As described in Part 1 (M2025), the CALIPSO retrieval uses IIR passive observations at 10.6 and 12.05 µm with description of the atmospheric column informed by perfectly collocated CALIOP observations. Results presented in this study are based on CALIPSO data during 2008, 2010, 2012 and 2013. Microphysical properties are retrieved for atmospheric columns that contain only one semi-transparent ice cloud layer that does not fully attenuate the CALIOP laser beam to ensure that cloud base is detected. We also require the cloud layer to be classified as ice with high confidence by the CALIOP ice/water phase algorithm (Avery et al., 2020). The CALIOP in-cloud 532 nm attenuated backscatter profiles with temperatures from MERRA-2 re-analyses (Gelaro et al., 2017) allows one to estimate the cloud layer radiative temperature, Tr (Garnier et al., 2021), and cirrus clouds are defined as those ice clouds having a Tr ≤ 235 K. The upper range of optical depth for semi-transparent single-layer cirrus clouds is approximately 3. The lower range is driven by retrieval uncertainties. In earlier work (M2018), global retrievals over both oceans and land have been in clouds having a CALIOP 532 nm layer integrated attenuated backscatter (IAB) larger than 0.01 sr⁻¹, generally corresponding to clouds with visible optical depth τ > 0.3. These clouds will be hereafter called “thick” cirrus clouds (see Table 1). However, over oceans, uncertainties are lower due to the absence of highly variable land surface emissivities, allowing all cirrus clouds with valid retrieved effective emissivities at both 10.6 and 12.0 µm to be considered, yielding τ > ∼ 0.01 (M2025).

The geographical distribution of the cirrus cloud sampling is shown in Fig. A1 in Appendix A. In M2018, the sampled cirrus clouds were in situ cirrus clouds having base temperature, Tbase, colder than 235 K. In this work, we take advantage of the improved ice/water phase assignment in the Version 4 CALIOP products to also include cirrus clouds with Tbase warmer than 235 K (and Tr colder than 235 K), hereafter called warm base cirrus clouds or WBC clouds. Even though these WBC clouds are identified as high confidence ice cloud layers by CALIOP, this assessment does not rule out the possibility of liquid droplets in the lower part of the layer. This classification method is an attempt to qualitatively contrast the properties of in situ and LOC clouds, using WBC as a proxy for LOC clouds. A similar approach was used in Gasparini et al. (2018). This approximation may underestimate LOC clouds (overestimating in situ cirrus) since cloud condensate from below the 235 K isotherm may be advected across this isotherm upwind of the CALIOP nadir view when there is no cloud at nadir below this isotherm. In this case the cloud would be mistakenly classified as in situ cirrus. On the other hand, the modeling study by Wernli et al. (2016) estimates that approximately 50 % of in situ cirrus clouds occur on top of LOC, indicating a strong dynamical linkage. Relative to an air parcel back trajectory analysis as used in Wernli et al. (2016) and other LOC studies cited below, our approach should underestimate in situ cirrus if there is no clear layer separating in situ from WBC clouds. This classification scheme is evaluated in Appendix A in Figs. A2 and A3. Figure A2 shows the dependence of the in situ fraction on temperature, where this fraction is ∼ 0.5 (indicating a transition from in situ to WBC) at approximately 227 K over oceans when all clouds are considered. In the LOC studies by Dekoutsidis et al. (2023) and Luebke et al. (2016), which are both based on the same field campaign, this transition occurs around 221 K and between 218–222 K, respectively. This suggests that the WBC approximation overestimates the in situ fraction somewhat (shifting the transition temperature by ∼ 6 K relative to these measurement-based studies), but that WBC may still serve as a qualitative proxy for LOC. The blue curves in Fig. A3 indicate that while WBC clouds have relatively higher IWCs, the main difference between them and in situ cirrus clouds is the greater ice water path (IWP) and τ associated with WBC clouds.

