China's aerosol cleanup has contributed to the acceleration in global warming

RAMIP simulations and recent emissions changes in China

We first document the emissions perturbation applied in the RAMIP baseline and East Asia simulations ¹⁸ (see Methods), and compare them to the actual emissions reductions from the same region since around 2010. Briefly, RAMIP isolates the climate effects of aerosol emissions in one region by comparing two sets of transient emission simulations; one following a global, high emissions pathway (SSP3-7.0, which assumes weak air quality policies), and one where aerosol emissions in one region (East Asia, consisting mainly of mainland China emissions) have been replaced by those from a strong air quality policy trajectory (SSP1-2.6). See Methods, or ¹⁸, for a full description. In the present analysis, we use simulations from 8 global models, each with 10 ensemble members, for a total of 80 ensemble members. This simulation set effectively samples both model uncertainty and internal climate variability.

Figure 1a shows changes in aerosol optical depth (AOD) retrieved by MODIS Terra and Aqua between the two previous decades. Consistent with previous literature, we find a dipole pattern consisting of an increase over India, and a strong decrease over China following their air quality improvement initiatives. For comparison, Figure 1b shows the pattern of AOD change between the RAMIP East Asia and baseline simulations, for the simulated period 2035-2049. Figures 1c and 1d show the corresponding SO₂ emissions and AOD change, for observations and simulations, within the box labeled East Asia in Figure 1a (a geographical box that covers the main emission regions of mainland China). 

For observations, relative to the 2005-2010 period, we find an AOD change of -0.13 units for the period 2014-2023, resulting primarily from emissions reductions of around 20 Tg SO₂ / year (Supplementary Figure 1b). Emissions data are from the December 2024 release of the Community Emissions Data System (CEDS) ¹⁹. Concurrent changes in black carbon (BC) aerosol emissions are shown in Supplementary Figure 1; they are smaller, in absolute terms and in particle number, and are not expected to contribute strongly to the AOD change, though they may influence climate features through their strong atmospheric shortwave absorption ²⁰. 

RAMIP transient simulations start in 2015, but use CMIP6 emissions based on a CEDS version that projected a delayed reduction in East Asia emissions compared to the actual, realized changes. The RAMIP East Asia and baseline simulations however still have an emissions difference trajectory that broadly corresponds to recent observations (20 Tg SO₂/year), for a later range of years. We also find a multi-model mean AOD change trajectory and magnitude that broadly tracks MODIS observations (DAOD of -0.11 ± 0.05 units for the RAMIP period 2035-2049). We do note, however, that even though all models used the same emissions, the RAMIP 2035-2049 mean East Asia AOD change ranges from -0.08 to -0.28 (see Fig.S1). This is due to a combination of factors including the optical properties of the simulated aerosols, the cloud fields, wind and precipitation climatologies, and aerosol removal rates.  

Physically, AOD decreases are associated with less scattering of incoming solar radiation, and hence increases in downwelling surface solar radiation. Supplementary Figure 1 shows the corresponding changes in downwelling shortwave radiation at the surface, in response to aerosol emissions reductions. Here, we find a multi-model mean change of 7.7 ± 2.5 Wm^-2,over the East Asia domain, with inter-model variation and spatial pattern that broadly follow that of AOD. 

Based on Figure 1 and Supplementary Figure 1, we conclude that the RAMIP East Asia results for the 2035-2049 period can be used as a proxy for the response to the emission rate change that has occurred in the real world over the 2010-2023 period (i.e. a 20 Tg / year sustained reduction in SO₂ emissions).

