The ice water content (IWC) is derived during most campaigns from the measurements of H2Otot, which is the amount of total water (gas phase + evaporated ice crystals), and H2Ogas, the gas phase water amount. IWC is then calculated by using the following equation: IWC=(H2Otot-H2Ogas)/E, where E is the enhanced volume out of which the ice crystals are sampled in comparison to the sampling volume of the gas phase. For more details see and . H2Otot was measured by the total water instruments FISH and CLH (Table a and b), H2Ogas with the gas phase water instruments FLASH, OJSTER, HWV, JLH, SEAL and SHARC (Table c–h). For details about the instruments see the respective references in the table.

Since no total water measurements were available, IWC for COALESC, ML-CIRRUS and ACRIDICON-CHUVA is derived by integrating the ice crystal size distributions from NIXE-CAPS (see next section) using the mass-dimension (m–D) relation (see , modified m–D relation of ): m=a×Db with a=0.001902,b=1.802forD>240µma=0.058000,b=2.700forD=10–240µmice crystals are spheresforD<10µm. In a recent study, developed new, observation-based m–D relations. These new relations confirm and show that for the cirrus cloud temperature range T<-38 ∘C the m–D relations have nearly no dependence on temperature or cirrus type, thus demonstrating the robustness of the connection between cirrus ice crystal size and mass.

The relative humidity with respect to ice (RHice; shown in Part 2 of the study) is also derived from H2Ogas using RHice=100×H2OgasH2Osat, ice, where H2Osat, ice is a function of temperature . The accuracy of the water vapor instruments was debated during the last decade and thus a couple of studies were carried out to investigate their performance under laboratory and field conditions. The good agreement of the instruments deployed in the field campaigns investigated here is stated by the studies of , , and . These studies provide further details of this group of instruments. Together with the respective uncertainties of the temperature measurements, the accuracy of the RHice observations here is between 10 and 20 %.

Ice crystal number (Nice) size distributions were measured during the different campaigns by the cloud spectrometers FSSP-100/300Dice=3–30µmCAPSDice=3–930µmNIXE-CAPSDice=3–930µm2D-SDice=15–1280µm. Dice denotes the instruments cloud particle size ranges used for this study, although the respective ranges might be larger (see Table i–m and respective references). The upper size limits of the instruments differ. However, the largest contribution to the total ice crystal concentration comes from sizes between 3 and about 30 µm; the concentrations of larger ice particles is in general about three orders of magnitude lower. Hence, the total ice number is covered by most instruments. The 2D-S with the smallest size at 15 µm does not record the smallest ice crystals, whose contribution to the total ice crystal number is highly variable and depends on cloud age: in the early stage of a cirrus most of the ice crystals are small, but grow to larger sizes when the cirrus develops.

The quality of cloud spectrometer measurements is – as with humidity – under ongoing discussion. In particular, a boost of smaller ice crystals appearing through shattering of large crystals at the instrument tips was discovered as a source of error in many older observations see e.g.and references therein.

Out of the instruments used here, FSSP-100/300 and CAS during MidCix 2004 (part of the CAPS=CAS+CIP

Two instruments are integrated in the CAPS (Cloud and Aerosol Particle Spectrometer): (1) CAS (Cloud and Aerosol Spectrometer: 0.6–50 µm) and (2) CIP (Cloud Imaging Probe, CIP:15-930 µm), .

A contamination of the measurements from FSSP-100/300 by shattered ice crystals was discussed by , who stated that a significant effect from shattering is not expected at low temperatures where the ice crystals are smaller, but only at temperatures where the occurrence of larger ice crystals increases. From the analysis of the full data set of this study together with the cirrus simulations we can now conclude that during the campaigns where the FSSP-100/300 were deployed, the ice crystals were not as large as during MidCix or MACPEX, and the ice crystal numbers do not exceed the possible atmospheric range as during those WB-57 campaigns where ice crystal shattering became obvious (e.g., CRYSTAL FACE). Hence, though shattering cannot be completely ruled out, the ice crystal numbers appear to be dominated by natural ice production processes. Thus, we believe that the ice crystal number data set from the Geophysica/Learjet – campaigns is suitable for the study presented here.

