Raman lidar observations of aged Siberian and Canadian forest fire smoke in the free troposphere over Germany in 2003: Microphysical particle characterization

Authors


Abstract

[1] Dual-wavelength Raman lidar observations were regularly carried out at Leipzig (51.3°N, 12.4°E) from May to August 2003. The measurements showed that particle backscatter and extinction coefficients in the free troposphere were higher compared to values in 2000–2002. Backward dispersion modeling indicates that intense forest fires that occurred in Siberia and Canada in spring/summer 2003 were the main cause of these free tropospheric haze layers. Measurements on 3 days were selected for an optical and microphysical particle characterization of these well-aged particle plumes. Particle lidar ratios measured at 532 nm wavelength were higher than at 355 nm. This property seems to be a characteristic feature of aged biomass-burning particles observed over central Germany. Mean particle Ångström exponents calculated for the wavelength range from 355 to 532 nm varied from 0 to 1.3. Particle effective radii varied between 0.24 and 0.41 μm. Pollution advected from North America on 25 August 2003, in contrast, was characterized by considerably smaller particles. Mean effective radii were ≤0.2 μm, and Ångström exponents were 1.8–2.1. Lidar ratios in that case were lower at 532 nm compared to those at 355 nm. Such signatures are characteristic for anthropogenic particles. At the moment, however, it cannot be completely ruled out that extremely hot forest fires in western areas of Canada generated comparably small particles. Except for this specific case the forest fire particles were considerably larger than what is usually reported from in situ observations of biomass-burning smoke. Possible explanations for this difference could be the kind of burning process, which could generate much larger particles in the source region, condensation of organic vapors on existing particles, and coagulation processes during the long transport time of more than a week. Relative humidity measured in these layers was very low. Hygroscopic growth of the particles therefore seemed to have little influence on the size of the particles. The forest fire smoke consisted of moderately absorbing material. Real parts of the complex refractive index of the particles were mostly <1.5, and imaginary parts were <0.01i. Single-scattering albedo in all cases varied between 0.9 and 0.98 at 532 nm.

1. Introduction

[2] Forest fires in Russia are believed to be a major source of pollution in the Northern Hemisphere each year [Wotawa et al., 2001; Kajii et al., 2002]. For that reason, there has been increased interest in recent years to assess the impact of these fires on climate [Shvidenko and Nilsson, 2000; Shvidenko and Goldammer, 2001; Conard et al., 2002; Kasischke and Bruhwiler, 2003]. However, studies that deal with biomass-burning aerosols from boreal areas in the Asian continent are scarce, but are an essential corner stone for an improved assessment of the effects of aerosols on radiative forcing [Intergovernmental Panel on Climate Change, 2001]. More information is available in the case of Canadian forest fires. Most studies dealt with in situ sampling [Chung, 1984; Pueschel et al., 1988; Mazurek et al., 1991; Miller and O'Neill, 1997; Radke et al., 1991]. Satellite observations mainly focused on tracing smoke plume evolution over urbanized areas of eastern North America [Chung and Le, 1984; Ferrare et al., 1990; Westphal and Toon, 1991; Hsu et al., 1999; Fromm et al., 2000; Li et al., 2001]. Sun photometer observations provide column-averaged information on smoke properties [Pueschel et al., 1988; Pueschel and Livingston, 1990; Holben et al., 2001; O'Neill et al., 2002].

[3] A general drawback of these studies is the lack of measurements and data analyses with special emphasis on a characterization of biomass-burning particles in lofted layers in the free troposphere on the basis of long-term observations. Further, the measurements were mainly confined to regions comparably close to the source regions, and thus provide little information on the optical and microphysical properties of such particles, if they are transported in the free troposphere over thousands of kilometers. In that respect it should be kept in mind, that the local effect of radiative forcing of climate by free tropospheric particles may be small on average, because of low optical depths of the order of 0.05 at visible wavelengths far away from the source regions. However, because of the spread of particles over large parts of the hemisphere the direct radiative impact is nonnegligible. Furthermore, there may be a sensitive impact on water and ice cloud formation in the free troposphere, which generally is rather clean. Forest fire smoke contains a considerable amount of soot and mineral components that are known to be efficient ice nuclei [DeMott et al., 1999, 2003].

[4] Up to date we know of only one study that investigated in detail the optical, microphysical, and radiative properties of aged Canadian forest fire smoke, which was transported from western Canada to Europe in 1998 [Fiebig et al., 2002; Wandinger et al., 2002; Fiebig et al., 2003]. A large-scale event of free tropospheric transport of biomass-burning particles that occurred in 2003 provided an excellent opportunity to gain new insight on the properties of such lofted haze layers. Mattis et al. [2003] reported on strongly enhanced values of particle backscattering and extinction in the free troposphere over Germany in spring/summer of that year. The study was based on two-wavelength Raman lidar observations carried out in the framework of the European Aerosol Research Lidar Network (EARLINET) [Bösenberg et al., 2003].

[5] Intense forest fires in eastern Siberia most likely were the origin of the particle layers observed over Leipzig in May 2003 [Damoah et al., 2004]. Measurements from the Terra satellite indicated strongly enhanced concentrations of carbon monoxide and aerosol optical depth in the Northern Hemisphere as the result of intense fires in the southeast of Russia in the spring and summer of 2003 [Edwards et al., 2004]. The fires resulted in enhancements in summer background CO and O3 in Alaska, Canada, and the Pacific Northwest [Jaffe et al., 2004; Bertschi and Jaffe, 2005].

[6] According to the Global Fire Monitoring Center (GFMC, http://www.fire.uni-freiburg.de/), 18.9 million hectares of land burned in Siberia, which is the largest area reported in at least 10 years. Reason for these forest fires were the extraordinary dry conditions that prevailed in Siberia in 2003. The dry conditions caused exceptionally high and continuous fire activity from May to August 2003 (ATSR World Fire Atlas, http://shark1.esrin.esa.it/FIRE/AFATSR) [Jaffe et al., 2004]. The aerosol plumes circled around the Northern Hemisphere with the prevailing westerly winds within a period of less than three weeks [Damoah et al., 2004].

[7] The total area burnt in Canada during the fire season 2003 was below normal, but nevertheless there were periods of very strong burning, particularly in Manitoba and Ontario. Especially strong events were reported after mid-June (Canadian Forest Service, http://www.nofc.forestry.ca/fire/), and likely represented another major source of the lofted particle layers observed over Leipzig.

[8] In this presentation we extend the analysis of the Raman lidar observations of the particle optical properties discussed by Mattis et al. [2003]. We present a microphysical particle characterization of the lofted, aged biomass-burning aerosol plumes that had originated from boreal areas thousands of kilometers upwind of the lidar site. Section 2 describes the Raman lidar and the inversion algorithm used for the optical and microphysical particle characterization. Section 3 presents the measurement cases. Section 4 discusses the findings. Section 5 closes with a summary.

2. Methodology

2.1. Dual-Wavelength Raman Lidar

[9] A description of the Raman lidar and the techniques used for the retrieval of the optical particle parameters is given by Mattis et al. [2002, 2004]. A Nd:YAG laser is used for generating laser pulses at 355, 532, and 1064 nm. The pulse repetition rate is 30 Hz. Profiles of the particle volume extinction coefficients are derived at 355 and 532 nm with the use of nitrogen vibrational Raman signals detected at 387 and 607 nm, respectively [Ansmann et al., 1990]. Signal averaging was done such that the uncertainties remained below a level at which a successful inversion of optical into microphysical particle parameters could be carried out. Relative uncertainties were 15–20% in the center of the particle plumes discussed in this paper. Absolute errors were 1.5–4 Mm−1. This number was derived under the assumption of values of 10–20 Mm−1 (Mm−1 = 1 × 10−6m−1 = 1 × 10−3km−1) for the extinction coefficient. These errors are smaller than those presented by Mattis et al. [2003]. In that case we analyzed all measurements of forest fire smoke in 2003, including cases of rather low particle extinction, which cause much higher uncertainties.

