Extreme levels of Canadian wildfire smoke in the stratosphere over central Europe – Part 1 : AERONET , MODIS and lidar observations

Light extinction coefficients of 500 Mm−1, about 20 times higher than after the Pinatubo volcanic eruptions in 1991, were observed with lidar in the stratosphere over Leipzig, Germany, on 22 August 2017. A pronounced smoke layer extended from 14–16 km height and was 3–4 km above the local tropopause. Optically dense layers of Canadian wildfire smoke reached central Europe 10 days after injection into the lower stratosphere caused by rather strong pyrocumulonimbus activity over western Canada. The smoke-related aerosol optical thickness (AOT) was close to 1.0 at 532 nm over Leipzig during the noon 5 hours. We present detailed observations of this record-breaking smoke event in a series of two articles. In part 1, we provide an overview of Aerosol Robotic Network (AERONET) sun photometer observations and Moderate Resolution Imaging Spectroradiometer (MODIS) retrievals of AOT and show lidar measurements documenting the aerosol layering and the very high particle extinction coefficients. In part 2 (Haarig et al., 2018), observations with three polarization/Raman lidars are presented, performed at Leipzig after sunset on 22 August to elucidate the optical and microphysical properties of the aged smoke. As 10 shown in this part 1, smoke particles were found throughout the free troposphere (532 nm AOT of 0.3). A pronounced 2–km thick stratospheric smoke layer occurred from 14–16 km height (AOT of 0.6). AERONET and lidar observations indicate peak mass concentrations of 70–100 μg m−3 in the stratosphere around noon and a well-defined (accumulation mode) smoke particle size distribution characterized by a large effective radius of 0.3–0.4 μm and the absence of a particle coarse mode.


Introduction
Exceptionally dense Canadian wildfire smoke layers causing an aerosol optical thickness (AOT) close to 1.0 at 532 nm crossed central Europe (Leipzig, Germany) between 3 and 16 km height on 22 August 2017.Stratospheric particle extinction coefficients at 532 nm reached 500 Mm −1 at 15-16 km height, about 3-4 km above the local tropopause, and were thus 20 times higher than the maximum extinction values observed in the stratosphere over central Europe in the winters of 1991 and 1992 after the strong Mt.Pinatubo eruption in June 1991 (Ansmann et al., 1993(Ansmann et al., , 1997;;Jäger, 2005).Record-breaking intensive fires combined with the formation of exceptionally strong and well organized pyrocumulonimbus clusters in western nucleation and cirrus life cycle via homogeneous freezing (Jensen and Toon, 1992;Sassen et al., 1995;Liu and Penner, 2002).
The smoke (soot) particles are solid and thus can serve as ice-nucleating particles in heterogeneous (deposition) freezing processes (Hoose and Möhler, 2012).The stratospheric smoke particles with similar sizes as the volcanic particles may enter the upper troposphere from above by slow gravitational settling lasting over months or even years.During every fire season, the stratospheric smoke aerosol particle reservoir may be filled with new particles again.The strength of injection is expected to increase during the upcoming decades as a response to changing climate conditions.Thus modeling of the complex influence of smoke on the global radiation field, chemical processes, and cloud and rain formation (hydrological cycle) will probably become more and more important.High quality and trustworthy modeling is however only possible in close connection with observations as presented here.It is especially necessary to monitor fire smoke events in terms of injection heights, smoke burden, microphysical and optical properties, and smoke decay and removal behavior.Our article is a contribution to this effort.
