Aerosol-radiation interaction in the cloudless atmosphere during LACE Measured and calculated broadband solar and spectral surface insolations

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 107, NO. D21, 8124, doi: /2000jd000226, 2002 Aerosol-radiation interaction in the cloudless atmosphere during LACE Measured and calculated broadband solar and spectral surface insolations Manfred Wendisch, 1 Andreas Keil, 1,2 Dörthe Müller, 1 Ulla Wandinger, 1 Peter Wendling, 3 Armin Stifter, 3 Andreas Petzold, 3 Markus Fiebig, 3 Matthias Wiegner, 4 Volker Freudenthaler, 4 Wolfgang Armbruster, 5,6 Wolfgang von Hoyningen-Huene, 7 and Ulrich Leiterer 8 Received 6 December 2000; revised 29 October 2001; accepted 1 November 2001; published 13 September [1] Vertical profile measurements of aerosol particle size distributions and of meteorological parameters (obtained from aircraft, radiosondes, and lidar) are used as input to a spectral radiative transfer model to calculate broadband solar and spectral surface insolations. The calculated values are compared to measured ones gathered with broadband solar pyranometers and pyrheliometers, and a fixed-grating photodiode array spectroradiometer with 512 spectral channels between 500 and 920 nm wavelength. The measurements were obtained during the joint field campaign Lindenberg Aerosol Characterization Experiment (LACE) 98 near Berlin/Germany in the summer of Two cases (days with high and low aerosol loading, respectively) are investigated in detail. Furthermore, a measurement-based sensitivity analysis was carried out focusing on the influence of particle composition (complex refractive index) and of microphysical and humidity growth uncertainties on the calculated surface insolations. Assuming a spectral refractive index of ammonium sulfate for the aerosol particles, on average the global component of the broadband solar surface insolations is W m 2 (2 3%) greater than the measured values; the direct portion is W m 2 (4 5%) higher, and its diffuse component is 6 7 W m 2 (4 10%) lower in comparison to the measurements. The measured and calculated spectral surface insolations (global portion) agree well in the central visible spectral region ( nm wavelength). Toward larger wavelengths (near infrared) the calculated spectral surface insolations are increasingly higher than the measured ones. INDEX TERMS: 0305 Atmospheric Composition and Structure: Aerosols and particles (0345, 4801); 0345 Atmospheric Composition and Structure: Pollution urban and regional (0305); 0365 Atmospheric Composition and Structure: Troposphere composition and chemistry; 0368 Atmospheric Composition and Structure: Troposphere constituent transport and chemistry Citation: Wendisch, M., A. Keil, D. Mueller, U. Wandinger, P. Wendling, A. Stifter, A. Petzold, M. Fiebig, M. Wiegner, V. Freudenthaler, W. Armbruster, W. von Hoyningen-Huene, and U. Leiterer, Aerosol-radiation interaction in the cloudless atmosphere during LACE 98, 1. Measured and calculated broadband solar and spectral surface insolations, J. Geophys. Res., 107(D21), 8124, doi: /2000jd000226, Institute for Tropospheric Research, Leipzig, Germany. 2 Now at Met Office, Farnborough, UK. 3 Institute for Physics of the Atmosphere, Oberpfaffenhofen, Germany. 4 Meteorological Institute, University of Munich, Munich, Germany. 5 Institute for Space Research, Free University of Berlin, Berlin, Germany. 6 Now at Department of Optics of the Atmosphere and Meteorology. Research Society for Applied Natural Sciences, Ettlingen, Germany. 7 Institute for Remote Sensing, University of Bremen, Bremen, Germany. 8 Meteorological Observatory, German Weather Service, Lindenberg, Germany. Copyright 2002 by the American Geophysical Union /02/2000JD Introduction [2] Enhanced atmospheric absorption of broadband solar radiation, both in the cloudy and the cloudless atmosphere, is discussed in a controversial way in the literature [e.g., Wiscombe, 1995]. (In our notation the term solar covers the wavelength range nm. The term broadband denotes spectrally integrated radiation.) The problem, alsocalled excess or anomalous absorption, consists of a possible systematic underestimation of atmospheric absorption of radiation in the solar spectral region by radiative transfer models (compared to measurements) of about 20 to 40 W m 2, which introduces serious effects into climate modeling results. Thus the radiative transfer models tend to systematically overestimate atmospheric transmission and LAC 6-1

2 LAC 6-2 WENDISCH ET AL.: MEASURED AND CALCULATED SURFACE INSOLATIONS broadband solar downwelling irradiances at the surface compared to measurements. (The term irradiance (W m 2 ) is defined as the radiative energy (J = Ws) incident per second upon a horizontal unit area.) In both cloudy and cloudless skies there are investigations in literature showing either agreement or disagreement between measurements and modeling. This confusion is not surprising as entirely different types of radiative transfer models are used, ranging from simple d-two stream techniques typically used in current GCM s (Global Circulation Models) and detailed spectral approaches. Furthermore, entirely different kinds of measurements (ground-based, aircraft, satellite) are compared with the calculated radiation. Thus also the different experimental approaches to measure atmospheric absorption are subject of critical discussion. [3] The output of the large variety of the used radiative transfer models considerably varies, especially due to a different treatment of atmospheric gas absorption. Preliminary results of the ICRCCM-III (InterComparison of Radiation Codes in Climate Models, Phase III) show that the majority of one-dimensional (1D) codes in use today systematically underestimate atmospheric absorption of solar radiation compared to highly credible benchmark calculations (H. Barker, private communication, 2001). It looks like that most of this bias rests with lack of water vapor continuum absorption in the models. For solar zenith angles q s with cos q s 0.2 and in the fundamental case that only molecular scattering and absorption takes place (i.e., no aerosol particles and cloud droplets are considered) the range of clear-sky surface absorptances produced by the twenty-five participating 1D codes varies between 4 9% of the incoming TOA (top of atmosphere) irradiance. Corresponding RMS (root mean square) differences relative to the median are between about 2 4%. As an example, at cos q s = 0.5 the range of the results from the