Dust mass, CCN, and INP profiling with polarization lidar: Updated POLIPHON conversion factors from global AERONET analysis

The POLIPHON (Polarization Lidar Photometer Networking) method permits the retrieval of particle number, surface area, and volume concentration for dust and non-dust aerosol components. The obtained microphysical properties are used to estimate height profiles of particle mass, cloud condensation nucleus (CCN) and ice-nucleation particle (INP) concentrations. The conversion of aerosol-type-dependent particle extinction coefficients, derived from polarization lidar observations, into the aerosol microphysical properties (number, surface area, volume) forms the central part of the POLIPHON 5 computations. These conversion parameters are determined from Aerosol Robotic Network (AERONET) aerosol climatologies of optical and microphysical properties. In this article, we focus on the dust-related POLIPHON retrieval products and present an extended set of dust conversion factors considering all relevant deserts around the globe. We apply the new conversion factor set to a dust measurement with polarization lidar in Dushanbe, Tajikistan, in central Asia. Strong aerosol layering was observed with mineral dust advected from Kazakhstan (0-2 km height), Iran (2-5 km), the Arabian peninsula (5-7 km), 10 and the Sahara (8-10 km). POLIPHON results obtained with different sets of conversion parameters were contrasted in this Central Asian case study and permitted an estimation of the conversion uncertainties.

in addition (Kanitz et al., 2013;Bohlmann et al., 2018). However, as mentioned the POLIPHON technique can be applied to any available polarization lidar observation around the world.
The paper is organized as follows. A brief overview of the POLIPHON methodology is given in Sect. 2. with focus on mineral dust and the determination of dust conversion factors from worldwide AERONET observations. We analyzed longterm sun/sky photometer observations of 20 AERONET sites in or close to important mineral dust source regions around the 5 world (AERONET, 2019). The stations are shown in Fig. 1. The results (conversion factors) of the in-depth AERONET data analysis are presented in Sect. 3. In Sect. 4, we discuss a Polly observation at Dushanbe, Tajikistan, with mineral dust up to the tropopause advected from Central Asia (at heights below 2 km above ground), from Iran and the Arabia peninsula (2-7 km height range), and the Sahara (above about 8 km height). The case study is used to demonstrate the full potential of the POLIPHON method for mineral dust profiling with the updated AERONET-based dust conversion factors and also how to 10 estimate the conversion uncertainties in the POLIPHON products. Concluding remarks are given in Sect. 5.

Summary of the POLIPHON method with focus on dust
The POLIPHON method is described in detail by Mamouri and Ansmann (2014 and with respect to the INP concentration retrieval also by Marinou et al. (2018). The main part of the POLIPHON data analysis deals with the conversion 15 of aerosol-type-dependent particle extinction coefficients into respective particle microphysical properties. Table 1 provides an overview of POLIPHON dust products and the respective conversions. Similar conversions for non-dust aerosols such as maritime particles or continental fine-mode aerosol pollution (urban haze, biomass burning smoke) can be found in Ansmann (2016, 2017).