As in M2018, global retrievals over both oceans and land are in thick clouds having τ > ∼ 0.3, and we are now sampling both in situ cirrus and WBC clouds. This raises the question: “Is this subcategory of cirrus cloud adequate for testing cirrus clouds in climate models?” It was argued in Sect. 6.3 of M2018 that cirrus clouds in this τ range should dominate the overall cirrus cloud net radiative forcing, but more can be said about this. Table 4 in Hong et al. (2016) gives global mean top-of-atmosphere net ice cloud radiative effect (CRE) values for the following five τ categories: τ < 0.03; 0.03 < τ < 0.3; 0.3 < τ < 3.0; 3 < τ < 20; τ > 20. The corresponding net CRE values in W m⁻² are: 0.05, 0.96, 4.98, -0.49 and -0.43. These CRE values are weighted by ice cloud frequency of occurrence. Relative contributions of net CRE can be calculated by taking absolute values, in which case ice clouds having 0.3 ≤ τ ≤ 3.0 contribute 72 % of the total “absolute” CRE. Moreover, the global mean net CRE for all ice cloud categories in their Table 4 was 5.07 W m⁻² whereas the global mean net CRE for 0.3 ≤ τ ≤ 3.0 is 4.98 W m⁻², further suggesting that this thick cloud category is representative for all ice clouds. However, Hong et al. (2016) addressed all ice clouds (i.e., Tr < 273 K) whereas our study addresses only cirrus clouds (Tr ≤ 235 K).

To further investigate this question for our cirrus sampling, we examine the distributions of retrieved τ (M2025) over oceans for the high- and midlatitudes and the tropics. The tropics are defined here as ±30°, while the mid- and high latitudes are 30–60 and 60–90° in each hemisphere, respectively. The τ distributions are built using bins equal to 0.1 in log⁡(τ) space. The number of IIR 1 km² cloudy pixels, p and the product p × τ, are plotted against τ in Fig. 1. This is done over oceans for all sampled cirrus clouds and for thick clouds that have lower retrieval uncertainties. The p × τ product serves as a proxy for the radiative impact of cirrus clouds since τ and frequency of occurrence are the principal factors determining CREs. The right-hand panels in Fig. 1 show the cumulative distributions over τ (panel b) and for this proxy or radiative weighting (panel d). The cumulative distributions are normalized with respect to all cirrus clouds sampled. The cumulative distributions for thick clouds indicate that cirrus clouds having τ > ∼ 0.3 represent 31 % to 48 % (depending on latitude band) of the cirrus clouds sampled. However, they represent 78 % to 87 % of the radiative weight for all cirrus clouds sampled (i.e., cirrus clouds having τ < 0.3 represent 13 % to 22 % of the radiative weight). This compares favorably with results from Hong et al. (2016) described above, where the ice cloud CRE fraction for τ < 0.3 relative to τ < 3 is 0.17, or 17 %. The occurrence frequency of cirrus having τ > 3 is negligible here because only clouds that are semi-transparent to CALIOP are sampled.

Global maps for December–January–February (DJF) and June–July–August (JJA) during 2008, 2010, 2012 and 2013 are shown for median Ni, De, IWC and cirrus cloud Tr for thick cirrus clouds (i.e., ∼ 0.3 < τ < ∼ 3) in Figs. 2, 3, 4 and 5, respectively. Global maps for the two other seasons are given in the Supplement as Figs. S1 to S4. From Sect. 2.2, this sampling criteria appears to provide cloud property values of cirrus clouds whose radiative properties are representative of all cirrus clouds.

Relatively high Ni are found at high latitudes over land, as well as over mountainous regions in the midlatitudes, especially during winter. These observations appear consistent with hom since hom tends to produce higher Ni in stronger updrafts (e.g., OGWs induced by mountains) and during winter (Joos et al., 2014). The seasonal dependence of Ni in Fig. 2 was also found by Sourdeval et al. (2018, their Fig. 8) at mid-to-high latitudes for winter vs. summer. The regions with large Ni in the tropics in Sourdeval et al. (2018) are absent in Fig. 2 possibly because thick cirrus here have τ < ∼ 3 and thus deep convective cores (where hom is often active) are not sampled.

