Impacts of emission reductions on aerosol radiative effects

Introduction Conclusions References


Introduction
The net radiative forcing caused by atmospheric aerosol particles originating from human activities 20 is currently negative, thereby offsetting a major, yet poorly-quantified fraction of the global warming caused by anthropogenic greenhouse gas emissions (Boucher et al., 2013;Smith and Mizrahi, 2013). The lifetime of atmospheric aerosol particles is relatively short, which has two major implications. Firstly, the climatically important aerosol properties vary greatly in both space and time in the atmosphere (e.g. Kaufman et al. (2002)). Secondly, and perhaps even more importantly, atmo-25 spheric aerosol concentrations respond rapidly to any changes in emissions of either primary aerosol particles or aerosol precursor gases.
Overall increases in aerosol emissions during the past decades have contributed to the so-called global dimming, i.e. the reduction of shortwave radiation reaching the surface, followed by some brightening due to later emission reductions in many regions of the world (e.g. Wild (2009) ;Cermak 30 et al. (2010); Haywood et al. (2011)). In near future, there is a pressure for further aerosol and aerosol precursor emission reductions due to the adverse health effects by atmospheric aerosol particles (e.g. Pope and Dockery (2006); Rao et al. (2012)). This has raised concerns about losing a significant fraction of the current aerosol cooling effect (Brasseur and Roeckner, 2005;Arneth et al., 2009;Raes and Seinfeld, 2009), and generated discussions on how to optimally realize future emission 35 reductions (Löndahl et al., 2010;Shindell et al., 2012;Shoemaker et al., 2013;Smith and Mizrahi, 2013;Partanen et al., 2013).
The discussed mitigation strategies focus on reduction of black carbon (BC). While BC itself has an apparent warming effect in the present-day climate (e.g. Jacobson (2010); Jones et al. (2011); Bond et al. (2013); Boucher et al. (2013)), the usually co-emitted sulphur and organic compounds 40 are effective cooling agents, substantially complicating the design of optimal emission reductions (Kopp and Mauzerall, 2010;Ramana et al., 2010;Wang et al., 2015). Furthermore, besides having a direct radiative effect on solar radiation, particles containing BC can act as cloud condensation and ice nuclei (Prenni et al. (2009) ;Leaitch et al. (2010)). The influence of BC emission changes on clouds and climate is potentially important yet poorly quantified (Chen et al., 2010a;Bahadur et al., 45 2012; Bond et al., 2013).
The relation between future aerosol emission changes, radiative forcing and climate has been investigated both globally (Kloster et al., 2008;Menon et al., 2008;Unger et al., 2009;Chen et al., 2010b;Bellouin et al., 2011;Makkonen et al., 2012;Gillett and Salzen, 2013;Levy et al., 2013;Smith and Bond, 2014;Wang et al., 2015) and over some continental regions (Mickley et al., 2012;50 Péré et al., 2012;Sillmann et al., 2013). While demonstrating potentially large regional effects, very few of these studies have simultaneously considered the following issues together: the direct and indirect aerosol effects, the role of different world regions' emissions in these effects, and contrasting emission changes reflecting alternative emission control strategies. In this paper, we aim to bring new insight into these issues by investigating near-future changes in the aerosol direct and indirect 55 2 radiative forcing globally as well as over a number of selected world regions as a result of emission changes according to four recently-developed emission scenarios. The specific questions we are searching answers for are the following: how much is the aerosol radiative effect, or the radiative forcing by aerosols, expected to change during the next couple of decades compared with the present day value? 60 how do these changes differ over different world regions?
what are the relative roles of direct and indirect effects?
to what extent are these patterns influenced by targeted emission reductions?
The paper is structured as follows: first, the model and the emission modifications are described in Section 2; Section 3 presents a detailed analysis of the results and explains the emission reductions 65 influences to the climate, followed by Section 4, where the main conclusions are listed and further steps are discussed.