In the first part of the POLIPHON data analysis, the polarization lidar observations are analyzed to obtain height 20 profiles of dust and non-dust backscatter coefficients. Here we assume that pure dust causes particle linear depolarization ratios of 0.3-0.35 around the globe, disregarding the dust source region. This is corroborated by numerous studies (see the reviews in Tesche et al. (2009); Ansmann (2014, 2017)) and also during recent field campaigns (Groß et al., 2015;Veselovskii et al., 2016;Haarig et al., 2017;Hofer et al., 2017). More details to the aerosol type separation procedure (including the separation of fine and coarse dust by the use of fine-mode and coarse-mode-related 25 depolarization ratios) can be found in Mamouri and Ansmann (2017). The derived total, fine, and coarse dust backscatter coefficients β d , β df , and β dc are then converted to respective dust extinction values σ d , σ df , and σ dc by means of appropriate dust extinction-to-backscatter ratios or lidar ratios S d , S df , and S dc . As shown in Table 2, the 532 nm dust lidar ratio S d may vary from about 30 to 60 sr for different mineral dust types (Müller et al., 2007;Tesche et al., 2011;Mamouri et al., 2013;Nisantzi et al., 2015;Groß et al., 2015;Veselovskii et al., 2016;Haarig et al., 2017;Hofer et al., 30 2017;Shin et al., 2018). However, for most dust regions, except the western Sahara, the typical dust lidar ratio is 40 sr at 532 nm. We therefore recommend to use 40 sr as dust lidar ratio and to select 50 sr only in cases with airflow from the western Sahara. The best option is however to use actual Raman lidar observations of the dust lidar ratio. We further assume that S d = S df = S dc (see lines 2-4 in Table 1). A relative uncertainty in the dust lidar ratio assumptions of 10% is considered in the estimation of the relative uncertainties (error propagation) in Table 1.
In the second part of the POLIPHON data analysis (see Table 1, lines [5][6][7][8][9][10][11][12][13][14][15], the height profile of the dust mass concentration M d (z) is derived from the dust extinction coefficients σ d (z), also separately for coarse dust (M dc considering particles with radius>500 nm) and fine dust (M df considering dust particles with radius<500 nm) from respective coarse and fine dust 5 extinction coefficients σ dc and σ df . The dust extinction coefficients are converted into dust particle volume concentrations v d , v dc , and v df by means of extinction-to-volume conversion factors c v,d , c v,df , and c v,dc , and afterwards multiplied by the dust particle density ρ d of 2.6 g/cm −3 (Ansmann et al., 2012) to obtain the respective dust mass concentrations. The required conversion factors are determined from AERONET observations as described in Sects. 2.2 and 3.1.
Further POLIPHON conversion products listed in Table 1 (lines 8-15) are needed in the estimation of the cloud-relevant 10 aerosol parameters such as the cloud condensation nucleus concentration (CCNC) and ice-nucleating particle concentration (INPC). The number concentrations n 100,d (considering particles with radius >100 nm) is a good proxy for the dust CCNC (Mamouri and Ansmann, 2016;Lv et al., 2018). However, CCNC depends on the water supersaturation at cloud base where aerosol particles mainly enter the cloud and serve as CCN. A typical water supersaturation value is 0.2% (Siebert and Shaw, 2017) and occurs when air parcels are lifted into the base of a liquid water cloud by weak updrafts, e.g., in the case of fair 15 weather cumuli. Water supersaturation values may exceed even 1% in strong updrafts. For the conversion of σ d into number concentration n 100,d , the conversion parameters c 100,d and exponent x d as shown in Table 1 are used and obtained from the AERONET observations (see Sects. 2.2 anbd 3.2).
We introduce the factor f ss,d to consider the water supersaturation dependence. With increasing supersaturation at cloud base an increasing number of dust particles (i.e., particles with lower radius) can be activated as CCN. For a 20 supersaturation value of 0.4% even dust particles with radius of 70-80 nm become activated. According to Shinozuka et al. (2015), f ss,d = 2 is appropriate when using n 100,d as the basic aerosol parameter in the CCNC estimation but the supersaturation is 0.4% (see Mamouri and Ansmann (2016) for more details). For completeness, in Table 1, f ss,d is 1.0, and the respective equation holds for a liquid-water supersaturation level of 0.2%.
The particle number concentration n 250,d (considering particles with radius >250 nm, Table 1, line 9) and the dust  surface-concentrations, the temperature profile T (z) and an assumed ice supersaturation value S ice (in the case of deposition-freezing INPC, U17-D) are input in the INPC estimation. The ice supersaturation S ice is set to a typical value of 1.15.
