Variability of cirrus cloud properties using a Polly Raman Lidar over high and tropical latitudes

Measurements of cirrus clouds geometrical and optical properties, performed with a multi-wavelength Polly Raman Lidar, during the period 2008 to 2016 are analysed. The measurements were performed with the same instrument, during sequential periods, in three places at different latitudes, Gual Pahari (28.43◦N, 77.15◦E, 243m a.s.l) in India, Elandsfontein (26.25◦S, 29.43◦E, 1745m a.s.l) in South Africa and Kuopio (62.74◦N, 27.54◦E, 190m a.s.l) in Finland. The lidar dataset has been processed by an automatic cirrus cloud detection algorithm. In the following, we present a statistical analysis of 5 the lidar derived geometrical characteristics (cloud boundaries, geometrical thickness) and optical properties of cirrus clouds (cloud optical depth, lidar ratio, ice crystal depolarization ratio) measured in different latitudes that correspond to subtropical and subarctic regions as well as their seasonal variability. The effect of multiple-scattering from ice particles to the derived optical products is also considered and corrected in this study. Our results show that, over the subtropical stations, cirrus layers, which have a noticeable monthly variability, were observed between 7 to 13km, with mid-cloud temperatures ranging from 10 -60◦C to -21◦C and a mean thickness of 1295 ± 489m and 1383 ± 735m for Gual Pahari and Elandsfontein respectively. The corresponding overall mean cirrus optical depth at 355nm is calculated to be 0.59 ± 0.39 and 0.40 ± 0.33, with lidar ratio values at 355nm of 26 ± 12 sr and 25 ± 6 sr, respectively. A more extended dataset was acquired for the subarctic area of Kuopio Finland, between 2012 and 2016. The estimated average geometrical thickness of the cirrus clouds over Kuopio is 1200 ± 585m and the temperature values vary from -71◦C to -21◦C, while the mean cirrus optical depth at 355nm is 0.25 ± 0.2, 15 with an estimated mean lidar ratio of 33± 7 sr, similar to the lidar ratio values observed over middle latitude stations. The kind of information presented here can be rather useful in the cirrus parameterizations required as input to radiative transfer models, and can be a complementary tool to satellite products that cannot provide cloud vertical structure. In addition, a ground-based statistics of the cirrus properties could be useful in the validation and improvement of the corresponding derived products from satellite retrievals. 20 1 https://doi.org/10.5194/acp-2019-565 Preprint. Discussion started: 15 July 2019 c © Author(s) 2019. CC BY 4.0 License.

