Radiative properties of mid-latitude cirrus clouds derived by automatic evaluation of lidar measurements

Cirrus, i.e. high thin clouds that are fully glaciated, play an important role in the Earth’s radiation budget as they interact with both longand shortwave radiation and determine the water vapor budget of the upper troposphere and stratosphere. Here, we present a climatology of mid-latitude cirrus clouds measured with the same type of ground-based lidar at three mid-latitude research stations: at the Swiss high alpine Jungfraujoch station (3580 m a.s.l.), in Zürich (Switzerland, 510 m a.s.l.) and in Jülich (Germany, 100 m a.s.l.). The analysis is based on 13’000 hours of measurements from 2010 2014. To 5 automatically evaluate this extensive data set, we have developed the "Fast LIdar Cirrus Algorithm" (FLICA), which combines a pixel-based cloud-detection scheme with the classic lidar evaluation techniques. We find mean cirrus optical depths of 0.12 on Jungfraujoch and of 0.14 and 0.17 in Zürich and Jülich, respectively. Above Jungfraujoch, subvisible cirrus clouds (τ < 0.03) have been observed during 7% of the observation time, whereas 10 above Zürich and Jülich significantly less. From Jungfraujoch, clouds with τ < 10−3 can be observed three times more often than over Zürich and Jülich, and clouds with τ < 2 × 10−4 even ten times more often. Above Jungfraujoch, cirrus have been observed to altitudes of 14.4 km a.s.l., whereas only to about 1 km lower at the other stations. These features highlight the advantage of the high-altitude station Jungfraujoch, which is often in the free troposphere above the polluted boundary layer, thus allowing to perform lidar measurements of thinner and higher clouds. In addition, the measurements suggest a change in 15 cloud morphology at Jungfraujoch above ∼13 km, possibly because high particle number densities form in the observed cirrus clouds, when many ice crystals nucleate in the high supersaturations following rapid uplifts in lee waves above mountainous terrain. The retrieved optical properties are used as input for a radiative transfer model to estimate the net cloud radiative forcing, 20 CRFNET, for the analysed cirrus clouds. All cirrus detected here have a positive CRFNET. This confirms that these thin, high cirrus have a warming effect on the Earth’s climate, whereas cooling clouds typically have lower cloud edges too low in altitude 1 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-32, 2016 Manuscript under review for journal Atmos. Chem. Phys. Published: 2 February 2016 c © Author(s) 2016. CC-BY 3.0 License.

