2.1 AerChemMIP models

Overall, nine coupled ocean–atmosphere climate models performed the SSP3-7.0 and SSP3-7.0-lowNTCF simulations, including CNRM-ESM2-1 (Séférian et al., 2019; Michou et al., 2019), MIROC6 (Takemura et al., 2005, 2009; Tatebe et al., 2019), MPI-ESM1-2-HAM (Mauritsen et al., 2019; Neubauer et al., 2019; Tegen et al., 2019), NorESM2-LM (Seland et al., 2020), BCC-ESM1 (Wu et al., 2019, 2020), GFDL-ESM4 (John et al., 2018; Horowitz et al., 2018; Dunne et al., 2020; Horowitz et al., 2020), CESM2-WACCM (Emmons et al., 2020; Gettelman et al., 2019; Tilmes et al., 2019), UKESM1-0-LL (Sellar et al., 2019), and MRI-ESM2-0 (Yukimoto et al., 2019). However, the first four models (CNRM-ESM2-1, MIROC6, MPI-ESM1-2-HAM, NorESM2-LM) lack interactive tropospheric chemistry schemes and therefore include identical ozone evolution in both SSP3-7.0 and SSP3-7.0-lowNTCF simulations (as recommended by AerChemMIP). As NTCF mitigation only includes the effects of aerosols in these four models, we refer to these models as “Aer”. The remaining five models, including BCC-ESM1, GFDL-ESM4, CESM2-WACCM, UKESM1-0-LL, and MRI-ESM2-0, include interactive atmospheric chemistry and aerosols, and therefore both aerosol and ozone reductions are included. These models are referred to as “Aer+O3”.

In addition to coupled simulations, models also performed analogous fixed sea surface temperature (SST) experiments to quantify the ERF. The ERF is calculated from the top-of-atmosphere (TOA) flux differences between atmosphere-only simulations with identical SSTs but differing composition (Forster et al., 2016; Pincus et al., 2016). The above scenarios (SSP3-7.0 and SSP3-7.0-lowNTCF) are repeated with prescribed SSTs. These SSTs (and sea ice) are taken from the monthly mean evolving values from one ensemble member of the coupled SSP3-7.0 ScenarioMIP run (Collins et al., 2017). MPI-ESM1-2-HAM used daily mean SST and sea ice. The differences in radiative fluxes between the weak and strong air quality control scenarios yield the TOA transient ERF due to NTCF mitigation.

2.2 Model data and methodology

All models performed at least one realization each of SSP3-7.0 and SSP3-7.0-lowNTCF. CNRM-ESM2-1, MIROC6, UKESM1-0-LL, NorESM2-LM, CESM2-WACCM and BCC-ESM1 performed three realizations of each experiment. For these models, the model mean response (average over the three realizations) is shown. The multi-model mean (MMM) is obtained by averaging each model's mean response (i.e., each model has the same weight). Only one realization exists for the corresponding fixed-SST experiments. Unless otherwise mentioned, all analyses are based on annual means. All data are spatially interpolated to a $\mathrm{2.5}{}^{\circ }\times \mathrm{2.5}{}^{\circ }$ grid using bilinear interpolation.

Model trends are calculated using least-squares regression, and the corresponding trend significance is based on a two-tailed Student's t test, where the null hypothesis of a zero-regression slope is evaluated. Multi-model mean trends and their significance are calculated using two different methods. In the first method, MMM trends are calculated from the multi-model mean time series using a weighted least-squares regression, where each value in the multi-model mean time series is weighted by $\mathrm{1}/{\mathit{\sigma }}_{\mathrm{m}}^{\mathrm{2}}$, where σ_m is the standard deviation across models. We note that the MMM trends and significance are very similar with and without weighting the regression. Autocorrelation of the time series is also accounted for by using the effective sample size, defined as $n(\mathrm{1}-r_{\mathrm{1}})/(\mathrm{1}+r_{\mathrm{1}})$, where n is the number of years and r₁ is the lag−1 autocorrelation coefficient.

