Particles originating from biomass burning activities are known to have a significant effect on radiation and climate (Kaufman et al., 2002). The factors affecting the optical properties of smoke are mainly the black carbon fraction and the impact of the ageing processes (Amiridis et al., 2009). Various findings from field measurements suggest that the smoke particles' surface may serve as highly effective cloud nuclei (Ackerman, 2000; Hoose and Möhler, 2012; Koch and Del Genio, 2010; Marinou et al., 2019; Nichman et al., 2019), modifying cloud properties and lifetime and thus indirectly affecting the radiative budget. Their various impacts also depend on their lifetime, since they tend to alternate their properties, i.e. become less absorbing or more hydrophilic due to atmospheric processes (Amiridis et al., 2009; Adachi and Buseck, 2011).

Smoke particles in the atmosphere can be identified with lidar measurements which provide valuable information on the optical properties of aerosols, such as the depolarization of the backscattered light in terms of the particle linear depolarization ratio (PLDR). Spherical particles do not depolarize the incident radiation; hence, the PLDR can be used to derive information on morphologically complex particles such as smoke. Fresh smoke tends to form fluffy, mostly hydrophobic aggregates composed of many small single monomers. As the particles age in the atmosphere, this aggregate structure collapse; the particles become more hydrophilic and are frequently found covered by cells composed of water-soluble components such as sulfates or organic materials (Worringen et al., 2008; Wu et al., 2016). Due to the aforementioned processes, the PLDR of smoke particles may present a large variability related to the age of the particles (Baars et al., 2019), the presence of other aerosol types found inside the smoke layers (Tesche et al., 2009; Groß et al., 2011, 2013) or even the particle water uptake due to different humidity conditions (Cheng et al., 2014). These processes alter smoke particle shape, size and composition, resulting in PLDR values that may vary from 2 % to 10 % at 532 nm for aged and fresh smoke. These values can be even lower/higher in cases of mixtures with low/high depolarizing components, respectively (i.e. marine/dust particles) Müller et al. (2005) carried out an extensive study on the optical properties and the effect of atmospheric ageing of long-range-transported smoke from Siberia and Canada and found that PLDR at 532 nm did not exceed 1 %–3 % for 10 d old plumes. This is comparable to findings by Nicolae et al. (2013), showing that smoke plumes up to 4 d old present PLDR values of almost 4 % at 532 nm. Moreover, measurements conducted in South Africa (Giannakaki et al., 2016) showed that for pure smoke the PLDR values at 355 nm are less than 6 %. On the other hand, smoke PLDR has been found to reach values up to 12 %–14 % at 532 nm if significant concentrations of highly depolarizing components (i.e. soil or dust particles) exist inside lofted smoke layers (Tesche et al., 2009; Veselovskii et al., 2016).

Lately, there has been observational evidence of smoke originating from large-scale fires with PLDR values that exceed the typical range. For example, in Sugimoto et al. (2010), values of 12 %–15 % at 532 nm are presented for both tropospheric and stratospheric smoke plumes reaching from Mongolia to Nagasaki and Tsukuba in 2007. Nisantzi et al. (2014) reported values of 9 %–18 % at 532 nm for smoke originating from Turkish fires and observed above Cyprus after 1–4 d of transport. A spectral dependence of smoke PLDR with decreasing values from UV to near-IR wavelength was presented for the first time by Burton et al. (2015). The measurements were performed above Denver, Colorado, with an airborne High Spectral Resolution Lidar (HSRL) instrument during the DISCOVER-AQ (Deriving Information on Surface Conditions from Column and Vertically Resolved Observations Relevant to Air Quality) field mission. This particular smoke plume was found at 8 km height, originating from Pacific Northwest wildfires, and exhibited PLDR values of 20 %, 9.3 % and 1.8 % at 355, 532 and 1064 nm, respectively.