Model description
The main tool in this work is the global aerosol-climate model ECHAM-HAMMOZ (version ECHAM5.5-70 HAM2.0) (Zhang et al., 2012). This model version has the HAM aerosol module (Stier et al., 2005), which includes the M7 aerosol microphysical module by Vignati et al. (2004). ECHAM-HAMMOZ simulates all the major aerosol sources (both natural and anthropogenic), microphysical processes and sinks. It predicts the evolution of seven interacting internally-and externally-mixed aerosol modes in terms of their size distribution and composition. The simulated aerosol components are 75 sulphate, BC, organic carbon (OC), sea salt and mineral dust. The aerosol module is coupled with the host model's large scale cloud scheme (no influence on convective microphysics) and radiation module; thus, both the direct and indirect aerosol effects are simulated online (Lohmann and Hoose, 2009). The cloud droplet activation is calculated using a parametrization by Abdul-Razzak and Ghan (2000). 80 The aerosol characteristics simulated by ECHAM-HAMMOZ have been evaluated in several previous studies. For example, ECHAM-HAMMOZ was included in the AeroCom model intercomparison exercise analyzing the life cycles of dust, sea salt, sulphate, black carbon and particulate organic matter in 16 global aerosol models (e.g Huneeus et al., 2011;Mann et al., 2014;Tsigaridis et al., 2014). Furthermore, Zhang et al. (2012) evaluated the ECHAM5-HAM2 version, which is 85 used in this study, against the AeroCom models and a large range of atmospheric measurements.
These studies have shown that ECHAM-HAMMOZ can reproduce the main aerosol characteristics realistically. There are, however, still some deficiencies in the model, as was pointed out by the study from Zhang et al. (2012): "(i) positive biases in AOD over the ocean, (ii) negative biases in AOD and aerosol mass concentration in high-latitude regions, and (iii) negative biases in particle number con-90 centration, especially that of the Aitken mode, in the lower troposphere in heavily polluted regions." However, in this study, we do not concentrate on model evaluation as such (this has been already partly done in Henriksson et al. (2014)), although we do compare our simulated aerosol burdens, lifetime and radiative effects to several previous model studies.

95
In this work, new emission modules were implemented to ECHAM-HAMMOZ and some of the old ones were updated. In the following sections, the new and modified modules are described in more  Table 1.

Continental anthropogenic emissions
For continental anthropogenic emissions, we applied gridded datasets based on the GAINS (Greenhouse gas -Air pollution Interactions and Synergies) model (Amann et al., 2011), operated by the International Institute for Applied Systems Analysis (IIASA, http://gains.iiasa.ac.at). Globally, the and 2015 for BC and OC (Granier et al., 2011). Over the same time period, SO 2 emissions have been estimated to slightly decrease globally (10-15%), although regionally, e.g. in India and China, the emissions may have increased . For comparison of GAINS emissions against for example Representative Concentration Pathways (RCP), see Granier et al. (2011). 120 In addition to the reference simulation, we considered four scenarios drawing on the energy projections presented in the World Energy Outlook 2009 (IEA, 2009) and including different assumptions of legislative and technological developments in the next few decades. The CLEC scenario 4 includes all currently agreed air pollution policies and legislation and estimates impacts on emissions in 2020 and 2030 (simulations CLEC2020 and CLEC2030, respectively). The CLECC scenario 125 includes these same policies, but is further designed to keep the total forcing due to long-lived greenhouse gases at 450 ppm CO 2 -equivalent level by the end of the century via CO 2 mitigation measures mostly targeting the energy and industrial sectors (simulations CLECC2020 and CLECC2030) -this scenario relies on the 2 • C (450 ppm) energy scenario developed by IEA (IEA, 2009). The main reductions in aerosol species between CLEC and CLECC occur in the residential, transport, energy and 130 industry sectors and are the result of shifts away from the use fossil fuels as well as improvements in energy efficiency (IEA, 2009). In addition, two more scenarios for 2030 were used. The BCAdd scenario targets the short-lived climate forcers (SLCFs) by including a portfolio of most important measures that could yield the largest reductions in their global radiative forcing in 2030 (simulation BCadd2030). The details of such scenario has been described in UNEP (2011)  sures include advanced emission standards on diesel engines (including diesel particulate filters), clean cookstoves, pellet stoves and boilers, more efficient brick kilns, and ban of agricultural burning. Thus, in terms of species used here, the reductions target BC and OC emissions. Measures with a relatively small net impact or increase in radiative forcing have been excluded from this portfolio. Lastly, the maximum technically feasible reduction (MTFR) scenario implements the maximum 145 reduction potential of anthropogenic aerosol and SO 2 emissions with currently available technologies by the year 2030 (simulation MTFR2030). The MTFR scenario introduces the best available technology to a maximum extend while ignoring any potential economic and political barriers. In this scenario, no consideration is given to the direction of the change in aerosol radiative forcing, so also measures that reduce strongly the emissions of SO 2 , e.g., fuel gas desulphurization, are in-150 cluded. The emission model used includes the end of pipe measures that remove pollutants from the exhaust. This means that it assumes that the use of most advanced particulate filters will reduce emissions of primary particular matter (PM), selective catalytic reduction (SCR) installations will bring NOx emissions down from industrial boilers, etc. For more detailed description of the current legislation and the MTFR scenarios, see e.g. Cofala et al. (2007) and Klimont et al. (2009). More 155 information about an overall emission scenario comparison can be found from Amann et al. (2013).
In this study, the detailed GAINS sectoral emissions were aggregated into six key categories: 1) agriculture (waste burning on fields), 2) residential and commercial combustion, 3) power plants, energy conversion, extraction, 4) industry (combustion and processing), 5) surface transportation and 6) waste. In addition, an extra sector for other SO 2 emissions not covered separately in GAINS 160 5 was included (mainly industrial sources not included in the 4th category). Each of the sectors were allocated into a 0.5 • × 0.5 • grid. The emissions from agriculture, residential and commercial combustion, surface transportation and waste sectors were emitted at the surface level. The energy sector emissions were released into the following model levels: 51.25% to 2 nd lowest level, 45.3% to 3 rd lowest level and 3.45% to 4 th lowest level. The industrial sector and the extra sector for SO 2 emis-165 sions had the same vertical emission height distribution: 95% to surface and 5% to 2 nd lowest level.
The emission heights were based on Bieser et al. (2011).
By default, GAINS provides only the total annual emissions for all sectors. Considering the importance of temporal resolution for few key sectors, we developed monthly estimates for power plants and residential combustion. Specifically for the latter, we applied the method developed by Streets 170 et al. (2003), who calculated the operating hours for stoves based on monthly mean temperature, i.e., < 0 • C ⇒ 16 hr/d, 0−5 • C ⇒ 12 hr/d, 5−10 • C ⇒ 6 hr/d and > 10 • C ⇒ 3 hr/d. In our approach, the monthly mean temperatures were obtained from the Climatic Research Unit (CRU) TS 3.1 dataset (Harris et al., 2014) and the calculations were done in each gridbox separately. Since our aim was to study the scenarios in current day climate conditions, the temperatures from 2005 were used also 175 for the scenarios.

