The Community Atmospheric Model, version 6 (CAM6), is the atmospheric component of CESM. The present study uses the Single Column Atmospheric Model, version 6 (SCAM6) which represents a single grid-box column of the global model and utilises the CAM6 physics package . Large-scale tendencies for SCAM6 are prescribed from observations or global simulations. SCAM6 is a valuable tool for developing and testing parameterisations generally.

In CAM6, the original version of the stratiform cloud microphysics scheme followed a two-moment bulk microphysics approach (Morrison and Gettelman, 2008; Gettelman and Morrison, 2015), hereafter “MG08”. The scheme represented four cloud hydrometeor species: cloud liquid, cloud ice, snow and rain. The activation of cloud droplets followed . Initiation of ice crystals followed . The ice number concentration in the stratiform microphysics scheme was limited so as not to exceed the so called “prescribed” value calculated at about -36 °C. For SIP, the original stratiform scheme (MG08) represented only the HM process of rime-splintering, and other SIP processes were omitted.

In this paper, we have modified the stratiform cloud microphysics in the CAM6 global model by including new microphysical processes and changing the representation of certain included processes of stratiform microphysics. The changes are summarised as follows:

The microphysics scheme now represents five cloud hydrometeor species: cloud droplets, rain, cloud-ice, snow, and graupel/hail. Graupel/hail mass and number mixing ratios are treated diagnostically instead of prognostically for the purpose of treating microphysical processes such as SIP (HM process, breakup in ice-ice collisions). Since there is no prognostic variable for graupel/hail in the global model (CAM6), its amount is diagnosed for the purpose of treating microphysical processes, according to a look-up table for graupel mass (as a function of cloud-liquid and snow mass mixing ratios and temperature) from high-resolution cloud simulations with the aerosol-cloud model (AC).

The treatment of stratiform microphysics is qualitatively consistent with the new convective microphysics scheme also for CAM6 (Jadav et al., 2025). In that scheme, deep convection follows the Zhang–McFarlane framework but with the modifications introduced by Jadav et al. (2025): a 6 h convective adjustment time scale, revised triggering including convective inhibition (CIN), replacement of SZ11 microphysics by a detailed aerosol–cloud convection (ACC) microphysical scheme with a high-resolution 1-D plume model, updated entrainment–detrainment formulations, and explicit transfer of detrained cloud liquid, ice, rain, snow, and graupel to the large-scale cloud scheme as prognostic source terms.

The new ACC treatment involved embedding a parcel in which hybrid bin-bulk microphysics routines are applied to treat coagulation, with ascent of the parcel in 1D through each bulk plume, using vertical velocity information derived from the convecective available potential energy (CAPE) in the ZM framework. The known and empirically characterised mechanisms of initiation of droplets and ice particles, pertinent to deep convective ascent, were treated, including three SIP mechanisms, homogeneous freezing of drops, heterogeneous ice nucleation in terms of aerosol conditions of size, loading and composition, and separate explicit treatment of cloud-base and in-cloud droplet activation. The supersaturation in the cloudy parcel was explicitly predicted throughout the ascent, enabling this detailed treatment.

The microphysical treatment of both schemes is related to that in a high-resolution aerosol-cloud (AC) model . Both schemes for large-scale cloud, described here, and for deep convection (Jadav et al., 2025) are included in the control simulations of the present paper.

The MC3E campaign was carried out over the Atmospheric Radiation Measurement (ARM) Southern Great Plains (SGP) site in Oklahoma to study mesoscale convective systems (MCSs) from April to June 2011. The campaign consisted of a Central facility (CF) and 20 extended facilities, which covered an area with a 150 km radius. The campaign incorporated ground-based and in situ aircraft observations .

The observed case of the MCS was simulated from 10 to 13 May 2011 in SCAM6 with a global time step of 20 min and grid size of 100×100 km. Aerosol concentrations in each species are initially determined from the Goddard Chemistry Aerosol Radiation and Transport (GOCART) global model . These vertical profiles of aerosol concentration were then adjusted at all levels based on the averaged measurements near the ground from the Interagency Monitoring of Protected Visual Environments (IMPROVE) from 9 to 12 May 2011. More details of the initial aerosol conditions are provided in .