; Sassen and Benson (2001) USA Sassen and Comstock (2001); Sassen et al. (2003Sassen et al. ( , 2007) ) Punta Arenas 53 The range-corrected signal r 2 P (r) detected by the ALS 450 can be described with the lidar equation (Kovalev and Eichinger, 2004;Wandinger, 2005): where β m and β p describe the backscatter from molecules and particles and α m and α p specify molecular and particulate extinctions, i.e. light attenuation by scattering and absorption, and take changes of scatterer density with altitude into account.
Instrumental properties are described by the constant C. O(r) is the overlap function which describes the overlap between the laser footprint and the telescope field of view.For the ALS 450 the complete overlap is achieved at a distance of 450 m from the lidar.As we analyze cirrus clouds that occurred entirely at greater heights above the lidar, we do not need to consider the overlap function.
The Leosphere ALS 450 measures the co-and cross-polarized components of the return signal.In order to solve equation 1 it is required to obtain the total signal from these two components.We calculate the total signal based on both channels as described by Rolf (2012).
From the detected co-and cross-polarized signal components the depolarization ratio can be obtained (Schotland et al., 1971).
Light that is scattered back by non-spherical particles changes its polarization state, whereas spherical particles do not change the state of polarization of the returned light.Therefore, the depolarization ratio provides information about the sphericity of the detected particles (Schotland et al., 1971;Kovalev and Eichinger, 2004).The cross-polarized signal from aspherical ice particles in thin cirrus often provides the better contrast than the parallel signal, a property we will use in our cloud retrieval algorithm.The lidar is pointed to 5 • off-zenith to avoid the effect of specular reflections of horizontally oriented ice crystal plates on the measured backscatter signal and depolarization ratio (Platt et al., 1978;Westbrook et al., 2010).
For our cloud detection scheme elaborated in Subsection 2.2, we use the backscatter ratio (BSR) defined as: To solve the lidar equation (Eq. 1) with four unknowns (β m , β p , α m and α p ) and only one measurement r 2 P (r), we need to make use of best current knowledge.The molecular quantities β m and α m are calculated from analysis data of the numerical weather prediction model (NWP) COSMO-2.We use pressure (p) and temperature (T ) from COSMO-2 (COSMO, 2015) to calculate the molecular density of air and determine β m and α m using Rayleigh theory (Bucholtz, 1995).
For the solution of Eq. ( 1) we use a lidar retrieval as described in Kovalev and Eichinger (2004).To ensure stable solutions, we use a far end boundary condition (Klett, 1981).Further, we need to define the extinction-to-backscatter ratio (hereafter referred to as lidar ratio).We derive = 0.234, the anisotropy of the molecules present in the atmosphere, from Eq. ( 6) in She (2001) and Table 1 in Bucholtz (1995) for our lidar wavelength of 355 nm.The lidar ratio of the molecular part is evaluated as: where σ R , given by Eq. ( 6) of She (2001), is divided by the expression for σ C π , provided in Eq. (4) of She (2001), as the receiver optical bandpass spectral width of 0.3 nm (<24 cm −1 at 28170 cm −1 ) suppresses the rotational Raman wing spectral contribution (Arshinov and Bobrovnikov, 1999).Note that our is called R A by She (2001).The particulate lidar ratio is defined as: Several studies have been performed to measure the particulate lidar ratio of cirrus clouds (e.g., Ackermann, 1998;Immler and Schrems, 2002;Larchevêque et al., 2002;Seifert et al., 2007).It can be obtained directly from Raman lidars that allow for an independent measurement of particle extinction and backscatter coefficients (Cooney, 1972;Giannakaki et al., 2007;Radlach et al., 2008;Reichardt et al., 2002;Achtert et al., 2013).In our retrieval we determine the lidar ratio such that BSR=1 above and below the cirrus cloud (e.g., Rolf, 2012).
The lidar equation (Eq. 1) assumes single-scattering of the emitted light in the direction 180 • to the emitted direction only.In reality, this is not strictly the case.As seen in Fig. 1 in Wandinger (1998), cloud particles produce strong forward scattering.(5) We use the multiple scattering model by Hogan (2008) as described by Wandinger (1998) and Seifert et al. (2007) to derive γ.The effective radius of the cirrus particles is taken from a climatology provided by Wang and Sassen (2002).For particles much larger than the detection wavelength, as it is the case for ice crystals observed with lidar, about 50% of the scattering occurs into the forward direction.In this study we find an average value for γ of 0.56 for Jungfraujoch and 0.52 (0.54) for Zürich (Jülich).
The lidar retrieval poses several uncertainties.Using NWP-data to calculate the molecular properties results in a maximal error of 2 %.However, there are uncertainties pertained to the data themselves.The lidar detector counts photons, and we calculate the counting error by means of poisson statistics.The assumed lidar ratio is also an error source.Here, we use lidar ratios that deviate ±5 sr from the determined lidar ratios to assess for the uncertainty caused by determining a lidar ratio.To assess the total, maximum uncertainty, we sum up the individual contributions to the uncertainty.Seifert et al. (2007) estimated the error in the multiple-scattering correction in the order of 10%.The signal-to-noise ratio (SNR) is determined by the variation within the 5-min-average profiles.As the Leosphere lidar does not allow to retrieve the photon counts directly from the data, we calculate the SNR from the original lidar profiles as where mean(r 2 P (r)) is the mean range-corrected signal and std(r 2 P (r)) the standard deviation of the originally retrieved lidar profiles over N = 600 shots (following a suggestion by P. Royer, Leosphere, private communication 11.8.2014).

