First forcing estimates from the future CMIP6 scenarios of anthropogenic aerosol optical properties and an associated Twomey effect

anthropogenic aerosol optical properties and an associated Twomey effect Stephanie Fiedler1, Bjorn Stevens1, Matthew Gidden2, Steven J. Smith3, Keywan Riahi2, and Detlef van Vuuren4,5 1Max Planck Institute for Meteorology, Bundesstrasse 53, 20146, Hamburg, Germany 2International Institute for Applied Systems Analysis, Schlossplatz 1, A-2361 Laxenburg, Austria 3Joint Global Change Research Institute, Pacific Northwest National Laboratory, College Park, MD, USA 4Utrecht University, Vening Meineszgebouw A, Princetonlaan 8a, 3584 CB Utrecht, The Netherlands 5PBL Netherlands Environmental Assessment Agency, The Hague, The Netherlands Correspondence to: Stephanie Fiedler (stephanie.fiedler@mpimet.mpg.de)


Introduction
Projections of future climate change require plausible assumptions on socio-economic pathways. The sixth phase of the Coupled Model Inter-comparison Project (CMIP6, Eyring et al., 2016) uses the socio-economic pathways described in O'Neill et al. (2014) and quantified by Riahi et al. (2017). Nine different emission scenarios have been defined for CMIP6 and are described in the framework of the Scenario Model Inter-comparison Project (ScenarioMIP, O'Neill et al., 2016). These emissions 5 have been harmonised and downscaled by Gidden et al. (submitted.). The scenarios include projections of the anthropogenic aerosol emissions that we interpret here with the simple plumes parameterisation MACv2-SP (Fiedler et al., 2017;Stevens et al., 2017).
MACv2-SP parameterises radiative effects of anthropogenic aerosol, e.g., in the atmospheric component of Earth system models at the Max-Planck-Institute for Meteorology (Giorgetta et al., 2018;Müller et al., 2018;Mauritsen et al., submitted). 10 Some MIPs endorsed by CMIP6 and other projects require the MACv2-SP parameterisation to perform the requested simulations. Such endeavours are motivated by the benefit of a consistent treatment of aerosol forcing when exploring origins of model differences in radiative forcing (RFMIP, Pincus et al., 2016) and a computationally cheap representation of aerosol effects on climate both in high-resolution simulations (HighResMIP, Haarsma et al., 2016) and in decadal climate predictions (Miklip and DCPP, Boer et al., 2016;Marotzke et al., 2016). The MACv2-SP parameterization induces aerosol-radiation in-15 teraction by prescribing anthropogenic aerosol optical properties and aerosol-cloud interaction in form of a Twomey effect by perturbing the cloud droplet number concentration. The historical development of the anthropogenic aerosol optical depth, τ , has been derived by scaling with the anthropogenic aerosol emission of the past (Stevens et al., 2017).
In the present article, we scale MACv2-SP's τ in the period 2015−2100 with the gridded CMIP6 emission scenarios of SO 2 and NH 3 . The resulting scenarios in MACv2-SP are compared and classified by categories, describing the strength in forcing 20 magnitude and spatial asymmetries. Based on the extremes in projected τ , we present the first CMIP6 estimates of the radiative forcing of anthropogenic aerosol for the mid-2090s, using the atmosphere component of MPI-ESM1.2 (Mauritsen et al., submitted). The scaling method and the MPI-ESM1.2 experiments are described in Section 2, followed by the derived temporal developments of τ , the scenarios classification, and the associated difference in aerosol radiative forcing from MACv2-SP.