participating models for the calculated clear-sky absorbed irradiances at the surface is about 40 W m 2 and the RMS is about 13 W m 2. For the clear-sky atmospheric absorptance, differences among the 25 participating 1D codes are even larger with a range of the participating model results of 14 16% and an RMS relative to median of 5 6% for cos q s 0.2 (H. Barker, private communication, 2001). [4] On the other hand, the radiation measurements themselves are affected with inherent uncertainties, which translate into the resulting absorbed solar irradiances. Solar absorption measurements, based on collocated aircraft and/or satellite data, are particularly disputed due to severe uncertainties in the way how net irradiance are determined from current instrumentation, from which the layer absorptance is derived. (Net irradiance is defined as the difference between down- and upwelling irradiance.) For example, airborne irradiance measurements are distorted by absolute calibration uncertainties, deviations from the ideal cosine angular response of the instruments and problems related to horizontal sensor leveling (balancing of pitch and roll movements of the aircraft during flight), which cannot adequately be eliminated by postflight software correction procedures. Wendisch et al. [2000] show that a deviation of only 0.2 of the radiation sensor from the horizontal reference plane at q s =60 causes an uncertainty of the measured downward irradiance of about 1%, which transfers into a 4% deviation of the absorbed irradiance. If a more realistic 1 horizontal misalignment is assumed, then the irradiances deviate by 3% and the absorbed irradiances by 22% at q s =60. This strong sensitivity of the absorbed irradiance from horizontal misalignments is due to the fact that this quantity is derived from the residual of net irradiances, i.e., a small difference of large quantities. Thus only small measurement errors in the downwelling irradiances give rise to serious distortions in the derived absorbed irradiances and the layer absorptance. It seems to be primarily advisable to improve the accuracy of the solar irradiance measurements before looking at the derived absorption properties. The measurement-model comparison should focus on irradiances and should first be attempted to less complicated cases, i.e., clear sky scenarios and using ground-based radiation measurements under well-described conditions. The model input should be based on reliable measurements of the crucial input parameters, which need to be figured out by respective sensitivity tests. [5] However, even in the easy-looking cloudless cases and using ground-based measurements, there is no general consensus in the atmospheric radiation community about how well measured and calculated broadband solar surface insolations or broadband solar absorption (atmospheric or surface) agree. (The terms surface insolation and downwelling irradiance at the ground are equivalent. In the subsequent text the shorter term surface insolation is used. Additionally it is differentiated between the diffuse, direct, and global (= diffuse + direct) portions of surface insolation.) Discrepancies beyond model and measurement uncertainties under clear sky conditions are reported by Arking [1996], Charlock and Alberta [1996], Kato et al. [1997], Halthore et al. [1998], Kinne et al. [1998], Pilewskie et al. [1998], Arking [1999a, 1999b], Wild [1999], Halthore and Schwartz [2000]. Good agreement within measurement and model uncertainties are found by Chou and Zhao [1997], Conant et al. [1997, 1998], de La Casinière et al. [1997], Zender et al. [1997], Jing and Cess [1998], Valero and Bush [1999], Valero et al. [2000]. This list of publications is limited to results obtained in cloudless skies and does not claim to be complete. It reflects the large variety of partly contradicting conclusions reported in literature. In the following two subsections a short survey of these references is given. It is distinguished between model-measurement comparisons of broadband solar and spectral surface insolations (section 1.1), and of broadband solar atmospheric absorption (section 1.2) Broadband Solar and Spectral Surface Insolations [6] Charlock and Alberta [1996] publish significant overestimation of measured broadband solar insolations by respective calculations in both clear-sky and cloudy conditions. The calculations used satellite data for cloud optical thickness and ozone loading, and also for identifying intervals as cloudless or cloudy. [7] Similarly, Kato et al. [1997] show that their results of a d-two stream radiative transfer model lie about 5% (34 W m 2 ) above ground-based broadband solar pyranometer measurements. The discrepancies could not be explained by uncertainties in the model assumptions and are mainly related to the diffuse portion of the irradiances. Since in-situ measurements of aerosol properties were missing, the authors apply inversion algorithms based on Sun photo-

3 WENDISCH ET AL.: MEASURED AND CALCULATED SURFACE INSOLATIONS LAC 6-3 meter data to estimate the particle size distribution in the boundary layer (fixed thickness of 1 km). Only the unrealistic assumption that the particles consist of pure soot yields a good agreement between measurements and calculations. Missing atmospheric gas absorption components, not included in the d-two stream model, are postulated as a possible source for the discrepancy. [8] In contrast, Chou and Zhao [1997] and de La Casinière et al. [1997] find that clear-sky broadband solar surface insolations can be reliably computed from their radiation models. [9] Kinne et al. [1998] find that their simulations exceed respective measurements by 10% (up to 50 W m 2 ), mainly in the diffuse portion. Poor cosine angular response of the instruments and uncertain aerosol absorption assumptions have possibly contributed to the disagreement. Furthermore, a model sensitivity study is carried out with regard to doubling water vapor and ozone concentrations (small effect), spectrally varying surface albedo (small effect), adding subvisible cirrus and aerosol particles (decrease of direct insolation, strong increase in diffuse, but nearly no effect on global insolation), and some other parameters. Including 2% black carbon in the assumed sulfate aerosol particles exerts a small effect of less than 1 W m 2 for the diffuse component of the surface insolations. [10] Halthore et al. [1998] show that their models overestimate diffuse pyranometer data by 9 40%, whereas the direct component is well