[10] Particle backscatter coefficients at 355, 532, and 1064 nm were calculated with the Raman method [Ansmann et al., 1992]. Relative uncertainties were ≤5% in the center of the plumes. Maximum errors were 20%. Absolute uncertainties were on the order of 0.01 Mm−1sr−1.

[11] The lidar ratios directly follow from the ratios of the extinction and backscatter coefficients at 355 and 532 nm, respectively. Errors in the worst case added up to 60%. Typical backscatter and extinction values in the center of the forest fires plumes resulted in errors on the order of 20%.

[12] The Ångström exponent [Ångström, 1964] was calculated from the extinction coefficients at 355 and 532 nm wavelength. This parameter describes the slope of the spectrum defined by the extinction coefficients at the two wavelengths. The uncertainty is determined by the errors of the individual profiles, and varies from 20 to 60%.

[13] The depolarization ratio of the particles is calculated from the total signal and the signal cross-polarized with respect to the state of polarization of the light emitted at 532 nm [Cairo et al., 1999]. Values in the haze plumes were close to the Rayleigh value; that is, they varied from 1.5 to 3%. The statistical uncertainty was <5%.

[14] The relative humidity was determined from measurements of the water vapor mixing ratio and temperature with lidar. The error is mainly determined by the uncertainty of the temperature measurements. The ratio of the signals from Raman scattering from water vapor and nitrogen molecules was used for calculating the water vapor mixing ratio [Melfi et al., 1969; Ansmann et al., 1992]. Average statistical errors of the water vapor mixing ratio were 5–15% (for relative humidities around 20%) in the center of the particle plumes. The temperature profiles were determined from measurements of the signals from rotational Raman scattering at several wavelengths around 532 nm [Arshinov et al., 1983, 2005; Mattis et al., 2002]. A detailed description of error analysis of temperature profiles is given by Mattis et al. [2002]. The accuracy of the temperature profiles is ∼1 K in the free troposphere for the cases presented here. Relative humidity thus was determined with 10–15% statistical uncertainty.

2.2. Inversion Algorithm

[15] A description of the inversion code is given by Müller et al. [1999a, 1999b] and Veselovskii et al. [2002, 2004]. Particle backscatter coefficients at the three laser wavelengths and particle extinction coefficients at 355 and 532 nm are used as input information. The inversion code provides approximations of volume size distributions from which effective radius, volume and surface-area concentration, as well as complex refractive index are derived. This information is used to calculate the particle single-scattering albedo with a Mie-scattering code [Bohren and Huffman, 1983].

[16] Uncertainties in general are <30% for effective radius. Errors can become as large as 50% for volume and surface-area concentration. The real part of the complex refractive index is derived to an accuracy of better than ±0.1. The imaginary part is found to its correct order of magnitude, if it is <0.01i. For larger values of the imaginary part the uncertainty is <50%. The single-scattering albedo can be calculated with an accuracy of ±0.05, if uncertainties of the input optical data on average are ∼10%. The uncertainty of the extinction coefficients was around this error level or higher. In combination with the much lower uncertainty of the particle backscatter coefficients which was less than 10% in the center of the haze plumes, the requirement of ∼10% was rather well fulfilled. A detailed error analysis is given by Müller et al. [1999b, 2001] and Veselovskii et al. [2002, 2004].

2.3. Backward Dispersion Model

[17] To explore where the observed aerosols have come from, the Lagrangian particle dispersion model FLEXPART [Stohl et al., 1998; Stohl and Thomson, 1999] was used. FLEXPART had already been used by Damoah et al. [2004] for the characterization of the transport pattern of smoke from the Russian boreal forest fires observed in the period from 10 to 31 May 2003. The model simulates the transport and dispersion of linear tracers by calculating the trajectories of a multitude of particles. FLEXPART was driven by global model-level data from the European Centre for Medium-Range Weather Forecasts with a temporal resolution of 3 hours (analyses at 0000, 0600, 1200, and 1800 UTC; 3-hour forecasts at 0300, 0900, 1500, and 2100 UTC), a horizontal resolution of 1° × 1°, and 60 vertical levels. Particles were transported both by the resolved winds and by parameterized subgrid motions. The model parameterizes turbulence in the boundary layer and in the free troposphere by solving Langevin equations [Stohl and Thomson, 1999]. To account for convection, a parameterization scheme is used [Emanuel and Živković-Rothman, 1999] which is based on the buoyancy sorting principle.

[18] In the present case, FLEXPART was run backward in time from along the lidar profiles; see Stohl et al. [2003] for a description of the backward mode. 40,000 particles were released every 500 m height and tracked for 20 days backward in time. The model output, which has a daily resolution, consists of a gridded response function to emission input, which is proportional to the residence time of the particles in a given volume. The response function, or emission sensitivity function has the unit ps/kg (ps = pico second). If it is multiplied with a known emission flux field in kg/m3/ps and integrated over the volume of the atmosphere, a mixing ratio at the receptor location is obtained. In analogy to forward simulations, we may call the plumes of the emission sensitivity function emerging from the release location retroplumes.

3. Observations

3.1. Measurement From 26 June 2003

[19] From May until August 2003 regular lidar observations (at least once per week) showed high particle backscattering and extinction in the free troposphere, see Figure 2 of Mattis et al. [2003]. Figure 1 shows the time-height plot of the range-corrected backscatter signal at 1064-nm wavelength for the measurement from 26 June 2003. Also shown are the 1-hour mean profiles of the particle backscatter and extinction coefficients measured on that day. The measurement serves as example to explain important properties of the forest fire plumes.

Figure 1.

Time-altitude plot of the range-corrected backscatter signal measured at 1064 nm wavelength from 2032 to 2133 UTC on 26 June 2003. Also shown are the 1-hour mean profiles of the particle backscatter coefficients at 355, 532 and 1064 nm, and the particle extinction coefficients at 355 and 532 nm. The profiles of the backscatter coefficients were smoothed with 60 m. The profiles of the extinction profiles were smoothed with 300 m up to 2.2-km height, 660 m from 2.2- to 4.4-km height, and 1260 km above 4.4-km height. Typical errors of backscatter and extinction coefficients are given in section 2.

[20] Significant values of particle backscatter coefficients were found up to 10-km height. Background values were reached around 12-km height. A characteristic feature of plumes advected to our site over transcontinental distances are stratified layers, sometimes stacked upon each other, of a few hundred meters in depth; see Plate 1 and Figure 1 of Mattis et al. [2003].

[21] The backscatter coefficients strongly decrease around 2.25-km height, which represents the height of the planetary boundary layer on that day. Below that height it is rather likely that a mixture of biomass burning and European anthropogenic pollution prevailed. In order to have a clear separation of these aerosol types we focused our analysis on data acquired above 2 km height. Particle extinction coefficients were as high as 0.02 km−1 at 532 nm wavelength around 6-km height, and dropped to background values around 9 km height. The particle lidar ratios varied between 40 and 60 sr at 532 nm and 30 and 55 sr at 355 nm.

[22] Figure 2 presents the results of the backward simulation for the altitude range from 5.5 to 6 km, corresponding approximately to the altitude of the observed layer of strongly enhanced backscatter. The upper panel of Figure 2 shows the column-integrated emission sensitivity integrated over the full 20-day time period.

Figure 2.

