langley plot method is best suited to determine the VCD, in particular, because it makes use of the information of several individual measurements. Thus we apply this method in this study. [12] For satellite measurements (e.g., by the GOME instrument aboard ERS-2 [Bednarz, 1995; European Space Agency, 1996]) the situation can, however, be completely different because usually an extraterrestrial sun spectrum can serve as Fraunhofer reference spectrum. Such spectra contain no atmospheric absorptions and the VCD can easily be obtained from a single observation [see, e.g., Wagner et al., 1998b; Wagner, 1999]. 2.3. Modeling of the Radiative Transport [13] In this study the AMFs for direct light and for zenith-scattered light observations are calculated using the radiative transfer model AMFTRAN [Marquard, 1998; Marquard et al., 2000]. It is based on a multiple scattering Monte Carlo code for full spherical geometry taking into account atmospheric refraction. Multiple scattering is calculated for Molecules (Rayleigh scattering), aerosol particles (Mie scattering) and for the Earth s surface (reflection). Aerosols are characterized by their extinction and scattering properties. [14] AMFs for direct light observations are independent of wavelength; they only depend on the (relative) atmospheric concentration profile of the considered species. In the case of O 4 the profile can be easily determined from temperature and pressure profiles measured, for example, by radiosondes. Because of the well-defined geometry of direct light observations the accuracy of such AMFs is much better than for scattered light observations. From internal comparisons between different radiative transport models we estimate the accuracy of the AMFs for direct light O 4 observations used in this study to about 1%. [15] In contrast, the AMFs for scattered light observations are subject to relatively large uncertainties. Because of the wavelength dependence of Rayleigh and Mie scattering and the atmospheric trace gases (mainly O 3 ) they strongly vary with wavelength. In addition they depend on the atmospheric trace gas concentration profile and on the atmospheric aerosol load. They especially depend strongly on clouds: for (mainly) stratospheric absorbers the effect of an increasing altitude of the scattering surface (from ground to cloud top height) and the change of the albedo are the dominant effects [Wagner et al., 1998a; Richter and Burrows, 2002]. For tropospheric absorbers like O 4, however, the changes can become even much larger due to multiple Mie scattering inside clouds and/or reflection between several cloud layers and the earth s surface. Compared to clear-sky conditions these changes can be up to several hundred percent [Erle et al., 1995; Wagner et al., 1998a; Pfeilsticker et al., 1999]. [16] To date, realistic 3-dimensional modeling of clouds is very difficult, and in addition, appropriate cloud parameters for a given atmospheric situation are generally not available. Thus in this study zenith-scattered light measurements of O 4 are quantitatively compared with modeled AMFs only for clear-sky conditions. However, also for clear-sky conditions Mie scattering on aerosols can significantly affect the AMFs for tropospheric absorbers like O 4. Thus the comparison between measured and modeled AMFs for clear-sky conditions allows us (1) to test the modeling of the atmospheric radiative transfer in general and (2) to determine the atmospheric aerosol extinction profile which yields the best agreement between the O 4 measurements and the modeled O 4 AMFs (see section 2.3.3). 