MACv2-SP parameterisation
MACv2-SP is the simple plumes parameterisation for anthropogenic aerosol optical properties and an associated effect on clouds (Fiedler et al., 2017;Stevens et al., 2017). The mid-2000s values are informed by the optical properties from the aerosol climatology of the Max-Planck-Institute for Meteorology version two, MACv2, which includes additional observational data and improved regional corrections (Kinne et al., 2013;Kinne, in review). For obtaining values for the entire historical period, future projections.
MACv2-SP parameterises the optical properties and a relative change in the cloud droplet number concentration associated with anthropogenic aerosol as function of the geographical position, height above ground level, time, and wavelength. It is designed for the implementation in a model's radiative transfer calculation. The development of MACv2-SP was inspired by 5 the desire for having a computationally in-expensive and transparent representation of anthropogenic aerosol, an approach with easily changeable settings for facilitating experimentation, and a method that is flexible enough for usage in a hierarchy of model complexity and resolution.
For achieving these aims, MACv2-SP approximates the spatio-temporal distribution of τ at 550 nm from MACv2 with analytical functions. These are a superposition of two rotated Gaussian distributions at each of the nine plume centres repre-10 senting regional pollution maxima. Figure 1 shows the annual mean of the aerosol optical depth of MACv2-SP for 2005 and the location of the nine plume centres for constructing the spatial distribution. We make herein a distinction between purely industrially polluted plumes and those that are additionally affected by seasonally active biomass burning. These regions differ by the level of aerosol absorption such that a different single scattering albedo is assigned, namely 0.87 for biomass burning and 0.93 for industrial plumes with the same asymmetry parameter of 0.63. The vertical distributions of the aerosol extinction 15 are approximated with beta functions, tuned to match the averaged profiles at the centre of the plumes from MACv2. Properties at wavelengths other than 550 nm are derived with an assumed Ångstrom exponent α = 2.0.

Construction of aerosol scenarios in MACv2-SP
MACv2-SP has month-to-month and year-to-year changes in τ . We adopt the same annual cycle as used for the historical reconstruction (Stevens et al., 2017). The year-to-year changes in the future scenarios are derived from the gridded aerosol We construct time series of τ i for each of the nine aerosol plume centres i using emission scaling factors, E i (t) as function of year t of the Gregorian calendar: where τ i in 2005 are the reference values from the MACv2 climatology (Section 2.1). The temporal scaling, E i (t), for each of 25 the nine aerosol scenarios in Tab. 2 are required input of the MACv2-SP parameterisation. They are available as MACv2-SP input files in netCDF format in the supplementary material. The resulting anthropogenic aerosol τ i (t) for each scenario are presented in Section 3.1.
We construct E i (t) from the gridded CMIP6 emissions, , of the chemical species, k: . (2)

30
The weight w k describes the relative contribution of the two species, namely w 1 = 0.645 for SO 2 and w 2 = 0.355 for NH 3 , motivated by the present-day ratio between sulphate and ammonia forcing as in Stevens et al. (2017). Following the approach by Stevens et al. (2017), we assume that the emission of all anthropogenic aerosol species scale with the emission of SO 2 and NH 3 , and therefore use these two species for scaling the anthropogenic aerosol optical depth of MACv2-SP over time.
We herein use NH 3 emissions in addition to SO 2 for considering that not all dominant aerosol emission changes over time scale with the SO 2 development. The approach is a simplification and is meant for facilitating experimentation and a better understanding of model errors (Stevens et al., 2017). Both emissions from open burning and otherwise classified anthropogenic 5 sources are taken into account. Aerosol absorption is represented by the single scattering albedo (Section 2.1).
We use here the gridded CMIP6 emission scenarios for the scaling, which is different from the emission scaling with ISO country codes for the historical reconstruction for MACv2-SP. For the scenarios, we average the anthropogenic emissions ik in a twenty by twenty degree box around each plume centre, marked in Figure  We test the reproducibility of the regional evolution of τ i by scaling with emissions averaged around the plume centers. For doing so, we derive E i from a pre-existing aerosol emission database adopting the same spatial averaging, and compare the results to the corresponding E i directly derived from the aerosol optical depth in a simulation with the aerosol-climate model ECHAM-HAM that uses the same aerosol emissions as boundary data ( Figure A1). Using spatial averages of aerosol emissions 15 around the plume centres gives E i similar to the direct scaling from the time-evolving aerosol optical depth from the complex model. The results for E i are herein not strongly sensitive to the choice of the number of grid boxes, e.g. a box of ten by ten degree around the plume centres only weakly modifies E i in most cases.  The future changes in τ of the scenarios further scale MACv2-SP's magnitude of the induced Twomey effect. We mimic a Twomey effect by the pre-factor, η N , that is parameterised as function of latitude, φ, and longitude, λ: Multiplying η N with the cloud droplet number concentration of the host model changes the cloud optical properties with the anthropogenic aerosol perturbation. The background aerosol optical depth, τ bg , is herein an idealised plume-wise approximation 30 consistent with the setting for the historical time period (Fiedler et al., 2017;Stevens et al., 2017): The components τ pl refer to a plume-shaped background and τ gl to a global constant. In the standard setup of MACv2-SP, τ gl is set to 0.02. This background is intended for MACv2-SP's aerosol-cloud interaction only and should not be confused with a natural aerosol pattern from observations. As emphasised by Stevens et al. (2017), this approach has been adopted so as to allow models to use their own natural aerosol for representing aerosol-radiation interaction, to optionally tune the radiation balance of models, and to keep a simple formulation of the Twomey effect for adjusting the magnitude. The Twomey effect (Twomey, 1974) is qualitatively understood, but the magnitude of aerosol-cloud interactions remains uncertain (Bellouin et al., in prep.). Reasons for the difficulties to constrain the magnitude are, for instance, a shortage of suitable observations, model 5 biases affecting radiative forcing, as well as the co-variability of meteorology and aerosol (e.g., Stevens and Feingold, 2009;Rosenfeld et al., 2014;Bony et al., 2015;Fiedler et al., 2016;Bellouin et al., in prep.). In result, different changes in N with aerosol have been proposed (Quaas et al., 2006;Andreae, 2009;Carslaw et al., 2013;Stevens et al., 2017). In the present work, we choose the original formulation by Stevens et al. (2017) for having a consistent treatment from pre-industrial to the future projections. Note that stronger aerosol-cloud interactions are also plausible (Stevens, 2015;Bellouin et al., in prep.) and can be

