Interactions between aerosol
particles and clouds contribute a great deal of uncertainty to the scientific
community's understanding of anthropogenic climate forcing. Aerosol particles
serve as the nucleation sites for cloud droplets, establishing a direct
linkage between anthropogenic particulate emissions and clouds in the climate
system. To resolve this linkage, the community has developed
parameterizations of aerosol activation which can be used in global climate
models to interactively predict cloud droplet number concentrations (CDNCs).
However, different activation schemes can exhibit different sensitivities to
aerosol perturbations in different meteorological or pollution regimes. To
assess the impact these different sensitivities have on climate forcing, we
have coupled three different core activation schemes and variants with the
CESM-MARC (two-Moment, Multi-Modal, Mixing-state-resolving Aerosol model for Research of Climate (MARC) coupled
with the National Center for Atmospheric Research's (NCAR) Community Earth
System Model (CESM; version 1.2)). Although the model produces a reasonable
present-day CDNC climatology when compared with observations regardless of
the scheme used,
Interactions between aerosol and water in different phases contribute
significant uncertainty towards the assessment of anthropogenic climate
change. Much of this uncertainty arises from the role of aerosol particles as
nuclei which seed the formation of clouds. Changes in the ambient particle
burden influence the microstructure of clouds and their optical properties,
leading to an “indirect effect” (AIE) on climate
The difficulty in constraining the indirect effect's magnitude on
contemporary climate change arises from two different but complementary
sources.
Details concerning the representation or parameterization of aerosol and
cloud processes in global climate models can also influence the magnitude of
their simulated AIE.
Here, we consider a fundamental component of aerosol–cloud interactions
(ACIs
hereafter) included in contemporary global climate models – the
nucleation of cloud droplets from the ambient aerosol population (also known
as “aerosol activation”). Droplet nucleation plays a key role in setting
the climatology of CDNC simulated within global models by providing the
initial inputs to cloud microphysical processes. In this manner, activation
schemes provide a direct linkage between otherwise independently modeled
aerosol and microphysical processes, enabling an explicit representation of
the indirect effect. This explicit representation has been implicated as a
critical component necessary to resolve regional aerosol impacts on both
warming trends over the 20th century
Several activation schemes have previously been developed for use in global
climate models
This work extends these previous literature efforts by quantifying the
influence of the representation of activation on estimates of the indirect
effect using a suite of state-of-the-science parameterizations coupled to an
aerosol-climate model. We include in our suite of parameterizations a
sophisticated emulator of droplet nucleation based on an adiabatic cloud
parcel model
This paper is organized as follows. Section
MARC aerosol mode size distribution and chemistry parameters. The
MOS mode (
In order to assess aerosol impacts on climate, we have utilized the
two-Moment, Multi-Modal, Mixing-state-resolving Aerosol model for Research of
Climate (MARC) coupled with the National Center for Atmospheric Research's
(NCAR) Community Earth System Model (CESM; version 1.2), which includes the
Community Atmosphere Model (CAM; version 5.3). In this CESM-MARC model (MARC
hereafter) we replace the default aerosol scheme
MARC explicitly simulates the evolution of a complex mixture of aerosol species, each with an associated lognormal size distribution. Within MARC, the aerosol species are divided into a set of externally mixed modes, including three distinct sulfate modes (nucleation or “NUC”, Aitken or “AIT”, and accumulation “ACC”), pure black carbon (BC), and organic carbon (OC). Additionally, MARC resolves two internally mixed modes, consisting of sulfate–black carbon (MBS) and sulfate–organic carbon (MOS). With the MBS mixture, particles are assumed to consist of a black carbon core coated with a sulfate shell; within the MOS mixture, particles are totally internally mixed according to the volumetric ratio of sulfate and organic carbon present. For each mode, MARC tracks the evolution of total number and mass concentrations. Additionally, MARC tracks the partitioning between carbon and sulfate for both the MOS and MBS modes.
