2.1 Lidar system description

The CIMEL CE372 lidar belongs to a new generation of lidar derived from the previous CE370 model operating in the visible (Dieudonné et al., 2013) and from the one especially developed for the IAOOS project (Mariage et al., 2017). The CE372 is a single wavelength system using a laser diode emitting 200 ns pulses at 808 nm, the temperature of which is regulated by a Peltier device. The maximum output power is 18 mW with a repetition rate of 4.72 kHz (3.8 µJ energy). The energy of the laser diode is recorded continuously with a photodiode and a 30 nm filter centred at 808 nm, but the energy measurement was only reliable at night because the background solar radiation is still too high on the photodiode to make daytime measurements possible. The optical receiver includes a 10 cm diameter lens and a 0.6 nm filter to reduce background light. The detection unit is based on an Avalanche photodiode (APD) used in Geiger mode (Single Photon Counting Module, SPCM, from EXCELITAS) and a standard high-speed sampling and averaging electronic card from Cimel Electronique. The photocounting signal is delivered by the SPCM with a maximum frequency around 35 MHz and detection gate of 100 ns (15 m vertical resolution). Lidar profiles are recorded with an integration time of 1 min. The signal is corrected from saturation due to APD detector dead time (22 ns) using the methodology of Mariage et al. (2017). The background correction uses the average signal recorded between 20 and 30 km.

For each day, three periods of 30 min are selected between 00:00 and 12:00 UT (day), 12:00 and 20:00 UT (night), and 20:00 and 24:00 UT (day) for the analysis of vertical aerosol profiles. The selection of the best interval of 30 min to average the 1 min lidar profiles is based on the elimination of very cloudy profiles. Data filtering with lidar radiometric detection of a cloud (day only), with a search for layers showing very strong backscatter below 5 km (day and night) and for high opacity of the 0–3 km atmospheric layer (day and night), very efficiently selects the 30 min time periods with no cloud layers below 5 km. The following criteria are then applied to eliminate the lidar data that are considered too cloudy: all the profiles with a daytime sky level (SB) greater than 7000 counts s⁻¹ or with a 150 m layer where the backscatter ratio is greater than 17 between 0 and 4.5 km, or with attenuated backscatter smaller than 10⁻⁴ km⁻¹ sr⁻¹ between 3 and 8 km cloud layers below 5 km.

A total of 540 averaged profiles are thus available for aerosol profile analysis over the period April 2015 to September 2016 with 300 daytime profiles and 240 night-time profiles.

An example of the attenuated backscatter vertical profile for a 30 min night-time and daytime averaging in June 2015 is shown in Fig. 1. The signal is normalized to the molecular attenuated backscatter at night at 9 km below a cirrus cloud observed above 10.5 km. The signal-to-noise ratio SNR is good enough to detect aerosol layers up to the tropopause at night. Only aerosols below 3 km are detected during the day and the molecular reference signal cannot be accurately measured during the day.

Figure 1Vertical profiles of the attenuated backscatter signal (PR2) for daytime (a) and night-time (b) averaged over 30 min on 29 June 2015 using a calibration constant K to normalize the night-time PR2 to attenuated molecular backscatter at 10 km below the cirrus layer. The red curve is the attenuated molecular backscatter signal. Daily evolution of the vertical profiles of the log10 of the attenuated backscatter (c) and the background signal due to solar radiation (d).

Download

As the alignment of the lidar remains very stable over time, the geometric overlap factor (OF) between the laser and the receiver is estimated between the surface and 500 m by averaging the profiles with a mean attenuated backscatter ratio <1.1 at 500 m and by assuming a constant scattering ratio between the surface and 500 m. This provides a sufficiently accurate geometric overlap factor to correct for the underestimation of the contribution of this altitude domain to the AOD assessment (Fig. 2) between 100 and 500 m. Below 100 m, OF retrieved with this method is not accurate enough and we will assume a constant backscatter ratio between the surface and 100 m. This assumption induces a 2 % error in the AOD assuming a constant extinction layer that is 1000 m deep and a 20 % error in the scattering ratio below 100 m.

Figure 2Geometrical overlap function used for the correction of the lidar data in log10 scale between 100 and 500 m.

Download

2.2 Lidar calibration

Owing to the low SNR of daytime lidar signal above 2–3 km altitude and the difficulty of finding an altitude zone where aerosol backscatter is negligible compared to molecular backscatter, we propose a specific methodology to determine the evolution of the lidar calibration factor. Indeed, a precise calibration of the lidar first allows the determination of the daytime integrated backscatter, assuming very low variation in the calibration factor during the day. Daytime integrated backscatter is then used to derive the integrated lidar ratio using independent AOD measurement from a sun photometer. During the night, calibrated lidar measurements are also useful to reduce the uncertainty in the calculation of the extinction profile to the relative error in the range-corrected signal (PR2) and to that on the determination of the lidar ratio (Appendix A).