Simulations are performed with the new stratiform scheme referred to as “LS 24”. LS24 includes the new version of convection scheme (Jadav et al., 2025), for consistency in the treatment of microphysics. The predicted cloud hydrometer profiles are conditionally averaged over the simulation period and are validated against aircraft data. Regarding the aircraft observations, the stratiform region is considered to be where vertical velocity is less than 3 m s⁻¹. We also compare the results from the model simulation with the original convective and stratiform cloud schemes (the “MG08” run).

Figure a shows that the LS 24 run predicts the two precipitation peaks from deep convection at 21:00 UTC 10 May and 18:00 UTC 11 May with adequate timing, although the major convective peak is predicted about 6 h late for several reasons. First, the Zhang-McFarlane convection scheme in CAM6 uses a CAPE-based trigger with dilute CAPE closure. Previous studies have shown this tends to delay convection by 1–3 h in mid-latitude systems compared to convection-permitting models (), and the same must be even more true in SCM simulations with no resolved 3D dynamics. Second, the VARANAL forcing dataset () represents domain-averaged conditions of large-scale advection into the simulated area (one global model grid-box), while the triggering processes in reality are sub-gridscal. Third, the forcing omits advection of condensate.

Moreover, both predicted peaks of precipitation rate (10 and 11 May) have insufficient intensity by a factor of 2 (weaker initial peak) and by about 20 % (main peak) respectively (Fig. a). The contributions from stratiform cloud are only minor during both predicted peaks. However, a consequence is that a surplus of humidity in the environment remains to allow too much deep stratiform cloud after the main peak on 12 May, especially during the subsequent 12 h. Then there is too much stratiform precipitation with weak peaks of 2 or 3 mm h⁻¹ (12 May), approaching intensities of both prior convective peaks (10 and 11 May) noted above. This stratiform precipitation on the final day (12 May) compares with practically none observed.

Regarding the cumulative surface precipitation (Fig. b), this is slightly over-predicted by only about 10 % at the end of the simulation for LS24, predicting 26.5 mm versus 24.0 mm observed. Stratiform precipitation overall contributes 65 % to the predicted accumulated precipitation with the LS24 run, in agreement with detailed simulations of the case by AC (). Thus, the bias noted above of stratiform precipitation on the final day (12 May), predicted but not observed, is explicable as a delay of about 12 to 15 h in the fall-out of most of the simulated stratiform precipitation, consistent with under-prediction of convective outflow of deep precipitating stratiform cloud. Whereas in nature, deep convection generates stratiform cloud by detrainment from cores in any MCS simultaneously, the model tends to generate the stratiform cloud from the environment independently (a similar decoupling problem is evident in the far weaker precipitation from MG08 in Fig. b). This is a cloud-dynamics bias that appears to be a feature of the treatment of convective outflow in CAM, unrelated to the microphysics. It is also possible that the mixed-phase stratiform cloud produces precipitation too slowly by the ice crystal process.

Table  shows that the LS24 run has an error of about 20 % to 30 % in the radiative fluxes at TOA and at the surface when compared to the satellite observations. An under-prediction by 17 % of net shortwave radiative flux at TOA entering the climate system is consistent with the model over-estimating the amount of cloud condensate in layer-cloud. This bias is consistent with that noted above of stratiform precipitation persisting for too long after the major convective peak on the final day. Similarly, insufficient outgoing longwave radiation at TOA by 20 % is explicable in terms of cold layer-cloud that is too high or similarly too extensive. It can be inferred from Fig. , a significant amount of ice crystals exists at upper levels because of homogeneous freezing. Higher-level clouds with abundant ice crystals act to reduce the longwave radiation emitted to space. Nevertheless, the radiation predictions are more accurate than for the MG08 scheme.

Figure a shows that the LS24 run predicts the cloud base is at 16 °C, in agreement with detailed simulations by our high-resolution AC model . The LWC predicted by the LS24 run, agrees well with the aircraft observations and falls within the 90 % confidence intervals for mean values. The LS24 run predicts a maximum LWC of 0.5 g m⁻³ at 14 °C. The observed LWC data-points differ less from the LS24 simulation than they do from each other. The LS24 average bias is about 10 % and MG08 is +30 %.

