Figure 3a, b, c shows the time variations in daily average values of AOD, single scattering albedo ω440, and asymmetry factor g440 in the summer period of 2012.

The AOD data indicate that there are two waves of high atmospheric smoke turbidities, from 17 June to 5 July and from 19 July to 6 August. Maximal AOD values (500 nm) reaching 3–5 were observed on 3–6 July and 25–29 July.

The single scattering albedos did not go out of the interval 0.93≤ω440≤0.99 in almost all cases. The asymmetry factor g440 varied from 0.66 to 0.74, i.e. in the same range as multi-year average values in the territory of Siberia (Sakerin, 2012; Sakerin et al., 2009, 2014). From the results presented here it also follows that, on those days when SSA and AF were retrieved by two methods, the SSMART and AERONET codes gave consistent estimates of aerosol properties within their retrieval uncertainties.

The arrival of smoke plumes was also accompanied by a substantial increase in the mass concentrations of black carbon (MBC) and aerosol (Ma) in the near-ground atmospheric layer. Kozlov et al. (2014) showed that the average values of these characteristics at the Fonovaya observatory in the period from 17 June to 4 August varied in the ranges MBC= 1.9–10.3 µg m-3 and Ma= 103–425 µg m-3 given the background levels of ∼ 0.45 and ∼ 20 µg m-3 respectively for 10–31 August. Similar to AOD, the maximal concentrations of MBC and Ma were observed in the period of 25–29 July.

The examples of daily average aerosol volume size distribution dV/dln⁡r for moderate to high AOD values are presented in Fig. 3d. A noticeable salient feature of the dV/dln⁡r distribution is the increase in the modal radius and in the width of the fine aerosol fraction during extreme turbidity on 28 July compared to periods on 27 June and 14 July, when AOD was much lower.

A consequence of anomalously high atmospheric turbidities was substantial changes in solar radiation reaching the earth's surface. Figure 4 illustrates the daytime behaviour of direct and diffuse radiative fluxes, measured with an MS-53 pyrheliometer and MS-802 pyranometer (0.305–2.8 µm) under usual (background) conditions on 24 July 2011 (τ500=0.07) and in the highly turbid atmosphere on 28 July 2012 (τ500=2.14). Under the influence of smoke plume, there was a 2- to 4-fold decrease in direct radiation, which was partially compensated for by an approximately equal increase in diffuse radiation. The diffuse radiation was considerably larger (a factor of ∼ 1.8) than the direct radiation throughout the day (28 July 2012).

The spatial distributions of AOD at 550 nm over Western Siberia and partially over Eastern Siberia (60–100∘ N, 52–70∘ E), averaged over the period of 15 June–10 August 2012, are presented in Fig. 5. These data show that the territory of the Tomsk region in this period was subjected to an intensive influence of large-scale smoke pollution and characterized by high values of AOD.

An analysis of multi-year AOD variations in a few regions of Russia according to data from photometric observations showed that, after eruption products of Pinatubo volcano had sunk out of the stratosphere, the interannual AOD variations were small and no statistically significant trend was observed in the past two decades (Sakerin, 2012; Sakerin et al., 2009; Sakerin and Kabanov, 2015). For instance, the annually average AOD values (500 nm) in Tomsk during the period of 2003–2015 (except in 2012) varied from 0.13 to 0.21 (Fig. 6a).

Under the influence of massive forest fires in Siberia during summer 2012, the annual AOD value exceeded the multi-year norm by almost a factor of 2 and became the highest over the entire period of photometric observations in Tomsk (since 1992) (Sakerin and Kabanov, 2015). Average summertime AOD values changed even more: in particular, the average value of τ500 in July increased by a factor of 4.5 compared to multi-year data (Fig. 6b). Even if we omit anomalously high turbidities (τ500>1), which may be partly due to clouds that are invisible through haze, the average AOD values in summer months of 2012 are outside the values defined when taking into account standard deviations (SDs).