can be undetectable from cloud radars (Comstock et al., 2002) or from passive instruments. However, lidar beam attenuates strongly in liquid water clouds, and therefore, it is likely that in the case of multiple cloud layers reliable detection of cirrus clouds cannot be ensured.
In the last decades, observations of cirrus clouds properties have been conducted both in terms of field experiments (e.g. Seifert et al., 2007) and systematic observations (e.g. Dionisi et al., 2013;Pandit et al., 2015) from groundbased lidar systems, 50 providing an estimation of their dependence on the geographical location. Dionisi et al. (2013) presented a methodology for identification and characterization of cirrus clouds properties, applied to the multiwavelength Rayleigh Mie-Raman (RMR) lidar in Rome. The study classified the detected cirrus clouds in different categories, based on their optical properties. Specifically, the analysis showed that 10% of the detected cirrus were subvisible clouds (τ < 0.03), 49% thin (0.03 < τ < 0.3) and 41% opaque cirrus (τ > 0.3). The overall mean value of cirrus optical depth was calculated 0.37 ± 0.18, while the mean LR ef f value 55 was 31 ± 15 sr. Another statistical analysis on optical and geometrical properties of upper-tropospheric cirrus clouds based on a lidar dataset, was conducted in Amazonia (Gouveia et al., 2017). The frequency of occurrence of cirrus clouds classified as subvisible was 41.6%, whilst 37.8% was for thin cirrus and 20.5% for opaque cirrus. The correction of the multiple scattering effect to the optical products in this study was made following the model of Hogan (2008). Lakkis et al. (2015) revealed that the most commonly observed cirrus were characterized as optically thin cirrus, rather than opaque ones, with a mean opti-60 cal depth value of 0.26 ± 0.11, over Buenos Aires (34.6 • S, 58.5 • W). There are also satellite based studies from either lidar (Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP), Dupont et al., 2010) or cloud radar (CloudSat) or combined lidar and cloud radar (e.g. Sassen et al. 2008) retrievals that provide a global view concerning the seasonal frequencies of cirrus clouds and their geometrical and optical properties and their variabilities.
However, there are only few long-term studies based on ground-based lidar systems, while these have a limited geographical 65 distribution. This kind of observations that correspond to different areas and atmospheric conditions are crucial to reveal information of the latitudinal dependence of the cirrus properties and can provide indications about the aerosol effect on the geometrical and optical characteristics of the detected cirrus layers. On top of that, these observations can be further used in the validation and improvement of the satellite retrievals, which provide global distribution of cirrus clouds (Sassen et al., 2008).
Given that for satellite retrievals, the main input parameter to the optical processing of the cirrus layers is the lidar ratio, the 70 selected lidar ratio value can introduce errors on the retrieved extinction and optical depth values of the cirrus layers, as it is illustrated by Young et al., (2018). The optical depth comparison of the Version 4.10 (V4) of the CALIOP optical depths and the optical depths reported by MODIS collection 6 show substantial improvements relative to earlier comparisons between CALIOP version 3 and MODIS collection 5, as a result of extensive upgrades of the extinction retrieval algorithm. New apriori information of the lidar ratio value for the cirrus layers, included in Version 4.10 (V4) of the CALIOP data products, led to 75 improvements of the extinction and optical depth estimates of the cirrus cloud layers. Thus, ground based lidar observations of the cirrus properties, that correspond to different areas and atmospheric conditions, are crucial to verify and eventually improve the satellite retrievals.
The aim of this work is to retrieve and analyze the cirrus geometrical, intensive and extensive optical properties at different latitudes (subtropical and subarctic), from observations derived with the same ground based lidar system, which partly fills the 80 gap concerning the latitudinal coverage of existing ground-based lidar studies. Then the observed differences are discussed in order to identify the possible causes. The information of the lidar ratio is an important parameter for the inversion of lidar signals in instruments that do not have Raman channel and space-based lidars, such as CALIPSO, and depend on a parameterization that may vary with location. Thus, information provided by well-calibrated ground based measurements is quite critical. Analysis of the lidar ratios values derived from lidar measurements in different parts of the world, where different 85 atmospheric and aerosol conditions prevail, will provide results that are more representative of the actual conditions and thus their use will lead to reductions in the uncertainties of the satellite retrievals.
The manuscript is structured as follows: after a brief description of the portable lidar system (Polly XT ) and the measuring sites in Section 2, we present the data analysis algorithm and the methods applied for the optical products retrievals in Section 3. The lidar derived statistical analysis and seasonal variations of geometrical and optical properties of cirrus clouds in both 90 subtropical and subarctic areas over the period 2008-2016 are presented and discussed in Section 4. Concluding remarks are presented in Section 5.

2 Instrument and Measuring Sites
A multi-wavelength depolarization Raman lidar Polly XT of the Finnish Meteorological Institute (FMI) performed automated measurements during the period 2008-2016 in three different geographical regions. The system is based on a compact, pulsed 95 Nd:YAG laser, emitting at 355, 532 and 1064 nm, at 20 Hz repetition rate. The laser beam is pointed into the atmosphere at an off-zenith angle of 5 • , so the impact of the specular reflection by ice crystals into cirrus layers on the backscattered signals is negligible. The backscattered signal is collected by a Newtonian telescope, with 0.9m focal length. The vertical resolution of the signal profiles is equal to 30m and the temporal resolution is 30s. The setup of the system includes two Raman channels at 387 and 607 nm, three elastic channels at 355, 532 and 1064 nm, a depolarization channel at 355nm (for India and South 100 Africa), a water vapour channel at 407nm and depolarization channel at 532nm (cross-polarization with respect to the initial emitted polarization plane) for Kuopio. Detailed description of the system is provided in Althausen et al., 2009 andEngelmann et al., 2015. All measurements processed within the period 2008-2016 are available online at http://polly.tropos.de. A more detailed description of the system components is presented in Table 1.
The Polly XT has participated in two campaigns in two subtropical areas, within the framework of the EUCAARI (European