Modeled temperature and precipitation changes

In Figure 2, we show the resulting global, annual mean temperature responses to a 20 Tg/year reduction in SO₂ emissions from East Asia. For 2035-2049, we find a multi-model mean global warming of 0.07 ± 0.05 ºC, where the uncertainty is the standard deviation of the eight individual model results. The signal evolves smoothly, along with the emission reduction, with a rate-of-change of 0.02 ºC/decade for the full 2015-2050 period. Note, however, the strong inter-model variability (Fig. 2b), with one model (NorESM2-LM) showing an ensemble mean warming of 0.15 ºC, while another outlier (GISS-E2-1-G) even shows a slight cooling (-0.02 ºC). We link these model differences primarily to Arctic amplification and aerosol-cloud interactions in the North Pacific; see Supplementary Figure 2 and further discussion below. There is also a strong contribution from internal variability, with marked diversity between ensemble members (Figure 2b). This illustrates the difficulties of quantifying the climate impacts of the recent East Asian aerosol emission reductions, and other notable emissions changes like those resulting from the recent IMO shipping regulations ²¹ and the importance of conducting large ensemble simulations when investigating climate forcings that are strong regionally but weaker on a global scale ¹⁸.

Geographically, the seasonal temperature change is strongest near the source (East Asia, notably Eastern and Northen China) both in boreal summer (JJA, Fig. 2c) and winter (DJF, Fig. 2d). However, we also find significant warming (>0.2 ºC; paired Student’s t-test, p<0.05) over much of the North Pacific, in both seasons. For DJF, we also find a significant warming of North America, and throughout the Arctic. A wintertime cooling patch in Central Europe that has been reported by previous studies ^22,23 is however not visible in our dataset. For annual mean responses, including for individual models and ensemble members, see Supplementary Figure 2.

Supplementary Figure 3 shows the corresponding precipitation response, which broadly tracks that of surface temperature. We find an overall global wettening of 0.009 ± 0.004 mm/day (0.3 ± 0.1 %), yielding an overall hydrological sensitivity of 4 ± 2 %/ºC, which is broadly consistent with previous estimates of the climate impacts of aerosol emissions changes (20). Geographically, we find a strong summertime (JJA) precipitation increase in East China, and along the East Asian coastline, as well as a wettening along the North Pacific storm tracks extending well into North America. We also find a northward shift of the ITCZ, consistent with expectations from preferential warming of the Northern Hemisphere relative to the Southern Hemisphere ²⁴ The regions with statistical significance are smaller than for temperature, as expected due to the higher internal variability and greater model diversity in precipitation simulations. 

Influence on recent global warming and radiative imbalance

We now put the RAMIP results in the context of recent trends in global warming, in Figure 3. In panel 3a, we show the global mean surface temperature anomaly (GSTA) relative to 1850-1900 since 1980, as the average of four observational reconstructions (HadCRUT5, NOAA, GISTEMP and Berkeley Earth; see Methods). For the 30-year period of 1980-2009, the average observed warming rate is 0.18 ºC / decade. For illustration, we also show the time series from (7), where internal variability from Pacific ENSO and other ocean modes of variability has been filtered out (our conclusions do not rely on the usage of this dataset). For the subsequent period of 2010 - 2023, we find an elevated observed warming rate of 0.33 ºC / decade, and 0.25 ºC / decade when interannual variability is filtered, consistent with previous studies ^2,3. 

In the main panel (Fig. 3b), we zoom in on the latter period. Most post-2010 GSTA values fall above the continuation of the 1980-2009 trend (dotted black line), in the reconstructions (dots) and in the reduced variability time series (solid black line), indicating a recent increase in the global warming rate. To estimate the contribution of East Asian aerosol emissions changes to this increase, we take the RAMIP quantified global warming of a 0.07 ± 0.05 ℃, and convert it to a warming rate over the 2010-2023 period (0.05 ± 0.04 ºC / decade). See the box-and-whisker on the right of Fig. 3b, which shows how we’ve added this aerosol cleanup induced warming rate (red line and range) to the continuation of the 1980-2009 trend. 

Assuming, for now, no change in the underlying greenhouse gas induced global warming rate, this suggests a combined post-2010 warming rate, from greenhouse gas increases and East Asian aerosol cleanup, of 0.18 ℃/decade + 0.05 ℃/decade = 0.23 ℃/decade, approaching the 0.25 ºC / decade found after filtering the effects of internal variability. For context, see further discussion below of other sources of recent warming. 