We introduce a new parameter (not considered in Mamouri and Ansmann (2016)), namely the surface area s 100,d as an alternative input parameter in the estimation of immersion freezing INPC. In the case of immersion freezing, liquid 35 droplets form first before freezing occurs. As discussed above, appropriate dust CCN for typical water supersaturation values of 0.2% have a radius >100 nm. Only these particles (immersed in the liquid droplets) can then serve as INP so that the surface area s 100,d may be a more appropriate aerosol proxy in the INP estimation by using the immersion freezing parameterization U17-I (Ullrich et al., 2017) than the total surface area concentration s d . However, both parameters (s d , s 100,d ) are required in the INP parameterization. For deposition nucleation (heterogeneous ice nucleation 5 by water vapor deposition directly on dust particles, without any liquid phase formation), s d is the relevant aerosol input parameter. All this is described in detail in Mamouri and Ansmann (2016). More details to the INPC retrieval are also given in Sect. 4. Table 1 also provides an overview of the uncertainties in the POLIPHON products Ansmann, 2016, 2017). The very large uncertainties in the estimation of n 100,d , n CCN , and n INP,d (factor of 2-5) are obtained when 10 taking all potential error sources into consideration. INPC parameterizations developed from field observations (for sometimes not well characterized aerosol types) and from laboratory experiments with fresh dust particles rather than aged, i.e., chemically and cloud processed dust particles (as they frequently occur in the atmosphere), must always we be handled with care and may not be fully applicable to atmospheric conditions with predominantly aged dust so that uncertainties of the order of a magnitude can not be excluded. However, meanwhile a variety of studies indicate

POLIPHON dust conversion parameters
Trustworthy and climatologically robust conversion parameters obtained from AERONET observations are of central importance for the applicability and attractiveness of the POLIPHON method. For our study, we downloaded the following data sets of AERONET products (single measurements, inversion products, version 3, level 2.0) (AERONET, 25 2019): 1) The particle volume size distribution resolved in 22 size classes from 50 nm (bin 1) to 15 nm (bin 22), 2) the corresponding data sets of total, fine-mode, and coarse-mode-related volume concentrations and effective radii (from which also surface area concentrations can be calculated), and 3) the corresponding AOTs for 8 wavelengths (denoted as extinction AOT in the AERONET data base) together with respective Ångström exponents AE for the 440-870 nm wavelength range. Details to the AERONET data processing steps are given in Mamouri and Ansmann (2014, 30 2016. To obtain climatologically representative dust conversion factors for a given AERONET station, we filtered out all AERONET data sets fulfilling the constraints of an Ångström exponent AE<0.3 and a 532 nm AOT>0.1. The AOT for 532 nm (in the following equations simply denoted as τ d ) is obtained from the 500 nm AOT τ 500 and the Ångström exponent a, stored in the AERONET data base, by τ d = τ 500 (500/532) a . (1) More information to the dust selection criteria are given in Sect. 3.
It is noteworthy to mention that recent airborne in situ observations of dust size distributions over the Sahara and remote dust outflow regions by Ryder et al. (2019) corroborate the high quality and consistence of the overall AERONET 5 optical and microphysical data sets and the applicability of the AERONET data analysis and inversion concept. The AERONET inversion method required to obtain the aerosol microphysical from the measured optical properties assumes that only particles with radius ≤15 µm are responsible for the observed dust-related optical effects. The presence of larger dust particle is ignored. Ryder et al. (2019) now show that dust particles with radius >15 µm contribute by only 1-3% to the particle extinction coefficient at 550 nm. This means that this size cutoff effect has practically no 10 impact on the AERONET inversion products and thus on the derived POLIPHON conversion factors.
In the following, we use the example of the dust mass concentration retrieval to explain the basic idea of the derivation of conversion factors from the AERONET data base. The mass concentration for the aerosol type dust (index d) is given by with the dust particle density ρ d of 2.6 g cm −3 and the dust volume concentration v d . The required dust volume concentration 15 in Eq.
