Estimating cloud condensation nuclei concentrations from CALIPSO lidar measurements

. We present a novel methodology to estimate cloud condensation nuclei (CCN) concentrations from spaceborne CALIPSO lidar measurements. The algorithm utilizes (i) the CALIPSO-derived backscatter and extinction coefﬁcient, depolarization ratio, and aerosol subtype information, (ii) the normalized volume size distributions and refractive indices from the CALIPSO aerosol model, and (iii) the MOPSMAP optical modelling package. For each CALIPSO height bin, we ﬁrst select the aerosol-type speciﬁc size distribution and then adjust it to reproduce the extinction coefﬁcient derived from the CALIPSO 5 retrieval. The scaled size distribution is integrated to estimate the aerosol number concentration which is then used in the CCN parameterizations to calculate CCN concentrations at different supersaturations. To account for the hygroscopicity of continental and marine aerosols, we use the kappa parameterization and correct the size distributions before the scaling step. We have studied the (cid:58)(cid:58)(cid:58) The (cid:58) sensitivity of the thus derived CCN concentration to the effect of (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) derived (cid:58)(cid:58)(cid:58)(cid:58)(cid:58) CCN (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) concentrations (cid:58)(cid:58)(cid:58) to variations of the initial size distributions (cid:58)(cid:58) is (cid:58)(cid:58)(cid:58)(cid:58) also (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) examined. It is found that the uncertainty associated with the algorithm can 10 range between a factor of 2 and 3. We have also compared our results with the POLIPHON and found comparable results (cid:58)(cid:58)(cid:58) Our (cid:58)(cid:58)(cid:58)(cid:58)(cid:58) results (cid:58)(cid:58)(cid:58) are (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) comparable (cid:58)(cid:58) to (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) results (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) obtained (cid:58)(cid:58)(cid:58)(cid:58)(cid:58) using (cid:58)(cid:58)(cid:58) the (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) POLIPHON (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) method for extinction coefﬁcients larger than 0.05 km − 1 . An initial application to a case with coincident airborne in-situ measurements for independent validation shows promising results and illustrates the potential of CALIPSO for constructing a global height-resolved CCN climatology.

In the present work, we utilize the CALIPSO aerosol model to calculate the extinction coefficient by using Mie scattering for spherical particles (continental and marine aerosols) and a combination of T-matrix and improved geometric optics method for non-spherical particles (dust aerosols). We then modify the NVSD by preserving its shape (mode radii and standard deviation remain constant) until a closure is achieved between the extinction coefficient inferred from CALIPSO measurements and The CALIPSO version 2 aerosol types include dust, smoke, clean continental, polluted continental, clean marine, and polluted dust. The microphysical properties of these six aerosol subtypes constitute the CALIPSO aerosol model (CAMel). The 95 lidar ratios used in the retrieval of extinction coefficient for each aerosol type were modelled using these microphysical properties. Of the six aerosol subtypes, the properties of smoke, polluted continental, and polluted dust were obtained directly from a cluster analysis of long term cloud screened AERONET measurements (Omar et al., 2005). The dust model was derived from Kalashnikova and Sokolik (2002) and the clean marine model was derived from the dry measurements taken during Shoreline Environment Aerosol Study (SEAS) campaign (Masonis et al., 2003;Clarke et al., 2003). The clean continental model was 100 formed by adjusting the properties of the background continental aerosol cluster from Omar et al. (2005) to measurements of Anderson et al. (2000). The aerosol model has evolved with time. In version 4, a new aerosol subtype namely the dusty marine (dust + marine) was introduced. Further, the polluted continental and smoke subtypes were renamed to polluted continental/smoke and elevated smoke, respectively (Kim et al., 2018). The lidar ratios were also modified leading to an increase in mean AOD by 52% (40%) for nighttime (daytime) retrievals, making it more comparable with MODIS derived AOD. In 105 our algorithm, we use the microphysical properties of 5 aerosol subtypes namely marine, dust, polluted continental/smoke, clean continental, and elevated smoke. :::: Note ::: that ::: the :::: lidar ::::: ratios ::::: used :: in :::::: version : 4 :: of ::: the ::::::::: CALIPSO ::::::: retrieval :::: have ::::: been ::::::: adjusted :::: from :::::: earlier ::::::: versions ::::: based ::: on ::: the ::::::: findings :::: from :::::::::: atmospheric :::::::::::: measurements :::::::::::::::: (Kim et al., 2018) ::: and :::: don't :::::::::: necessarily ::::::: connect :: to ::: the :::::::: CALIPSO ::::::: aerosol :::::: model. Since the changes in lidar ratio from version 2 to version 4 are minor (≤ 1%) for all aerosol types except for clean continental (51%), we believe the aerosol models ::::: model : can still be used in our algorithm. However, 110 for the case of clean continental aerosol subtype, further study is required to estimate the effect of change in lidar ratio on its microphysical properties. Having said that, we do not exclude it from our analysis for the completeness of our algorithm, leaving a scope of future validation study to examine its applicability in estimating the CCN concentrations from CALIPSO.