For a given season, regions having relatively high Ni in Fig. 2 have relatively small De in Fig. 3. An exception is in the tropical western Pacific during winter where De is relatively small and Ni is relatively low. The deepest convection occurs in this region during winter (Danielsen, 1993), producing the coldest cloud top temperatures where ice crystal growth rates are bound to be minimal, probably resulting in smaller De. This phenomenon is more evident in Fig. 4, showing retrieved median IWCs, where IWC is lowest during winter in the tropical western Pacific due to these minimal temperatures. The correspondence between Tr (Fig. 5) and IWC (Fig. 4) is clearly seen as expected from the Clausius–Clapeyron equation.

Figures S5 and S6 show global maps of median Ni and De for in situ cirrus clouds only, where Ni is higher and De considerably smaller relative to Figs. 2 and 3 (where WBC are included). To interpret these figures, microphysical properties of in situ cirrus and WBC clouds are investigated in Fig. A3. Although corresponding to a specific season and latitude zone, these Fig. A3 results are globally representative. Median De increases as Tr increases, with no marked difference between in situ and WBC cirrus clouds at a given Tr. This suggests that the De differences seen between Fig. 3 (for thick WBC and in situ cirrus) and the thick in situ cirrus in Fig. S6 (residing at a lower Tr on average) are due to temperature differences. Temperature could also be a factor in explaining the lower Ni in Fig. 2 (for thick WBC and in situ cirrus) relative to the thick in situ clouds in Fig. S5, since the latter tend to form at lower temperatures (Fig. A2) and hom depends strongly on temperature (e.g., Lamb and Verlinde, 2011). Figure A3 shows that both the geometric thickness and the equivalent thickness seen by IIR are on average greater for WBC clouds than for in situ cirrus clouds. This is the main reason that IWP and τ tend to be substantially greater for WBC clouds. Finally, IWC and Ni are both higher in WBC, in agreement with Luebke et al. (2016), only when clouds with τ < ∼ 0.3 are included (blue curves). Figure A3 highlights the importance of cloud sampling in optically thick vs. thin cirrus when interpreting cirrus cloud field campaign measurements of Ni and IWC, and ultimately in terms of the contribution of het and hom to cloud formation.

The main purpose of this section is to justify the use of the extinction coefficient αext as a means of separating cirrus clouds formed primarily through het from those formed primarily through hom. As discussed in M2025, when all IIR samples over oceans were used (∼ 0.01 < τ < ∼ 3) retrieved median Ni, Rv (mean volume radius), and IWC were in reasonable agreement with corresponding values in the cirrus cloud property climatology of Krämer et al. (2020) which is based on many field campaigns throughout the world. However, when the CALIPSO retrievals were restricted to thick clouds, the agreement degraded for Ni and IWC. Again, this raised the question of whether cirrus clouds having τ smaller and larger than 0.3 tend to be characterized by different physical properties and processes.

Figures 6 provides a means of investigating this question, evaluating Ni, IWC, De and the sampling densities for optically thin (∼ 0.01 < τ < ∼ 0.3) and thick (∼ 0.3 < τ < ∼ 3) cirrus clouds. Only retrievals over ocean are considered since variable land emissivities preclude retrievals over land for the thin clouds. These quantities are plotted in Tr vs. Tr–Ttop space where Ttop is temperature at cloud top and Tr is most of the time near mid-cloud (M2025). Thus, Tr–Ttop is an index for cloud geometrical thickness. Figure 6a shows that for thick cirrus, the highest Ni (resulting from hom due to its magnitude) is found in relatively geometrically thin clouds (consistent with Fig. 10 in M2018), while Fig. 6d for optically thin cirrus is almost featureless with Ni < 100 L⁻¹ in general. Note that the retrievals characterize cloud layers (i.e., no vertical profiling within a cloud is possible).