Figure b shows the cloud droplet number concentration (CDNC) in comparison with the probe data (King, Nevzorov and CDP probes). CDNC predicted by the LS24 run agrees with aircraft observations at most levels. Both LS24 and MG08 have a similar average bias of about 10 %.

If left uncorrected, the artificial bias from ice shattering on the probe would increase the measured number concentrations of ice particles by an order of magnitude or more, especially at smaller sizes (Korolev et al., 2011). Therefore the observations were corrected with anti-shattering tips and by considering data only for sizes >200 µm and 1 mm. The same size threshold is applied to the simulation results to be consistent.

Figure a shows agreement of the predicted average of ni₂₀₀ (ice number concentration of particles with sizes >200 µm) from LS24 with the aircraft observations, which vary logarithmically in the vertical. The predicted concentration of ice particles has the correct order of magnitude and differs from the observed values by less than the measurement error of the observations, with different probes at any given level differing from each other by almost an order of magnitude. Near the freezing level, ni₂₀₀ predicted by the LS24 run differs by only about 30 % from observations. LS24 average bias is -58 % and MG08 is -84.0 %. LS24 has a lower absolute bias.

Figure b shows that the prediction of ni₁ for precipitation (ice number concentration of particles >1 mm) from LS24 has the same order of magnitude as the aircraft observations in the lower half of the “mixed-phase region” of temperature (0 to about -36 °C). This is where both liquid water content (LWC > 0.01 g m⁻³) and ice water content (IWC > 0.01 g m⁻³) may co-exist. In the MC3E case, the mixed-phase region spans altitudes of 4–9 km (corresponding to -5 to -30 °C). It is where supercooled liquid droplets and ice particles may interact through either the Bergeron-Findeisen process, riming, and other microphysical processes. Although it is lower by an order of magnitude in the upper half, only one probe (HVPS) for the larger sizes is available and the instrumental error is difficult to quantify (Fig. b). If there is an uncertainty by an order of magnitude in the measurement of these concentrations, as noted above for the smaller sizes (>0.2 mm) in view of the spread among various probes, then there is little evidence of any bias with the larger sizes. the average bias is -86 % for LS24 and for MG08 is -95 %, so LS24 is more accurate by about half an order of magnitude. However, LS24's bias ranges from −55 % at very cold temperatures to +100 % at warm temperatures, representing bidirectional errors that partially cancel.

Figure c shows the ice water content (IWC) predicted by the LS24 run agrees with the observations at most levels, with the correct order of magnitude. However near the -20 °C level, the IWC is under-predicted by half an order of magnitude. Overall with the new scheme, the distribution of predicted IWC is much closer to the observations than is the MG08 scheme, as for the filtered ice concentrations noted above. The average bias is -32 % for LS24 and for MG08 is -66 %. LS24 demonstrates decisive superiority with a lower bias. Remarkably, LS24 achieves 0 % bias at −26 °C (perfect agreement at this optimal temperature) and maintains good performance through −11 °C (−18.2 % bias), with accuracy degrading only near 0 °C where mixed-phase conditions create measurement ambiguity. MG08 shows consistently poor performance across all temperatures (−65 % to −83 %), suggesting a fundamental calibration deficiency.

The MC3E storm was simulated with the original unmodified version of the model, and the run is referred to as the MG08 run. Accuracy of results is compared between the MG08 and LS24 runs.

The original SCAM6 model (MG08) predicts adequately the timing of the total precipitation peaks at the end of 10 and 11 May, but under-estimates the intensity of the second (main) peak. The first peak is predicted to be purely convective when according to the detailed simulation it was mostly stratiform (Gupta et al., 2023). The cumulative surface precipitation from MG08 is 60 % lower than observations by the end of the simulated period, which is much less accurate than that for LS24 noted above. Generally, only 30 % of all precipitation is stratiform for MG08, while detailed high-resolution simulations with AC, comprehensively validated by Gupta et al. (2023), show that this fraction is 80 %. Thus LS24 appears to agree better with observations in the cause and intensity of precipitation than does MG08.