2.4. Direct Light Observations [17] According to the astronomical moon parameters direct moonlight observations were possible within two periods during the measurement campaign in January to March 1994 in Kiruna (northern Sweden, 68 N, 21 E). Please note that poleward from the polar cycle the moon stays above the horizon for several days around full moon. After the moon was sufficiently bright (after 22.01. and 18.02) our moon observations started; they ended after the moon set below the horizon on (04.02. and 04.03.). The most sensitive measurements were possible around 28.01 and 25.02. when the moon was full (31.01. and 28.02.) and the range of LZAs includes small as well as large values. For the quantitative determination of the O 4 cross sections we chose the observation during the night from January 28 to 29 because during this night the height profiles of the atmospheric pressure and temperature above the measurement site were measured by an ozonesonde (launched shortly after midnight). Also during this night a large number of individual measurements at different LZAs was possible. 2.4.1. Spectra [18] In Figure 1a the result of a spectral evaluation of all six O 4 absorption bands between 360.5 and 630.0 nm is shown for the night of 28./29.01. The O 4 absorption spectrum from Greenblatt et al. [1990] is scaled to the O 4 absorptions in the measured spectrum. The reference spectra of all trace gases which show structured absorptions in the

AAC 3-6 WAGNER ET AL.: UV-VISIBLE OBSERVATIONS OF ATMOSPHERIC O 4 ABSORPTIONS respective wavelength ranges were simultaneously fitted over the whole wavelength ranges (347 389 nm for the UV and 415 681 nm in the visible spectral range) as described in section 1.2. This is possible because for direct light measurements the AMF does not depend on the wavelength, and correspondingly the atmospheric O 4 absorptions are in general well described by the laboratory cross section measured by Greenblatt et al. [1990] over the whole wavelength range. Also some small deviations are found: The relative strengths of the different O 4 bands in the atmosphere seem to be slightly different from those of the Greenblatt spectrum. The bands at 360.5 and 577.2 nm are by about 5 6% larger compared to those at 360.5 and 477.3 nm, respectively. The band at 630.0 nm is about 1 2% larger compared to that at 477.3 nm. 2.4.2. O 4 Absorption Cross Sections [19] To derive absolute values of the O 4 absorption cross sections one needs both the O 4 VCD (e.g., derived from a radiosonde) and the corresponding O 4 absorption derived from the measured spectra. To determine the optical density (corresponding to the O 4 VCD), the Langley plot method (equation (3)) can be applied: instead of the column density SCD diff here we use the measured (differential) optical density (OD diff ) of a specific O 4 absorption. That means that we replace the column densities in equation (3) by the respective optical densities (OD diff = SCD diff s, OD VCD = s VCD, with s the O 4 absorption cross section): OD diff ðjþ ¼ OD VCD AMFðJÞ OD Fraunh ðj F Þ ð4þ The OD VCD derived from the slope of the Langley plot represents the optical density of the vertical absorption path through the atmosphere. In Figure 2 the respective Langley plots of the O 4 absorptions at 360.5, 380.2, 477.3, 532.2, 577.2, and 630.0 nm for the night from 28 to 29 January 1994 are shown. For all bands the measured absorptions fall on the expected straight lines within their errors. Due to the strong extinction by Rayleigh scattering in the UV these observations were only possible for LZA 80. [20] Dividing the derived OD VCD by the O 4 VCD yields the O 4 absorption cross section: s ¼ OD VCD =VCD ð5þ Figure 2. Langley plots for the direct moonlight observations during the night 28./29.01. 1994. The absorptions of six O 4 bands are plotted as a function of the calculated AMF. From the slope (fitted line) the absorption corresponding to the vertical atmospheric absorption path is derived. The LZA range for the observations in the visible is between 60 and 85. Because of the strong attenuation of the moonlight by Rayleigh scattering the measurements in the UV were only possible up to about LZA = 85. It should be noted that in this study we express the O 4 absorption cross section with respect to the quadratic O 2 concentration, because the equilibrium constant between O 4 and (O 2 ) 2 is not known [see also