The cirrus cloud remote sensing results from Dekoutsidis et al. (2023) and the cirrus cloud modeling results of Spichtinger and Gierens (2009a, b) both show that hom is primarily active only near cloud top, with the later showing relatively low Ni in the mid- to lower cloud where sedimenting ice crystals quench hom by lowering the RH_i. Based on this, geometrically thicker cirrus should have vertically averaged Ni much lower than Ni in the upper cloud and in geometrically thinner cirrus (assuming comparable thicknesses for the ice generating regions). These findings are consistent with our results for optically thick cirrus. IWC for these cirrus (Fig. 6b) is higher than for optically thin cirrus (Fig. 6e) clouds which exhibit much lower values, typically smaller than 4 mg m⁻³. Jensen et al. (2024) show that Ni and IWC are higher in younger tropical cirrus clouds due to mesoscale temperature fluctuations from gravity waves that act to decrease them over time. This may help explain these results if hom cirrus are associated with young cirrus. Differences in De between the two τ categories are considerably less, as shown in Fig. 6c and f. For a given Tr, De tends to be quasi-constant, although decreasing for the geometrically thinnest clouds in both τ categories. To summarize, these results show that hom is active in optically thick cirrus (due to the high Ni in geometrically thin cirrus) but not in optically thin cirrus. Moreover, they show that hom mostly affects Ni and IWC, with perhaps a secondary impact on decreasing De (referring to Figs. 2, 3, S5 and S6 as well as Fig. 6). This suggests that the ratio IWC / De and therefore the extinction coefficient (αext = 3 IWC / (ρiDe), where ρi = bulk density of ice = 0.917 g cm⁻³) may be sensitive to hom.

As described in M2018 and M2025, one of the cloud properties retrieved by this method is the cloud extinction coefficient for visible light, αext. Here we use this retrieved αext as a means for measuring the activity of hom since it contains the ratio IWC / De, which was shown in Sect. 3.1 to be sensitive to hom activity. For a given Tr, relatively high values of αext should correspond to a substantial impact from hom on the cloud microphysics.

To test this idea, Ni is plotted in Tr vs. log⁡(αext) space in Fig. 7 for retrievals over ocean using all IIR samples, with τ ranging from ∼ 0.01 to ∼ 3, for the tropics, midlatitudes and high latitudes during winter (for both hemispheres) for in situ cirrus and WBC clouds. Light gray regions in these plots are where the number of samples is smaller than 10. Characteristic in all plots is a broad region on the left-hand side (relatively low αext) where Ni < 30 L⁻¹. Although hom can produce such low concentrations at warmer temperatures and low updrafts (Krämer et al., 2016), hom tends to produce much higher Ni (Barahona and Nenes, 2009). To the right of this region is a gradient of increasing Ni, culminating in values of Ni > 1000 L⁻¹. This gradient region is probably produced by varying degrees of hom activity, although het may also contribute to this gradient under conditions of relatively high INP concentration. It is evident that hom-affected cirrus clouds are common in both in situ cirrus and WBC. For WBC, hom appears quite active even at the highest temperatures. This can be understood if the cloud is glaciated by het freezing at T > 235 K and ice supersaturation si is already near the hom threshold. Then some cloud parcels are transported upwards across the 235 K isotherm, increasing si, with haze solution droplets freezing as they cross this isotherm. The main difference between these cloud types is in the tropics where in situ cirrus often appear to the right-hand side of the region predicted for pure hom (i.e., the triangles or squares), which will be discussed below. This may be due to deep convection overshooting the temperature level predicted for hom to activate, depositing ice mass at lower temperatures where in situ cirrus may subsequently form. Results like Fig. 7 are shown in Fig. S7 for land where this “overshooting effect” is more evident for in situ cirrus outside the tropics, perhaps due to stronger orography-induced updrafts over land. Also to be noted in Fig. S7 is the strong contribution from polar stratospheric clouds over Antarctica at very cold temperatures. To the right of this Ni gradient in Fig. 7, the boundary between peak Ni values and the “no data” region has a curious diagonal orientation (as well as the gradient) that may be explained from first principles.