105
Integrated project on Aerosol Cloud Climate and Air Quality interactions) project (Kulmala et al., 2011) (Giannakaki et al., 2015). Figure 1 presents  technical problems with the laser, the data coverage from September to January was limited. Measurements could not be done in October 2008 and January 2009, and in September and November-December only a few usable profiles were measured . Measurements in Elandsfontein were performed almost continuously, as two periods were dedicated to the maintenance of the system (the one from December 23rd to January 26th 2009 and the second one, from October 23rd to November 23rd 2010). Since November 2012 the Polly XT is operating in Kuopio (62.74 • N,27.54 • E,190m a.s.l) in Finland 115 providing continuous measurements and information for the presence of clouds of the subarctic area (Filioglou et al, 2017). The three measurement sites constitute regions with different atmospheric conditions and different sources of aerosol and evaluation of the cirrus dataset in these latitudinal experimental sites provides valuable information of the regional characteristics of the measured cirrus properties. 4 3 Geometrical and optical retrievals of cirrus clouds 120 3.1 Description of the cirrus retrieval algorithm Several steps were followed for the processing of the signal at 1064nm derived by the Polly XT , needed for the estimation of the cirrus boundaries. These are illustrated in Figure 2. Firstly, the signal to noise ratio (SNR, Eq. 1) is calculated according to the following equation (Georgousis et al., 2015): The SNR is selected above 3.5 (above this threshold value the boundary layers estimation found independent from SNR), since the lidar signal is strongly attenuated at higher altitude levels and the noisy parts of the signal should be rejected. Then, the zero and background levels are subtracted and the range-corrected signal is calculated. In the next step, we normalize the range-corrected signal by its maximum value found below 1.5km, so as to enhance the applicability of the method in various atmospheric conditions. Given that lidar signals are uncalibrated and signal levels from one lidar system to another can be 130 rather different, the normalization ensures the applicability of the criteria used by Baars et al., 2008. After these corrections are made, the Wavelet Covariance Transform (WCT) is applied to the range corrected signal. The method used (Eq. 2), detects discontinuities in the lidar signal, such as the top of the boundary layer, elevated aerosol layers or cloud boundaries, allowing the detection of cirrus cloud base and top (Brooks, 2003).
In Eq. 2, P(z) is the product profile where the WCT is being applied to. WCT is the result of the transformation, z is the altitude, b is the height at which a noticeable change in the normalized signal occurs, and α is the dilation chosen. A critical step to the accurate WCT application to the signal is the selection of an appropriate value of the window (dilation), so as to distinguish cloud layers from aerosol layers. In our case, a dilation of 225m, is chosen, proportional to the cirrus geometrical depth (Baars et al., 2008). Another critical step is the threshold WCT value for the determination of the cirrus boundary. A 140 threshold value of 0.1 is selected as a detection limit for both the base (-0.1) and the top (+0.1) of cirrus cloud (Baars et al., 2008) after sensitivity studies. The WCT transformation has already been applied successfully on cirrus cloud detection (Dionisi et al., 2013).
Finally, cloud retrievals from the algorithm are classified as cirrus clouds when the following four criteria were met: i) the particle linear depolarization value is higher than 0.25 (Chen at al., 2002;Noel et al., 2002), ii) the altitude is higher than 6km 145 and iii) the base temperature is below -27 • C ( Goldfarb et al., 2001;Westbrook et al., 2011) and iv) the top temperature is below -38 • C (Campbell et al., 2015). The application of these criteria is made so as to avoid the presence of liquid water. It should be pointed out that lidar measurements were processed only in the absence of lower tropospheric (below 4 km) thick clouds.
The application of the WCT on a case of cirrus layer observed on July 20th 2016 at Kuopio station, for a time period between 00:00 and 01:00 UTC is presented. In our study, 60-min averages are computed and the respective mean value are 150 taken as cloud base and top height. The hourly mean wavelet applied to the corrected 1064 signal and the hourly mean particle depolarization ratio and the backscatter coefficient profile of the cirrus evolution are presented in Fig. 3. The temperature values are also plotted with white line and the threshold values are marked with red lines.