Greenblatt et al., 1990]. The O 4 absorption cross section therefore includes also this equilibrium constant and it has the units [cm 5 /molec 2 ]. In order to indicate this speciality we will, in this study, from here on, use the symbol s* for the absorption cross section of O 4. Accordingly also the O 4 VCD will be expressed with respect to the quadratic O 2 concentration VCD* O 4 ¼ Z 1 ð z o O 2 Þ 2 dz It has the units [molec 2 /cm 5 ] and we will use the symbol VCD* from here on. From the temperature and pressure profiles measured by the ozonesonde launched shortly after midnight we calculated the O 4 VCD* = 1.26 10 43 molec 2 / cm 5. This calculation takes into account that the observation site (485 m above see level) is located at a higher altitude than the launch site. For our observations we calculated an effective temperature T eff by weighting the atmospheric temperature profile with the quadratic O 2 concentration; we derived T eff = 242 K. Please note that this linear averaging does not take into account a possible non linear dependence of the O 4 absorption cross section on temperature. [21] The errors of our O 4 absorption cross sections result from different contributions: (1) the errors of the spectral retrieval, (2) the errors of the calculated AMFs, (3) the errors of the Langley-plots, and (4) the errors of determining the O 4 VCD* from the radiosonde data. The 1-s-errors of the spectral analysis are between <1% and several 10%. They are displayed as error bars in Figure 3. The error of the AMF for direct light observations is assumed to be <1%

WAGNER ET AL.: UV-VISIBLE OBSERVATIONS OF ATMOSPHERIC O 4 ABSORPTIONS AAC 3-7 Figure 3. O 4 absorption cross sections s* determined from our direct moonlight observations compared to those derived from previous studies plotted as a function of temperature. The lines represent least squares fits to all measurements except those from Perner and Platt [1980]. (see also section 1.3). From the Langley-plots the error of the slope of the fitted line was determined by varying the slope within the error bars of the measurements. The error of the O 4 VCD* calculated from the pressure and temperature data of the radiosonde is estimated to be about 2%. For the O 4 bands at 360.5, 380.2, and 532.2 nm the total error is dominated by the error of the slope of the Langley-plot. It varies between 9% for 360.5 nm and 30% for 532.2 nm. For all other bands the error is dominated by the uncertainty of the O 4 VCD*. The total error of these bands is about 3%. [22] In Figure 3 and Table 4 the temperature dependent O 4 absorption cross sections for the six O 4 bands are compared to those of previous studies. The lines in Figure 3 represent a least squares fit to all values (except those of Perner and Platt [1980] because their uncertainties significantly exceed those of the other measurements). Please note that many measurements errors as well as the spread of the data points is large indicating the poor knowledge about the temperature dependence. For all considered O 4 absorption d Table 4. O 4 Absorption Cross Sections s* Determined From the Direct Moonlight Observations Compared to Those Derived From Previous Studies a Naus and Ubachs [1999] (294K) Newnham and Ballard [1998] (283K) Newnham and Ballard [1998] (223K) Osterkamp et al. [1998] (256K) Osterkamp et al. [1998] (204K) Volkamer [1996] (296K) Perner and Platt [1980] (279K) Greenblatt et al. [1990] (196K) This Work (242K) Greenblatt et al. [1990] (296K) O 4 Band, nm 360.5 5.70 ± 0.5 4.1 ± 0.4 5.7 ± 0.6 5.4 ± 1.5 5.42 ± 0.7 380.2 2.44 ± 0.4 2.4 ± 0.2 3.7 ± 0.4 < 1.4 2.4 ± 0.2 477.3 7.80 ± 0.2 6.3 ± 0.6 7.6 ± 1.3 5.9 ± 1.8 6.1 ± 0.3 7.9 ± 0.3 7.0 ± 0.3 6.99 ± 0.35 8.34 ± 0.83 532.2 1.74 ± 0.5 1.3 ± 0.3 1.4 ± 0.2 1.2 ± 0.2 1.31 ± 0.2 1.23 ± 0.38 577.2 13.50 ± 0.4 11 ± 1 16 ± 6 10.3 ± 0.3 12.2 ± 0.4 13.6 ± 0.4 12.61 ± 0.11 11.75 ± 0.2 11.41 ± 0.5 630.0 9.61 ± 0.3 7.2 ± 0.7 6.2 ± 0.6 8.8 ± 0.13 7.9 ± 0.15 7.55 ± 0.5 a Here s* values are given in 10 46 cm 5 /molecule 2.

