Based on the 328 aerosol profiles, we analyzed the annual cycle of aerosol layering in the lower, middle, and upper troposphere over Dushanbe. By visual inspection, we found two main regimes: (a) the main aerosol layer that typically extends from the surface to about 3–6 km height and contributes to 500 nm AOT by usually more than 90 % and (b) frequently occurring thin dust layers between 5 and 10 km height that mainly contained aerosol from remote source regions such as the Arabian deserts and Sahara. We call the top height of the highest layer containing dust the uppermost aerosol layer top in the following discussion. A measurement example of this layering is shown in Fig. .

Besides the visual inspection of the lidar backscatter profiles, we tested several automated top-height detection methods. The most useful approaches (three in total) are considered in Fig. b and c. The first technique searches for the height (of the main layer) at which the backscatter coefficient (bsc) at 1064 nm wavelength drops below a threshold value of 2.5×10-5 m-1 sr-1 (in Fig. b) for the first time above a starting height of, for example, 500 m above the lidar. The second approach analyses the 1064 nm backscatter ratio (ratio of total-to-Rayleigh backscatter) profile and uses a threshold value (bsc ratio) of 1.8 (in Fig. c). The agreement between the different results is good.

In Fig. b, the blue dashed horizontal line shows the height level at which the integrated backscatter coefficient (IB, column backscatter) from the surface up to this specific height reaches 90 % of the total column backscatter value IB (third method). The 90 % IB height level of about 4 km means that most of the aerosol is in the main aerosol layer reaching up to about 5 km on this day. The depolarization ratios for 355 and 532 nm in Fig. c are far below the depolarization ratios for pure dust of 0.25 (355 nm) and 0.30 (532 nm) and thus indicate a mixture of mineral dust and air pollution.

Figure  provides an overview of the 18-month observations in terms of top heights of the main aerosol layer and the uppermost lofted aerosol layer, obtained from the visual inspection. In 184 cases out of the 328 available nighttime observations, a lofted aerosol layer (above the main layer) was present. As can be seen, the top height of the main layer increases from about 2.5–4.5 km in March to about 4–6.5 km in July 2015. From August 2015 the top heights decrease to minimum heights of about 1–2 km in December 2015 and January 2016. During February and March 2016 the top heights increased steeply. Later in the year 2016, the top heights are rather variable from day to day.

The top heights of the detected uppermost dust layer also vary strongly and indicate the frequent occurrence of dust traces up to the upper troposphere from late winter (February) to autumn (October). Such lofted layers are seldom during November to January. Although optically thin, dust layers in the middle and upper troposphere may have a sensitive impact on ice formation in mixed-phase and ice clouds . Figure  provides a statistical overview of the top heights of the main and uppermost layer for the entire 18-month measurement period.

Table  summarizes the seasonal mean top heights of the main aerosol layer and the highest detected layer in the troposphere. The results obtained by case-by-case visual inspection and by applying the automated retrievals for the main layer depth are in good agreement (mostly within a range of 10 % deviation). The seasonal means for the 90 % IB level height indicate that most of the time the aerosol within the main layer (Fig. b) contributes 90 % or even more to the overall IB or AOT (as will be shown later). As a consequence, we may conclude from the height-resolved observations that many snow-covered (glaciated) regions of Central Asia (Pamir mountains, Tien Shan) located at heights below 5 km are continuously exposed to dust and aerosol pollution during the summer half year.

Figure  provides an overview of the main air flow and aerosol transport towards Dushanbe. During the winter season (Fig. a), local sources and regional aerosols (clusters 1, 2, and 5, 84 %) contribute to the aerosol conditions in the main aerosol layer over Dushanbe. Source regions are northern Afghanistan, southwestern Tajikistan, Turkmenistan, southern Uzbekistan, and the region downwind the Caspian Sea and the Aralkum Desert.

During the summer season, again regional air mass transport prevails (Fig. b, clusters 2, 3, and 4, 80 %) in the main aerosol layer. Source regions are Uzbekistan, western Tajikistan, Aralkum Desert, and southern Kazakhstan. Regarding the upper tropospheric air mass transport, Fig. c shows that the aerosol originates mainly (62 %) from Middle Eastern deserts (cluster 5) and northern Africa (clusters 2 and 4) but also from polluted Mediterranean (cluster 1) and eastern European regions (cluster 3). This also holds for the summer season.