Cirrus Detection Algorithm FLICA
For an efficient evaluation of this extensive data set, the automated data evaluation algorithm FLICA (Fast LIdar Cirrus Algorithm) was developed.The algorithm is based on a classical lidar retrieval (e.g., Klett, 1981;Kovalev and Eichinger, 2004) combined with a cloud detection scheme.FLICA analyzes profiles over 5 minutes (i.e.averages over 5 min ×60 s/min × 20 shots/s = 6000 shots).This time range was chosen to be short enough to ensure that all clouds could be detected by the algorithm while long enough to provide profiles smooth enough for the lidar retrieval to function.The output of the cloud detection scheme has been visually inspected for individual days and was found not to show any apparent artifacts.There is a trade-off between detecting cloud structures small enough and avoiding misclassifying noise as a cloud, especially for daytime measurements.The combination of the criteria below represents a rather conservative approach, which might result in missing some particularly small/thin clouds.The conservative approach ensures that no noise is misclassified as a cirrus cloud.An example of the resulting cloud detection can be seen in Fig. 1.
The FLICA algorithm contains the following steps: I Cloud top detection.The cloud top is needed as an upper boundary for the subsequent lidar retrieval.Our cloud top detection averages individual lidar profiles so that the resolution of one pixel is 5 min in time and 30 m in altitude.Areas of 3 × 3 pixels are examined, with the pixel to be checked for cloudiness in the center.At least 8 of the 9 examined pixels have to have a volume depolarization larger than 0.007(0.006)for day(night)-time measurements.At least 8 of 9 pixels also have to have larger co-and cross-polarized raw signals than empirical thresholds.
II Setting the far-end boundary condition for the lidar retrieval.The mean of the co-polarized signal at altitudes from detected cloud top to 500 m above cloud top is computed for each profile individually and used as far-end boundary condition for the lidar retrieval as described in Klett (1981).At this boundary, we assume a BSR of 1.This assumption introduces no error if the aerosol density above the cloud is the same as the one of the interstitial aerosol.If these densities are different we estimate from our in-situ observations (http://www.iac.ethz.ch/groups/peter/research/Balloon_soundings/COBALD_sensor) that the error introduced is of the order of 1 -2 %.
III Lidar retrieval.The lidar retrieval is performed as described in Chapter 5 in Kovalev and Eichinger (2004) to solve the lidar equation (Eq. 1) and calculate the extinction coefficients and BSR of the cirrus cloud.The retrieval is performed for a set of lidar ratios (5:5:40) and the best choice was determined such that BSR is closest to 1 below the cirrus cloud.
The BSR is corrected during the retrieval such that the mean BSR in the range 500 m above the cloud top is equal to 1.
IV Cirrus cloud detection.The cloud detection scheme is based on the retrieved BSR and volume depolarization as follows: Resolution of one pixel: 5 min in time, 30 m in altitude.
Areas of 3 × 3 pixels are examined, with the pixel to be checked for cloudiness in the center.At least 8 of the 9 examined pixels have to have a volume depolarization larger than 0.007(0.006)and a BSR larger than 1.08(1.03)for day(night)time measurements.
Temperature has to be lower than -38 • C to ensure pure ice clouds (this is checked using COSMO-2 or COSMO-7 analysis data).
The detection is applied to each pixel at each time independently.(2008) as described in Wandinger (1998) and Seifert et al. (2007).
VII Optical depth.The optical depth τ of the detected cirrus cloud is calculated by integrating over the retrieved extinction profiles.
VIII Radiative effect.The optical depth τ combined with temperatures from COSMO-2 or COSMO-7 are used to calculate the radiative effect of the cirrus cloud by means of the model of Corti and Peter (2009).

Measurement sites
Here we present the retrieved lidar cirrus climatology.First, a description of the different measurement sites shown in Fig. 2 is provided.Subsequently, we present the climatology of the cirrus cloud properties.The section ends with a comparison of our data with previous mid-latitude climatology studies.