Emission sensitivity obtained from the 20-day backward simulation with FLEXPART for the layer between 5500 and 6000 m from 2032 to 2133 UTC on 26 June 2003. Shown are vertically integrated emission sensitivity (upper panel) and emission sensitivity for a layer from the surface to 3-km height and only for days 8 (middle panel) and 7 (lower panel) backward in time. Numbers shown on each of the plots are the days backward in time. They are plotted at the retroplume centroid location on that day.

[23] The air mass traveled from the Atlantic Ocean to the lidar site. Because of the modeling approach the plumes are moving backward in time from the lidar site to the source region. About 2 days back in time and close to Iceland the backward plume split into two branches, one of which traveled toward Cuba and the larger one crossed North America. The vertically integrated emission sensitivity shows a maximum over Alberta, Saskatchewan and Manitoba, where the retroplume was located approximately on days 7 and 8 backward in time, i.e., around 19 and 20 June 2003. Figure 2 (middle and lower panels) shows the emission sensitivity averaged over the layer from the surface to 3 km height for days 8 (approximately 19 June) and 7 (approximately 20 June) back in time. This layer was chosen because the emissions from forest fires may occur not only at the ground but may also be injected at higher altitudes. However, a similar result was also obtained with a layer extending only up to 150 m above ground. The retroplume reached low altitudes in two separate regions, in the western North Atlantic and over North America. On 20 June, the highest values over North America are found just west of James Bay, whereas on 19 June they are located farther toward to the southwest.

[24] Exactly in the regions with the largest emission sensitivity, very strong fire activity was reported. The weekly fire report from the Canadian Forest Service (http://www.nofc.forestry.ca/fire/) from 25 June notes that for a 10-year average “the weekly fire activity was above normal this past week with 520 fires (129% above average) burning over 383,000 ha (180% above average). The dates of 19 and 20 June accounted for almost half of all the new fire starts. Manitoba and Northern Ontario were hardest hit with several evacuations occurring over the weekend in Ontario.” Many of the new fires on 19 and 20 June were started by lightning due to strong thunderstorm activity. Because of the high wind speeds under these conditions it is likely that these fires were very intense crown fires. Figure 3 shows hot spot maps for 19 and 20 June 2003. Many fires burned in the region with the largest emission sensitivity and the difference between the maps for 19 and 20 June indicates that many of the new fires were located underneath the retroplume from Leipzig.

Figure 3.

Maps of hot spots of forest fires (red points) issued by the Canadian Forest Service for 19 June and 20 June 2003.

3.2. Optical Properties

[25] In this section we discuss the results obtained for the free tropospheric particle layers observed on 29 May, 26 June, 10 July, and 25 August 2003. Haze layers observed above 2-km height on 29 May 2003 rather likely originated from the boreal areas of central and eastern Siberia. According to Damoah et al. [2004] the air masses had traveled with westerly winds across Canada and the North Atlantic to Central Europe in less than 3 weeks. Part of the haze observed on 29 May likely stemmed from another plume that traveled in the opposite direction across the Ural mountains to Europe.

[26] The plumes on 10 July 2003 were probably generated by fires in two different source regions. Forest fires injected into the atmosphere over North America may have been responsible for the particle load observed below 7-km height. According to the ATSR World Fire Atlas and reports by the Canadian Forest Fire Service the fires that started on 19–20 June continued burning at least until 25 June 2003. The FLEXPART simulations showed high emission sensitivity in the layer from 0- to 150-m height, and also in the layer from 0- to 3-km height over Canada for backward simulations that had started from Leipzig below 6-km height. The largest values were found in the simulations that had started from below 3-km height over Leipzig. Travel time of the plume in this altitude range was about 2 weeks, such that the aerosols likely originated from the same fires as for the measurement case from 26 June 2003, but from a few days later, around 25 June 2003.

[27] Siberian fires may still have contributed to the particle load >7-km height. For retroplumes started between 7- and 9-km height, the FLEXPART emission sensitivity below 150-m height was zero everywhere on the globe, except for a patch of high sensitivity in central Asia, approximately over southern Kazakhstan, northwestern China, and Mongolia about 2–3 weeks back in time. Satellite images of Russia show strong fire activity to the southwest of Lake Baikal for the period around 25 June 2003. Although FLEXPART simulations put the emission sources a little farther to the south of the area of intense fire, the model results agree reasonably well given the large transport times and overall simulation uncertainties.

[28] In contrast, the lofted haze plume observed on 25 August was probably caused by anthropogenic emissions in the North American continent. According to FLEXPART results pollution from the northwest coast of the United States may have contributed to the particle load. The air masses then travelled over the Great Lakes and the east coast where additional uptake of pollution may have occurred. The lidar ratio as well as the mean particle size are completely different from the values found on the other 3 days, and strongly resemble results from measurements of anthropogenic pollution (see section 4).

[29] FLEXPART simulations however do not provide a conclusive result. According to the simulations forest fires also contributed to the observed particle load. Strong forest fire activity was noted in British Columbia and Alberta in mid-August. In fact, at that time the second peak of forest fire activity after the strong fires in June was reached. According to maps of the Canadian Forest Service up to 100% of the tree crowns were burned in areas affected by forest fire. Such fires create extremely high temperatures, and thus may generate particles quite different from the ones observed on the other 3 days; see section 4. FLEXPART simulations indicate a relatively small contribution of CO from anthropogenic sources to the overall CO burden, which also points to a significant contribution by forest fires. Transport time of the plume observed on 25 August 2003 was approximately 5 days, and thus short, compared to the other cases.

[30] The profiles of the optical properties were split into sublayers of variable height. The optical data of these layers were averaged and subsequently used for the data inversion. The separation into sublayers was done such that data averaging across strong gradients of the intensive parameters, i.e., particle lidar ratio and particle Ångström exponents, was avoided as good as possible.

[31] Figure 4 shows the Ångström exponents and the particle lidar ratios for the selected layers. Also shown are the profiles of relative humidity for the selected measurement times. Relative humidity was mostly <50% above the planetary boundary layer. Extremely dry conditions with relative humidity <20% were found in large sections of the profiles. Table 1 lists the data shown in Figure 4.

Figure 4.

(a–d) Ångström exponents for the wavelength range from 355 to 532 nm and (e–h) particle lidar ratio at 355 (solid squares) and 532 nm (open squares). Shown are the mean values derived from averaging the optical data across height ranges. The optical data were subsequently used in the data inversion. Vertical error bars denote these height ranges, respectively. The height ranges are given in Table 1. Horizontal error bars denote standard deviation of the Ångström exponents and lidar ratios in each height interval, respectively. Also shown are the profiles of the particle backscatter coefficient at 532 nm (thick lines), and relative humidity (thin lines). Measurement times were 2003–2104 UTC on 29 May, 2032–2133 UTC on 26 June, 2059–2200 UTC on 10 July, and 1944–2050 UTC on 25 August 2003. The vertical resolution of the profiles of the backscatter coefficient is 60 m. The profiles of relative humidity were smoothed with 60 m from 0- to 4-km height, 300 m from 4- to 6-km height, and 660 m above 6-km height. Uncertainties of the backscatter profiles and relative humidity profiles is given in section 2. The dashed vertical lines denote 50% relative humidity.

Table 1. Mean Values of Particle Backscatter Coefficients, Ångström Exponents, Lidar Ratios, and Relative Humidity for the Selected Height Rangesa
DateHeight, mβ532, Mm−1sr−1ålr355, srlr532, srRH, %
  • a

    See Figure 4. Uncertainty denotes the height range across which optical data were averaged for the data inversion. Here, β532 denotes the particle backscatter coefficient at 532 nm wavelength, å denotes the Ångström exponent calculated for the wavelength range from 355 to 532 nm, lr355 is the lidar ratio at 355 nm, and lr532 is the lidar ratio at 532 nm. The lidar ratios are given in unit of sr. RH denotes the relative humidity, given in percent.