We also quantify the significance of the multi-model mean trend relative to each individual model mean trend. Here, the MMM trend is calculated as the average of the individual model mean trends and its uncertainty is calculated as plus/minus twice the standard error (i.e., the 95 % confidence interval), which is $\mathrm{2}\mathit{\sigma }/\sqrt{n_{\mathrm{m}}}$, where σ is the standard deviation of the trends and n_m is the number of models. If this confidence interval does not include zero, then the multi-model mean trend is significant at the 95 % confidence level. Both methods yield similar conclusions as to the magnitude and significance of the MMM trends.

We note that there are several sources of uncertainty, including model differences as well as internal climate variability. Our quoted uncertainties include both. However, if we had three realizations for each model, the role of internal climate variability could be better isolated. Although three realizations are probably not enough to truly quantify this source of uncertainty, which is the goal of large ensemble projects. Similarly, additional models would also allow improved quantification of the uncertainty due to model differences. For a given variable that we analyze, the uncertainty across realizations for models with multiple runs is comparable to the uncertainty across models. For example, in terms of the total global surface temperature change (Sect. 3), we quote a multi-model mean 95 % confidence interval of 0.12 K. Models with multiple realizations yield corresponding values of 0.02, 0.08, 0.10, 0.11, 0.15, and 0.22, which yields an average of 0.11 K. Similarly, in terms of the total global land PM_2.5 change, we quote a multi-model mean 95 % confidence interval of 0.32 µg m⁻³. Models with multiple realizations yield corresponding values of 0.08, 0.09, 0.16, 0.25, and 0.29, which yields an average of 0.17 µg m⁻³. Furthermore, if all models and realizations are used, our uncertainty estimates are reduced (due, in part, to more data). For example, the 95 % confidence interval for total global surface temperature change decreases from the quoted 0.12 to 0.07 K. The 95 % confidence interval for the total global land PM_2.5 change decreases from 0.32 to 0.24 µg m⁻³.

Climate variables analyzed include monthly mean surface temperature (T_s) and precipitation (Precip). Surface temperature and precipitation are analyzed as these are arguably two of the most important climate variables. Changes in surface temperature are particularly relevant in the context of climate mitigation, as the goal of the Paris Agreement is to keep the increase in global mean surface temperature to well below 2 ^∘C above preindustrial values (IPCC, 2018). Precipitation, and fresh water resources in general, is important to both human society and ecosystems.

As discussed in the introduction, both PM_2.5 and ozone are commonly used indicators of air quality, and both have been associated with adverse human health impacts (WHO, 2016; Cohen et al., 2017; Butt et al., 2017). Air quality is therefore quantified from surface PM_2.5 and surface O₃. These monthly mean fields are obtained from the model level closest to the surface. Unfortunately, few models archived sub-monthly aerosol or ozone data, so we are unable to analyze changes in daily or sub-daily maximum PM_2.5 or O₃ pollution. Furthermore, only four models directly archive PM_2.5 (with differing methodologies), and not all models include the same aerosol species (e.g., nitrate aerosol; see the Supplement). Thus, we approximate PM_2.5 in all models using the following equation (Fiore et al., 2012; Silva et al., 2017): PM_2.5= BC + OA + SO₄ + 0.1× DU + 0.25× SS, where BC is black carbon, OA is organic aerosol, SO₄ is sulfate aerosol, DU is dust, and SS is sea salt. This formula assumes 100 % of the BC, OA, and SO₄ is fine mode, whereas 25 % of the sea salt and 10 % of the dust is fine mode. The SS and DU factors will be dependent on the model and its size distribution. In the case of CNRM-ESM2-1, sensitivity tests were used to estimate a much smaller SS factor of 0.01. This smaller factor addresses the large SS size range of up to 20 µm in this model (Pierre Nabat, personal communication, 27 November 2019). Although this approach likely introduces some uncertainties (see Sect. 3.1), it provides first and foremost an estimate of PM_2.5 for all models, as well as a consistent estimate for all models.