Aviation emissions
We implemented into ECHAM-HAMMOZ the monthly aviation emission produced in QUANTIFY (Quantifying the Climate Impact of Global and European Transport Systems) project (Lee et al., 2005;Owen et al., 2010). Concerning the aerosol species and precursors of interest in our work, only

Wildfire emissions 185
The Global Fire Emissions Database (GFED) dataset for the wildfire emissions was updated to version 3 van der Werf et al., 2010). The data has a 0.5 • spatial resolution and is on a monthly time resolution. To make the emissions height dependent, the same approach as used by Dentener et al. (2006) with AeroCom emissions was applied. In this approach, based on location and type, the emissions are divided into six altitude regimes: 0-100 m, 100-500 m, 0.5-1 km, 1-2 km,  Table 1.

Shipping emissions
The international ship emissions used here were based on the improved ICOADS (International Comprehensive Ocean-Atmosphere Data Set) data by Wang et al. (2008). The ICOADS dataset presents only a proxy grid on a 0.1 • horizontal resolution, i.e. the dataset gives the fraction of total 200 global ship emissions that is emitted at each grid cell. The final gridded emissions were obtained by using the global proxy with the values from RCP 8.5 (Riahi et al., 2007(Riahi et al., ) (for 2005(Riahi et al., , 2020 and 2030 separately). The sensitivity of the results to the chosen RCP was tested by repeating the reference simulation (Refe2005) using RCP 2.6 emissions. However, the difference between the two RCPs was found to be so small that no further analysis will be shown from RCP 2.6 simulations. Since 205 the proxy does not include estimates on how the shipping routes will change in the future, the same emission pattern was used in all the simulations.
In the Arctic, we used an additional high resolution emission inventory by Corbett et al. (2010).
In this inventory, the data are given on a seasonal scale in a 5 km × 5 km horizontal grid for 2004, including 2020 and 2030 as scenario years. We used the emission values for 2004 in our reference 210 simulation for year 2005 without any modifications. It can be assumed that the error from this approach lies within the uncertainty limits of the emissions. For the scenario years 2020 and 2030, the Business As Usual (BAU) approach was chosen. The scenarios also include changes in the shipping route patterns (details in Corbett et al. (2010)). If there were overlapping grid boxes between ICOADS and Arctic emission datasets, the latter was chosen. The yearly shipping emissions are 215 represented in Table 1.

Simulations
Each simulation was run for 5 years (2003)(2004)(2005)(2006)(2007) preceded by a 6-month spin-up. In order to minimize the variation in the model meteorology, all the simulations were nudged, i.e. divergence, vorticity, surface pressure and temperature were nudged towards the ERA-Interim reanalysis data (Dee 220 et al., 2011). The sea surface temperatures (SST) were taken from the Atmospheric Model Intercomparison Project (AMIP II) (Taylor et al., 2000). The 5-year monthly data was furthermore averaged to one year monthly data (multi-year monthly mean), which minimizes the influence of the internal variability of the model. All simulations were conducted at a T63 horizontal resolution (∼200 km) with 31 vertical terrain following levels (top reaching 10 hPa). 225 We also made shorter simulations where the aerosol characteristics were compared to simulations with original emissions (not shown here). Based on these simulations, the new version reproduces closely the aerosol fields of the original model version.