AAC 3-8 WAGNER ET AL.: UV-VISIBLE OBSERVATIONS OF ATMOSPHERIC O 4 ABSORPTIONS bands an increase of the peak absorption with decreasing temperature is found: about 13%/100K at 477.3 and 532.2 nm, about 20% at 360.5 and 577.2 nm, and about 33% at 380.2 and 630.0 nm [see also Pfeilsticker et al., 2001]. Except for the band at 380.2 nm the O 4 absorption cross sections derived in this study are slightly larger compared to the line fitted to all measurements. [23] Our measurements add new information on the O 4 absorption cross sections for a temperature region (around 242 K), where only very few previous measurements exists. This temperature (and also pressure) best represents the conditions for O 4 absorption measurements of the whole atmospheric O 4 column performed from ground-based or space-borne instruments. Thus they are well suited for the use in atmospheric applications (in particular for polar latitudes). Within this study we apply these O 4 absorption cross sections to zenith sky measurements using the same instruments (see sections 2.2 and 2.3). 2.5. Zenith-Scattered Light Under Cloudy Sky Conditions 2.5.1. Spectra [24] Clouds affect the atmospheric light paths in several ways compared to clear-sky conditions. First, because of multiple Mie scattering inside the clouds the light path can be largely increased, leading to enhanced absorptions of trace gases that are present inside the cloud [Erle et al., 1995; Van Roozendael et al., 1994; Wagner et al., 1998a; Pfeilsticker et al., 1999]. In particular under heavy clouds the absorptions of O 4 and other tropospheric trace gases can be strongly enhanced. It should be noted that for the observation of zenith-scattered light especially thin high clouds can also reduce the absorption paths below the cloud. For details of this effect, see Wagner et al. [1998a] and Pfeilsticker et al. [1999]. Here we consider a vertically extended cloud (from 200 m to 11 km) on March 5, 1994, when a warm front passed over the measurement site (see Figure 6, left panel). This case is described in more detail by Wagner et al. [1998a]. [25] Because of the large size of cloud particles Mie scattering in clouds is expected to show no wavelength dependence [Van de Hulst, 1957]. Thus the modifications of the light path due to thick clouds are expected to be similar over a large wavelength range. The observation of the O 4 absorptions at different UV-vis wavelengths can be used to test this expectation. In addition, because of the large O 4 absorptions derived under heavy clouds the shape and spectral position of the atmospheric O 4 absorptions can be compared to that of laboratory measurements. [26] In Figure 1b the DOAS fitting results for O 4 are shown for two measurements under the heavy cloud cover during March 5, 1994. All reference spectra including O 4 are simultaneously fitted to the logarithm of the ratio of two spectra measured during this day. The small average intensity observed in the first spectrum (7:40 UT) indicates a relatively thick cloud cover (see Figure 4); the much higher intensity of the second spectrum (measured about two and a half hours later at 10:14 UT) indicates a relative thin cloud cover. Correspondingly the light path enhancement and thus the O 4 absorption differ strongly in both spectra, leading to a strong O 4 absorption in the difference spectrum (Figure 1b). It should be noted that compared to the moonlight observations especially in the UV the signal to noise ratio is much better. It is obvious that (similar to the moonlight observations) the laboratory spectrum of Greenblatt et al. [1990] is well suited to describe the atmospheric observations under cloudy sky conditions over a large wavelength range. This confirms that light path modifications due to the clouds are essentially independent of wavelength as expected from the Mie-theory. [27] It should be noted that (similar to the direct moon measurements) also here the relative strength of the different O 4 bands in the atmosphere seem to be slightly different from that of the Greenblatt spectrum. The bands at 360.5 and 577.2 nm are by about 5 6% larger compared to those at 360.5 and 477.3 nm, respectively. The band at 630.0 nm is about 2% larger compared to that at 477.3 nm. From the observations for cloudy sky conditions we investigated the spectral calibration of the O 4 cross section of Greenblatt et al. [1990] (see section 1.2). We found small changes for the different absorption bands (see Table 3). 