represented by the calculations which is more or less in agreement with Kato et al. [1997] and Kinne et al. [1998]. Halthore and Schwartz [2000] could remove part of the discrepancy by correcting the pyranometer measurements for thermal offset problems, inherent in mainly Eppley type of pyranometers. The calculations are scaled by Sun photometer measurements, therefore the agreement for the direct component is not surprising. The differences are resolved by decreasing particle optical thickness and artificially increasing continuum-like atmospheric absorption by about 5%. The authors do no find a clear correlation of the discrepancy with air mass character or particle optical thickness. [11] Pilewskie et al. [1998] show that measured spectral insolations are overestimated by respective calculations in the nm wavelength range. Again, the model aerosol input was constrained by matching to measured particle optical thickness. Particle absorption was neglected in these calculations. [12] In contrast, Valero and Bush [1999] find good agreement (within the range given by measurement and model uncertainties) between simulated and measured insolations (broadband and fractional solar, to some extent spectral) assuming a mineral aerosol type in the calculations. The correspondence is further improved by introducing absorbing particles in the calculations. Nevertheless, the calculated global insolations tend to be slightly larger compared to measurements, the diffuse data are overpredicted, and the direct portion is slightly underpredicted by the model. The conclusions drawn from the spectral comparison remain incomplete. [13] Harrison et al. [1999] discuss differences between measured and calculated diffuse spectral surface insolations obtained with their RSS (Rotating Shadowband Spectroradiometer) under low aerosol loading conditions. Mlawer et al. [2000] also compare RSS data with calculations, whereby the direct beam calculations are scaled to the particle optical thickness and no in-situ aerosol measurements are used Broadband Solar Atmospheric Absorption [14] Arking [1996] suggests, that the cloudless atmosphere absorbs significantly more solar radiation than predicted by models. This finding is not confirmed by Zender et al. [1997], who conclude that the enhanced absorption of solar radiation rather takes place in cloudy skies than under clear-sky conditions. Conant et al. [1997] support this conclusion. They show that their radiation models agree to within 5 W m 2 with the observed daily average clearsky solar absorption, if the atmospheric model input is constrained by profile measurements, and TOA albedo is used as boundary condition. Conant et al. [1998] conclude that excess absorption is less than experimental uncertainty. The difference is essentially independent of water vapor amount, i.e., there are no significant H 2 O vapor related biases in the clear-sky model. This statement diverges from the conclusion by Arking [1999a, 1999b], who reports 15 to 30 W m 2 enhanced clear-sky solar absorption. The observations show a much stronger dependence of absorption on column water vapor than the models do. The author concludes that this is an indication of possible problems with the water vapor continuum absorption in the models. [15] Jing and Cess [1998] find no evidence for enhanced absorption in clear skies. In their study two different radiative transfer models are used in combination with satellite and ground-based observations. [16] Wild [1999] shows significant underestimation of solar atmospheric absorption in three GCM s in areas and seasons with large aerosol burden, caused by extensive vegetation fires, i.e., absorbing particles. The author cannot exclude the possibility that part of the lack of absorption in the GCM s is due to cloud problems. However, Wild [1999] shows that the major contribution is in cloudless skies. Deficiencies of clear-sky radiative transfer models and underestimation of water vapor amount by GCM s are discussed as additional source for the discrepancies. Possibly, there is also some error cancellation between insufficient aerosol absorption and enhanced cloud absorption. On the other hand, the paper by Valero et al. [2000] reports no evidence for enhanced clear-sky absorption beyond model and experimental errors. Instead, significant enhanced absorption in cloudy skies is reported by these authors Our Approach [17] In our approach in-situ measured, vertically resolved meteorological and, in particular, microphysical (including humidity growth) and physicochemical (chemical composition) aerosol particle data are used in order to calculate broadband solar and spectral surface insolations with a sophisticated, spectral radiative transfer model. This is different to former studies, in which the radiative transfer calculations are matched with simultaneously gathered Sun photometer data of particle optical thickness at one or some specific wavelengths. Our approach is based on predominantly using measured data as model input in order to avoid unjustified model assumptions and simplification. Furthermore, in order to evaluate the differences between modeled and measured surface insolations, the influence of inherent

4 LAC 6-4 WENDISCH ET AL.: MEASURED AND CALCULATED SURFACE INSOLATIONS model input uncertainties on the model output is comprehensively investigated in this study. [18] In literature, mostly measured and calculated broadband solar radiative quantities are compared. If spectral data are used, then often with coarse resolution, distinguishing between visible (VIS) and near infrared (NIR), or limited to some specific wavelengths. There are only a few recent publications which present surface insolation measurements with higher spectral resolution. [19] In this context, this paper aims at closing two gaps in the present discussion by (1) using detailed airborne in-situ profile measurements of microphysical particle properties (mainly particle size distributions) as well as meteorological parameters in spectral radiative transfer calculations as model input, and (2) comparing the model results with respective measurements of both broadband solar and spectral surface insolations. Furthermore, the model is used to carry out a comprehensive sensitivity analysis of the model output with special emphasis on the influence of aerosol physicochemical (particle composition, i.e., complex refractive index), microphysical and humidity growth properties. This, on the other hand, allows to evaluate the significance of