A first guess of the calibration coefficient K is obtained from a normalization of the minimum backscatter ratio R to 1 at an altitude between 4 and 9 km for night profiles. The vertical profile of molecular backscatter is estimated from the pressure and temperature profiles after temporal and spatial interpolation of the four daily ERA-Interim ECMWF meteorological fields at 0.75^∘ (Dee et al., 2011). The lidar backscatter ratio is averaged over 150 m to reduce the uncertainty in the molecular signal below 2.5 %, i.e. 3.2 times less than the 8 % signal standard deviation shown in Fig. 1 at night at 9 km. A first guess of the aerosol two-way transmittance $T_{\mathrm{a}}^{\mathrm{2}}$ between the altitude 100 m and the reference altitude z_r, chosen for normalization to the backscatter profile, is then calculated after determining the extinction profile with an a priori lidar ratio and using the backward inversion method described in Appendix A. An a priori value of 60 sr is chosen for the vertically averaged lidar ratio S at 808 nm because it corresponds to biomass burning or pollution aerosols using the lidar ratio look up table at 532 nm of the CALIPSO mission aerosol climatology (Omar et al., 2009) and the spectral variability of the lidar ratio between 500 and 808 nm proposed by Cattrall et al. (2005).

The second step in our estimation of lidar calibration is to select the night-time profiles with two additional criteria: z_r>7.5 km and $T_{\mathrm{a}}^{\mathrm{2}}>\mathrm{0.89}$. There are 106 such profiles out of 540. This selection increases the probability of having a normalization zone with a good signal-to-noise ratio, a negligible contribution of particle backscatter (z_r>7.5 km) and minimization of the normalization error due to an error in $T_{\mathrm{a}}^{\mathrm{2}}$ when S is very different from 60 sr ($T_{\mathrm{a}}^{\mathrm{2}}>$0.89). This corresponds to the profile selection shown in the left part of Fig. 4. For these profiles, the calibration factors deduced from a normalization of the PR2 at z_r are the best proxies for the lidar calibration. The corresponding optimal values K_opt of the calibration factor are shown in Fig. 3 (black crosses). Since the error in the backscatter ratio at z_r is below 3 %, the error, ΔK, in this calibration factor depends mainly on the lidar ratio error, ΔS:

Assuming a 35 % relative uncertainty in S (i.e. expected lidar ratio in the range 35–60 sr assuming that all the aerosol types can be encountered except the clean marine or dusty marine types Omar et al., 2009) and the Cattrall et al. (2005) S spectral variability, the error in K_opt is less than 4 % according to Eq. (1) when $T_{\mathrm{a}}^{\mathrm{2}}>$0.89.

The third step is to replace the calibration factors K for non-optimal conditions (daytime profiles, and night-time profiles with either AOD >0.06 or clouds between 4 and 7.5 km) by interpolated values between the nearest K_opt values (see the four arrows labelled with K_opt in Fig. 4). If there are more than 10 days between two optimal calibration factors, the nearest value of K_opt is chosen. If the interpolated value is greater than 20 % of the calibration factor first guess divided by $T_{\mathrm{a}}^{\mathrm{2}}$, the latter is retained to take into account exceptionally lower optical transmission of the lidar (window icing, de-tuned filter) or a transient decrease in the emitted energy. Indeed, the use of the interpolated calibration factor would lead to a backscatter ratio much too low in the free troposphere ($<\mathrm{0.8}.T_{\mathrm{a}}^{\mathrm{2}}$). There are fewer than 20 such cases between December 2015 and June 2016; therefore fewer than 3 % of the cases studied have an unusually low calibration factor.

The time evolution of K in Fig. 3 shows that the overall transmission of the lidar system increased by 30 % when it was installed in the Fonovaya container in September 2015 and decreased again when it was operated again on the roof of the IAO for 1 month in September 2016. At the Fonovaya site the short-term variability (<10 days) is much higher (>15 %) than at the Tomsk site, where, in contrast, the calibration constant increases regularly by 30 % over 4 months. The short-term variability is mainly related to changes in the optical transmission of the air-conditioned container window, while the drift over 4 months with the initial conditioning of the CE372 on the roof of IAO is due to an improvement in the filter transmission at 808 nm during a gradual increase in outside temperatures. Analysis of the night-time energy measurements does not indicate any significant variation in the energy emitted by the laser diode (<15 %). To estimate our error in K values for non-optimal conditions (red points in Fig. 3), a good proxy is the difference between two optimal calibration factors derived for two observations made with a time difference <1 day. Changes in K_opt for such a short time period cannot be expected when aiming at calibration of daytime observations with night-time calibrated profiles. There are 23 pairs of K_opt values with a 1-day time difference and the standard deviation of their difference, ΔK_opt, is 2.5×10⁴. Such a variability is then a limiting factor in our ability to calibrate the lidar for daytime observations or night-time conditions with AOD >0.06 or clouds between 4 and 7.5 km. The corresponding accuracy of the calibration factor K is then of the order of 8 % ($\mathrm{2.5}\times {\mathrm{10}}^{\mathrm{4}}/\mathrm{3}\times {\mathrm{10}}^{\mathrm{5}}$).

Figure 3Time evolution from April 2015 to September 2016 of the lidar calibration factor (multiplied by 10⁻⁵). Dotted red lines correspond to major changes in lidar housing and expected change in calibration. Black crosses are for nocturnal profiles with molecular normalization at z_r>7 km and aerosol two-way transmittance $T_{\mathrm{a}}^{\mathrm{2}}>\mathrm{0.89}$ (106 values out of 540). Red dots are for calibration interpolated from optimal conditions. Blue dots are for the few cases (20 out of 540) when calibration cannot be interpolated from optimal conditions.

