The impact of aerosols on ice- and mixed-phase processes in deep convective clouds remains highly uncertain, and the wide range of interacting microphysical processes is still poorly understood. To understand these processes, we analyse diagnostic output of all individual microphysical process rates for two bulk microphysics schemes in the Weather and Research Forecasting model (WRF). We investigate the response of individual processes to changes in aerosol conditions and the propagation of perturbations through the microphysics all the way to the macrophysical development of the convective clouds. We perform simulations for two different cases of idealised supercells using two double-moment bulk microphysics schemes and a bin microphysics scheme. The simulations cover a comprehensive range of values for cloud droplet number concentration (CDNC) and cloud condensation nuclei (CCN) concentration as a proxy for aerosol effects on convective clouds. We have developed a new cloud tracking algorithm to analyse the morphology and time evolution of individually tracked convective cells in the simulations and their response to the aerosol perturbations.
This analysis confirms an expected decrease in warm rain formation processes due to autoconversion and accretion for more polluted conditions. There is no evidence of a significant increase in the total amount of latent heat, as changes to the individual components of the integrated latent heating in the cloud compensate each other. The latent heating from freezing and riming processes is shifted to a higher altitude in the cloud, but there is no significant change to the integrated latent heat from freezing. Different choices in the treatment of deposition and sublimation processes between the microphysics schemes lead to strong differences including feedbacks onto condensation and evaporation. These changes in the microphysical processes explain some of the response in cloud mass and the altitude of the cloud centre of gravity. However, there remain some contrasts in the development of the bulk cloud parameters between the microphysics schemes and the two simulated cases.
Deep convective clouds are an important feature of the Earth's atmosphere,
ranging from widespread convection dominating the atmosphere in the tropics
to mid-latitude convective systems
Over recent years numerous studies using cloud-resolving model simulations
(CRM) have investigated aerosol–convection interactions in various set-ups,
ranging from case study simulations to idealised simulations of squall lines
or supercells like the cases used in this study
Convective invigoration
Many studies have pointed out the representation of cloud microphysics in
models as one of the main sources of uncertainty in high-resolution model
studies of aerosol–cloud interactions or cloud feedbacks to a warming
climate, especially for mixed-phase and ice-phase clouds
Most currently used cloud microphysics schemes can be separated into two
approaches, bulk microphysics schemes and bin microphysics schemes
The separation of the hydrometeors into individual hydrometeor classes in
microphysics schemes brings with it specific challenges in resolving the
microphysical processes. In bulk schemes, liquid water in the cloud is
separated into cloud droplets and raindrops. The collision–coalescence
processes leading to the formation of rain from cloud droplets have to be
parameterised through the artificial process of droplet autoconversion and a
simplified treatment of accretion of droplets by raindrops. The
semi-empirical nature of these parameterisations has been shown to be the
source of major uncertainty in the assessment of aerosol–cloud interactions
in numerical model simulations
Bin microphysics schemes represent the different hydrometeors in the cloud
through a number of individual size bins per hydrometeor class, thus allowing
for more flexible representation of the actual size distribution and the
interaction between the different size bins
This study aims to unravel the underlying microphysical mechanisms responsible for the large diversity of simulated aerosol effects on convection through a comprehensive analysis of the propagation of aerosol perturbations through microphysical pathways in different microphysics schemes.
Tracking individual convective cells in the simulation makes it possible to
draw direct conclusions about the behaviour of individual convective cells in
the simulations, e.g. regarding their time evolution or the response to
changes in simulation parameters that go beyond the bulk average over the
simulation domain or the sum of all cloudy areas in the simulation. The
analysis of tracked cumulus clouds has been applied in previous studies
We have implemented detailed microphysical process-rate diagnostics for
pathway analysis in the two double-moment microphysics schemes of
In addition to the detailed process-rate diagnostics, we derive important
bulk cloud properties, such as the total cloud mass or the altitude of the
centre of gravity, and analyse their evolution over the life cycle of the
tracked cells. Our approach goes beyond previous studies with a similar
set-up
We use a well-documented idealised supercell set-up based on
We represent idealised aerosol perturbations through changes to a fixed cloud
droplet number concentration (CDNC) in each simulation with the two bulk
microphysics schemes. This allows us to isolate the actual cloud
microphysical pathways from uncertainties in the representation of the
activation of CCN in numerical models
We compare the results to simulations performed with a bin microphysics scheme (HUJI spectral-bin scheme) for a subset of the analyses to investigate whether the effects investigated in more detail through the microphysical pathway analysis for the two bulk microphysics schemes agree with the response of a bin microphysics scheme to perturbations of aerosol proxies.