In the past, many studies have used simpler or more complicated particle shape models in order to reproduce the lidar measurements of smoke. In Kahnert (2017), the PLDR of black carbon aggregates covered by a coating of sulfates was simulated by two different models: a closed cell model (i.e. each monomer in the aggregate is coated separately) and a coated aggregate model (i.e. the whole aggregate is coated). Their analysis showed that for thicker coating the coated cell model of volume-equivalent radius of 0.3 to 0.4 µm can provide PLDR values of the order of 15 % at 532 nm. Mishchenko et al. (2016) and Liu and Mishchenko (2018) used rather complex morphologies for smoke particles in order to reproduce the PLDR values measured by Burton et al. (2015). Amongst others, these morphologies included (a) a fractal aggregate partially embedded in a spherical sulfate cell, (b) two externally mixed spherical sulfate cells, each hosting an aggregate (models 6 and 11 in Fig. 1 in Liu and Mishchenko (2018) and (c) a high-density aspherical soot core, encapsulated in a circumscribing spheroid cell (with axial ratio of 0.9 to 1.2; model 4 in Fig. 2 in Mishchenko et al., 2016). All these morphologies reproduced successfully the smoke optical properties measured by Burton et al. (2015). Moreover, Luo et al. (2018) used 20 different configurations of coated fractal aggregates and showed that for a relatively small fractal dimension (i.e. relatively fresh aggregates), and for small black carbon fractions (i.e. densely coated aggregates; configuration C in Fig. 2 in Luo et al., 2018), the PLDR values can reach up to 40 %, 15 % and 6 % at 355, 532 and 1064 nm, respectively. Ishimoto et al. (2019) used fractal aggregates and artificial surface tension induced on the particles to mimic the effect of coating by water-soluble materials forming around the particles. This study present results for both the PLDR and the lidar ratio (LR), which is indicative of the composition of the particles. In Liu and Mishchenko (2019), tar-ball aggregates were used to model exceptionally strong PLDR as those measured by Burton et al. (2015). The aforementioned studies highlighted the fact that in order to reproduce significant PLDR values (higher than 20 % at 532 nm), the fractals need to be coated (i.e. shapes of “type-B, size 11, Vr=20” shown in Fig. 4 of Ishimoto et al., 2019). We should point out though that most of the aforementioned studies refer to monodispersed particles, and averaging over size could possibly suppress some of the observed features.

In the spotlight of the large-scale Canadian fires of 2017, the discussion regarding the high PLDR values and their spectral dependence for smoke has been opened also for stratospheric smoke. These wildfires inserted large amounts of smoke to the lower stratosphere by explosive pyrocumulonimbus activity (Khaykin et al., 2018). In fact, the smoke load in the stratosphere was found to be comparable to that of a moderate volcanic eruption (Peterson et al., 2018). The smoke plumes encircled the Northern Hemisphere in nearly 20 d, reaching Europe in less than 10 d. Above Europe, their properties were intensively studied by the European Aerosol Research Lidar Network (EARLINET; Pappalardo et al., 2014). Multi-wavelength lidar measurements in central (Ansmann et al., 2018; Haarig et al., 2018; Hu et al., 2019) and southern Europe (Sicard et al., 2019) revealed high PLDR values at 355 and 532 nm and a strong spectral dependence from the UV to the near-IR wavelength. However, despite the extensive analysis of this event, the microphysical characterization of the stratospheric smoke particles is not yet adequate and further analysis is imperative to draw conclusions. Most of the microphysical properties reported for the stratosphere are retrieved from lidar measurements using inversion algorithms and assumed scattering models that are applied in EARLINET (e.g. Dubovik et al., 2006; Veselovskii et al., 2002). For example, the derived microphysical properties presented in Haarig et al. (2018) and Hu et al. (2019) are based on the lidar backscatter and extinction coefficient profiles that were used as inputs to inversion schemes. However, the observed PLDR values could not be reproduced by these studies due to the assumed shapes.