Jungfraujoch
Jungfraujoch is the highest measurement site used in this study.It is located in the Swiss Alps (46.55 • N, 7.99 • E) at 3580 m a.s.l. at the top of the Aletsch glacier.Due to its high elevation, the research station is frequently situated in the free troposphere (Zieger et al., 2012), which is a great advantage for lidar measurements.The sphinx observatory, where our measurements took place, is one of the Global Atmospheric Watch (GAW) research stations.Therefore, long time series of meteorological measurement data are available for this site.Due to its high location and cold climate, the site poses challenges for instruments being able to run continuously.The Leosphere ALS 450 used in this study was built into a ventilated, temperature-controlled and regulated containment.Jülich is relatively flat.The Research Center itself is located in a rural area and therefore the lidar measurements might be less influenced by boundary layer aerosol than the Zürich measurements, despite nearby brown coal industry activity.

Cirrus Climatology
Following Section 2.2 we present the climatological evaluation of more than 13'000 hours of lidar measurements within the period 2010-2014 from the three mid-latitude measurement sites.The main information on the measurement statistics for the three sites is compiled in Table 2.More measurements are available from Jungfraujoch and Zürich than from Jülich.For Jungfraujoch, most of the data was retrieved in the spring, whereas during the summer only very limited data is available.In Zürich, on the other hand, a large number of the measurements took place during the summer, while the other seasons show similar coverage.The Jülich lidar was running predominantly during spring and summer, while the autumn and winter data is sparse.The amount of data has to be considered when judging seasonal variability.The retrieved cirrus properties listed in Table 2, indicate a temporal cirrus cloud coverage between 9 and 15 % for all stations, agreeing well with the CALIPSO measurements discussed by Sassen et al. (2008).0.17 +.02 −.08 (1) Refers to the number of hours lidar measurements with the ALS 450.Not included are times times when the ceilometer detected low-level clouds closer than 1.5 km.
(2) According to the specifications of the FLICA algorithm, see Subsection 2.2.
(4) Refers only to clouds at least 1 km above the lidar.
(5) As observed by the ALS 450.than 440 hPa) for January observed by the TIROS-N Operational Vertical Sounder, TOVS, averaged over 8 years, 1987-1995 (http://ara.abct.lmd.polytechnique.fr/index.php?page=clouds).For January, these data suggest decreasing amounts of high clouds when the air passes from the North Sea towards the Alps.Also, a time series of 40 years of measurements of turbidity in Jülich confirm the high cloud coverage during wintertime (personal communication A. Knaps, Forschungszentrum Jülich).However, there are large uncertainties in this part of our climatology, as the number of hours of measurements with the Leosphere ALS 450 available for the Jülich winter is small (Table 2), the specific winter might have had a particularly high cirrus cloud coverage, and the applied manual operation of the lidar might add bias.Another remarkable feature is that autumn maximum in cirrus coverage observed above Jungfraujoch.Also this feature is in qualitative agreement with the seasonal cycle suggested by the TOVS data set, but again interannual variability might be important but cannot be derived from our measurements.