29 May2.01 ± 0.120.5960.82678151
29 May2.61 ± 0.120.4431.10476149
29 May3.42 ± 0.510.2880.66213129
29 May4.59 ± 0.180.1160.71538712
29 May5.34 ± 0.090.1670.2732608
26 June2.52 ± 0.510.4090.37456145
26 June4.23 ± 0.30.2490.0032495
26 June5.91 ± 0.180.3720.64334610
26 June6.66 ± 0.210.3121.10495019
26 June7.53 ± 0.480.2571.00485534
10 July2.76 ± 0.270.3610.6935524
10 July3.51 ± 0.360.2750.7826364
10 July4.2 ± 0.270.2220.4734504
10 July5.46 ± 0.150.1750.64212615
10 July6.51 ± 0.180.1710.22264017
25 Aug.2.85 ± 0.240.2661.7746356
25 Aug.4.92 ± 0.510.3161.89453324
25 Aug.5.64 ± 0.210.3662.06412840
25 Aug.6.42 ± 0.510.3411.68473753
25 Aug.7.17 ± 0.420.2551.87553846

[32] Ångström exponents varied between 0 and 1.1 on 29 May, 26 June, and 10 July 2003. In contrast, the measurement from 25 August 2003 was characterized by larger Ångström exponents above 1.7, which indicates the presence of considerably smaller particles compared to particles observed on the other 3 days.

[33] The particle lidar ratios representing the biomass-burning plumes are lower at 355 nm compared to those at 532 nm; see also Mattis et al. [2003]. Mean lidar ratios at 355 nm vary from 21 to 67 sr, whereas 26–87 sr is found for the lidar ratio at 532 nm. A similar spectral behavior was also found from dual-wavelength Raman lidar observations of Siberian biomass-burning aerosol detected over Tokyo, Japan, on 21 May 2003 [Murayama et al., 2004], and for biomass-burning aerosol advected from Canada to Germany [Wandinger et al., 2002]. In contrast, lidar ratios of young forest fire smoke observed over Greece in August 2001 showed a different spectral behavior. Values of around 60 sr and 50 sr at 355 and 532 nm, respectively, were reported [Balis et al., 2003].

[34] Lidar ratios measured at 355 nm on 25 August 2003 fall within the variability of numbers found on the other 3 days. Lidar ratios at 532 nm tend to be smaller than the numbers derived for the forest fire smoke.

[35] Measurement errors cause an uncertainty of the lidar ratios of approximately 15–25% in the center of the pollution plume observed on 25 August 2003. Despite such uncertainties the mean lidar ratio at 355 nm was higher by about 30% compared to the mean lidar ratio at 532 nm. Long-term observations of anthropogenic haze over Leipzig also showed that on average the lidar ratio at 355 nm is larger than the one at 532 nm [Mattis et al., 2004].

[36] Optical depth was determined from observations with the institute's Aerosol Robotic Network (AERONET) Sun photometer [Holben et al., 1998]. Table 2 presents the numbers for optical depth measured at 500 nm wavelength. The table also presents optical depth of the free troposphere up to 10-km height measured with lidar at 532 nm. The height of the planetary boundary layer was 1.77 km on 29 May 2003, 2.25 km on 26 June 2003, 1.53 km on 10 July 2003, and 1.95 km on 25 August 2003. On the basis of these numbers it is possible to compute the contribution of the lofted haze layers to total optical depth.

Table 2. Optical Depth Measured With AERONET Sun Photometer at 500 nm Wavelengtha
Instrument29 May26 June10 July25 August
  • a

    The first row presents the approximate height of the planetary boundary layer (PBL height). The second row gives the optical depths that were measured with the Sun photometer; the second row gives the mean values and respective standard deviation obtained from all Sun photometer observations carried out on each day, respectively. The third row presents those values that were measured closest in time to the start of the respective lidar observations. The last Sun photometer observation was at 1542 UTC on 29 May, 1713 UTC on 26 June, 1734 UTC on 10 July, and 1616 UTC on 25 August 2003. Respective lidar observations were carried out from 2003 to 2104 UTC on 29 May, 2032 to 2133 UTC on 26 June, 2059 to 2200 UTC on 10 July, and 1944 to 2050 UTC on 25 August 2003. The fourth row presents the mean values of optical depth above the planetary boundary layer measured with the lidar at 532 nm wavelength.

PBL height1.77 km2.25 km1.53 km1.95 km
SPM (mean)0.182 ± 0.006 (32)0.248 ± 0.077 (4)0.122 ± 0.016 (3)0.144 ± 0.018 (23)
SPM0.1830.160.140.165
Lidar0.0830.0920.0980.075

[37] A considerable amount of optical depth (45–70%) was contributed by the free-tropospheric aerosol layers. This estimate rests upon the assumption that optical depth did not change too much between the Sun photometer observations carried out close to sun set and the following nighttime lidar observations. The time series of particle optical depth obtained with Sun photometer for 29 May 2003 (32 individual measurements) shows that this assumption is rather well justified. In the case of 26 June and 10 July 2003 only 3 and 4 individual measurements were available during each day, respectively. In the case of 25 August 2003 we note an increase of optical depth by approximately 20% in the final 15 min of the Sun photometer observations.

3.3. Microphysical Properties

[38] Figure 5 shows the results for particle effective radius and single-scattering albedo. Also shown are the profiles of the backscatter coefficient at 532 nm and relative humidity. Table 3 summarizes the results for particle effective radius, particle volume and surface-area concentration, and complex refractive index. The large uncertainty of the imaginary part exceeds the mean values, and results from the rather low light absorption of the particles. The data inversion becomes rather uncertain, if imaginary parts are <0.01i [Müller et al., 1999b]. Accordingly, errors easily add up to 100%.

Figure 5.

Particle effective radius (squares) and single-scattering albedo at 532 nm (circles) for the selected measurement days. Horizontal error bars denote uncertainty from data inversion. Vertical error bars denote height range across which optical input data were averaged. Profiles show particle backscatter coefficient at 532 nm (thick lines), and relative humidity (thin lines). The dashed vertical lines denote 50% relative humidity.

Table 3. Physical Particle Parameters and Respective Standard Deviations Derived From Inversion of Optical Dataa
DateHeight, mreff, μmv, μm3cm−3s, μm2cm−3mrealmimag
  • a

    Numbers behind the semicolon denote median of probability density distribution of solution space obtained for the imaginary part. Here, reff denotes the effective radius, v denotes the volume concentration, and s denotes the surface-area concentration. The term mreal is the real part, and mimag is the imaginary part of the complex refractive index. The refractive index is given as a wavelength-independent quantity.