MOPSMAP package
Modelled optical properties of ensembles of aerosol particles (MOPSMAP) package provides the aerosol optical properties of 115 arbitrary, randomly oriented spherical or spheroidal particle ensembles for size parameter ranging up to 1000 and refractive index range of [0.1, 3.0] and [0, 2.2] for real and imaginary parts, respectively (Gasteiger and Wiegner, 2018). It includes a data set of pre-calculated aerosol optical properties and a Fortran program which estimates the properties of user-defined aerosol ensembles. The optical properties of spherical particles are modelled using Mie scattering. While for spheroids, based on the aerosol size parameter, MOPSMAP uses a combination of the T-matrix method and improved geometric optics method.

120
MOPSMAP has been used to simulate the optical properties of different aerosol types such as mineral (silica and alumina) and ash aerosols (Jiang et al., 2021), and Martian dust aerosols (Chen-Chen et al., 2021). We apply the MOPSMAP package to model the aerosol extinction coefficient of different aerosol subtypes with the bimodal lognormal volume size distributions and refractive indices from the CAMel. The details of the MOPSMAP input parameters are discussed in the methodology section.

135
The CCN concentration at a certain supersaturation is estimated from the aerosol number concentration as where f ss = 1.0, 1.35, and 1.7 for supersaturations of 0.15%, 0.25%, and 0.40%, respectively. In this study, we use the conversion factors and extinction exponents for continental and marine aerosols from Mamouri and Ansmann (2016). For dust aerosols, we use the globally averaged values as suggested by Ansmann et al. (2019) for application to satellite data. For

Aerosol size distribution
The remote sensing of aerosol number concentration requires an initial assumption of aerosol microphysical properties (size distribution and refractive index). For instance, the MODIS algorithm over the ocean uses a combinations :::::::::: combination of 9 155 predefined aerosol size distributions and refractive indices, and selects the one for which the difference in the measured and modelled radiance is minimum (Appendix B of Remer et al. (2005)). In our study, we use the aerosol microphysical properties from CAMel and adopt a two-step algorithm to derive the aerosol size distribution: (i) select the appropriate initial normalized volume size distribution and refractive index, and (ii) scale the size distribution as per the CALIPSO measured extinction coefficient. In contrast to MODIS, the aerosol type in CALIPSO is set prior to the computation of the extinction coefficient.

160
This eases the selection of initial aerosol microphysics, which can now be done directly from the CAMel as per the aerosol subtype information included in the CALIPSO retrieval.
The next step is to scale the NVSD as per the CALIPSO measured extinction. The extinction coefficient (α) for a certain incident wavelength can be described as where r is the particle radius, V (r) is the volume of the particle with radius r and K α is the extinction cross-section which is a function of the complex refractive index (m) and r. dV (r)/d ln r is the log-normal volume size distribution which for a bimodal case can be given by Here, ν i , σ i , and µ i are the volume fractions, geometric standard deviations, and geometric mean radii of i th mode, respectively.