the observed differences between measured and calculated surface insolations and to figure out the crucial input parameters. Using the aerosol microphysical data avoids the scaling of the model input to the particle optical thickness as often done in previous studies. Thus our approach has the advantage that the model is not indirectly forced to correctly predict the direct component of surface insolations. Furthermore, consistency between the three major input quantities with regard to the particles (scattering/absorption coefficient, single scattering albedo, phase function) for the radiative transfer model is guaranteed by this method. The disadvantage of our approach is, that the needed airborne measurements (mainly the vertical profiles of particle size distributions, and their chemical composition) are much more complicated to obtain and more susceptible to measurement errors and sampling artifacts (e.g., not ideal vertical aircraft patterns). Column aerosol particle optical thickness is much easier to measure. Furthermore, validity of Mie theory (including the spherical particle shape assumption) is assumed to derive the radiative transfer model input. [20] This paper is part 1 of two contributions dealing with aerosol-radiation interaction in the cloudless sky during the field campaign LACE 98 (Lindenberg Aerosol Characterization Experiment 1998) carried out nearby Lindenberg (Germany) in summer Part 1 concentrates on surface radiation measurement-model comparison, whereas part 2 [Wendling et al., 2002] deals with the radiative effects of the particles (broadband solar radiative forcing) at different altitudes within the atmosphere. 2. Experimental and Instrumentation [21] Within the framework of a German aerosol re-search focus the joint field experiment LACE 98 was conducted during July and August An overview of LACE 98 is given by Ansmann et al. [2002]. The experiment took place at the Meteorological Observatory of DWD (Deutscher Wetterdienst, in English: German Weather Service) in Lindenberg (52.2 N, 14.1 E), which is situated about 50 km southeast of the eastern boundary of Berlin, and at Falkenberg (4 km south of Lindenberg). Three aircraft, frequent radiosonde profiling and an assemblage of tower and ground-based remote sensing (including lidar, light detection and ranging) and in-situ techniques were employed in order to characterize the microphysical and physicochemical aerosol properties and their radiative effects under summer-time, continental conditions Aircraft Measurements [22] From the three aircraft participating in LACE 98, data obtained by two aircraft (Partenavia and Falcon) are used in this paper. The Partenavia P68B D-GERY (owned and instrumented by enviscope GmbH, chartered by IfT, Institut für Troposphärenforschung) probed the lower troposphere up to a maximum altitude of about 4 km. The second aircraft, a Mystere Falcon-20 D-CMET (owned and operated by DLR, Deutsches Zentrum für Luft- und Raumfahrt), focused on measurements in higher altitudes up to the tropopause region. The complete scientific equipment of both aircraft is summarized by Ansmann et al. [2002]. Beside general meteorological devices for measuring static air pressure ( p), static air temperature (T), and relative humidity (RH), and broadband radiation sensors (Eppley pyranometers), several specific aerosol instruments were installed on board both aircraft. The following description is restricted to aerosol devices which delivered data for this study. [23] In the cabin of the Partenavia a PCASP-X (Passive Cavity Aerosol Spectrometer Probe, manufactured by PMS, Particle Measuring Systems, Boulder, CO, USA) for dry particle size distribution measurements (30 size channels between 0.1 and 10 mm diameter) and a three-wavelength nephelometer (manufactured by TSI, Inc., St. Paul, MN, USA) to determine the spectral particle volume scattering and backscattering coefficients (at the wavelengths l = 450; 550; 700 nm) are installed. Both instruments are supplied by an isokinetic particle inlet [Maser et al., 1994]. During their way to the sensors the particles are heated and dried. Generally, a relative humidity below 30% was measured just upstream of the sample volume of the aerosol instruments. Therefore the aerosol properties behind the inlet are measured in their dry state. The calibration and data extraction procedures for the PCASP-X are described by Keil et al. [2001]. From the measured particle size distributions integrated parameters like particle concentration or particle effective radius are derived. The particle volume scattering coefficient is calculated via Mie theory and compared with nephelometer data. The calibration of the nephelometer was done shortly before and after the flights with a commercial gas of known volume scattering coefficient (CO 2 ) and with filtered particle free air. The measurement accuracy of the nephelometer has been estimated to be ±7% by Anderson and Ogren [1998]. [24] The Falcon aircraft carried under their wings another optical particle counter of the type PCASP-100X (15 size channels between 0.1 and 3 mm particle diameter) and an FSSP-300 (Forward Scattering Spectrometer Probe, also manufactured by PMS). From these measurements composite size distributions are obtained covering the three typical aerosol modes (Aitken, accumulation, and coarse particles). Calibration issues, data analysis procedures and measure-

5 WENDISCH ET AL.: MEASURED AND CALCULATED SURFACE INSOLATIONS LAC 6-5 ment uncertainties as well as the strategy to deduce the composite particle size distributions are discussed in detail by Fiebig et al. [2002]. Further results, obtained with these data during LACE 98, are reported by Petzold et al. [2002] and Wendling et al. [2002] Ground-Based Data Sun Photometer [25] Data from a multiwavelength Sun photometer of the type BAS (Boden-Atmosphären-Spektrometer, in English: Surface-Atmosphere-Spectrometer) developed by DWD are analyzed. The instrument itself, as well as calibration and quality assurance issues are described by Leiterer et al. [1994]. The Sun photometer yields the column-integrated, spectral optical thickness of aerosol particles by measuring the extinction of the direct solar beam. The instrument was placed at Lindenberg. Altogether the Sun photometer comprises 69 spectral channels, from which six are selected for our analysis (l = 451.2; 492.2; 551.0; 653.8; 791.5; nm). The angle of view of the Sun photometer is about 1. The used interference filters are characterized by values of FWHM (Full Width at Half Maximum) of 13 nm below 700 nm wavelength and of 20 nm in the NIR. Climate chamber tests were carried out with a constant-output light source in the temperature range between 10 C and +36 C. These measurements have delivered temperature coefficients for the filters for each channel, which have been considered in the data analysis. The Sun photometer is regularly calibrated at mountain sites (e.g., Zugspitze, Germany) using the Langley plot technique. Absolute accuracy of the measured particle optical thickness is estimated to be ± Pyranometer and Pyrheliometer [26] In order to measure global (F glo ) and diffuse (F dif ) broadband solar surface insolations, unshaded (for F glo ) and shaded (for F dif ) pyranometers (manufactured by Kipp & Zonen, Delft, Netherlands) were used in parallel. The CM 21 version of the Kipp & Zonen pyranometers was running at the DWD measurement site Lindenberg. At Falkenberg pyranometers of the type CM 14 were installed. Here only CM 21 pyranometer data from Lindenberg are reported. Independently, a solar-tracking pyrheliometer of the type NIP (Normal Incidence Pyrheliometer, manufactured by The Eppley Laboratory, Inc., Newport, RI, USA) was used to measure the direct broadband solar surface insolations (F dir ) at Lindenberg. F dir was also estimated from the difference between shaded and unshaded pyranometer measurements. [27] Lindenberg is part of the BSRN (Baseline Surface Radiation Network). The BSRN operations manual (WMO/ TD-No. 879, 1998) calls for the use of a pyrheliometer to measure F dir and to sum up F dif measured by a shaded pyranometer and F dir supplied by a pyrheliometer to derive the global component of broadband solar surface insolations F glo. Here this suggestion is followed. Additionally the results of this method were compared with unshaded pyranometer measurements of F glo. [28] The pyranometers are regularly calibrated and maintained at DWD. Also comparisons with other pyranometers and with several pyrheliometers are done. The Kipp & Zonen pyranometers are fully compliant with all IOS (International Organization for Standardization) 9060 Secondary Standard Instrument performance criteria. Laboratory measurements have shown, that uncertainties caused by deviations from the ideal cosine angular response are negligible for the used pyranometers. Therefore, no corrections of this effect have been made. The wavelength range covered by the pyranometers lies between nm (50% points), however the calibration includes the complete solar spectral range ( nm). The temperature dependence of sensitivity of the CM 21 is given by the manufacturer to be ±1% between 20 C and +50 C, nonlinearity is ±0.2% for insolations less than 1000 W m 2. The overall measurement accuracy for the global pyranometer data is estimated to be better than ±2%. [29] The pyranometer measurements have been corrected for re-radiative thermal offsets, which have been shown to be responsible for part of measurement-calculation gap by Halthore and Schwartz [2000]. Instead of using the comprehensive correction procedure reported by Bush et al. [2000], a first-order correction has been performed, which was proven to be sufficiently accurate. The thermal offsets of the diffuse and direct surface insolations are determined in the night before the respective day between 0000 and 0300 UTC and then subtracted from the measurements during the day. Usually the thermal offset of the Kipp & Zonen pyranometer is much less than that of Eppley pyranometers. Values less then ±2 W m 2 have been observed for our Kipp and Zonen instruments [Behrens and Dehne, 2000]. The pyranometer have also been ventilated which further reduces thermal offset. [30] The NIP is a WMO (World Meteorological Organization) First Class Pyrheliometer designed for the measurement of solar radiation at normal incidence. By multiplying these measurements with cos q s, the direct component of the broadband solar surface insolation F dir is derived. The NIP is traceable to the WRR (World Radiation Reference). The temperature dependence of sensitivity is given by Eppley to be ±1% between 20 C and +40 C, nonlinearity is less than ±0.5% from 0 to 1400 W m 2. The overall measurement accuracy for the NIP data is estimated to be better than ±1%. [31] The NIP is internally calibrated each half hour by adjusting the NIP output to a self-calibrating Absolute Cavity Pyrheliometer, Model AHF (Automated Hicky Frieden), which is also manufactured by Eppley. The AHF is an absolute standard radiometer for direct broadband solar insolation measurements with an accuracy of ±0.5% Spectroradiometer [32] Our spectral surface insolation measurements have been performed at Falkenberg. The used spectroradiometer has been built by Meteorologie Consult GmbH (Glashütten, Germany). It consists of an MCS (Multi Channel Spectrometer) module with a fixed grating used as dispersion element and a diode array as detector. The MCS has been manufactured by Zeiss GmbH (Jena, Germany). It covers the wavelength range nm. The spectral pixel distance is 0.8 nm with a resolution of 2.4 nm (Rayleigh criterion). The absolute wavelength accuracy is given by the manufacturer to be <0.3 nm with a temperature drift of <0.005 nm K 1, the stray light is quantified to be 0.1%. The MCS module is connected via fiber optics with a cosine diffuser as optical inlet (designed and manufactured by Meteorologie Consult GmbH, Glashütten, Germany). An airborne version of the instrument with extended wavelength range (290

6 LAC 6-6 WENDISCH ET AL.: MEASURED AND CALCULATED SURFACE INSOLATIONS 1000 nm) for measuring of up- and downwelling spectral irradiances has been developed recently [Wendisch et al., 2000]. The ground-based instruments used by Harrison et al. [1999], Meywerk and Ramanathan [1999], and Mlawer et al. [2000] are similar to our spectroradiometer, they cover the wavelength range nm. [33] Absolute calibrations of our spectroradiometer have been done jointly with DWD in the laboratory using a 1000 W quartz halogen lamp (PTB-SL 120) that is traceable to an absolute standard, maintained at PTB (Physikalisch-Technische Bundesanstalt, Braunschweig, Germany), the German Metrology Institute. The lamp illuminated the sensor (including the diffuser) perpendicular to the horizontal sensor reference plane. A measurement uncertainty for this calibration lamp of ±1.6% ( nm wavelength) and ±3% ( nm) is quantified by PTB. The stability of the calibration has been verified before and after LACE 98 using secondary standard