CMIP6 model evaluation of air quality metrics, including surface O₃ and PM_2.5 (as approximated here), is quantified in a companion paper (Turnock et al., 2020). To summarize, CMIP6 models generally underestimate PM_2.5 over most regions relative to ground based observations from the Global Aerosol Synthesis and Science Project (GASSP) (Reddington et al., 2017). This in part is due to the absence of nitrate aerosol, and may also be related to misrepresentation of secondary organic aerosol. A similar PM_2.5 underestimation occurs over Europe and North America relative to the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA2) aerosol reanalysis product (Buchard et al., 2017; Randles et al., 2017). In contrast, CMIP6 models overestimate PM_2.5 relative to MERRA2 over south and east Asia, contrary to the evaluation using GASSP observations. Compared to surface O₃ measurements from Tropospheric Ozone Assessment Report (TOAR) (Schultz et al., 2017), CMIP6 models consistently overestimate surface ozone during both summer and winter across most regions, potentially due to the coarse resolution of global models simulating excess O₃ production.

Perhaps more important than changes in the mean of a climate variable are changes in its extremes. Heat waves, for example, are a major cause of weather-related fatalities. Thus, we also analyze climate extremes including the hottest day (monthly maximum value of daily maximum surface temperature), wettest day (monthly maximum 1 d surface precipitation), and consecutive dry days (CDDs), defined as the maximum annual number of consecutive days with surface precipitation <1 mm d⁻¹. We focus on these three extreme indices since they are frequently used metrics for temperature and precipitation extremes. Prior observational analyses have shown significant increases in the hottest and wettest days, and decreases in CDDs over the latter half of the 20th century (Donat et al., 2013a, b). Climate extremes are based on daily data and are calculated at each grid box and then spatially averaged.

2.3 Future scenarios: SSP3-7.0 and SSP3-7.0-lowNTCF

As part of ScenarioMIP, a set of future emissions pathways has been developed for CMIP6 (Eyring et al., 2016). These scenarios, referred to as Shared Socioeconomic Pathways (SSPs) (O'Neill et al., 2014; van Vuuren et al., 2014; Gidden et al., 2019), link socioeconomic and technological innovation to provide future trajectories of emissions, including different levels of controls on air quality pollutants. The medium strength of pollution control corresponds to current legislation (CLE) until 2030 and progresses three-quarters of the way towards maximum technically feasible reduction (MTFR) thereafter. Strong pollution control exceeds CLE and progresses ultimately towards MTFR. Weak pollution control assumes delays to the implementation of CLE and makes less progress towards MTFR than the medium scenario (Rao et al., 2017). The rate of progress is different for high-, medium-, and low-income countries. By encompassing a wide range of possible futures, these scenarios provide a large sample space of potential emissions through the 21st century.

To detect the largest signal, AerChemMIP uses the SSP3-7.0 “regional rivalry” without climate policy (∼7.0 W m⁻² at 2100) (Fujimori et al., 2017) as the reference scenario, which has the highest levels of NTCFs and “weak” levels of air quality control measures (O'Neill et al., 2014; Rao et al., 2017). The perturbation scenario SSP3-7.0-lowNTCF uses the same socioeconomic scenario but with “strong” levels of air quality control measures (Gidden et al., 2019). Basically, the emissions drivers (population, GDP, energy, and land use) are based on SSP3, but the emissions factors of air pollutants that are related to NTCFs are associated with a sustainability pathway represented by SSP1 in conjunction with the stringent climate policy equivalent of stabilizing the radiative forcing to around 2.6 W m⁻². Assumptions include the following: SSP3-7.0-lowNTCF can reduce CH₄ as if SSP1's stringent climate mitigation policy is implemented in the SSP3 world; SSP1's air pollutant legislation and technological progress can be achieved in the SSP3 world; other species (e.g., CFCs, HFCs, and SF₆) are identical to the SSP3 baseline. Although methane reductions are included in the lowNTCF scenario, they are not included in the lowNTCF experiment. This allows quantification of the aerosol and ozone effects alone; similar SSP3-7.0-lowNTCF experiments that are analogous to those presented here but also include the methane reductions will also be performed and analyzed in subsequent work. Differences between these two scenarios are designed to evaluate a SSP3 world in which NTCF-related policies are enacted in the absence of other GHG-related climate policies. Moreover, our results (e.g., the magnitude of the surface temperature increase) represent an upper bound as our baseline scenario lacks climate policy and contains the highest levels of NTCFs.