The 2D-S (MACPEX 2011) is a new generation cloud instrument, which is equipped with tips and software to minimize shattering effects. However, it starts to record ice crystals at larger sizes than the other instruments (15 µm in comparison to 3 µm, see above; note that technically it starts recording at 5 µm, but it is recommended not to use the size bin between 5 and 15 µm) and thus the total ice crystal number is lower. Nevertheless, as for the MidCix CAPS data, taking this offset into account, the MACPEX data provide a valuable contribution when looking at relative changes of ice crystal concentrations.

The CIP (part of CAPS used for TC-4) as well as NIXE-CAPS (ML-CIRRUS 2014) are also new generation cloud instruments so there are no restrictions for these data sets. NIXE-CAPS is a further development of the CAPS instrument; the new features and data evaluation procedure are described in and . One improvement realized for NIXE-CAPS is a modification of the particle inlet of the CAS probe – which is part of NIXE-CAPS – to minimize shattering. The wall of the inlet entrance is now “knife edged”, greatly reducing the area susceptible for ice crystal shattering. Comparisons to particle size and concentration measurements using other instruments such as SMPS, APS, VIPS, SID-3, 2D-S, CPD, made at the cloud chamber AIDA and on aircraft yield good agreement between the probes .

The total IWC database from all campaigns shown in Table  and Fig.  is presented in Fig.  as a function of temperature (panel a: mixing ratio, panel b: concentration). Altogether, about 94 h of flight time was spent in cirrus clouds, thus considerably extending the IWC climatologies of (27 h) and (38 h). Nevertheless, the median and core IWC – functions derived by and confirmed by are also valid for the new, extended IWC climatology.

In Fig.  the IWC measured during the individual field campaigns are shown. The color code is the ice crystal number Nice. The time of the simultaneous measurements of IWC and Nice is indicated in the panels and sums up to 85.7 h for all campaigns, a little less than the 94 h quoted earlier because of the reduced overall sampling time of Nice. In Fig. c, the part of the IWC climatology of where IWC and Nice ice particle measurements were performed simultaneously is shown. Simultaneous IWC and Nice measurements from TC-4 and MidCix are part of the climatology of and are here shown in the right column. Recent studies are MACPEX, COALESC, AIRTOSS-ICE, ML-CIRRUS and ACRIDICON-CHUVA. The quite different ranges of IWC and the distribution of Nice within these ranges will be discussed by comparing the measurements with a simulated cirrus climatology (Cirrus Guide, Sects.  and ) in Sect. .

MAID simulates parcels of ice clouds where the ice crystals form directly from the gas phase in the temperature range below about 235 K. The ice nucleation processes implemented in MAID are heterogeneous freezing after and homogeneous freezing after . The heterogeneously freezing ice nuclei (IN) can vary in concentration as well as freezing threshold (RHice, see Fig. ): RHiceMD represents very efficient IN with a low freezing threshold (MD: mineral dust), while RHiceCS (CS: coated soot) are quite inefficient IN having a high freezing threshold . The homogeneously freezing aerosol particles are assumed to be supercooled binary solution particles with a concentration of 300 cm-3 and a mode size of 200 nm. The threshold for homogeneous ice nucleation is higher than that of heterogeneous freezing (see Fig. , middle). Thus, in the case where both IN and supercooled solution particles are present, heterogeneous freezing occurs first.

Once the ice particles have formed at the water activity where the ice nucleates, they grow by diffusional growth in separate size bins (Lagrangian ice particle tracking). When the temperature becomes warmer and the air parcel subsaturates, the ice crystals sublimate and return their water vapor to the air parcel.

Ice crystal sedimentation is treated in MAID following the sedimentation scheme of , which uses ice mass and number-weighted terminal velocities after to simulate sedimentation of ice crystals. The ice flux through the model box is defined by the “sedimentation factor” (sedi-f) and represents the ratio of the flux through the top divided by the flux through the bottom. sedi-f = 0 means there is no flux from above into the model box and all ice particles will fall out through the bottom (cloud top), sedi-f = 1 represents the cloud bottom where the flux from above is equal to the downward flux from the bottom, which simulates a no sedimentation scenario.