10
The first property of interest is the distribution of the optical depths of the detected cirrus clouds, which we classify according to Sassen and Cho (1992)  hence termed subvisible.Cirrus clouds with an optical depth τ in the range 0.03 ≤ τ < 0.3 are termed thin, while clouds with τ ≥ 0.3 are referred to as opaque.The upper limit of detection for our lidars is τ ≈ 3, as for larger optical depths the light is almost fully extinguished within the cloud.Under these conditions no molecular signal from above the cloud can be detected (Immler and Schrems, 2002), as would be required for an inversion.Therefore, we are not able to specify the optical thickness of the thickest cirrus clouds.Chen et al. (2000) classified clouds with tops above 440 hPa (≈ 6500 m) and optical depths larger than 3.6 as cirrostratus.These cirrus clouds may have a negative cloud radiative effect CRF NET , but cannot be considered here because of the detection limits of our lidar instrument.
Figure 4b shows the optical depth of the retrieved cirrus clouds for different seasons.The grey dashed lines indicate the categories defined by Sassen and Cho (1992).The optical depth averaged over the whole data sets for each measurement site is displayed in Figure 4a.The occurrence frequency of subvisible cirrus clouds is larger for Jungfraujoch than for the other two sites.Two reasons may be responsible for the observed differences.First, Jungfraujoch is located in the central Alps, where orography-driven lifting of air masses leads frequently to mountain-wave (lenticularis) cirrus.These clouds are thicker than large-scale cirrus clouds, but thinner than the cirrus formed as outflow of anvils or in warm conveyor belts.The second reason is the enhanced detectability of optically thin clouds at Jungfraujoch as a result of improved signal-to-noise ratios (SNR, see Eq 6).The alpine site is located at an altitude of 3500 m asl, 3000 m above Zürich and 3400 m above Jülich.According to Eq. 1 the received signal depends on the inverse of the squared range between lidar and target.In addition, the Jungfraujoch is frequently above the boundary layer.Therefore, measurements from Jungfraujoch avoid strong beam extinction due to boundary layer aerosols.
Figure 5 provides vertical profiles of the cloud-mean signal-to-noise ratio (SNR) of the three stations, where the noise is obtained from Eq. 6.From the profiles it can be seen that the SNR of Jungfraujoch extends to greater heights by about 3 km.This suggests that the increased detection rate of thin and subvisible cirrus clouds is a result of the increased signal-to-noise ratio.Furthermore, the SNR at Jungfraujoch increases at heights above 13 km a.s.l.This suggests that the morphology of the clouds at these heights differs from the morphology of the highest clouds observed at Jülich and Zürich.The backscattering efficiency appears to be enhanced in these clouds, possibly because a large amount of small crystals formed in the observed cirrus clouds, when many ice crystals nucleated in the high supersaturations in rapid uplifts as they occur in lee waves above mountainous terrain (Lin et al., 1998a, b;Kärcher, 2003).
The number of detected subvisible cirrus as function of optical depth and cloud top altitude is depicted in Fig. 6.As expected, Jungfraujoch displays a larger fraction of subvisible cirrus as well as higher cirrus cloud tops.Therefore, we have  It is interesting to see that at Jungfraujoch the lower detection limit in optical depth of a few times 10 −5 is approached in a few cases.However, by far the most subvisible cirrus stay clearly above τ = 10 −4 proving that physical mechanisms prevent clouds so thin to survive for appreciable times.Nucleation is one such mechanism.In case these clouds nucleate homogeneously, this is likely to happen in nucleation bursts, which will provide the newly formed clouds immediately a minimum optical depth.
The same is true for heterogeneous nucleation, if the nucleation barrier and the number of nuclei are at all significant.One mechanism that might lead to extremely low ice crystal number densities is the formation of fall-streaks and subsequent dispersion of the particles.The rare occurrence of clouds with τ < 5 × 10 −4 suggests that such mechanisms do not often lead to the formation of so thin clouds.
To ensure that the highest cirrus clouds observed above Jungfraujoch were not volcanic particles, we have examined satellite measurements and found no indication for volcanic influences.The effect of the high altitude of Jungfraujoch can also be seen in the cloud tops at the different measurement sites (see Fig. 7a).The cloud tops are higher above Jungfraujoch than above Zürich and Jülich.The retrieved cloud tops agree well with the observations by Sassen and Campbell (2001) in Salt Lake City (40 • N, 12 • W, 1520 m a.s.l.) as well as by Hoareau et al. (2013)  amongst themselves, which may influence the results.However, the cloud top altitudes are very similar for the five mid-latitude stations.As Salt Lake City is located further south than the other sites, the slightly higher cloud tops may be a result of a higher tropopause being present over Salt Lake City compared to the other sites.Similarly, we see in Fig. 7b that the tropopause over Jülich, which is located further north than Zürich and Jungfraujoch, generally is lower.Between Zürich and Jungfraujoch, the tropopause reaches similar altitudes with a larger spread over Jungfraujoch (possibly due to the Alpine heat low affecting the To quantify the net radiative effect, CRF NET , for the cirrus clouds observed here we use the radiation model of Corti and Peter (2009), which is a simplified model based on the more sophisticated Fu-Liou radiative model (Fu andLiou, 1992, 1993).The cloud radiative forcing due to shortwave radiation, CRF SW , is dependent on the surface albedo, the solar zenith angle as well as the cirrus cloud optical depth τ (see Eq. 13 of Corti and Peter (2009) for details).The longwave cloud radiative forcing, CRF LW , is mainly determined by the temperature difference between the Earth's surface and the cirrus cloud top temperature and by the cloud optical depth τ (see Eq. 6 of Corti and Peter (2009) for details).The net cloud radiative forcing, CRF NET , is calculated as a superposition of these two effects (i.e.CRF NET =CRF SW +CRF LW ).The following parameters are needed as input for the calculation of the radiative effect: The values of the solar constant S, multiplied by the fraction of the day that the sun is above the horizon, and the mean solar zenith angle Z are set to 684 Wm −2 and 60 • (daily mean conditions with zero incoming flux at nighttime), respectively, as suggested by the online version of the radiation model (Corti and Peter, 2009b).This results in an incident solar flux I = 684 × 0.5 = 342 Wm −2 .The amplitude of the solar background noise in the lidar signal profiles is used to distinguish between day and night time.We use an albedo of 0.3 (corresponding to the global average value).The cloud optical depth τ is automatically calculated in the FLICA for 5-min profiles as described in Section 2.2.The temperatures needed for the radiation model (T srf and T cld ) are extracted from the COSMO-2 (for Jungfraujoch and Zürich) and COSMO-7 (for Jülich) model.The radiation model of Corti and Peter (2009) is well suitable to be used with lidar data, as the model does not require further information, such as ice crystal sizes or shapes, which the lidar measurements could not provide.

