Figure 1 presents time series of single cloud core volume fractions (fvol) and cores' properties for three aerosol concentrations (clean, intermediate, and polluted). Also included are time series of instantaneous rain rates (mm h-1) at the domain surface. For all aerosol concentrations and during most of the clouds' lifetimes, the volume fraction of Wcore tends to be the largest and the volume fraction of Bcore the smallest. Exceptions to this finding are seen either at the initial time step for the polluted cloud or in the later stages of cloud lifetime for the lower-concentration clouds. In addition, we find that RHcore⊆Wcore for all stages of cloud lifetime while Bcore⊆Wcore, RHcore for all stages of the polluted cloud, but this only applies to the growing stages of lower-concentration clouds before precipitation production. Thus, the main finding from PT1 (i.e., Bcore⊆RHcore⊆Wcore) generally applies to all aerosol concentrations during the pre-precipitation stages of the clouds' lifetimes.

Lower-aerosol-concentration simulations produce more rain and at earlier stages of cloud lifetime due to efficient collision coalescence. The increase in the Bcore volume fraction at later stages of cloud lifetime in those simulations (clean and intermediate) coincides with initiation of precipitation production followed by a consequent increase in the Wcore volume fraction as well (more so for the intermediate concentration). This dissipating Wcore is mostly contained within the Bcore. The possible mechanism behind the increase in prevalence of buoyant parcels during precipitation is explored in Sect. 3.2. The lack of RHcore pixels at these stages indicates that the Wcore is composed of pixels with small vertical velocities, insufficient for supersaturation production. The RHcore is the only one which is not sensitive to rain and monotonically decreases during all clouds' lifetimes. Another clear aerosol effect seen in Fig. 1 is an increase in cloud lifetime with a decrease in aerosol concentration. This point will be further explored in Sect. 3.3.

The theoretical arguments in PT1 showed that Bcore should be the smallest of the three. This was shown for both the adiabatic cloud column case and also the non-adiabatic case, where entrainment mixing and consequent evaporation have a strong net negative effect on cloud buoyancy. Despite this fact, results show (see Fig. 1 and Fig. 2 in PT1) that pockets of positive buoyancy may form independently of the other cores during dissipation and precipitation stages, even though evaporation is to be expected then. Since positive buoyancy is the result of either higher temperature or vapor content (or both) than the surrounding environment, we choose to analyze these two terms during different stages of the single clouds' lifetimes. The liquid water content (LWC) buoyancy term (not shown here) is always negative and typically increases (in absolute value) with an increase in vertical velocity or total buoyancy.

Figure 2 shows the values of the temperature (BT) and humidity (BQv) buoyancy terms in pixel buoyancy vs. pixel vertical velocity phase space. The scatterplots include all cloudy pixels during all time steps for the three different aerosol concentration simulations. The distribution of points for the polluted simulation shows a positive linear dependence of buoyancy on vertical velocity. Negative vertical velocity is associated with negative buoyancy, and positive vertical velocity shows a transition from negative to positive buoyancy with an increase in magnitude. For this case both BT and BQv increase with an increase in vertical velocity, as is generally expected in convective clouds. The sign of pixel buoyancy is mostly dependent on BT, since all pixels have positive BQv and a negative water loading term. This behavior is also seen for lower aerosol concentrations, where the sign of buoyancy is for the most part determined by BT. Hereafter, we refer to positive buoyancy (both BT and BQv) production within updrafts as updraft buoyancy.

The clean and intermediate simulations show a similar dependence of buoyancy on vertical velocity; however, it is apparent that these simulations also include an outlier scatter region of pixels with positive buoyancy and weak negative vertical velocity which is absent in the polluted simulation (see white arrows; Fig. 2). Consistent with the rest of the cloudy pixels, these outlier pixels have positive BT but differ in that they show neutral BQv. It can also be seen that these pixels are only attributed to the stages after surface precipitation has commenced (indicated by black dots in markers). Precipitation is indicative of both downdraft motion and abundance of large drop sizes.