Download

2.3 Methodology for the lidar aerosol optical depth (AOD) retrieval

Daytime indirect aerosol optical depth retrieval from the calibrated PR2 is based on the well-known backward inversion of PR2 (Fernald, 1984; Klett, 1981) described in Appendix A, provided that an independent measurement of $T_{\mathrm{a}}^{\mathrm{2}}$ is available to constrain the lidar ratio, e.g. using a sun photometer (Chaikovsky et al., 2016; Cuesta et al., 2008). This work proposes a methodology for the AOD and backscatter ratio profile retrieval, taking into account the different observation conditions described in Fig. 4. It is necessary because the cloud-free lidar profile identified during the 18-month period cannot always be constrained by a sun-photometer AOD, e.g. for night-time observations or daytime observations without the synergy lidar/sun photometer. Two different cases are identified in Fig. 4 for night-time observations. First, a direct AOD retrieval is possible when a layer with molecular signal is found in the upper troposphere above 7.5 km and takes advantage of the good lidar calibration to measure the AOD from the lidar-attenuated backscatter (see Appendix A). For the second case (cloudy conditions 4–7 km or very low AOD <0.04, resulting in a large error in $T_{\mathrm{a}}^{\mathrm{2}}$ deduced from the attenuated backscatter above 7 km), we cannot rely on an independent estimate of $T_{\mathrm{a}}^{\mathrm{2}}$ to iterate over the proper lidar ratio. An assumption about the aerosol source must then be made, e.g. using FLEXPART simulations (see Sect. 3), and the lidar ratio is taken from a look-up table for the five main Siberian aerosol sources. This look-up table is representative of Siberia because it is built using all the daytime and night-time observations when independent $T_{\mathrm{a}}^{\mathrm{2}}$ are available.

Figure 4Flow chart of the lidar range-corrected signal (PR2) processing to derive the 800 nm aerosol optical depth (AOD) up to the reference altitude (z_ref). Four different calibrations and AOD calculations are used according to the measurement conditions. Iteration between the AOD calculation and the lidar ratio value is only possible when the AOD is compared to an external AOD reference (green two-sided arrows).

Download

Independent daytime $T_{\mathrm{a}}^{\mathrm{2}}$ at 808 nm can be obtained using the 870 nm AOD and the Ångström coefficient (AE) measured by the AERONET sun photometer located either at the Tomsk site (56.4^∘ N, 85.0^∘ E) or that of Tomsk22 (56.4^∘ N, 84.1^∘ E). Long-range transport of aerosol plumes are generally similar at both sites (Zhuravleva et al., 2017). There are 210 cases out of 539 lidar profiles with coincident lidar and sun-photometer observations.

To increase the number of lidar profiles constrained by an independent estimate of $T_{\mathrm{a}}^{\mathrm{2}}$, direct lidar measurements of the 808 nm AOD can also be obtained at night if the lidar is well calibrated and if the reference altitude is above 7.5 km, i.e. with a negligible contribution of particle backscatter (<10 % of molecular backscatter). Indeed, the value of the attenuated backscatter ratio at altitude z_r is then a direct measurement of the two-way transmittance $T_{\mathrm{a}}^{\mathrm{2}}(z_{\mathrm{r}}$) (Appendix A). The accuracy of the corresponding AOD is $\frac{\mathrm{1}}{\mathrm{2}}\frac{\mathrm{\Delta }\mathrm{K}}{\mathrm{K}}=\mathrm{4}$ % when using the 8 % accuracy for the calibration factor determined in Sect. 2.2. The analysis is limited to AOD >0.04 to avoid large relative errors in the retrieved AOD. There are 63 such cases, providing additional constraint for the lidar ratio retrieval.

3.1 FLEXPART aerosol tracer simulation

FLEXPART is a Lagrangian model designed for computing backward or forward long-range transport, diffusion, dry and wet deposition of air pollutants or aerosol particles from point sources using a large number of particles (Stohl and Seibert, 1998; Stohl et al., 2002). Particle dispersion model calculations can be performed by assuming two modes of transport in the atmosphere: passive transport without removal processes and transport of aerosol tracers, including removal by dry and wet deposition in the cloud and under the cloud (Kristiansen et al., 2016; Stohl et al., 2012). For each lidar profile, the latter was chosen using backward simulations of 10 000 particles released in two altitude zones: (i) 500 m to z_aer and (ii) z_aer to z_max, z_max being the highest altitude with a scattering ratio R>2 and z_aer being the aerosol-weighted altitude calculated with the aerosol backscatter vertical profile.

For dry removal, particle density, aerodynamic diameter and standard deviation of a log-normal distribution were assumed to be 1400 kg m⁻³, 0.25 µm and 1.25, respectively following Stohl et al. (2013). Below-cloud scavenging is modelled using a wet scavenging coefficient defined as λ=AI^B, where A is the wet scavenging coefficient, I the precipitation rate in mm h⁻¹, and B is the factor dependency. We set $A={\mathrm{2.10}}^{-\mathrm{5}}$ s⁻¹ and B=0.8. The in-cloud scavenging is simulated using a scavenging coefficient defined as $\mathit{\lambda }=\left(\mathrm{1.25}{I}^{\mathrm{0.64}}\right)H^{-\mathrm{1}}$, where H is the cloud thickness in metres. The occurrence of clouds is calculated by FLEXPART using the relative humidity fields. The meteorological fields used for the simulations (including precipitation rates) are ERA-Interim ECMWF field at T255 horizontal resolution (≈80 km) and 61 model vertical levels.