Results and discussion
Below, we concentrate mainly on the 2030 simulation results, and discuss briefly 2020 when it re-230 veals additional information about the time scale of the emission reductions. All the absolute and relative changes presented are calculated as the difference between the scenario and reference simulation (Refe2005) values. In addition to global results, we analyse the simulations separately for the 8 regions shown in Figure 1. The column burdens and aerosol radiative effects for these regions are summarized in Tables 2 and 3. The annual mean BC column burden is shown in Fig. 2. In all the simulations, the BC burden peaks in the Amazon region and central Africa (biomass burning areas), India (residential biomass burning area) and eastern China (industrial area). In these peak areas, changes in the BC burden are relatively 240 modest in most of the scenarios apart from CLEC2030 which shows a 32% increase over India, as well as BCAdd2030 and MTFR2030 which both show nearly 60% decreases over eastern China (Table 2). Over India, the increase comes mainly from the traffic sector, which approximately doubles in CLEC2030. Even though the CLEC scenario includes current legislation measures, i.e. after some time new vehicles complying with existing standards will be in use, emissions start eventually 245 to grow proportionally to the activity growth. However, it is noteworthy that the domestic sector will still have the biggest emissions over India. The decrease over eastern China in the two mitigation scenarios (BCAdd and MTFR) is primarily due to declining use of solid fuels (mostly coal) for cooking and heating in the residential combustion sector. The high BC burden areas in the biomass burning regions of South America and Africa show negligible change in all the scenario runs because the 250 GAINS scenarios do not predict reductions for this sector (and the wildfire emissions from GFED are the same for all simulated years).
In regions with lower absolute BC burden values, all the scenarios predict significant decreases by 2030 over Europe (-24% to -66%, mainly from residential combustion and traffic sectors) and North America (-3% to -54%, mainly traffic sector), although in CLECC2030 the burden slightly increases 255 over Mexico and southern parts of U.S. (increment over western U.S. 8%, caused by residential combustion sector). Furthermore, in the CLEC and CLECC scenarios, the BC burden increases over Africa (9 and 5%, respectively; from residential combustion sector) and western China (28 and 15%, respectively; from residential combustion, traffic and industrial sectors). In these scenarios, small increases are seen also in southern Argentina, the west coast and southern parts of Africa, and the 260 border area of Indonesia and Papua New Guinea. There changes are caused by the overall emission increases over land areas in the Southern Hemisphere, as can be seen in Fig. S1. Partly due to atmospheric transport from continental areas and partly due to increased shipping emissions, the BC 8 burden also increases over Antarctica as well as over most oceanic regions in the Southern Hemisphere. Although the absolute BC values in these regions are low, the increased burdens could lead 265 to changes in the surface albedo over snowy and sea ice covered areas. In the CLEC2030 scenario, the burden also increases over the Arctic region. This is due to transport coming from southeastern Asia (around India), where the increased emissions cause increased values of BC at higher altitudes (lifting) which are eventually transported to the Arctic regions. In our analysis (details not shown here, but for more information, please visit http://www.maceb.fi/result_viewer.html), we found that 270 the lower tropospheric BC burden decreases in CLEC and CLECC over the Arctic, but the transported BC from southeastern Asia makes the overall burden change quite small, or even positive in the case of CLEC2030. A similar pathway for upper tropospheric Arctic BC from southeastern Asia has been discussed already in a previous study by Stohl (2006). In any case, since the albedo change due to BC deposition is not included in the current model version, further investigation concerning 275 BC effects on snowy regions is left for future studies.
The two other scenarios (BCAdd and MTFR) show a decreased BC burden over the whole globe (-26 and -27%, respectively). The differences between the burdens in these two scenarios are quite modest also on regional scales (Table 2), which means that the targeted sectors (transport and especially residential combustion) in BCAdd include most of the reduction potential of BC and very little 280 further reductions can be obtained with additional technological measures (as in MTFR). These additional measures come from waste disposal and treatment as well as from agricultural waste burning.
In the latter sector, the MTFR scenario assumes that all activity can be stopped and thus the emissions are set to zero.
Our reference simulation can be compared with previous model estimates on the atmospheric 285 aerosol burden. Schulz et al. (2006) reported results from a multi-model comparison for global BC, organic aerosol (OA) and SO 4 burdens for the year 2000. In the sub-set of models using AeroCom emissions, the global ensemble mean for BC was 0.25 mg/m 2 (standard deviation σ = 0.08 mg/m 2 ), whereas in the sub-set of model using other emission inventories, the global ensemble mean was 0.37 mg/m 2 (σ = 0.08 mg/m 2 ). In addition, Bond et al. (2013) collected results from recent publica-290 tions (some the same as in Schulz et al. (2006), details in the paper) and calculated a mean burden of 0.26 mg/m 2 . These results are in good agreement with our result of 0.25 mg/m 2 (Table 2) The residence time of aerosols is also one factor worth mentioning. In the reference simulation, we get a residence time of 6.0 days for BC. This compares well with earlier results, as shown by Textor et al. (2006), who did a multi-model comparison of the AeroCom models (simulating the 305 year 2000). Authors reported that the BC residence time for ECHAM was 5.3 days while the mean for all the AeroCom models was 7.1 days (σ = 33%). In addition, Shindell et al. (2013)  Our results indicate BC residence times of 6.3 days in CLEC2020 and CLECC2020, 6.4 days 310 in CLEC2030 and CLECC2030, 6.6 days in BCAdd, and 6.8 days in MTFR. The higher residence times in the scenarios reflect mainly the decreased washout due to less clouds, mainly caused by sulphate reductions (Lohmann and Feichter, 1997).