Retrieval of the optical properties of cirrus
The integration of the extinction profile between the defined cloud base and the top of the cirrus layer is calculated to obtain 155 the cirrus optical depth (COD) from the lidar measurements as shown in Eq. (3).
The night-time measurements from Polly XT were processed by the Raman method, which allows the independent determination of the extinction and backscatter coefficients, thus providing the lidar ratio (extinction-to-backscatter ratio) (Ansmann et al., 1992). For the retrieval of the cirrus extinction coefficient profiles obtained from the daytime measurements, the inte-160 gration of the backscatter profile multiplied by the lidar ratio is calculated. The daytime measurements from Polly XT were processed using the Klett inversion (Klett, 1981;Fernald, 1984), with respect to the ratio of the extinction to the backscatter coefficient. These two unknowns have to be related using either empirical or theoretical methods in order to be able to invert the lidar equation. In our study, the lidar ratio was determined by comparing the forward and the backward solution of Klett and the effective lidar ratio value was chosen as the value for which the aforementioned profiles tend to coincide (Ansmann 165 et al., 1992). Both the daytime and night-time optical products were derived for each 1hour averaged profiles. The calculation of the corresponding molecular backscatter and extinction profiles was made based on temperature and pressure profiles obtained from radio soundings. Radiosondes launched daily at 06 and 18 UTC at the Jyvaskyla Airport, located to the southwest Africa were used in the processing of the other two sites. Another important lidar quantity to be calculated is the particle depolarization ratio. This ratio constitutes a qualitative way to discriminate particle shapes and to distinguish spherical from non-spherical particles. Cirrus generally cause enhanced particle depolarization values, higher than 0.25 (see for e.g. Chen et al., 2002), depending on the ice-particle shape and orientation (Lynch et al., 2001). The calibration of the depolarization measurements, needed for the calculation of the particle depolarization ratio, was determined by using the geometric mean of the 175 two ±45 • measurements, following the procedure described by Freudenthaler et al., (2009). The particle depolarization ratio is presented only for the dataset of Kuopio, as for the other two sites only the Rayleigh calibration method for the calibration measurements was available.

3.3 Multiple scattering correction on optical products
The lidar equation assumes single scattering from the hydrometeor, but eventually the received photons could have been scat-180 tered multiple times before reaching the telescope. This effect, named multiple scattering, is considerably important primarily to the measured extinction coefficient values of cirrus clouds, and secondly to the calculated cirrus optical depth and the estimated lidar ratio values. Multiple scattering depends not only on cloud optical depth and cloud extinction, but also on the lidar system components, such as the laser beam divergence and the full-angle field-of-view of the receiver.
The relative influence of multiple scattering decreases with increasing height within the cloud, and the errors of the extinction 185 coefficient can be even equal to 60% at the cirrus base (Lynch et al., 2001). As generally, multiple scattering effect cannot be negligible in a receiver field of view equal to 1mrad (Wandinger, 1998), this effect on cirrus clouds optical properties was considered and corrected in this study. In order to calculate the multiple scattering contribution to the calculated optical products, the Eloranta model (Eloranta, 1998) was used to estimate the ratio between the total received power and the contribution of the single scattering, the ratio P tot (z)/P (z) (Eq. 4). The single extinction coefficient a 1 is then related to the actual (multiple 190 scattering) coefficient a(z) through the parameter F as shown in Eq. (5) (Wandinger, 1998).
The model assumes cirrus consist of hexagonal ice crystals and the required inputs are: (i) the laser beam divergence, (ii) the receiver field of view, (iii) the cirrus effective radius, (iv) the measured single scattering extinction profile (or the lidar 195 ratio multiplied by the backscatter for the daytime measurements) and (v) the order of scattering. The estimation of the cirrus effective radius was taken from Wang and Sassen (2002), based on the linear relation of the effective radius with the cirrus cloud temperature derived from radio soundings. For the multiple scattering calculation, the code applies an iterative method including the following steps: i) The measured extinction profile of the cirrus layer is provided (a 1 ).

200
ii) With the provided effective radius profile of the cirrus layer (linear relation of the effective radius with the cirrus temperature derived from radio soundings) and the measured extinction coefficient, an iterative procedure provides the ratio iii) From (ii) a first value for the correcting factor F(z) can be worked out.
iv) The iterative procedure continues till the calculation of a stable correcting factor F(z) is found.

205
v) The corrected extinction can be then calculated from equation (5) and hence the value of lidar ratio.
The model has already been validated against other models (Hogan, 2006) in order to correct the derived optical characteristics of cirrus clouds and has already been applied in cirrus lidar applications (for e.g. Giannakaki et al. 2007). In the following sections, the cirrus optical properties (lidar ratio, extinction coefficient, and optical depth) derived in the frame of this study were corrected for multiple scattering.

4 Results and discussion
In the following section, we present the mean geometrical and optical properties of the detected cirrus layers within the period 2008-2016 for the three measurement sites, which correspond to subtropical and subarctic regions, and we further discuss the differences between the retrieved properties.