The cluster analysis suggests that the air masses are transported further to the east, crossing eastern Asia, continuously diluting, but mixing with, new dust and pollution over China, and traveling across the Pacific. The upper tropospheric dust and aerosol pollution mixtures as observed over the lidar station at Dushanbe will become part of the northern hemispheric upper tropospheric aerosol background reservoir that influences cirrus formation and precipitation processes on continental to hemispheric scales.

The characterization of the aerosol optical properties is based on the lidar and AERONET sun photometer observations. Figure  provides an overview of the basic optical properties obtained from the lidar measurements. For the comparison with AERONET products (column-integrated values), the lidar-derived 532 nm AOT was determined from the particle extinction profile in Fig. b. By means of the Raman lidar method the 355 and 532 nm particle extinction coefficients are directly computed from the respective 387 and 607 nm nitrogen Raman signal profiles for heights >1 km above the lidar. The large uncertainty in the correction of the incomplete laser-beam receiver field-of-view overlap prohibits a trustworthy extinction coefficient retrieval from the nitrogen Raman signal profiles in the near range, i.e., for the lowest 1000 m above the lidar. To extend the particle extinction profiles towards the ground, we use the 532 nm particle backscatter coefficient in Fig. a. This quantity is obtained from the ratio of the elastic backscatter signal to the respective nitrogen Raman backscatter signal so that overlap effects widely cancel out. The 532 nm particle backscatter coefficients are trustworthy down to about 100 m above the lidar. We estimate the respective particle extinction coefficient for heights below 1–1.5 km by multiplying the backscatter coefficient with the lidar ratio of the actual measurement, in the present case measured at about 1.4 km height as shown in Fig. c and indicated by a dashed magenta line. The 532 nm AOT is finally obtained by the integration of the entire extinction profile up to 6 km height. In this way, we obtained 276 nighttime extinction profiles. Residual AOT contributions from heights above 6 km were usually <0.02.

To check the accuracy of lidar-derived AOT values, we compared the lidar-derived AOTs with respective AERONET 500 nm AOTs measured with the sun photometer in the afternoon, preferably close to sunset. For this purpose, 192 lidar extinction profiles were computed. The averaging time ranged from 15 min to 1 h 40 min with an average of about 1 h. The temporal distance to the last AERONET measurement ranged from 2 h to about 5 h. Figure  shows the comparison. The agreement is acceptable. A small bias is observed and most probably related to the fact that the AERONET observations are performed at relatively low sun elevation angle across the city center of Dushanbe and lidar observations were performed in the vertical direction. The combined AERONET and lidar field site was located in a less urbanized area, about 4 km east of the city center. The variability in the data is caused by temporally and spatially varying aerosol conditions and variations in the wavelength dependence of AOT in the 500–532 nm range and thus aerosol size distribution changes. The uncorrected wavelength dependence can cause differences of the order of 5 %, only. It is interesting to note that a similar bias (deviation from the Dushanbe AERONET observations) was found when comparing MODIS (Moderate Resolution Imaging Spectroradiometer) long-term observations of 550 nm AOT with the AERONET data shown in the supplementary material of . One of the reasons could be an overestimation of multiple scattering effects (caused by mineral dust) in the AERONET data analysis.

AOT histograms obtained from all 276 Dushanbe lidar nighttime profiles and all sun photometer AOT measurements (from March 2015 to August 2016) are shown in Fig. . Similar AOT distributions are observed with both instruments. Most AOTs are <0.3. Large values (>0.5) indicate dust outbreak events. The specific value of a polarization lidar is the potential to separate the dust from non-dust backscattering and thus to obtain an accurate estimate of the dust optical thickness (DOT). The distribution of DOT is given in Fig. c. Most DOT values are ≤0.2. This is in agreement with findings of . These authors concluded from in-depth analysis of long-term spaceborne passive remote sensing over Central Asia that most DOTs are below 0.2 at 550 nm. Higher AOTs over western than over eastern Tajikistan were observed, and, as a general finding, more dust was observed in the western parts of Central Asia (closer to the Caspian Sea) than in the eastern parts.