The trajectory parcel simulations do not capture dynamical processes such as shear, entrainment, and cloud radiation-dynamics interactions, which might reduce the ice number concentrations below the initial values obtained by nucleation .

In one MAID Cirrus Guide scenario the cirrus temperature space is scanned in 10 K steps between 190 and 230 K (see Fig. , left). For each of the five temperatures, simulations with constant vertical velocities of 0.01, 0.1, 0.5, 1.0, 3.0 ms-1 are performed, which are initialized with a water amount of 90 % RHice (T). This sums up to 25 simulations. Further, as at 190 and 200 K the vertical velocity of 0.001 ms-1 is added to enclose the range of large-scale upward motions at the cold temperatures in the Tropical Transition Layer (TTL). Altogether, one scenario contains 27 simulations. They represent the formation and evolution of cirrus clouds during air parcel ascent, covering the atmospheric range from slow frontal large-scale updrafts up to fast lifting in jet streams or convection. The sublimation phase of the clouds is not considered in the simulations. Also, ice crystals sedimenting into subsaturated air as, e.g., fall streaks are not included in the simulations.

For the complete MAID Cirrus Guide, 36 scenarios (972 model runs) were performed altogether, each varying in the initial and boundary conditions. First of all, pure homogeneous freezing (HOM) or heterogeneous followed by homogeneous freezing (HET + HOM) is allowed to occur. Second, the temperature course of the trajectory can be chosen to be a constant updraft or the updraft can be superimposed with temperature fluctuations. Further, the IN number and freezing threshold and the sedimentation parameter are prescribed for each scenario (see Fig. , right).

For the constant updrafts, two HOM scenarios with varying sedi-f (0.9 and 0.5: moderate and strong ice particle sedimentation) are performed, while for HET + HOM 16 scenarios are realized for each combination of four possible IN numbers (0.001, 0.01, 0.1, 1.0 cm-3: very low to very high), two different freezing thresholds (MD, CS) and two sedimentation parameters (sedi-f =0.9, 0.5). That means that each constant updraft scenario is simulated 18 times under different conditions, summing up to a total of 486 model runs. The scenarios are abbreviated as follows: the 16 HET + HOM scenarios are HET + HOM|sedi-fIN-MD/CS, the two pure homogeneous freezing scenarios: HOM|sedi-f.

Finally, the 18 scenarios are repeated by superimposing temperature fluctuations on the constant vertical velocities. The fluctuations are superimposed with five different frequencies (2×10-4, 8×10-4, 2×10-3, 4×10-3, 8×10-3 Hz) to cover those fluctuations significant for cirrus formation and to reproduce the typical turbulence spectrum found in the atmosphere. The amplitudes are statistically distributed, whereas the maximum amplitudes increases with height and range from 0.3 K for the slowest constant vertical velocity to 4 K for the fastest constant vertical velocity. The frequency spectra and the resulting power spectral density (PSD) are in accordance with and also with in situ measurements of temperature fluctuations during several aircraft campaigns e.g. MACPEX, see.

We would also like to mention that the computer time needed for the complete MAID Cirrus Guide sums up to several weeks where multiple PCs are operated simultaneously. The reason is the exact partitioning of the water vapor between gas and solid phase using the different water vapor partial pressures.

As an example, Fig.  shows the Cirrus Guide scenario HET + HOM|sedi-0.90.01-MD, which we consider as the “middle case” with assumed mean IN conditions of 0.01 cm-3, MD, and a moderate sedimentation of sedi-f = 0.9

Here and in the rest of our study we present the results of our simulations with constant updraft (“nofluct”), since the cirrus evolution can be seen more clearly in these scenarios. In Sect.  it is shown that the general cirrus patterns are well represented by the “nofluct” simulations.