Cirrus cloud cover detection 215
Cirrus cloud detection over the three regions is presented in Fig. 4. The detected cirrus clouds over Gual Pahari cannot provide any monthly trend and cannot be representative of an annual pattern. The time periods with technical issues (mentioned above) and the occurrence rate of low clouds observed between March and September leaded to a limited dataset of cirrus observations.
Concerning the annual pattern observed over Elandsfontein, the maximum detection of cirrus layers is reported during May and December. No data processing could be performed during unfavourable weather conditions, such as the presence of low 220 cloud, observed mainly the months between January to April with a percentage of ∼ 30% of the total measurement period. The analysis of measurements over Kuopio showed that the cirrus cloud cover was found to vary both diurnally and seasonally. From the available data, the detection of cirrus clouds appears to exhibit an annual pattern with the maximum detection from February to September and minimum occurrence during the period between October and January, given the favorable meteorological conditions. Layers of low water clouds were present all year long, with the peak of monthly occurrence between April (28 225 cases) and November (27 cases). This monthly pattern of low clouds existence seems to follow the annual temperature cycle over the region (Jylha et al., 2004), with maximum temperature values observed during the period April to October. Concerning the diurnal pattern, the number of detected cirrus clouds during nighttime is higher from March to Semptember, and lower in the period from October to January.

Geometrical properties of cirrus clouds over the sub-arctic and tropical sites 230
Mean cirrus cloud geometrical thickness reported in literature from satellite retrievals is about 2.0 km globally (Sassen et al., 2008), while a broad distribution of geometrical boundaries from ground-based systems have been reported in literature (e.g., Gouveia et al., 2017;Seifert et al., 2007;Hoareau et al., 2013). Figure 5 shows the monthly variations of cirrus base height and the cirrus top height (displayed in monthly boxplots) derived with the automated algorithm, with the corresponding mean temperatures above each site. The cirrus geometrical properties show a broad monthly distribution ranging from 6790m to 235 13070, having the larger variability in the two subtropical sites compared to the subarctic site.
The cirrus lidar dataset in Gual Pahari (28.43 • N, 77.15 • E,243m a.s.l -Northern hemisphere) region is the less extensive one compared to the other two sites and limitations due to the low signal to noise ratios exist. Indeed the sampling might not be statistically representative of the cirrus cloud properties, but some first results can be discussed. Specifically, during the one-year-long measurement period, Polly XT was measuring on 183 days, corresponding to 2500h in total. The mean value of 240 cirrus base is calculated 9000 ± 1580m, whilst mean top is found to be 10600 ± 1800m, with mean geometrical thickness of 1500 ± 700m. The temperature varied from -27 • C to -50 • C. Our results are consistent with another study over North China (Min et al., 2011), based on CALIOP satellite measurements. In this study a value of 1600 ± 1015m is reported for the cirrus geometrical thickness. According to this study, the cirrus top temperatures were found lower than -50 • C and higher than - Infrared Pathfinder Satellite Observations) measurements (Sassen et al., 2008). The averaged geometrical properties between daytime and nighttime are found to be nearly identical above all sites, with differences less than 0.3km.