analyzed satellite (MODIS) observations from 2002 to 2017 and found for the southern part of Central Asia (for the two southern countries Turkmenistan and Tajikistan; see Fig. ) seasonal mean 500 nm AOT values of 0.18 (Tajikistan) to 0.22 (Turkmenistan) in spring, 0.2–0.22 in summer, 0.16–0.18 in autumn, and 0.14–0.17 during the winter months. In contrast, the Dushanbe AERONET observations revealed seasonally mean values of 0.19 (spring), 0.36 (summer), 0.23 (autumn), and 0.15 (winter) for the CADEX period from March 2015 to August 2016. The lidar observations provided seasonal mean DOT/AOT ratios of 0.5 in spring, 0.8 in summer, 0.6 in autumn, and 0.1 in winter. Thus, a considerable part of the AOT is caused by anthropogenic fine-mode aerosol pollution. Our findings are in reasonable agreement with the MODIS observations of the Ångström exponent with seasonally mean values of 1.0–1.2 for Turkmenistan and 1.3–1.4 for Tajikistan. From the Dushanbe 2015–2016 AERONET photometer observations, we obtained, however, lower seasonal mean Ångström exponents (describing the AOT wavelength dependence in the spectral range from 440 to 870 nm), namely 0.7 (spring), 0.5 (summer), 0.8 (autumn), and 1.15 (winter). The dust and non-dust aerosol fractions of the total aerosol burden is further discussed in the next sections.

Figure  shows the different contributions of different height ranges to the measured DOT. For AOT (not shown), the features are similar. In Figs.  and , it is shown that the main aerosol layer reaches, on average, to 4–5 km height during the summer half year. The planetary boundary layer (convective mixing layer) typically covers the lowermost 1.5 km (red AOT contribution) and contributes about 50 % to AOT in spring and summer and 80 % during the winter months. The remaining part of the main aerosol layer (from about 1.5 to 4.5 km height) causes almost the entire residual AOT contribution. Only 10 % of the AOT is caused by particles in the middle and upper troposphere during spring.

The 18-month climatology for the 532 nm extinction coefficient is shown in Fig. . The figure is based on the 276 height profiles discussed above. Typical particle extinction values are 25–50 Mm-1 in spring and autumn, 50–100 Mm-1 in summer, and 50–150 Mm-1 during the winter season. According to the shown seasonal mean extinction profiles, the main aerosol layer reaches up to 5–5.5 km in spring, summer, and autumn and to about 2 km in winter. A moderate atmospheric variability in terms of particle extinction is observed in spring, autumn, and winter, but a strong variability is found during the summer season. This is the result of partly major dust storms with extreme particle extinction values of the order of 1500 Mm-1 and related horizontal visibilities of 2 km or less. Near-surface extinction values are highest in winter during the domestic heating period. The pollution is then trapped in the shallow aerosol layer with a depth of 2 km only.

The POLIPHON method permits the conversion of dust and non-dust extinction coefficients into height profiles of dust and non-dust mass, CCN, and INP concentration. The full retrieval procedure, starting from the basic data sets of 532 nm particle backscatter and linear depolarization ratio profiles is described in Sect.  and shown in Fig. . Recently, the required conversion parameters for mineral dust were updated and now include Central Asian and Middle Eastern dust conditions . The measured particle depolarization ratio at 532 nm was close to 0.3 at heights above 3 km and indicated the presence of an almost pure dust layer up to 8 km height. Only in the lower part (below 2.5 km height) the depolarization ratio dropped below 0.2 and indicates a mixture of mineral dust, aerosol pollution (urban industrial particles, biomass burning smoke), and continental background aerosol. This case was already discussed by .

In the shown example, the dust mass concentration was low (<25 µg m-3) in the polluted layer and >150 µg m-3 in the center of the lofted dust layer. The mass concentration of continental aerosol pollution was much lower with values <5 µg m-3 throughout the troposphere. The estimated profiles for the dust and non-dust CCN concentrations show values of up 300 cm-3 in the center of the lofted dust plume and a total CCN concentration of about 150 cm-3 in the polluted layer below 2 km height (see Fig. c). By means of the derived height profiles of dust particle number concentration considering particles with radii >250 nm only and the dust particle surface area concentration in Fig.  together with the respective GDAS temperature profile, the profile segments for the INP concentrations are obtained in Fig. e. We distinguish profiles relevant for immersion freezing and deposition nucleation of ice crystals . Immersion freezing dominates in mixed-phase clouds at temperatures >-30∘C, whereas deposition nucleation is the relevant heterogeneous ice nucleation process at temperatures <-30∘C, e.g., in cirrus layers. We only show the dust INP values because non-dust aerosol components (such as soot particles) are inefficient INPs at the given temperatures.