Although not usually the focus of studies, the existence of positively buoyant downdrafts in convective clouds has been reported in both observations (Igau et al., 1999; Wei et al., 1998) and simulations (Xu and Randall, 2001; Zhao and Austin, 2005a, b). A possible explanation for this can be deduced from previous theoretical studies predicting mixing-induced downdrafts in cumulus clouds (Betts and Silva Dias, 1979; Betts, 1982). It was shown that in some cases cloud–environment mixtures are negatively buoyant (while still containing liquid water), and the consequent downdrafts can sometimes descend only partway down to the cloud base before reaching neutral buoyancy. Similar to convective overshooting, parcels with negative vertical momentum may then undershoot the downdraft equilibrium level and turn positively buoyant while the downdraft weakens. One can therefore expect the magnitude of positive buoyancy within the downdraft to reach a maximum when the velocity approaches zero. Hereafter we refer to positive buoyancy production within downdrafts as downdraft buoyancy.

Downdraft buoyancy production occurs frequently in cumulus fields because the negatively buoyant downdrafts follow a warming lapse rate which is more unstable than the environmental one, which is typically between the dry adiabat and moist adiabat (as is the case for the Hawaiian and BOMEX profiles simulated in this work). On one extreme, a descending parcel is least buoyant (i.e., coolest) when evaporation (after mixing) keeps it just barely saturated (Paluch and Breed, 1984, also PT1) so that the lapse rate of descent tends to be moist adiabatic and may remain negatively buoyant. On the other extreme, if little to no evaporation of liquid water occurs, the descent will follow the dry adiabat and switch to neutral (and then positive) buoyancy rapidly. Thus, the ability of a negatively buoyant cloudy downdraft to sustain itself depends on continuous inflow of liquid water (by mixing) and its consequent evaporation (Knupp and Cotton, 1985).

Indeed, the results in Fig. 2 match the hypothesis explained above, where positively buoyant downdrafts are warmer than the environment and tend to show larger buoyancy values for weaker downdraft velocities (especially for the intermediate case). Further analysis also shows that the more unsaturated the downdrafts (indicated also by low BQv), the larger the positive buoyancy. Moreover, the occurrence during precipitating stages and for lower aerosol concentrations indicates that slow evaporation due to larger drop sizes is crucial for downdraft buoyancy production, enabling a near-dry adiabatic lapse rate during descent.

Here we evaluate how aerosol effects within the core and margin (using the three core definitions) affect the cloud characteristics, focusing on two main parameters: size (or volume) and mass. In Fig. 3 we follow the evolution of cloud, core, and margin mass and volume for different aerosol concentrations, using only the RHcore definition. We choose the RHcore, since it is the most stable out the core types, generally decreasing monotonically (see Fig. 1). A non-monotonic dependency of total cloud mass on aerosol concentration is seen, showing a maximum for the intermediate concentration. This type of dependency has been previously reported for warm cumulus clouds (Dagan et al., 2015b; Savane et al., 2015).

One can generally expect an increase in diffusion and decrease in collision–coalescence processes efficiency with an increase in aerosol concentration (Hudson and Yum, 2001; Jiang et al., 2009; L'Ecuyer et al., 2009; Pinsky et al., 2013), affecting both condensation and evaporation processes. The intermediate concentration shows the highest total mass as a result of being an optimal case, with higher condensation efficiency than the clean case and lower evaporation efficiency than the polluted case. It is convenient to represent the condensation and evaporation efficiencies by the RHcore and RHmargin mass, respectively. The intermediate cloud has almost identical core mass as the polluted cloud but retains higher mass in its margin as well. The clean cloud shows the lowest core mass but manages to accumulate the largest mass in its margin that dissipates slowly in subsaturated conditions. By comparing the total cloud mass evolution with the core and margin mass evolutions, it becomes clear that the total mass is primarily dependent on the cloud core. Another way to see this is by plotting the core mass fraction (Fig. 3g), which shows that clouds are core dominated (core fraction >0.5) with respect to mass for most of their lifetimes and for all aerosol concentrations.

With respect to cloud total volume, the lower the concentration, the larger the total cloud volume. We note that the cloud volume here excludes regions of precipitation below the initial cloud base height. By separating to core and margin regions, one can see that the total cloud volume is primarily dependent on the volume of the margin, which increases significantly with decreasing concentration. This is especially true during the dissipating stages of cloud lifetime, when the cloud is margin dominated. Although increasing the aerosol concentration does initially yield an increase in core volume (as was seen for the mass), the extents of the core size are typically smaller than those of the margin. There are large differences in the relative core volume percentage for the different clouds. The clean (polluted) cloud is margin (core) dominated with respect to volume for most of its lifetime. Excluding time of formation, the clean cloud shows the lowest core volume fractions but manages to maintain its core for the longest time span.

