Geophysical Research Letters

Multiwavelength Raman lidar observations of particle growth during long-range transport of forest-fire smoke in the free troposphere

Authors


Abstract

[1] We present particle effective radii and Ångström exponents of aged free-tropospheric forest-fire smoke. The particle plumes were observed with different multiwavelength Raman lidars downwind of the fires that burned in boreal areas of the northern hemisphere. We find an increase of particle size, respectively decrease of the Ångström exponent with transport time which was more than two weeks in some of the investigated cases. Mean effective radii were as large as 0.4 μm. Mean Ångström exponents were as low as 0.04 for the wavelength pair at 355/532 nm. A fit curve that describes particle growth with time is derived. Particle growth levels off after approximately ten days of transport time.

1. Introduction

[2] Despite the importance of free-tropospheric particles on climate and air quality their geometrical features, and temporally and vertically resolved optical and microphysical properties are only poorly understood. The main difficulty of characterizing particles in the free troposphere is the limitation of observational techniques. Satellite passive sensors and Sun photometers cannot separate between particles in the boundary layer and pollution in the free troposphere. The retrieved parameters rather describe a mixture of particle types. Airborne observations are limited to short periods during field campaigns.

[3] Knowledge on free-tropospheric smoke particles from fires in boreal areas is particularly scarce. Such particles have attained specific interest in recent years. Particle concentrations were significantly enhanced over background conditions during the severe forest-fire season in Canada and Siberia in 2003 [Mattis et al., 2003]. Air quality along the west coast of North America was impaired during that time [Jaffe et al., 2004].

[4] These particles were carried around the globe [Damoah et al., 2004], but only little information is available on particle transformation processes such as increase of particle size during the transcontinental transport. First hints on such an effect may be found in Penndorf [1953]. Satellite observations of Canadian smoke delivered inconclusive results [Ferrare et al., 1990]. Radke et al. [1995] present some evidence on particle growth. The authors investigated smoldering wildfires in the Pacific Northwest of the United States. However, the observations were limited to three days of transport time. Further evidence regarding particle growth is presented by O'Neill et al. [2002] who note an increase of particle size in Canadian forest-fire smoke observed with Sun photometer. Transport distances of the plumes were <2000 km, which is equivalent to transport times of <4–5 days.

[5] A major step forward was made by Fiebig et al. [2003] who used in-situ data of Canadian smoke particles detected by aircraft in the free troposphere over Germany. A particle growth model was applied to reproduce the measured size of the particles which were six days of age at the time of observation. No data were available along the track of the plume. According to the model the median diameter of the measured particle size distributions increased between 20%–80% during transport.

[6] Our study adds further experimental weight to the previous results. Multiwavelength Raman lidar observations of forest-fire smoke over Germany, Spitsbergen, and Japan provide particle size at different distances from the source regions, in part in excess of six days of transport time. In contrast to the column-integrating Sun photometer measurements discussed by Eck et al. [2003], Colarco et al. [2004], and Taubman et al. [2004] our lidar measurements only take account of free-tropospheric pollution. The height-resolved information on aerosol properties from lidar and backwardtrajectory analysis give us rather good indication that the impact of local anthropogenic sources was comparably low. On the basis of these data sets we obtain for the first time a parametrization of particle growth in dependence of transport time for forest-fire smoke in the free troposphere of the northern hemisphere.

2. Methodology

[7] Data were collected with four different multiwavelength Raman lidar systems. The stations were located at Tokyo (35.66°N, 139.8°E) in Japan [Murayama et al., 2004], at Koldewey station (78.55°N, 11.56°E) on Spitsbergen [Ritter et al., 2004], and at Leipzig (51.3°N, 12.4°E) [Mattis et al., 2004] and Lindenberg (52.2°N, 14.1°E) in Germany [Schäfer et al., 1997]. All systems use Nd:YAG lasers for generating laser pulses at 355, 532, and 1064 nm. Technical details of the systems are given in the aforementioned literature.

[8] 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. Profiles of the particle volume backscatter coefficients are derived with the Raman method [Ansmann et al., 1992]. A detailed description of the analysis of optical data collected at the Leipzig site is given by Mattis et al. [2003]. Data analysis for the Tokyo site is described by Murayama et al. [2004]. The data of the Spitsbergen and the Lindenberg sites were analyzed according to the aforementioned methods. The accuracy of the derived backscatter profiles is approximately 20%. The extinction profiles were derived with an accuracy of 10%–30% depending on the lidar system. Ångström exponents Ångström [1964] are calculated from the extinction coefficients for the wavelength pair 355/532 nm. The mean uncertainty determined from all data used in this study is ±0.3.

[9] Particle effective radius was derived with an inversion algorithm [Veselovskii et al., 2002]. The input data were generated from the profiles of the optical properties. We calculated mean coefficients of particle backscatter (355, 532, and 1064 nm) and particle extinction (355 and 532 nm) for height-ranges of the haze layers, respectively. We carried out ten inversions for each optical data set with different realizations of statistical error (10%–20%). So the uncertainty for the retrieval of the particle effective radius could be determined. The results of all ten inversion runs were averaged to yield the final solutions, respectively.