2.5.2. Diurnal Variation [28] Besides from meteorological observations (in particular satellite images) information about the cloud cover over the measurement site can be derived also from the measured spectra. From additional Mie scattering the observed intensities are usually enhanced by clouds for zenith scattering measurements. Moreover, because of the different wavelength dependence of Mie and Rayleigh scattering the scattered light is affected differently at different wavelengths; the ratio of these intensities, the so-called color index (CI), is thus a very sensitive indicator of the appearance of clouds. The diurnal variation of the average intensity, the O 4 absorptions for 380.2, 477.3, 577.2 and 630.0 nm, as well as the CI (here defined as the ratio of the intensities at 670 and 388 nm; see, for example, Sarkissian et al. [1991] and Enell et al. [1999]) is displayed in Figure 4. In order to allow a direct comparison for all wavelengths here the fit coefficients of the O 4 reference spectrum [Greenblatt et al., 1990] are displayed rather than the optical densities (which are different for several O 4 absorption bands). It is obvious that the variation of the O 4 fit coefficients for all selected bands is nearly identical during the whole cloudy day, which is another confirmation of the wavelength independence of Mie scattering in clouds. [29] It should be noted that in addition to the O 4 absorptions also the absorptions of O 3 (in the Chappuis and Huggings bands), NO 2 and H 2 O show a similar variation during that day, also caused by the light path modifications due to the clouds. The derived average light path enhancement (compared to clear-sky conditions) due to multiple Mie scattering was determined as up to 100 km for these conditions [Wagner et al., 1998a]. 2.6. Zenith-Scattered Light Under Clear-Sky Conditions 2.6.1. Spectra [30] Zenith-scattered light observations of O 4 absorptions under clear-sky conditions allow us to test the understanding and modeling of the atmospheric radiative transport. Because of the strong wavelength dependence of Rayleigh scattering on air molecules the optical depth of the atmosphere varies strongly with wavelength. As a consequence for large solar zenith angles (SZA > 75 ) the AMF for the various O 4

WAGNER ET AL.: UV-VISIBLE OBSERVATIONS OF ATMOSPHERIC O 4 ABSORPTIONS AAC 3-9 Figure 4. Diurnal variation of the color index (CI, ratio of the intensities at 670 and 388 nm) and the average intensity (both in the upper panel), as well as the O 4 absorptions (expressed as fit coefficients, see text) for 380.2, 477.3, 577.2 and 630.0 nm (lower panel) for measurements under a heavy cloud cover (Kiruna, 5.3.1994). absorption bands differs strongly. This finding can be clearly seen in Figure 1c, where the DOAS analysis for O 4 is shown for a clear day (22.03.1994) at Kiruna. All reference spectra including O 4 are simultaneously fitted to the logarithm of the ratio of two spectra measured at different SZA (89.5 and 75.3 ) on that day. However, in contrast to the direct moon measurements or the observations under cloudy sky conditions, in this case it is not possible to analyze the whole spectral range simultaneously, because of the strong wavelength dependence of the clear-sky AMF for tropospheric species. For example, in Figure 1c the absorption of the O 4 band at 630.0 nm is similar to that at 577.2 nm although the respective O 4 absorption cross sections differ by nearly a factor of two. This finding can be attributed to the difference of the AMFs for both wavelengths. Also the absorption band at 477.3 nm is much weaker than expected from the laboratory cross sections, and the bands at 360.5 and 380.2 are even inverted indicating a larger O 4 absorption in the spectrum measured at smaller SZA. 2.6.2. Diurnal Variation [31] In Figure 5 the diurnal variation of the average intensity, the O 4 absorptions for 380.2, 477.3, 577.2 and 630.0 nm, and the CI (similar to Figure 4) at Kiruna for a mostly clear day (22.03.1994) are shown. The Fraunhofer spectrum used for the analysis of the O 4 absorption was taken at a SZA of 67. From the measured CI and the average intensity as well as from satellite images of that day (Figure 6) we conclude that during the morning the sky above Kiruna was clear while during the afternoon thin clouds appeared. Correspondingly during the morning the difference between the O 4 absorptions for different wavelengths is large, while during the evening the difference becomes smaller. 2.6.3. Modeling of the Radiative Transport Under Clear-Sky Conditions [32] We selected the morning measurements of March 22, 1994 as a test case for the modeling of the atmospheric radiative transport under clear-sky conditions (our measurements during that winter were stopped a few days later and no totally clear day appeared within that period). In the upper panel of Figure 7 the morning O 4 absorptions (as shown in Figure 5) are plotted, now as a function of the SZA. In order to allow a direct comparison with the model results we here express the measured O 4 absorptions as