A backward run of the model initialized from the receptor point (the lidar location) provides every 6 h potential emission sensitivity (PES) field in seconds with a vertical resolution of 1000 m and a horizontal resolution of $\mathrm{1.75}{}^{\circ }\times \mathrm{1}{}^{\circ }$ (Seibert and Frank, 2004). These PES fields are generally recombined over a 9-day period, either in the first vertical layer (0–1000 m) to obtain PES_surf or over the first five vertical layers (0–5000 m) to obtain PES_0–5 km. The first 12 h before release are excluded to avoid a strong bias by the high PES due to recent local emissions which will mask high PES from remote sources. Examples of PES_0–5 km fields are shown in Sect. 5.1.

3.2 Distribution of aerosol sources

Several potential aerosol sources have already been identified for Siberia: (1) urban pollution (Dieudonné et al., 2017; Raut et al., 2017), (2) flaring in the oil and gas industry (Huang and Fu, 2016; Stohl et al., 2013), (3) biomass burning (Teakles et al., 2017; Warneke et al., 2009), (4) dust from central Asian deserts (Gomes and Gillette, 1993; Hofer et al., 2017) and (5) organic aerosols emitted by taiga (Paris et al., 2009). The position of these source zones are coupled with the PES maps calculated by FLEXPART for the aerosol source attribution to a given lidar observation.

The role of urban pollution will be identified by the position of cities of more than 500 000 inhabitants in Russia, Mongolia and Kazakhstan without including emission inventory or seasonal variation of the emissions. We are aware it is a crude assumption for a true aerosol modelling exercise but it is a reasonable criteria for testing the potential effect of urban aerosol on the lidar data.

The biomass-burning emission zones are derived from the daily fire radiative power (FRP) maps provided by NASA Fire Information for Resource Management System (FIRMS) using MODIS (Giglio et al., 2003) and the Visible Infrared Imaging Radiometer Suite (VIIRS) (Schroeder et al., 2014). The FRP is estimated from both MODIS and VIIRS hot spots of the brightness temperature measurements. MCD14ML collection 6 standard quality products and VNP14IMGTDLNRT are used for, respectively, MODIS and VIIRS. The FIRMS data set then provides daytime (MODIS, VIIRS) and night-time (VIIRS) measurements with spatial resolutions of 1 km (MODIS) or 0.375 km (VIIRS). Only FRP values >0.3 GW for MODIS and >0.1 GW for VIIRS are used to identify biomass-burning zones.

To identify continental regions covered by forests and deserts, we use the built-in United States Geological Survey (USGS) 24 category land-use database in the WRF (Weather Research and Forecasting) model. This global land cover database is derived from the Advanced Very High Resolution Radiometer (AVHRR) data with a resolution of 1 km spanning a 12-month period (April 1992–March 1993) (Sertel et al., 2010). The role of dust plumes can be overestimated when using only the land-use map, so it is only considered if neither urban pollution nor biomass burning have been identified.

Russia and Nigeria are the two biggest contributors to gas flaring used at oil and gas production and processing sites. The location of flaring sources is based on the anthropogenic emissions ECLIPSEv4 database (Evaluating the Climate and Air Quality Impacts of Short-Lived pollutants) described in Klimont et al. (2017). This inventory includes, in particular, the gridded methane emissions from gas flaring in the Russian Arctic at a $\mathrm{0.5}{}^{\circ }\times \mathrm{0.5}{}^{\circ }$ degrees horizontal resolution. A threshold of 50 moles km⁻² h⁻¹ has been applied to the methane emissions to select areas that could potentially be defined as flaring sources. Owing to the strong variability of flaring emissions, the role of flaring may be overestimated, so, as for the dust emission, it is only considered if anthropogenic and biomass-burning sources are not identified.

Figure 5Map of the 2015 (a) and 2016 (b) aerosol sources coupled with a FLEXPART PES gridded map: grid cells with large cities (blue square), central Asia desert (yellow), biomass burning (light brown), gas flaring (dark brown) and taiga (green).

Download

The map of the main aerosol emission sources are shown in Fig. 5 for 2015 (top) and 2016 (bottom). The lidar measurement site corresponds to the blue squares at 56^∘ N, 85^∘ E. In 2016 forest fires were very numerous in central Siberia, whereas they are much further east of Lake Baikal in 2015. When PES_surf>1500 s for at least one grid cell with a large city or flaring emissions, the type of aerosol is classified, respectively, as urban aerosol or flaring aerosol. When PES_0–5 km> 1500 s for at least one grid cell with fires or desert soils, the type of aerosol is classified, respectively, as biomass-burning aerosol or dust aerosol. PES_0–5 km is chosen for dust and biomass-burning plumes which can be quickly uplifted in the free troposphere up to 5 km. If none of the above conditions are fulfilled, the remaining significant source is the contribution of oxygenated aerosol emission from the very large area covered by the taiga (Zhang et al., 2007).