Figures , , and summarize the findings regarding dust and non-dust mass concentrations and dust mass fraction for the 18-month lidar campaign. According to Fig. , typical (mean) dust mass concentrations in the lowermost 2.5 km of the atmosphere are 100 (spring), 200–600 (summer), 100–300 (autumn), and 20–50 µg m-3 (winter). The season mean values for anthropogenic haze and biomass burning smoke are an order of magnitude lower with typical values (up to 2.5 km height) of 10–20 µg m-3 (spring to autumn) and 20–50 µg m-3 during the domestic heating period.

Figure  shows the mean dust mass fraction as a function of height. Each mean profile is based on a different set of single profiles. In each of the eight computations, we considered only profiles with significant particle backscatter and depolarization ratio up to the given top height from 3 (in panel a) to 10 km (in panel h). This means we did not consider measurements when the backscatter coefficient indicated clear air in the upper part of the profile, i.e., below the defined top height. The aim was to determine the mean dust fraction in the occurring dust layers. By excluding all cases with clear air in the upper part of the averaging height range, the number of considered night-by-night observations decreases from 310 profiles (out of a total set of 328 profiles) in the case of the top height of 3 km (Fig. a) to nine profiles when considering and averaging all profiles that show dust from the surface up to 10 km height (Fig. h). As can be seen, the profile-mean dust mass fraction is in most cases 70 %–80 %. The dust mass fraction typically decreases with height from 90 % to about 50 % in the uppermost part of the profile. The other way around, the anthropogenic aerosol mass fraction is always of the order of 20 %–30 % in the main aerosol layer up to 4–5 km height, and also in the lofted dust layers higher up (5–10 km height range). This is consistent with the DOT and AOT mean values and DOT/AOT ratios discussed above.

Figure  highlights the decreasing number of dust cases (available for dust profile averaging up to a given top height) with increasing height. Figure  is similar to Fig. , but considers a higher resolution concerning the defined top heights in the averaging procedure. The main result is a very smooth, decreasing curve for the frequency of occurrence of dust layers. In terms of the number of observed dust cases, close to 100 % (with 80 % mean dust mass fraction) out of all 328 nighttime profiles show dust up to 3 km top height, as already mentioned above; 65 % in the case of the 5 km top height (mean dust fraction of 74 %); and only a few cases of dust (<10 %) when the defined top height is >9 km.

Figures  to summarize the results concerning the cloud-relevant dust and non-dust aerosol parameters. As a general impression, CCN and INP concentrations decrease with height because of the decreasing dust particle number concentration. To facilitate the comparison of CCN and INP concentration levels for the different seasons, we define typical height ranges as representative reservoirs for CCN and INPs. The 3–4 km height range can be regarded as the main CCN and immersion-freezing INP reservoir for convective cloud formation (liquid-water and mixed-phase clouds) over Dushanbe during the spring, summer, and autumn seasons. The INP levels at 7–8 km height may be representative of a deposition INP reservoir in the upper troposphere and thus relevant for cirrus formation. Table  summarizes the layer mean values of the cloud-relevant aerosol quantities for the 3–4 and 7–8 km height ranges and for the different seasons.

The results for CCN concentrations in Fig.  show that dust particles lead to mean dust CCN concentrations of the order of 50 cm-3 (spring), 100 cm-3 (summer), and 30–50 cm-3 (autumn) and non-dust CCN concentrations of the order of 200 cm-3 (spring), 400 cm-3 (summer), and 250 cm-3 (autumn) in the 3–4 km height layer during the summer half year (spring to autumn). The shown SD values (difference between the solid and dotted lines in Fig. ) indicate an atmospheric variability according to a factor of 2 (thus a value range from almost zero to 2× mean). For comparison, summarized European CCN observations (for a supersaturation values of 0.2 %) and compared them with other measurements in Alaska and Amazonia. They found mean CCN concentrations of 1200–1500 cm-3 in central Europe, 100 cm-3 at the Atlantic coast in Ireland, <100 cm-3 in Alaska during the summer half year, and 100 cm-3 in the wet season to 700 cm-3 in the dry season in the Amazon, Brazil. reported CCN concentrations of 140–270 cm-3 in the lofted Saharan air layer (containing pure dust) over Barbados in the Caribbean during the summer season of 2013 and 400–500 cm-3 in spring (March 2014) for a mixture of dust and pollution over Barbados originating from Africa. Thus, the CCN levels over Dushanbe indicate moderately polluted continental aerosol background conditions.

