The interactions between aerosols and convective clouds represent some of the greatest uncertainties in the climate impact of aerosols in the atmosphere. A wide variety of mechanisms have been proposed by which aerosols may invigorate, suppress or change the properties of individual convective clouds, some of which can be reproduced in high-resolution limited-area models. However, there may also be mesoscale, regional or global adjustments which modulate or dampen such impacts which cannot be captured in the limited domain of such models. The Convective Cloud Field Model (CCFM) provides a mechanism to simulate a population of convective clouds, complete with microphysics and interactions between clouds, within each grid column at resolutions used for global climate modelling, so that a representation of the microphysical aerosol response within each parameterised cloud type is possible.
Using CCFM within the global aerosol–climate model ECHAM–HAM, we demonstrate how the parameterised cloud field responds to the present-day anthropogenic aerosol perturbation in different regions. In particular, we show that in regions with strongly forced deep convection and/or significant aerosol effects via large-scale processes, the changes in the convective cloud field due to microphysical effects are rather small; however in a more weakly forced regime such as the Caribbean, where large-scale aerosol effects are small, a signature of convective invigoration does become apparent.
The indirect effects of atmospheric aerosol via interactions with cloud and
precipitation remain some of the most uncertain contributors to anthropogenic
radiative forcing of the Earth's climate
However, although these mechanisms may be captured
in idealised cloud-resolving models (CRMs) and large-eddy simulations (LESs) of
individual clouds
Studying the effects of aerosol on convection using models at the global scale
is challenging, however, as we are some way from having the computer power to
routinely run convection-resolving global climate models for more than short
time periods. Current models therefore all require some form of
sub-grid-scale parameterisation of sub-grid-scale physical processes,
including convection
There have been a number of previous attempts to represent aerosol–convection
interactions in parameterised convection by a variety of approaches.
The Convective Cloud Field Model (CCFM), introduced by
We use CCFM here to study the impact of anthropogenic aerosol emissions on convective clouds in a model which can represent both local variations in the convective cloud field and feedbacks via the global atmosphere (though still not those via the ocean, without moving to a coupled atmosphere–ocean GCM). In this way, we quantify the responses of the parameterised convective cloud fields to changing aerosol in several regions, and we ask which physical processes are dominating these responses.
The modelling framework used here is as in
Developed at the Max Planck Institute for Meteorology, ECHAM6
The aerosol module HAM2
As in many models, clouds are divided into large-scale stratiform clouds and
convective clouds. The former use a two-moment microphysics scheme
In this study, we use version ECHAM6.1–HAM2.2–MOZ0.9 in the
ECHAM–HAM configuration (i.e. with the MOZ chemistry switched off) at
T63L31 resolution. This corresponds to about 1.875
CCFM is a spectral convective parameterisation which aims to represent the
large-scale effects of an ensemble of multiple convective cloud types within
each GCM column. This is based on the framework of
The individual cloud types are represented by a steady-state entraining plume
model following
The set of possible cloud types in CCFM is specified according to a range of starting radii between 200 m and the depth of the planetary boundary layer (PBL); an ensemble of 10 different entraining plumes is run over this range, each of which will develop differently as it rises in terms of its radius, velocity, moisture content, microphysics, etc.
Further details of CCFM can be found in
There are a number of mechanisms by which, without CCFM, ECHAM–HAM already supports aerosol effects on climate: direct radiative effects by scattering and absorption, semi-direct effects as cloud (both large-scale and convective) adjusts to the modified thermal profile, the cloud albedo effect due to changes in CDNC in the large-scale cloud scheme, and effects on (large-scale) cloud microphysics due to changes in CDNC, e.g. enhanced liquid water path due to rain suppression.