4.1 Night-time direct AOD measurements

The probability density function (PDF) of the night-time lidar AOD using the direct retrieval method described in Sect. 2.3 is compared with the PDF of the AOD measured during the day by the sun photometer (not including one-third of the sun-photometer AOD <0.04 since AOD <0.04 are not considered in the direct night-time AOD retrieval). The comparison shows that our night-time retrieval using the backscatter ratio at z_r gives a realistic distribution of the AOD with similar median and 90th percentiles of the AOD (Fig. 6a, b). A direct comparison between night-time lidar AOD and photometer AOD is not possible in Tomsk because lunar photometry is not available. The alternative solution is to use a sun-photometer AOD with a time difference <6 h with the lidar observations and to include the observed daily variability of the sun-photometer AOD. The correlation plot is also shown in Fig. 6c, showing no clear bias and a satisfactory agreement considering the daily variability of the sun-photometer AOD (error bar in Fig. 6c).

Figure 6(a) PDF of daytime AOD from a sun photometer for lidar measurement days and (b) of night-time AOD calculated with the lidar-attenuated backscatter ratio measurement at the reference altitude z_r. N is the number of observations, while p50th and p90th are respectively the median and 90th percentile of the AOD distribution. (c) Correlation plot of night-time lidar AOD versus sun-photometer daytime AOD when the time difference between the two measurements is less than 6 h (35 cases out of 63 night-time lidar AOD). The error bar is the daily variability of the sun-photometer AOD.

Download

4.2 Integrated lidar ratio retrieval

According to Fig. 4, backward inversion of the calibrated lidar-attenuated backscatter can be done iteratively using different lidar ratios when the optical thickness calculated with the extinction profile is compared with the independently obtained AOD. The final solution is always obtained after six iterations. Starting with the largest expected lidar ratio allows a fast convergence towards the true value (e.g. see Young, 1995). Thirteen S₈₀₈<45 sr out of the 15 FLEXPART dust cases could be retrieved with this method, even though iteration starts with 60 sr. A set of 273 lidar ratios constrained by daytime observations with sun-photometer or by night-time measurements where $T_{\mathrm{a}}^{\mathrm{2}}(z_{\mathrm{r}}$) is then available to build the lidar ratio look-up table (Table 1) for each of the five aerosol types determined with the FLEXPART analysis described in Sect. 3 and for three seasons: the cold season (15 October to 15 March), spring (15 March to 30 June) and warm season (30 June to 15 October). The standard deviation of the lidar ratio for each class is a good proxy for the error in the 273 S₈₀₈ values retrieved with this method. Since the 10 sr error remains significant, it is important to discuss the lidar variability obtained in Table 1. First, as expected (Burton et al., 2012; Omar et al., 2009), the lowest values (40 sr) are indeed obtained for the desert aerosol class, while the highest values (>60 sr) are characteristic of pollution aerosols (flaring and urban pollution in winter). Using the spectral variability of the lidar ratio proposed by Cattrall et al. (2005) to calculate the equivalent S values at 532 nm, S₅₃₂ is 50 sr for the lower limit of our lidar ratio and 80 sr for the lidar ratio of pollution aerosol. This is consistent with the Burton et al. (2012) analysis, but the lower limit is higher than the average lidar ratio obtained by Hofer et al. (2017) (35 sr) in the deserts of Tajikistan. Aerosol growth and mixing during long-range transport are likely responsible for higher values of S in Tomsk (Ancellet et al., 2016; Nicolae et al., 2013).

For the remaining 267 lidar profiles, where the lidar ratio cannot be constrained by the sun photometer or a good calibrated lidar measurement above 7.5 km, the FLEXPART analysis and the lidar ratio look-up table (Table 1) are used to retrieve the backscatter ratio and the extinction profile (see right- and left-hand side cases in Fig. 4). The relative error in the AOD calculated with the extinction profile is then mainly related to the relative error in the lidar ratio derived from the look-up table, i.e. of the order of 25 %.

Table 1Lidar ratio at 808 nm in sr for the five FLEXPART-derived aerosol types and three seasons (cold, spring and warm) when using independent AOD measurements.

Download Print Version | Download XLSX

4.3 Lidar AOD seasonal variability

The whole time series of the median of the backscatter ratio R₈₀₈ between 0–2.5 and 2.5–5 km are shown in Fig. 7. As expected, the mean backscatter ratios >3 are seen mainly in the lowermost troposphere below 2.5 km (22 % of the 540 profiles), while only 5 % are observed for the altitude range 2.5–5 km. Elevated backscatter ratios (>3) are observed from February to September below 2.5 km and from April to September in the free troposphere. The latter is more or less in phase with the start and end dates of dust storm and forest fire periods in Eurasia.

Figure 7Time evolution from April 2015 to September 2016 of the median of the 808 nm Tomsk lidar backscatter ratio for two altitude ranges: 0–2.5 km (b) and 2.5–5 km (a). A, B and C are the cases analysed in Sect. 5.

Download

The time series of the AOD calculated from the extinction vertical profiles is then compared to the AOD from the sun photometer (Fig. 8). The agreement is generally good between the two time series of AOD and elevated AODs (>0.2) are clearly visible at about the same periods. More short-term variability is obtained for the sun-photometer AOD since all 10 min cloud-free observations are shown in Fig. 8. The elevated AOD are not only observed in summer (June to September), which indicates that biomass-burning episodes are not solely responsible for the strong AOD. A strong difference between AOD₅₅₀ for warm (AOD =0.3) and cold season (AOD =0.08) has been also reported by Chubarova et al. (2016) for the city of Moscow. The corresponding time evolution of the aerosol-weighted altitude calculated with Eq. (2) shows an average altitude of 1.5 km, meaning that the major contribution of the extinction profile to AOD is within the altitude range 0–2.5 km, defined hereafter as the planetary boundary layer (PBL). For periods with elevated AOD, e.g. A, B and C in Fig. 8, z_aer=2, 3.5, 1 km, respectively. So z_aer>2 km is not only related to an aerosol extinction profile with low AOD.