These results with respect to cloud volume can be attributed to the smaller drop sizes and higher diffusion efficiencies with an increase in aerosol concentration. Additionally, lower collision–coalescence efficiencies also maintain a narrow droplet spectrum of small droplets in the polluted cloud. During the growing stage a higher aerosol concentration may permit the cloud to condense more water, release more latent heat, and promote cloud growth. This explains the larger core volume sizes. However, after the cloud exhausts its convective potential (i.e., the growth of the convective core terminates and reaches its peak in mass), its main method of expansion is by mixing with the environment (i.e., detrainment and dilution). We note that precipitation can also be considered a method of expansion; however our choice to focus on volume above initial cloud base excludes this effect. Detrainment and mixing with the environment result in subsaturation conditions and evaporation of LWC. A clear indication for dilution is seen in Fig. 3, where between 30 and 35 min of simulation time, both the clean and polluted clouds lose total mass but only the clean cloud increases in total volume. The polluted cloud is composed of small drops, evaporates its margin regions efficiently, and is thus limited in horizontal growth by detrainment. The clean cloud is composed of larger drops, less efficient in evaporating its margins, and hence can grow by dilution of its LWC upon a larger volume. This large margin “shields” the core during dissipation stages and enables it to the live for a longer time.

The mechanism behind the results in Fig. 3 is demonstrated in Fig. 4, where horizontal cross-sections of mean (taken in the vertical dimension) cloud RH are shown for different stages during the clouds' lifetimes. For the polluted cloud, supersaturated or subsaturated conditions are rare. The RH throughout the cloud is near 100 % (almost always between 99.8 % and 100.2 %) except for a few pixels at its far edges which are a bit below 99 %. The polluted cloud resembles what one would expect to see using a moist adiabatic approximation (i.e., saturation adjustment), where all excess water vapor above saturation is converted to liquid water, mimicking infinitely efficient condensation (and evaporation).

The clean cloud shows opposite behavior, with extremes of large supersaturation during cloud growth (initial stages) and large subsaturation during cloud dissipation (final stages). The large supersaturation can be explained by slow diffusional growth, but the large subsaturation also takes into consideration the larger drop sizes, which take more time to evaporate. This enables the clean cloud to expand to larger horizontal extents (by dilution and mixing with the environment without fully evaporating) and live for longer times. The intermediate aerosol concentration shows a midway scenario, where the supersaturation is consumed more efficiently than the clean case and at the same time much larger values of subsaturation may exist than those seen for the polluted case.

Here CvM space representations (see Sect. 2) are used to observe the core volume fractions of all clouds in BOMEX cloud field simulations. The rows in Fig. 5 represent different aerosol concentrations, while the columns represent different core type definitions. Different aerosol concentrations produce a vastly different scatter of clouds in the CvM space, as was previously discussed in depth (Heiblum et al., 2016b). The clean simulation (25 cm-3) shows two disconnected regions of cloud scatter: one which is adjacent to the adiabatic approximation and one of mainly small mass and high-COG clouds. The former region includes both clouds during their growth stages (smaller masses, LWP < 10 g m-2) and large precipitating entities (larger masses, LWP > 10 g m-2) which form due to merging processes (Heiblum et al., 2016b). The latter region (small mass and high COG) includes clouds at their dissipating stage, which form by the shedding mechanism off the large cloud entities. We note also the existence of small-mass elements well below the adiabat, representing precipitation cloud segments which shed off large precipitating clouds.

The polluted simulation (2000 cm-3) shows a much more homogeneous scatter of clouds. The lower part of the scatter (closest to the adiabat) represents the cloud growing branch, while the rest of the scatter represents dissipating clouds, either by the gradual process of rising cloud base or by an immediate process of shedding off larger cloud entity (see Fig. 1, PT1). Precipitating cloud segments below the adiabat are absent from this simulation. The intermediate simulation (250 cm-3) shows a scatter which generally resembles the polluted case. However, the existence of relatively disconnected (from the main cloud scatter) small-mass cloud segments below the adiabat and near the inversion base height resembles the clean simulation as well. It should be noted that horizontal dashed lines in Fig. 5 represent the inversion base height after 5 h of simulation (approximately middle of simulation), where an increase in the inversion base height is seen with a decrease in aerosol concentration. This is due to increased net warming in the upper cloudy layer (i.e., release of latent heat during condensation with reduced local evaporation) with an increase in precipitation (Dagan et al., 2016; Heiblum et al., 2016b), which raises the inversion base.