Sulfate particles are formed in MARC via binary nucleation of
Dust and sea salt are computed in MARC using a sectional, single-moment
(fixed-size) scheme (with mean size bins of 0.16, 0.406, 0.867, and
1.656
The aerosols simulated by MARC fully couple and interact with both the CESM
radiative transfer model and its cloud microphysics scheme (through droplet
nucleation). Particles from all modes can be lost through dry deposition,
gravitational settling, and impaction scavenging via precipitation, and each
mode undergoes these processes with different efficiencies related to their
size and hygroscopicity
MARC adopts the stratiform cloud microphysics scheme from CAM5.3
We perform a set of simulations with MARC using different activation schemes
(see Appendix
For the
For each scheme, we performed a pair of 6-year simulations using a horizontal
grid resolution of
For all simulations, we output monthly mean fields and analyze the final
3 years of output for both the PD and PI cases. The change in SW cloud
radiative forcing between the two cases is diagnosed using a decomposition
which takes into account impacts due to surface albedo change
Additionally, we output a suite of instantaneous cloud micro- and
macrophysical variables sampled at either the cloud top or 1 km above
the surface and saved every 3 h over the duration of the simulations.
For consistency with the radiative transfer calculations in the model, the
maximum-random overlap hypothesis is used to derive cloud-top quantities
To assess MARC's performance in simulating present-day cloud and radiation
fields, we use a climatology of observations derived from satellite-based
sensors. Cloud micro- and macro-physical fields were derived from the
MODerate Resolution Imaging Spectroradiometer (MODIS; Collection 5.1). Cloud
droplet number is derived from Level 1 data from the same instrument using a
technique employing an adiabatic cloud assumption
Locations of in situ observational data reported by
Because the MODIS-derived cloud droplet number concentration retrievals have
a high degree of uncertainty, we also evaluate simulated droplet numbers
against a large collection of in situ observations previously compiled by
We supplement our simulations by further analyzing an additional set of
climate model output from the Aerosol Comparisons between Observations and
Models (AeroCom) Indirect Effects Experiment. This intercomparison includes
five independent aerosol-climate models (CAM5, ECHAM6-HAM2, ModelE-TOMAS,
SPRINTARS, and HadGEM3-UKCA), as well as several variations of the core
models adjusting the cloud microphysical scheme (CAM5-MG2), the turbulence
closure (CAM5-CLUBB), and the autoconversion scheme (SPRINTARS-KK). Similar
to the experiment conducted here, pairs of integrations (using present-day
and preindustrial emissions scenarios) were performed with each model, using
the same IPCC emissions scenarios for primary aerosol and precursors
One-to-one comparisons between observed and simulated cloud droplet
number concentrations from regions across the globe. Panels
Predicted cloud droplet number concentrations (CDNCs) from each variant of MARC
are compared against observations sourced from around the globe
(Fig.
Distributions of relative error in model-simulated CDNC versus observations
aggregated by region are shown in Fig.
Distributions of relative error between observed and simulated CDNC
(
For a more rigorous assessment of simulated CDNC, we compare MARC fields to
CDNC derived from MODIS observations
Global distribution in relative error of MARC-simulated CDNC versus MODIS-derived satellite observations (bottom right).
Zonal average aerosol, cloud, and radiation fields under present-day emissions scenario. Colored lines correspond to configurations of MARC using different activation schemes; black lines are derived from CERES-EBAF (SW cloud radiative effect, CRE) and MODIS (all other panels) observations. The shaded gray area corresponds to the inter-model spread for all available models participating in the AeroCom Indirect Effects Experiment; dashed white lines are the zonal averages for each participating model. Cloud droplet number is computed at the cloud top, using only grid cells over the ocean. Here, the SW CRE is computed using the difference between clear-sky and all-sky fluxes.
Although using different activation schemes does not directly perturb the
simulated aerosol distributions in MARC, the two-way coupling facilitated by
nucleation scavenging can indirectly influence average aerosol number
concentrations. In these simulations, the PD accumulation-mode number
concentration over the oceans is 31–40 % smaller in the simulations
using the
Compared to the original version of CESM/CAM5.3, the inclusion of an
alternative aerosol formulation does not substantially change the model's
simulated cloud and radiation fields, as illustrated in
Fig.
Present-day zonal average cloud macrophysical properties are summarized in
Fig.
The zonal averages of liquid cloud optical depth in
Fig.
Following the discussion in Sect.
Same as Fig.
Difference in SW CRE between
preindustrial and present-day emissions scenarios
Figure
Aerosol direct and indirect effects (in
Global-average effective radiative forcing for aerosol direct radiative effects (ERFari) and indirect effects (ERFaci) in both the shortwave and longwave. The total effect is computed as the sum of the direct and both indirect components.