29 May2.01 ± 0.120.31 ± 0.0511 ± 2100 ± 141.37 ± 0.040.001 ± 0.003; 0.001
29 May2.61 ± 0.120.24 ± 0.055.3 ± 1.765 ± 151.47 ± 0.010.004 ± 0.005; 0.001
29 May3.42 ± 0.510.29 ± 0.041.4 ± 0.214 ± 11.6 ± 0.060.004 ± 0.005; 0.0025
29 May4.59 ± 0.180.38 ± 0.072 ± 0.516 ± 1.21.38 ± 0.050.002 ± 0.004; 0.001
29 May5.34 ± 0.090.4 ± 0.132.3 ± 116 ± 41.45 ± 0.130.007 ± 0.011; 0.001
26 June2.52 ± 0.510.34 ± 0.053.6 ± 0.831 ± 21.46 ± 0.040.006 ± 0.005; 0.005
26 June4.23 ± 0.30.41 ± 0.142.1 ± 115 ± 1.71.45 ± 0.070.005 ± 0.006; 0.0025
26 June5.91 ± 0.180.36 ± 0.133.8 ± 1.830 ± 61.45 ± 0.090.003 ± 0.005; 0.001
26 June6.66 ± 0.210.33 ± 0.084.1 ± 1.437 ± 91.39 ± 0.070.001 ± 0.003; 0.001
26 June7.53 ± 0.480.33 ± 0.13.4 ± 1.331 ± 81.4 ± 0.090.003 ± 0.006; 0.001
10 July2.76 ± 0.270.34 ± 0.114.1 ± 1.835 ± 71.45 ± 0.090.003 ± 0.004; 0.001
10 July3.51 ± 0.360.29 ± 0.081.7 ± 0.717 ± 31.54 ± 0.060.003 ± 0.003; 0.0025
10 July4.2 ± 0.270.33 ± 0.091.7 ± 0.616 ± 1.41.49 ± 0.060.006 ± 0.006; 0.001
10 July5.46 ± 0.150.3 ± 0.050.8 ± 0.28 ± 11.56 ± 0.030.003 ± 0.003; 0.0025
10 July6.51 ± 0.180.38 ± 0.051.1 ± 0.29 ± 0.61.51 ± 0.030.004 ± 0.005; 0.005
25 Aug.2.85 ± 0.240.17 ± 0.041.7 ± 0.629 ± 51.52 ± 0.060.003 ± 0.004; 0.001
25 Aug.4.92 ± 0.510.16 ± 0.042.3 ± 143 ± 91.53 ± 0.070.001 ± 0.002; 0.001
25 Aug.5.64 ± 0.210.16 ± 0.042.3 ± 143 ± 91.52 ± 0.050.003 ± 0.003; 0.001
25 Aug.6.42 ± 0.510.2 ± 0.032.5 ± 0.835 ± 51.52 ± 0.050.003 ± 0.004; 0.0025
25 Aug.7.17 ± 0.420.18 ± 0.052.9 ± 1.348 ± 131.53 ± 0.070.001 ± 0.002; 0.001

[39] The probability density distribution of the solution space of the imaginary part was quite skewed. It showed a pronounced maximum of number of solutions toward low imaginary parts, which thus allows one to define the so-called most likely value of the imaginary part; see also Müller et al. [2000, 2001]. This value also defined the count median value in almost all the investigated data sets. For that reason, Table 3 also lists those imaginary parts at which the maximum of the density distribution was found. The count median imaginary parts reflect the grid of values used in the inversion algorithm [Müller et al., 1999b]. Imaginary parts were tested at 0i, 0.001i, 0.0025i, 0.005i, 0.0075i, and 0.01i–0.1i with step width 0.025i.

[40] Effective radii varied from 0.24 to 0.4 μm on 29 May 2003. Similarly large values of particle effective radius were found for the measurement on 26 June and 10 July 2003. In that respect there does not seem to be a significant difference in size for particles advected from the Siberian fires and the ones transported from Canada. The particle sizes are in strong contrast to particle effective radii of 0.16–0.2 μm found for the anthropogenic pollution plume on 25 August 2003. One of the main results of this study is that particles of the forest fire smoke are also much larger than particles of biomass-burning plumes observed close to source regions; compare section 4.

[41] There is little variation of single-scattering albedo at 532 nm on all 4 days, i.e., in that respect the biomass-burning particles do not differ from the case of anthropogenic pollution. It has to be observed, however, that the uncertainties of single-scattering albedo are comparably high.

[42] According to Table 3, volume concentrations were <11 μm3cm−3. The majority of cases showed volume concentrations <4 μm3cm−3. Volume concentration varied by a factor of 2–8 within the plumes on the 4 days. A similar variability was found for the surface-area concentration.

[43] The real part of the complex refractive index varied from 1.37 to 1.6. Real parts of the anthropogenic particles are at the upper end of values retrieved for the biomass-burning particles. Mean imaginary parts in all cases were ≤0.007i. As mentioned before, the uncertainty is quite high, but still allows us to state that the light absorption of the particles was rather low. This property is in strong contrast to the highly absorbing biomass-burning particles observed with lidar during the Indian Ocean Experiment in 1999/2000 [Franke et al., 2003; Müller et al., 2003].

[44] Figure 6 shows a high correlation coefficient of 0.94 for mean effective radius versus mean Ångström exponent, if the data from all 4 days are considered. The correlation coefficient is 0.42 in the case of mean single-scattering albedo versus mean imaginary part. The correlation increases to 0.54, if only data points describing the forest fire smoke are considered.

Figure 6.

Correlation plot of (a) Ångström exponent versus effective radius and (b) single-scattering albedo versus imaginary part of the complex refractive index. Open circles denote cases of forest fire smoke. Solid circles describe the case of anthropogenic pollution. Regression lines denote best linear fit. In the case of Figure 6b, fit is shown under consideration of all data points (solid line) and if only data points representing forest fire smoke are considered (dash-dotted line).

4. Discussion

4.1. Particle Size

[45] Figure 4 shows that the anthropogenic particles observed on 25 August 2003 were much smaller than the biomass-burning particles. The smoke particles were also considerably larger than European haze particles observed in the free troposphere with six-wavelength lidar during the Second Aerosol Characterization Experiment (ACE 2) [Müller et al., 2002]. Effective radii of aged particles that were generated by anthropogenic sources in the European continent were 0.15 ± 0.06 μm. The forest fire particles were also larger than mixtures of anthropogenic and biomass-burning particles found in free-tropospheric plumes during the Indian Ocean Experiment (INDOEX) [Müller et al., 2003]. In that case, particle sizes were 0.2 ± 0.08 μm. Lidar observations of an aged biomass-burning plume that had originated from western Canada in 1998 and which was observed over Germany during the Lindenberg Aerosol Characterization Experiment (LACE) 98 showed effective radii from 0.11 ± 0.03 μm to values as large as 0.27 ± 0.04 μm; see Tables 1 and 2 of Wandinger et al. [2002] and Tables 3 and 4 of Veselovskii et al. [2002]. In situ observations carried out by aircraft in the same plume showed particle effective radii as large as 0.25 ± 0.07 μm [Fiebig et al., 2002; Wandinger et al., 2002]. The Ångström exponent was as large as 0.06 which puts it in the range of values found in this study. Dual-wavelength Raman lidar observations of Siberian forest fire smoke were carried out at Tokyo on 21 May 2003 [Murayama et al., 2004]. Effective radii were smaller than what was found at the Leipzig site on 29 May 2003. Numbers varied around 0.21 μm, which is equivalent to an Ångström exponent on the order of 1.35.

[46] Effective radii of the particles are considerably larger than what is usually observed near source regions of forest fires (e.g., Table 2 of Fiebig et al. [2003, and references therein]). Because of the unknown burning conditions that prevailed at the source of the Siberian and Canadian fires, and insufficient information regarding the physical, chemical, and meteorological processes that occurred along the transport path of the plumes to our site, we can merely speculate on possible reasons that generated these large particles.

[47] In situ observations of particles in South America during the Smoke, Clouds and Radiation–Brazil (SCAR-B) field project indicated that the kind of fires, i.e., flaming versus smoldering combustion, has a significant influence on the size of the particles [Reid and Hobbs, 1998; Reid et al., 1998]. Burning of biomass at higher temperatures, i.e., flaming fires, generates smaller particles than smoldering fires; see, e.g., Figure 12 of Reid et al. [1998]. In forest fires, the flaming stage is followed by a longer period of smoldering fires [Ward et al., 1996; Ferek et al., 1998]. The large smoke particles we observed may thus have originated from the smoldering phase of the fires.