The simulations are performed with the Weather and Research Forecasting model
(WRF) version 3.7.1
The detailed analyses of the process rates in this paper are carried out for
simulations using the two bulk microphysics schemes. To investigate how the
results obtained from the detailed analysis of the two bulk microphysics
schemes hold for a bin cloud microphysics scheme, we also include additional
simulations with the Hebrew University cloud model (HUCM) spectral-bin
microphysics scheme
Both bulk microphysics schemes make use of saturation adjustment, removing
all water vapour exceeding the saturation vapour pressure in each time step
and instantaneously condensing it to cloud water at each time step. This
prevents a build-up of supersaturation in strong updrafts and can thus impact
effects of perturbations in the microphysics
We simulate two different idealised supercell cases. The first set of
simulations (CASE1) is based on the default WRF quarter-circle shear
supercell case representative of a supercell case over the southern Great
Plains of the United States
To test the representativeness of the results for different cases of
idealised deep convection, a set of simulations for a second supercell case
(CASE2) is based on an observed supercell storm over Oklahoma in 2008
Both cases are simulated without a boundary layer scheme and the calculation
of surface fluxes or radiation. The horizontal grid spacing of the
simulations is 1
We analyse the effects of varying the CDNC in the two bulk microphysics
schemes to isolate the impact of microphysical pathways. We use a CDNC of
250
For the simulations with the spectral-bin microphysics scheme, activation of
aerosols to cloud droplets is calculated from a CCN spectrum following the
equation
Overview of the 52 simulations performed in this study, including
the two cases simulated and the different CDNC/CCN values for each of the
microphysics schemes. The CDNCs for the SBM simulations are the median values
for grid points with a cloud water mixing ratio larger than
10
We have extended two double-moment bulk microphysics schemes, the Morrison
scheme
For most analyses in this study, the individual microphysical processes are
grouped into a consistent set of classes according to their contribution to
the hydrometeor mass transfer in the model. This includes the six different
phase transitions between frozen hydrometeors, water drops and water vapour
(
We have developed a tracking algorithm focussed on the tracking of individual
deep convective cells in CRM simulations, but flexible enough to be extended
to other applications, e.g. simulations of shallow convection or based on
geostationary satellite observations using brightness temperature data. The
initial tracking of features is performed on the column maximum vertical
velocity at each output time step using Python tracking library trackpy
Based on these trajectories, a three-dimensional watershedding algorithm,
A separate watershedding is performed for both liquid water content (cloud
droplets and rain drops) and ice water content (all ice hydrometeors). This
allows for the determination of the centre of gravity and the mass, for the
entire cloud as well as for the in-cloud liquid and frozen phases,
respectively. The evolution of the centre of gravity has been studied mainly
for warm convective clouds
The tracking algorithm does not explicitly treat splitting and merging of convective cells. In all simulated cases in this study, the initial convective cell splits into two separate counter-rotating cells early into the simulations. In CASE1 this leads to a relatively symmetric situation with similarly strong individual cells. In both cases, one of the cells develops more directly out of the initial cell: in CASE1 this is the right-moving cell, while in CASE2 this is the stronger left-moving cell. In each simulation, this stronger cell gets picked up as a continuation of the initial cell by the tracking algorithm. The second cell has been analysed following the same methodology and showed very similar results in all major aspects. We have thus decided to focus on the analysis of the first cell in this paper and to not discuss the results from the second cell in more detail.
Microphysical process rates, latent heating rates and other cloud
microphysical parameters such as hydrometeor mixing ratios are summed up for
regularly spaced altitude intervals in the volume of the individual cells to
get representative profiles for each cloud. We interpolate the microphysical
process rates and other variables used in the analysis to slices along and
perpendicular to the line of travel of the cell
(Fig.
Cloud microphysical morphology along a slice parallel
to the cell track for a cloud droplet number concentration of
250
Cloud microphysical morphology along a slice parallel
to the cell track for a cloud droplet number concentration of
250
The simulations with CDNC
During the initial phase of the formation of the convective cloud in the
simulation using the Morrison bulk microphysics scheme
(Fig.