Figure a shows the LWC predicted by both schemes does not differ so greatly, with a higher LWC from MG08 by up to a factor of 3 relative to the new scheme (LS24). LS24 is more in agreement with the aircraft observations, though MG08 is not inconsistent with these in view of the variability among the limited number of aircraft traverses. Figure b, shows that both runs (MG08, LS24) are similarly in agreement with the observations of CDNC at most levels.

Figure a shows that at all levels, ni₂₀₀ values predicted by both schemes mostly agree with the observations, though less so for MG08 than for the new scheme (LS24). Below the -10 °C level, the value from MG08 is up to an order of magnitude too low compared with the observations. Generally, the ice concentrations from MG08 are lower than from LS24 by up to an order of magnitude. Above the homogeneous freezing level (-36 °C), ni200 predicted by MG08 is up to an order magnitude lower than the LS24 run. Figure b shows that the precipitation concentration, ni₁, predicted by the MG08 run, is at least an order of magnitude lower than observations at all flight levels, being lower than LS24 by at least half an order of magnitude at most subzero levels.

Figure c shows that the original MG08 run under-predicts IWC by about half an order of magnitude throughout the mixed-phase region compared to observations. At most subzero levels, it is less than LS24 by up to an order of magnitude.

The MG08 run predicts net radiative fluxes slightly less accurately than the LS24 run. At the TOA the net shortwave and longwave fluxes are 30 % and 15 % too low respectively for MG08 compared to observations.

The simulations show the various sensitivities to different pathways of ice initiation. The mechanisms were prohibited as discussed in Table  in both the stratiform and convective microphysics schemes.

Figure shows that excluding all SIP mechanisms (no-SIP case) produces relatively modest changes in ice water content and ice number concentration relative to the control run, with differences of 15 %–25 % rather than the order-of-magnitude changes sometimes reported in other studies. Prohibiting all SIP reduces the snow number concentration by up to a factor of two in the mixed phase region (Fig. c) relative to the control. Excluding breakup in ice-ice collisions reduces the snow number concentration by half an order of magnitude in the lower half of the mixed phase region. The corresponding changes in the snow mass content are minimal.

This muted sensitivity stems from several factors specific to the MC3E case and the model configuration. At temperatures below about -36 °C in cloud-top regions, homogeneous freezing produces explosive ice concentrations (103 to 104 L⁻¹) that swamp the SIP contribution. Figure a shows homogeneous freezing contributes about 99 % of the ice particle concentration at T<-36 °C. Since most of the ice particles originate at upper levels where homogeneous freezing dominates and are downwelled, disabling SIP has limited impact on column-integrated IWC. Comparison with high-resolution simulations of the same case by reveals that the homogeneous ice is downwelled too far into the lower half of the mixed-phase region in the SCM runs, which tends to evaporate supercooled cloud-liquid. This artificially inhibits graupel production and SIP. Moreover, as noted above, too little production of deep stratiform cloud by convective outflow, as noted above (Sect. 4.1), would tend to mute the sensitivity with respect to SIP, which produces most of the ice in the convective cores ().

As noted above, Fig. a shows without homogeneous aerosol freezing in the large-scale clouds (no-HOMO STRATO case), the ice concentration is reduced by two orders of magnitude in the mixed phase region of temperature (0 to -36 °C) and by even more aloft in cirriform clouds. Further removal of all other homogeneous freezing causes only a slight reduction by up to 30 % relative to the no-HOMO STRATO case. Regarding SIP, the no-SIP run reveals little systematic impact on stratiform ice concentration from all SIP in the control run, which is explicable in terms of the prevalence of homogeneous ice throughout all sub-zero levels. Switching off individually each SIP mechanism perturbs the ice concentration by up to ±50 %, due to alteration of precipitation amounts aloft and supercooled cloud liquid properties, affecting homogeneous freezing in a non-linear way.

Figure b shows a marked reduction of IWC by about an order of magnitude in the upper half of the mixed-phase region when homogeneous aerosol freezing in the large scale cloud scheme is prohibited. There is a similar reduction in the cirriform cloud at upper levels. This is all consistent with the decrease in ice number concentration noted above. The complete removal of all other homogeneous freezing (no-HOMO) further decreases IWC slightly (by 20 %) relative to no-HOMO STRATO, with a comparatively small incremental change. The no-SIP run results in minor changes to the IWC profile relative to the control, reaffirming the dominance of homogeneous freezing in driving ice mass in the simulations. Switching off individual SIP processes introduces variations in IWC of up to ±40 %, attributed to their perturbation of mixed-phase cloud microphysics, especially through indirect effects on alterations of vapor growth, riming, and sedimentation of ice.