Note that we assume here that the full warming due to the observed recent East Asian emissions reduction has already been realized, so that it is justifiable to compare warming in observations during the 2010-2023 period to warming in the RAMIP simulations during the 2035-2049 period. This is supported by recent modelling exercises using sustained step changes in SO₂ emissions (21), finding that the majority of subsequent global mean surface temperature change has been realized within the first 24 months, while the rest develops slowly at a multi-decadal timescale. The transient RAMIP simulations also do not show any appreciable delay in the climate response to SO₂ reductions. However, our estimate should still be taken as an upper limit, to take this limitation into account.

Since aerosol changes have regionally heterogeneous climate influences, as shown above, we next investigate the correspondence of simulated changes to observed regional warming rate. In Figure 3c, we show the observed difference between the 2010-2023 and 1980-2009 warming rates, in the four reconstructions. See Supplementary Figure 4 for individual time series and the two trend periods in isolation. While a 13-year trend will be strongly influenced by decadal scale variability, we do find a very clear pattern of observed warming in the North Pacific, with two distinct maxima: one along the East Asian coastline, and the other following the west coast of North America, extending west to the center of the Pacific. 

In Figure 3d, we show the annual mean surface warming pattern from East Asian aerosol emission reductions in RAMIP. As documented above, we also here find a two-maxima pattern along the East Asian and western North American coastlines. This shows that the simulated increased global mean warming rate comes from a geographical region where observations also find an elevated warming rate since 2010, relative to previous decades. See Supplementary Figure 7 for the transient evolution of regional means (Western and Eastern North Pacific) for individual models. 

Top-of-atmosphere radiative imbalance

The next question is what physical processes lead to this warming rate. In Figure 4, we investigate the top-of-atmosphere (TOA) all-sky change in radiative imbalance (shortwave plus longwave) in response to East Asian aerosol emissions reductions in RAMIP, and compare them to observations (CERES) and reanalysis (ERA5). Fig. 4a-b show the time series and 2035-2049 means of the TOA all-sky radiative imbalance, which has a mean of 0.06 ± 0.04 Wm^-2. While there is substantial inter-model and ensemble member variability, the overall evolution is very similar between the eight RAMIP models. Using the same logic as above, this corresponds to an evolving increase in the TOA imbalance since 2010 of 0.05 ± 0.03 Wm^-2 / decade.

Fig. 4c-d shows the geographical distributions of clear sky and all-sky radiative imbalances. Again, for all-sky conditions, we find a geographical pattern displaying two clear maxima, near the source region, and in the western North Pacific, with peak values exceeding 2 Wm^-2. There is little influence on other regions. For clear sky conditions, only the East Asian and eastern North Pacific influence remains, indicating it has a major contribution from the direct interaction of aerosols with incoming sunlight. The maximum west of North America, however, is likely primarily a result of aerosol-cloud interactions, seen in a region with high prevalence of low clouds (stratocumulus decks). Supplementary Figure 5 shows individual models, while Supplementary Figure 6 shows a further breakdown into shortwave and long wave components. 

This result also explains part of the inter-model diversity in RAMIP responses; the models that have low overall temperature response to East Asian aerosol changes (notably CNRM and GISS), also have weak responses in this region, related to their cloud climatologies and their simplified treatment of aerosol-cloud interactions. See also Supplementary Figure 7, which shows the transient evolution of Western and Eastern North Pacific means of temperature and surface shortwave fluxes. We further note that that for clear sky conditions, there is a strong positive anomaly over Eastern China in most models. This indicates the presence of compensating aerosol-cloud effects, and perhaps other aerosol processes such as wet removal by precipitation, over this region in the RAMIP models.