Differences in climate, effective radiative forcing, chemical composition, and air quality between the two scenarios will be solely due to the alternative air quality control measures. These experiments cover the time frame from 2015 to 2055, as this is when reductions in aerosol and ozone precursor emissions are expected to be significant, particularly in some world regions. Here, we define NTCF mitigation as the difference between the strong (low NTCF) and weak (high NTCF) air quality control scenarios (i.e., SSP3-7.0-lowNTCF minus SSP3-7.0). Although methane reductions are included in the strong air quality control scenario, AerChemMIP protocol specifies unchanged levels of well-mixed greenhouse gases (WMGHGs), including methane, between the strong and the weak air quality control simulations (Collins et al., 2017). Thus, our results quantify non-methane NTCF mitigation (aerosols and ozone only).

Figure 1 shows the 2015–2055 global mean time series of CO₂, aerosol species, and gaseous precursor emissions for SSP3-7.0 (weak air quality control) and SSP3-7.0-lowNTCF (strong air quality control). Emissions shown here come directly from the CMIP6 forcing datasets, which were downloaded from the input datasets for Model Intercomparison Project (input4MIPS) served by the Earth System Grid Federation. We note that the emissions data are decadal after 2015, with monthly values for the years 2015, 2020, 2030, 2040, 2050, 2060, etc. We estimate the emissions in 2055 as the mean of the emissions in 2050 and 2060 at each grid box. Only weak air quality control CO₂ and CH₄ emissions are shown, as AerChemMIP simulations include the same change in CO₂ and CH₄ emissions based on the weak air quality control scenario. By 2055, CO₂ and CH₄ increase by 65 % and 50 % (relative to 2015), respectively. In contrast to CO₂ and CH₄, however, very different non-methane NTCF evolution occurs. Under weak air quality control, global emissions of all aerosols and gaseous precursors (except SO₂) increase by 5 %–15 % by 2055. In contrast, strong air quality control yields strong emission reductions in all species, ranging from ∼30 % for VOCs to 55 % for SO₂. Thus, NTCF mitigation (SSP3-7.0-lowNTCF – SSP3-7.0) yields emission reductions of all aerosols and gaseous precursors by ∼40 %–55 %.

Figure 1The 2015–2055 global mean CO₂, NTCF, and precursor gas emissions. Panels show (a) black carbon (BC); (b) sulfur dioxide (SO₂); (c) organic carbon (OC); (d) carbon monoxide (CO); (e) nitrogen oxides (NO_x); (f) volatile organic compounds (VOC); (g) methane (CH₄); and (h) carbon dioxide (CO₂) emissions for weak (red) and strong (blue) air quality control. Also included is the percent change (relative to 2015) for weak (red solid) and strong (blue solid) air quality control, and NTCF mitigation (black solid). Emission units for species X are Mt X yr⁻¹. Percent change units are %. Only weak air quality control CO₂ and CH₄ emissions are shown, as AerChemMIP simulations include the same change in CO₂ and CH₄ emissions based on the weak air quality control scenario. Emissions data come directly from the CMIP6 forcing datasets, which were downloaded from the input datasets for Model Intercomparison Project (input4MIPS).