Figure a and b show how the cloud liquid properties are influenced by SIP mechanisms and homogeneous freezing. The no-SIP run shows an increase in the supercooled LWC throughout the mixed phase region by 10 % relative to the control, because the growth of fewer ice particles causes less depletion of liquid by evaporation and riming. For similar reasons, in the lower half of the mixed phase region of the cloud, prohibiting all homogeneous freezing boosts the LWC by about half an order of magnitude relative to the control. The cloud droplet number concentration is mostly increased by excluding either homogeneous freezing or SIP, because the descent of homogeneous ice into the mixed-phase region creates more subsaturation with respect to water, evaporating supercooled cloud-liquid.

Figure c,d shows that the mass and number concentrations of rain are perturbed but not in a systematic manner at all levels by the exclusion of most of the mechanisms of SIP and homogeneous freezing. An exception to this is the exclusion of breakup in ice-ice collisions, which reduces the rain concentration by about an order of magnitude below the freezing level and by about half an order of magnitude above. Yet there is much less impact from excluding all SIP processes, indicating cancellation among these. Another exception is exclusion of all homogeneous freezing, which boosts the supercooled rain concentration by half an order of magnitude in the lower half of the mixed-phase region, due to the impact on supercooled LWC noted above.

Figure a, b shows that the two major convective peaks seen in the observations (10 May, 21:00 and 11 May, 21:00) are predicted to be boosted by about 20 % from the SIP being prohibited, but the stratiform precipitation following the major peak on the final day is slightly weakened (by 20 %). Switching off all the homogeneous freezing drastically weakened the stratiform precipitation on the final day by a factor of about 2, similarly diminishing the cumulative surface precipitation. The no-breakup case run had a cumulative surface precipitation very similar to the no-SIP run, indicating that the breakup in ice-ice collisions dominates the overall SIP impact on precipitation.

Figure c, d shows the impact on the top of the atmosphere radiation, also detailed in Table . The no-SIP run has a TOA Net Shortwave flux entering the atmospheric column that is 3 W m⁻² lower than control due to more cloud condensate aloft at sub-zero levels with a corresponding reduction in the outgoing long-range radiation, relative to control. Without homogeneous freezing, there is a stronger reduction (by 12 W m⁻²) in SW flux entering the climate system than without SIP due to more cloud condensate aloft. However, at upper levels there is less homogeneous ice and more emission of longwave radiation to space from warmer average emitting level when homogeneous freezing is excluded.

Figure a illustrates the response of ice number concentration in the sensitivity experiment where the concentration of ice nucleating particles (INPs) in the environmental aerosol population is increased by a factor of 100 at all levels. Despite already high baseline loadings of INPs, the Figure reveals that the ice number concentration exhibits more than an order-of-magnitude variation across the temperature range. In the control simulation at subzero levels, homogeneous freezing is the dominant mechanism for ice formation overall, since it occurs at temperatures colder than about -36 °C, and humidities approaching water saturation (high supersaturations with respect to ice). This leads to a high number of ice crystals aloft at upper levels which are then downwelled somehow.

In contrast, the high-INP case (100× INP numbers) activates heterogeneous nucleation more intensely during (e.g., large-scale) stratiform ascent – at warmer temperatures and lower supersaturation. This early onset of ice formation limits the buildup of supersaturation necessary for homogeneous freezing, thereby suppressing it. As a result, the homogeneous nucleation pathway and the related downwelling of very numerous ice particles are effectively “switched off” in the high-INP case, leading to an overall reduction in total ice number concentration throughout the mixed-phase region of temperature, as is evident in Fig. . Thus, the model represents the known competition between homogeneous aerosol freezing and heterogeneous ice nucleation in cirriform clouds . The early consumption of water vapor by heterogeneous nucleation prevents further crystal formation homogeneously and shifts the microphysical regime.

