Significant climate impacts of aerosol changes driven by growth in energy use and advances in emissions control technology

. Anthropogenic aerosols have increased significantly since the industrial revolution, driven largely by growth in emissions from energy use in sectors including power generation, industry, and transport. Advances in emission control technologies since around 1970, however, have partially counteracted emissions increases from the above sectors. Using the fully-coupled Community Earth System Model, we quantify the effective radiative forcing (ERF) and climate response to 10 1970-2010 aerosol emission changes associated with the above two policy-relevant emission drivers. Emissions from energy use growth generate a global mean aerosol ERF of -0.31 W m -2 , and result in a global mean cooling of -0.35 K and a precipitation reduction of -0.03 mm day -1 . By contrast, the avoided emissions from advances in emission control technology, which benefit air quality, generate a global mean ERF of +0.21 W m -2 , a global warming of +0.10 K and global mean precipitation increase of +0.01 mm day -1 . The total net aerosol impacts on climate are dominated by energy use growth, from 15 Asia in particular. However, technology advances outweigh energy use growth over Europe and North America. Also, the Arctic climate is significantly affected by aerosol emission changes from Europe, North America, and Asia. Various non-linear processes are involved along the pathway from aerosol emissions to radiative forcing and ultimately to climate responses, suggesting that the diagnosed aerosol forcing and effects must be interpreted in the context of experiment designs. Further, the temperature response per unit aerosol ERF varies significantly across many factors, including location and magnitude of 20 emission changes, implying that ERF, and the related metrics, need to be used very carefully for aerosols. Future aerosol emission pathways have large temporal and spatial uncertainties; our findings provide useful information for both assessing and interpreting such uncertainties, Using a coupled composition-climate these investigated the influence of past avoided aerosol emissions due to legislation and technology 20 measures on European air quality, health and Here we employ the fully-coupled Community Earth System Model (CESM1) to examine the impacts of aerosol emission changes associated with the above two major competing drivers (energy use growth and technology advances) at both global and regional scales. The aerosol scenarios used the best estimates of past emissions. Therefore, compared to idealized experiments where aerosol emissions/concentrations are scaled rather arbitrarily, the implications of this work be more informative for future decision-making. The scenarios, CESM1 model overview and experiment design, as analysis the the large uncertainties (both spatially and temporally) in our and estimates of future aerosol emission trajectories, the possibility that very different future emission pathways may be adopted by different countries to compromise between climate/air pollution impacts and economic growth. The large impacts of present-day aerosol 25 emissions from the two competing drivers, as reported in this work, therefore suggest that the major drivers (e.g., future energy structure and efficiency, air pollution control measurements, as well as technology progresses) of aerosol emission changes may play even more important but uncertain roles in future climate projections. Nevertheless, our findings may help better assess and interpret such uncertainties in future climate projections. community to verify and further develop our findings in terms of aerosol-climate interactions and projections, as well as policy-relevant aerosol emission changes and their influences on air quality and associated socioeconomic impacts.


Introduction
Climate change is driven by changes in a combination of natural and anthropogenic factors (Stocker et al., 2013). The increasing atmospheric abundance of greenhouse gases (GHGs) associated with human activities has long been recognized as the major driver of global warming since the industrial revolution. Additionally, anthropogenic emissions of aerosols and their precursor gases have also led to significant climate impacts (Boucher et al., 2013), in addition to their detrimental effects on 5 atmospheric visibility, human health and ecosystems. Aerosols can influence climate by absorbing/scattering shortwave radiation (aerosol-radiation interactions; Haywood and Ramaswamy (1998)), and by modifying cloud microphysics and precipitation processes (aerosol-cloud interactions;Fan et al. (2016)). Overall, aerosol emissions cause a net cooling of the Earth; almost a third of the warming from increases in GHGs is thought to have been counteracted by cooling due to increased anthropogenic aerosols since the 1950s (Stocker et al., 2013). Yet, despite extensive research in the last decade that has led to 10 significant progress in our understanding of the effects of aerosols (Ming and Ramaswamy, 2009;Shindell and Faluvegi, 2009;Allen and Sherwood, 2011;Bollasina et al., 2011;Boucher et al., 2013;Hwang et al., 2013;Wilcox et al., 2013;Xie et al., 2013;Shindell, 2014;Wang et al., 2015), there are still major uncertainties associated with their impacts on climate (Carslaw et al., 2013a;Fan et al., 2016;Lee et al., 2016;Fletcher et al., 2018).