Organic aerosol burden
The absolute values of OA burden in the reference simulation (Fig. 3) are higher than those for the 315 BC burden (almost by a factor of 10), but overall the burden maps are very similar. This reflects the fact that these two compounds are often co-emitted from the same sources, although organic emissions dominate in magnitude, especially in the residential combustion sector. The OA burdens differ less between the different scenarios and show overall much smaller relative changes from the reference run than the BC burdens (Figures 2 and 3). The main reason for this is that the current 320 legislation measures do not have a major impact on domestic and agricultural sectors, which are two biggest sectors emitting OC (domestic is 5 times bigger than agricultural sector). This, together with unperturbed natural emissions, diminishes the differences seen in Fig. 3. On the other hand, the domestic sector will change quite dramatically (down to one fifth of the reference) in the BCAdd and MTFR scenarios, which mainly explains the larger differences in the OA burden for these scenarios.

325
Furthermore, the difference between BCAdd and MTFR can be explained by the agricultural sector, which, as was mentioned before, does not include any emissions in MTFR.
The CLEC2030 and CLECC2030 scenarios predict the largest changes in the OA burden over eastern China (-25 and -31%, respectively), mainly from the residential combustion sector due to reduction of solid fuel use and effective decline of stove emissions. On the other hand, changes over 330 India, Europe and North America are very small, in contrast to the BC burden changes. The differing behaviour of BC and OA burdens can be explained by the traffic sector, in which BC emissions are impacted much more strongly than OC emissions. This is because the reductions in the traffic sector are targeted to diesel emissions, which is a high BC emitter.
In the BCAdd simulation, the OA burden decreases globally and the highest reductions are over 335 Europe (-25%, mainly from residential combustion and traffic sectors), India (-50%, mainly resi-dential combustion sector), western China (-47%, residential combustion sector) and eastern China (-53%, residential combustion and energy sectors). The geographical pattern of the change is similar in MTFR, although the decrement is higher; the highest reductions occur over China, Japan, India, Middle-East and Europe reaching a -21% decrement globally (all sectors decrease, residential com-340 bustion sector having the biggest reductions). In these two scenarios, the pattern of the OA burden change is again quite different from the pattern of the BC burden change (Figures 2 and 3). The OA burden change is much larger over India due to a very large contribution from both stoves and agricultural burning, and these two sources have a high share of OC. On the other hand, larger BC changes are seen over Europe and North America as there are less stoves with high OC emissions, 345 and instead most mitigation will be in diesel controls with a high BC share and some in the residential combustion sector. It is also noticeable that changes over the Southern Hemisphere are small in all the scenarios.
The values for the global OA burden from Schulz et al. (2006)  The residence time of OA in our reference simulation was 5.8 days. Textor et al. (2006) reported for ECHAM a residence time of 5.4 days and overall AeroCom multi-model mean of 6.5 days (σ = 360 27%), whereas Kloster et al. (2008) got 5.7 days. This means that, similar to BC, the residence times of OA in our simulations are in good accord with previous studies. Our future estimates show an OA residence time of 5.8 days in CLEC2020 and CLECC2020, 5.8 days in CLEC2030, 5.9 days in CLECC2030, 5.9 days in BCAdd and 6.0 days in MTFR. These are similar to Kloster et al. (2008) estimates: 5.8 days in CLE and 5.9 days in MFR.