Moreover, with fewer ice crystals forming overall in the high-INP case, the competition for water vapor during growth is reduced. This enables existing crystals to grow larger, contributing to increased snow production. The enhanced snow growth can lead to greater meltwater generation and subsequently more rainfall through the ice crystal process of precipitation (or “cold-rain process”), consistent with findings reported by . Finally, Fig.  also reflects a shift in the equilibrium saturation conditions – ice formation occurs at a lower saturation threshold in the high-INP scenario

Figure b shows that the ice water content (IWC, g m⁻³) follows a pattern consistent with the response of ice number concentration to increased INP loadings under the 100× INP sensitivity scenario. In the control simulation, where homogeneous freezing dominates at lower temperatures and high supersaturation, the rapid formation of numerous ice crystals leads to elevated IWC values. In contrast, in the high-INP case, earlier initiation of heterogeneous nucleation results in fewer crystals forming, thereby lowering IWC in regions that would otherwise favor homogeneous freezing. However, the larger size of ice particles in the high-INP case noted above partially compensates and maintains moderate IWC values across a portion of the temperature range. This balance between fewer but larger crystals shapes the overall IWC response and underlines a shift in the ice growth regime.

Figure c also illustrates the snow number concentration (L⁻¹), providing insight into how ice particles evolve into precipitation-sized snow. In the control case, where homogeneous nucleation yields a high number of small ice crystals, subsequent growth processes (e.g., riming and aggregation) produce a relatively large number of snow particles. In the high-INP simulation, however, the fewer initial crystals noted above cause the downstream formation of snow particles, leading to lower snow number concentrations. The reduced competition for water vapor, and the greater abundance of supercooled cloud-liquid from fewer crystals in the large-scale cloud (Fig. ), in this scenario favors the growth of fewer, but larger, snow particles. This further highlights the shift in microphysical pathways due to increased INP concentrations.

The snow mass content (g m⁻³) in Fig. d shows that, while the snow number concentration is reduced in the high-INP scenario relative to the control, the snow mass content does not decline proportionally. This is because the reduced number of snow particles grow more efficiently, reaching larger sizes as with the ice crystals. In some temperature regimes, the snow mass mixing ratio is even enhanced relative to the control case, signifying more efficient precipitation development. This outcome supports the interpretation that the high-INP regime promotes a shift toward fewer but more massive snow particles, enhancing cold-rain production from melting of ice precipitation (Fig. ), in line with the findings of .

Figure a shows that in the control case, where homogeneous nucleation dominates, surface precipitation occurs in intermittent bursts with moderate peak intensities. By contrast, the high-INP case has more frequent and stronger peaks in surface precipitation rate. This outcome arises from earlier activation of heterogeneous nucleation, which limits homogeneous ice formation but favors the rapid growth of fewer ice crystals into precipitation-sized hydrometeors. The cumulative precipitation (mm) in Fig. b further highlights this difference. Both simulations initiate precipitation at nearly the same time, yet the high-INP case produces considerably more accumulated surface precipitation. This is consistent with a microphysical regime where fewer but larger ice and snow particles grow more efficiently, ultimately enhancing cold-rain production through melting, as also noted in . Figure c–d show the radiative implications of these changes at the top of the atmosphere (TOA). The net shortwave flux (Fig. c) is slightly reduced in the high-INP case, reflecting enhanced cloud shading due to more persistent condensate and precipitation. Meanwhile, the net longwave flux (Fig. d) reveals systematically lower outgoing fluxes at TOA for the high-INP simulation, consistent with deeper, longer-lived cloud layers that trap infrared radiation, partially re-emitting it at a colder temperature aloft to space. Taken together, Fig.  demonstrates that the increase in INP concentrations not only modifies the microphysical pathways of ice initiation – shifting from homogeneous to heterogeneous nucleation – but also amplifies precipitation production and alters the radiation budget at the TOA.

In summary, the higher numbers of INPs act to shift the dominant pathway of ice initiation from homogeneous to heterogeneous ice nucleation in the simulations, as expected. This occurs by their effect on lowering the humidity aloft. Such changes in nucleation pathways not only influence the number of hydrometeors but also their mass, with significant implications for precipitation.