halogen lamps, which are connected to the PTB lamp. The wavelength settings provided by the manufacturer have been checked using several low-pressure spectral light sources (Ne: and nm; Kr: and nm; Xe: nm; Ar: nm) and were confirmed by comparing with Fraunhofer lines in the TOA spectral insolations. Additionally, the spectroradiometer performance has been checked by outdoor comparisons in cloudless skies with data of a spectroradiometer of the type OL 754/10 [Feister and Grewe, 1995] supplied by DWD yielding agreement within the estimated measurement errors. The DWD spectroradiometer is calibrated with a tungsten halogen lamp (200 W), traceable to NIST (National Institute of Standards and Technology). Problems related with such type of lamp calibrations are discussed by Kiedron et al. [1999]. These authors found that even NIST standard lamps may disagree with each other beyond their stated accuracy. These uncertainties transfer into NIST traceable secondary standards. Nevertheless, the lamp calibrations in the laboratory have been used in this study, because it was not possible to carry out Langley plot calibrations similar to those of Harrison et al. [1999] and Mlawer et al. [2000], which require long term operation or calibration measurements under very stable optical atmospheric conditions only assured at high mountain stations. [34] Measurements of global (F l,glo,m ) as well as diffuse (F l,dif,m ) spectral surface insolations have been collected during the LACE 98 campaign. The diffuse portion was determined by shading the solar disc with a metal ring which is automatically turned into the solar beam from time to time. The difference between both components yields an estimate of the direct spectral surface insolations (F l,dir,m ). It should be noted that in this way the three components of the spectral surface insolations (global, direct, and diffuse) are not measured independently. Furthermore, the Sun photometer measured spectral atmospheric optical thickness (particle + gas extinction) could not perfectly be reproduced from the estimated direct spectral surface insolations. This problem is related to uncertainties in the direct-diffuse separation by the applied shadowing technique. Therefore, only global measurements of spectral surface insolations are discussed in this paper. The estimated direct and diffuse spectral surface insolations have only been used to correct the global measurements for nonideal cosine angular response of the diffuser. This has been done in the following way. [35] First, spectral functions of the direct cosine correction factors c l (q) (with q the zenith angle and l the wavelength of direct incident light, the azimuth angle of the incident light was fixed) of the cosine diffuser, have been determined for selected wavelengths l in the laboratory. c l (q) is measured via the direct cosine error function f l (q) = 1/c l (q) 1 which is defined as: f l ðþ¼ q E l ðþ q E l ðq ¼ 0 Þcos q E l is the incident spectral irradiance (only direct component) from the laboratory calibration lamp. The laboratory measurements were repeated for several azimuth angles. The differences between the resulting cosine error functions for the different azimuth angles were small (2%). [36] Second, spectral functions of the diffuse cosine correction factors k l are obtained from averaging of c l (q), whereby in practice the integration is performed until q = 85 only: Z 90 k l ¼ c l ðþ¼2 q c l ðq 0 Þsin q 0 cos q 0 dq 0 0 Finally, using the direct and diffuse cosine correction factors, the corrected global spectral surface insolations are obtained as follows: F l;glo ¼ F l;dir;m c l ðq s ÞþF l;dif ;m k l This procedure to correct for nonideal cosine characteristics of the optical diffuser involves the assumption that the diffuse part of irradiance is isotropic Lidar [37] During LACE 98 the vertical and horizontal structure as well as temporal changes of the aerosol particle distribution was also observed by several lidars [Wandinger et al., 2002]. While the airborne lidar of the DLR, flying at the Falcon, was available for special intensive observation phases only, several ground-based lidars were continuously measuring throughout the whole LACE 98 period. One of which was the mobile lidar of the Meteorological Institute of the University of Munich (MIM). It is a backscatter lidar emitting at the three wavelengths 1064, 532, and 355 nm. The atmosphere was observed under different zenith angles, in particular, horizontal measurements were made to determine particle extinction coefficients very close to the surface. 3. Measurement Cases [38] Altogether, twelve Partenavia and ten Falcon coordinated flights were conducted during LACE 98. The flights made on 1 and 10 August 1998 are selected for a detailed discussion. On these two days different meteorological and aerosol features with comparably high aerosol loading on ð1þ ð2þ ð3þ

7 WENDISCH ET AL.: MEASURED AND CALCULATED SURFACE INSOLATIONS LAC August (daily average of particle optical thickness at l = 551 nm: d 551 = 0.25; air mass from the Northwest to West) and low aerosol burden with an elevated aerosol layer on 10 August (d 551 = 0.09; air mass from the North) were observed [Ansmann et al., 2002]. Data of two other days (31 July and 11 August) were also investigated, but are not discussed here. The results were similar compared to the selected two days of 1 and 10 August Meteorology [39] The meteorological situations for the measurement cases 1 August (morning) and 10 August (early afternoon) are displayed in Figures 1a and 1b. In both plots the vertical profiles of static air temperature (T ) and relative humidity (RH ) are presented, as measured by aircraft and radiosonde. In Figure 1a the effect of a distinct temperature inversion at about 3 km altitude is obvious in the RH profile. Below this inversion two layers are present: One between the inversion and about 0.9 km altitude with a more or less constant RH, and another one below 0.9 km with a vertical RH gradient, which indicates good mixture within this near-surface layer. Looking at the potential temperature profiles [Ansmann et al., 2002] confirms the existence of the two layers below the inversion. The radiosonde and aircraft data agree within their measurement uncertainties. This correspondence is a result of the close spatial and temporal collocation of aircraft and radiosonde and the low temporal variability of the atmosphere in the morning hours of 1 August, which is also confirmed by the lidar observations [Ansmann