Cloud microphysical morphology along a slice through the
cloud parallel to the track of the cell for simulations with three
different CDNC values (
We first investigate changes to the right-moving cell in CASE1 due to a
variation of CDNC. We focus on three different CDNC values (clean, baseline,
polluted; see Fig.
In the hydrometeor mass mixing ratios
(Fig.
Figure
Hydrometeor mass mixing ratios in a slice along the line of travel
of the cell for the cleanest
Time evolution of the microphysical process rates for the
cleanest
A more detailed analysis of the processes involved in the formation of rain
over the lifetime of the cell in the different cases
(Fig.
The processes transforming liquid to frozen water can be further broken down
into processes representing the freezing of individual cloud droplets or
raindrops and riming processes, in which existing ice-phase hydrometeors
accrete liquid water (Fig.
The riming processes are spread out over a much larger altitude range in the
cloud, between the melting level at about 4
The evolution of the deposition and sublimation processes
(Fig.
In the simulations with the Thompson microphysics scheme
(Fig.
Time evolution of the microphysical process rates
relevant for rain formation processes (autoconversion, accretion
of cloud droplets by rain and melting of ice hydrometeors) as in
Fig.
Time evolution of the microphysical process rates
of freezing and riming processes as in Fig.
Latent heating constitutes a key feedback of the microphysics scheme onto the
model dynamics along with changes to the buoyancy due to changes in
condensate loading. The vertically resolved latent heating over the lifetime
of the tracked cell in CASE1 is shown in
Fig.
Latent heat release from condensation is the dominant contribution to the
latent heating and about a magnitude stronger than the other contributions,
thus determining the general shape of the latent heating profile
(Figs.
The same limitation applies to the evaporation of cloud droplets, which also
cannot show any direct effect from changes in CDNC due to the use of
saturation adjustment. However, the evaporation shows much stronger
differences between the two microphysics schemes and also a stronger effect
of a variation in CDNC (Fig.
Time evolution of the microphysical process rates of deposition
and sublimation as in Fig.
All three microphysics schemes show a small shift of latent heating to higher
altitudes superimposed on that in the range between 7
There are large differences between the microphysics schemes in the latent
heating and cooling from sublimation and deposition and their response to
changes in CDNC. The Morrison scheme shows a significant decrease in both
sublimation and deposition with increased CDNC
(Fig.
In the Thompson scheme, sublimation of ice hydrometeors is weak and barely
affected by changes in CDNC (Fig.
In contrast to the increased latent heating from freezing or melting, changes in condensation and evaporation, as well as in sublimation and deposition, are linked to a change in condensate loading, which affects the buoyancy of the cloud and thus at least partially buffers the impact of latent heating and cooling on the dynamics of the clouds.
The changes to the vertically integrated latent heating in the cloud for all
three microphysics schemes do not show a significant trend with increasing
CDNC (Fig.
Profiles of the sum of latent heating over the lifetime of the dominant tracked cell for the three microphysics schemes in CASE1.
Profiles of the components of the latent heating and cooling over the lifetime of the tracked cell for the two bulk microphysics schemes in CASE1.
Total water mass, liquid water mass and frozen
water mass in the analysed right-moving cell for the three different
microphysics schemes (Morrison:
The tracking and watershedding allow for a determination of the cloud mass inside the identified cloud volumes and the centre of gravity of the hydrometeors in the cloud. These analyses are also performed separately for the liquid-phase and ice-phase hydrometeors in the cloud, which allows us to relate the changes in the properties for the entire cloud to changes in the individual phases.
The evolution of the cloud mass and the mass of the two water phases in the
cloud (Fig.
Altitude of the centre of gravity of the cloud and the individual phases
in the analysed right-moving cell for the three different microphysics schemes (Morrison:
There are, however, marked differences in the response to changes of the
aerosol proxy between the microphysics schemes. The Morrison scheme shows a
decrease in total cloud mass and ice-phase mass by about 10 %–15% over
the range in which we increase the CDNC and no significant changes in the
liquid phase. This decrease in ice-phase mass can be directly linked to the
changes in the microphysical process rates analysed in
Sect.
The altitude of the centre of gravity is affected by the choice of
microphysics scheme, with an overall higher centre of gravity for the SBM
scheme (Fig.