Finally, in Fig. 4e-f, we show recent all-sky TOA imbalance trends from observations (CERES), and its difference from a reanalysis product (ERA5). While the CERES observations (Fig. 4e) have a strong influence of internal variability, and will also be influenced by other recent changes in the climate system, we do find a positive anomaly in the North Pacific. The ERA5 reanalysis itself shows a very similar pattern (Supplementary Figure 3). However, when taking the difference between CERES and ERA5 (Fig. 4f), we find two striking features. One is a positive anomaly in the low cloud region of the eastern North Pacific, the other is a negative anomaly over Eastern China. ERA5 does include a treatment of aerosols, but it uses a CMIP5 era combination of historical emissions up to 2009 and subsequently the RCP emissions ²⁵. These pathways do not include the recent reductions in East Asian emissions. Hence, the difference between CERES and ERA5 may be indicative of regions where the recent aerosol changes are important for the reconstruction, and have an influence on observed rates of change of TOA radiative fluxes. This is supported by RAMIP finding that these two regions are strongly influenced by aerosol-cloud interactions. We do note, however, that trends in sea surface temperatures will also influence ERA5 reconstructions, meaning that we cannot make a clear attribution of the observed changes to aerosol emissions changes with this method. 

Other sources of recent warming

We have shown how recent aerosol emissions reductions in East Asia may have had a strong influence on post-2010 elevated rates of surface warming, both globally and in the North Pacific. To put these results in context, we here discuss some other, concurrent changes that we cannot consider in the same framework. 

One anthropogenic factor is the accelerated increase in atmospheric CH₄ concentrations over the same period. As an estimate, using recent global near-surface concentrations from NOAA ²⁶ and the IPCC AR6 ²⁷, combined with the forcing estimation methods from (22), we find a global mean CH₄ radiative forcing (RF) of 0.06 Wm^-2 for the 2010-2023 period, corresponding to a rate of 0.047 Wm^-2 / decade. This is a marked increase over the previous decade (2000-2010), where we find 0.01 Wm^-2 / decade. However, for the full 30-year period of 1980-2010, we estimate a forcing of 0.043 Wm^-2 / decade. This means that while changes in CH₄ atmospheric concentration growth rates may have contributed to decadal variability, and clearly can enhance the overall rate of global warming, the recent decade has not seen markedly strong influence from CH₄ increases compared to recent history. We also note that the above numbers are for RF, while the Effective Radiative Forcing (ERF), and the temperature influences of methane, may be muted due to rapid adjustments (23). 

Another, more recent, anthropogenic factor is the post-2020 reduction in SO2 emissions from the shipping sector, following the recent regulations of the International Maritime Organization. Here, a range of studies have concluded that the global mean ERF from the 80% reduction in emissions, corresponding to around 9 Tg SO₂/year, is in the range of 0.05-0.10 Wm^-2 (10, 11). While studies using ensembles of simulations from fully coupled models do find a surface temperature response to this emissions change over time, the magnitude and detectability relative to internal variability for the years 2020-2023 is still disputed ^21,28. Coming in at the end of the time period studied here, the IMO regulations are unlikely to have had a major influence on the above conclusions regarding the influence of Chinese aerosol emission changes.

ERF estimates are unfortunately not available from all RAMIP models. For those that have delivered the required simulations (see Methods), we find an ERF from the emissions changes in China discussed above ranging from 0.06 – 0.21 Wm^-2 (15).  

Recently, in the context of the 2023 record global mean surface temperatures, Goessling et al. ¹⁵ identified a record-low planetary albedo as a contributing cause. Using the same datasets as here (CERES and ERA5), they highlight low cloud cover over the North Pacific as a key component of this, and note that the role of aerosols in this change is still unclear. Their results are well in line with the present study, both geographically and in terms of physical processes, however their main focus is on reduced cloud fraction in selected years while our results concern a reduced aerosol-cloud induced albedo on a decadal scale. The regional similarity, however, opens the possibility that the results of Goessling et al. are, at least in part, interpretable as a sustained reduction in aerosol-cloud interactions.