When using the standard Tiedtke–Nordeng convection scheme, there is no
explicit coupling of aerosols and convection; however both the direct
radiative effects and those on the large-scale cloud scheme
Activation at cloud base for each CCFM cloud type is calculated using the
parameterisation of are due to changes in droplet number propagating
through the convective cloud microphysics, including changes in
autoconversion rates, glaciation and associated latent heat release, etc.
are due to changes in the number (and thus size) of
droplets and ice particles detrained from CCFM into the large-scale cloud
scheme, implemented as in
In addition to these effects arising from aerosol interacting directly with
convective cloud, there are the semi-direct effects due to aerosol–radiation
interactions altering the thermodynamic profile and hence the CAPE and
convective inhibition (CIN) in the environment. Particularly
in the context of heating due to absorbing aerosol, these effects have been
shown to have a significant effect on convective behaviour
In both the CCFMall_
CCFM is only able to represent a subset of the possible aerosol effects on
convective microphysics through changes in rain formation rates
(autoconversion and accretion) and hence the amount of cloud water available
to freeze. However, this pathway appears to be key to aerosol effects on
convective clouds
In order to quantify the role of each of these mechanisms, including rapid
adjustments and feedbacks, we have run the model in a number of different
aerosol-coupling configurations which differ in the inclusion or exclusion of
one of these mechanisms, as shown in Table
Configurations used for aerosol effects in the ECHAM–HAM
simulations. “Convective microphysics effects” refers to changes in
microphysical process rates (autoconversion, accretion, freezing, etc.) within
the convective cloud as aerosol effects on droplet number propagate through
the convective microphysics. “Convective anvil effects” refers to changes in the size distribution of droplets and/or ice particles detrained to the large-scale cloud scheme. (See Sect.
Each configuration is run with both present-day (PD, year 2000) and
pre-industrial (PI, year 1850) climatological emissions of aerosols and
precursors, following the AeroCom Phase II/ACCMIP recommendations
(
The simulations have each been run for a period of 10 years (plus 15 months of
spin-up). To illustrate the aerosol perturbation that may lead to any
aerosol–convection interactions, Fig.
Annual mean CCN concentrations at the surface (left), large-scale cloud base (centre) and convective cloud base (PD–PI) from simulations using CCFM convection under PD (top row) and PI (middle row) aerosol emission scenarios, and the difference (bottom row). Specifically, these come from the CCFMall_ari simulations (see Table
The set of model configurations, with aerosol coupling processes successively
activated, allows the response of any given model output to each process to be
quantified via the difference between two of these configurations as detailed
in Table
The separate contributions of each mechanism to the total aerosol effect are extracted by taking the difference between pairs of simulations. (ARI: aerosol–radiation interactions; LS ACI: large-scale aerosol–cloud interactions; CCFM microphysics: changes to autoconversion in convective cloud; CCFM anvil: changes to size distribution of detrained condensate). See Sect.
We analyse this information from three different perspectives. Firstly, we look at the precipitation fields averaged over the whole 10-year period to identify the processes which affect the climatological distribution of precipitation in the model.
Secondly, we construct histograms showing the joint distribution of the radii
and tops of the individual convective plumes represented in CCFM
In order to distinguish meaningful signals indicating aerosol effects from the
noise due to the limited number of years in the simulations, we have carried
out statistical significance testing at the 95 % level using a two-sided
paired-sample
The 10-year mean precipitation response to anthropogenic aerosol in the
simulations with both Tiedtke–Nordeng and CCFM convection is shown in
Fig.
Annual mean precipitation response (PD–PI) in ECHAM–HAM with Tiedtke (top) and CCFM (below) convection, with the latter decomposed into large-scale and convective mechanisms as listed in Table
Figure
The 10-year mean PD
It is important however to appreciate that the constraints on total precipitation are stronger in an atmosphere-only model with fixed SST such as this than they would be in a coupled atmosphere–ocean model where the SST is able to vary in response to a perturbation, providing additional mechanisms for feedbacks including changes to global evaporation rates.