There is a consistent response in the cloud heights for all three
microphysics schemes. The microphysics schemes show an increase in the height
of the centre of gravity of the entire cloud, which is more pronounced using
the Thompson scheme (about 1.5
All three microphysics schemes show a clear saturation in the effect of
changes in the CDNC/CCN concentration. Variations above 2000
Temporal evolution of the microphysical process rates
in CASE2 for the cleanest
Total water mass, liquid water mass and frozen water mass in the
analysed left-moving cell for the three different microphysics schemes
(Morrison:
Altitude of the centre of gravity of the cloud and the individual
phases in the analysed left-moving cell for the three different
microphysics schemes (Morrison:
To investigate the representativeness of the results and the response of the
deep convective clouds to the variation of aerosol proxies CDNC and CCN, the
same set of simulations and analyses have been performed for a second
idealised supercell case (CASE2) with different forcing and initial
conditions (Sect.
The time evolution of the cloud-averaged process rates for the two bulk
microphysics schemes (Fig.
Despite these differences in the evolution, CASE2 shows very similar changes in the microphysical processes due to a variation of CDNC to CASE1 for both microphysics schemes. The formation of rain due to autoconversion of cloud droplets and accretion by rain is smaller and restricted to lower heights in the polluted case using the Morrison microphysics scheme. For the Thompson microphysics scheme, the formation of rain is decreased and shifted to higher levels in the model under polluted conditions. Furthermore, the freezing and riming processes predominantly occur at higher altitudes than in the clean case for both bulk microphysics schemes.
In line with these changes to the microphysical process rates, the evolution
of the cloud mass in CASE2 (Fig.
The effects of a variation of CDNC are quite similar to the ones seen in
CASE1 for the two bulk microphysics schemes
(Fig.
The changes in the altitude of the centre of gravity show less clear
relationships with changes in the aerosol proxies CDNC/CCN in this case for
the two bulk microphysics schemes. The Morrison scheme
(Fig.
We investigated the effects of changes in cloud droplet number concentration
(CDNC) and cloud condensation nuclei (CCN) concentrations on the development
of idealised simulations of deep convection to test proposed aerosol effects.
This includes different mechanisms of convective invigoration
An increase in cloud droplet number concentration from values representing
clean conditions (CDNC
A more detailed analysis of the different components of the latent heating
for the two bulk microphysics schemes shows a complex superposition of
changes to the different phase changes in the tracked cells. This confirms
results from previous studies on the effects of aerosols on supercells
There are significant differences between the two bulk schemes in the profiles of sublimation and deposition as well as in the response of these processes to changes in CDNC. This can be attributed to different parameter choices in the schemes. The strongest differences result from the fact that deposition onto graupel hydrometeors is not allowed to occur in the Thompson microphysics scheme, which leads to a strong increase in deposition due to the replacement of graupel by the other ice-phase hydrometeors on which deposition occurs. This strong increase in deposition additionally drives changes in condensation and evaporation in the mixed-phase region of the cloud via the Wegener–Bergeron–Findeisen process. By effectively removing water vapour, this leads to a noticeable feedback on the evaporation and condensation on cloud droplets that are intrinsically not affected by changes in CDNC because of the use of saturation adjustment. It was also shown that the melting of frozen hydrometeors contributes significantly to the formation of raindrops, especially under high CDNC conditions, which forms an additional important feedback of changes in the ice-phase onto the warm-phase processes.
The changes to the individual components of integrated latent heating in the cloud due to a variation of CDNC compensate each other in the two bulk microphysics schemes. Hence, there is no significant change in the total integrated latent heating in the cloud with changes in CDNC/CCN and no thermodynamic invigoration from changes in the microphysics due to the change in the aerosol proxies. This result is confirmed in the SBM simulations, that also do not show any significant change in vertically integrated latent heating for a variation of CCN. Therefore, the absence of convective invigoration in the bulk microphysics schemes cannot be solely attributed to the application of saturation adjustment.
The analysis of the clouds with respect to the total cloud mass and the
altitude of the centre of gravity showed some contrasting results between the
different microphysics schemes. There is a clear signal of a lifting of all
parts of the clouds to higher altitude under polluted conditions, probably
associated with the changes in the ice-phase hydrometeor partition. This
agrees with findings from, e.g.