In fact, aerosol forcing remains the dominant uncertainty in current estimates of radiative forcing on climate since pre-industrial 15 times (Myhre et al., 2013). This is because of compounding uncertainties associated with the large spatial and temporal variability of aerosols, their short lifetimes, their diverse physical and chemical properties, and complex interactions with radiation and microphysical processes (Boucher et al., 2013;Carslaw et al., 2013b;Fan et al., 2016). For example, the sign and magnitude of the effect of aerosols on clouds and precipitation can vary substantially depending on emission locations, forcing types as well as meteorological conditions (Rosenfeld et al., 2008;Stevens and Feingold, 2009;Yu et al., 2014;20 Malavelle et al., 2017;Kasoar et al., 2018). Also, there are large uncertainties due to the incomplete knowledge of both historical aerosol emissions and how they will evolve in the future (Gidden et al., 2018). All these uncertainties make it challenging to project future climate and to quantify the associated impacts on a range of sectors. More importantly, despite ongoing debates (Feichter et al., 2004;Xie et al., 2013;Wilcox et al., 2018), a large body of studies indicate that, per unit of forcing/warming, aerosols have significantly larger impacts than GHGs on both global mean climate (Hansen et al., 2005;25 Shindell, 2014;Shindell et al., 2015), as well as global/regional climate extremes (Perkins, 2015;Xu et al., 2015;Lin et al., 2016;Samset et al., 2018a;Zhao et al., 2018;Zhao et al., 2019).
Emissions of anthropogenic aerosols have followed opposite trends between developed (decreases) and developing (increases) regions during the past few decades. For example, emissions of SO2 from Asia increased steadily since the 1950s, while emissions from Europe and North America started to decline after the 1970s Wang et al., 2015;Crippa et 30 al., 2016) (Silver et al., 2018;Zheng et al., 2018). As a result, India has recently overtaken China as the largest present-day emitter of SO2 (Li et al., 2017). Anthropogenic aerosol emissions will be further significantly reduced worldwide during the 21 st century (Markandya et al., 2018). Aerosol mitigation, however, may lead to adverse climate impacts, such as the increased risk of climate extremes (Kloster et al., 2010;Samset et al., 2018a;Zhao et al., 2018;Zhao et al., 2019). A 5 number of equally-plausible future emission pathways have been designed to seek a compromise between the impacts of air pollution on environment and climate following aerosol abatement in the near-, medium-, and long-term (Gidden et al., 2018).
The uncertainty in the emission pathway alone represents a key limiting factor to a robust quantification and isolation of the overall aerosol impact on climate. Yet, possible differences in the climate response to varying aerosol emissions trajectories, all the other forcings being the same, have been mostly overlooked so far (e.g., Sillmann et al. (2013); Pendergrass et al. 10 (2015); Bartlett et al. (2016)).
Emission changes described above are primarily associated with three important and largely regulated sectors (industry, power generation and transportation), while the residual contribution to emissions from residential and agricultural sectors is relatively stationary in time Hoesly et al., 2018). Also, such changes originate primarily from two competing emission drivers: economic growth and policy-driven emission controls . The former is 15 associated with energy use growth within the three sectors described above, while the latter includes both air pollution abatement measures and technology advances (hereinafter technology advances for short). To quantify the impacts of these factors,  developed the Emission Database for Global Atmospheric Research (EDGAR) retrospective air pollution emission scenarios for the period 1970-2010 (Sect. 2.1). Using a coupled composition-climate model and these scenarios, Turnock et al. (2016) investigated the influence of past avoided aerosol emissions due to legislation and technology 20 measures on European air quality, human health and climate.