The consideration of smoke aerosols in atmospheric models is however complicated.As will be shown in a follow-up article (Baars et al., 2018a), many of the stratospheric smoke layers were found to ascend with time.This was already noticed by Khaykin et al. (2018).During August 2017 the layers were about 2-4 km above the tropopause, weeks to months later they were mostly observed above 20 km or even above 25 km, and thus more than 10 km above the tropopause.Upward movements of soot containing layers can be the result of heating of the environmental air masses by solar absorption by the soot particles (de Laat et al., 2012).However, smoke AOTs considerably higher than 1.0 are required to force the warmed, soot-containing air masses to rise.Renard et al. (2008) pointed out that soot particles can also reach the middle stratosphere owing to the gravitophotophoresis effect (Rohatschek, 1996;Pueschel et al., 2000;Cheremisin et al., 2005).By absorbing solar and terrestrial radiation, coarse soot particles with diameters D>1 µm can ascend with time without an air-heating (convection) effect, and thus even at rather low AOT<0.05.Favorable atmospheric conditions for a gravito-photophoresis-related upward motion are given around 20 km height which is in agreement with the existence of a black carbon aerosol layer from 18-21 km as reported by Strawa et al. (1999).Renard et al. (2008) found from ballon-borne and satellite observations that soot particles, originating from biomass burning around the globe, contribute to the aerosol population in the height range from 22-30 km at all latitudes.This height range is clearly above the altitudes directly reached by fire plumes in cases of strong pyrocumulonimbus activity.
The ascend and descend behavior of the smoke layers and changes in the optical and microphysical properties of the aged smoke during the second half of 2017 will be illuminated in two further papers which are in preparation (Baars et al., 2018a, b).
Here, we present the observations taken on 22 August 2017 in two articles.In part 1, we discuss Aerosol Robotic Network (AERONET) sun photometer AOT observations at Leipzig and Lindenberg (about 180 km to the northeast of Leipzig), satellite AOT retrievals (MODIS), and lidar observation of aerosol layering and extinction coefficients with focus on the smoke conditions observed during the noon hours of 22 August, when the thickest smoke layers crossed Leipzig.In part 2 (Haarig et al., 2018), we highlight our unique triple-wavelength polarization/Raman lidar observations.For the first time, particle extinction coefficients, depolarization ratios, and extinction-to-backscatter ratios (lidar ratios) were measured at all three important lidar wavelengths of 355, 532 and 1064 nm in aged tropospheric and stratospheric smoke layers.Very different smoke optical and microphysical properties were found in the troposphere and stratosphere.

Polly lidar
On 22 August 2017 a five-channel single-wavelength 532 nm Polly (Portable lidar system) (Althausen et al., 2009;Engelmann et al., 2016;Baars et al., 2016) was run at the Leipzig EARLINET lidar station (51.3 • N, 12.4 • W, 125 m above sea level) and provided continuous time series of aerosol profiles throughout the day.In addition, a triple-wavelength Polly of the Leibniz Institute for Tropospheric Research (TROPOS) was continuously measuring at Kosetice (49.6 • N, 15.1 • E, 500 m above sea level), Czech Republic.This lidar was involved in a three-month field campaign.Kosetice is located 275 km southeast of Leipzig.During the smoke event northwesterly winds prevailed in the stratosphere and the air masses crossed Leipzig about four hours before reaching Kosetice.
The so-called Fernald method (Fernald, 1984) was used to derive height profiles of particle extinction coefficient from the lidar observations at daytime.The reference height was set around 10-11 km height (tropopause region) on 22 August and as input lidar ratio we used a value of 70 sr for 532 nm.This lidar ratio of 70 sr was measured with our Raman lidars after sunset on 22 August 2017 (see part 2).To reduce the influence of signal noise the signal profiles have to be smoothed.We used vertical gliding-averaging window length of 185 m in the boundary layer (up to 2.5 km height) and 750 m (above the boundary layer up to 16 km height).The sensitivity tests with different smoothing lengths of 175 m, 350 m, and 750 m in the free troposphere revealed that the main layering features are well resolved by using the comparably large vertical window length of 750 m.The large smoothing length was necessary because the densest smoke layers crossed the lidar at Leipzig during the noon hours when the signal noise by sunlight was highest.
Temperature and pressure profiles are required in the lidar data analysis to correct for Rayleigh extinction and backscattering.This information is taken from the GDAS (Global Data Assimilation System) data base which contains profiles of temperature and pressure from the National Weather Service's National Centers for Environmental Prediction (NCEP) (GDAS, 2018) with a horizontal resolution of 1 • .We ignore a minor ozone absorption effect at 532 and 607 nm in the determination of smoke extinction coefficient.The neglect is of the order of a few percent.