Download

The corresponding 2015–2055 regional emission trends (relative to 2015) are shown in Fig. 2. As with climate and air quality trends, emission trends are estimated using least-squares regression. Consistent with the global mean time series of emissions (Fig. 1), CO₂ emissions increase under weak air quality control (and in both sets of AerChemMIP simulations), with larger increases in south and north Africa, south Asia, and southeast Asia. Similarly, CH₄ emissions increase in all world regions under weak air quality control (and in both sets of AerChemMIP simulations), with larger increases in south and north Africa, and south Asia. Most world regions also show increases in BC, SO₂, and organic carbon (OC) under weak air quality control but strong decreases under strong air quality control. NTCF mitigation (strong minus weak air quality control) shows large (∼20 % decade⁻¹) BC decreases in Central America, central and north Asia, east Asia, and southeast Asia. Most world regions exhibit a 10 %–20 % decade⁻¹ reduction in SO₂ emissions under NTCF mitigation, with a large decrease in south Asia at −28 % decade⁻¹. Similarly, OC and CO emissions decrease by ∼10 %–20 % decade⁻¹. Relatively large OC reductions also occur in east Asia, south Asia, and southeast Asia. NO_x and VOC emissions also decrease in all world regions under NTCF mitigation (although this is only a decrease relative to non-mitigated emissions for NO_x in south Asia and for VOC in east Asia).

Figure 2The 2015–2055 regional mean CO₂, NTCF, and precursor gas emission trends. Regional 2015–2055 emission trends for (a) black carbon (BC); (b) sulfur dioxide (SO₂); (c) organic carbon (OC); (d) carbon monoxide (CO); (e) nitrogen oxides (NO_x); (f) volatile organic compounds (VOCs); (g) carbon dioxide (CO₂); and (h) methane (CH₄) for weak (red asterisks) and strong air quality control (SSP3-7.0-lowNTCF; blue triangles) and NTCF mitigation (SSP3-7.0-lowNTCF−SSP3-7.0; black diamonds). The center map shows the corresponding color-coded world regions, based on Seneviratne et al. (2012). The following abbreviations are used: Canada: 1 (Can; black), United States: 2 (US; magenta), Central America: 3 (cAm; sky blue), South America: 4 (sAm; purple), south Africa: 5 (sAf; yellow), north Africa: 6 (nAf; green), Europe: 7 (Eu; pink), central and north Asia: 8 (cnA; orange), east Asia: 9 (eA; navy), south Asia: 10 (sA; red), southeast Asia: 11 (seA; gray), and Australia: 12 (Au; beige). The average over these 12 land regions is abbreviated as “L”. Trend units are % decade⁻¹ (relative to 2015). Only weak air quality control CO₂ and CH₄ emission trends are shown, as AerChemMIP simulations include the same change in CO₂ and CH₄ emissions based on the weak air quality control scenario. Emissions data come directly from the CMIP6 forcing datasets, which were downloaded from input4MIPS.

3.1 Global climate and air quality trends

Figure 3 shows the 2015–2055 global annual mean time series for air quality under NTCF mitigation. By design, NTCF mitigation leads to significant decreases in air pollution, in terms of both surface PM_2.5 and O₃. All models yield significant global mean decreases in both quantities, with an overall MMM decrease of −0.23 µg m⁻³ decade⁻¹ for PM_2.5 and −1.19 ppb decade⁻¹ for O₃ (Table 1). Over the 2015–2055 time period, these rates of change correspond to global mean decreases of −0.92 µg m⁻³ and −4.76 ppb, respectively. Larger PM_2.5 decreases occur over land only at −2.20 µg m⁻³, whereas similar O₃ decreases occur over land only at −4.55 ppb. Similar PM_2.5 trends occur in Aer+O3 and Aer models over land only (−0.59 versus −0.44 µg m⁻³ decade⁻¹, respectively), as well as over both land and ocean (−0.26 versus −0.16 µg m⁻³ decade⁻¹, respectively). Note that the MMM over all models for O₃ does not include Aer models, as they yield negligible change in surface ozone (by design).

Figure 3The 2015–2055 time series of global annual mean air pollution due to NTCF mitigation. Panels show (a) surface particulate matter (PM_2.5) (µg m⁻³) and (b) surface ozone (ppb) for NTCF mitigation. The multi-model mean time series and the corresponding trend estimated using a weighted least-squares regression are included as thick black lines. The multi-model mean (MMM) trend, its significance, and R² value are also included, as is the 95 % confidence interval (CI). Individual model mean trends are also included as defined by the legend.