[48] The kind of burnt material as well as the combustion efficiency also influence the size of the particles. For that reason, considerably larger particles than what usually are found in, e.g., South America and southern Africa (see below) may have already been generated in the source regions in Siberia and Canada. Fire radiative power determined from measurements with the MODIS sensor aboard the Aqua and Terra satellites indicates that vegetation fires burn less intense in Russia compared to Canadian fires [Wooster and Zhang, 2004]. The lower fire intensity is probably due to the fact that the burning of surface fuels is predominant over the more intense burning of tree crowns [Furyaev, 1996; Conard and Ivanova, 1997].

[49] There is strong evidence from observations of biomass-burning plumes in different regions of the world that particles grow in size during the aging of the plumes, e.g., Table 2 of Fiebig et al. [2003]. Processes that lead to the increase of particle size are gas-to-particle conversion of inorganic and organic vapors [Reid and Hobbs, 1998; Reid et al., 1998], condensation of large organic molecules from their gas phase in the first few hours of aging [Reid and Hobbs, 1998; Pósfai et al., 2004], particle growth due to coagulation [Westphal and Toon, 1991; Fiebig et al., 2003], and photochemical and cloud-processing mechanisms. Figure 4 shows that relative humidity generally was <50% in the haze plumes observed on 29 May, 26 June, and 10 July 2003. In many cases it varied between 10% and 20%. Thus hygroscopic growth [Hobbs et al., 1997] seems to be a minor factor responsible for the rather large particle size.

[50] Ångström exponents on the order of 1.8 for rather fresh forest fire smoke (distances up to 600 km from the sources) were determined from Sun photometer observations carried out in north central Canada during the Boreal Ecosystem-Atmosphere Study (BOREAS) [Markham at al., 1997], and during long-term observations carried out in the same area from 1994 to 1999 [Holben et al., 2001]. Observations of smoke from intense forest fires in Canada in 1998 were carried out with Sun photometer located in different spots downwind of the fire events [O'Neill et al., 2002]. The authors noted a decrease of Ångström exponent with distance to fire source. According to regression analysis the Ångström exponent decreased from 1.9 at 30 km distance to 1.4 at 2000 km distance. It has to be kept in mind that local sources of anthropogenic pollution in the planetary boundary layer contributed to the overall signal. Effective radii were retrieved from the optical measurements. Regression analysis showed that particle size slightly increased from 0.13 μm at 30 km distance to 0.15 μm at 2000 km distance. Such a distance is equivalent to a transport time of smoke of up to 5 days. The cases presented here describe even longer transport times. Fiebig et al. [2003] presented model calculations of the growth of particles, which were observed with airborne in situ instruments during LACE 98. The authors showed that particle coagulation processes can sufficiently well explain the large effective radii observed, if transport over a period of 6 days was assumed. An increase of particle size of up to 50% was found from the scenarios considered in their study.

[51] The evolution of a massive haze plume that originated from strong forest fires in central Quebec (Canada), and which then spread across northeastern parts of the United States, was analyzed with lidar, Sun photometer, and airborne in situ instrumentation [Colarco et al., 2004]. Mean Ångström exponents obtained with aircraft in the plume up to 1000 km downwind of the source varied from 0.83 to 1.23 [Colarco et al., 2004; Taubman et al., 2004] for the wavelength range from 450 to 700 nm, and thus point to relatively large particles. The measurements were carried out at ≤20% relative humidity. Taubman et al. [2004] found the peak of the size distribution at particle diameters from 0.3 to 0.6 μm, which points to much larger particles than what is usually found for anthropogenic pollution. Average particle size may have been even larger, as the authors could not rule out loss effects of the aircraft inlet used in their study. Model simulations showed that particle coagulation processes occurring along the way from the source to the observation sites could explain the low Ångström exponents [Colarco et al., 2004]. Sun photometer observations of that event resulted in values of ∼1.24 (440–670 nm), which is at the upper end of numbers obtained from the aircraft observations [Eck et al., 2003]. Anthropogenic pollution in the planetary boundary layer rather likely caused these higher values.

[52] Particle size measurements were carried out with a balloon-borne optical particle counter in combination with lidar observations in the framework of the Mildura Aerosol Tropospheric Experiment (MATE 98) in southern Australia [Rosen et al., 2000]. At times lidar detected free tropospheric layers at heights up to 13 km. Effective radius was 0.18 μm for one of the observed free-tropospheric particle size distributions. The measurement was characterized by advection of haze plumes with the prevailing westerly winds from southern Africa and/or South America. In view of the transport times the particles probably were rather aged.

[53] Liousse et al. [1995] found a significant increase of particle size with age of plumes that originated from savanna fires in Ivory coast in Africa. A decrease of the Ångström exponent (450–550 nm wavelength), which is, as mentioned, equivalent to an increase of particle size was observed for fires in tropical forest and cerrado during SCAR-B [Reid et al., 1998]. In situ observations showed that Ångström exponents were on the order of 2.2 ± 0.2 for fresh smoke and 1.2 ± 0.2 for aged smoke. The numbers for aged smoke are at the upper end of values reported here.

[54] Ångström exponents of particle plumes generated by strong fires events in Mexico in 2002 were 1.28–1.58 for the wavelength range from 440 to 670 nm [Kreidenweis et al., 2001]. Observations were carried out with Sun photometer. These numbers also are in the upper range of values presented here, and describe plumes from 2 days up to 10 days of age. Care has to be taken in this comparison, as the column mean values most certainly do not describe pure biomass-burning particles. The plumes were transported from Mexico to the United States, and thus most likely were affected by anthropogenic pollution. In contrast, the free-tropospheric particle layers we observed were rather likely much less influenced by anthropogenic pollution in the planetary boundary layer.

[55] Table 4 summarizes the results for particle effective radius and Ångström exponents of particles generated by fires in boreal regions. There is almost a factor 3 difference between particles detected near the sources [see O'Neill et al., 2002] and particles we detected far away from the fire events. In addition there seems to be a systematic shift of particle size toward larger values with increasing distance from the source of boreal fires. It has to be emphasized once more that the fires described here and the fires discussed in the cited literature certainly had different properties. Thus definite conclusions regarding the increase of particle size with duration of transport have to be treated with caution, and are subject to future studies.

Table 4. Particle Effective Radius, Ångström Exponent, Single-Scattering Albedo, and Imaginary Part of Complex Refractive Index Found for Forest Fire Plumes From Boreal Firesa
Source RegionObservational Area (Distance)reff, μmåssamimagapReference
  • a

    Numbers are discussed in the text. Meaning of symbols is as before. Wavelength ranges for which the Ångström exponents were calculated are given in the text. Here, ssa is the single-scattering albedo at 532 nm, and ap means that observed plumes were possibly affected by anthropogenic pollution in the planetary boundary layer.

  • b

    Distance estimated under the assumption that the plume took the direct (shortest) path from the source to the lidar site in a westward direction.

  • c

    The reff, ssa, and mimag obtained from inversion of lidar optical data with IfT inversion algorithm; see Table 1 of Wandinger et al. [2002].

  • d

    Here, n/a means that information on travel distance or age of the particles could not be estimated.

  • e

    Numbers denote range of radii around which maximum number concentration of particles was found.