To explore this further, two formulations of the Clausius–Clapeyron equation and the supersaturation threshold equation for hom, first derived in Lamb and Verlinde (2011), were used: 1esi=es0exp⁡LsRg1T0-1T, where esi is water vapor pressure at ice saturation, Ls is latent heat of sublimation, Rg is the gas constant, T is temperature and es0 and T0 refer to water vapor pressure and temperature at the ice-point of water (6.11 hPa and 273.15 K, respectively). Reformulating Eq. (1) as described in Sect. 3.3.1 of Lamb and Verlinde (2011) gives an expression useful in calculating the supersaturation required for homogeneous ice nucleation: 2esesi=exp⁡LfRg1T-1T0, where es = water vapor pressure at water saturation and Lf = latent heat of fusion. Finally, the threshold supersaturation ratio where hom occurs, Sif, was developed in Lamb and Verlinde (2011) from results in Koop et al. (2000): 3Sif=1.0+0.305esesi. Noting that the water vapor pressure at Sif, ehom, is simply ehom = Sif esi, water vapor densities are obtained from the gas law (ρhom = ehom/(RvT) and ρsi = esi/(RvT), Rv = gas constant for water vapor) to predict the maximum IWC resulting from hom: 4IWChom=ρhom-ρsi. But to relate IWChom(T) to retrieved IWC in Tr vs. αext space, a corresponding αext is needed which requires De. For the tropics, the PSD mean ice particle maximum dimension Dmean was related to T using the CEPEX PSD scheme described in Mitchell et al. (1999), where only the large ice particle mode was used. For outside the tropics, a Dmean–T relationship based on the SPARTICUS field campaign (Mishra et al., 2014) was used, which is log⁡(Dmean) = 0.0307 T + 3.169, with Dmean in micrometers and T in °C. Using the ice particle mass- and projected-area-dimensional relationships from Erfani and Mitchell (2016) and an exponential gamma PSD, De was calculated from Dmean as described in that study. This gives De under “normal” conditions. To estimate De under maximum hom conditions (corresponding to Eq. 4), we use panels (h), (i) and (l) of Fig. 10 over land for high latitudes and for midlatitudes during winter which show that when hom is most active, De decreases from “normal values” of ∼ 55 to ∼ 35.5 µm for T between 219 and 235 K, or by a factor of 0.646. Thus, we apply this factor to De estimated from Dmean to estimate De for maximum hom conditions. Finally, we calculate Ni by converting this De back to Dmean for an exponential PSD using Eq. (7) in Mitchell et al. (2020) and then calculating the PSD slope parameter λ from Dmean (λ = 1/Dmean). For the region where hom is most active, Ni is Nmax⁡ calculated as 5Nmax⁡=Γν+1IWChomλβαΓβ+ν+1, where ν is a parameter in the gamma PSD N(D)=N0Dνexp⁡-λD and α and β define the ice particle mass relationship m = αDβ, where ν = 0 here. The values of α and β were obtained from Erfani and Mitchell (2016) in the same way that De was calculated; Nmax⁡ predicted from Eqs. (5) and (1) of Schneider et al. (2021) is overplotted in Fig. 7 as shown by the triangles and squares, respectively, with the color inside the triangles/squares indicating the concentration range (based on the color bar). Schneider et al. (2021) provide an empirical expression for Sif based on experiments in the AIDA cloud chamber.

Figure 8a to c show the retrieved median IWC plotted in Tr vs. log⁡(αext) space for all samples over oceans (both in situ and WBC), with triangles predicted from Eq. (4) overlaying these retrieved IWCs. The different Dmean–T relationships for the tropics and for outside the tropics yield slightly different αext(T) associated to IWChom(T) (since αext = 3 IWC / (ρiDe)). The color within the triangle indicates the approximate IWC_hom value via the color bar legend. It is seen that the color within the triangles matches the background color in almost all cases, showing good agreement between basic theory and the retrieved IWC where hom is expected to be most active. IWC_hom was also calculated from Eq. (1) in Schneider et al. (2021), shown by the squares in Fig. 8. There is close agreement between the theoretical (i.e., Eq. 3) and experimental methods, where both methods address the homogeneous freezing of solution haze droplets based on the activity of water in a solution droplet (Koop et al., 2000) and do not address the homogeneous freezing of activated cloud droplets. In panels (d) to (f), retrieved median De is plotted in Tr vs. log⁡(αext) space for all samples over oceans, with theoretical and experimental hom De indicated by the triangles and squares, respectively, with their color based on the same color legend as used for the retrievals. The predicted and retrieved De values are in general agreement for this hom-active region. Results like Fig. 8 but over land are shown in Fig. S8.

