CESM2 features the Community Atmosphere Model version 6 (CAM6) as its atmosphere model. CAM6 uses a two-moment stratiform cloud microphysics scheme MG2 with prognostic liquid, ice, rain, and snow hydrometeor classes. As a two-moment bulk scheme, MG2 has prognostic variables for number- and mass-mixing ratios for each of the hydrometeor classes. It also permits ice supersaturation and is described further below.

MG2 is also coupled to a four-mode aerosol model with liquid activation, following , and features natural and anthropogenic aerosols. CAM6 includes a physically based mixed-phase ice nucleation scheme and accounts for preexisting ice for nucleation . MG2 is coupled to a unified moist turbulence scheme, Cloud Layers Unified By Binormals CLUBB;, implemented in CAM by . CLUBB treats boundary layer moist turbulence, stratiform clouds, and shallow cumulus. CAM6 uses an ensemble plume mass flux deep convection scheme with simple microphysics. The radiation scheme is the rapid radiative transfer model for general circulation (RRTMG) models .

MG2 is an update of the original MG scheme . The original MG scheme was implemented in CAM5, as described by . MG2 added prognostic rain and snow and is used in CAM6. Many of the process rate treatments are similar to the bulk two-moment microphysics scheme of developed for mesoscale models. further updated the MG2 scheme to MG3 by adding rimed ice (with a switch to represent either graupel or hail). Thus, MG3 includes two additional prognostic variables for graupel/hail number- and mass-mixing ratios. In a separate work, and updated the scheme to use a single category for cloud ice and snow. This has not been implemented yet in MG2 or MG3, and the integration is a subject for future work.

Versions of MG microphysics have been adapted for use in the NASA GEOS Goddard Earth Observing System with an integrated Earth system model;, the Geophysical Fluid Dynamics Laboratory (GFDL) Atmosphere Model AM;, and the NASA Goddard Institute for Space Studies (GISS) model . The scheme is also available in the idealized kinematic driver KID;. MG1 and MG2 are also used in a suite of Earth system models derived from CESM1 and CESM2. Analyses have been conducted against observations for sub-grid-process rate formulations and against in situ aircraft flights for supercooled liquid water . There has been significant work on numerical time-stepping issues detailed by and .

There are now a large number of people contributing to the community development of the MG scheme. Thus, the MG name is no longer appropriate, and we are now naming the scheme the Parameterization for Unified Microphysics Across Scales (PUMAS). This work describes the updates and changes from MG3 to PUMAS (version 1).

As part of this update, we have tested and implemented several science and software changes to PUMAS beyond the MG3 scheme . These changes are as follows:

refactoring the ice number limiter,

We conduct several 6-year-long simulations using near-present-day cyclic boundary conditions for the year 2000. Boundary conditions used to force the model are climatological monthly averages over 1995–2000, including greenhouse gases, atmospheric oxidants, and emissions of aerosols. The climatological annual cycle is repeated each year. We also use averaged sea surface temperatures (SSTs) from observations . Data are compiled monthly and smoothly interpolated between months. Model resolution for CAM6 is 0.9∘ latitude ×1.25∘ longitude, with 32 levels in the vertical up to 10 hPa, a standard physics time step of 1800 s, and three coupling periods between microphysics and cloud macrophysics (turbulence and large-scale condensation), for a default microphysics time step of 600 s. A total of 6 years is sufficient so that the differences between past, present, and future configurations (see below) are much larger than the interannual standard deviation of global quantities (e.g., the signals are larger than the internal variability).

For each of the cases described below in Sect. , we conduct the following experiments. The first is a simulation with present-day (PD; year 2000) boundary conditions including SSTs, greenhouse gases (GHGs), and aerosol emissions. Next, we conduct simulations with (a) 1850 estimated aerosol emissions, termed preindustrial (PI), and (b) SSTs increased uniformly by +4 K (SST4K). All other boundary conditions (especially GHGs) are the same. When compared with the PD simulation, the PI assesses the impact of aerosol forcing, and SST4K assesses the impact of cloud feedbacks, which are representative of the full model response to warming, as first noted by and verified for CESM2 by .