To provide a better understanding and a clearer view of cirrus microphysics in the temperature parameter space, some trajectories of cirrus formation and development are selected from Fig.  and are shown in Fig. . Since the cloud processes are much more dependent on updraft than on temperature, we discuss two trajectories at slower updrafts (green line: 1 cms-1 at 230 K and blue line: 10 cms-1 at 220 K), three at higher (turquoise, purple and red: 0.5, 1, 3 ms-1 at 210 K) and one at a very low vertical velocity representative of the large-scale ascent in the TTL (0.01 cms-1 at 190 K). The trajectory time proceeds from higher to lower temperatures, as marked by the black arrow. Note here that the times of the scenarios greatly differ: to cool the air by 1 K, 170/17/0.6 min are needed by the green/blue/red air parcels (low/middle/high updrafts). It should be kept in mind that the trajectories apply to the ice nucleation zone of cirrus clouds and not to regions that sedimenting ice crystals fall into. We discuss the two groups of scenarios – slow and fast updraft cirrus – mentioned in Sect. .

A representative sampling of all 36 cirrus scenarios in the IWC-T parameter space is shown in Fig. . Each panel includes all 27 trajectories, covering the full cirrus temperature and vertical velocity range, as described in Sect.  (see Fig. ). Eight scenarios are shown here: the top row represents simulations with efficient IN like mineral dust (MD, low freezing threshold) and the bottom row inefficient IN like coated soot (CS, high freezing threshold), while the columns show varying IN concentrations. For all scenarios shown here the sedimentation factor is assumed to be moderate (sedi-f=0.9).

It is obvious from Fig.  that the differences between the IN initial conditions (concentration and freezing threshold) do not greatly influence the general pattern of the IWC-T portrayal of cirrus. In other words, it means that the IWC is a quite stable parameter, particularly in comparison with the variability of Nice and Rice in dependence to the initial atmospheric conditions (see Sect. , Fig. ).

In Fig. , the scenario HET + HOM|sedi-0.90.01-MD (same as in Fig. : red box; see also Figs.  and ) is shown three times. Figure a depicts the IWCs when temperature fluctuations are superimposed on the constant vertical velocities (“fluct” runs). It can be seen that the general IWC distribution is preserved, looking less structured, while the individual model runs are generally shorter in comparison to the “nofluct” cases (= constant updrafts). The IWC oscillations are parallel to the temperature course, reflecting growing and shrinking ice crystals with increasing and decreasing temperature. The shorter lifetimes can be explained by the fact that the temperature fluctuations contain warming phases so that the cirrus clouds simply evaporate. From Fig. b (“nofluct” scenario, color coded by time from blue to red) it can be seen that the parts of the simulations that have disappeared in the “fluct” scenario are cirrus older than about 18 h (reddish colors). Since the general pattern of IWC evolution is more clear in the “nofluct” simulations, we prefer to show those in this study, though the scenarios with temperature fluctuations better reflect atmospheric conditions.

Next, in Fig. c the scenario “nofluct” is shown again, but with enhanced sedimentation (sedi-f=0.5). Here, a decrease of IWCs can be seen, in particular at warmer temperatures, where the ice crystals are large and preferentially fall out. However, from comparison of the Cirrus Guide scenarios with the meteorological situations found in the observations (see Sect. ), we can conclude that the observations are better represented by the simulations with moderate sedimentation (sedi-f=0.9).

From the previous sections we conclude that two types are not only found in selected trajectories (Sect. ), but are a general feature of mid-latitude in situ origin cirrus (205–235 K, Sect. ).

We can conclude that cirrus microphysics is visible in the IWC-T parameter space up to a certain degree, i.e., depicting measurements of IWC vs. temperature gives an impression of the nature of observed cirrus without the need of more detailed microphysical measurements. This is summarized in the sketch shown in Fig. , where we plotted some of the MAID scenarios together with the median, core and maximum IWC-lines to guide the eye.

IWCs below the median line mostly stem from slow updraft cirrus in low- or high-pressure systems. They consist of a few, but large ice crystals which are – due to the increasing amount of available water vapor – larger the warmer the cirrus is. The first part of these cirrus is determined by HET freezing. However, if the cloud life time is long enough, HOM freezing will start as well. Due to the slow updrafts, the number of ice crystals produced by HOM freezing is on the same order of magnitude as the HET nucleated ice crystals and thus also have low IWCs. We emphasize here that in slow updraft regions the difference in microphysics between homogeneous and heterogeneous freezing is small. In this cirrus type, sedimentation plays an important role, controlling the microphysical properties and might produce fallstreaks, which causes some vertical redistribution of water.