Comparison of CRFs with previous studies
We compare our computational results to satellite data, which have been averaged zonally at 50 • N or globally and combined with a radiative transfer model (Chen et al., 2000).The results of this comparison are listed in Table 3 together with maximum possible uncertainty ranges (see last paragraph of Subsection 2.1).The "overcast values" (i.e.taking only cloudy values into account) consider the radiative effect under conditions with cirrus clouds, while the "all sky values" include also conditions  (Chen et al., 2000) is similar to our observations, the ISCCP category of cirrus clouds show much larger cloud radiative forcings in the short-and longwave, CRF SW and CRF LW (by more than one order of magnitude).This can only be explained in terms of a much larger optical depth τ of the clouds observed by the satellites.
The reason for this at first sight surprising differences are the different definitions of "cirrus".First, FLICA detects only clouds with lower cloud edge colder than -38 • C, which is typically above 7-8 km.Chen et al. (2000) instead used a pressure threshold of 440 hPa to separate clouds, which corresponds to an altitude of 6.3 km (standard atmosphere).The clouds in the range 6.3-7.5 km are missing in our study.Especially in this altitude range thick cirrus stratus (still with τ < 3.6).Second, although our criteria allow for thick clouds up to τ = 3.6 at altitudes clearly above lower edge, they cut clouds once their lower edge gets warmer than -38 • C, which is more likely for thicker clouds (see rounded edge in the red boxes in Figure 6).Third, while we count vertically distinct cirrus layers as separate clouds, the geostationary ISCCP weather satellites add the signal of vertically staggered layers, which increases τ .Furthermore, it should be noted that satellite data reveal discrepancies amongst themselves (ISCCP, MISR, MODIS) with differences of 20-30% in coverage of cirrus with τ < 3.6 (Marchand et al., 2010).
Finally, the distribution of thicker cirrus with τ > 0.3 is zonally inhomogeneous, with clouds preferentially occurring at the continental east coasts.For the other parts of China lower values of 20 Wm −2 are found.The radiative forcing of the lateral boundary of cirrus clouds observed with CALIOP is investigated by Li et al. (2014).The lateral boundary with optical depths less than 0.3 is found to have still a radiative forcing of 10 Wm −2 .This value is similar to our mean overcast radiative forcing of Zürich and Jülich and demonstrates also the sensitivity of cirrus cloud inhomogeneity on cloud forcing as also found by Gu and Liou (2006).However, all these studies investigate single cases or different regions compared to our study.There is no study which investigates the cirrus radiative forcing over Europe.
In a study of Dupont et al. (2010) cirrus cloud observations over 2 years from the CALIPSO satellite lidar CALIOP are compared to four ground based lidar stations (two sites in the US and two in France) for their consistency of macrophysical and optical properties.They found larger discrepancies in the frequency distributions of cloud base, top and thickness.They point out that the significant part of the deviations can be attributed to different sampling (seasonal, irregular sampling of ground based stations, opaque low level clouds).However, they found that for high cirrus clouds the optical depth distribution τ >0.1 from ground stations and CALIOP is consistent within 10% using the same retrieval method.This shows that our optical depth distribution of all three stations is most likely not or only less affected by sampling issues of ground base lidar compared to satellite measurements.Further, we more closely examined the CRF NET categorized by their optical thickness τ as in Sassen and Cho (1992).