For a detailed analysis, we also conduct single-column model simulations with the Single Column Atmosphere Model (SCAM) that is available as a part of CAM . SCAM runs exactly the same atmospheric physics parameterization code as the full 3D simulations. We focus on the Atmospheric Radiation Measurement (ARM) program case from June 1997 for a month over the Central Great Plains (Oklahoma, USA), called the ARM97 case, in addition to the Mixed-Phase Arctic Cloud Experiment (M-PACE) case from October 2004 over the North Slope Borough of Alaska. The cases are detailed in . The standard time step in SCAM is 1200 s, with four microphysics sub-steps for a microphysics time step of 300 s.

Sensitivity tests are grouped sequentially, as indicated in Table , with the testing many of the items listed in Sect.  and with the item numbers also listed in that section. The first is a control case, using the standard CAM6 and the MG3 microphysics. All of these tests include graupel. Next, we test a PUMAS case with modifications nos. 1–7 listed in Sect. , except for no. 8 (optional switches to replicate processes in the ECMWF IFS). This includes (1) refactoring the ice number limiter, (2) using a reduced aerosol number for ice nucleation, (3) allowing the vapor depositional growth of snow, (4) applying fall speed corrections, (5) noting implicit sedimentation, (6) seeing newly autoconverted rain with accretion, and (7) adding ACC directives. An additional modification is to increase the specified diameter of ice crystals detrained from deep convection (from 25 to 60 µm; Control-Di60). We test the impact of this change in a separate simulation based on the control case. The PUMAS simulation in which we use the CNT immersion freezing without a reduced aerosol number to act as INPs (no. 2) is illustrated in Fig.  (PUMAS-Imm). We then look at the impact of removing vapor deposition onto snow (no. 3) in PUMAS, as described in Sect.  (PUMAS-VapDepSn). Next, we test removing the fall speed modifications (no. 4) and implicit sedimentation (no. 5) in PUMAS simulations in the PUMAS-Fall simulations. Because the fall speed modifications mainly impact the explicit rather than implicit treatment of sedimentation, we also tested the fall speed modifications along with the explicit sedimentation in PUMAS (e.g., testing explicit sedimentation with the fall speed correction). The results are basically the same as for the explicit sedimentation without the fall speed correction in PUMAS-Fall (not shown), indicating that the fall speed correction has little impact. Finally, we also explore a simulation in which we adjust (tune) uncertain microphysical parameters to increase cloud water and ice to come close to the control case (PUMAS-Tune). This tuning is accomplished by scaling the autoconversion of cloud to rain by a factor of 0.5 from the control case and increasing the threshold diameter for ice autoconversion to 1000 µm (from the control value of 500 µm).

To look in detail at the effects of changes to the microphysics scheme, we focus on SCAM simulations for the ARM97 summer case. Figure  illustrates liquid (Fig. a–c), ice (Fig. d–f), rain (Fig. g–i), and snow (Fig. j–l) process rates. Graupel was included in these simulations but is not shown and discussed briefly below. Figure a, d, g, and j (left column) are the control simulation, the center column (Fig. b, e, h, k) is the PUMAS modified code (changes nos. 1–7 from Sect. ), and the right column (Fig. c, f, i, l) is the difference between the two. Differences in liquid at pressures less than 600 hPa mainly come from changes in ice and mixed-phase processes (vapor deposition). In particular, there is less accretion of liquid onto snow (blue dotted–dashed line) and vapor deposition onto snow (Bergeron snow; solid teal line), reducing the liquid loss rates at higher altitudes. At lower altitudes (pressures >600 hPa), there are small changes to liquid, with less liquid sedimentation (solid blue line) and more loss from the accretion of cloud by rain (accretion; dashed turquoise line).

Ice microphysics sees larger changes, with a decrease in ice mass loss from sedimentation (solid blue line), leading to a positive net change. This is caused by a larger ice number that comes from removing the NIMAX limiter and results in more ice mass. Cloud ice remaining longer in the atmosphere increases the availability for sublimation and deposition (dotted–dashed red line) as well.