East/Central Siberia and central CanadaGermany (∼7000–13,000 km)b0.36 ± 0.050.71 ± 0.440.93 ± 0.030.003 ± 0.003 this study
West CanadaGermany (∼8000 km)b0.27 ± 0.04∼0.060.83 ± 0.060.05 ± 0.02 Wandinger et al. [2002]c
East/central SiberiaJapan (∼3000–4000 km)b0.21 ± 0.041.35 ± 0.280.94 ± 0.070.003 ± 0.003 Murayama et al. [2004]
North/central CanadaEast Canada (∼2000 km)0.151.4>0.9xO'Neill et al. [2002]
North AmericaNorth America (n/ad)∼0.940.009 ± 0.003xDubovik et al. [2002]
Quebec, Canadanortheast United States (∼1000 km)0.15–0.3e0.83–1.230.93 ± 0.02 Colarco et al. [2004] and Taubman et al. [2004]
Quebec, Canadanortheast United States (∼1000 km)∼1.240.97 ± 0.03xEck et al. [2003]
North/central Canadanorth/central Canada (≤600 km)∼1.8xMarkham et al. [1997]
North/central Canadaeast Canada (∼30 km)0.131.9xO'Neill et al. [2002]
North AmericaNorth America (<5 min old)0.8–0.95xRadke et al. [1988] and Hobbs et al. [1997]
CanadaCanada (n/ad)∼0.01 Westphal and Toon [1991]

4.2. Single-Scattering Albedo

[56] The mean single-scattering albedo was >0.9 at 532 nm in the forest fire plumes, and indicates moderately absorbing particles. The single-scattering albedo of the Siberian biomass-burning plume that was observed with Raman lidar over Tokyo on 21 May 2003 was similar to the value for the Siberian smoke observed over Leipzig on 29 May 2003 [Murayama et al., 2004]. Wandinger et al. [2002] and Fiebig et al. [2002] found a considerably lower value of 0.78–0.83 at visible wavelengths for an aged Canadian biomass-burning plume. We can merely speculate on the reasons why those particles were much more absorbing than the ones discussed here.

[57] Black carbon, which is the main source for absorbing aerosols is emitted primarily in the hot flaming stage of vegetation fires. As mentioned before, the flaming stage of forest fires is followed by a smoldering phase. During that time more organic carbon than black carbon is emitted [Ward et al., 1996; Ferek et al., 1998]. Temperatures of the fires in the source regions in 1998 may have been higher than in the present case, and thus responsible for higher-absorbing particles. Fromm et al. [2000] noted increased values for particle optical depth in the lower stratosphere in August 1998. Strong fire activity may have partly been responsible for the increased aerosol load, and thus also for a comparably high amount of absorbing material. The amount of low-absorbing organic vapors on existing soot particles may have been lower. Fiebig et al. [2002] noted that a small amount of highly absorbing iron oxide may have been present in the particles observed during LACE 98.

[58] Aircraft observations of smoke from the boreal fires that occurred in central Quebec in summer 2002 showed single-scattering albedos of 0.93 ± 0.02 at 550-nm wavelength approximately 1000 km downwind of the source region [Colarco et al., 2004; Taubman et al., 2004]. Dubovik et al. [2002] reported a value of around 0.94 at 440–670 nm wavelength for North American boreal forest fires. This number was derived from Sun photometer observations. No information on the age of the particle plumes is available. O'Neill et al. [2002] showed that single-scattering albedo at 500 nm was >0.9 at distances around 2000 km downwind of Canadian forest fires. Ferrare et al. [1990] estimated an increase of single-scattering albedo with distance (up to 3000 km downwind) from boreal forest fires on the basis of satellite observations.

[59] On average, lower single-scattering albedos than reported here were obtained for fresh smoke (less than 5 min old) from boreal fires in North America. Numbers varied from 0.8 to 0.95 in the green spectrum of light [Radke et al., 1988; Hobbs et al., 1997]. These numbers are 0.05–0.1 lower than those for young flaming or smoldering fires (less than 4 min old) in the tropical forests of Amazonia observed during SCAR-B [Reid and Hobbs, 1998].

[60] A rather high single-scattering albedo of around 0.97 at 670 nm was reported for aged biomass-burning plumes advected from Mexico to the United States [Kreidenweis et al., 2001]. These column-averaged values were explained by water uptake of the particles, and a lower content of absorbing carbon compared to African or South American biomass-burning particles. Our measurements were characterized by extremely low relative humidities. As mentioned before, the measurements of the Mexican haze layers may have been affected by anthropogenic pollution in the boundary layer.

[61] Single-scattering albedo measured in tropical regions tends to be lower than what we found in our observations. Measurements of biomass-burning aerosols with AERONET Sun photometer in a savanna region in south central Africa in 1997 yielded a single-scattering albedo of 0.82–0.85 at 550 nm wavelength [Eck et al., 2001]. In situ observations of tropical forest fire smoke carried out in the Amazon region during the Large-Scale Biosphere-Atmosphere Experiment in Amazonia (LBA-EUSTACH) in September-October 1999 [Guyon et al., 2003] showed single-scattering albedos of 0.9 ± 0.03 at 545 nm for ambient relative humidities <80%.

[62] Haze from tropical fires in South America and South Africa was characterized by aircraft during the Transport and Atmospheric Chemistry near the Equator–Atlantic (TRACE-A) experiment. Column mean values for the single-scattering were <0.9 at 500 nm during all research flights [Anderson et al., 1996]. This number refers to dry as well as to humidity-corrected particle conditions. There was no significant difference of single-scattering albedo measured during flights over the continents, i.e., close to the source regions, and in the outflow areas along the coastlines of the two continents. However, the aircraft did not follow any air mass over large distances, so that possible aging effects could not be identified.

[63] Aircraft measurements of haze observed during intense fires in Indonesia in 1997 showed a mean single-scattering albedo of 0.93–0.95 at 532 nm, which puts them in the range of numbers reported here. Smoldering peat fires rather likely were the reason for the observed high single-scattering albedos. In contrast, flaming fires in northern parts of Australia showed considerably lower single-scattering albedos of 0.85 [Gras et al., 1999].

[64] Measurements indicate that particle single-scattering albedo increases with age of the pollution plume. Magi et al. [2003] found a mean value of 0.81 ± 0.02 at 550 nm in various areas of southern Africa during the dry biomass-burning season. The aircraft-borne in situ measurements represent the properties in the planetary boundary layer up to 4-km height, and were corrected for relative humidity. Background conditions which most likely were determined by much more aged biomass-burning aerosol showed a considerably larger mean value of 0.89 ± 0.03.

[65] Abel et al. [2003] observed agricultural fires in southern Africa, and found an increase of single-scattering albedo from 0.84 at 550 nm wavelength at the source to 0.9 after a transport time of 5 hours. Changes due to hygroscopic growth could be neglected as the mean relative humidity was approximately 22% within the smoke plumes [Magi and Hobbs, 2003]. Aircraft-based in situ observations were carried out by Reid et al. [1998] during forest fires and cerrado fires in Brazil during SCAR-B. The authors found an increase of single-scattering albedo by 0.06 after a transport time of 2–4 days. The change of single-scattering albedo was likely caused by an increase of scattering material due to the condensation of volatile organics during the particle aging process [Reid et al., 1998]. Condensation of, e.g., organic carbon away from the fire source can produce significant amounts of new particles [Hobbs et al., 2003; Pósfai et al., 2003].

[66] Table 4 summarizes the results for single-scattering albedo of particles generated by fires in boreal regions. In contrast to the result obtained for particle size, i.e., that there seems to be a systematic shift toward larger particle size with increasing distance from fire source, single-scattering albedo does not seem to undergo any systematic change with travel time.

4.3. Complex Refractive Index

[67] Table 3 shows that the mean value of the real part of the complex refractive index varied from 1.37 to 1.6, and the mean imaginary part was always <0.01i. The real parts on average are lower than the values derived from multiwavelength lidar observations of the aged Canadian biomass-burning plume observed during LACE 98. Values were estimated to be 1.56–1.77 at 532 nm [Wandinger et al., 2002]. Imaginary parts in that case varied between 0.04i and 0.07i, which is roughly an order of magnitude higher than what was found in the present study.