The explicit sub-grid-scale cloud fields of CCFM allow us to look more closely
at the effects on simulated cloud morphology in a way that is not possible
with a bulk mass-flux scheme like Tiedtke–Nordeng convection. The left column of
Fig.
Response of the CCFM cloud top–radius distribution in four different regions, as shown in the map (top, along with four additional regions included in the Supplement). Note that the “India” region used for analysis is restricted to the land points in the box. The left column shows the distribution under PD aerosol emissions; the right column shows the difference from PI aerosol emissions (with all effects included). The radius on the
The change in these distributions between PI and PD aerosol conditions does
not show a consistent pattern but varies considerably between regions and
regimes. Although there is some noise, many coherent features in the
responses do show statistical significance. In the Amazon, we see a shift
from broad deep clouds to narrower and
shallower ones, either due to reduced development of the clouds or due to the
triggering of clouds from smaller initial parcels. As there is no direct
mechanism by which aerosol can retard the development of an individual cloud
in CCFM, this is likely to be the result of either reduced CAPE in the
environment or reduced CIN allowing smaller clouds to
trigger. India is similar, but with only a slight lowering of the
deepest clouds. In China, the dominant effect is simply an overall reduction
in the amount of convective cloud (with the deep clouds most affected and only
a slight increase in the smallest and shallowest clouds),
suggesting a reduction of the large-scale convective forcing in the region
– perhaps associated with changes to the monsoon circulation
In order to disentangle these different responses, we turn to look at the
separate contributions of the different mechanisms.
Figure
Regional cloud field response decomposed into its different
mechanisms as listed in Table
In the three deep convective regions (Amazon, India and China), the total
aerosol effect on the cloud field is clearly dominated by that which occurs
when only the large-scale mechanisms are active, as shown in the LS row of
Fig.
In the shallow-to-deep transition environment of the Caribbean, however, the
weaker large-scale forcing leaves room for further vertical development of the
convective cloud, and the total effect appears to be driven by the convective
microphysics (as shown in
the CCFM row of Fig.
The idea that aerosol effects on convective microphysics are easily obscured
by changes to the large-scale forcing is consistent with idealised studies
The changes to the vertical profile of rain and ice production (and
associated latent heat release) within the CCFM-parameterised convective
clouds (Figs.
Regional change in the vertical profile of rain production within CCFM clouds, weighted by mass. The top row shows the PD distributions, while the bottom row shows the PD–PI difference (both with all effects included); the middle rows show the contributions from large-scale and CCFM (convective) mechanisms. Note that the colour scales are identical between mechanisms to allow comparison of their magnitude, but not between the different regions. A further breakdown into individual process effects, along with additional regions, is included in the Supplement. (PD: present-day aerosol; PI: pre-industrial aerosol; black dots indicate histogram bins where the PD–PI difference is statistically significant at the 95 % level.)
Regional change in the vertical profile of ice production (and consequent latent heat release) within CCFM clouds, weighted by mass. The top row shows the PD distributions, while the bottom row shows the PD–PI difference (both with all effects included); the middle rows show the contributions from large-scale and CCFM (convective) mechanisms. Note that the colour scales are identical between mechanisms to allow comparison of their magnitude but not between the different regions. A further breakdown into individual process effects, along with additional regions, is included in the Supplement. (PD: present-day aerosol; PI: pre-industrial aerosol; black dots indicate histogram bins where the PD–PI difference is statistically significant at the 95 % level.)
The direct and indirect responses of parameterised convective clouds to aerosol have the potential to contribute positively or negatively to the overall effective radiative forcing due to aerosol–cloud interactions (ERFaci). These cannot in general be captured by the bulk mass-flux parameterisations commonly used in global climate models, and thus CCFM provides a novel and potentially useful tool for investigating their role from a modelling perspective at the global scale.