The results in Fig. 5 show a consistent behavior of the core volume fractions for all aerosol concentrations, where the Wcore type shows the largest fractions and the Bcore type shows the smallest fractions. The Wcore and RHcore generally show a decrease in core fractions along the growing branch, while the Bcore fraction initially increases with cloud growth and then decreases for the large-mass growing clouds. The percentages in the panel legends (Fig. 5) indicate the fraction of clouds (out of the scatter) which are core dominated with respect to volume (fvol > 0.5). For all concentrations, less than 7 % of clouds are Bcore dominated, while more than 55 % are Wcore dominated (with RHcore percentages somewhere in between). The Bcore typically occupies a small portion of a typical cloud volume, while the Wcore typically occupies most of the cloud. The mean cloud area (proportional to scatter point size) shows an increase with the increase in mean cloud LWP.

These results are consistent with PT1 and the single cloud simulations in Sect. 3.1. Nevertheless, some significant aerosol effects on the partition to core types can be seen. Focusing on the growing branch first (i.e., clouds located near the adiabat), we note the following:

For the RHcore type, the core volume fractions of clouds after formation (i.e., with small mass) increase with decreasing aerosol concentration. This effect was also seen for the single cloud simulations and can be explained by the reduced efficiency of supersaturation consumption for fewer aerosols.

In Fig. 6 the core mass and volume fractions (fmass and fvol, using the RH definition) of all growing branch clouds are sorted by mass for the three aerosol concentrations. We note that the higher cloud masses reached by lower-aerosol-concentration simulation can be explained by cloud field organization effects due to precipitation (i.e., increased merging of clouds) rather than increased cloud condensation (Heiblum et al., 2016b; Seigel, 2014). The clean case starts off with the highest core fractions (both mass and volume), which decrease steadily with an increase in mass (or increase in lifetime). For all concentrations, most of the cloud mass is concentrated in the core region. The polluted case shows a slight increase in core mass fractions with increase in mass, while the other two cases show decreases in core mass fractions.

The core volume fractions show lower values than the mass fractions. The clean clouds are margin dominated for most masses, and the polluted clouds are core dominated for all masses. The intermediate case is generally confined to values between the other two cases. Figure 6 can be considered comparable with the lower panels in Fig. 3g and h, but excluding the dissipating part of those time series. The similar findings in both figures indicate the robustness of the aerosol effects on core properties in clouds.

Following the analyses of Sect. 3.1, we next test how aerosol concentration affects the subset properties of one core type within another for all clouds in a field (Fig. 7). We focus only on the typically smaller-sized cores (Bcore, RHcore) within larger-sized cores. Out of the three permutations, the RHcore inside Wcore shows the lowest sensitivity to aerosol. All three growing branches (for the different aerosol concentrations) consistently show that the RHcore is a subset of Wcore (i.e., RHcore⊆Wcore), while the dissipation branches show much lower overlap fraction between the two cores.

For the dissipating clouds in general, the lower the mass and the higher the COG, the smaller the overlap. The dissipating branches do include a scatter of small cloud for which RHcore⊆Wcore, comprised of small cloud segments which shed off the main core regions of larger clouds. These findings slightly differ from those of the single cloud simulations that show RHcore⊆Wcore for their entire lifetimes, while for cloud fields this property breaks down during dissipation. This difference highlights the importance of cloud interactions (i.e., splitting and merging) and cloud field air flow patterns (i.e., organized advection, updrafts, and downdrafts) in determining the relationships between core types, enabling supersaturation and downdrafts to coincide in small dissipating clouds.

The other two permutations (i.e., Bcore inside RHcore, Wcore) show significant changes due to aerosol. For the polluted case, Bcore⊆Wcore for nearly all clouds, including clouds at initial stages of dissipation. Similar results are seen for Bcore inside RHcore, but with slightly lower pixel fractions. The polluted case thus illustrates the case of buoyancy production due to convective updraft. For the lower aerosol concentrations, two main aerosol effects are seen:

The lower the concentration, the lower the chance that Bcore is a proper subset of the other cores for large growing branch clouds.