Here we employ the fully-coupled Community Earth System Model (CESM1) to examine the impacts of aerosol emission changes associated with the above two major competing drivers (energy use growth and technology advances) at both global and regional scales. The aerosol scenarios used here represent the best estimates of past emissions. Therefore, compared to idealized experiments where aerosol emissions/concentrations are scaled rather arbitrarily, the implications of this work can 25 be more informative for future decision-making. The EDGAR scenarios, CESM1 model overview and experiment design, as well as analysis methods, are introduced in Section 2, Section 3 presents the results followed by a discussion in Section 4, and a summary in Section 5.

Model and experiment design
We carry out time-slice simulations for the period 1970-2010 (Table 1) using the fully-coupled Community Earth System Model (CESM1; Hurrell et al. (2013)) at the nominal 1-degree resolution. The motivation of carrying out time-slice model simulation for this particular period has been justified in Zhao et al. (2019b). The atmosphere component of CESM1 is the Community Atmosphere Model 5 (CAM5) in which concentrations of CO2 and CH4 are prescribed with seasonal cycles and 20 latitude-gradients . CAM5 includes a three-mode (Aitken, accumulation, and coarse) aerosol scheme (Modal Aerosol Mode 3). Several aerosol species (sulphate, organic carbon (OC), black carbon (BC), sea-salt, and dust) are simulated and their number concentrations and mass are prognostically calculated. Simple gas-phase chemistry is included for sulphate species: SO2 is converted into SO4 through both gas-phase OH oxidation and aqueous-phase oxidation by H2O2 and O3 (Liu et al., 2015a;Liu et al., 2015b;Tilmes et al., 2015). BC is emitted into the accumulation mode and ages, which allows 25 it to be coated with soluble species (e.g., SO4) and to nucleate cloud droplets Liu et al., 2012). Long-lived GHGs, natural aerosols and other reactive gases emissions/concentrations are obtained from (Lamarque et al., 2010) 2015)). Also, we have a number of perturbation runs (Table 1). For each case, we have a paired set of simulations: one with sea surface temperature and sea ice fixed (hereinafter Fsst), and the other with a fully coupled ocean (Fcpd). All Fcpd simulations were integrated into equilibrium (i.e., where the climate system equilibrates to imposed permutations but the deep ocean) after the 5 initial perturbation, with repeated annual cycles of the forcings. For example, the baseline simulations were integrated into equilibrium under constant 1970 and 2010 forcings (denoted as B70 and B10). Note carefully that the length of each integration is different, and is deemed sufficient for analysis once the top-of-the-atmosphere radiation imbalance does not show significant trends any more (stabilizing at ~0.3 W m -2 in this work) during the last few decades of each run, following recent works Myhre et al., 2017;Samset et al., 2018b). We analyse the last 30 years of each equilibrium simulation 10 and show differences between the baseline and perturbed simulations. Specifically, we denote 'best estimate' as the response to the best estimate of 1970-2010 total net anthropogenic aerosol-related emissions, 'energy use growth' as the response to emissions increases due to growth in energy use, and 'technology advances' as the response to avoided emissions from advances in emission control technology. The statistical significance of the difference between each two (baseline and perturbed) sets of 30-yr model runs is estimated by the two-sided student t-test (p-value <0.05). 