Sulphate burden
The absolute sulphate aerosol (SA) burden map in Fig. 4 differs from the BC and OA maps, because the anthropogenic emission sources are more similar between BC and OC compared with SO 2 . For BC and OC, the biggest source is the residential combustion sector, whereas SO 2 is mainly emitted from the industrial and energy sectors. In Europe, it is well known that sulphate precursor (SO 2 ) emissions have decreased over the last 2-3 decades (Hamed et al., 2010, and references therein). The same decreasing trend is also visible 375 in the current legislation based simulations, which have reductions from 26% (CLEC2030) to 35% (CLECC2030) over Europe. In North America, the reductions in the SA burden are even higher, especially over eastern and central parts of U.S. CLEC2030 gives -33% decrement over western U.S. and -40% over eastern U.S., whereas in CLECC2030 the values are -41% and -48%, respectively. These significant decreases in both Europe and North America are mainly from the energy sector, 380 although the industrial sector has also reductions that influence the results.
Quite the opposite can be seen over India, where the SA burden will increase in all the scenarios, except in MFTR. The increment is smallest in the CLECC2030 scenario (12%) and the highest in the CLEC2030 scenario (62%), although almost as high increase (58%) is simulated in the BCAdd scenario. On the other hand, in the MTFR scenario, the SA burden decreases by 60%. These fea-385 tures come from the industrial and energy sectors and mean that the SA burden over India could be controlled with technical measures, such as flue gas desulphurization.
The global SA burden was also reported by Schulz et al. (2006). For model using AeroCom emissions, the global mean burden was 2.12 mg/m 2 (σ = 0.82 mg/m 2 ) and for the other models 2.70 g/m 2 (σ = 1.09 mg/m 2 ). The SA burden from our simulation is slightly lower, being 1.85 mg/m 2 . For sulphate, the residence time in the reference simulation was 3.8 days. From Textor et al. (2006), ECHAM sulphate residence time was the same 3.8 days, while the AeroCom multi-model mean was 4.1 days (σ = 18%). Shindell et al. (2013) reported that their multi-model mean for sul-

Aerosol burdens in 2020
In order to explore the timeline of the emission reductions, we will shortly summarize the current legislation scenarios changes between 2005 and 2020. Details about the burden changes between these years are shown in Table 2 and Figure S4.
Regarding the BC burden, the same general features which were visible in the CLEC2030 sim-410 ulation can also be seen in CLEC2020. While the changes from 2005 through 2020 to 2030 do not follow a linear path, the CLEC2020 simulation shows overall the same global pattern as the CLEC2030 simulation (Fig. S4) Table 2, where over western U.S. the BC burden decreases 13% by 2020, but increases 8% by 2030. This feature comes from the residential combustion sector, which is estimated to increase quite significantly by 2030. The reason for this is that in CLECC the underlying idea is to move from fossil fuels to bio fuels and residential 430 burning, which takes place mainly between 2020 and 2030. Another region with a large difference in CLECC between 2020 and 2030 is eastern China, where the BC burden change (with respect to 2005) will increase from -9% to -25% and this comes from the reductions in residential combustion and energy sectors. Similarly as in CLEC, the reduction in the energy sector are roughly balanced out by the increased traffic sector.

435
The global OA burden changes are small in both scenarios. However, in the CLEC scenario, the burden increases 1.0% between 2005 and 2020, and 0.9 % between 2005 and 2030, indicating a slight reduction during the 2020s. On the other hand, a much stronger reduction after 2020 takes place in the CLECC scenario as the OA burden change is -0.05% by 2020 and -1% by 2030. Regionally, the largest differences are over eastern China and the Mexico-U.S. border. The decrement over eastern 440 China increases between 2020 and 2030 in CLEC from -10% to -25% and in CLECC from -15% to -31% and mainly comes from the residential combustion sector. Over the Mexico-U.S. border, 13 the scenarios show no signal by 2020, but by 2030 both have a strong positive sign; over western U.S. the burden change in CLEC is -2% by 2020 and 4% by 2030, and in CLECC -2% and 13%, respectively. As explained above, this is caused by the increases in residential combustion sector. In

Radiative effects
We will next investigate how the simulated changes in the aerosol burden translate into aerosol radiative effects. As the radiative effects presented in the following sections are mostly negative, i.e. they have a cooling effect, positive changes in radiative effects translate into a weaker cooling by aerosols, and vice versa. This should be kept in mind when the radiative effect plots are analysed.

465
Additionally, the values given in the following sections refer to the top of the atmosphere and are obtained directly from the radiation scheme (parallel calls with and without aerosols/clouds).