Figure 8Time evolution from April 2015 to September 2016 of the 808 nm AOD for lidar (red) and sun-photometer (black) observations (a) and corresponding aerosol altitudes z_aer given by Eq. (2) (b). A, B, and C are the cases analysed in Sect. 5.

Download

5.1 Lidar data daily variability and comparison with sun-photometer AOD

From 14 to 15 April 2016 (case B in Fig. 8), AOD₈₀₈ varies between 0.05 and 0.3 for both the lidar and for the sun photometer (Fig. 9a). The vertical profiles of the backscatter ratio (Fig. 10) show a tripling of the aerosol content in the PBL in 12 h which is consistent with the daily variability of the sun-photometer AOD. The comparison of the two night-time lidar profiles also shows a similar increase in the aerosol content between 2.5 and 5 km altitude, which suggests that long-range transport in the aerosol layer took place above the PBL. S₈₀₈ decreases from 60–70 to 42 sr along with the AOD increase, showing that the increase in aerosol concentrations is the driving factor for the tripling of the AOD. The FLEXPART simulation (Fig. 10) shows strong PES values (>1500 s) northwest of Tomsk over the Ob industrial valley between Tomsk (56^∘ N, 85^∘ E) and Surgut (62^∘ N, 73^∘ E) for aerosols detected below 2.5 km. The strong PES values are much more scattered for the upper layer above 2.5 km with aerosol sources both from the lower Ob valley and from a large part of Kazakhstan. Indeed, according to our classification of the type of aerosol, measurements below 2.5 km have been classified as flaring on 14 April, urban on 15 April 03:00 UT and dust on 15 April 16:00 UT. Measurements above 2.5 km were classified as dust emissions. The decrease in S₈₀₈ is also consistent with a decreasing fraction of pollution aerosol when dust is advected from Kazakhstan (Burton et al., 2012; Hofer et al., 2017).

Figure 9Diurnal evolution of the lidar AOD and the sun-photometer AOD at 808 nm for A (b), B (a) and C (c) cases shown in Fig. 8.

Download

Figure 10Vertical profiles of the scattering ratio on 14 April 2016 at 02:05 and 22:30 UT, and on 15 April 2016 at 03:20 and 16:05 UT (a) and a map of the PES distribution for FLEXPART backward simulation initialized in the PBL (b) and above the PBL (c).

Download

From 30 June to 2 July 2015 (case A in Fig. 8), the measured AOD₈₀₈ values also gradually increase from 0.08 to 0.16 for both the lidar and the sun photometer (Fig. 9b). The vertical profiles of the backscatter ratio (Fig. 11) show up as in the previous case, with high values (5–7) in the PBL but lower values (≈4) above 2.5 km. S₈₀₈ increases from 35 to 47 sr, implying that the AOD increase is due both to a change in the aerosol type (40 %) and an increase in the aerosol load (60 %). PES maps indicate an origin still associated with the lower Ob valley for measurements in the PBL (Fig. 11), while high PES values are observed between Tomsk and Lake Baikal for the layer observed in the free troposphere. The Baikal region was impacted by forest fires at the end of June 2015 (see Sect. 5.2), so our classification indeed indicates biomass-burning aerosol for the layer above 2.5 km and a mixture of aerosol produced by flaring (30 June and 1 July) and by biomass burning (2 July) in the PBL. Although the S₈₀₈ increase is consistent with the advection of biomass-burning aerosol, S₈₀₈ is surprisingly low (35 sr) for the plume advected at the beginning of the period from the flaring region. One explanation is the strong daily variability of flaring emissions, which cannot be taken into account for our flaring-type attribution only based on advection from the flaring region. S₈₀₈ of 47 sr is also in the lower range of the expected value for biomass burning, in accordance with the dual air mass origin for the upper layer in Fig. 11, which implies some aerosol mixing.

Figure 11As Fig. 10 on 2 July 2015 at 14:20, 18:20 and 21:20 UT.

Download

Figure 12As Fig. 10 on 19 September 2016 at 12:30 and 16:25 UT and on 20 September 2016 at 12:25 and 14:40 UT.

Download

From 19 to 21 September 2016 (case C in Fig. 8), AOD₈₀₈ values decreased from 0.4 to 0.1 according to the sun photometer and the lidar (Fig. 9c). The vertical profiles of the backscatter ratio (Fig. 12) actually show a very strong decrease in the PBL (20 to 5), with the high values being confined in the 0–800 m altitude range. The aerosol content above 1 km is lower with R₈₀₈ ranges between 3 and 5. S₈₀₈ always remains at about 60 sr, except at the end of the 3-day period, where it drops to 40 sr for AOD equal to 0.1. The PES distributions are different from the two previous cases with a large horizontal extension of the area with strong PES values for the PBL (Fig. 12). This area includes a 500 km circle around Tomsk and two branches extending to Lake Baikal and to Kazakhstan. On the contrary, aerosol sources are now confined, for the free troposphere, to a southwest sector above Novosibirsk and Kazakhstan (Fig. 12). For the entire period of 19 to 21 September, aerosols were classified as biomass-burning aerosols due to the presence of forest fires over a large area to the east and north of Tomsk. The 60 sr high values of S₈₀₈ are consistent with the transport of the biomass burning aerosol from eastern Siberia (Burton et al., 2012), while even the S₈₀₈ drop to 40 sr is explained by the mixing with air coming from Kazakhstan.