15 The paired Fsst simulation is under the same forcings as the corresponding Fcpd simulation, and are integrated for 40 years from the initial condition. The last 30-years of each Fsst simulation is used to diagnose the effective radiative forcing (ERF) at the top-of-the-atmosphere (top of the model in this case, ~3.6 hPa) following Forster et al. (2016). Additionally, we carried out similar Fsst simulations to diagnose the ERFs of the best estimate of 1970-2010 changes in the three major anthropogenic aerosol species (BC, OC and sulphate). For example, we have a perturbation Fsst run in which only emissions of sulphate 20 species are changed back to 1970 levels while all other forcings are the same as B10 to diagnose the ERF due to 1970-2010 sulphate aerosol changes. anti-correlation between OC burdens and ERFs. The global mean ERF of sulphate aerosols is small in magnitude because of the partial cancellation between the positive forcing from sulphate aerosol reductions over Europe and North America and the marked negative forcing from sulphate aerosol increases over Asia (Fig. 1f). Yet, it can be seen that regional ERF values are dominated by sulphate. It is worth noticing that the individual ERF values of each aerosol species do not add up to that due to the simultaneous changes in all these at the global scale (Fig. 2d). A further discussion on this is provided in Sect. 4.1. 5

Effective radiative forcing
The spatial patterns of the changes in the 550-nm Aerosol Optical Depth (AOD) are strongly correlated with those of aerosol burden (compare Fig. 1a-c to Fig. 2a). Therefore, instead of aerosol burdens, we turn to change in the total AOD of all aerosol species for the three scenario experiments where all aerosol species change simultaneously. The total net 1970-2010 AOD changes ( Fig.2a), not surprisingly, display a sharp contrast between Asia (+0.034) and Europe (-0.022) and North America (-0.003). This, as described above, is mainly driven by changes in sulphate species (Fig. 1c). The 1970-2010 aerosol-related 10 emission changes produce a global mean ERF of -0.11 W m -2 , with marked regional values over Europe (+2.29 W m -2 ), North America (+0.24 W m -2 ) and Asia (-1.06 W m -2 ; Fig.3b). Emissions from energy use growth lead AOD to increase almost worldwide (Fig.2b), resulting in a global mean ERF of -0.31 W m -2 , with the most noticeable negative forcing of -0.88 W m -2 over Asia followed by -0.51 W m -2 over North America (Fig. 3b). By contrast, the avoided emissions due to technology advances lead AOD to decrease predominately over the Northern Hemisphere (Fig. 2c), and generate a global mean positive 15 forcing of +0.21 W m -2 (Fig. 2f). The most noticeable changes are found over Europe (+1.16 W m -2 ) and North America (+0.55 W m -2 ). warming over both Europe (+0.83 K) and the Arctic (+0.26 K), as well as a weak temperature increase over North America.

Temperature responses
On the contrary, cooling is found over Asia (-0.03 K), the Indian Ocean and the central North Pacific Ocean. The cooling is very likely due to Asian aerosol emissions, contrasting the widespread warming elsewhere related to aerosol reductions over Europe and North America. The zonal mean temperature profile displays weak warming in the Southern Hemisphere, but insignificant changes over the Northern Hemisphere tropics and mid-latitudes. However, the temperature response (warming) 25 is noticeably amplified over polar regions (e.g., +0.7 K over the Arctic). The Arctic warming amplification, primarily due to local sea-ice albedo positive feedbacks (Kay et al., 2012;Najafi et al., 2015;Sand et al., 2015;Navarro et al., 2016;Dobricic et al., 2019), is likely to be associated with the positive forcing (Fig. 2d)

from aerosol reductions over Europe and North
America (Navarro et al., 2016).