To better illustrate the sources of differences in simulated CRE, changes in
aerosol and cloud microphysical properties between the PD and PI emission
scenarios are shown in Fig.
Regardless of which activation scheme is used, compared to
Fig.
Figure
Each perturbed component in the top-of-atmosphere (TOA) radiative budget is
decomposed in Table
Furthermore, we note that the indirect effect in the longwave is critically
sensitive to the baseline ice crystal number burden simulated in the model.
Additional tests using an alternative, aerosol-coupled ice nucleation scheme
The majority of the difference in the indirect effect and net TOA radiative
flux thus arises from changes in cloud interactions with SW radiation
via cloud optical thickness. For the SW CRE alone, the spread between
the different activation schemes is larger than the net effect itself at
1.1
The difference in SW radiative forcing between the
The previously discussed changes in the indirect effect and net TOA SW
radiative flux potentially have implications for the model meteorology
observed in our simulations. Although modifying the activation schemes only
directly influences the cloud microphysics, the resulting changes in
radiative forcing could impact both the larger scale circulation and locally affect processes such as convection. To highlight this,
Fig.
The majority of the global-average convective precipitation simulated by MARC
occurs in the intertropical convergence zone, extending from the Indian
Ocean basin into the Pacific around the equator. This region plays an
important role in the global SW radiative budget, as evidenced by both
the localized enhancement in cloud optical depth and SW CRE previously noted in Fig.
Relationship between regionally averaged PD–PI change in SW
CRE and aerosol optical depth
Annual average convective precipitation rate in the present-day
These local increases of up to 10 % of the reference simulation
convective precipitation rate suggest that local changes in meteorology might
play a larger role in the observed changes in cloud optical depth and other
radiative forcing diagnostics than local aerosol effects and their
derivatives alone. For instance, the localized increases in convective
precipitation rate observed in several of the simulations in
Fig.
Because of these effects, the potential role of the meteorological response in contributing towards the observed changes in cloud optical depth and thus SW CRE in the simulations presented here confounds to some extent the purely activation-driven changes.
We highlight in Fig.
The largest AOD increases occur over land, and these are also associated with
larger perturbations to SW CRE. The MARC simulations and most of the
AeroCom models do not simulate major increases in AOD over the ocean, even
while there is considerable spread in the magnitude of
Absolute difference between present-day annual average convective
precipitation for each of the indicated simulations versus the
Similar to
Fig.
CCN also directly increases with anthropogenic emissions. However, in
contrast with AOD, small PD–PI changes in CCN are associated with a larger
(more negative) indirect effect (Fig.
Comparison between estimates of ERFari
To assess this influence, we plot similar relationships between CDNC, liquid
water path (LWP), and liquid cloud fraction in Fig.
The liquid water path and cloud fraction exhibit a different relationship with
the indirect effect (Fig.
Using different activation cases produces larger differences in simulated PI CDNC versus either LWP or cloud fraction. This suggests that the large-scale cloud properties in MARC are insensitive to the background aerosol level. Instead, changes in the simulated indirect effect arising from the different activation schemes are dominated by the first indirect effect and a change in the ambient CDNC burden, which is driving microphysical changes leading to the observed perturbations in both cloud optical properties and their spatiotemporal distribution.
To contextualize the influence of aerosol activation on the indirect effect
in the simulations presented here, we plot estimates of the indirect effect
(ERFari
In our simulations with MARC, differences in aerosol activation produce a spread in estimates of the indirect effect comparable in magnitude to the total inter-model diversity. Furthermore, our estimates – especially for the configurations with lower CDNC – tend to cluster in the higher end of estimates compared to previous intercomparisons. The same is true for the AeroCom models considered here, although we note that four of the AeroCom models are closely related variants of the same parent model as MARC (the NCAR CAM5.3), and therefore the estimates are not totally independent of one another.
Our range of indirect effects induced by different activation treatments is
much larger than the few others reported in the literature. By reordering
the droplet activation calculation in each model timestep,
Additionally, we note a non-negligible meteorological response to the changes in activation in some of our simulations, particularly with regards to convective activity in the tropics. These changes in meteorology might imprint on the estimated sensitivity of MARC to changes in the activation scheme, since they can produce effects in cloud optical depth or other fields which influence the SW CRE.