Figure 13Five-day average of AOD at 532 nm from $\mathrm{1}{}^{\circ }\times \mathrm{1}{}^{\circ }$ MODIS observations (a, b, c) and of CO in molecules cm⁻² from IASI observations (d, e, f) for July 2015 (a, d), April 2016 (b, e) and September 2016 (c, f). The red circle is Tomsk and the thick black lines are the CALIOP overpasses shown in Figs. 14 to 16.

Download

Figure 14Latitudinal cross section of CALIOP 532 nm scattering ratio R₅₃₂ (a), aerosol depolarization ratio δ₅₃₂ (b), aerosol optical depth AOD₅₃₂ (c) and lidar ratio S₅₃₂ (d) for 14 April 2016.

Download

5.2 Satellite observations

5.2.1 Description of data products

Available satellite observations for these three periods were selected to identify the aerosol source regions. The horizontal distribution of strong AOD is documented by the 550 nm MODIS AOD maps averaged over 5 days. AOD maps are made using the Level-3 MODIS Atmosphere Daily Global Product, which contains roughly 600 statistical data sets sorted into 1 by 1^∘ cells on an equal-angle grid that spans a 24 h interval (Levy et al., 2013; Platnick et al., 2015). The role of biomass burning or fuel combustion can be described with a satellite tropospheric CO column measured, for example, by the Infrared Atmospheric Sounding Interferometer (IASI) instrument on Metop A and B. Because a large fraction of atmospheric CO is also related to the oxidation of hydrocarbons including methane, flaring will be a source of CO. The IASI CO data used in this paper have been processed at LATMOS using a retrieval code, FORLI (Fast Optimal Retrievals on Layers for IASI), developed at ULB (Université Libre de Bruxelles) by Hurtmans et al. (2012). Validation for the Siberian and Arctic regions is described in Pommier et al. (2010). The vertical distribution of aerosol layers is inferred from CALIOP overpasses. In this work 532 nm backscatter and depolarization ratios are calculated using the CALIOP level 1 (L1) version 4.10 attenuated backscatter coefficients because they correspond to a better calibration of the lidar data (Vaughan et al., 2012; Winker et al., 2009). They are averaged using a 10 km horizontal resolution and a 60 m vertical resolution. Before horizontal or vertical averaging, the initial 333 m horizontal resolutions (1 km above the altitude 8.2 km) are filtered to remove the cloud layer contribution. This cloud mask makes use of the version 3 level 2 (L2) cloud layer data products (Vaughan et al., 2009) and measurements of the IR imager on the CALIPSO platform. Our scheme for distinguishing between cloud and aerosol is described in Ancellet et al. (2014). To calculate the extinction profile and the optical depth, we use the lidar ratio S₅₃₂ from the CALIOP version 3 L2 aerosol layer data products (Omar et al., 2009), unless we can calculate the aerosol layer transmittance to constrain S₅₃₂. To reduce the error when using high horizontal resolution CALIOP profiles, the attenuated backscatter is averaged over 80 km to compute the layer transmittance whenever it is possible. The aerosol depolarization ratio δ₅₃₂ is also calculated using the perpendicular- to the parallel plus perpendicular polarized aerosol backscatter coefficient (see Appendix B). Whenever it is possible, the use of night-time overpasses are preferred to improve the signal-to-noise ratio (SNR).

Figure 15Same as Fig. 14 for 2 July 2015.

Download

5.2.2 Results

From 14 to 18 April 2016, the AOD MODIS and CO IASI maps (Fig. 13b, e) show maxima around the town of Tomsk and more generally in the lower Ob valley (only for IASI insofar as the cloud cover and snow cover do not allow MODIS to be used above 58^∘ N). No forest fires were detected during this time period and a predominant role of flaring emissions seems a likely hypothesis for the aerosol layers observed at Tomsk. A CALIPSO overpass with low cloud cover between 50 and 60^∘ N is available on 14 April 2016 (thick black line in Fig. 13). The AOD₅₃₂ observed by CALIOP (Fig. 14) in the range 0.05–0.15 and the associated backscatter ratio (≈2) are lower than the highest values observed by the Tomsk lidar (AOD₈₀₈≈0.3, e.g. corresponding to AOD₅₃₂≈0.4 using the sun-photometer AE =0.87), but it is consistent with the range 0.07–0.4 of AOD₅₃₂ when using the AOD₈₀₈ observed by the TOMSK lidar. The CALIOP AOD is also lower than the 5-day average MODIS AOD₅₅₀ (≈0.5) near Tomsk (Fig. 13b), because the CALIOP track was at the edge of the MODIS AOD maxima. The CALIOP observations, however, provide the vertical (0–2 km) and latitudinal (52 to 57^∘ N) extent of the aerosol layer due to flaring/urban emissions (Fig. 14), similarly to the Tomsk lidar observations. At latitude below 52^∘ N, an aerosol layer is identified as dust by CALIOP with AOD₅₃₂≈0.2, an upper boundary up to 3 km and depolarization ratio >12 %. The lidar ratio attributed by CALIOP is 55 sr, being consistent with dust emission from Kazakhstan that is responsible for the increasing AOD and advection of the aerosol layer observed in Tomsk above 2.5 km on 15 April 2016.