With Tiedtke–Nordeng convection, there is only a single class of aerosol–cloud interactions represented in the large-scale cloud and precipitation; using CCFM this contribution (ERFaci_ls) is joined by that due to interactions between aerosol and the convection scheme itself (ERFaci_cv) if these are activated. These two combine to produce the total ERFaci. The extra ERFaci_cv seen in CCFM when these effects are activated is small and of marginal statistical significance from 10 years of simulation (95 % confidence interval
By explicitly considering convective microphysics and the sub-grid-scale heterogeneity of convective cloud, the Convective Cloud Field Model (CCFM) allows a physically based parameterisation of aerosol–convection interactions to be included in a global atmospheric model. This extends the more usual state of the art, where only aerosol interactions with large-scale liquid clouds are explicitly represented in global models.
Using 10-year ECHAM–HAM–CCFM simulations with each of the interaction mechanisms (de-)activated in turn, we have shown how the different processes and feedbacks typically interact to produce an overall response. The global mean precipitation response is not dominated by one process, but results from a combination of convective and large-scale microphysics, and feedback from aerosol–radiation interactions. To a large extent, these tend to counter one another, as expected based on the energetic control of global mean precipitation, especially in an atmosphere-only mode with fixed SST. Investigation of how aerosol affects the distribution of precipitation intensity in the model, even if the total remains fixed, may be worth further study, as would the future extension to a coupled atmosphere–ocean model.
The impacts on cloud field morphology are also a combination of large-scale and convective mechanisms, with considerable regional variation. In the deep convective regions, the overall response is dominated by the combination of large-scale radiative and cloud effects (including their feedbacks on circulation), with a smaller countering contribution from convective microphysics (which is often not statistically significant) via changes in the vertical profiles of process rates within the CCFM cloud model. In the Caribbean shallow convection region, however, the response of the convective parameterisation itself to the aerosol dominates and is statistically significant, with rain suppression, enhanced glaciation and deeper clouds indicative of convective invigoration.
These results are consistent with previous more idealised studies which have
suggested that shallower regimes with weaker forcing may be more susceptible
to aerosol-induced invigoration than strongly forced deep convection and that
aerosol microphysical effects become apparent only when they are not
overpowered by the greater effect of changes to the large-scale forcing
However, the results also show that, allowing for feedbacks on convective forcing, the traditional invigoration hypothesis does not apply globally. This has implications in particular for nested convection-resolving simulations in which the large-scale forcing remains fixed, suppressing these feedbacks which may be key to the total response of the system to increased aerosol.
From an effective radiative forcing perspective, a small additional effective forcing is seen from the aerosol–convective interactions captured in CCFM, but with 10 years of data this is of marginal statistical significance.
CCFM currently only represents a subset of the possible aerosol–convection
interactions, especially in the context of mixed-phase microphysics; however
these interactions appear to be of particular importance for the overall
aerosol effect on convection
The relevant model source code can be accessed via the HAMMOZ SVN repository (
The supplement related to this article is available online at:
ZK designed and conducted the experiments and data analysis and wrote the bulk of the manuscript. LL developed some of the additional model code used, in particular relating to microphysics, and provided insight during the analysis and additional text for the manuscript. PS provided supervision and guidance from the original concept through the experiments and analysis to the drafting of the manuscript.
The authors declare that they have no conflict of interest.
The ECHAM–HAMMOZ model is developed by a consortium composed of ETH Zürich, Max Planck Intitut für Meteorologie, Forschungzentrum Jülich, the University of Oxford and the Finnish Meteorological Institute, and it is managed by the Center for Climate Systems Modeling (C2SM) at ETH Zürich. The authors would like to acknowledge the use of the University of Oxford Advanced Research Computing (ARC) facility in carrying out this work (
This research has been supported by the European Commission, Seventh Framework Programme (ACCLAIM (grant no. 280025), BACCHUS (grant no. 603445)) and H2020 Research Infrastructures (RECAP (grant no. 724602)).
This paper was edited by Pedro Jimenez-Guerrero and reviewed by two anonymous referees.