More detailed characteristics of atmospheric AOD for three wavelengths in UV, visible, and near-infrared ranges during the extreme smoke in comparison with the multi-year average data for July are presented in Table 1. Also given are the average characteristics for various conditions of ordinary smoke, which are observed every year in Siberia in the warm period. The comparison showed (see also Fig. 6c) that AOD exceeds the background values by a factor of 5.5–6 in the entire spectral range during an extreme fire event in 2012. In absolute value, AOD increased more significantly in the shortwave part of spectrum, i.e. due to fine aerosol fraction. The slight increase in the Ångström exponent α440–870, which depends mainly on the interrelation between contributions of fine and coarse aerosols to AOD, also indicates that small particles predominate in smoke aerosol. Despite the temperature, aerosol, and other anomalies, the total water vapour content of the atmosphere in the period of strong fires differed little from the multi-year norm.

Average AOD characteristics in ordinary smoke incorporate data for fires of different types and distances from the region of measurements and, as such, occupy an intermediate position between background conditions and extreme smoke. Independent of their intensity, a common feature of all smoke is an identical exponent α, which characterizes a higher content of fine aerosol.

It is noteworthy that the multi-year average values of α440–870 and AOD in the near-infrared range, calculated for background conditions and total data set, differ little (see columns 2 and 3 in Table 1), primarily because of relatively small number (about 8 %) of smoke conditions in the total data set.

Under the assumption that aerosol size distributions are bimodal, O'Neil et al. (2001, 2003) have developed a spectral deconvolution algorithm (SDA) to infer the component fine- and coarse-mode optical depths from the spectral dependence of AOD. In June–August 2012, the coarse-mode τ500c is less than fine-mode τ500f and is typically low (only 3 days show that τ500c∼ 0.2), whereas τ500f is high and exhibits very large day-to-day variability (Fig. 7a). The parameter η500, which characterizes the contribution of fine fraction to AOD at 500 nm, exceeded 0.7 and approached 0.9 at τ500≥0.6 (Fig. 7b). These results are consistent with earlier published data, according to which the content of fine aerosol fraction predominately increases during vegetation burning in the absence of any other significant sources (Reid et al., 2005b; Eck et al., 2009; Giles et al., 2012).

The analysis of spatio-temporal AOD variations was confined to the consideration of the central part of Western Siberia. A salient feature of this territory is a uniform landscape (without mountainous terrain) and the absence of large sources of anthropogenic pollution. Under usual conditions, the AOD values, retrieved from satellite observations, are characterized by small (background) values and by quite a uniform spatial distribution (Zhuravleva et al., 2009; Sakerin, 2012).

Since the formation and evolution of smoke plumes are strongly affected by large-scale atmospheric dynamics, spatial distributions of AOD may substantially change from one day to another. The main specific features of variations can be identified through the analysis of AOD fields, averaged for a few 5-day periods, selected by taking into account the ground-based observations (Fig. 3a). On 6–10 June there were background conditions (before beginning of forest fires), on 1–5 and 24–28 July there were maximal smoke turbidities, and on 10–14 July there were relatively low AOD values observed between two turbidity maxima. Hereinafter, for calculations of the AOD values averaged over the selected time interval (5 days), we used the daily satellite product.

Spatial AOD distributions, presented in Fig. 8, show that the Tomsk-22 observation site was either at the centre (Fig. 8b) or on the periphery of smoke plumes (Fig. 8a) in different periods of time. The periods of small turbidities of the atmosphere (on 6–10 June and 10–14 July) were characterized by quite a uniform AOD distribution, and the largest spatial inhomogeneities and values τ550>1 were observed over the central part of Western Siberia on 24–28 July (Fig. 8 and Table 2).