Download

Table 1Air pollution and climate responses to NTCF mitigation. Annual mean 2015–2055 trends in surface particulate matter (PM_2.5), ozone (O₃), surface temperature (T_s), precipitation (Precip), hottest day, wettest day, CDDs, and the effective radiative forcing (ERF) for NTCF mitigation. First set of numbers is the global mean trend; second set of numbers is the land-only trend. Trends significant at the 95 % confidence level are denoted by bold font based on a t test. Trend units are K decade⁻¹ for T_s and hottest day; mm d⁻¹ decade⁻¹ for Precip and wettest day; µg m⁻³ decade⁻¹ for PM_2.5; ppb decade⁻¹ for O₃; days per year decade⁻¹ for CDDs; and W m⁻² decade⁻¹ for ERF. The first five models include both aerosol and ozone changes (Aer+O3 models); bottom four models include only aerosol changes (Aer models). MMM is the multi-model mean and the last row (“MMM total”) shows the total change over the entire 2015–2055 time period based on all models. n/a indicates not applicable.

Download Print Version | Download XLSX

As mentioned in the methods section, as only four models directly archive PM_2.5 (with differing methodologies), and not all models include the same aerosol species (see the Supplement), we approximate PM_2.5. Comparing estimated PM_2.5 trends to those from the actual PM_2.5 as calculated and archived by four models (GFDL-ESM4, NorESM2-LM, MRI-ESM2-0, and MPI-ESM1-2-HAM) yields reasonably good results. The global annual multi-model mean trend in estimated (actual) PM_2.5 for this four-model subset is −0.24 (−0.28) µg m⁻³ decade⁻¹ (Fig. S1). Over land only, the corresponding trends are −0.56 (−0.65) µg m⁻³ decade⁻¹. Thus, the estimated global mean and land-only PM_2.5 trends are about 85 % as large as those based on archived PM_2.5 (underestimation by a similar amount exists in all four models, with the largest underestimation in GFDL-ESM4). Larger differences exist in some world regions, particularly south Asia, where the estimated (actual) PM_2.5 is $-\mathrm{4.08}\pm \mathrm{0.70}$ ($-\mathrm{4.71}\pm \mathrm{1.36}$) µg m⁻³ decade⁻¹ (Fig. S2 in the Supplement). Some of this underestimation is due to the aforementioned lack of nitrate and ammonium aerosol in our estimated PM_2.5. However, other factors also contribute, as the estimated PM_2.5 trends in all four models underestimate the actual PM_2.5 trends, but not all of these models include nitrate and ammonium species.

GFDL-ESM4 is the lone model that archived nitrate aerosol data. Globally (over land only), nitrate decreases by −0.04 (−0.12) µg m⁻³ decade⁻¹, with maximum decrease over east Asia and in particular south Asia (Fig. S3). These trends are 17 % and 20 % (13 % and 15 %) of the magnitude of the estimated (actual) global and land-only PM_2.5 trend. GFDL-ESM4 also archives ammonium, and similar changes occur (Fig. S3). Globally (over land only), ammonium decreases by −0.05 (−0.12) µg m⁻³ decade⁻¹, with maximum decreases over both south Asia and east Asia. These trends are 21 % and 20 % (16 % and 15 %) of the magnitude of the estimated (actual) global and land-only PM_2.5 trend. Thus, excluding nitrate and ammonium in GFDL-ESM4 leads to ∼30 %–40 % underestimation of the global and land-only PM_2.5 trend. The relatively large decreases in nitrate and ammonium in south Asia helps to explain the relatively large difference in estimated and actual PM_2.5 trend in this region (Fig. S2). In addition to GFDL-ESM4, CESM2-WACCM also archives ammonium (Fig. S3). Here, however, the global and land-only ammonium trends are an order of magnitude smaller than those in GFDL-ESM4, which leads to ∼1 % underestimation of the corresponding (estimated) PM_2.5 trends.