[10] Caution has to be exercised with the Lindenberg data, because signals were only detected at 355 and 532-nm wavelength. It has been shown previously [Veselovskii et al., 2002] that an accurate retrieval of microphysical particle parameters requires a third particle backscatter coefficient. We determined the backscatter coefficient at 1064-nm wavelength on the basis of the Ångström exponent of particle extinction and the lidar ratios measured at 355 and 532 nm. On the basis of these calculations we could roughly estimate particle effective radius.

[11] Source areas of the smoke plumes were identified on the basis of satellite imagery and information from web sites on forest-fire activity, e.g., Canadian Forest Service (available at http://www.nofc.forestry.ca/fire/) and Global Fire Monitoring Center (available at http://www.fire.uni-freiburg.de/). Backward trajectories were calculated with HYSPLIT (Hybrid Single-Particle Lagrangian Integrated Trajectory) [Draxler and Hess, 1998]. Some measurements at Leipzig [Damoah et al., 2004; Müller et al., 2005] were analyzed with the Lagrangian particle dispersion model FLEXPART [Stohl et al., 1998], which provides information on source regions of forest-fire smoke.

[12] We analyzed eleven different particle layers observed on ten individual days of lidar measurements. Backward trajectory and FLEXPART simulations indicated that in one case particles in different height levels had been emitted from different source regions. Each particle layer extended over several hundred meters, respectively. We selected several data sets within each particle layer. In summary we analyzed 47 optical data sets that describe different heights of the eleven particle layers. The individual results for the different particle layers were then averaged into one mean data point for each particle layer. The trajectory in the center of each individual height layer was computed, from which mean travel times for each measurement day were derived. These mean properties are shown in Figure 1. We added three data points from Sun photometer observations in Canada [O‘Neill et al., 2002; Markham et al., 1997].

Figure 1.

Ångström exponents (closed symbols and thin trace from curve fit) and effective radius (open symbols and thick trace from curve fit) of forest-fire smoke versus transport time. The symbols denote the following Raman lidar sites and measurement days: Tokyo (star), 21 May 2003 [Murayama et al., 2004], Lindenberg (hexagon), 21 May 2003, Lindenberg (inverted triangle), 9 August 1998 [Wandinger et al., 2002], Spitsbergen (triangle), 8 May, 11 June, and 23 June 2003, Leipzig (diamond), 29 May, 26 June, and 10 July 2003 [Müller et al., 2005], and 22 July 2004. The error bars denote the uncertainty of distance and travel time from the source regions to the field sites, respectively. Results of Sun photometer observations were taken from O'Neill et al. [2002] (circle), and Markham et al. [1997] (square; effective radii are not available for that case). The wavelength range for which the Ångström exponents were calculated are provided in the cited literature.

3. Results

[13] Figure 1 shows that effective particle radius increases and Ångström exponents decrease with duration of transport. Particle growth seems to level off after approximately 10–15 days of transport time. The data collected at the four lidar stations describe free-tropospheric plumes that originated from Siberia, areas to the north of the Caspian Sea and west of the Ural mountains. The smoke was transported at least 3–4 days before observation, respectively.

[14] The data indicate a change of optical properties with transport time, as shown for instance by the Ångström exponents for the Tokyo and the Leipzig sites, see Figure 1. That finding becomes much clearer if we add the data points obtained with Sun photometers. The data were taken considerably closer to the source regions, i.e., transport time was <4 days. We assume that smoke dominated optical depth, and that the impact of anthropogenic pollution was accordingly low. For this reason we had to reject many Sun photometer observations of forest-fire events that occurred along the east coast of the United States.

[15] A curve describing a first-order exponential decrease of the Ångström exponent, respectively an increase of effective radius with transport time was fitted to the data. The fit parameters are summarized in Table 1. To our knowledge this is the first time that experimental data have been used to parameterize the growth of free-tropospheric smoke particles from boreal areas. Table 1 also lists the same parameters in dependence of distance of the plumes from their presumed sources, which gives us an additional error estimate. We consider the curve fit to travel time slightly more robust than the curve fit to travel distance. The distances were taken from backward trajectories calculations and the respective travel times of Figure 1. We also carried out a curve fit without using the data of the Lindenberg site because of the high uncertainty of those data. The fit parameters remain within the uncertainty reported in Table 1.

Table 1. Parameters Describing Exponential Fit of First Order to Ångström Exponents and Effective Radius Versus Travel Time and Distancea
Correlationåend ± Δåendreff,end ± Δreff,enda ± Δab ± Δbχ2R2
  • a

    Ångström exponents, å(= åend + a expx/b); effective radius, reff( = reff,end + a expx/b; in μm); travel time, x = t (in days); distance, x = d (in km). Parameter a is dimensionless in the case of the fit to the Ångström exponents. In the case of the fit to effective radius parameter a has the unit μm. Parameter b has the unit day (km), if it describes the fit to travel time (travel distance). The table also lists Chi-squared (χ2) and the correlation coefficient (R2).