Direct radiative effect
Aerosols scatter and absorb the incoming solar radiation and the sum of these is called the direct radiative effect (DRE). Investigating changes in the DRE between two time periods, or years, tells 470 us how the direct radiative forcing by aerosols changes between these years in different emission scenarios. Besides short wave radiation permutations, aerosols (especially large particles, for example dust) can also influence the long wave radiation through absorption and emissivity. However, this is of minor importance for the small anthropogenic aerosols (Ramanathan and Feng, 2009). We have conducted tests to estimate the magnitude of the long wave component in our simulations and, 475 based on the results, the impact was found to be not important. Thus, the DRE in our analysis is only 14 calculated for the short wave radiation. It should also be noted that the DRE values are clear-sky values, which means that they are calculated assuming a zero cloud cover. Figure 5 shows the annual mean DRE for the reference run and the difference plots for the scenarios. The reference run shows that overall, the DRE is negative around the world (global mean -3.94 480 W/m 2 ). Previous studies have shows similar estimates. For example, Yu et al. (2006) presented a review of DRE estimates and concluded it to be −4.9 ± 0.7 W/m 2 over land and −5.5 ± 0.2 W/m 2 over oceans. Since many of the satellite measurements only give estimates over oceans, we have also calculated the equivalent value and got -4.68 W/m 2 (globally). This can be compared with Zhao et al. (2008), who estimated an oceanic DRE of −4.98 ± 1.67 W/m 2 , and with Forster et al. (2007), 485 who estimated from satellite remote sensing studies a value of -5.4 W/m 2 (σ = 0.9 W/m 2 ) over the oceans. Therefore, our simulations seem to give realistic values and are in good agreement with previous studies.
In the reference simulation, the strongest cooling effect caused by DRE takes place over the Atlantic ocean near the coast of East-Africa; this is mainly because of the dust transport from Sahara.

490
The overall aerosol burden is high over the polluted areas, including eastern China, where it leads to a DRE of -5.16 W/m 2 . Over Europe, India, Africa and eastern U.S., the values are quite close to the global mean, whereas in western China and western U.S., they are only approximately half of it. Over limited regions, the DRE can also be positive (Fig. 5). This happens when the underlying surface has a high albedo and the aerosols above are absorbing. This occurs mainly over Sahara, Antarctica 495 and Greenland. Seasonally, positive DRE could also be seen over the Arctic and other snow-covered regions. Note that DRE could be positive also if the absorbing aerosol are above clouds, but here we use only clear-sky values.
Consistent with reductions in aerosol emissions, all the scenario simulations predict a decreasing cooling effect by aerosols due to DRE over both Europe and North-America. The decrease in the 500 magnitude of the DRE is predicted to be 0.5-1.0 W/m 2 over Europe, 0.9-1.3 W/m 2 over eastern U.S., and 0.5-0.8 W/m 2 over western U.S. The smallest changes are seen in the CLEC and CLECC scenarios, and the largest in the MTFR scenario. These changes are mainly caused by reductions in SO 2 emissions, which lead to lower aerosol concentrations and thus, decrease the cooling effect.
The main sector causing these reductions is the energy production and distribution sector, which has 505 the highest reductions in the CLECC and MTFR scenarios. These reductions are also visible over eastern China, where BCAdd and CLEC scenarios show modest reduction in the cooling effect due to DRE change (0.07 and 0.29 W/m 2 , respectively), whereas they are much higher in the CLECC and MTFR scenarios (1.18 and 2.38 W/m 2 , respectively).
The simulated changes over India show significant variation between the different scenarios. Our 510 simulations predict that the cooling effect will increase in BCAdd and CLEC (-1.32 and -0.84 W/m 2 , respectively), no significant changes will occur in CLECC, whereas in MTFR, the cooling effect 15 will decrease (1.15 W/m 2 ). The reason for this behavior can be searched from the changes in aerosol component burdens (Figs. 2 -4).
It was shown in Section 3.1.1 that over India, the BC burden increases in the CLEC and CLECC 515 scenarios and decreases in the BCAdd and MTFR scenarios. Since the DRE change does not follow this pattern, it is obvious that its sign does not directly follow the changes in the BC burden. The OA burden changes over India are fairly similar with the BC burden changes, but overall, both changes are so small that they do not influence the DRE significantly. On the other hand, the SA burden increases in the BCAdd (58%) and CLEC scenarios (62%), has small changes in CLECC (12%) and 520 decreases in MTFR (-60%). It is clear that apart from CLECC, the changes in the DRE follow quite systematically the changes in the SA burden. In the CLECC simulation, the increased absorption coming from the increased BC burden eliminates the cooling entirely (absorption maps are in the supplementary material; Fig. S5). This means that, based on our model simulation predictions, the sign of DRE change over India is a combination of a warming component, for which the changes are 525 mainly caused by the residential combustion sector, and a cooling component, for which the changes are mainly due to energy production and distribution sector. Naturally, the same counteracting effects from absorbing BC and scattering sulphate can occur in other locations, but is particularly obvious over India in our simulations.
It is not straightforward to compare the simulated DRE changes to previously published estimates