AAC 3-10 WAGNER ET AL.: UV-VISIBLE OBSERVATIONS OF ATMOSPHERIC O 4 ABSORPTIONS Figure 5. Similar to Figure 4, but for measurements under a mostly clear sky (Kiruna, 22.3.1994). It should be noted that while in general (for medium and small SZA, see, for example, the measurements around noon) clouds enhance the CI, for large SZA the dominant effect of clouds can also become their attenuation of the light before it reaches the zenith. In such cases the CI can also be reduced by clouds compared to clear-sky conditions. This is, for example, the case for the CI in the evening of March 22. AMF. For that purpose a value for the O 4 VCD* (1.27 10 43 ) was used which was calculated from a radiosonde launched at Sodankylä (about 300 km in the east of Kiruna), because on that day at Kiruna no radiosonde data were available. [33] Also shown in Figure 7 are the modeled AMFs for a pure Rayleigh atmosphere as well as for different tropospheric aerosol loadings. Both the measured and modeled O 4 AMFs are expressed as differences to the respective values for a SZA of 67 ; at this SZA the Fraunhofer spectrum was measured. [34] While the relative SZA dependence for the O 4 absorptions at different wavelengths is already qualitatively well described by the modeled AMFs for a pure Rayleigh atmosphere, especially for 477.3, 577.2, and 630.0 nm the absolute values are by about a factor of up to four larger than the measured ones. This discrepancy can be attributed to the influence of aerosols (or sub visible clouds) on the measurements. In order to investigate the influence of aerosol on the radiative transport in more detail we calculated O 4 AMFs for several cases including the effects of aerosols by varying both the altitude range and the optical depth of the assumed aerosol extinction. It should be noted that taking into account only the influence of background aerosols (various profiles with different total optical depths) it was possible to reduce the absolute values of the O 4 AMFs but it was impossible to simulate the correct SZA dependence for large SZA (unfortunately, no data on the aerosol extinction profiles were available for the day of our measurements). The respective O 4 -AMFs are very similar to those for assumed layer heights of 0 1 km, see Figure 7. [35] The resulting AMF (-differences) are presented in the lower part of Figure 7. Several general findings can be obtained: (1) In general, an increase of the optical depth of the aerosol extinction at a given altitude reduces the corresponding O 4 AMFs (as an exception an aerosol layer near the surface can also slightly enhance the AMF for small SZA). (2) Increasing the altitude of the assumed aerosol layer shifts the maximum of the modeled AMFs

WAGNER ET AL.: UV-VISIBLE OBSERVATIONS OF ATMOSPHERIC O 4 ABSORPTIONS AAC 3-11 Figure 6. Satellite images (IR, 10.3 11.3 mm) for Scandinavia on March 5, 1994, when a heavy cloud band passed over Kiruna (left) and for 22.3.1994, when the sky was mainly clear (right). Kiruna is indicated by the circle. The images are obtained via the Dundee Satellite Receiving Station, Dundee University, Scotland (http://www.sat.dundee.ac.uk/). towards higher SZA. Finding 1 can be explained by the increase of the scattering altitude of the observed light (from the zenith) for an increased aerosol extinction (see also section 2.2.1). Finding 2 can be explained by the fact that at large SZA (when the sun is below the horizon) the observed light can have already penetrated near to the surface before it is scattered in the zenith. This is, however, only possible if the light extinction by aerosols near the ground is small. For a large aerosol absorption near the ground, such light paths are strongly attenuated and thus can hardly contribute to the measured signal. Assuming in contrast that an aerosol absorption of similar optical depth occurs near to the tropopause, the near surface light paths are only weakly attenuated. In that case near surface light paths can contribute significantly to the measured signal leading to large O 4 AMFs at large SZA. [36] From the results displayed in Figure 7 it becomes evident that the