Aerosol-related emissions from energy use growth result in a relatively homogenous cooling that is statistically significant 30 almost worldwide (Fig. 4b), with a global mean cooling of -0.35 K. The cooling is enhanced over the Arctic (-0.92 K), very likely to be related to Asian emissions due to energy use growth, on top of the smaller contributions of the relatively smaller https://doi.org/10.5194/acp-2019-616 Preprint. Discussion started: 19 July 2019 c Author(s) 2019. CC BY 4.0 License. changes in aerosol emissions over Europe and North America (Fig. 2b). The zonal mean temperature response displays significant cooling across all latitude bands, with peak values found at the North Pole (up to -1.5 K). The avoided aerosolrelated emissions from technology advances (Fig. 2c) lead the globe to warm by +0.10 K, with the most pronounced responses over the Arctic (+0.22 K) and North America (+0.18 K). The zonal mean temperature response is only distinguishable from zero over the Northern Hemisphere mid-latitudes (~30°N) and the polar regions. It is worth pointing out the noticeable cooling 5 pattern over Europe despite the large positive forcing (+1.16 W m -2 ; Fig. 4f). The responses in sea level pressure and 850 hPa winds (not shown) suggest a large contribution to the European temperature response from the atmospheric circulation adjustments, resulting in anomalous cold advection from higher latitudes that offset the effects of local positive radiative forcing (Undorf et al., 2018).
As described above, despite the broad consistency between the patterns of aerosol ERF (Fig. 2) and temperature responses 10 ( Fig. 4), there are also notable dissimilarities. This is particularly true at regional scales such as over Asia, Africa and Europe.
To further investigate this, we calculate the temperature response per unit aerosol ERF (temperature sensitivity) over various domain (Fig. 5). It can be seen that the relationship between ERF and temperature response is far from being linear even at the global scale and over latitudinal bands. Also, note the negative temperature sensitivity values, suggesting that forcing and temperature response can be of opposite signs for aerosols. For example, the global mean sensitivity is estimated to be +1.17 15 and +0.48 K (W m -2 ) -1 for energy use growth and technology advances experiment, respectively, but -0.81 K (W m -2 ) -1 for the best estimate case. Further, even considering positive temperature sensitivity values only, there are substantial variations between regions and emission scenarios. A representative example is Europe, with both the largest regional temperature sensitivity of 1.28 K (W m -2 ) -1 in the energy use growth experiment, and the smallest value (0.02 K (W m -2 ) -1 ) in the technology advances experiment. Note also the very large temperature sensitivity value over the Arctic (~ 2 K (W m -2 ) -1 ) because of the 20 strong sea-ice albedo positive feedbacks which act to enhance the temperature response to a given forcing.

Precipitation responses
Changes in precipitation show complex spatial patterns (Fig. 6) and much larger uncertainties (Fig. 3d) compared to temperature responses. Overall, the 1970-2010 aerosol-related emission changes result in a global drying trend (-0.01 mm day -1 ), with the most pronounced changes over Asia (-0.13 mm day -1 ) and adjoining oceans. By comparison, Europe gets wetter 25 (+0.05 mm day -1 ; Fig. 6a). Also, there is a southward shift of the rain belt over the tropical oceans, indicating a zonal-mean southward shift of the Inter-Tropical Convergence Zone (ITCZ; Fig. 6).
The globe, especially land areas, gets drier in response to aerosol emissions from energy use growth (Fig. 6b). The precipitation change in Asia (-0.11 mm day -1 ) is close to that associated with the best estimate of 1970-2010 aerosol changes (-0.13 mm day -1 ). This suggests that aerosol emissions from energy use growth exert the predominant control on precipitation reduction 30 over Asia. The drying is also notable over Europe (-0.05 mm day -1 ). Along with precipitation decreases at almost all latitude bands and the tropics in particular, zonal mean precipitation changes show a weak but further southward shift of the ITCZ, https://doi.org/10.5194/acp-2019-616 Preprint. Discussion started: 19 July 2019 c Author(s) 2019. CC BY 4.0 License. leading to a weak precipitation increases over the Southern Hemisphere subtropics (10-30°S). On the contrary, technology advances lead precipitation to increase globally (+0.01 mm day -1 ) and particularly in the Northern Hemisphere. Nevertheless, precipitation decrease can still be seen over Southeast Asia, East China, northern Europe and USA. Meanwhile, the zonal mean precipitation profile shows a marked northward shift of the ITCZ with notable precipitation reductions over the Southern Hemisphere tropics. 5 Generally, in all the three aerosol perturbation experiments, precipitation changes with temperature at a rate of ~0.09 mm day -1 K -1 . This is slightly larger than the estimate (~28.6 mm yr -1 K -1 , i.e., ~0.08 mm day -1 K -1 ) for the slow climate response component derived from the Precipitation Driver Response Model Intercomparison Project (PDRMIP; ).