[68] Westphal and Toon [1991] report complex refractive indices on the order of 1.45–1.55 for the real part, and 0.01i for the imaginary part at 500 nm wavelength for biomass burning in Canadian forests in July 1982. No information on relative humidity was reported. Dubovik et al. [2002] found 1.5 ± 0.04 for the real part, and 0.009i ± 0.003i for the imaginary part from Sun photometer observations. Taubman et al. [2004] estimated a real part of 1.58 for Canadian fires in 2002, which is about 0.1 larger than what we found on average. Aged biomass-burning plumes advected from Mexico to North America resulted in real parts of 1.41–1.45 at 670 nm [Kreidenweis et al., 2001], which is rather close to our numbers. As mentioned before, high relative humidities and contamination by anthropogenic pollution could be responsible for the low real parts. Imaginary parts were not reported. However, on the basis of the large single-scattering albedo and the comparably large size of the particles it has to be assumed that the imaginary part was well below 0.01i.

[69] Most of the information on complex refractive indices is available for biomass-burning particles in tropical areas of South America and Africa. In general, numbers are larger that what we found in our measurements. A mean refractive index of 1.54–0.018i was derived for aged regional haze observed with aircraft during the Southern African Regional Science Initiative (SAFARI 2000) [Haywood et al., 2003a]. Similar values were found from Sun photometer observations by AERONET in the same area [Haywood et al., 2003b]. Observations during the period of intense biomass burning in South America showed a value of 1.41 ± 0.05 for the real part and 0.013i ± 0.005i for the imaginary part. Measurements were carried out at 545 nm wavelength, and hold for relative humidities from 48 to 80% [Guyon et al., 2003]. Again the imaginary part is much higher than what is reported here. Sun photometer observations carried out by AERONET in Amazonian forests in Brazil (1993–1994) and in Bolivia (1990–1999) showed column mean values of 1.47 ± 0.03 for the real part and 0.0093i ± 0.003i for the imaginary part [Dubovik et al., 2002]. The numbers hold for the wavelength range from 440 to 1020 nm. Biomass burning in South American cerrado in Brazil resulted in values of 1.52 ± 0.01 for the real part and 0.015i ± 0.004i for the imaginary part. Previous observations in these areas with AERONET radiometers during SCAR-B resulted in slightly larger values of 1.53 ± 0.04 for the real part at 440 and 670 nm wavelength. Haze that resulted from intense forest fires in Southeast Asia in 1997 was characterized by real parts <1.55 in the wavelength range from 560 to 870 nm [von Hoyningen-Huene et al., 1999]. Table 4 summarizes once more the results for the imaginary part of particles generated by fires in boreal regions. As in the case of single-scattering albedo it is not possible to derive any conclusion regarding the effect of transport time on the change of the absorption behavior.

5. Summary

[70] Particle backscatter and extinction coefficients measured in the free troposphere over Central Europe (Leipzig, Germany) in spring/summer 2003 were a factor of 4–5 above background values observed in previous years [Mattis et al., 2003]. The most likely reason for this enhancement was a high load with biomass-burning particles that had spread over the Northern Hemisphere. Backward dispersion modeling with FLEXPART showed that intense forest fires in the boreal areas of Siberia probably were the main source of the increased aerosol concentrations in the free troposphere of the Northern Hemisphere [Damoah et al., 2004]. Fires in Canada additionally injected particles into the free troposphere in summer 2003. For the first time such a long-lasting event was documented with an advanced Raman lidar that measures humidity, temperature, and aerosol properties at several wavelengths, so that microphysical particle characterization can be carried out.

[71] Particle optical and microphysical properties of these aged biomass-burning plumes were discussed on the basis of three measurement examples. One case represented Siberian fires, and one case described Canadian fires. In the third case both source regions likely contributed to the observed particle load. Lidar ratios at 532 nm were larger than those at 355 nm. This property of aged biomass-burning particles has already been noted in a study about aerosols from Canadian forest fires [Wandinger et al., 2002] and was also found from dual-wavelength Raman lidar observations over Tokyo in May 2003 [Murayama et al., 2004]. Particle Ångström exponents in these haze layers were comparably large. They varied from 0 to 1.1. Particle effective radii were 0.24–0.41 μm. To our knowledge, particle sizes of 0.41 μm have not been reported to date for aged forest fire smoke in the free troposphere. For comparison, pollution advected from North America on 25 August 2003 showed considerably smaller particles from 0.16 to 0.2 μm, higher Ångström exponents of 1.8–2.1, and a different spectral slope of the lidar ratio, i.e., larger values at 355 nm than at 532 nm. On the basis of these properties we assume that anthropogenic sources were mainly responsible for the lofted particle plume detected on that day. However, fire activity in Canada reached a second maximum in the first half of August, and the observed air masses crossed these areas on their way to Europe. Thus it cannot be completely ruled out that the detected particles also originated from forest fires. In that case there must have been a fundamental difference of physical, chemical, and meteorological processes that lead to the significant difference in particle size.

[72] Relative humidity was low (<50%) in the center of the biomass-burning plumes. Thus hygroscopic growth seemed to be a minor factor responsible for the large particles. Possible reasons for the large particles may have been aging effects, such as particle coagulation and condensation of gases on existing particles, as well as injection of large particles as the result of the kind of burning process. Forest fires are characterized by a protracted smoldering phase. Flaming fires are linked to higher burning temperatures, which may create smaller particles than smoldering fires.

[73] Mean single-scattering albedo from all cases considered here was 0.93 ± 0.03 at 532 nm. Such values are considerably higher than what has usually been observed near the source of fires. Condensation of nonabsorbing gases on existing particles, as well as particle aging effects during the long transport times of the particles of more than 5 days may have been the reason for the comparably large single-scattering albedo.

[74] Table 5 summarizes the results for the microphysical particle parameters in terms of mean values for the selected cases of biomass-burning haze observed on 29 May, 26 June, and 10 July 2003. Also shown are numbers that were previously obtained from six-wavelength lidar observations of lofted plumes during ACE 2 and INDOEX. Effective radii are much larger than those of anthropogenic pollution (ACE 2) or mixtures of anthropogenic pollution with biomass-burning aerosols (INDOEX). Single-scattering albedo falls in between the case of rather nonabsorbing European pollution, and the much more absorbing pollution observed over South and Southeast Asia. The same holds true for the imaginary part of the refractive index. Volume and surface-area concentration are considerably lower than respective numbers found during ACE 2 and INDOEX.

Table 5. Mean Values of Effective Radius, Volume and Surface-Area Concentration, Real and Imaginary Parts of the Complex Refractive Index, and Single-Scattering Albedo of the Biomass-Burning Particles Observed on 29 May, 26 June, and 10 July 2003a
Experimentreff, μmv, μm3cm−3s, μm2cm−3mrealmimagssa
  • a

    For the calculations we considered only the mean values from each height layer. The standard deviation accordingly is a measure of the variability of the individual mean values. For comparison this table also lists mean values that describe lofted plumes of anthropogenic origin observed during ACE 2 [Müller et al., 2002, Table 2] and mixtures of biomass burning and anthropogenic pollution observed during INDOEX [Müller et al., 2003, Table 2]. The meaning of symbols is as in Tables 1 and 4.

Leipzig0.36 ± 0.053.2 ± 2.529 ± 241.47 ± 0.070.004 ± 0.0020.93 ± 0.03
ACE 20.15 ± 0.0613.6 ± 5.3360 ± 2701.56 ± 0.070.009 ± 0.010.95 ± 0.06
INDOEX0.2 ± 0.0821 ± 11350 ± 1901.54 ± 0.110.022 ± 0.0140.9 ± 0.06

Acknowledgments

[75] EARLINET has been funded by the European Commission under grant EVR1-CT-1999-40003. We thank ECMWF for data access.

Ancillary