High IWCs above the median line are mostly produced by fast updraft cirrus caused by atmospheric waves. The cirrus is only briefly at low IWC during the HET phase since the HOM freezing quickly pushes the IWC above the median line. With their large number of small, non-sedimenting ice crystals, these cirrus are microphysically very different from the slow updraft cirrus.

An impression of the radiative properties in terms of extinction of the two cirrus types is given in Fig. d, where the ice crystal number Nice is shown vs. the corresponding mean mass radius Rice for all temperatures. The IWCs are plotted as black isolines. Figure b shows the same, but with color coded IWC, in Fig. c the color code is vertical velocity. The extinction of all simulated cirrus is calculated from the empirical relationship between IWC and extinction provided by . Those cirrus with a low number of large crystals (slow updraft, low IWC, see Fig. c and b) have a small extinction, i.e., they are optically thin, while the cirrus with the many small ice crystals (fast updraft, large IWC, see Fig. c and b) have larger extinctions and are optically thicker.

The majority of cirrus clouds in the atmosphere are of the first “slow updraft” type, i.e., they appear on a larger scale and have a longer lifetime, while the second “fast updraft” type occurs less frequently since the fast updrafts are limited in space and time (note here that “fast updraft” does not include convection in the simulations; “convective” or “anvil” cirrus will be addressed later in Sects.  and ).

In this section, we compare observed and simulated cirrus clouds to demonstrate that the simulated Cirrus Guide shown in Fig.  is confirmed by measurements and thus is indeed appropriate for an impression of the cirrus microphysics.

In the last paragraph of Sect.  it is mentioned that high IWCs above the median IWC together with large ice crystals (Fig. ) are not represented in the Cirrus Guide. Nevertheless, this type of cirrus is found in the observations, especially during MidCix, MACPEX and ACRIDICON-CHUVA, as well as in the other campaigns shown in Fig.  (except AIRTOSS-ICE). Here, a larger number of very high IWC are detected than was expected based on the MAID simulations. From further analysis, relating meteorological situations to cirrus types, it can be concluded that these high IWC cirrus stem from lower altitudes and are not produced by in situ ice nucleation directly from the gas phase like in our simulations for detail see. Instead, they are formed by heterogeneous freezing of liquid droplets at temperatures >235 K or possibly as low as 235 K by homogeneous drop freezing. The ice crystals were then lifted to the temperature range <235 K and are therefore regarded as cirrus clouds. This can happen in regions with mesoscale convective activity (observed during MidCix and MACPEX), but also in warm conveyor belts (observed during ML-CIRRUS). In tropical convective systems, which are observed in this study during ACRIDICON-CUVA, this cirrus type is known as “anvil cirrus” or “anvil outflow”.

Note that in addition to a high IWC, other indicators for cirrus originating from liquid drops are high Nice together with the appearance of larger ice crystals for more detail see. The high ice crystal concentrations may stem from an originally larger frozen drop number, or from an additional freezing event that may happen on top of the preexisting ice in case the updraft is fast enough. The nucleation mechanism will probably be homogeneous freezing, since the IN are already consumed in the previous liquid cloud. The high supersaturation that is needed can be reached in fast updrafts, when the reduction of RHice by uptake of water on the ice surfaces is less than the RHice enhancement caused by the decreasing temperature.

Altogether, these thick liquid origin cirrus have microphysical, and thus optical, properties quite different from the in situ cirrus formed directly via the gas phase. Further, they appear together with mixed-phase and often liquid clouds below. This is shown in detail for liquid origin clouds stemming from warm conveyor belts in comparison to in situ cirrus observed during ML-CIRRUS by .

It is a surprising result from this study that these “liquid origin cirrus” seem to appear in the complete cirrus temperature range with a non-negligible frequency. As a consequence, for a complete understanding we recommend devoting further research to assessing their properties and including them in model studies.

Cirrus cloud classifications already exist and are used in the literature e.g. . Here we intend to compare our findings of the two in situ origin types and the liquid origin cirrus with the present classifications.