In Sect. 3.2 it was seen that for single clouds, positive buoyancy results from two main mechanisms: (i) convective updrafts – where updrafts promote supersaturation and latent heat release, and thus always positive BQv and frequently positive BT – and (ii) dissipative downdrafts – where subsaturated cloudy downdrafts promote a positive BT and neutral BQv. The latter case is dependent on low evaporation efficiency and hence seen mostly for precipitating stages of low aerosol concentration simulations. In Fig. 8 we perform a similar test for the cloud field scale. Instead of analyzing pixel by pixel, we check whether each buoyancy core within a cloud is BT or BQv dominated. To quantify this we use a normalized buoyancy dominance parameter ΣpixelBT>0-ΣpixelBQv>0ΣpixelB>0, where a core comprised of only BT>0 (BQv>0) pixels yields 1 (-1). Hence, we expect negative (positive) values to indicate dominance of updraft buoyancy (downdrafts buoyancy).

Analysis of the buoyancy components in the CvM space (Fig. 8c, f, and i) shows that the large majority of clouds are BQv dominated. For all concentrations, clouds initiate with all pixels showing BQv>0. As clouds develop along the growing branch the Bcore becomes more abundant with BT>0 pixels. This is expected with increasing release of latent heat during cloud growth. During dissipation BQv again becomes the dominant component for the majority of clouds. The polluted simulation shows an extreme case where all buoyancy cores in the simulation are BQv dominated, while for the lower concentrations a portion of the dissipating and precipitating clouds are BT dominated.

Thus, we hypothesize that the polluted simulation only permits buoyancy cores of the updraft type which intersect with the other core types (i.e., Bcore∈RHcore, Wcore), while the lower concentrations also permit buoyancy cores of the downdraft type which do not intersect with the other core types (i.e., Bcore∉RHcore, Wcore). We test this by observing the relation of cloud maximum absolute vertical velocity (Fig. 8a, d, and g) and mean drop size (Fig. 8b, e, and h) with the relative dominance of the buoyancy terms. Absolute vertical velocity is chosen to represent both updrafts and downdrafts. The data are further separated into independent (Bcore∉RHcore, Wcore) and dependent (Bcore∈RHcore, Wcore) buoyancy subsets of the data by that separating to buoyant cores within updrafts and downdrafts. Clear aerosol effects are seen on cloud mean drop size and maximal |W|. As expected, there is a decrease in drop size with increase in aerosol concentration and increase in maximal velocity. Regarding cloud field buoyancy, as predicted, the independent buoyancy cores are more frequently BT dominated than the dependent buoyancy cores.

The polluted case is populated with dependent cores (white scatter) and shows a classic pre-precipitation convective growth scenario, where relative dominance of the BT term increases linearly with increase in cloud mean drop size. A logarithmic dependence of BT dominance on maximal |W| is seen, which saturates at high maximal |W|. This can be explained by the fact that increased convection mainly increases the abundance of pixels with BT>0, but without changing the fact that the entire cloud is BQv>0, so that BT is unlikely to become the dominant term. The lower concentrations show a more complex scenario. These simulations show a superposition of dependent core convective growth behavior (i.e., the scatter pattern seen for the polluted case) and additional populations of both dependent (other white scatter points) and independent (black scatter) cores.

The independent cores span all the range of possibilities of BT and BQv relative dominance. They tend to have larger cloud mean drop sizes, and a near-zero maximum |W|, indicating that they only form at late non-convective stages of cloud development. Furthermore, a trend is seen for the subset of scatter that is BT dominated, where a positive (negative) correlation between mean drop size (maximal |W|) and BT dominance is seen. This again stresses the importance of drop size on the formation of positive buoyancy within downdrafts and highlights the fact that BT should be largest (and most abundant) below the downdraft equilibrium level, when the |W| approaches zero. The independent cores that are BT dominated thus fulfill the characteristics of the downdraft buoyancy production process, while the independent cores that are BQv dominated may originate from larger clouds (shedding mechanism) with high humidity content, have weak |W|, and are slow to evaporate.

The intermediate simulation shows an additional scatter area of dependent core clouds with increasing BT relative dominance for lower maximal |W|, located between the independent core clouds and the convective growth core clouds. These clouds may represent a gradual transition from BQv dominance to BT dominance during dissipation which is only possible in the intermediate simulation. This scatter area is absent from the clean and polluted simulation, in the former case due to absence of the gradual dissipation pathway and in the latter case due to efficient evaporation eliminating Bcore during dissipation. We note that the intermediate case shows a slightly higher percentage of clouds that are BT dominated (see legends in Fig. 8) than the clean case. This can be due to stronger convection in this simulation (i.e., increased |W| range), which favors increased mixing with the dry environment (see Fig. 9) and the formation of unsaturated strong downdrafts that descend below the level of neutral buoyancy.