170
V t is the total volume of the size distribution. The above size distribution is normalized when V t = 1. Substituting Eq. (4) in Eq.
(3), we get Thus, the extinction coefficient is a function of the size distribution parameters (V t , ν i , σ i and µ i ) and the extinction cross section (K α ). Out of these parameters, under ideal conditions, only V t is an extensive property, while the rest are intensive and 175 independent of aerosol amount or concentration (Omar et al., 2005). Eq. (5) can be simplified to Where α n is the normalized extinction coefficient corresponding to the NVSD. If we consider α as the CALIPSO measured extinction, V t would be the scaling factor for the NVSD to compute the actual aerosol size distribution. From Eq. (6), we can compute V t if the value of α n is known. 180 We estimate α n for each aerosol subtype by using the NVSDs and refractive indices from CAMel as input to the MOPSMAP optical modelling package. In the MOPSMAP input, we consider dust as spheroids and use the axis ratio distribution from  (also used in the AERONET inversion). Other aerosol subtypes are considered spheres. We then compute V t from the ratio of α and α n (Eq. 6). On multiplying V t with the NVSD, we get the final scaled aerosol size distribution.
Since the algorithm principally relies on the optical modelling of CALIPSO aerosol microphysics, we hereafter refer to it as 185 OMCAM.

Aerosol hygroscopicity
The hygroscopic aerosol particles in the atmosphere can uptake water and grow in moist conditions. The hygroscopic growth needs to be accounted for before deriving the aerosol size distributions discussed in the previous section :::: CCN :::::::::::: concentrations.
We consider continental (clean continental, polluted continental/smoke and elevated smoke) and marine aerosols as hygro-190 scopic. We assume dust aerosols to be hydrophobic in accordance with previous studies (Mamouri and Ansmann, 2016;Ansmann et al., 2019). The hygroscopicity correction can be either applied to the ambient extinction coefficient measured by CALIPSO or to the initial NVSD in the retrieval algorithm. We consider the latter approach and modify the initial NVSD before modelling the extinction coefficient. There is an inbuilt functionality in the MOPSMAP package to account for the hygroscopicity using the Kappa parametrization scheme (Petters and Kreidenweis, 2007;Zieger et al., 2013) as where RH is the relative humidity and κ is the hygroscopic growth parameter. The r min , r max , and µ of the log normal ::::::::: log-normal size distribution (Eq. 5) are multiplied with this ratio whereas the standard deviation (σ) remains unchanged. The refractive index of the hygroscopic aerosol is also modified following the volume weighting rule (Gasteiger and Wiegner, 2018). The κ value is set to be 0.3 for continental and 0.7 for marine aerosols. The values are global averages and are suggested 200 by Andreae and Rosenfeld (2008).

CCN parametrizations
We use the parameterizations listed in Mamouri and Ansmann (2016) to estimate CCN concentrations from the dry aerosol number concentration. The final scaled aerosol volume size distribution obtained from the scaling procedure is first converted to number size distribution. The number size distribution is integrated starting at 50 or 100 nm to compute n 50,dry or n 100,dry 205 depending on the aerosol type. Finally, substituting the values in Eq.
(2) results in the required CCN concentration at different supersaturations.