Influence of subvisible cirrus on the net radiative forcing by cirrus clouds
Subvisible cirrus clouds generally are not considered in numerical weather prediction models as their optical depths are considered to be too small.The overcast net radiative effect CRF NET , divided into the categories defined by Sassen and Cho (1992), is shown in Fig. 9.We see that the subvisible cirrus clouds indeed have an effect on the net cirrus cloud radiative forcing CRF NET .
On average they contribute about 4 % of the total CRF NET of cirrus clouds at Jungfraujoch and in Zürich, and 3 % in Jülich.
The maximal effect of 6 % is reached in Zürich during spring.As seen in Fig. 9, the thin and opaque cirrus clouds are the main contributors to CRF NET of cirrus clouds above each of the three stations, both by roughly equal shares (with a small dominance of opaque clouds).
Jungfraujoch displays the lowest CRF NET -values throughout the whole year.This pattern is also seen in the optical depths CRF NET : as more subvisible cirrus are observed at Jungfraujoch (cf.Table 2) and their contribution to CRF NET is smaller than the contribution by thin and opaque cirrus, this leads to a smaller CRF NET on Jungfraujoch.Summing up the percentages listed in Table 2, Jungfraujoch shows a fraction of 57 % thin and opaque cirrus while Zürich and Jülich both occurrence frequencies of 65 and 68% thin and opaque clouds, respectively.These different sums result in the CRFs listed in Table 3 (taking note of the fact that the thresholds for the cirrus cloud categories (Sassen and Cho, 1992) are on a logarithmic scale).positive, i.e. warming, effect.We calculate that subvisibe cirrus contribute about 5%, thin cirrus about 45% and opaque cirrus about 50% of the total cirrus radiative forcing.In order to exert a negative forcing, i.e. a cooling effect, clouds need to be either optically much thicker or in altitude much lower, or both, but we excluded these clouds by demanding that the lower edge of the cloud needs to be colder than -38 • C (cf. Fig. 8).
One important difference between the high ice clouds measured at Jungfraujoch compared to Jülich (with Zürich intermediate) is the possibility to measure thinner clouds above Jungfraujoch, which emphasizes the enhanced suitability of the high alpine measurement station to achieve a high signal-to-noise ratio.Reasons for this are that the objects of interest are closer (and the backscattered signals scales with the square of the distance) and that the polluted boundary layer stays often below the Jungfraujoch station.The Jungfraujoch data show that the lower detection limit in optical depth of a few times 10 −5 is approached in a few cases, but by far the most subvisible cirrus stay clearly above τ = 10 −4 .We argue that this indicates that physical mechanisms prevent clouds to become and stay so thin for appreciable times.After formation, clouds will typically grow quickly and assume higher optical thicknesses.Conversely, evanescent clouds-once having become so thin-will evaporate quickly, not leaving much time for their detection.This leads us to speculate that the Jungfraujoch measurements enable us to explore the very onset of cirrus formation and to possibly learn from the lidar measurements about the relative importance of homogeneous and heterogeneous ice nucleation.
Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-32,2016   Manuscript under review for journal Atmos.Chem.Phys.Published: 2 February 2016 c Author(s) 2016.CC-BY 3.0 License.This causes some of the scattered photons to remain within the field of view of the lidar, where they can be scattered back to the lidar receiver during a subsequent scattering process.These additional backscattered photons cause an underestimation of the particle extinction.The strength of multiple scattering depends mainly on the laser divergence, the telescope field of view and the effective radius of the scattering particles.In order to provide extinction values that are comparable to other lidar systems and cloud conditions, the measured apparent, multiple-scattering affected extinction coefficient α obs p needs to be corrected with the correction factor γ to obtain single-scattering related values α single p Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-32,2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 2 February 2016 c Author(s) 2016.CC-BY 3.0 License.Clouds extending less than 150 m in altitude during daytime conditions are not further taken into account (as noiselimiting measure), whereas nighttime clouds are allowed to be as thin as 3 × 30 m = 90 m.Cloud pixels separated vertically by less than 150 m are merged into one cloud layer.The detected cloud top and cloud base, h top and h base , are stored.VI Multiple scattering correction.The single-scattering extinction coefficients are derived from the apparent, multiplescattering affected extinction coefficients as described in Section 2.1.We use the multiple scattering model of Hogan