et al., 2002]. The situation on 10 August (Figure 1b) is different in the sense that the atmosphere is dryer and warmer compared to 1 August. The planetary boundary layer (PBL) with a top height of about 1.6 km is well mixed, which is indicated by a strong gradient of RH within the PBL. However, there is more temporal variability in the PBL height compared to 1 August. This is demonstrated by time-resolved backscatter profiles from the MIM (ground-based) and DLR (airborne) lidars (Figure 1c). In the course of the day a rapid change of both the height of the PBL and its structure occurred [Ansmann et al., 2002] on 10 August. While in the morning the top of the PBL was well pronounced around 0.8 km altitude, this sharp boundary disappeared and a new top was built up at 1.7 km in the early afternoon. The temporal variability on 10 August was sometimes very high as evident from the lidar measurements between 1133 and 1140 UTC where strong convection was present (not shown). This fact is confirmed by horizontal lidar measurements which show a lot of transient aerosol particle features around the measurement site. Despite the high temporal variability of the boundary layer on 10 August, there are only small differences between the aircraft and the radiosonde measurements in Figure 1b Particle Properties (Dry State) [40] The different features between both days are strongly reflected in the vertical profiles of the particle microphysical quantities. In Figures 2 and 3 vertical profile measurements of integrated (number concentration, spectral volume scattering coefficient, see section 3.2.1) and size-resolved (number size distribution, see section 3.2.2) dry particle microphysical properties are presented for the two days discussed Integrated Microphysical and Optical Parameters [41] In Figure 2a the particle number concentrations (integrated over the optically relevant size range covered by the PCASP-X) in their vertical distribution are plotted. On the more polluted 1 August the particle number concentration is mostly higher compared to the rather pristine 10 August. In the layer between the temperature inversion (about 3 km) and the height of 0.9 km the particle number concentration on 1 August is more or less continuously decreasing with increas-ing altitude, which is in accordance with the statement that this layer is not well mixed (see Figure 1a). Below 0.9 km (within the well mixed near-surface layer) particle number concentrations remain rather constant. The observations on 10 August are completely different. Below 1.6 km within the PBL the particle number concentration remains low and constant, above it drops off sharply. This is in accordance with the fact that the PBL was well mixed on 10 August (Figure 1b). In about 3 km altitude a 200 m thick layer of enhanced particle number concentrations with maximum values of 500 cm 3 is observed (indicated by an arrow). The spatial and temporal evolution of this interesting elevated aerosol layer is investigated in detail by Ansmann et al. [2002], Fiebig et al. [2002], and Wandinger et al. [2002]. [42] From the measured (dry) particle size distribution, the (dry) spectral volume scattering coefficient of the particles is calculated via Mie theory and compared to the respective nephelometer measurements for both days. The results for the wavelength of 550 nm are plotted in Figure 2b. The assumed refractive indices for the Mie calculations for 550 nm (1 August: i; 10 August: i) are determined by Wex et al. [2002] from ground-based data during LACE 98. These complex refractive indices are also in agreement with measurements by Bundke et al. [2002] and Ebert et al. [2002] carried out during LACE 98. The real part of the chosen refractive index is close to that of ammonium sulfate. Neusüß et al. [2002] have shown by means of chemical analysis of size-segregated impactor samples (major ions) that ammonium sulfate is the very dominant soluble particle species during these two LACE 98 days. In Figure 2b the maximum deviations between measured and calculated spectral particle volume scattering coefficients is in the order of ±20%. Mostly the data agree within the measurement error bars of the nephelometer (which is ±7%). A more detailed analysis of this type of closure experiment from the LACE 98 aircraft and ground-based data (inclusive a sensitivity analysis) is given by Wex et al. [2002]. The different vertical features of 1 and 10 August are similar to those discussed for the particle number concentrations in Figure 2a. From the maximum value of the spectral (dry) volume scattering coefficient (at 550 nm) of the particles in the elevated aerosol layer on 10 August a lower limit of the particle optical thickness at 550 nm of 0.02 can be estimated which is in agreement with Ansmann et al. [2002] and Wandinger et al. [2002] Particle Size Distribution [43] For the radiative transfer calculations particle size distribution data are required as a major input. The Falcon measurements have been used to derive composite particle number size distributions from the PCASP-100X and FSSP- 300 data [Fiebig et al., 2002]. The original data have been

8 LAC 6-8 WENDISCH ET AL.: MEASURED AND CALCULATED SURFACE INSOLATIONS Figure 1. (a) Vertical aircraft (Partenavia, one second averages) and radiosonde profile measurements of static temperature (T, lower axis) and relative humidity (RH, upper axis) on 1 August The aircraft data were taken during a vertical descent between 0800 UTC and 0824 UTC, the radiosonde was launched 0600 UTC and reached the altitude of 5 km around 0617 UTC. (b) The same as Figure 1a, but for 10 August The aircraft data were taken during a vertical descent between 1350 UTC and 1415 UTC, the radiosonde was launched 1327 UTC and reached the altitude of 5 km around 1339 UTC. (c) Vertical profiles of lidar backscatter coefficients at 532 nm wavelength derived from the airborne (DLR) and ground-based (MIM) lidars on 10 August. averaged for certain altitude ranges and then fitted to a threemode lognormal type of function of the following form: dn dlogd p ¼ X3 i¼1 " 2 # N i logd p logd N;i pffiffiffiffiffi exp 2p log si 2 log 2 s i ð4þ with dn dlogd p the logarithmic particle number size distribution; D p the particle diameter; N i the particle number concentration in mode i; s i the geometric standard deviation of mode i; and D N,i the geometric mean particle diameter of mode i. The three modes correspond to the Aitken (i = 1), accumulation (i = 2), and coarse (i = 3) particle size modes, respectively.