Optical properties of cirrus clouds over the sub-arctic and tropical sites 260
This section presents the cirrus optical properties for the three regions and Figure 6 shows the monthly variations of the cirrus optical properties (displayed in monthly boxplots) above each site.
The COD values over the three sites are presented in Figure 6a and 6b. For the subtropical region of Gual Pahari the mean COD 355 is 0.59 ± 0.25 and the mean COD 532 is found to be 0.45 ±0.30. The classification of clouds according to Sassen and Cho (1992) the percentage of 2% is categorized as subvisible cirrus, 61% as thin cirrus and 37% as opaque cirrus. For the Kuopio, the column-integrated mean corrected COD at 355nm is 0.25 ± 0.2, and is found to vary between 0.018 and 1.53, while the mean COD 532 is found to be 0.24 ± 0.20. The highest values of COD are found between January and March, with the highest value to be 0.37 ± 0.18 ( Dionisi et al., 2013). Reichardt (1998) reported that cirrus clouds optical depth values were lower than 0.3 280 for 70% of the cases processed for northern midlatitude cirrus. The classification of cirrus clouds according to Sassen and Cho (1992)  In what follows we proceed into finding a connection between the COD values derived in the different sites and the AOD load over the regions which are exposed to different aerosol sources. Table 4 lists the predominant aerosol type over each region and the results from the analysis of AOD at 355 nm in the free troposphere and the calculated COD values. We can conclude that there is an indication of the relationship of the aerosol load on the derived cirrus statistics, as the higher AOD values are linked with the higher COD values calculated for the two subtropic regions. More specifically, the one year aerosol analysis of 290 lidar observations in Gual Pahari  showed that in the summer, the measured air masses were slightly more polluted and the particles were a bit larger than in other seasons (higher Angstrom exponent values), with the main aerosol sources to be the local and regional biomass and fossil fuel burning. The annual averages revealed a distinct seasonal pattern of aerosol profiles, with aerosol concentrations slightly higher in summer (June -August) compared to other seasons, and particles larger in size. During the summer and autumn, the average lidar ratios were larger than 50 sr, suggesting the presence 295 of absorbing aerosols from biomass burning. The lidar observations that were performed at Elandsfontein and used for aerosol characterization for the corresponding study period (Giannakaki et al. 2016) showed that the observed layers were classified as urban / industrial, biomass burning, and mixed aerosols using the information of backward trajectories, MODIS hotspot fire products and in situ aerosol observations. The analysis of the seasonal pattern of vertical profiles of the aerosol optical properties showed that the more absorbing (higher lidar ratio at 355 nm) biomass particles were larger on August and October, sr, using a micropulse lidar. For the Elandsfontein site, the mean LR 355 value is found to be 26 ± 6 sr and the mean LR 532 is 25 ± 6 sr and the lidar ratios reach their highest values during April. A mean value lidar ratio of 33 ± 7 sr at 355nm is observed for the whole period studied over Kuopio site, with higher variability observed on June, while the corresponding 310 mean value LR 532 is calculated to be 31 ± 7sr, without any obvious seasonal cycle. Specifically, the mean LR 355 for the corresponding months are calculated as follows: 33 ± 7 sr (MAA), 34 ± 7 (JJA), 33 ± 7 (SON) and 34 ± 5 (DJF). For opaque, thin and sub-visible cirrus clouds the means are 31 ± 6 sr, 34 ± 7 sr, and 35 ± 7 sr, respectively. Gouveia et al. (2017) found a mean LR 355 value of 23.9 ± 8.0sr (SD) for the tropical region of Amazonia, while Giannakaki et al., (2007), reported a corresponding value of 30 ± 17sr for a mid-latitude station. Josset et al. (2012) and Garnier et al. (2015) analyzed spaceborne 315 CALIOP (Cloud Aerosol Lidar with Orthogonal Polarization) lidar observations. Both studies concluded that cirrus lidar ratio (corrected for multiple scattering effects) around the globe has typically values of 30-35sr ± 5-8sr at 532 nm. Nevertheless, the lidar ratio values may vary greatly depending not only on the altitude and composition of the cirrus clouds (Goldfarb et al., 2001), but also on the correction of the multiple scattering effect (Platt, 1981;Hogan, 2008). The aforementioned depends on the ice crystals effective radius and the associated uncertainty could range from 20 to 60% (Wandinger, 1998). Lidar ratio for 320 cirrus clouds was assumed to be constant with altitude and season with a value of 25 sr using the CALIOP extinction retrieval algorithm (Young et al., 2013;Young and Vaughan, 2009), but this value has changed in the upgraded algorithm, as illustrated by Young et al. (2018).
Concerning the monthly variability of the depolarization values ( Figure 6e) over Kuopio, no clear tendency is observed. The higher monthly mean value was observed on July, but the variability was less than 0.04 between months, with a mean value of 325 0.38 ± 0.07.
As the assumption that the backscatter and the extinction coefficients for sufficiently large cirrus particles are spectrally independent; the color ratio (ratio of backscatter profiles, CR) at 355 and 532 is supposed to be equal one. This assumption is also used in satellite processing schemes. However, it is reported that the measured variability of cirrus color ratios is much larger than previously realized and that measured color ratios are higher in the tropics (Vaughan et al., 2010). For the Kuopio 330 station, mean CR is found 1.1 ± 0.8, while for the less extensive dataset of New Delhi the mean value is found 1.5 ± 0.8 and for Elandsfontein the mean value is 1.4 ± 1.1. Table 3 summarizes the mean optical values discussed above, for the three sites, separating daytime and nighttime observations. Generally, the averaged optical properties values are found to be nearly identical, except one site (New Delhi), where average nighttime optical properties found higher (∼ 4sr) than that of daytime.