The 2015–2055 global annual mean time series for climate variables under NTCF mitigation are shown in Fig. 4. All but one model (MIROC6) show significant global annual mean surface warming in response to NTCF mitigation (Table 1 lists the trends for each model). Averaged over all models, global mean surface warming is 0.06 K decade⁻¹, or 0.25 K over the 2015–2055 time period (Table 1). We note that this warming will continue past 2055, as these transient simulations have not reached radiative equilibrium. Similar conclusions exist over land only, where the multi-model mean (MMM) warming is even larger at 0.36 K over the entire time period (Table 1). Enhanced land warming is consistent with the land-sea warming contrast (Sutton et al., 2007; Joshi et al., 2008), which may also act to increase aerosol burden itself (Allen et al., 2016, 2019b), implying a climate change penalty to air quality. Interestingly, models that include both aerosol and ozone reductions (Aer+O3) yield similar surface warming relative to the models that include aerosol reductions (Aer) alone (0.07 versus 0.06 K decade⁻¹, respectively). Although this could be due to several factors (small sample size, internal climate variability, different model parameterizations, feedbacks, etc.), it suggests weak surface cooling due to reductions in ozone. Such an interpretation is consistent with the negative forcing from aerosol increases dominating the positive forcing due to ozone increases over the historical period (Naik et al., 2013). Simulations with a single model, running both coupled and uncoupled chemistry experiments, would help isolate this effect.

Figure 4The 2015–2055 time series of global annual mean climate variables due to NTCF mitigation. Panels show (a) surface temperature (K); (b) hottest day (K); (c) precipitation (mm d⁻¹); (d) wettest day (mm d⁻¹); (e) consecutive dry days (annual number of days); and (f) ERF (W m⁻²) for NTCF mitigation. The multi-model mean time series and the corresponding trend estimated using a weighted least-squares regression are included as thick black lines. The multi-model mean (MMM) trend, its significance, and R² value are also included, as is the 95 % confidence interval (CI). Individual model mean trends are also included as defined by the legend.

Download

Warming in response to NTCF mitigation is consistent with the corresponding increase in ERF. All but two models (BCC-ESM1, GFDL-ESM4) yield a significant increase in ERF, with a MMM of 0.44 W m⁻² over the entire time period (Table 1). Over land only, this increases to 0.59 W m⁻². Although not significant, Aer+O3 models yield a weaker trend in global mean ERF than Aer models, at 0.07 versus 0.17 W m⁻² decade⁻¹. This is consistent with ozone reductions driving a decrease in ERF in Aer+O3 models, offsetting part of the ERF increase due to aerosol reductions (Turnock et al., 2019).

All models also yield a significant increase in global annual mean precipitation (Table 1), with an overall MMM of 0.008 mm d⁻¹ decade⁻¹. Aer+O3 and Aer models yield similar increases in global mean precipitation at 0.009 and 0.005 mm d⁻¹ decade⁻¹, respectively. Somewhat less robust results occur over land only. Although all models yield an increase in precipitation over land, it is only significant in four models.

Similar but less robust responses also occur in climate extremes, particularly those based on precipitation. Globally significant increases in the surface temperature of the hottest day occur in all but one model (MIROC6 is the exception). The multi-model mean also yields a significant trend at 0.06 K decade⁻¹. The wettest day significantly increases in about half of the models and in the overall MMM at 0.053 mm d⁻¹ decade⁻¹. A mixed signal exists for CDDs, with four models yielding a positive trend and three models yielding a negative trend. The overall MMM yields 0.08 days per year decade⁻¹ but lacks significance.

Thus, from a global mean perspective, NTCF mitigation leads to significant improvements in air quality based on both PM_2.5 and O₃ but also significant climate change in most metrics. This includes increases in surface temperature and precipitation, as well as corresponding increases in most climate extremes, particularly the hottest day and to a lesser extent the wettest day. Except for surface temperature and the hottest day, less robust results generally occur over land only. CDD yields a mixed signal, with lack of significance in the multi-model mean.