(2) can be obtained from the following conversion: with the extinction-to-volume conversion factor c v,d,λ (derived from the AERONET long term observations) and the particle extinction coefficient σ d,λ measured with lidar at wavelength λ. We concentrate on lidar observations at 532 nm in this study and omit the wavelength index λ in the following. The conversion factor is obtained from the AERONET observations of the 20 vertically integrated particle volume concentration V d (denoted also as column volume concentration) and the aerosol optical thickness τ d (AOT at 532 nm, see Eq. 1), To provide a link to the lidar-derived height profile of σ d (z) (see Eq. 3), we introduce an aerosol layer depth with an arbitrarily chosen vertical extent D. With D, Eq. (4) can be written as with the layer mean volume concentration v d and the layer mean particle extinction coefficient σ d . For simplicity, we assume that all aerosol is confined to the introduced layer with vertical depth D. We may interprete this layer as the dust containing boundary layer or as a lofted dust layer with a vertical extent D. The introduced layer depth D has no impact on the further retrieval of the conversion factors and is only required to move from column-integrated values and AOT to more lidar-relevant quantities like concentrations and extinction coefficients.
To obtain climatologically representative dust conversion factor for a given AERONET station, we selected all individual dust observations (from number j = 1 to J d collected over many years), as mentioned defined by an Ångström exponent AE<0.3 and 532 nm AOT>0.1. For each dust observation j we computed c v,d,j and then determined the mean value, which As before, indices df and dc denote fine-mode and coarse-mode dust fractions, respectively. In Ansmann (2015, 2016), we explain how we calculate n 250,d,j , s d,j as well as s 100,d,j (discussed below) from the downloaded AERONET size 15 distribution data sets.
In the retrieval of the conversion parameters required to obtain n 100,d (Table 1, line 8), we used a different approach.
Following the procedure suggested by Shinozuka et al. (2015), we applied a log-log regression analysis to the log(n 100,d )log(σ d ) data field for each of the considered 20 AERONET station and determined in this way representative values for c 100,d and x d that fulfill best the relationship, as will be shown in the next section.
3 Conversion parameters from the AERONET data base Table 3 contains the list of AERONET stations considered in our effort to determine dust conversion factors for different desert regions around the globe. Measurement periods, numbers of available individual observations and of dust-related observations 25 (J d ) in Eqs. (6)-(11), and mean aerosol and dust conditions are given as well in Table 3. The locations of the AERONET sites are shown in Fig. 1. We preferred stations in Africa, Middle East and Asia with long data records and large numbers of observations. As can be seen in Table 3, the number of useful dust observations (AE<0.3, AOT>0.1) ranges from 218-4199 for 13 out of the 20 sites and is thus sufficiently high enough for the statistical analysis. The first six stations (from Tamanrasset to Ilorin) in Table 3 are exclusively influenced by Saharan dust, the next six stations (Limassol to Mezaira) by Saharan and Middle 5 East (mainly Arabian desert) dust, followed by three stations in Central and East Asia (Dushanbe to Dalanzadgad), which are influenced by long-range transport from the Sahara and western Asian deserts (including deserts in Iran and Kazakhstan) but also strongly by desert dust from Taklamakan and Gobbi deserts (Langzhou, Dalanzadgad). The Limassol data sets belongs to the Sahara group because the majority of dust outbreaks contain Saharan dust (Nisantzi et al., 2015). Sun photometer observations in North America (Great Basin, Tuscon, White-Sands), South America (Patagonian desert, Trelew), South Africa 10 (Kalahari desert, Gobabeb) and in the central Australian desert (Birdsville) complete our global AERONET dust data set.
It was difficult to find AERONET stations in North and South Amercia with a useful number of cases indicating pure dust observations.