Figure  provides an overview of the observations of dust properties that are used as aerosol input in the INP concentration retrievals. Dust is the main INP-relevant aerosol component. The uncertainty in these input profiles is only 30 %. The large uncertainty in the INP concentration estimates of a factor of 3–5 is caused by the use of the published INP parameterization schemes. Figure  shows seasonal mean INP concentration profiles for an air temperature of -25 ∘C relevant for immersion freezing and for -50 ∘C and thus a temperature relevant for deposition nucleation of ice crystals.

The seasonal mean values in the 3–4 km aerosol layer range from 0.9 to 3.3 cm-3 (n250,d), 20–75 µm2 cm-3 (dust particle surface area concentration), and 2.3–12 L-1 (immersion freezing INP concentration), during the three non-winter seasons (spring to autumn). The atmospheric variability is again roughly described by a factor of 2. In the case of cloud formation and the evolution of the ice phase, significant lifting of cloud parcels and cooling of the air parcels occur. Note that the INP concentration then increases by an order of magnitude for the given aerosol concentration when the air temperature decreases by 5 K triggered, for example, by the strong lifting in updrafts.

To compare the Dushanbe INP concentration levels with the ones in other dust regions, we checked the literature. performed airborne in situ measurements and found INP concentrations (immersion freezing, -25 ∘C) from 10 to 1000 L-1 in dust layers from 300 to 3500 m height over Cabo Verde in the summer of 2015 in the outflow regime of Saharan dust, less than 1000 km west of the African coast. The maximum values of the large particle number concentration (coarse mode fraction) reached 200 cm-3 and the maximum particle surface area concentrations were as high as 1500 µm2 cm-3. reported in situ measured values of n250,d=15–20 cm-3 and INP concentrations (immersion freezing, -25 ∘C) of 60–100 L-1 in a lofted dust layer around 2 km height over Cabo Verde in July 2011. discussed observations taken in the Saharan dust layer between 2 and 4 km height in the Caribbean in the summer of 2013, more than 5000 km west of the African dust sources, and found INP concentrations of 30–60 L-1 (immersion freezing, -25 ∘C) and 10 L-1 in the spring season (March 2014, 2–3 km height). reported a mixed-phase cloud layer (at 5–6 km height) occurring in desert dust over Cyprus in the Eastern Mediterranean in March 2015, and the INP concentration was estimated to be of the order of 1 L-1 at -20 ∘C, which means about 10 L-1 at -25 ∘C. Thus the seasonal mean Dushanbe INP concentration levels at 3–4 km height indicate typical INP conditions in regions influenced by long-range transport of dust.

The INP reservoir between 7 and 8 km height contains significant amounts of INPs over Dushanbe only during the spring and summer seasons. The atmospheric variability is high and characterized by a factor of 4–6 (1 positive SD). The seasonal mean particle surface area concentrations ranged from 0.7 to 3.3 µm2 cm-3 in spring and summer and were around 0.057 µm2 cm-3 in autumn, and they indicate seasonal mean INP concentrations (deposition nucleation, -50 ∘C, ice supersaturation of 1.15) of 2.4 (spring), 0.55 (summer), and 0.041 L-1 (autumn). For comparison, found over Cyprus in lofted dust at cirrus level (10–11 km height, -50 ∘C) INP concentrations (deposition nucleation for an ice supersaturation value of 1.1) of 5 L-1 and corresponding particle surface area concentrations around 30 µm2 cm-3. The dust had a strong influence on the cirrus evolution and life cycle. Thus, during times with strong dust advection to Dushanbe at greater heights, predominantly during spring and summer, a significant impact of dust on ice formation in the upper troposphere can be expected.

As a final result, we present the INP concentration profiles for the found aerosol conditions in combination with the actual pressure and temperature profiles from in Fig.  to give an impression of typical (actual) INP concentration values for an ice supersaturation of 1.15. It can be seen that only in spring and summer significant levels of INP concentrations (0.01–1 L-1) occurred in the 6–8 km height range (mixed-phase cloud and immersion freezing regime) and from 8 to 10 km height (ice cloud and deposition nucleation regime). In autumn and winter, the seasonal means indicate a rather low potential for heterogeneous ice formation at the given height levels.