Figure 16Same as Fig. 14 for 20 September 2016.

Download

From 28 June to 2 July 2015, the MODIS AOD and IASI CO maps (Fig. 13a, d) indicate two maxima with both elevated AOD₅₅₀ and CO values: a forest fire zone of 3.10⁵ km² at 51^∘ N, 97^∘ E (AOD₅₅₀>0.7, i.e. AOD₈₀₈≈0.3 with AE =2) and the flaring zone in the lower Ob valley between 56 and 65^∘ N (AOD₅₅₀≈0.3, i.e. AOD₈₀₈≈0.13 with AE =2). This is in rather good agreement with our analysis of aerosol sources, which indicate a mixture of fire and flaring emissions for aerosol layers observed below 2.5 km at Tomsk and the role of fires in the free troposphere above 2.5 km. There is only one CALIPSO overpass on 2 July 2015 (thick black line in Fig. 13) across the fire plume west of Lake Baikal. Elevated AOD₅₃₂>0.5 are indeed observed by CALIOP at 54^∘ N, 97^∘ E in smoke layers with very low depolarization ratio (<5 %) and backscatter ratio >5 up to an altitude of 6 km (Fig. 15). The corresponding AOD at 808 nm of the order of 0.18–0.45, when using the 1.9 sun-photometer AE over Tomsk on 2 July 2015, shows that the Tomsk lidar AOD is 2 times lower after being transported from Lake Baikal and mixed with background aerosol. It also explains the 47 sr moderate S₈₀₈ for a biomass-burning event as discussed in Sect. 5.1. The satellite data analysis for July 2015 is therefore consistent with the results of the Tomsk lidar data processing, both for the AOD range and for the aerosol-type assumption and related lidar ratio.

From 17 to 21 September 2016, the AOD MODIS and CO IASI maps show a very large area impacted by the numerous forest fires (see https://www.fire.uni-freiburg.de/GFMCnew/2016/09/28/20160928_ru.html, last access: 24 November 2017) that took place in Siberia in September 2016. MODIS AOD₅₅₀>0.7 and CO columns $>\mathrm{3}\times {\mathrm{10}}^{\mathrm{18}}$ mol cm² are observed over an area of 1000 km × 1000 km at 57–67^∘ N, 95–115^∘ E Fig. 13c, d). Tomsk lies just at the edge of this wide plume. The influence of biomass-burning aerosol found in our analysis of Tomsk lidar observations is linked to this event. The CALIPSO track passing over Tomsk and over the fire plume between 56 and 70^∘ N (black line in Fig. 13c, d) shows AOD₅₃₂ in the range 0.4–0.8 between 56 and 60^∘ N (Fig. 16). The CALIOP AOD increase corresponds to the MODIS AOD anomaly, although it is 30 % lower than the average MODIS AOD maxima (Fig. 13c). The MODIS AOD is consistent with AOD₈₀₈>0.4 observed in Tomsk, i.e. a corresponding AOD₅₅₀=0.75 using AE =1.5 as measured by the Tomsk sun photometer during this time period. The CALIOP depolarization ratio (7 %) is higher than for the July 2015 fire event indicating soil aerosol vertical transport simultaneously with the production of biomass-burning aerosol for these late summer fires (Nisantzi et al., 2014). Similarly the CALIOP lidar ratio (≲70 sr) and the sun-photometer AE (1.5) are lower than the values obtained for the July 2015 fires (S₅₃₂≈75 sr, AE =1.9) even if these values remain characteristic of a combustion aerosol. The corresponding S₈₀₈≲53 sr for the 20∕9 CALIOP cross section is lower than S₈₀₈=60 sr measured for the biomass-burning plume in Tomsk, but the difference is well within the 10 sr expected uncertainty for the Tomsk lidar S₈₀₈ and the known uncertainties for the CALIOP lidar ratio assessment (Omar et al., 2009). It is also interesting to see that the vertical extent of the fire plume observed by CALIPSO remains fairly low (<1.5 km) between 55 and 60^∘ N, i.e. an aerosol plume thickness similar to the Tomsk lidar measurement.

Figure 17PDF of the 808 nm AOD lidar according to the different aerosol types: all (a), urban (b), flaring (c), natural emissions from Siberian forest and grasslands (d), biomass burning (e) and dust (f). Blue PDF is for aerosol-weighted altitude $z_{\mathrm{aer}}<=\mathrm{1.75}$ km and red for z_aer>1.75 km. N_PBL and N_FT are respectively the number of PBL only and PBL plus FT observations, while p50th and p90th are respectively the median and 90th percentile of the AOD distribution for both altitude ranges.

Download

The overall conclusion of this section (Sect. 5) is that (1) our approach to attributing an aerosol type to Tomsk lidar observations is validated by a more in-depth study of aerosol sources based on available satellite observations, (2) the AOD and lidar ratio calculated for the Tomsk lidar observations are comparable to the sun-photometer daily AOD variability and satellite AODs in the aerosol source regions identified by the FLEXPART analysis. This will allow the statistical analysis of the lidar measurements according to aerosol types for the 18-month database.