In Sect.3, we will also show height-time displays of the volume depolarization ratio.This quantity is defined as the ratio of cross-to-co-polarized backscatter coefficient.Co and cross denote the planes of polarization (for which the receiver channels are sensitive) parallel and orthogonal to the plane of linear polarization of the transmitted laser pulses, respectively.The volume depolarization ratio enables us to identify non-shperical particles such as ice crystals and irregularly shaped smoke particles.
The volume depolarization ratio is comparably high when the particles are non-spherical in shape and very low (almost zero) if the particles are spherical such as soot particles with a liquid shell.

AERONET sun/skyphotometer
The EARLINET station at Leipzig is collocated with an Aerosol Robotic Network site (Holben et al., 1998).The AERONET sun/sky photometer is operated since 2001 and measures AOT at eight wavelengths from 339 to 1638 nm (AERONET, 2018).
Sky radiance observations at four wavelengths complete the AERONET observations.From the spectral AOT distribution for the wavelength range from 440 to 870 nm the wavelength dependence of AOT expressed in terms of the Ångström exponent AE is obtained.Furthermore, the fine mode fraction FMF (fraction of fine-mode AOT to total AOT), and particle size distribution for the entire vertical column can be derived (O'Neill et al., 2003).Fine mode particles have per definition a diameter of ≤1 µm.
The full set of AOT and sky radiance measurements are used to derive the particle size distribution (Dubovik et al., 2006).
We further included AERONET observations at Lindenberg, Germany, about 180 km northeast of Leipzig in our studies.
The size distribution of the smoke particles could be derived from the Leipzig AERONET observations in the early morning of 22 August and at Lindenberg in the early morning hours of 23 August 2017.

MODIS
MODIS aboard the NASA Terra and Aqua satellites have been in operation since 2000 and 2002, respectively, providing retrieval products of aerosol and cloud properties with nearly daily global coverage (Remer et al., 2013).The MODIS dark target algorithms over land (with spatial resolution of 10 km and 3 km) perform a simultaneous inversion of the measured top of the atmosphere reflectance in three channels (centered at wavelengths of 466, 645, and 2113 nm) to retrieve total spectral AOT, fine model weighting parameter, and surface reflectance at 2113 nm (Remer et al., 2013) from the MODIS data.The most recently released MODIS Collection 6 product MOD04_3K (for Terra) and MYD04_3K (for Aqua) contains AOT at a 3 km horizontal resolution in addition to the L2 10 km product (Remer et al., 2013;Levy et al., 2015).The retrieval algorithm of the higher resolution product is similar to that of the 10 km standard product with several exceptions (for more details, see http://modis-atmos.gsfc.nasa.gov/MOD04_L2).Validation against surface sun photometer shows that two-thirds of the 3 km retrievals fall within the expected error on a regional comparison but with a high bias of 0.06 especially over urban surfaces.
The uncertainty in the retrieved AOT is 0.05±0.15×AOT for AOT≤1.0 (Levy et al., 2010(Levy et al., , 2013)).In our study (Sect.3), we used the MODIS Collection 6 (C006) AOT retrievals at 3 km×3 km (at nadir) spatial resolution collected with Terra (10:30 local equatorial crossing time) and Aqua (13:30 local equatorial crossing time) over Leipzig on 22 August 2017 (MODIS, 2018).We show AOT for 500 nm.The wind velocity decreased with height from the tropopause to 16 km height (GDAS, 2018) which may explain the apparent upward motion of the stratospheric layer.Wind shear may have tilted the smoke plume so that the base of the extended smoke field was detected with lidar first and the upper part of the plume about one day later.The volume depolarization ratio is highest in the cirrus clouds (containing strongly light-depolarizing hexagonal ice crystals) and is also significantly enhanced in the stratospheric smoke layer (caused by irregularly shaped smoke particles).The tropospheric smoke-related volume depolarization ratios were comparably small, indicating the dominance of spherical (compact) particles.The different light-depolarizing features of the tropospheric and stratospheric smoke particles will be discussed in detail in part 2 (Haarig et al., 2018).