In contrast to prior studies, for our investigation for the stratospheric smoke originating from the Canadian wildfires, we do not adopt morphologically complex shapes of bare or coated smoke aggregates, which are associated with excessive computations. Instead, we propose a much simpler model of compact near-spherical particles. Our starting point and main assumption is that the particle near-spherical shape can be highly depolarizing, as shown in the work of Mishchenko and Hovenier (1995) and Bi et al. (2018). Our analysis shows that for the Canadian stratospheric smoke observed above Europe in August 2017, the PLDR and LR measurements along with their spectral dependence can be successfully reproduced with the proposed model of compact near-spherical particles. The size and refractive index of the particles are estimated as well and seem to agree well with past observations for aged smoke. We further examine the capability of this model to be used on an operational level and in particular as an extension to the Aerosol Robotic Network (AERONET) operational aerosol retrieval (Dubovik et al., 2006), since it provides a much simpler and faster solution with respect to more complicated shapes for stratospheric smoke particles (e.g. Mishchenko et al., 2016; Ishimoto et al., 2019).

Our paper is organized as follows: in Sect. 2, we discuss the methodology followed for the retrieval of the microphysical properties of stratospheric smoke by constructing look-up tables of PLDR and LR at 355, 532 and 1064 nm, assuming (a) near-spherical shapes and (b) more complicated Chebyshev particle shapes. In Sect. 3, we provide a brief description of the Canadian wildfires during August 2017, describing the mechanism that introduced the smoke particles into the lower stratosphere and the route of the smoke plume from Canada to Europe. The lidar measurements performed over Leipzig, Germany, are presented in this section. In Sect. 4, we provide the results of our microphysical retrieval. The discussion of these results and the future perspectives of our work are found in Sect. 5. Conclusions are summarized in Sect. 6.

For the retrieval of the smoke microphysical properties from the measured PLDR and LR at 355, 532 and 1064 nm, we constructed appropriate look-up tables using near-spherical shapes and more complicated shapes (i.e. Chebyshev particles), along with a range of size distributions and refractive indices based on values reported in the literature for smoke particles (Dubovik et al., 2002; Müller, 2005; Müller et al., 2007a; Nicolae et al., 2013; Giannakaki et al., 2016). For the construction of the look-up tables, we used the T-matrix code (Mackowski and Mishchenko, 1996; Mishchenko and Travis, 1998). The T-matrix outputs are used to calculate PLDR and LR as shown in Eqs. (1) and (2): 1PLDR(λ)=P11(180∘)-P22(180∘)P11(180∘)+P22(180∘)2LR(λ)=4πCext(λ)Csca(λ)P11(180∘), where Pij are the elements of the scattering matrix, Cext and Csca are the extinction and scattering cross sections, and λ is the wavelength (Fig. 1).

The extreme pyroconvection (Fromm et al., 2010) that was recorded in the area of British Columbia (western Canada) during summer 2017, resulted in particularly strong updrafts that penetrated and released large amounts of smoke particles into the lower stratosphere (Peterson et al., 2018). Here, we use an ensemble of satellite observations from MODIS (Moderate Resolution Imaging Spectroradiometer) aboard Terra and Aqua, OMPS (Ozone Mapping and Profiler Suite) aboard the Suomi National Polar-orbiting Partnership (Suomi-NPP) and CALIOP (Cloud-Aerosol Lidar with Orthogonal Polarization) aboard the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) to identify the dispersion and vertical distribution of the plume above Canada. The combination of these observations is shown in Fig. 4, where True Color images from MODIS are overlaid with the fire active regions and thermal anomalies (red dots) from Suomi-NPP, and CALIPSO (green lines) overpasses on 8 and 15 August 2017.

Figures 5 and 6 show the backscatter coefficient and PLDR curtain plots at 532 nm from CALIPSO measurements. Based on these observations, smoke plumes were found above the regions of fire activity starting from the beginning of August (Fig. 4a), when the plumes remained in the troposphere, below 5–6 km (39–45∘ N, 123–125∘ W) (Fig. 5a, dashed red lines), exhibiting low PLDR values of the order of 3 %–4 % (Fig. 5b) at 532 nm. Then, on 12 August 2017, the unprecedented buoyancy force caused by the strong fire activity started lifting the plumes up towards the tropopause, while already on 15 August 2017 smoke covered a large part of northern Canada (Fig. 4b). CALIPSO observations on 15 August reveal that the plume lies in the stratosphere at 11–14 km height (63–69∘ N, 89–94∘ W) (Fig. 6a, dashed red lines) and PLDR values exceed 15 % at 532 nm (Fig. 6b).