“In situ”, “synoptic” and also “lee wave”, “gravity wave” or “orographic” cirrus are used as definitions for cirrus cloud forming directly from the gas phase. Since “in situ” is the most common name – and “lee/gravity wave” and “orographic” cirrus are already very specific – we followed this approach in our classification. On the other hand, the term “convective” or “anvil” cirrus is used to classify cirrus clouds representing glaciated, but originally liquid clouds lifted to the cirrus temperature regime in the prevailing updraft. Here, we name this type “liquid origin cirrus”, since we found that the lifting of clouds to temperatures T<-38 ∘C happens not only in convective systems connected to anvils, but also in warm conveyor belts, in mesoscale convective systems and even in gravity wave clouds.

Thus, compared to the grouping based on specific meteorological situations, the cirrus classification scheme we present here is based on (i) the formation mechanism (directly ice or frozen liquid droplets), which is tied to the temperature threshold of -38 ∘C below which liquid water does not exist, and, (ii) the vertical velocity, which determines the thickness of the cirrus.

A thorough study on the impact of large-scale dynamics on the microphysical properties of midlatitude cirrus has been performed by . The study is based on cirrus cloud observations during the field campaign SPARTICUS in 2010 see also in the vicinity of the ARM SGP site (Atmospheric Radiation Measurement, Southern Great Plains) and analysis of atmospheric states using a combination of ECMWF ERA-Interim reanalysis data with continuous observations from a millimeter-wavelength cloud radar.

Four cirrus types are defined by : (a) ridge-crest, (b) frontal, (c) subtropical jet stream and (d) anvil. Striking differences in the cirrus microphysics (IWC and Nice) for different large-scale environments were found in this study. The authors suggest that vertical velocities are a poor predictor for explaining the microphysical variability in cirrus.

Before comparing the findings of with this study, we like to note a difference in the observed IWC ranges: here, IWCs between about 0.001 and ≈ 300 mg m-3 are detected in the SPARTICUS temperature range > 210 K (see Fig. b), while in the observed IWC range spans from 1 to ≈ 400 mg m-3 (their Fig. 6). From Fig. b it becomes obvious that the in situ cirrus in slow updrafts with low IWCs (in general < 1 mg m-3, i.e., below the median IWC line) are barely included in the SPARTICUS observations. Thus, the definitions of “low” and “high” IWCs differ between the two studies: the peaks of the frequency distributions of ridge-crest cirrus are discussed as low IWC (≈ 8 mg m-3) in connection with high Nice (≈ 0.2 cm-3) . However, in the parameter space of the Cirrus Guide, 8 mg m-3 is well into the high IWC range, so the classification would be high IWC and high Nice. Frontal and subtropical jet stream cirrus, which are described by to have middle IWC and Nice are also in the range of high IWC and Nice in the context here. Anvil cirrus have high IWC and Nice in both studies.

Taking this into account, we can relate (a) ridge-crest to fast updraft in situ origin cirrus – a type of cirrus that is quite rarely found in our observations and (b) frontal cirrus most probably to liquid origin cirrus in slower frontal updrafts (like WCBs in our study). Subtropical jet stream cirrus (c) also seem to be of liquid origin – because of the higher IWCs – with a possible subsequent weak in situ homogeneous freezing event, and, (d) anvil cirrus are classified as liquid origin in fast updrafts. With respect to a vertical velocity classification, we think that the lack of significant differences detected by between the cirrus types is a consequence of observing cirrus classes that are all situated in the fast updraft range above the IWC median. The slow updraft in situ cirrus classified in this study exhibit IWCs below the median IWC line and are not included in .

Altogether, the four classifications provided by for the high IWC range agree very well with the cirrus classes of this study. They are more specific than the two types provided here (fast updraft in situ or liquid origin cirrus), which we have not further specified due to a lack of observations of fast updraft in situ cirrus. From the few observations we have, we can confirm that fast updraft in situ cirrus (ridge crest in ) have a lower IWC but higher Nice than liquid origin cirrus (frontal, subtropical jet stream and anvil cirrus in ).