The statistical characteristics in Table 2 were calculated using measurements from AQUA and TERRA platforms (MODIS collection 6, http://giovanni.sci.gsfc.nasa.gov/giovanni). These data suggest that, for three out of four periods considered here, the average AOD values at the Tomsk-22 site are close to spatio-temporal averages across the entire territory of Western Siberia: the difference between satellite and ground-based data is much less than between the standard deviations, except in the period of 24–28 July, when the average AOD at Tomsk-22 site was τ550=2.74 and exceeded the average value for Western Siberia (MODIS) τ550=1.06. At the same time, the satellite data in the (1∘ × 1∘) region of Tomsk-22 site agree well with the results of photometric observations. Thus, the results of ground-based measurements in the region of Tomsk-22 can be considered to reflect not only the local AOD features, but also the regularities of variations for the entire territory of Western Siberia.

During the extreme smoke, the average values of volume median radii for fine and coarse fractions were rvf=0.18 µm and rvc=3.3 µm, a little larger than these characteristics for ordinary smoke, equalling 0.16 and 2.9 µm respectively (Table 3). The rvf value, in both cases, was in the range of values obtained for biomass burning in other regions of the globe (Reid et al., 2005a, b; Eck et al., 2009; Chubarova et al., 2012), as well as in the range of measurements for aged smoke, performed using an optical particle counter (0.1–3 µm) and a differential mobility particle sizer (0.01–0.6 µm) in the boreal zones of Europe and North America (Reid et al., 2005a). As in Siberia, a wider size distribution of fine particles than for ordinary smoke (Table 3; Fig. 3d) was also noted by other authors for severe fires in the boreal zone of Alaska (Bonanza Creek, 2004–2005, Eck et al., 2009).

The noteworthy salient features of fine aerosol fraction during summer 2012 are as follows. The contribution of fine fraction to the total volume (Cvf/Cvt), on average, increased to 0.8±0.065 compared to the value 0.65±0.21 according to multi-year observations. Volume concentrations and volume median radii of the fine mode are plotted in Fig. 9a as functions of AOD at 440 nm. This figure shows that there is a general increase in volume concentrations Cvf as τ440 increases (Fig. 9a). The median radius of the fine mode increased from about 0.15 to 0.22 µm as τ440 changed from 0.4 to 1.5 and larger (Fig. 9b). The smoke particles have large radii, and rvf and τ440 are closely correlated, possibly because of a high concentration of aerosol, which leads to greater coagulation, condensation, and gas-to-particle conversion (Reid et al., 1998, 2005a; Eck et al., 2009). From this figure it also follows that the interrelation between median radius of fine aerosol fraction and AOD at 440 nm in ordinary smoke is much less pronounced, primarily because of the variety of properties of biomass burning aerosol combined in this data set. For instance, in May 2004, the median radius was relatively small (rvf∼ 0.14 µm) and did not practically change with varying τ440. Most probably, this was because the burning of the previous year's vegetation was one of the sources of smoke particles in this period of time. An analogous feature was also observed during African savanna fires (Dubovik et al., 2002).

In contrast to data from multi-year observations, in summer 2012 an increase in the volume concentration and median radius of the coarse mode was observed with growing AOD. Most probably, soil particles, having been suspended by saltation of surface dust driven by fire generated winds, were also present during this period of time in the composition of coarse mode in addition to carbon aggregates.

Results of retrieval of disperse composition of fine-mode aerosol in the atmospheric column agree with measurements in the near-ground layer (Kozlov et al., 2014). Data from spectral-polarization nephelometric measurements were used to show that the volume median radius of the fine mode under the conditions of weak turbidity of the atmosphere (10–13 July 2012) increased from 0.1 to 0.4–0.5 µm when smoke plumes intruded into the region of observations (25–29 July 2012).

Average values of the real and imaginary parts of the refractive index in extreme and ordinary smoke are presented in Table 3. In ordinary smoke, the imaginary part of the refractive index κλ changes weakly with increasing wavelength and is approximately 0.01. However, the imaginary refractive index during smoke pollution in 2012 shows low values and relatively large decrease in κλ as the wavelength grows from 440 to 675 nm. The κλ variations in the wavelength interval of 675–1020 nm are not so large.