We defined an ambitious, quite demanding criterion to filter out the pure dust cases for our study. The constraints AOT>0.1 at 532 nm and Ångström exponents AE<0.3 for the 440-870 nm wavelength range guarantee that interference by anthropogenic 15 pollution, biomass burning smoke, and marine particles are of minor importance. The mean values and standard deviations for dust AOT in Table 3 Table 4), we start with basic correlations between the dust microphysical properties and the dust extinction coefficient. Figure 2 provides an overview 25 of the relationship between the dust particle number concentration of larger particles n 250,d and the dust extinction coefficient  We set the layer depth D in Eq. (5) simply to 1000 m so that σ d (in Mm −1 in Fig. 2) divided by 1000 yield the basic AERONET 532 nm AOT value. We selected different colors to distinguish Saharan dust observations (green), Middle East measurements (orange) and data collected in Central and East Asia (red). We used bluish colors (blue, cyan) for the American and Australian stations, respectively, and blue-green for the African site (in the southern hemisphere) of Gobabeb.
As can be seen in Fig. 2, there are no large differences in the correlation features for the different AERONET stations. The East AERONET sites are influenced by both Saharan as well as Middle East (mainly Arabian) dust.
The spread in the data mainly reflects variations in the dust aerosol characteristics (size distribution, refractive index) as a function of varying mixtures of freshly emitted local dust and long-range-transported aged dust. Fresh and aged dust mixtures may have occurred in different dust layers above each other (as in the case study in Sect. 4). Uncertainties in the AERONET data inversion procedure applied to obtain the microphysical properties from the measured AOT and sky radiance observations However, it should also be mentioned that dust extinction coefficients in lofted layers above the boundary layer (in the free troposphere) seldom exceed 200-300 Mm −1 . For σ d <500 Mm −1 the scatter in the data is comparably low in Fig. 2. Figure 3 indicates that the relationship between the surface area concentration s 100,d , i.e., the CCN-related particle 15 surface concentration, and the particle extinction coefficient σ d at 532 nm is much more robust (less variable) than the one for s d vs σ d in Fig. 2b. The reason for this less noisy relationship is probably that the AERONET inversion analysis (for coarse-mode dominated particle ensembles) is not very accurate for the small-particle fraction (radius classes from 50-100 nm) and this inversion-related uncertainty is then reflected in the variability of the s d values considering all particle classes. With increasing minimum particle radius in the surface area computation the variability in the 20 relationship between respective surface area concentration and extinction coefficient decreases.
However, as will be shown in the next section, in contrast to the s 100,d vs σ d relationship, the correlation between n 100,d and σ d is strongly variable. One of the reason for this difference is that particles with large geometrical cross section (coarse-mode particles) have a higher weight in the surface area computation (integral over all sizes classes) and thus control the s 100,d values. In the n 100,d calculation, on the other hand, the size classes with highest particle number 25 concentration (fine-mode classes) dominate the n 100,d values.
3.2 Relationship between n 100,d and dust extinction coefficient σ d A different way of data analysis is used for n 100,d . As suggested by Shinozuka et al. (2015) we correlated log(n 100,d ) vs log(σ d ). Figure 4 shows the relationship between particle number concentration n 100,d and the dust extinction coefficient σ d at 532 nm for two stations (Mezaira, Dushanbe) in logarithmic scale. As outlined in Sect. 2, the particle number concentration 30 n 100,d , considering only the particles with dry radius >100 nm, represents very well the CCN reservoir in the case of dust particles for a typical water supersaturation of 0.2% (Mamouri and Ansmann, 2016;Lv et al., 2018).
In Fig. 4, we highlight the difference in the correlation when using all available data (532 nm dust AOT from 0.1 to 3.0 or σ d from 100-3000 Mm −1 ) and when using only observations with AOT<0.6. By detailed inspection of all data sets (station by station), we observed that the correlation strength significantly decreases with increasing AOT and is no longer clearly visible for all measurements with AOT from 1.0 to 3.0. The Dushanbe data set shown in Fig. 4b is a good example for this observation.