Calculation strategy for the aerosol forcing
We perform climate simulations for estimating the anthropogenic aerosol forcing. For doing so, we use the atmosphere-only The first year of each simulation is considered as the spin-up period and not used in the data analyses.
The setup of the experiments are summarised in Table 1. We choose three projections of τ in the mid-2090s for characterising the difference in the anthropogenic aerosol forcing, namely the scenarios SSP1-26, SSP3-70, and SSP5-85 that are introduced in Section 3.1. For these scenarios, we perform experiments with the standard settings for η N consistent with the historical reconstruction of Stevens et al. (2017). Additionally, we perform experiments with increased η N for each of these scenarios We determine the instantaneous radiative forcing (RF) of anthropogenic aerosol for the scenarios in Table 1. RF is computed 30 online by calling the radiative transfer calculation twice, i.e., once with and once without τ of MACv2-SP. Since we have three simulations for each setup with τ of MACv2-SP, we have thirty estimates of RF for each of the anthropogenic aerosol patterns.
We further estimate the regional forcing efficiency as: for assessing the co-location of RF with τ .
For calculating the effective radiative forcing (ERF) of anthropogenic aerosol relative to pre-industrial, we perform experiments without τ of MACv2-SP. For this reference setup, we run an ensemble of six simulations for 2000−2010 without 5 anthropogenic aerosol, but otherwise the same initial and boundary conditions as for the simulations with τ of MACv2-SP, for efficiently increasing the number of estimates for ERF. ERF is determined as annual differences in the top of the atmosphere shortwave radiation balance between the three simulations with additionally τ from MACv2-SP and the six simulations without τ from MACv2-SP. Since each simulation provides ten years for the analysis, we yield a total of 180 annual estimates of ERF for each anthropogenic aerosol pattern. The net contribution of rapid adjustments is the difference between ERF and RF 10 of anthropogenic aerosol.

Results
3.1 Scenarios of future anthropogenic τ

Regionally averaged projections
The results for τ i (t) are shown in Fig. 2 for each of the nine CMIP6 scenarios, listed in Tab. 2. Anthropogenic τ projections 15 for East Asia are decreasing and reach levels comparable to the present-day conditions in Europe in the scenarios of SSP1, SSP2, and SSP4 by the middle of the 21st century or later. The scenarios of SSP3 and SSP5 also show decreasing τ by the end of the 21st century, but the level first increases and does not reduce as drastically as in the other SSPs. The development for Africa's anthropogenic τ is typically an increase by 2100. This projection is particularly pronounced in SSP3 and SSP4 with an increase from 2015 to 2100 by factors around two to four. The scenarios of SSP1 and SSP2 have slight τ i increases in Africa 20 only, whereas the scenarios of SSP5 assume an increase in Africa's τ towards the 2040s and 2080s and a subsequent decrease, respectively. Anthropogenic τ in Europe and the Americas remain comparable to the low present-day levels.
The analytical functions in MACv2-SP construct the temporally changing τ patterns from the time series of τ i . Examples of the resulting spatial distributions of τ are shown for the mid-2050s and mid-2090s from all nine scenarios in the appendix ( Fig.   A2 and A3). We separate in the following MACv2-SP's scenario interpretation into differences in global mean magnitudes and 25 spatial patterns of τ . A summary of the global mean τ and the scenario categories is provided in Tab. 2.
Magnitude differences are shown as globally averaged τ and η N in Fig. 3. SSP3-70 stands out as the high-end scenario for both τ and η N with values exceeding the historical reconstruction for the mid-2000s and the mid-1970s. This scenario depicts socio-economic development failures associated with increasing air pollution (Riahi et al., 2017). Strong aerosol forcing is also expected from SSP4-60 and SSP5-85. In contrast, steep decreases in aerosol forcing are expected in SSP1, although τ reaches 30 minima around 0.012 only after the 2030s. This scenario reflects the assumption of stringent pollution controls (Riahi et al., to 36% of the τ in 2015 (Tab. 2). We therefore build our estimates of the radiative forcing of anthropogenic aerosol for the mid-2090s on these scenarios (Section 3.2).