Cloud radiative effect
The cloud radiative effect (CRE) is a sum of the short wave and long wave cloud radiative effects.
Since the short wave radiative effect is more dominant, the following analysis only includes the short wave component and makes the CRE analysis more consistent with the DRE analysis. Therefore, as 570 was with DRE, we only include the short wave component when discuss the CRE.
The CRE was calculated based on the method proposed by Ghan (2013), which removes the effects of aerosol scattering and absorption. The double-moment cloud scheme used in this work takes into account cloud droplet activation (Sec. 2.1). Freshly emitted insoluble BC may act as ice nuclei and thus, influence ice clouds directly. In case of warm clouds, only soluble aerosols have the 575 potential to act as cloud condensation nuclei (CCN). BC is emitted as insoluble, but can in our model become hygroscopic through condensation of sulphuric acid and coagulation with soluble particles. Furthermore, some error is introduced by the nudging method because it restricts some of the feedback processes. For example, if emission reductions change the regional or global cloud features in a way that it should impact the overall circulation, these feedback processes will not be fully realized in our simulations. Nevertheless, our approach does show how clouds and their properties react to 615 emission changes in current climatological conditions and gives indications on how the future cloud radiative effect might change.

Changes in aerosol radiative effect by the year 2020
We investigated the changes in radiative effects realized by year 2020 by looking into the current legislation simulation results (CLEC2020 and CLECC2020). The results are summarized in Table   620 3) and Figure S6.

Summary and Conclusions
We used the global aerosol-climate model ECHAM-HAMMOZ to evaluate how changes in the aerosol radiative effects, and hence forcing, are expected to decrease during the next couple of decades, and how they are influenced by emission reductions. This was done by modifying the The technical maximum potential for DRE reductions is globally 0.4 W/m 2 by 2030. Regionally, the cooling effect from DRE changes can also increase, for example over India and western China.
These changes follow mainly the BC and SA concentrations, which cause DRE changes of different signs. SA that has higher concentrations is more dominant and causes a cooling effect through scat-670 tering, while BC has the ability to absorb solar radiation and causes a heating effect. For example, over India, the cooling effect from DRE was estimated to increase due to the increased SA burden, although in one of the current legislation simulations the warming effect coming from the increased BC burden canceled out the cooling effect.
Our simulations suggest that the magnitude of the cloud radiative effect (CRE) will decrease Globally, the changes in the CRE cooling effect are roughly double the changes in DRE in most scenarios, but regionally large variability in the relative changes can be seen. For example, over India and western China the DRE change is larger than the CRE change. The changes in CRE take 680 place over oceans, whereas the DRE changes are seen mostly over the continents. Regionally, India and western China are the only areas where the cooling effect from DRE and CRE is expected to increase. This is because of the aerosol burden increases over these two regions.
Our simulations predict a notable positive radiative forcing change in the current day climate conditions, up to about 1 W/m 2 globally and > 5 W/m 2 regionally, due to the reductions in aerosol and 685 their precursor gas emissions that will take place during the next couple of decades. The magnitude of this forcing depends strongly on the chosen emission pathway. We have shown that targeted BC emission reductions are clearly the most beneficial for climate, making it even possible to achieve further enhancements in the negative direct radiative forcing (i.e. cooling effect) in some of the world regions (e.g. India and western China). To the contrary, reducing aerosol and their precursor 690 emissions as much as it is technically feasible could be harmful for climate practically in all continental regions, although potentially beneficial from human health protection point of view. Finally, our simulations suggest that more than half of the near-future aerosol forcing change is due to the 20 radiative effects associated with aerosol-cloud interactions. Noting this and the large uncertainties associated with this phenomenon (Boucher et al., 2013), more work is clearly needed for investi-695 gating the sources of cloud active aerosol particles into the atmosphere, aerosol-cloud-precipitation interactions and associated feedbacks in the climate system. Moreover, the use of coupled aerosolchemistry models with more detailed aerosol description (e.g. including nitrates) would give more detailed estimates of the future forcing of aerosols.
Acknowledgements. This work was supported by the EU Life+ project (LIFE09 ENV/FI/000572 MACEB), the     Table 3. The global and regional mean clear-sky direct radiative effect (DRE) and cloud radiative effect (CRE) at the top of the atmosphere for the reference simulation, and the changes in these (i.e. changes in aerosol radiative forcing) from the reference simulation to the future in different emission scenarios.