The HYSPLIT backward trajectories (Stein et al., 2015;Rolph et al., 2017;HYSPLIT, 2018) in Fig. 4 provide an impression of the air flow during the last 10 days (12-21 August 2017) for heights in the middle and upper troposphere.Strong convective processes as required to lift smoke up to the upper troposphere and lower stratosphere can not be resolved by global circulation models.So, the HYSPLIT trajectories should be interpreted with caution when pyrocumulonimbus activities come into play and influence the vertical smoke exchange and long-range transport.According to the backward trajectories, the smoke traveled about 7-10 days at tropospheric as well as stratospheric height from western Canada to central Europe.This is in good agreement with the travel time derived from the spaceborne lidar observation presented by Khaykin et al. (2018).The AOT values derived from the Polly lidar observations after sunset (20:40-23:00 UTC) are also in good agreement with the AERONET observations.The Leipzig and Lindenberg AERONET stations measured smoke-related AOTs of about 0.4 just before sunset (Leipzig) and after sunrise on the next morning at 6:00 UTC (Lindenberg, not shown).The two lidar AOT values for the 11:00-12:00 UTC period were computed by using a low, but not unrealistic smoke lidar ratio of 50 sr in the lidar retrieval and the lidar ratio of 70 sr as measured with our Raman lidars after sunset on 22 August 2017 (see part 2).
The boundary-layer 500 nm AOT was around 0.1-0.15 on 21 and 23 August 2017 (before and after the smoke period) and about 0.15-0.2 on 22 August as the lidar observations at Leipzig and Kosetice indicated.Thus, the fire smoke layers caused a 532 nm AOT close to 1.0 during the noon hours of 22 August 2017.Satellite (MODIS) observations of AOT corroborate the lidar and sun photometer measurements.The cloud fields in Fig. 6a provide an impression of the cumulus cloud distribution in the morning of 22 August 2017 (10:15 local time) which hampered the AERONET observations and the MODIS retrieval efforts.Only a few AOT values for 3 km×3 km pixels could be retrieved from the MODIS observations.However, these few AOT values in Figs.6a and 6b clearly point to AOT values of the order of 1.0 at 500 nm in the Leipzig area, north and south of the Leipzig AERONET station.Northwesterly winds prevailed on this day in the free troposphere and lower stratosphere.
Figure 7 shows the lidar profiles for the 11:00-12:00 UTC period and the nighttime hours.At noon, the entire free troposphere showed traces of smoke aerosol.The 532 nm AOT was 0.3 in the free troposphere (from about 2.5 km height up to the tropopause) and 0.6 in the stratosphere for the height range from the tropopause up to 16 km height.The smoke-related AOT was significantly lower in the evening hours (blue curve) with a free tropospheric contribution of 0.08 and a stratospheric contribution of 0.2-0.25.
The high stratospheric AOT of 0.6 over Leipzig is in good agreement with CALIOP measurements (Khaykin et al., 2018).
The maximum stratospheric smoke AOT measured with CALIOP was of the order of 1.0 at 532 nm.These values occurred over northeastern Canada on 17-19 August 2017 and thus a few days upstream of central Europe.Khaykin et al. (2018) reported maximum AOTs of 0.7, but these values were directly estimated from the height profiles of the attenuated backscatter coefficients and were not corrected for particle extinction influences.If we take smoke extinction (according to an AOT of the order of 0.7-1.0)into account, the true profile of the particle backscatter coefficient multiplied by a smoke lidar ratio of 70 sr leads to an AOT about a factor of 1.5 higher than the apparent one given by Khaykin et al. (2018) and thus to values of the order of 1.0.
Figure 8 shows AERONET and lidar inversion products.The AERONET size distributions were downloaded from the AERONET data base (AERONET, 2018).Details to the lidar data inversion are given in part 2 (Haarig et al., 2018).According to the AERONET size distributions, the smoke-dominated aerosol column showed a pronounced accumulation mode centered at a radius of 200-300 nm at the beginning of the smoke period (Leipzig, 22 August, 5:48 UTC) and at 300-400 nm at the end of the smoke event (Lindenberg, 23 August, 5:42 UTC).The size distribution was clearly shifted to larger sizes compared to The 500 nm AOT rapidly decreased over Lindenberg from 0.4 (at 6 UTC) to about 0.1 (at 8 UTC).In the late afternoon of 23 August, anthropogenic haze dominated.The dense smoke layers had left Germany.The effective radius decreased to typical values for anthropogenic pollution (r eff =0.24 µm, r eff,f =0.14 µm).