There is a close correspondence between the predicted triangle (or square) color and adjacent retrieval colors in Figs. 8 and S8, indicating reasonably good agreement between theory and observations. The predicted triangles and squares in Figs. 7 and S7 show the maximum Ni expected for a given Tr (thus indicating the region of maximum hom activity). The predicted colors (triangles and squares) tend to indicate higher Ni than the adjacent background retrieved Ni colors, which is consistent with expectations that hom events are diluted through mixing and entrainment (Jensen et al., 2012). The diagonal orientation of this hom region is a consequence of the Clausius–Clapeyron equation that predicts lower saturation water vapor pressure and density as T decreases. In the tropics, the predicted colors indicate considerably higher Ni values than the adjacent retrieval values. This may indicate that hom is less active in the tropics for our cloud sampling.

The black contours in Figs. 7, S7, 8 and S8 show the pixel count densities in Tr vs. log⁡(αext) space. Outside the tropics, the triangles/squares coincide with relatively low pixel counts, indicating the frequency of occurrence for hom events is quite low, consistent with theoretical expectations (e.g., Krämer et al., 2016).

It has now been shown that αext seems to be a suitable metric to identify varying degrees of hom activity, but how can this be used to quantify the contribution of hom-affected cirrus clouds in terms of frequency of occurrence and radiative impact? To address this question, a method is presented that estimates the fraction of hom-affected cirrus clouds such that these clouds are strongly altered microphysically relative to cirrus associated with relatively low Ni, such as cirrus formed through het and aged, dissipating/sublimating cirrus clouds (formed either through hom or het). Henceforth the former will be referred to as hom cirrus and the latter as het cirrus. These dissipating cirrus clouds are microphysically described in Fig. 6 of Krämer et al. (2020) where RH_i < 100 %. These two categories will now be described and finally quantified.

The median Ni and median De corresponding to various temperature intervals in Figs. 7 and 8 were plotted against αext in Fig. 9 (panels a–f and g–l, respectively) for both winter and summer. These retrievals are over oceans for only in situ cirrus clouds, and a similar plot, Fig. S9, is given for WBC clouds. When median De was set equal to the sensitivity limit (M2025), it was not plotted, which rarely occurred. Colors indicate the 4 K temperature interval corresponding to the retrievals. Outside the tropics, it is seen that two populations exist, characterized by different dlog⁡(Ni)/dlog⁡(αext) slopes in the case of Ni or trends in the case of De, mostly for the higher temperatures. If one were to describe the “warm-colored” (T > 215 K) Ni data with two linear fits, a change in slope occurs around αext ≈ 0.3 km⁻¹. The increased slope for αext > 0.3 km⁻¹ is where median De increases slowly or starts decreasing. For Tr < 215 K, these trends are less coherent. This behavior suggests a transition from het cirrus to hom cirrus. For Ni in the tropics, there is no apparent change in slope, which may indicate that hom is less active in the tropics. Note that this analysis is based on clouds having τ < ∼ 3 and thus does not consider thick cirrus clouds originating from deep convection where τ > 3 and hom is probably active in the strong updrafts. Note that these observations apply to both in situ and WBC clouds, suggesting that similar physics applies to both cloud types.

A maximum in De is expected to occur after a slope change regarding Ni due to the competition for water vapor among a relatively high concentration of newly formed ice crystals. This competition reduces the growth rates of these ice crystals, keeping them relatively small. Other than hom, we know of no process that can explain the decrease in De at higher extinction shown in Fig. 9. Since for spherical ice the spherical volume radius Rv is proportional to (1/Ni)1/3, it is apparent that even for non-spherical ice crystals, it will take a larger change in Ni to affect a significant change in Rv or De through vapor competition effects. Thus, regarding De and Ni, a decrease in De caused by hom would be a more conservative criterion for defining hom cirrus. This would be more pragmatic if one were primarily interested in the impact of hom on radiation since cloud radiative properties are parameterized in terms of De and IWC (e.g., Gettelman et al., 2010; Emde et al., 2016). For this reason, we define hom cirrus as cirrus having an extinction exceeding that at the De maximum.