The quantitative differences and specific features of spectral dependence of the imaginary refractive indices stem from different properties of atmospheric carbonaceous particles, i.e. black carbon (BC) and organic aerosol (OA). Recent studies showed that OA components can contribute substantially to light absorption. In contrast to BC, which absorbs light throughout the UV-visible spectrum, an OA component such as brown carbon (BrC) absorbs mostly at the ultraviolet wavelengths and less significantly in the visible spectral range (Kirchstetter et al., 2004; Bergstrom et al., 2007; Chen and Bond, 2010; Zhong and Jang, 2014; Chakrabarty et al., 2016). The majority of BrC is emitted to the atmosphere through low temperature, incomplete combustion of biomass, bio- and fossil fuel (Bond, 2001; Kirchstetter et al., 2004; Bergstrom et al., 2007; Lewis et al., 2008; Chen and Bond, 2010; Zhong and Jang, 2014).

The absorption efficiency and spectral dependence vary depending on the type of BrC origin. Lu et al. (2015) reviewed available measurements (laboratory and field observations) of light-absorbing primary organic aerosols, and quantify the wavelength-dependent imaginary refractive indices (κOA) for the bulk primary OA emitted from biomass/biofuel, lignite, propane, and oil combustion sources. Based on generalized information, they suggested parameterizing κOA of biomass/biofuel combustion sources as a function of BC-to-OA ratio. Analysis of imaginary refractive indices showed the stronger wavelength-dependent κOA for lower BC-to-OA ratio conditions (smoldering combustion), while the greater κOA values at λ > 350 nm are observed for higher BC-to-OA ratio (flaming combustion).

These results allow us to hypothesize that the reason for the κλ decrease in the interval of 440–675 nm during summer 2012 may be due to the contribution of compounds absorbing radiation in UV wavelength region (brown carbon). To confirm this hypothesis, we considered OMI (Ozone Monitoring Instrument) observations of the ultraviolet aerosol index (UVAI), the value of which is sensitive to aerosol absorption in the ultraviolet wavelength region (Torres et al., 1998; Jethva and Torres, 2011; Hammer et al., 2016). Satellite data indicate that UVAI values were quite high during the smoke pollution in 2012, and exceeded 4–5 on individual days, thus explaining the above-mentioned κλ decrease from UV to the visible spectral region (Fig. 10).

The relatively constant imaginary refractive index in ordinary smoke is due to averaging over many situations and differs, in particular, in the type of burning biomass (peat, forest, grass). For instance, in May 2004, when the characteristics of smoke aerosol were quite homogeneous, the average spectral behaviour of κλ was weakly manifested, with κλ≈0.013 in the interval of 440–1020 nm. At the same time, the UVAI value did not exceed 1.2 (http://giovanni.sci.gsfc.nasa.gov/giovanni/), suggesting that BC was seemingly the main absorbing substance in this period of time.

The main results, presented in this section, were obtained using the algorithm from Dubovik and King (2000). At the same time, we present the SSA and AF values, retrieved using our SSMART algorithm in separate atmospheric conditions. The SSA and AF values were retrieved directly from the data from AOD and sky radiance measurements through solution of radiative transfer equation (Way 1, Bedareva et al., 2013a, 2014).

Detailed analysis of spectral SSA in 2012 has revealed two types of dependence: monotonically decreasing and monotonically increasing ωλ with growing wavelength (Fig. 11a).