We can only speculate about the reason for the weak relationship for AOT>0.6. When the AOT is too large, the coarse-mode dust fraction may control the measured optical properties and respective inversion results so much that a trustworthy retrieval of the particle fraction with radii from, e.g., 100-200 nm is no longer possible. Another explanation is related to the observational 5 procedure. Most inversion computations are based on AERONET observations in the early morning and evening hours when the effective impact of aerosols is strongest (so that the effective dust AOT is even higher by a factor of two and more than the one for the vertical column stored in the AERONET data base). At these low-visibility conditions, the short-wavelength AERONET channels (340 and 380 nm) may have problems to correctly measure the overall AOT (Rayleigh AOT plus particle AOT). The short-wavelength AOT values are, however, especially important in the inversion retrieval of small dust particles 10 and thus have a strong influence on the n 100,d retrieval results.
As a consequence of the low correlation between log(n 100,d ) and log(σ d ) for large AOT we restricted the determination of the conversion parameters c 100,d and x d (see Eq. 12) by means of a regression analysis to AOT values from 0.1-0.6 (or respective σ d from 100-600 Mm −1 ). Furthermore, the relationship between n 100,d and σ d as found by Shinozuka et al. (2015) for dusty field sites is presented.

20
As mentioned above, most of the dust-related lidar observations in the free troposphere show dust extinction coefficients (σ d ) <200-300 M −1 . For a moderate dust extinction value of 100 Mm −1 , the POLIPHON retrieval yields n 100,d ≈ 150 cm −3 and 250 cm −3 when using c 100,d and x d numbers as derived from the Cabo Verde and Mezaira AERONET observations (AOT<0.6), respectively. Thus, a maximum overall error of a factor 2 in Table 1 (for n 100,d and n CCN,d ) also concluded by Shinozuka et al. (2015) and corroborated by Mamouri and Ansmann (2016) is justified.

Overview of AERONET-derived conversion parameters
In Table 4, the AERONET-based conversion parameters for all stations are presented. Regional mean sets of conversion parameters are given as well. Figures 6 and 7 provide a station-by-station overview of the conversion parameters (mean and SD values). Systematic differences from region to region are visible in the case of c 250,d and also weakly for c v,d . The conversions parameters for the American, Australia, and southern Africa need to be handled with caution because the number of available 30 observations is relatively low and the mean 532 nm AOT of these observations was low as well with values from 0.15-0.25.
A decrease in c 250,d and a slight increase in c v,d (and c v,dc ) from African to East Asian AERONET stations suggests that, for the same measured extinction coefficient (σ d ), the accumulation mode particle number concentration (in our case particles with radius from 250-500 nm) is slightly larger and the coarse mode dust particle number concentration, dominating the dust volume concentration, is lower in the case of Saharan dust compared to East Asian dust. This behavior may indicate that the African AERONET stations, e.g., in Cabo Verde, Izana, and Dakar observe predominantly dust after long-range transport (which leads to a bit enhanced fine dust fraction because of size dependent sedimentation and removal of particles), whereas the East Asian AERONET stations may be influenced more frequently by the occurrence of local, freshly emitted dust with the relatively strong contribution of coarse-mode particles. Similar conditions as suggested for Central and East Asia may hold for  This exponent is then linked to c 100,d values of 5-6 cm −3 (at σ d =1 Mm −1 ).