Spatial τ asymmetries
We assess the regional characteristics in the scenarios by quantifying hemispheric differences in τ , rather than comparing mean 5 maps of τ . For doing so, we calculate the hemispheric asymmetry from zonally averaged τ for each latitude, φ: In a second step, we divide A by the global mean of the same scenario, τ , for screening out magnitude differences. The results are shown for three years in Fig. 4. We find similarly large A for SSP2-45, SSP3-70, and SSP3-LowNTCF that are closest to the value for 2015. For all scenarios, A is particularly large in the tropics and sub-tropics, here defined as regions equatorwards 10 of 36 • and referred to as low latitudes in the following.
All scenarios project a gradual decrease in the averaged hemispheric A in the course of the 21st century. This implies that the zonally averaged τ becomes increasingly symmetrically distributed about the equator in stark contrast to the historical reconstruction when most anthropogenic aerosol was in the northern hemisphere. We show these temporal evolutions of A and also the ratio, A lowφ /A highφ , of the mean A in low latitudes relative to the high-latitude mean in Fig. 3. The mean A lowφ 15 exceeds the value of A highφ by at least a factor of two (Fig. 3), indicating that most τ is in latitudes equatorwards of 36 • .
SSP4-34 has herein by far the strongest contrast between the low and high latitudes with a factor of roughly 11 in 2100, but the overall smallest hemispheric A. This behaviour reflects the relatively symmetric τ about the equator (Fig. A3). The scenarios of SSP1 also has a stronger concentrations of anthropogenic aerosol in the low latitudes than further polewards, but a moderate hemispheric A compared to more extreme scenarios. Overall, the scenarios of SSP5 have the smallest differences between the 20 low and high latitudes.
The scenarios in MACv2-SP are constructed with the same scaling values for the annual cycles in τ , shown in Fig. A4. Their main differences in the annual cycle of τ are associated with the variety in the spatial patterns discussed aloft. Similarities amongst the scenarios are marked (1) tropical τ maxima in the scenarios of SSP4 and SSP5-85 for June to January, and (2) northern high-latitude τ maxima in SSP2, SSP3, and SSP5. The former maxima are herein associated with anthropogenic 25 aerosol from biomass burning, whereas the latter is dominated by industrial emissions.
3.2 Radiative forcing of anthropogenic aerosol

Global means
We choose three scenarios for assessing the differences in the radiative forcing of anthropogenic aerosol in the mid-2090s associated with the choice of the emission pathway (Table 1). These are SSP3-70 as high-end scenario and SSP1-26 as a lower 30 bound for the differences in τ of 0.009 to 0.027 at the end of the 21st century. The third scenario choice is SSP5-85 that also has a high τ of 0.022, but interesting differences in the spatial patterns compared to SSP3-70. SSP5-85 projects most aerosol in Africa, while SSP3-70 most in Asia. The aerosol in Africa is seasonally dominated by biomass burning (Section 2.1), whereas aerosol in East Asia is primarily associated with industrial emissions (Section 2.1). SSP3-70 has herein the largest hemispheric A (Fig. 4), whereas the annual cycle in SSP5-85 is more strongly pronounced due to the seasonally active biomass burning in Africa (Fig. A4). Estimating ERF requires accounting for variability internal to the model. Fig. 6 shows the distribution of yearly estimates of 15 ERF from the ensemble of simulations for the mid-2090s. The year-to-year standard deviations of around 0.3 Wm −2 are herein comparable between the mid-2000s and all our projections for the mid-2090s. This behaviour reflects that a precise estimate of ERF for any given aerosol distribution and strength requires averaging over several decades (Fiedler et al., 2017). Particularly when the ERF is small, e.g., for SSP1-26, the year-to-year standard deviation is even larger than ERF itself (Tab. 3). Compared to ERF, the year-to-year standard deviation of RF is small, indicative of comparably stable estimates such that a one-year mean 20 is typically sufficient for a precise estimate of a model's RF. It implies that the model-internal variability in ERF is primarily associated with the variability in the net contribution of rapid adjustments (Fig. 6).