Precipitation changes in Fig.  are also significant. Rain has a negative sedimentation tendency (solid blue line) all the way to the surface in the revised code (Fig. h), with reduced fall speeds higher up. Snow has a positive sedimentation tendency (solid blue line) from 800–600 hPa, as snow falling from above enters a layer below. In PUMAS, this continues all the way to the surface (Fig. k) because the nonzero fall speed was added below the lowest level of precipitation. The implicit sedimentation does not seem to introduce any numerical issues with rain or snow.

Graupel changes (not shown) do not significantly affect snow or other hydrometeors and come mostly from differences in the graupel-collecting snow term (dashed turquoise line), which is larger (more negative) in PUMAS (Fig. l).

We have explored in detail how the different parts of the PUMAS code create these differences, by plotting some of the changes separately (not shown). The upper-level increases in cloud ice deposition and sublimation come from the ice number limiter change. The limiter also reduces the low-level accretion of cloud liquid (qc) to rain (qr) and snow (qs). The implicit sedimentation results in more accretion of cloud liquid. The new accretion increases the accretion of cloud liquid as well but reduces cloud ice deposition/sublimation. Thus, not all the changes act in the same way. The changes due to the accretion, including the newly autoconverted rain, have a smaller impact (limited to increasing low-level accretion) than the explicit sedimentation or ice limiter changes.

We have also used the SCAM cases as a way to explore the time step sensitivity of the schemes. To focus on the microphysics, we set a 1200 s physics time step in SCAM, with four sub-steps for the CLUBB-unified turbulence scheme. We then can sub-step the microphysics within the 300 s sub-step. This takes the same condensation from turbulence and passes it to the microphysics, which we run from 300 to 10 s (e.g., 1 to 30 microphysics sub-steps). The results are illustrated in Fig.  for the June 1997 (ARM97) case. We focus on the control case (MG3), all the PUMAS modifications (PUMAS), and the PUMAS without the implicit sedimentation and fall speed corrections (PUMAS-Fall). Other sensitivity cases are similar to PUMAS. Figure  illustrates the same tests but for the Mixed-Phase Arctic Cloud Experiment (M-PACE) in October 2004 over the North Slope Borough of Alaska.

Several results stand out. First, cloud liquid and rain (Fig. a and d) are fairly sensitive to time step changes for the SCAM ARM97 case in the control (MG3) configuration. The same is true for supercooled liquid (Fig. c). The PUMAS code reduces the time step sensitivity of most of the hydrometeors. This is particularly true for the M-PACE case in Fig. , where cloud (Fig. a), supercooled liquid (Fig. c), and cloud ice (Fig. b) have much less variability across time steps than in the control MG3 case. Most of this improvement is not seen if the fall speed and sedimentation changes are removed, indicating that implicit sedimentation contributes substantially to the reduced sensitivity of solutions to time step (as expected). Figure b and c indicate reductions in supercooled liquid water and ice with the microphysical changes in PUMAS.

Impacts of the implicit sedimentation and fall speed changes are seen in the rain mass (Fig. d). The peak rain mass for PUMAS in the ARM97 case is less variable in PUMAS than in the control or in PUMAS without implicit sedimentation. However, there is a reduced gradient below the peak for PUMAS, which may be caused by the more diffusive nature of the formulation. This is not an issue in the M-PACE case (Fig. d). It is not seen to be a problem in the 3D simulations.

The M-PACE case (Fig. ) has less sensitivity to the time step than ARM97 for the control MG3 case and even less sensitivity for PUMAS. Reduced sensitivity in all cases may have to do with the absence of deep convective cloud and deeper cloud in the M-PACE, while such cloud is present throughout the column in ARM97. Cloud liquid (all supercooled; Fig. a and c) has little sensitivity in PUMAS, less than with MG3 or PUMAS-Fall, again indicating the importance of implicit sedimentation. There is less cloud ice (Fig. b) but more snow in the PUMAS simulations compared to MG3. Overall, besides the reduced time step sensitivity, PUMAS and PUMAS-Fall give similar results for these SCAM cases.