The lidar-derived size distribution (from the nighttime measurements) fits very well into the AERONET observations at Lindenberg in the early morning of 23 August.The effective radius r eff of the smoke particles in the stratospheric layer was 0.32 µm and thus very close to the Lindenberg values.To obtain a column-integrated quantity that can be comparable with the AERONET observations we multiplied the size-resolved volume concentrations by a vertical extent of the stratospheric layer of 1000 m (see Fig. 7).This volume size distribution for the stratospheric smoke layer (column) is shown as black curve in Fig. 8.
Further lidar inversion products are given in Fig. 8 for the evening smoke conditions.Because the peak values of the particle extinction coefficient at 532 nm was a factor of 2 higher during the noon hours, peak mass concentrations were in the range of 70-100 µg m −3 in the stratosphere around noon of 22 August 2017.The retrieved single scatter albedo (see part 2 for more explanations concerning the retrieval) of 0.8 at 532 nm clearly indicates strongly light-absorbing smoke (soot) particles in the stratosphere.
Similar size distributions with a pronounced smoke accumulation mode shifted to larger sizes were found during several airborne in situ measurements of North Amercian wildfire plumes in the European region (Fiebig et al., 2002;Petzold et al., 2007;Dahlkötter et al., 2014).All these airborne in situ observations indicated that super micrometer particles (coarse-mode particles) were almost absent in the aged smoke plumes and that the accumulation mode was enhanced and shifted towards larger mode diameters.
After the illumination of the optical and size-distribution aspects, Fig. 9 finally deals with the shape properties of the stratospheric smoke particles.As will be explained in detail in part 2 a surprisingly strong wavelength dependence of the particle linear depolarization ratio was observed with lidar in the stratosphere.As mentioned above, the particle linear depolarization ratio is rather sensitive to changes in shape properties.The depolarization ratio is almost zero for spherical particles and significantly enhanced for non-spherical, irregularly shaped particles.On 22 August 2017, the stratospheric smoke depolarization ratio was highest with 22% at 355 nm and lowest with 4% at 1064 nm, about a factor of 4-5 lower.We hypothesize that this strong wavelength dependence is the consequence of a missing particle coarse mode.As shown in Fig. 9, a similar wavelength dependence is also observable in the case of non-spherical desert dust particles in the absence of a dominating coarse mode (Järvinen et al., 2016;Mamouri and Ansmann, 2017).Especially, the 355 nm and 532 nm particle depolarization ratios for stratospheric smoke and fine-mode dust particles are rather similar.
Extreme levels of Canadian fire smoke were observed in the stratosphere on 22 August 2017.Extinction coefficients reached values of 500 Mm −1 and were thus about a factor of 20 higher than maximum extinction values found over central Europe after the Pinatubo eruption.Such high stratospheric extinction coefficients were never observed over Leipzig before and are related to the record-breaking fire season in western Canada in 2017.A pronounced stratospheric smoke layer extended from 14-16 km height and was about 3-4 km above the local tropopause.We analyzed AERONET, MODIS and lidar observations to document this record-breaking stratospheric smoke event.Maximum smoke-related AOTs were close to 1.0 and the stratospheric particle mass concentrations reached 70-100 µg m −3 .The smoke particles formed a well-defined aged accumulation mode characterized by a large effective radius of 0.3-0.4µm.
This extreme case presented here demonstrates that significant amounts of wildfire smoke can reach the stratosphere and can significantly disturb the stratospheric aerosol conditions (Khaykin et al., 2018) and may thus have a sensitive influence on chemical processes, radiative fluxes, and even heterogeneous ice formation in the upper troposphere and this over months after each summer fire season.The black carbon aerosol partly enriches the natural soot particle reservoir between 20-30 km by upward motions, probably caused by gravito-photophoresis forces (Renard et al., 2008).
We will continue and deepening the characterization of the smoke optical and microphysical properties in the troposphere and stratosphere in part 2 (Haarig et al., 2018).Very different smoke optical and microphysical properties were found in the troposphere and stratosphere.Based on the results of part 1 and part 2, an extended conclusion section is given in part 2.