9 WENDISCH ET AL.: MEASURED AND CALCULATED SURFACE INSOLATIONS LAC 6-9 Figure 2. (a) Vertical aircraft profile measurements (one second resolution) of total particle number concentration measured by PCASP-X (carried by Partenavia aircraft) on 1 (thick dashed line) and 10 (thin solid line) August. The same aircraft descends as in Figures 1a and 1b are analyzed. The arrow indicates an elevated aerosol layer in about 3 km altitude present on 10 August. (b) Vertical aircraft profile data (averaged over 20 m height intervals) of spectral particle volume scattering coefficient (l = 550 nm) measured by the nephelometer (1 August: Thick dashed line; 10 August: Solid line) and calculated via Mie theory (1 August: Open triangles; 10 August: Open circles), on the basis of the size distributions measured by the PCASP-X. Both instruments were carried by Partenavia. The same aircraft descends as in Figures 1a and 1b are shown. Again the arrow indicates an elevated aerosol layer in about 3 km altitude present on 10 August. For the nephelometer data of 1 August the measurement uncertainty of ±7% is indicated by horizontal bars. Figure 3. (a) Particle size distributions derived from PCASP-100X and FSSP-300 data measured on 1 August for three selected altitude ranges ( UTC). Both instruments were carried by the Falcon aircraft. For the lowest altitude range, additionally, the averaged data points of the original measurements are included in the figure (open triangles: PCASP-100X, open circles: FSSP-300) together with vertical error bars representing standard deviation of the data. (b) The same as Figure 3a but for the 10 August ( UTC).

10 LAC 6-10 WENDISCH ET AL.: MEASURED AND CALCULATED SURFACE INSOLATIONS Table 1. Parameters of Three-Mode Lognormal Particle Number Size Distribution a Altitude, km Aitken Mode Accumulation Mode Coarse Mode N 1,cm 3 D N,1, mm s 1 N 2,cm 3 D N,2, mm s 2 N 3,cm 3 D N,3, mm s 3 1 August , , , August , , , a See equation 4. The data are retrieved from PCASP-100X and FSSP-300 measurements on board the Falcon aircraft on 1 August ( UTC) and on 10 August ( UTC). The results of the fitting procedure for the two days 1 and 10 August are summarized in Table 1. This table indicates all parameters needed to reconstruct the composite average particle size distributions for certain altitude ranges with equation 4. Some minor differences to the data in Figure 2a result from the different times at which the flights were conducted. In particular, the elevated aerosol layer observed on 10 August subsided during the course of the day from about 3.8 km (Falcon morning flight: UTC, Table 1) to about 3 km (Partenavia, early afternoon flight: UTC, arrows in Figures 2a and 2b). [44] In Figures 3a and 3b the fitted particle number size distributions for the two investigated days and for three selected altitude ranges are shown. In Figure 3a (1 August) a continuous decrease of the total particle number concentration of the size distribution especially in the Aitken and coarse particle modes with increasing altitude is obvious. A coarse particle mode is evident in all altitudes, which cannot be neglected in the radiative transfer calculations, as proven by respective sensitivity tests (not shown). This fact also holds for the data obtained on 10 August (Figure 3b). The coarse particle mode is evident in all altitude ranges. The generally lower pollution level on 10 August becomes most obvious in the suppressed accumulation mode, whereas there are more particles in the Aitken mode on 10 August compared to the 1 August data. The particle size distribution of the distinct aerosol layer in about 3.8 km altitude differs from the neighboring layers above and below by an enhanced number of particles of 1 mm diameter. 4. Radiative Transfer Simulations 4.1. General Features of the Model [45] The measured vertical profiles of meteorological and aerosol quantities (from aircraft and radiosondes) are used as input for spectral radiative transfer simulations in order to calculate broadband solar and spectral surface insolations for a comparison with the measurements. The one-dimensional approach is based on the matrix operator method. The radiative transfer equation for a vertically inhomogeneous atmosphere is solved by a combination of the Discrete Ordinate Method for each homogeneous sublayer and the Adding Method for linking the sublayers [Nakajima and Tanaka, 1986]. In our calculations the upper hemisphere is divided into eight Gaussian quadrature points (eight streams). The model is numerically accelerated (optional) by different delta approximations for the phase function and the asymmetry parameter [Nakajima and Tanaka, 1988]. Beside multiple scattering, the model accounts for absorption by seven major molecular gases (H 2 O; CO 2 ; O 3 ; N 2 O; CO; CH 4 ; O 2 ) and 21 further trace gas species. The gas absorption coefficients are mostly described with Lowtran 7 data, which can be downloaded (as part of Modtran) from The composite extraterrestrial spectral insolations at the TOA between wave number zero and cm 1 (l = 174 nm) with 4350 grid points and respective interpolation schemes are also adapted from Lowtran 7. The extraterrestrial spectral insolations are adjusted to the actual Earth-Sun distance (corresponding to the date of calculations) by factors from literature ( and for 1 and 10 August, respectively). The vertical resolution of the model is fixed to 20 m below 5 km altitude and 1 km from 5 to 12 km altitude. Additional altitudes are 20, 40 and 100 km. In total, 260 atmospheric layers are defined Model Input and Output [46] The measured vertical profiles of the meteorological parameters ( p, T, RH, and the derived H 2 O volume mixing ratio) as well as the O 3 volume mixing ratio (from radiosonde) are projected on the 260 layer grid. In those altitudes, where no meteorological or O 3 measurements are available, standard vertical profiles from McClatchey et al. [1971] ( Midlatitude Summer ) are used. For the uniformly mixed