Spatial patterns
The regional pattern of the clear-sky contributions to RF resembles the distribution of τ , shown in Fig. 7. Negative radiative effects herein correlate with increasing τ , shown by the similarity of the patterns in forcing efficiency for all three scenarios of 25 different forcing strengths. Most regions show the expected negative radiative effects associated with anthropogenic aerosol.
The only exception is North Africa, where the more strongly absorbing aerosol at the edge of the biomass burning plume induces weakly positive radiative effects over the strongly reflective desert surface. In all-sky conditions, the patterns are to a great extent similar to the ones in clear-sky, but clouds mask parts of the negative radiative effects such that the regional all-sky contributions to RF are typically less negative (Fig. 8).

30
Including rapid adjustments strongly impairs the detectability of significant radiative effects for the mid-2090s aerosol patterns. Fig. 8 and Fig. 9 show the similarly strong impact of atmospheric variability on ERF for both high (SSP3-70) and low (SSP1-26) aerosol forcing categories as well as different strengths of the Twomey effect. The impact of natural variability is consistent with findings for the patterns of the mid-1970s and mid-2000s (Fiedler et al., 2017). An interesting feature in the projection for the mid-2090s is the positive forcing in parts of central Africa. This pattern emerges primarily from rapid adjustments in the atmosphere with a relative smaller regional contribution from RF (Fig. 8). In our model, the radiative forcing of anthropogenic aerosol from aerosol-radiation interaction and the Twomey effect induce heating perturbations. The associated change in the air temperature affects for instance the static stability of the atmosphere and thereby the circulation and embedded clouds. Such rapid adjustments cause the difference between ERF and RF, and are here summarised as net 5 contribution.

Conclusions
The present article presents the MACv2-SP interpretation of the future CMIP6 emissions of anthropogenic aerosol. We show the construction of the scaling parameter for the aerosol optical depth, τ , for 2015−2100 and the resulting spatio-temporal distribution of τ . The highlights of the projected aerosol developments for the 21st century are (1) a continuous stabilisation or 10 further decrease in τ in Europe and the Americas, (2) a long-time decrease of τ in East Asia stretching over the next decades in many scenarios, and (3) steep increases in τ in Africa's biomass burning regions in most scenarios. We rank the scenarios with respect to their strengths in the aerosol forcing magnitude, the hemispheric asymmetry and the low-to high-latitude asymmetry, summarised in Fig. 10.
We estimate the differences in the radiative forcing of anthropogenic aerosol at the end of the 21st century that are associated 15 with the choice of the future aerosol emission scenario (Fig. 11). For doing so, we choose three aerosol forcing scenarios that forcing. Research initiatives such as the radiative forcing model inter-comparison project (RFMIP, Pincus et al., 2016) can help to make progress in understanding model biases causing diversity in aerosol forcing. RFMIP adopts MACv2-SP including the here-presented high-end scenario SSP5-85 and will provide aerosol forcing estimates from 1850 to 2100. All scenario input files for MACv2-SP are freely available in the supplementary material of this publication. We hope MACv2-SP's scenarios will be useful for advancing our understanding of climate change and supporting impact studies for informing stakeholders.  , 9, 1937-1958, doi:10.5194/gmd-9-1937-2016, 2016.  Summarised are the instantaneous (RF) and effective radiative forcing (ERF) from aerosol-radiation interaction (ARI) and aerosol-cloud interaction (ACI), based on the high-end scenarios SSP3-70, SSP5-85, and SSP1-26 as low-emission scenario for the upper bound. The bars illustrate the difference in forcing associated with the emission pathways of CMIP6.