Application of OMCAM to CALIPSO retrieval
Figure 1 outlines the OMCAM retrieval algorithm for estimating CCN concentrations from CALIPSO measurements. In order to apply the OMCAM algorithm to CALIPSO level 2 version 4.20 data, we first start by preprocessing the dataset. To begin 210 with, we apply all the quality filters listed in Tackett et al. (2018, Table 1). The CALIPSO aerosol typing algorithm consists of dust mixtures (dusty marine and polluted dust). In such a case, we separate the dust and non-dust extinction coefficients by using the methodology given in Tesche et al. (2009). This is a rather simple and accepted dust separation technique also used by Ansmann (2015, 2016) for lidar based ::::::::: lidar-based CCN retrieval. It uses the particle depolarization ratio (δ p ) to separate the particle backscatter coefficient (β p ) into dust (β d ) and non-dust (β nd ) contributions. β d can be calculated as Where the values of δ 1 and δ 2 are 0.31 and 0.05, respectively. The aerosol mixture is assumed to be pure dust (non-dust) when δ p > 0.31 (< 0.05). When 0.05 ≤ δ p ≤ 0.31, we first estimate β d from Eq. (8) and then calculate β nd by subtracting β d from β p . We compute the dust and non-dust extinction coefficient by multiplying the backscatter coefficient with the respective lidar ratio. The lidar ratios of dust, polluted continental and clean marine aerosol subtypes are taken from Kim et al. (2018) and are 220 equal to 44, 70, and 23, respectively. The extinction coefficient of polluted dust is separated into polluted continental/smoke and dust, while that of dusty marine is separated into dust and marine contributions. Finally, the extinction coefficient, relative humidity, and aerosol subtype information are passed to the CCN retrieval algorithm.
In the CCN retrieval part, we first select the normalized size distribution and refractive index as per the aerosol subtype and modify them as per the RH value so as to account for the hygroscopicity of aerosols. In the next step, we model the extinction 225 coefficient using the MOPSMAP package and calculate V t from Eq. (6). Multiplying V t with the initial dry normalized size distribution gives the final dry aerosol size distribution which is used in the CCN parameterizations (Eq. 2) to estimate the CCN concentrations at different supersaturation values. This methodology is applied to every bin of the CALIPSO profile. In the case of dust mixtures, the separated dust and non-dust extinction coefficients are passed through the CCN retrieval algorithm individually, and the results are finally added to compute the net CCN concentration for that bin. It is worthwhile to note 230 that this algorithm can in principle be used to derive INP concentration from CALIPSO measurements. This can be done by first estimating n 250 from the modified size distribution (Section 4.1 :: 3.1) and then using the INP parameterizations (DeMott et al., 2010(DeMott et al., , 2015 to estimate INP concentrations. However, in the present study, we limit our focus on retrieving the CCN concentrations.