As shown in Figs. 9 and 10, De exhibits a maximum where d(De)/d(log⁡(αext)) ≈ 0. This maximum generally occurs for αext > 0.3 km⁻¹ and serves as a more conservative estimate of the transition from het to hom-affected conditions. Figure 11 shows plots of Ni in Tr vs. log⁡(αext) space for the tropics, midlatitudes and high latitudes over oceans for DJF and JJA in each hemisphere, where ∼ 0.01 < τ < ∼ 3. The transition from het to hom cirrus at each Tr value is where De is maximum, and this maximum is marked by the black “+” signs. However, in the tropics and in the summer NH midlatitudes, De maxima were ambiguous and the transition from het to hom cirrus at a given Tr was determined by a Ni threshold, Nthres. This threshold is the mean value of Ni that correspond to all the De maxima (51 in total) in the extra-tropics. The value and standard deviation of Nthres is 118 ± 57 L⁻¹ where the standard deviation is probably subject to variations in INP concentration (Haag et al., 2003). The hom threshold was assigned to the αext having the corresponding Ni closest to Nthres. In two extra-tropical datasets (DJF MID SH and DJF MID NH), Nthres was also used to estimate the hom threshold for Tr < 219 K and Tr < 207 K, respectively.

The horizontal dashed line in Fig. 11 indicates the temperature Tmin⁡, below which the number of cirrus cloud samples was too low for statistically meaningful analysis. The region of hom-affected cirrus clouds is thus defined as the region bounded by the + signs in the vertical and the dashed line in the horizontal, with the right-hand vertical boundary defined by the absence of data and the lower horizontal boundary defined by the 235 K isotherm. Also plotted in Fig. 11 are alternating solid and dashed curves indicating the fraction of samples, in increments of 0.10 from right to left, having a larger extinction than the extinction corresponding to the curve (at a given Tr). This provides estimates of the fraction of hom cirrus for a given Tr. For reference, the triangles predicted from hom theory are also shown (same as in Fig. 7). Overall, αext at het–hom transitions is found between ∼ 1.8 km⁻¹ (in the tropics) and 0.02 km⁻¹ (in the winter Arctic) and tends to decrease as Tr decreases (similar to the temperature dependence of Ni predicted from hom theory). Similar results were found for cirrus clouds over land (∼ 0.3 < τ < ∼ 3), which are shown in Fig. S12.

These hom region boundaries were incorporated into look-up-tables that relate 4 K temperature intervals to latitude band, season and surface type (oceans or land) for determining whether a sample was het cirrus or hom cirrus. When De > sensitivity limit, we set the sample as het cirrus. Applying this De sensitivity limit to all cirrus cloud samples over oceans reduces the estimated hom fraction by less than 8 % at Tr < 211 K. The largest changes are for Tr within the 223–235 K interval, where the hom fraction is reduced by 50 % in the tropics, and by 22 %–25 % in the extra-tropics.

The results in this study appear generally consistent with those reported in Krämer et al. (2016), titled “A microphysics guide to cirrus clouds – Part 1: Cirrus types” as shown in their Fig. 5 and summarized in their Fig. 9. That study combined measurements of cirrus cloud IWC from 17 aircraft campaigns (94 sampling hours) with predictions of cirrus cloud properties from a detailed microphysical box model. The modeling revealed that the measured median IWC effectively divided cirrus clouds into two updraft (w) regimes; slow (0.01 to 0.1 m s⁻¹) and fast (0.5 to 3.0 m s⁻¹). The slow w regime where het dominates was characterized by IWC < median IWC, relatively low Ni and relatively large mean mass volume radius Rv. Conversely, the fast w regime where hom dominates was characterized by IWC > median IWC, relatively high Ni and relatively small Rv. Interpreting these w regimes as het and hom regimes, respectively, this study reports similar findings in Figs. 9 and 10 (where αext correlates strongly with IWC). Figure 17 compares the temperature dependence of the retrieved median IWC (and variance as percentiles) where the median IWC usually lies near the temperature-dependent het–hom transition (shown by the black triangles) as described in Sect. 3.4.1 for cirrus over oceans (∼ 0.01 < τ < ∼ 3). In most cases there is a close correspondence between the median IWC and the het–hom transition IWC. The tropics and JJA MID NH are exceptions where the De maximum was ambiguous (suggesting hom is less active) and the hom threshold was determined by Nthres. Excluding these exceptions, the median IWC tends to partition the het and hom cirrus cloud regimes. This description is different from the explanations given in Krämer et al. (2016), where the median IWC roughly separates in situ cirrus from LOC (their Fig. 13).