( 6 )Figure 3 .
Figure 3. Seasonal cycle of cirrus cloud coverage for the three measurement sites.Dashed lines: annual means.
evidence for both:(a) the advantage of location of the higher Jungfraujoch as evidenced by the ability to measure thinner subvisible clouds (by about a factor 5);

Figure 4 .Figure 5 .
Figure 4. Optical depths of the three measurement sites.(a): Means across whole data set.(b): Seasonal cycle of optical depth.Horizontal line in box: median.Boxes: the upper and lower quartile.Whisker: extremes.Gray horizontal lines: Cirrus categories bySassen and Cho (1992).

Figure 6 .Figure 7 .
Figure 6.Scatter plots of cloud optical depths and cloud top altitudes for the cirrus detected above Jungfraujoch (JFJ), Zürich (ZRH) and Jülich (JUL).The red lines provide an indication of the range of data accessible by the lidar measurements: AODmin = 4 × 10 −5 , AODmax = 2.6, and Altmin = 5.8 km.The lower edge of the accessible altitude is determined from T < -38 • C. Thicker clouds are more likely to extend into lower, warmer levels and therefore are more likely to be excluded from the analysis.

-
Solar constant S -Solar zenith angle Z -The surface albedo α -The cloud optical depth τ -The surface temperature T srf -The cloud top temperature T cld

Table 3 .
Cirrus radiative forcing at the Top of Atmosphere in Wm −2 for Jungfraujoch, Zürich and Jülich, as compared to 50 • N zonally averaged and globally averaged values provided by Chen et al. (2000).Small numbers in CRF-values indicate uncertainty ranges according to the last paragraph of Subsection 2.1.Small numbers in global cloud coverage indicates variability in zonal averages.considering the cirrus occurrence frequency.While the cirrus cloud coverage at 50 • N from the satellite-based climatology ISCCP

Figure 8 .
Figure 8. Cloud Radiative Forcing (CRF) in Wm −2 for the different sites.Magenta/cyan/blue isolines: positive/zero/negative values in Wm −2 from the CRF model (Corti and Peter, 2009).Color coding: Occurrence frequency of cirrus clouds as function of optical depth and cloud top altitude.First row: CRFLW.Second row: CRFSW.Third row: CRFNET.

Table 1 .
Lidar stations that have been used for systematic climatological studies of cirrus clouds in the mid-latitudes

Table 2 .
Properties of the cirrus clouds detected between 2010-2014.