335
To further investigate the distribution of the cirrus lidar ratio values over Kuopio, we present a histogram of the values derived in Fig. 7. The most frequent observed lidar ratio values ranging between 28 and 36 sr for 355 nm and 20 and 36 sr for 532 nm. Similar results have been retrieved regarding the variability of LR 532, which is constant from one month to another, as shown. This figure can provide an evidence that although the lidar dataset are not continuous (due to not favour weather conditions the winter months), the frequency distributions are close to normal and thus the statistics shown here have 340 a significance. In addition we can claim that with this scarce sample of data we observe consistent results with a number of other literature studies.
In Figure 8, we examine the dependence of the LR 355 with the COD 355 values on intervals of 5 sr. The dashed lines indicate the categories defined by Sassen and Cho (1992). The most common lidar ratio values from 25 to 40 sr are found for quite low COD values (corresponding to thin cirrus) for the subarctic station.

Cirrus classification at Kuopio
The classification of cirrus clouds according to Sassen and Cho (1992) is made based on the COD values. Ground-based lidars are well suited for thin cirrus layers observations, due to their sensitivity to thin atmospheric features, in contrast to spaceborne lidar observations (Martins et., 2011). For this reason, additional analysis on each cirrus category is also conducted for Kuopio site as measurements in this station represent the most extensive dataset between November 2012 and December 2016. . Their mean COD 355 is calculated 0.021 ± 0.0031, their mean LR 355 is 34 ± 7sr and their mean particle depolarization value is 0.45.

Category "Thin"
As mentioned previously, thin cirrus is the most predominant type of cirrus in our study, with 152 observations. Thin cirrus 365 can also be undetectable by passive remote-sensing satellites, especially the ones with COD less than 0.2, and have so far not systematically been characterized. Their geometrical thickness found to be 1100 ± 586 m. Their mean COD 355 is calculated 0.16 ± 0.07, their mean LR 355 is 34 ± 7 sr and their mean particle depolarization value is 0.3 ± 0.13.

370
Opaque cirrus are the one with the highest value of optical depth that contribute the most of the total radiative forcing (Kienast-Sjogren et al., 2016). In our study, a total of 55 measurements of opaque cirrus are processed. Their mean geometrical thickness is found to be 1462 ± 659 m, higher than the value of all cirrus categories. Their mean COD 355 is calculated 0.5 ± 0.21, their mean LR 355 is 31 ± 6 sr and their mean particle depolarization value is 0.33 ± 0.12.