O 4 absorptions under clear-sky conditions are very sensitive to variations of the tropospheric aerosol extinction. This sensitivity in turn can also be utilized to derive aerosol properties from O 4 measurements under clear-sky conditions. [37] To illustrate this possibility we tried to determine a set of O 4 AMFs (according to a specific aerosol extinction profile), which best describes the selected O 4 measurements for 380.2, 477.3, 577.2 and 630.0 nm. For that purpose we varied the tropospheric aerosol profiles over a wide range including various altitude distributions and aerosol extinctions. We varied the height of the assumed aerosol layer from the surface to 14 km and the width from 1 km to 14 km. The aerosol extinction was varied between 0.01/km and 0.5/km. For all AMF calculations we assumed a background aerosol extinction according to a clear remote atmosphere (taken from the LOWTRAN database [Kneizys et al., 1988]), and at specific altitudes we added the layers of a stronger aerosol extinction. The best agreement between the model results and our measurements was obtained for an assumed additional aerosol layer between 11 and 12 km with an extinction of 0.1/km. This finding might indicate that during the observations a thin (sub visible) high cirrus cloud layer might have been present; some hints of such a cloud layer could also be concluded from the satellite images (Figure 6, right panel). [38] Langley-plots using the above-described AMFs for the measurements during the morning of March 22 are shown in Figure 8. As in Figure 7 the measured O 4 are expressed as AMF. The measurements for 630.0 nm fall closest to the expected straight line. However, for the observations at 477.3 and 577.2 nm still significant deviations from this line appear. In addition, the values of the slope deviate from the expected value of 1 by ±17%. We attribute these discrepancies to two major causes: First, the aerosol conditions might have varied throughout the observations. Second, the aerosol properties are only poorly captured by the parameters (aerosol extinction and phase function) in our radiative transfer model. In general, these parameters depend, for example, on wavelength, which is not considered in our model calculation. Also for high clouds Mie scattering might not be well suited, because the cloud particles are not spherical. A part of the discrepancies of the modeled AMF and the measured O 4 absorptions might also be caused by differences of the actual and the assumed value (80%) for the ground albedo. From our modeling results it follows that reducing the ground albedo from 80% (for a snow surface) to about 5% (for a snow free surface) reduces the AMFs for all investigated wavelengths by about a factor of two (for SZA < 90 ). [39] The remaining differences between our measurements and modeling results might be resolved in future model calculations, which will use a more detailed aerosol parameter input. Further clear-sky O 4 observations, especially for low ground albedo during summer might also be helpful. Nevertheless, from our studies we already conclude that the comparison of measured and modeled clear-sky O 4 absorptions is a very sensitive tool to derive aerosol

AAC 3-12 WAGNER ET AL.: UV-VISIBLE OBSERVATIONS OF ATMOSPHERIC O 4 ABSORPTIONS Figure 7. (top) Comparison of the measured and modeled O 4 absorptions at 380.2, 477.3, 577.2 and 630.0 nm. To allow a direct comparison between the measurements (left) and model results (for a pure Rayleigh atmosphere, right), both results are expressed as AMF (differences to the Fraunhofer spectrum). For the conversion of the measured O 4 absorptions into AMFs we calculated the O 4 VCD* for that day from a radiosonde (see text). (bottom) Modeled AMF (-differences) for various atmospheric aerosol extinction profiles. The strength of the aerosol absorption (horizontal direction) as well as the altitude of the aerosol layer (vertical direction) are varied systematically. The scales are similar to those of the upper panel. For all conditions and SZA the O 4 AMFs decrease systematically with wavelength. parameters from zenith sky ground-based measurements. It should be noted that this method can be expected to be even more sensitive for off-axis-viewing geometry (the viewing direction is slightly above the surface) or the novel multiaxis-viewing geometry (scattered light from several viewing directions is observed). For such observation geometries the tropospheric light paths and thus the O 4 absorptions are significantly larger. Our method can thus provide indepen-