Most of the global and regional mean responses follow to some extent the linear increase (compare Fig. 3c to 3d), yet Asia and Europe deviate drastically from the linear relationship. This may suggest that regional precipitation responses are not 10 simply linked to temperature through energy budget constraints, but also depend on other factors such as prevailing circulation patterns and remote teleconnections Wilcox et al., 2018;Lewinschal et al., 2019). Overall, the above indicates the importance of changes in aerosol emissions in both global and regional precipitation changes. This is particularly true for Asia and Europe where represent the major sources of present-day aerosol emissions. In addition, aerosol changes are shown to have important influences on the ITCZ that tends to shift toward the warmer hemisphere (Allen and Sherwood, 2011;15 Hwang et al., 2013;Acosta Navarro et al., 2017;Liu et al., 2018).

Nonlinearities and the importance of background aerosol levels
Instead of linearly attributing the total aerosol emission changes into individual contributing factors, a "what-if" approach was adopted to develop the EDGAR retrospective emission scenarios . This design is useful to assess the 20 effectiveness of major drivers of aerosol emissions and allows us to show explicitly the policy-choice driven impacts, while accounting for nonlinear interplays between individual drivers. However, this approach adds extra nonlinearities to the results presented here in that, as discussed throughout this work, aerosol emissions from energy use growth and technology advances do not add up to the total net 1970-2010 emission changes. This may suggest the existence of other factors taking effects, yet it is difficult to attribute the residuals to such factors. Nevertheless, even when total emissions are linearly decomposed into 25 individual contributing factors, it is reasonable to expect both the radiative forcing and climate responses to not linearly add up because of a variety of intertwined mechanisms. For example, the location-dependent lifetime of different aerosol species (Liu et al., 2012), and the forcing efficacies (Kasoar et al., 2016;Aamaas et al., 2017).
In Sect. 3.1, we diagnosed the ERF associated with changes in each individual aerosol species as the differences between the baseline Fsst simulation (B10) and the ones where the targeted species (e.g., BC) are kept constant at their 1970 levels while 30 the others are as prescribed in B10. We note that changes in the burdens and AOD of the three aerosol species are identical to https://doi.org/10.5194/acp-2019-616 Preprint. Discussion started: 19 July 2019 c Author(s) 2019. CC BY 4.0 License. those in the experiment where all the three species change simultaneously (B10-B70). However, the ERF estimates do not linearly add up to the total. In fact, the residual (0.14 W m -2 ) is even larger in magnitude than the 1970-2010 total net aerosol ERF (-0.11 W m -2 ). This reflects partly the nonlinear effect associated with the mixing states of aerosol species as well as the importance of background aerosol loadings. This is particularly important for BC whose effects depend also on the presence of sulphate and organic aerosols (Ramana et al., 2010). More specifically, BC particles tend to be coated with other species 5 (e.g., sulphate, ammonium, and organic carbon) during ageing processes, thereby enhancing the absorption of BC and amplifying their radiative forcing (Haywood and Ramaswamy, 1998;Kim et al., 2008;Chung et al., 2012;Wu et al., 2016).
That is, the radiative forcing of BC may change with the ratio of BC to soluble aerosol species. In our case, the ERF of BC is diagnosed as the difference between the baseline experiment (B10) and that with BC held at the 1970 levels, leading the latter experiment to have a smaller ratio of BC to SO4 and therefore smaller ERF. As a consequence, the ERF estimate due to the 10 1970-2010 changes in BC emissions may be overestimated and may contribute to nonlinearities in the ERF of individual species. Note that these nonlinearities can be further enhanced by processes related to aerosol-cloud interactions, which is difficult to be quantified (Fan et al., 2016;Forster et al., 2016).