The results for the first case (CASE1), based on
The pathway analysis developed for this study also includes the process rates
for the number concentrations of the different hydrometeors. This includes
processes like ice multiplication that could play an important role in better
understanding some of the possible pathways of aerosol effects on convective
clouds
This work focused on the analysis of microphysical pathways of aerosol
effects on deep convective clouds in an idealised framework. To test the
robustness of the results under realistic scenarios, including potential
buffering mechanisms, we are currently applying our analysis framework to
large case study simulations of isolated convection over the area around
Houston, Texas, as part of the ACPC initiative (Aerosol, Cloud,
Precipitation, and Climate Working Group,
The understanding of the detailed structure of microphysical processes in
individually tracked deep convective clouds and the analysis of the pathways
through which aerosol perturbations affect the deep convective clouds advance
our understanding of aerosol–cloud interactions. This can be used to inform
the parameterisation of microphysical processes and aerosol–convection
interactions in global climate models. Recent developments in the use of
global cloud-resolving models in climate research
The WRF model is publicly available at
The tracking algorithm tracks individual convective cells and their volume
based on the model output fields of vertical velocity and total condensate
mixing ratio. The tracking of maxima in the column vertical velocity field is
performed using trackpy
The volume of the convective clouds is determined by a watershedding
algorithm using a fixed threshold to determine the extent of the individual
clouds based on the tracked updrafts. We use a threshold of
1
Tables
In the Thompson scheme, some of the process rates are defined as signed variables representing two opposed processes. In these cases, we have used the process-rate variable with the positive sign for the respective process and ignored the values with the negative sign, which are covered by the opposing process (e.g. PRG_RCG for riming of rain on graupel and PRR_RCG for melting of graupel due to the collection by rain). Condensation/evaporation processes and deposition/sublimation processes are only defined through one combined process rate variable in the code. We have thus added the process rates with a negative sign as a variable in our diagnostics (e.g. E_PRW_VCD for the evaporation of droplets in addition to PRW_VCD for condensation) to allow for independent analyses of these, e.g. when aggregating the variables in space or time.
Ice multiplication according to the Hallet–Mossop process is implemented differently in the two bulk microphysics schemes. In the Morrison scheme, this is implemented as a direct transfer of water mass from the liquid phase to ice particles and considered to be contributing to riming. In the Thompson scheme, however, it forms a transfer from the frozen hydrometeor to new ice particles and is thus part of the “ice processes”. Hence, these processes are found in different categories in the two tables presenting the process rates. As the actual mass transfer is negligibly small, this difference between the schemes is not relevant for the analyses performed in this study.
In the Morrison microphysics scheme as used in this study, the autoconversion
of cloud droplets and accretion by rain are parameterised based on
The two bulk microphysics schemes differ in important parameters regarding
the different hydrometeor classes. The Morrison microphysics scheme is used
in its configuration that treats the dense frozen hydrometeors as hail with a
density of 900
Mass transfer process rates for the Morrison microphysics scheme
Mass transfer process rates for the Thompson microphysics
scheme
MH, BW, LL and PS designed the experiment, MH and BW implemented the microphysical pathway analysis in WRF, MH set up the simulations and developed the data analysis including the tracking algorithm, MH wrote the manuscript and BW, PS and LL contributed to the analysis and the manuscript.
The authors declare that they have no conflict of interest.
This article is part of the special issue “BACCHUS – Impact of Biogenic versus Anthropogenic emissions on Clouds and Climate: towards a Holistic UnderStanding (ACP/AMT/GMD inter-journal SI)”. It is not associated with a conference.
Max Heikenfeld acknowledges funding from the NERC Oxford DTP in Environmental
Research (NE/L002612/1). The research leading to these results has received
funding from the European Union's Seventh Framework Programme (FP7/2007-2013)
project BACCHUS under grant agreement no. 603445 (Max Heikenfeld, Philip
Stier and Laurent Labbouz). The authors acknowledge funding from the European
Research Council project ACCLAIM (Philip Stier and Bethan White) under grant
agreement no. 28002 from the European Union's Seventh Framework Programme
(FP7/2007-2013). Philip Stier and Max Heikenfeld acknowledge funding from the
European Research Council project RECAP under the European Union's Horizon
2020 research and innovation programme with grant agreement 724602. Laurent
Labbouz has also been supported by CNES. Bethan White has also received
funding from the Australian Government through the Australian Research
Council. The authors would like to acknowledge the use of the University of
Oxford Advanced Research Computing (ARC) facility
(