Sensitivity analysis
The performance of OMCAM in retrieving CCN concentrations primarily relies on the initial NVSD given in the CAMel.
The aerosol size distributions may change depending on the age and composition of aerosols (region and type dependent), and the ambient meteorology. As most of the size distributions used in the CAMel are derived from cluster analysis of the long term AERONET measurements (see Section 2.2 ::: 2.1), they incorporate the errors associated with the AERONET inversion 240 algorithm. Dubovik et al. (2000) found that the relative error in the AERONET-retrieved volume size distribution for dust, biomass burning, and water-soluble aerosols can go beyond 50% for both small (r < 0.1 µm) and large (r > 7 µm) particles.
In order to account for such errors and natural variability, we analyzed the sensitivity of CCN concentrations to the initial normalized size distributions considered in our retrieval algorithm.
For each aerosol subtype, the initial NVSD can be perturbed by changing the size distribution parameters such as the volume , and mean radii (µ f & µ c ) of fine and coarse modes. Since the sum of the volume fractions is unity, this leads to 5 independent size distribution parameters. We first study the individual effects of varying these parameters on the output n j,dry (j = 100 for dust and 50 for other aerosol subtypes), as they are the main input to the CCN parameterizations. Figure 2 depicts the effect of varying these size distribution parameters by ± 50% on the n j,dry relative to that of unperturbed size distributions from CAMel, for a preset α = 0.1 km −1 and RH = 0 for different 250 aerosol subtypes. The results show fine mode as the primary contributor to the output aerosol number concentration. A certain change in the volume size distribution in the fine mode will have a larger impact on the number concentration compared to the coarse mode, as a much larger number of small particles is needed to produce the same change in volume. Out of the 5 parameters, µ f has the maximum effect (≈ 800%) on the output number concentration followed by σ f (≈ 150%). This is because both µ f and σ f modify the distribution of volume across different radii in the fine mode. Decreasing (increasing) µ f 255 shifts the fine mode towards a smaller (larger) radius thereby resulting in a comparatively larger (smaller) number of particles for a constant fine mode volume. However, for dust, the effect is opposite when µ f is decreased. This is because the minimum cut-off radius for dust is set to be 100 nm and the fine mode moves out of this limit when µ f is reduced leading to a decrease in the output number concentration. Increasing (decreasing) σ f leads to an increase (decrease) in the fraction of smaller particles within the fine mode. This results in an increase (decrease) in the output number concentration for all aerosol subtypes except 260 dust. The output number concentration is comparatively less sensitive to coarse mode parameters (µ c & σ c ), as they contribute primarily to the optical properties of the aerosol volume rather than the number concentration. When we change the value of α, the aerosol number concentration scales as per the ratio between α and α n , resulting in no change in the relative n 100,dry and n 50,dry .
The size distributions formed by varying the size distribution parameters separately may not be sufficient enough to capture 265 the natural variability. Thus to imitate the natural variability in a better way, we further consider combinations of the variations of all the parameters. We don't expect extreme shifts in the size distribution parameters as well. For instance, reducing µ f by 50% results in abnormal size distributions with 30%-50% of the fine mode moving out of the AERONET size limits (0.05 ≤ r ≤ 15 µm). Therefore, in order to exclude the non-physical size distributions, we limit the variations of the parameters in terms of the actual volume size distributions. To implement these constraints, we first vary the size distribution parameters 270 linearly with a uniform spacing of 0.01 and then consider all possible combinations of the variations. The NVSDs generated from all the combinations forms the input NVSD set for the sensitivity analysis. We further fix the maximum limits of bimodal NVSD to ± 50% of the amplitude of each of its modes and do not consider the NVSDs that fall outside this domain in the sensitivity studies. The resulting input NVSD space for each aerosol type is shown by the shaded region of Figure 3. The maximum and minimum values of all the size distribution parameters considered in the sensitivity analysis are given in Table  As we have kept a constant spacing for varying the size distribution parameters, the number of NVSD in the input space directly depends on the volume of particles present in each mode. While it is minimum for clean marine subtype because of its almost non-existent fine mode (which reduces the range of variation), it is maximum for polluted continental and elevated smoke subtypes. The output ensemble ::::::::: ensembles of number concentrations for an extinction coefficient of 0.1 km −1 and 280 relative humidity of 0% are shown in the violin plots of Figure 4. The percentiles of the output n j,dry set are given in Table 3.
The number concentration of the output ensemble is primarily dependent on the fine mode of the input size distributions. The variations in the output ensemble relative to the output from unperturbed NVSD from CAMel is minimum (about a factor of 1) for dust mainly because we only consider particles with a radius > 0.1 µm. For clean marine, the spread is about a factor of 2 (95 th percentile; 200%). However, for polluted continental and elevated smoke, the output ensemble is bi-modal. For the 285 first mode, the values can go up to a factor of 1.5 for polluted continental and to around 1 for elevated smoke. The second mode is relatively small and is related to the size distributions whose fine mode mean radii are shifted to low values (extreme left in Figure 2). For this mode, the values can go up to a factor of 3 for polluted continental and 2.5 for elevated smoke. The largest spread in the output ensemble is found for clean continental (95 th percentile; factor of 2.7). This might be because the bi-modality of the NVSD is not well defined for the clean continental aerosol subtype, thereby increasing the input space of 290 variation. Neglecting the long tail of the distribution, we can assume the uncertainty to ::: due :: to ::: the :::::: initial :::::: NVSD :: to be about a factor of 2.
We have also estimated the effect of change in RH on the output ensemble of n 100,dry and n 50,dry (not shown). Increasing RH decreases the spread of the output ensemble slightly, with a significant decrease for RH > 90% except for dust which is assumed to be hydrophobic. At RH = 99%, the bi-modality of polluted continental and elevated smoke subtypes disappears. The

295
variations in the relative number concentrations decrease to less than a factor of 2 for all subtypes. This might be a result of the decrease in the absolute number concentration, as the particle size increases with RH and fewer particles are needed to produce the same extinction. At a constant RH value, when α is modified, the output ensemble of aerosol number concentrations scales as per the ratio between α and α n resulting in no change in the relative n 100,dry and n 50,dry (not shown). To summarize, if we neglect the contributions of extreme shifts in the size distribution (i.e., the long tails in the violin plots) and consider the effect 300 of RH, we can assume the overall uncertainty in the retrieval algorithm due to the initial NVSD is likely to range between a factor of 1.5 and 2.5.
However, OMCAM incorporates additional uncertainties due to the hygroscopicity correction. Studies have found that the conversion factors used in the POLIPHON technique for dust and smoke aerosols vary with the source region and the age 315 of aerosols (Ansmann et al., 2019;?) :::::::::::::::::::::::: (Ansmann et al., 2019(Ansmann et al., , 2021b. Such factors further increase the uncertainties associated with the retrieval algorithm when applied to satellite/global data sets.