Overall, the above discussion illustrates the importance of background aerosol concentrations in estimating the radiative forcing of aerosols. For example, diagnosing the ERF of BC the other way round, namely, keeping all other aerosol species at 15 1970 levels while changing BC to 2010 levels, would likely result in different ERF estimates. Therefore, it is important to carefully bear in mind the method used when interpreting the ERF and climate responses associated with aerosol changes. For example, the single forcing experiments in the Coupled Model Intercomparison Project (CMIP5) (Taylor et al., 2012), the PDRMIP and other idealized aerosol perturbation experiments (Wang et al., 2015;Kasoar et al., 2018;Liu et al., 2018;Persad and Caldeira, 2018), as well as the upcoming AerChemMIP  model experiments 20 all need to be interpreted in the context of their experiment designs.

Caveats on the use of effective radiative forcing for aerosols
The ERF is generally deemed to be a useful indicator of temperature changes (Shindell and Faluvegi, 2009;Myhre et al., 2013;Shindell et al., 2015;Forster et al., 2016;Lewinschal et al., 2019). Based on ERF, many metrics have been proposed to facilitate comparing the effectiveness of various forcing agents. Also, these metrics are appealing to quickly assess the climate outcomes 25 of possible future emission pathways, and may hence provide useful information to policy-makers (Aamaas et al., 2017;Lewinschal et al., 2019). However, it is known that forcing and temperature response are not necessarily collocated, due to many other climate processes and feedbacks such as the atmospheric and oceanic heat transport, and atmospheric circulation adjustments (Boer and Yu, 2003;Shindell et al., 2010;Bellouin et al., 2016;Persad and Caldeira, 2018). Specifically, ERF and the associated metrics may work for well-mixed forcing agents such as GHGs (Zhao et al., 2019b). However, they are 30 misleading and open to dangerous miss-interpretation when used for aerosols and some other short-lived climate forcers (having a lifetime shorter than CO2, notably aerosols and ozone, and their precursor gases). We stress here again that temperature responses do not necessarily follow the ERF of aerosols. The range (0.02-2 (W m -2 ) -1 ) of the global mean temperature response per unit ERF is even larger than that (0.1-1.4 (W m -2 ) -1 ) reported by Persad and Caldeira (2018). Also, our results suggest that the model simulated temperature response per unit aerosol ERF can differ considerably with even subtle differences in experiment design (e.g. with different amount of aerosols emitted in different locations at different timings). Further, due to the fact that aerosol schemes are represented differently across present generation 5 climate models; it is highly likely that the sensitivities will differ further upon the choice of climate models. Therefore, as also pointed out by recent works (Persad and Caldeira, 2018;Lewinschal et al., 2019), the large divergence in the temperature response per unit ERF from aerosols highlights the need to use ERF and derivative metrics carefully for aerosols.

Implications for future climate projection
Reliable projections of future climate under different but equally plausible emission pathways are of utmost importance to 10 better constrain the range of possible societal risks and response options. Unfortunately, there are still considerable challenges due to limitations and uncertainties in our understanding of many aspects of the earth system (Knutti and Sedláček, 2013;Northrop and Chandler, 2014;Marotzke, 2019). Aerosols represent one of the largest sources of uncertainty (Boucher et al., 2013;Lee et al., 2016;Fletcher et al., 2018). Present-day anthropogenic aerosol emissions are largely influenced by sectors including power generation, industry and transport. However, in some of the future emission pathways, for example the Tier-15 1 Shared Socioeconomic Pathways scenarios (SSP1; Gidden et al. (2018)), aerosol emissions are expected to decline drastically worldwide as we transit to non-fossil-fuel-based fields together with rapid implementation of air pollution control measures and new technologies. However, the timing and rate of such transitions are largely uncertain. On the other hand, it is also likely that aerosol emissions will increase, especially over some developing regions, under scenarios where high inequality exist between and within countries. For example, in SSP3, expanding industrial sectors over Southeast Asia will rely continually on 20 traditional energy sources such as coal. As a consequence, aerosol emissions from energy use are expected to increase and therefore offset aerosol reductions elsewhere.