Lidar measurement example: Case study of a dust observation in Tajikistan
We used the updated set of conversion parameters to analyze a dust measurement performed with a Polly system deployed fine-mode fraction (FMF) of 0.2 (just before sunset close to 13:00 UTC). Thus, fine dust contributed about 20% to the overall (fine and coarse) dust extinction coefficient. According to the backward trajectories in Fig. 9, mineral dust in the polluted 30 boundary layer (0-2 km height) originated from Kazakhstan and local dust sources. The dust particles in the thick dust layer from to 2-5 km height were mostly emitted in Iran and Oman. Higher up (above 5 km) long range transport of dust from the Arabian peninsula (5-7 km height) and even the Sahara (8-10 km) prevailed. More details to the long-range transport features in comparison with aerosol transport modeling is given in Hofer et al. (2017). Figure 10a shows the basic lidar profiles used in the POLIPHON data analysis. The height profiles of the particle (dust + non-dust) backscatter coefficient and the related particle linear depolarization ratio are used to derive the dust and nondust extinction profiles Ansmann, 2014, 2017). The dust extinction coefficients are then converted into the 5 dust mass concentrations in Fig. 10b by means the dust conversion factor c v,d in Table 4 for Dushanbe (red profiles). The mass computation is performed in the way described in Table 1. The corresponding dust mass fraction (ratio of dust mass concentration to total particle mass concentration) is presented in Fig. 10b as well. To provide an estimate of the uncertainty in the dust mass concentration introduced by the conversion uncertainty two conversion factors for Dushanbe and for Cabo Verde, representing a relatively high and low value of all conversion factors listed in Table 4, were applied in Fig. 10. The resulting 10 differences in the POLIPHON results are well covered by the overall uncertainty in the POLIPHON mass retrieval of 30% (see the error bars in Fig. 10) which also includes the uncertainty in the dust extinction determination. Figure 11 presents the POLIPHON results in terms of several CCNC profiles obtained with conversion parameter sets for Cabo Verde, Mezaira, and Dushanbe (see Table 4). As mentioned, n CCN,d ≈ n 100,d for a water supersaturation value of 0.2%.
According to the discussion in Sect. 3.2 and the uncertainty information in Table 1 the overall uncertainty in the regression 15 analysis of n 100,d with σ d is of the order of 50-200%. In Fig. 11, an uncertainty factor of 2 is considered by the dashed lines.
Compared to this factor-2 uncertainty margin, the impact of the applied different conversion parameter sets is likewise small. which is relatively low and may explain the short-lived thin ice cloud features (occurring after 16:15 UTC) and the absence of large cirrus fields with extended virga zones.

5
An extended global AERONET analysis has been performed to create a global data set of dust-related POLIPHON conversion factors. We analyzed AERONET observations for all relevant desert regions in Africa, Middle East, Central and East Asia, America, and Australia and provide respective regional conversion parameter sets. Significant differences in the obtained conversion parameters caused by potentially different dust composition and size distribution characteristics for different desert regions were not found. Furthermore, the presented Tajikistan case study showed that the use of different, contrasting con-10 version parameters did not have large (dominating) impact on the overall uncertainty in the POLIPHON results. This is of advantage for spaceborne lidar applications when one wants to use, e.g., one set of conversion parameters in global observations. This universal conversion parameter set may be the mean of all Saharan, Middle East, and Asian dust conversion parameters given in Table 4. For ground-based observations it is however always advisable to make use of the specific, regional conversion parameters and to check the uncertainty caused by the conversion by using different 15 conversion parameter sets listed in Table 4.
In conclusion, we can state that appropriate conversion parameters are now available for mineral dust around the globe. In addition, conversion parameters representing pure marine conditions are available from marine Barbados AERONET observations Ansmann, 2016, 2017). As an outlook, it remains to investigate in detail the conversion parameters for anthropogenic aerosol particles (urban haze, rural background aerosol, forest fire smoke, and free tropospheric smoke and haze 20 by using mountain stations). A detailed study for anthropogenic aerosol conversion parameters has only be done so far for the urban, highly polluted AERONET stations of Leipzig and Limassol Ansmann, 2016, 2017).

Data availability
All data used in this work can be accessed through the AERONET home page at https://aeronet.gsfc.nasa.gov/ (last access: 22 Februray 2019). Polly lidar observations (level 0 data, measured signals) are in the PollyNET data base (http://polly.rsd.tropos.de/). 25 All the analysis products are available at TROPOS upon request (info@tropos.de).

Author contributions
AA and REM worked on the applied methodology and prepared the manuscript. JH provided the Dushanbe case study results.
DA, JH, and SFA took care of the excellent performance of the Polly lidar and AERONET photometer during the 18-month CADEX field campaign. 30 The authors declare that they have no conflict of interest.