The main group of SSA data (45 cases out of 65) is characterized by a weakly decreasing spectral dependence: the range of the differences Δωλ=ω440-ω1020 did not exceed 0.03. Practically all sets of retrieved SSA of this type lie between curves 2 (ω440=0.926) and 3 (ω440=0.996). In other cases, the ωλ decrease with growing wavelength is larger, but does not exceed 0.06 (curve 4, Fig. 11a). We note that the values of the imaginary part of the refractive index are ∼ 0.006 for this subset of data. In a few (14 out of 65) situations with high atmospheric turbidities (τ440>0.9), there was an increase in SSA with growing wavelength (curve 1 in Fig. 11a). The monotonic increase in ωλ agrees with a decrease in κλ (on average, from 0.01 at 440 nm to ∼ 0.006 at 675 nm), presumably due to additional absorption of brown carbon in the UV wavelength region.

We will compare the single scattering albedo and asymmetry factor, retrieved using AERONET and SSMART algorithms (Table 4, Fig. 11b, c). Table 4 presents the ωλ and gλ values for three individual atmospheric conditions. These data illustrate that both approaches ensure a consistent spectral behaviour of SSA: the increase (28 July) and decrease (4 August) of the single scattering albedo with growing wavelength, as well as a near-neutral spectral dependence of SSA (1 August). The difference in ωλ values in these examples does not exceed 0.03 and is within the uncertainty of retrieval estimates of both algorithms. The difference in asymmetry factor is ∼ 0.03, increasing to 0.05 at λ= 1020 nm.

The reason for the differences in the optical characteristics may be the following circumstances. When measurements of solar radiation in solar almucantar are used in the solution of the inverse problem with the help of the SSMART algorithm (Way 1), the scattering phase function of aerosol particles ga(θ) can be retrieved only in the region of scattering angles θ≤2θ0 (θ0 is the solar zenith angle); while in the remaining angular range the values of aerosol scattering phase function are extrapolated. The extrapolation procedure may lead to relatively large (larger than 20–30 %) uncertainties of ga(θ) retrieval for θ > 2θ0, and may influence the accuracy of the retrieval of the asymmetry factor and, to a lesser degree, the single scattering albedo. We discussed these issues in our earlier work (Bedareva and Zhuravleva, 2011). Another reason for the differences may be the fact that the SSMART algorithm retrieves the optical characteristics for each spectral channel singly, rather than calculating them for all channels simultaneously on the basis of retrieved complex refractive index and particle size spectrum, as the AERONET algorithm makes. Therefore, the results of the solution of the inverse problem with the use of the algorithm from Dubovik and King (2000) are characterized by more consistent spectral dependence of SSA and AF compared to the SSMART software package.

We will consider the results of retrieval of single scattering albedo and asymmetry factor from 28 July 2012 (Fig. 11b, c). The solutions of the inverse problem, based on application of two independent approaches, agree not only qualitatively but also quantitatively. At the same time, the daily average SSA (AF) values, retrieved using the SSMART algorithm, are overestimated (underestimated) relative to AERONET data by about 0.01, except for λ= 1020 nm, when the differences increase to 0.02 and 0.04 respectively.

The consistency of results can be considered as an indirect evidence that the spectral SSA dependence, atypical for the smoke aerosol, is not so much due to methodic features of retrieval procedures, but to specific features of smoke aerosol over the AERONET site in Tomsk on 28 July (possibly, chemical composition and particle sizes). In Sect. 4.2.2 we hypothesized that the increased absorption by aerosol particles in the ultraviolet spectral range may be due to the presence of brown carbon; indeed, the largest value UVAI = 4–6 over the entire period of extreme smoke was observed on 26–28 July (Fig. 10). However, the limited data set considered in this work still does not allow us to explain the observed spectral behaviour of SSA. We can now only state that the increase in the single scattering albedo with growing wavelength was observed not only in Tomsk, but also in certain situations during severe fires in Moscow during 2010 and in Alaska during 2004 (http://aeronet.gsfc.nasa.gov). In addition, the laboratory analysis showed that brown carbon, with a negligible amount of black carbon, predominated in the composition of aerosol particles emitted from the burning of Alaskan and Siberian peatlands (Chakrabarty et al., 2016). A more detailed analysis of the spectral dependence of SSA of smoke aerosol under the conditions of extraordinary fires will be the subject of further studies.