In this study, we have quantified the influence of the representation of droplet activation in global models on the sensitivity of the aerosol indirect effect. Using a suite of state-of-the-science activation parameterizations incorporated into our global aerosol-climate model, MARC, we performed simulations under both preindustrial and present-day aerosol emissions scenarios to estimate the magnitude of the indirect effect and its relationship to changes in both cloud and aerosol fields. Previously, few studies exploring the indirect effect focused explicitly on the role of droplet activation, instead concentrating on either the processes that produce ambient aerosol itself (emissions and atmospheric chemistry) or the results of changes occurring purely in cloud droplet number concentration (such as imposed minimum values for cloud droplet number or in microphysical processes which modify it).
Beyond assessing three unique activation schemes, we supplement our analysis
by considering three additional, idealized droplet activation schemes which
use a heuristic to simplify accounting for competition between different
aerosol modes for moisture during the nucleation process. Including these
heuristics provides more than just additional variability in the activation
schemes studied here. Previous work has shown that for many
aerosol–meteorology parameter combinations arising in a global model, a
single dominant mode (typically the accumulation mode, especially if it
is mostly comprised of sulfate) tends to be a good predictor for the
activation dynamics of the full aerosol population
The relationship between cloud droplet number concentration and aerosol in
MARC is critically influenced by the representation of droplet activation.
Estimates of CDNC in the present-day climate are up to 40 % higher in
polluted regimes when using the most-sensitive activation scheme, and the
increase from preindustrial to present-day emissions is up to twice as large.
CDNC in regimes dominated by natural aerosol, especially remote marine
regions with prevalent sea salt, is also impacted by the activation scheme.
Using the advanced droplet activation schemes included here, which explicitly
account for biases due to giant CCN particles, helps reduce the
underprediction in maritime regimes compared to satellite observations and in
polluted regimes compared to in situ observations. However, MARC
systematically produces CDNC that is too low in most parts of the globe.
While this could be due to misrepresentation of aerosol–cloud processes, we
emphasize that it could also be fundamentally related to the simulated
aerosol size distribution within MARC and how it apportions aerosol number
and mass in the size ranges where likely CCN reside. However, evaluations of
previous versions of the model
We note that MARC's underprediction of CDNC may contribute to an
oversensitivity of the indirect effect to perturbations in aerosol
emissions. This is best understood in the context of
Compared to available satellite measurements and the models participating in the AeroCom intercomparison, though, MARC does well at capturing the present-day climatology of cloud and radiation fields, likely because its parent model, the NCAR CESM, is itself well tuned towards this end. However, the details of activation and how it influences cloud microphysics plays a major role in setting the SW CRE. Under present-day emissions, the differences between that effect for each of the different activation schemes is as large as the change from the preindustrial case for each scheme. This leads to large differences in the modeled indirect effect in each model, almost entirely occurring due to the SW CRE. The resulting spread in indirect effect estimates is twice as large as that previously reported by studies considering activation, and about as large as the inter-model spread from both historical and recent model intercomparisons, which consider models including a variety of different aerosol effects.
We note that the preindustrial CDNC burden is a very strong predictor of the
strength of the indirect effect, but not necessarily the change between
preindustrial and present-day emissions; this hints at the previously hypothesized
buffering effect of clouds on aerosol perturbations
Additionally, we note that in our simulations with MARC, changes to the activation scheme seem to elicit local meteorological responses beyond those that might be directly accounted for by changes in cloud microphysics alone. In particular, we noted changes in convective precipitation in the tropics that might hint at changes in the frequency and/or intensity of convection which imprint on the local climatology and sensitivity of cloud optical depth to aerosol perturbation. Model meteorological responses confound to at least some extent the changes in the AIE arising from the initial changes to the activation schemes. The largest changes in convective precipitation rate are associated with a smaller AIE, though, potentially pointing to a buffering effect when considering that these simulations also have the largest PI CDNC burdens. To address this potential confounding impact, future work should address whether or not this is an idiosyncrasy of MARC or a more general result by carrying out similar simulations with alternative global aerosol-climate models.