å vs. t0.38 ± 0.12 2.1 ± 0.34 ± 10.0370.89
å vs. d0.35 ± 0.15 1.5 ± 0.26300 ± 18000.0370.89
reff vs. t 0.36 ± 0.05−0.28 ± 0.067 ± 40.0020.71
reff vs. d 0.40 ± 0.14−0.26 ± 0.1316000 ± 160000.0030.63

4. Discussion

4.1. Uncertainties

[16] Despite the relatively large uncertainties of the fit parameters, we find high correlation of the curve fits. The correlation coefficients are larger for the curve fits to the Ångström exponents than for the fits to the effective radii. That effect results from the lower error of the Ångström exponents, which is determined from extinction coefficients measured directly with Raman lidar. Effective radius, in contrast, follows from the data inversion, which generates additional uncertainties.

[17] Many assumptions in our data analysis, however, are affected with uncertainties that cannot be accurately quantified. The data set is very inhomogeneous. We link information of different observational days, lidar stations, and source regions. According to our information none of the forest-fire plumes has been observed by several lidar stations during its travel.

[18] We neglect particle scavenging. Müller et al. [2005] and references given therein discuss processes that may be responsible for the large particles, which are: (1) the kind of fire, i.e., flaming versus smoldering combustion, which influences the size of the emitted particles; (2) the kind of burned material and its combustion efficiency; (3) hygroscopic growth of the particles; (4) gas-to-particle conversion of organic and inorganic vapors during transport, and condensation of large organic molecules from gas phase in the first few hours of aging; (5) particle growth due to coagulation; and (6) photochemical and cloud-processing mechanisms. An apportionment of the different mechanisms responsible for particle growth is impossible, and we neglect these factors in our analysis.

[19] We assume that the impact of relative humidity on particle growth was of secondary importance as it was comparably low in most of the cases analyzed. Concomitant observations of relative humidity with the Leipzig lidars showed values of 5%–60% in the lofted layers [Wandinger et al., 2002; Müller et al., 2005]. Radiosonde observations of smoke plumes over Spitsbergen showed relative humidity <50%. Relative humidity of 7%–17% was derived for the Lindenberg data shown in Figure 1.

[20] Travel times and distances from the presumed fire sources to the observational sites are affected with considerable uncertainty. For the Sun photometer observations in Canada travel distances were estimated on the basis of source region mentioned in literature [O'Neill et al., 2002; Markham et al., 1997]. Travel times were estimated on the basis of different mean horizontal wind speeds between 2–20 m s−1. For the Leipzig site we could track the plumes on the basis of FLEXPART simulations [Wandinger et al., 2002; Damoah et al., 2004; Müller et al., 2005]. Backward trajectory analysis was used to calculate transport times and distances for the Tokyo, Spitsbergen, and Lindenberg sites.

[21] We do not know the residence time of the individual smoke plumes in their source regions before they were transported off. We added an error of ±20% to the transport time, with the additional constraint that the minimum uncertainty is at least ±2 days for cases in which transport time is at least 4 days.

[22] In summary it is very difficult to combine the different data sets, given the many unknown factors involved in our study. However we find high correlation of the curve fits, which indicates that some source-independent effect may act on particle growth.

4.2. Comparison to Model Studies

[23] We compared our results to the model studies of Fiebig et al. [2003] who used the theory of particle aging processes described by Reid et al. [1998]. Briefly, condensational growth dominates increase of particle size in the first two days after emission of a plume. Thereafter coagulation becomes the dominating process. This behavior can be explained by the reduction of condensable precursor gases inside the emitted plume, after it leaves the source region. Fiebig et al. [2003] considered different scenarios of aging due to particle coagulation. In-situ data on Canadian forest-fire smoke observed after six days of transport, as well as assumptions on particle size at the presumed area of smoke release served as input to the model studies. The authors showed that the median diameter of the measured particle size distributions increased between 20%–80% during transport.

[24] We may draw the following conclusions: (1) The uncertainties of our approach make it impossible for us to favor any of the aging scenarios assumed by Fiebig et al. [2003]; see also the respective conclusions by Fiebig et al. [2003]. (2) Our experimental data indicate that at least in the first 5–6 days of transport particles grow faster than what is implied by the model.

5. Summary

[25] We presented experimental data that allow us to parameterize the growth of free-tropospheric particles which originate from boreal forest fires. Plumes in excess of six days of transport time were observed with multiwavelength Raman in different areas of Earth. The derived growth curve indicates that the speed of particle growth may be higher than what has been obtained from model studies. Our results may be useful for further modeling efforts that deal with particle modification processes in the free troposphere. Our study shows the potential of carrying out coordinated observations with Raman lidar and Sun photometer on the global scale. In the coming forest-fire season we shall carry out coordinated observations of forest-fire smoke transported across EARLINET (European Aerosol Research Lidar Network) Raman lidar stations.

Acknowledgments

[26] David Kaiser acknowledges funding by the German Environmental Foundation (DBU). He thanks F. Immler and O. Schrems for the operational and analytical work conducted in the framework of the measurement campaign MARL@MOL/LITFASS2003 in Lindenberg. The present publication would have been impossible without their support.

Ancillary