Figure  illustrates the zonal mean climatologies from 6-year global simulations of CAM6 with control microphysics (MG3) and updated PUMAS code. Also shown are observations of the radiative fluxes and clouds from the CERES (Clouds and the Earth's Radiant Energy System) satellite Energy Balanced and Filled (EBAF) products , liquid water path (LWP) from the CERES SYN product , and precipitation from the Global Precipitation Climatology Project (GPCP; ). We test the code following the experiments in Table . PUMAS (blue) is the full set of modifications. PUMAS-VapDepSn (orange) removes the vapor deposition onto snow, PUMAS-Fall (green) removes the fall speed changes and uses an explicit rather than implicit sedimentation calculation, and PUMAS-Tune (red) increases liquid and ice mass with parameter changes. The full PUMAS ice modifications (blue) reduce the overall cloud radiative effect (CRE) magnitude (less negative shortwave (SW) and less positive longwave (LW) effect) from the control MG3 case by about 2.5 Wm-2 (Fig. g and h). The difference tends to reduce SW biases but increase LW biases with respect to CERES. This is due to the lower LWP (Fig. a) and ice water path (IWP; Fig. b). Cloud fraction, cloud-top liquid number, and precipitation (Fig. c, d and f) are similar across all simulations. Note that the satellite data for LWP (Fig. a) and cloud fraction (Fig. c; based on CERES and MODIS clouds) are not reliable at high southern latitudes over the ice sheet. It is clear that the highest water contents are not over the South Pole. The PUMAS-Tune case is designed to increase liquid and ice (Fig. a and b) in order to increase the magnitude of the cloud radiative effects so that they are more similar to the control (Fig. g and h), but PUMAS-Tune also decreases the ice number (Fig. e). The resulting cloud radiative effects (Fig. g and h) are of a lower magnitude than CERES LW observations but generally within the variability in the observations (purple shading). Note that the change to detrained ice size (included in all PUMAS simulations) is tested on its own with the control model (Control-Di60), and it increases LWP (Fig. a) and decreases IWP (Fig. b), likely through the faster sedimentation of larger ice crystals. Also note that the revised immersion freezing (present in all the PUMAS simulations) have higher ice numbers (Fig. e) from 30–60∘ N latitude.

Figure  illustrates the ice fraction in the control and PUMAS simulations. The ice fraction is calculated at each location and time step as the ice mass-mixing ratio divided by the total cloud water (ice and liquid) mass-mixing ratio. An increase in ice fraction reduces the CRE from supercooled liquid water but increases CRE from ice. Cloud ice provides a negative cloud-phase feedback when it is reduced under climate warming. The detrained ice size change does not impact the ice fraction in the control simulation (Fig. b). PUMAS features an increase in the ice fraction in the tropical middle troposphere (600–300 hPa) relative to the MG3 control simulation (Fig. c). It appears that the implicit sedimentation change is the reason, as removing the implicit sedimentation and the fall speed correction (Fig. d) produces ice fractions that resemble the control case. Our tests showed that adding the fall speed correction to explicit sedimentation does not matter (not shown). The sedimentation likely impacts the evolution of ice near the freezing level. There are also some impacts at high latitudes. The tuning of clouds to increase liquid and ice water path modifies the ice fraction due to substantial differences in the liquid and ice water path (Fig. ), generally resulting in more moderate changes to the ice fraction relative to the control than with the standard PUMAS (with less condensate mass; Fig. c).

Cloud feedback is determined using a kernel-adjusted cloud radiative forcing, as described by and implemented in CESM by also used by.

Figure  illustrates the kernel-adjusted cloud feedback estimates, weighted by the cosine of latitude (so that the area under the curve is the contribution to the global mean). Table  presents the global averages. PUMAS slightly increases the cloud feedback, lowering the LW and increasing the SW. Some of this results from the change to the detrained ice size (Control-Di60) and not PUMAS itself. The PUMAS-Fall SW feedbacks are very close to the Control-Di60 case, indicating that the differences in SW feedbacks are perhaps due to the implicit sedimentation. The change to the fall speed was tested with explicit sedimentation and does not impact feedbacks. The mechanism might be through the changes to the ice phase seen in Fig. . The net cloud feedback (Fig. c) is reduced from 30–70∘ latitude in both hemispheres but offset with changes at higher latitudes. The PUMAS-Tune simulation with enhanced cloud water has larger SW cloud feedbacks, from 30∘ S–30∘ N, but there was some offsetting of lower LW feedbacks. This is likely due to the significantly larger ice mass changing high (cirrus) cloud feedbacks, as LW feedbacks are lower in the tropics in regions of higher ice mass and number, from 20–40∘ N (Fig. ).