The weight of these results suggests an important role of activation in
setting the sensitivity of the indirect effect. However, we caution that our
approach is not able to disentangle the influence of activation from that due
to the underlying aerosol model and its implicit aerosol size and CCN
distributions. This is not meant to diminish the influence of cloud
microphysical treatments on the indirect effect;
To test this idea, additional work following this and
In order to better understand contemporary climate change and account for its future trajectory, the aerosol community must continue to seek constraints on the aerosol indirect effect. Although epistemic uncertainty due to unknown preindustrial emissions complicates this task, the role of droplet activation illustrated in this work highlights an additional path that the community may explore to provide indirect or emergent constraints on the AIE via the basic aerosol–CDNC relationship.
A Git repository archiving the scripts and build files
used to process the MARC and AeroCom output and perform the analyses
presented in this work can be found at
Output from the simulations used in this analysis are available upon request.
Droplet nucleation, or aerosol activation, refers to the process through
which aerosols, which are entrained through the base or sides of a cloud, grow
into a nascent cloud droplet population. Assessing this process is
complicated by the fact that latent heat release from condensation on the
surface of aerosol within an adiabatically ascending (and therefore cooling)
parcel provides a strong feedback, limiting the development of
supersaturation (relative humidity over 100 %) and thus the potential for
some particles (usually referred to as cloud condensation nuclei, or CCN) to
grow into droplets. Contrary to its common usage in the field, CCN is not
necessarily a stand-alone, diagnostic measure of a given aerosol population;
instead,
The aerosol size distributions predicted by MARC are explicitly used to
constrain droplet activation in the stratiform cloud microphysics scheme; the
shallow and deep convection schemes do not include the prognostic droplet number.
With respect to stratiform clouds, activation is driven by a characteristic
sub-grid-scale vertical velocity derived from the turbulent kinetic energy
(TKE) predicted by the University of Washington shallow convection and moist
turbulence parameterization
All of the aerosol species described in Table
In this work, we have implemented several additional activation schemes and
associated variants. Fundamentally, each activation scheme attempts to
simplify the calculation of the maximum supersaturation achieved in a parcel
under the simultaneous influence of both cooling from adiabatic ascent and
warming from latent heat release as water condenses on particles contained
within the parcel. The total of this physical process can be summarized in a
single, integro-differential equation:
Here,
It is immediately useful to simplify Eq. (
The first two activation schemes employed in this study utilize this formulation of the activation equation; the third implicitly uses the equation, albeit in an alternative form as a system of coupled ordinary differential equations.
Owing to its flexibility, the iterative scheme has been successively modified
over the course of several follow-on papers.
However, there are two drawbacks to this method. First, the need for a
reference parcel model for tuning is not completely eliminated; the branch
of the iterative scheme accounting for kinetic limitations still requires an
empirical relationship. Second, the iterative scheme can be much more
computationally expensive than the
An alternative approach to parameterizing droplet activation calculations
involves building look-up tables for inclusion in global models. However,
this approach is not widely used; in a modern model, the parameter space
influencing activation is very large, and covering such a space in a look-up
table is intractable. Fundamentally, a look-up table is a cache of results
from a higher complexity model (in this case, a detailed parcel model), which
are used to generate a piecewise-planar response surface on the fly during
model run time. With respect to activation, look-up table emulation has been
successfully employed with cloud-resolving models featuring simple
aerosol/CCN distributions
Unlike the
Additionally, we supplement the
While there is no substitute for detailed calculations of droplet activation which
take into account all potential factors, using just the dominant mode for
assessing activation provides a simple heuristic for widening the pool of
potential activation schemes to couple in our model. For the
The supplement related to this article is available online at:
The authors declare that they have no conflict of interest.
The work in this study was supported by the National Science Foundation Graduate Research Fellowship Program under both NSF grant 1122374 and NSF grant AGS- 1339264, the National Research Foundation of Singapore through the Singapore–MIT Alliance for Research and Technology and the interdisciplinary research group of the Center for Environmental Sensing and Modeling, and the US Department of Energy, Office of Science (DE-FG02-94ER61937). We thank Steve Ghan (PNNL) and Athanasios Nenes (Georgia Tech) for reference implementations of their activation parameterizations, and Natalie Mahowald (Cornell) for assistance in tuning the natural dust simulations in MARC. Edited by: Graham Feingold Reviewed by: two anonymous referees