375
The three presented datasets are derived from different latitudinal and climatic sites. In this section we firstly examine the latitudinal dependence of the cirrus geometrical and optical properties. The reported values in literature from previous studies based on lidar grounbased dataset and the retrievals of the current one are listed in Table 5 and plotted in Figure 9 for comparison. We can note, that the cirrus geometrical properties and the lidar ratio values may vary greatly depending on the latitude and an decreasing trend of the geometrical boundaries with the rise of the distance from the equator is obvious, also reported by satellite 380 observations (Sassen et al., 2008). Generally, cirrus layers have been observed up to altitudes of 13km above the subtropical sites, whereas they have only been detected to about 1km lower at the subarctic region and this conclusion is in accordance with the Cloudsat observations (Sassen et al., 2008). Based on the satellite information, the derived cirrus cloud thicknesses was found to be larger in the tropics and decreasing toward the poles. Also from the values reported from groundbased studies, a pattern can be concluded: cirrus cloud geometrical properties peaks around the equator and at midlatitudes sites, with generally 385 decreasing amounts as the poles are approached. On the other hand, the lidar ratio values seem to follow a diverse relation, showing greater values moving to the poles. In our study, lower LR are observed for Gual Pahari and Elandsfontein and higher mean value for Kuopio. The larger variability of the optical properties at the two subtropical regions, relative to Kuopio, could be related to the larger and variable aerosol load over these regions. Overall, our results seem to demonstrate that subarctic cirrus clouds are colder, lower and optical thinner than subtropical cirrus clouds. However, a more extended database is needed 390 to strengthen these indications.
The dependence of geometrical and optical properties on mid-cirrus temperature is also examined in Fig. 10. In order to investigate this dependence, we have grouped cirrus clouds temperatures into 5 • C intervals. The number of cases per temperature bin are also labeled. Temperature values are obtained from radio soundings, as mentioned above. Thicker clouds ( ∼ 1.5 km) are observed at temperatures between ∼ -45 • C and ∼ -35 • C, with decreasing thickness reported for lower temperatures, 395 for both the subtropic and subarctic regions and a second peak is found in the range between ∼ -75 • C and ∼ -65 • C for the subarctic station. A similar trend has been reported for a midlatitude region by Hoareau et al. (2013), where thickest cirrus layers were found about -42.5 • C, and thinner ones at both colder and warmer temperatures. Another study (Pandit et al., 2015) reports that the geometrical thickness increases from 1 to 3.5 km as mid-cloud temperature increases from -90 to -60 • C, while for the further increase in temperature from -60 to -20 • C, the geometrical thickness decreases to less than 1 km. Concerning 400 the optical properties shown in Fig. 10, a steady increase of lidar ratio from -25 • C to -40 • C is noticed for the two subtropical stations,while the variability of this parameter is relatively constant across months for the subtropic station, with a slightly increase at warmer temperatures ( Figure 10b). There are indications that the cloud optical depth increases with the increasing cirrus mid temperature for the two subtropical sites (Fig. 10c). At cold temperatures (∼ -65 • C), optical thickness for cirrus layers of the subarctic station is high, compared to warmer temperatures and also cloud thickness for this temperature is sim-405 ilar high ( ∼ 1.5 km). The dependence of the particle depolarization values on base temperature is also examined (Fig. 10d).
No clear tendency is found, as the variability of this parameter is relatively constant, with a slightly increase of the particle depolarization with the increasing mid temperature. This behaviour indicates a relation between cirrus ice crystal shape and temperature, however, more studies should be done in order to examine this behaviour on various geographical locations. Figure 11 presents the color ratios values on 5 • C intervals of cirrus mid temperature, indicating an almost stable behavior 410 with temperature. Generally, we can conclude that for higher altitudes, lower spectral dependence is noticed, taking also into account the number of measurements performed at each site.
The dependency of the mid temperature with the lidar ratio values at 355nm and the particle depolarization values is further examined (Figure 12). Fig. 12 shows that the highest values of cirrus lidar ratio (>40) correspond to higher values of cirrus depolarization (>0.4) and warmer cirrus. Moreover, it can be seen the variety of depol values that correspond to the mean value 415 of lidar ratio (∼ 31). A similar behavior is reported in Chen et al. (2002) for lidar ratio values higher than 30 sr. In his study, the relationship between the depolarization ratio and the lidar ratios shows the former split into two groups for lidar ratios higher than 30. The first group has high depolarization ratios about 0.5 and the second one has 0.2.

Conclusions
Observations of cirrus clouds geometrical and optical properties, performed with a ground-based multi-wavelength Polly  optical properties on mid temperature shows quite similar tendency, but less variability for the subarctic dataset. Cirrus found geometrical and optical thickest at temperatures between -45 • C and -35 • C. At temperatures below -55 • C, the optical thickness of cirrus layers becomes again high and this trend appears only for the subarctic station. However, we should keep in mind that the number of samples corresponding to temperatures below -60 • C is limited. The lidar ratio is found to be quite constant with temperature, with a slightly increase in the warmer mid temperatures, showing larger variability for the subtropic datasets, 440 while the particle depolarization values seem almost constant at temperatures between -27 • C and -60 • C.
The geometrical and optical properties of cirrus layers are studied in detail, providing information useful in the validation of the cirrus parameterizations in models. Furthermore, our results could be useful for lidar ratio selection schemes needed by satellite optical properties retrievals of cirrus layers over different locations, e.g., the upcoming EarthCARE (Earth Cloud Aerosol and Radiation Explorer) mission. The spectral dependence discussed above, is another important issue for the satellite 445 algorithm schemes, given the different wavelengths applied among the different satellites.
In any case, further cirrus observations must be conducted, so as to investigate whether differences in the background aerosol load contribute to potential differences in the cirrus cloud geometrical and optical properties and which are the different atmospheric mechanisms leading to these differences over the different regions. Competing interests. The authors declare that they have no conflict of interest. Temperature base (C) -33 ± 6 -34 ± 5 -43 ± 10 Temperature top (C) -45± 4 -45± 6 -57 ± 9       Table 5 for references), stars denote estimations from this study and lines correspond to CLOUDSAT estimations according to Sassen et al.
(2008) (a), latitudinal dependence of cirrus temperature base and top. Circles denote estimations from groundbased lidar from the literature (see Table 5 for references), stars denote estimations from this study and (b) same as above, but for latitudinal dependence of lidar ratio values (c).