The above reflects the large uncertainties (both spatially and temporally) in our understanding and estimates of future aerosol emission trajectories, given the possibility that very different future emission pathways may be adopted by different countries to compromise between climate/air pollution impacts and economic growth. The large impacts of present-day aerosol 25 emissions from the two competing drivers, as reported in this work, therefore suggest that the major drivers (e.g., future energy structure and efficiency, air pollution control measurements, as well as technology progresses) of aerosol emission changes may play even more important but uncertain roles in future climate projections. Nevertheless, our findings may help better assess and interpret such uncertainties in future climate projections.

Summary and conclusions
Using CESM1, time-slice simulations were carried out to investigate the ERF and climate impacts of 1970-2010 aerosol emission changes, focusing on two major policy-relevant emission drivers that compete: energy use growth and advances in emission control technology. The 1970-2010 anthropogenic aerosol emission changes generate an ERF of -0.11 W m -2 . This is dominated by sulphate species, but the ERF estimates resolved into each individual species do not add up linearly to the 5 total. The residual may be associated with the mixing states of different aerosol species (Kim et al., 2008), as well as many other intertwined nonlinear processes linking aerosol emissions to radiative forcing, and to temperature and precipitation responses. These nonlinearities highlight the importance that one must bear aerosol experiment designs carefully in mind when interpreting aerosol forcing and effects. Particularly, the background concentration of both GHGs and aerosols may have strong influences on isolated aerosol effects using climate models (Regayre et al., 2018;Grandey and Wang, 2019). 10 Despite the global mean negative forcing from 1970-2010 aerosol changes, the global mean temperature increased by +0.09 K, while precipitation decreased by -0.01 mm day -1 . Energy use growth leads aerosols to increase over the Northern Hemisphere and Asia in particular, giving a global mean ERF of -0.31 W m -2 , and resulting in a homogenous global cooling (-0.41 K) and drying (-0.03 mm day -1 ). On the contrary, the avoided aerosol emissions due to technology advances generate a global mean ERF of +0.21 W m -2 , and result in global warming (+0.10 K) and wettening (+0.01 mm day -1 ). Change in aerosol 15 emissions and the resultant climate impacts are dominated by energy use growth over Asia but by technology advances over Europe and North America, while the global changes are a competition of these two drivers. Compared to the rest of the world, temperature responses in the Arctic are noticeably amplified because of positive feedback processes (Navarro et al., 2016;Wobus et al., 2016;Dobricic et al., 2019). The large temperature responses are likely to be related to changes in aerosol emissions over Europe and North America, while our results demonstrate that aerosol emissions from Asia may also play an 20 important role (Westervelt et al., 2015;Wang et al., 2018;Dobricic et al., 2019). The caveat is that all findings here may be model dependent, which is particularly the case for aerosols, given the high degree of parameterisation and divergence in aerosol schemes across present generation climate models. These findings, therefore, need to be verified using other models, while identifying the possible differences and reasons behind.
In conclusion, energy use growth and technology advances represent two major drivers of present-day aerosol emission 25 changes, and have strong and competing impacts on present-day climate. We anticipate that there will be significant but uncertain changes in aerosol emissions over the coming decades driven by these two drivers. Also, there are a variety of nonlinearities in the effects of aerosols, originating from many factors including aerosol experiment design. All these uncertainties and nonlinearities may translate into even larger uncertainties in future climate projections and associated impacts. Given all the findings and implications laid out above, we strongly encourage model groups to better constrain the 30 nonlinearities and uncertainness associated with aerosols in their climate models. Also, we encourage the wider research