365
The profiles of CALIPSO measured extinction coefficient, aerosol subtype, and the n 50,dry concentration calculated from the OMCAM algorithm for the CALIPSO overpass over Thessaloniki on 9 September 2011, are shown in Figure 6. Over the land areas (latitude from 40.6 • -41.2 • N), CALIPSO aerosol typing algorithm identifies the presence of elevated smoke and polluted continental aerosols (Figure 6b). However, for retrieving the extinction coefficient for polluted continental aerosol layer, the lidar ratio was modified and, thus, is not considered in our present comparison (not shown). The presence of smoke over the 370 land region was also identified by T17. The CALIPSO measured extinction coefficient over land is highly variable in space ranging from 0.07 km −1 to as high as ≈ 3 km −1 in the proximity of cloud. The OMCAM estimated n 50,dry correspondingly varies from 617 cm −3 to 40000 cm −3 . Over the sea region (latitude from 40 • -40.6 • N), T17 detected the presence of elevated smoke plumes. This was not detected by the aerosol typing algorithm of earlier version-3 CALIPSO data used in T17. However, with the modifications of version 4 used in this work, CALIPSO successfully detects elevated smoke, marine, and dust aerosols 375 with elevated smoke being the dominant one. The overall extinction :::::::: coefficient ::::: along :::: with ::: its ::::::::: variability over the sea area is less compared to land, with the values ranging from 0.026 km −1 to 0.36 km −1 . The corresponding OMCAM estimated n 50,dry concentrations vary from 33 cm −3 to 5000 cm −3 .
T17 estimated the n 50,dry at different altitudes over the land region corresponding to two 5 km cloud-free segments of CALIPSO retrieval with latitudes in between 40.85 • N and 40.95 • N. The average n 50,dry concentration estimated for the se-380 lected CALIPSO segments over land using OMCAM and POLIPHON (taken from G20) are plotted along with the in-situ measurements from T17 in Figure 7a and the values are listed in Table 4. On average, when no hygroscopicity correction is applied, the OMCAM and POLIPHON overestimate the n 50,dry concentration by 355% and 370%, respectively. A similar result from OMCAM and POLIPHON is expected given that elevated smoke was the dominant aerosol type over the land with extinction coefficient > 0.1 km −1 , for which both the algorithms yield a similar result ( Figure 5). Upon accounting for the 385 hygroscopic growth, the overestimation decreases to 167% (130% for POLIPHON). Note that the RH-corrected POLIPHON values in G20 are produced by using the in-situ dry to ambient extinction coefficient ratios (DAR) measured at different RH values during the aircraft measurements (Tsekeri et al., 2017). In contrast to the overestimation over the land, both the algorithms underestimate the n 50,dry concentrations over the sea (Figure 7b). When we don't account for the hygroscopic growth, both the OMCAM and POLIPHON algorithms underestimate the n 50,dry concentration by 22% and 38%, respectively. When 390 the RH growth is corrected, the underestimation further increases to 40% and 52%, respectively. Similar to land region :::::: regions, both the algorithms yield comparable results over the sea, as the dominant aerosol type is elevated smoke in both scenarios. seconds, it is around 2 hours for the aircraft. From Figure 6c, we can clearly see that the extinction coefficient along with the 400 n 50,dry concentrations is highly variable over the land region (ranging from 617 cm −3 to 40000 cm −3 ) compared to rather homogeneous concentrations over the sea. This might be the reason for such huge differences ::: the :::: large :::::::::: discrepancy : between in-situ and CALIPSO retrievals over the land region. Moreover, only two cloud-free CALIPSO 5 km profiles are considered for the comparison over land, which further increases the chances of disparity. Given the limited sample space, this comparison should not be considered as validation but rather a demonstration of the capability for retrieving CCN concentrations from 405 spaceborne lidar measurements. A detailed study comparing the CALIPSO retrieved aerosol number and CCN concentrations with ground-based and aircraft in-situ measurements is required to evaluate the reliability of OMCAM and POLIPHON algorithms in estimating the CCN concentrations.