We have examined maps of cloud feedback by regime for the LW and SW feedbacks, and there do not appear to be large and significant systematic differences in cloud feedback between the control and PUMAS simulations. On the whole, the impact of PUMAS on cloud feedbacks is moderate and not significant.

Finally, we assess the impact of the changes in cloud microphysics on the aerosol forcing of climate by running 6-year simulations identical to the present-day (PD) simulations but with the aerosol emissions set to 1850 emissions, i.e., the preindustrial (PI) era. The difference in PD–PI represents the impact of aerosols on climate. The indirect effect of aerosols on clouds is illustrated by the change in the cloud radiative effect (ΔCRE). For liquid clouds, this is mostly in the SW (ΔSWCRE), and for ice clouds, this is both the SW and LW (ΔLWCRE). The direct radiative effect of aerosols is the clear-sky shortwave flux change (ΔSWclr). The total effect is the all-sky residual top-of-model (ΔRESTOM) flux change. Changes in the column drop number (ΔCDNC) and liquid water path (ΔLWP) and ice water path (ΔIWP) are also assessed, which drive changes in CRE for clouds with liquid. For all quantities, the global average PD–PI differences are larger than twice the interannual standard deviation of global means of these quantities.

The CAM6 total radiative effect due to aerosols (ΔRESTOM) is on the high end of the range assessed by . Overall, the net aerosol forcing in the present day (PD–PI) defined by ΔRESTOM drops by less than 10 % with the PUMAS modifications (Table ). This occurs due to small reductions in the negative SW clear-sky forcing (direct effects of aerosols) and small reductions in the SW cloud radiative effect changes but is partially offset by reductions in the positive LW forcing as well. Thus, the magnitude of the components is slightly reduced. The PD–PI percent change in LWP is lower with PUMAS than the control (Table ). The cloud drop number increases by 30 %, while the cloud fraction increases by less than 1 %. The simulations are similar for the different sensitivity tests, but this masks regional differences.

Figure  illustrates the zonal mean changes in key quantities between pairs of PD and PI simulations. The gross aerosol forcing of climate in the SW (Fig. a) and LW (Fig. b) is similar in most of the PUMAS simulations compared to the control, where SW forcing is negative and becomes less so in the Northern Hemisphere. LW forcing is reduced in all three PUMAS configurations and the PUMAS-Tune simulation. The residual net change in the top-of-model (TOM) flux (Fig. c) has a lower magnitude (less negative) than the control in Northern Hemisphere high latitudes for the PUMAS-Tune simulations (Fig. c). There is a significant reduction in the LWP and column cloud drop number (CDNUMC) response from 25–90∘ N latitude (Fig. d and f), which drives the reductions in the magnitude of the SW CRE changes (Fig. a). Some of the difference is due to the smaller mean values. The percent change in LWP is reduced in PUMAS simulations (Table ), but the percent change in drop number is similar, except for the PUMAS-Tune simulation.

Since the forcing response of most of the PUMAS simulations is similar, Fig.  compares maps of the patterns of differences (PD–PI) for the control (left column) and PUMAS (right column) simulations. Maps of the pattern of differences in Fig.  show that these changes occur in most places across the globe. SW CRE changes (Fig. a and b) are reduced with PUMAS slightly throughout the Northern Hemisphere and even into the Southern Hemisphere. There are some positive values over northeastern Asia. Changes are similar over the Pacific, with large impacts in the eastern Pacific that are a common feature of CAM aerosol–cloud interaction (ACI) effects but lower in PUMAS. The LW CRE changes (Fig. c and d) have more positive values (Fig. d) than the control simulation (Fig. c), especially over the Indian Ocean, which shifted from the eastern Pacific, with little net effect. The LWP changes (Fig. g and h) are smaller with PUMAS and localized closer to the source regions compared to the control. This is true especially in the Arctic (Fig. h). Some of this is due to lower mean values, but the percent change in LWP is smaller (Table ), and the high-latitude LWP is reduced in PUMAS (Fig. a).