On the whole, the period of strong fires in Tomsk was characterized by weak spectral variations in SSA, the average being ∼ 0.96, consistent with previous results in other regions of the Eurasian boreal zone, e.g. in Moscow (2002 and 2010) and Alaska (2004–2005), where the smoldering phase prevailed over the flaming fire phase (Eck et al., 2009; Chubarova et al., 2011; Sayer et al., 2014) (Fig. 12a). The relatively constant spectral single scattering albedo during extreme smoke in 2012 may be also partly due to growth of aerosol particle sizes (Sect. 4.2.1, Table 3; see also Eck et al., 2009).

The SSA retrievals in the atmospheric column satisfactorily agree with data in the ground-layer (Kozlov et al., 2014), according to which the single scattering albedos in the visible range (λ= 510 nm) were in the interval of 0.95–0.98. At the same time, the imaginary and real parts of the complex refractive index of dry base of substance varied in the ranges of 0.01–0.02 and 1.36–1.47 respectively.

The overwhelming majority of results of individual retrievals in ordinary smoke also showed a monotonic SSA decrease with the increasing wavelength, with the opposite dependence being observed in only 6 cases out of 140. However, in contrast to extreme smoke, the spectral SSA behaviour was more pronounced: Δωλ reached 0.06–0.18 in 18 % of cases; Δωλ was 0.03–0.06 in 40 % of cases, and Δωλ was less than 0.03 in 34 % of cases.

The average single scattering albedos in ordinary smoke, presented in Fig. 12a, turned out to be about 0.02 lower than those observed in the boreal forests of USA and Canada (Dubovik et al., 2002). These differences may be due to both methodic and physical factors. Methodic-type differences might stem from different methods for selecting the smoke conditions. The specific features of the spectral dependence of SSA can also be explained by different combinations of the composition and age of smoke aerosol, which were recorded at AERONET observation sites. This is clearly illustrated by results obtained in May 2004 in Tomsk (Fig. 12a). Lower values and pronounced spectral variations of SSA (a decrease from 0.9 to 0.85) may be due to the previous year's grass and agricultural vegetation combustion products present in the smoke composition, as well as due to a flaming phase of combustion which predominated in that period of time. Close values and analogous spectral dependence of single scattering albedo were also noted in African savanna fires in Zambia (Dubovik et al., 2002; Sayer et al., 2014).

Aerosol absorption optical depth τabs,λ and absorption Ångström exponent αabs, computed for the spectral interval of 440–870 nm (Sect. 2), were considered as other characteristics of absorbing properties of aerosol particles.

Data presented in Fig. 12c show that the average values of τabs,λ during strong fires in 2012 were about a factor of 1.5–2 smaller than for ordinary smoke. In addition, there are differences in αabs, which are 1.6 ± 0.44 for extreme smoke and 1.2 ± 0.5 for ordinary smoke.

The AAE values can be used as an indicator of aerosol composition (Andreae and Gelencsér, 2006; Russell et al., 2010; Schuster et al., 2016). The relatively low values αabs∼ 1 are typical for aerosol absorption largely dominated by black or light absorbing carbon, while larger values suggest absorption by different material organic carbon (Bergstrom et al., 2002, 2007; Kirchstetter et al., 2004; Lewis et al., 2008; Russell et al., 2010; Schuster et al., 2016). The αabs increase in the 2012 fires may indicate that, on average, smoke absorption possibly shifted from domination by black carbon in ordinary smoke to an increasing influence of absorption by other materials, most likely organic carbon (see also Fig. 10).

In contrast to single scattering albedo, the spectral asymmetry factor differs insignificantly among the 2012 fires, ordinary smoke, and May 2004 smoke (Table 3, Fig. 12b).