Summary and conclusions
We present the OMCAM algorithm to derive the height-resolved cloud relevant CCN concentrations from CALIPSO mea-410 surements. The algorithm uses the normalized size distributions and refractive indices from CALIPSO aerosol models  as an input to MOSPMAP to calculate the extinction coefficient. The size distributions are then scaled to reproduce the CALIPSO measured extinction coefficient. In order to :: To : account for the hygroscopicity, we use κ parametrization (Petters and Kreidenweis, 2007) , and modify the size distribution and the refractive index before the scaling step. We then estimate the required aerosol number concentration by integrating the final scaled size distributions over the size ranges rel-415 evant for different aerosol types. Utilizing the aerosol type-specific CCN parameterizations from the POLIPHON method (Mamouri and Ansmann, 2016) ::::::::::::::::::::::::: Mamouri and Ansmann (2016), we convert the aerosol number concentrations to cloud relevant CCN concentrations for different supersaturation.
The OMCAM algorithm relies on the potentiality of the CALIPSO aerosol models to accurately describe the microphysical properties of the aerosol subtypes defined within the CALIPSO retrieval algorithm. We performed sensitivity tests by varying 420 the normalized size distributions by up to ± 50% of the amplitude of each mode and found that the uncertainty in the final aerosol number concentration ranges between a factor of 2 and 3.
We compared the CCN concentrations obtained from OMCAM with that of the POLIPHON method-the existing method for lidar-based CCN retrieval. For extinction coefficient > 0.05 km −1 , we found a good agreement for continental, dust, and smoke aerosols. However, as the extinction coefficient becomes smaller than 0.05 km −1 , the difference increases with the 425 POLIPHON values going as high as twice the OMCAM values. For marine aerosols, the CCN concentration derived using the POLIPHON method is always higher (4-6 times) than that of OMCAM.
For an initial evaluation of the OMCAM algorithm, we compared the thus obtained n 50,dry with in-situ measurements taken over the land and sea region around Thessaloniki during the ACEMED campaign (Tsekeri et al., 2017). For the retrievals over sea, we found that CALIPSO is underestimating the n 50,dry by about 40%. Over the land areas, however, CALIPSO overestimates n 50,dry by about 167%. The large discrepancies may be a result of the combination of highly variable n 50,dry over the land region and the instantaneous measurement by CALIPSO , in contrast to the in-situ measurement which were performed in a time period of 2 hours. All values remained within a factor of 2 which is in agreement with the estimated uncertainty. Moreover, the n 50,dry retrieved from CALIPSO using the OMCAM algorithm was comparable to that of POLIPHON (Georgoulias et al., 2020).

450
The ability of CALIPSO not only in measuring vertically resolved aerosol optical properties but also being able to detect the responsible aerosol type has facilitated the retrieval of global 3D type-specific aerosol properties. We have described a novel methodology to retrieve cloud relevant CCN concentrations from CALIPSO measurements illustrating the potential of CALIPSO to produce 3D global CCN climatology for ACI studies and climate model evaluations, opening new gates for further validation of the algorithm against ground-based and airborne in-situ measurements. Competing interests. The authors declare that they have no conflict of interest.   and elevated smoke (e) aerosol subtypes adopted from the CALIPSO aerosol model. The shaded region represents the ::: input ::::: space :::: along :::: with :: the : maximum and minimum limits of size distributions selected for the sensitivity analysis      Table 4. n 50,dry concentrations (in cm −3 ) from in-situ measurements (Tsekeri et al., 2017) and CALIPSO measurements by using OMCAM and POLIPHON (Georgoulias et al., 2020)