Due to the altitude of the smoke plume, one could attribute such PLDR values to the beginning of ice formation. Indeed, radiosonde temperature profiles from two stations located underneath the smoke plume (green stars in Fig. 4b) reveal that the temperature above 11 km drops below -40 ∘C, at which point homogeneous ice formation can occur (Wallace and Hobbs, 2006). However, the PLDR values of cirrus clouds are usually no less than 40 % at 532 nm (Chen et al., 2002; Noel et al., 2002; Voudouri et al., 2020), whereas the values observed in this case are mostly between 15 % and 25 % at 532 nm, and remain so during the months of August and September following the stratospheric injection (Baars et al., 2019; Hu et al., 2019). Further analysis of CALIOP data provides a mean (median) value of the backscatter-related Ångström exponent (BAE) at 532/1064 nm of 0.9 (0.9) with a standard deviation of 1.07. For cirrus clouds, BAE values close to zero are expected, although, as indicated by the large standard deviation, CALIPSO data are highly noisy at these altitudes. A recent study by Yu et al. (2019) also showed that the largest fraction of stratospheric smoke particles consisted of organic carbon (98 % compared to 2 % for black carbon). Particles of such high organic carbon content serve poorly as ice nuclei (Kanji et al., 2017; Phillips et al., 2013). Although the possibility of small ice crystals formed inside the smoke layers cannot be excluded (largely due to the absence of in situ measurements), the aforementioned characteristics indicate that this plume consists primarily of smoke particles rather than ice crystals.

Inside the lower stratosphere, unaffected by the intensive tropospheric interactions, smoke particles started drifting, following a northeasterly direction and first appeared over Europe approximately after mid-August (Khaykin et al., 2018; Ansmann et al., 2018; Hu et al., 2019).

Interestingly, even after 2 months of the initial stratospheric smoke injection, the plume seems to have sustained its high depolarization capability. During this period, the smoke plume has already encircled the Northern Hemisphere and it was detected by airborne lidar measurements performed above the Atlantic near the west coast of Ireland (Fig. 7b). Lidar observations showed PLDR values in the range of 10 %–14 % at 532 nm between 10 and 12 km (Fig. 7a). These observations were conducted in the framework of Wave-driven ISentropic Exchange (WISE) mission organized by the German Aerospace Centre (DLR) and support the high depolarization values detected for months over Europe by EARLINET, as shown in Fig. 7 in Baars et al. (2019).

The unique optical properties of transported stratospheric smoke, originating from the pyrocumulonimbus activity of the large Canadian fires in 2017, were reproduced using T-matrix simulations and assuming near-spherical shapes for smoke. This is consistent with results of past studies showing that near-spherical particles produce PLDR values that can reach up to 100 % depending also on their size and composition (Bi et al., 2018) and that smoke particles in particular, when heavily coated or even encapsulated with weakly absorbing materials, can produce large depolarization with a noticeable spectral dependence (Mishchenko et al., 2016; Ishimoto et al., 2019). As a next step, we examined whether the AERONET retrieval could possibly be benefited by taking into account the near-spherical shape for stratospheric smoke. Sun-photometer measurements from Lindenberg and Punta Arenas revealed that for the current algorithm configuration, AERONET retrievals for stratospheric smoke cases are associated with high residual errors (higher than 5 %) and are eventually rejected. The extension of the AERONET scattering model to include the near-spherical shapes could possibly improve the retrieval for these cases that seem to become frequent. Our analysis does not intend to generalize on the performance of the AERONET retrieval on tropospheric biomass burning cases. It is focused on the stratospheric smoke cases related to pyrocumulonimbus activity.

In conclusion, studying the stratospheric smoke from the Canadian wildfire activity provided us with the great opportunity to show the potential of remote sensing measurements in investigating and deducing new optical and microphysical properties for the stratospheric smoke particles. Our analysis also highlighted the need for coordinated ground-based lidar network measurements, such as the ones provided by EARLINET, as an exploratory tool in investigating unknown processes in the stratosphere.