Figure 2
Figure2shows four photographs taken in the evening of 22 August 2017, about 250 km northwest and thus upstream of Leipzig.Unusually long filament-like structures over the entire horizon could be observed by eye.Cirrus features are very different and much more heterogeneous.Similar indications for the presence of a massive stratospheric aerosol layer and a coherent large-scale aerosol transport in the stratosphere were observable in the first winter after the Mt.Pinatubo eruption,

Figure 3
Figure 3 provides an overview of aerosol layering over central Europe from 20-23 August 2017.We used the Kosetice observations because this lidar was continuously running on these days, while the Polly at Leipzig was sporadically operated only, e.g., not continuously on 21 and 23 August 2013.As can be seen, large coherent smoke structures were observed in the troposphere as well as in the stratosphere over more than one day.Cirrus clouds developed in the smoky environment on 20 and 21 August 2017.Unfortunately, boundary layer clouds (below about 2-3 km height) disturbed aerosol and cloud profiling considerably during the daytime periods.The tropopause height was mainly between 10 and 11.5 km from 20-23 August.

Figures 5
Figures 5 shows the Leipzig AERONET observations from 21-23 August 2017.Level 1.0 data are presented.In the case of Level 1.5 data, too many valid observations are removed by the automated AERONET cloud screening procedure.We performed a careful cloud screening of the Level 1.0 data by checking the Ångström exponent and the 1640 nm AOT which sensitively varies with the occurrence of cirrus.In this way we avoided that the automated AERONET cloud screening analysis removed, e.g., the observations of the maximum AOT of 1.1 around 11 UTC on 22 August 2017.The respective Ångström exponent was 1.2.The lidar observations (diamonds in Fig.5) conducted between 11:00-12:00 UTC are in good agreement with the extraordinarily high 500 nm AOT.According to our lidar observations, cirrus clouds were absent during the noon hours of 22 August 2017 so that the shown AERONET smoke observations were not affected by any cloud occurrence.As can be seen in Fig.5, FMF increased from values below 0.8 to close to 1 when the smoke layers dominated from noon on 21 August to the evening of 22 August.Accordingly the coarse-particle-related AOT c was almost zero, and the total AOT was almost equal to the fine-mode AOT f .This indicates that coarse particles were almost absent.The Ångström exponent (for the the fine mode caused by central European anthropogenic pollution as observed over Lindenberg during smoke-free conditions Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-357Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 6 April 2018 c Author(s) 2018.CC BY 4.0 License. in the afternoon of 23 August 2017 (blue curve in Fig. 8).The effective radius r eff of the particles was comparably large with values of 0.33-0.42µm and the fine-mode-related effective radii r eff,f of 0.23-0.32µm.Coarse mode particles were almost absent.The remaining weak impact of coarse particles on the volume size distribution for the entire vertical column is probably related to the occurrence of soil and road dust in the boundary layer.The smoke layers vanished in the morning of 23 August.

Figure 7 .
Figure 7. Height profile of particle extinction coefficient at 532 nm over Leipzig on 22 August 2017 measured with Polly close to noon(red profile), when the optically densest stratospheric smoke layers crossed the lidar site, and at nighttime (blue profile).The nighttime observations with three lidars will be discussed in part 2(Haarig et al., 2018).The Fernald method was applied to compute the extinction profiles.An input lidar ratio of 70 sr (as measured after sunset, see part 2) was used.The AOTs for the free troposphere and lower stratosphere are given as numbers.

Figure 8 .
Figure 8. Particle volume size distributions derived from AERONET sun photometer observations at Leipzig (red) and Lindenberg (blue, green).Effective radii (r eff for the total size distribution, r eff,f for the pronounced fine or accumulation mode) indicate that the aged smoke particles were comparably large.Accumulation-mode particles dominated the optical properties.The volume size distribution is considerably shifted towards larger sizes compared to the pure anthropogenic fine mode (in blue).The lidar-derived size distribution (in black) for the stratospheric layer alone agrees well with the AERONET observations and indicates that the pronounced accumutaion mode is caused by smoke particles.The retrieval of the size distribution and listed microphysical smoke values for the 15-16 km layer are discussed in the text and in part 2. Mean values of the particle number concentration N , volume concentration V , mass concentration m, effective radius r eff , and single scattering albedo SSA for the stratospheric layer are given as numbers.