From Fig. 3 it was demonstrated that a large part of the differences in single cloud characteristics (such as mass, volume, and the partition of these to core and margin regions) due to aerosols can be attributed to differences in vapor diffusion efficiencies. In Fig. 9 we check how these aerosol effects are manifested in the cloud field scale (using the CvM space) by observing the mean RH in the cloud core and margin of all clouds, where the core (margin) mean RH can be taken as a proxy for condensation (evaporation) efficiency. To gain additional intuition regarding the distribution of RH values within the clouds, vertical cross-sections (parallel to the prevailing wind direction) of the most massive clouds from each simulation are shown.

The vertical cross-sections demonstrate the large differences in the massive clouds for each of the simulations. In addition to the increase in precipitation production, lower aerosol concentrations yield much larger horizontal extents of clouds. The clean, intermediate, and polluted most massive clouds have a maximum radius of ∼3, ∼1, and ∼ 0.5 km, respectively. It is clear from the cross-section that the clean cloud is actually composed of two large clouds which merge together. For the clean case, the highest RH values are reached slightly below the cloud top. The edges of the clouds show subsaturation conditions, with the lowest RH values observed below the lifting condensation level (LCL; precipitation regions) and at the upper interface of the cloud with the environments.

The intermediate-case cloud shows lower maximal and minimal RH values and an increased dominance of the margin region. This cloud penetrates the inversion layer and entrains dry air into the cloud. In addition, the cloud produces significant precipitation which initiates downdrafts of dry entrained air through the cloud center. It can be seen that the increased vertical development of the intermediate-case cloud in comparison with the clean case increases the mixing with the environment. Thus, the dynamic effect of increased mixing and reduction in cloud RH overcomes the microphysical effect of increased evaporation and increase in cloud RH. The polluted-case cloud, on the other hand, shows a homogeneous RH pattern, with most of the cloud showing around 100 % RH and only a thin layer at the cloud edges (mainly at the upper regions) showing lower RH values. The polluted cloud penetrates the inversion layer as well, but this case lacks precipitation, and the microphysical effect of evaporation overcomes the dynamical effect of mixing.

Keeping in mind the insights obtained from comparisons of individual cloud, we move on to compare the RH characteristics of all clouds within the field. Looking first at core mean RH, a robust decrease is seen with increase in aerosol concentration. This decrease is seen for all cloud types and locations within the CvM space. The polluted case displays the most homogeneous pattern, with all clouds showing core mean RH values around 100 %, indicating efficient consumption of the supersaturation. The intermediate case displays a slightly less homogeneous pattern, with values ranging from 100 % to 101 % and with the higher values occurring along the growing cloud branch, especially for the largest clouds. The clean case shows the largest variance in core mean RH, ranging from 100 % for some cloud fragments that soon start to dissipate to 103 % in the cores of the large cloud entities. In addition to the low efficiency in consuming supersaturation, the high RH values in clean large clouds are due to the “protection” by large margin regions surrounding the core region.

The CvM patterns of mean margin RH show significant differences between the polluted case and the other two. The mean margin RH values of the polluted case are only marginally lower than 100 %, since subsaturated conditions within the cloud are quickly adjusted by efficient evaporation. Only the largest clouds in the polluted case permit lower mean margin RH values (∼95 %) due to the entrainment of very dry environmental pixels near the cloud tops (as seen in the vertical cross-section as well). The intermediate and clean cases show similar patterns. The smaller mass clouds (both growing and dissipating) show values above 95 %, while the larger mass clouds show values as low as 85 %. The larger clouds are most likely to reach low-RH areas near the inversion base and below the LCL (i.e., sub-cloudy layer) and entrain dry air and by that reduce the cloud margin RH.

As seen in the vertical cross-section examples, the largest clouds in the intermediate case have even lower margin RH values than for the clean case. This can be explained by the increased development of the large intermediate clouds to heights with lower RH and by more intense downdrafts for these large clouds. The lowest RH values in the domain are seen for the precipitating fragments (i.e., located below the adiabat). These fragments typically contain low concentrations of large drop sizes (precipitation drops) which are slow to evaporate and capable of surviving in low-RH conditions within the sub-cloudy layer.