3.2 Regional climate and air quality trends

Figure 5 shows the regional climate and air pollution trends for weak and strong air quality control and the effect of NTCF mitigation. We include both Aer and Aer+O3 models in this analysis to maximize the signal-to-noise ratio (except for ozone changes). The aforementioned response differences between these two model subsets are generally not significant. Consistent with increased aerosol and precursor gas emissions (Figs. 1–2), air quality metrics generally show significant increases under weak air quality control, particularly O₃ where all 12 world regions exhibit an increase. In contrast, strong air quality control yields decreases in both PM_2.5 and O₃ for nearly all world regions. The overall effect of NTCF mitigation is thus a robust decrease in air pollution, in terms of both PM_2.5 and O₃, over all 12 world regions, as well as the Arctic, Northern Hemisphere (NH) midlatitudes, and tropics. Over all land surfaces, the PM_2.5 decrease is $-\mathrm{0.55}\pm \mathrm{0.08}$ µgm⁻³ decade⁻¹. Regionally, decreases in PM_2.5 range from $-\mathrm{0.05}\pm \mathrm{0.01}$ µgm⁻³ decade⁻¹ over Canada to $-\mathrm{3.8}\pm \mathrm{0.69}$ µgm⁻³ decade⁻¹ in south Asia. Relatively large PM_2.5 decreases also occur over east Asia, southeast Asia, and north Africa at $-\mathrm{2.1}\pm \mathrm{0.27}$ µgm⁻³ decade⁻¹, $-\mathrm{0.78}\pm \mathrm{0.16}$ µgm⁻³ decade⁻¹, and $-\mathrm{0.82}\pm \mathrm{0.20}$ µgm⁻³ decade⁻¹, respectively. The relatively large PM_2.5 decreases over east Asia, southeast Asia, and south Asia are generally consistent with the relatively large reductions in aerosol species, including BC, SO₄, and OC (Fig. 2).

Figure 5Regional climate and air pollution responses to NTCF mitigation. Bar plots show regional 2015–2055 trends in (a) surface temperature (T_s); (b) hottest day; (c) precipitation (Precip); (d) wettest day; (e) CDDs; (f) ERF; (g) surface particulate matter (PM_2.5); and (h) ozone (O₃) for weak (red) and strong (blue) air quality control, and NTCF mitigation (black). Bar center (gray horizontal line) shows the multi-model mean trend, estimated as the average of each model's mean trend. Bar length represents the 95 % confidence interval, estimated as $\mathrm{2}\mathit{\sigma }/\sqrt{n}$, where σ is the standard deviation of the individual model mean trends and n is the number of models. Center map shows the corresponding color-coded world regions for each bar plot (as in Fig. 2). The average over these 12 land regions is abbreviated as “L”. Also included is the Arctic (“A”; 60–90^∘ N; light blue hatched region); NH midlatitudes (“N”; 30–60^∘ N; yellow hatched region); tropics (“T”; 30^∘ S–30^∘ N; beige hatched region); SH midlatitudes (“S”; 60–30^∘ S; red hatched region); and the global mean (“G”). Trend units are K decade⁻¹ for T_s and hottest day; mm d⁻¹ decade⁻¹ for Precip and wettest day; µg m⁻³ decade⁻¹ for PM_2.5; ppb decade⁻¹ for O₃; days per year decade⁻¹ for CDDs; and W m⁻² decade⁻¹ for ERF.

Similar results exist for O₃, with a robust decrease over land of $-\mathrm{1.11}\pm \mathrm{0.22}$ ppb decade⁻¹. Regionally, O₃ decreases range from $-\mathrm{2.41}\pm \mathrm{0.33}$ ppb decade⁻¹ over Central America and $-\mathrm{1.97}\pm \mathrm{0.20}$ ppb decade⁻¹ over southeast Asia to $-\mathrm{0.86}\pm \mathrm{0.11}$ ppb decade⁻¹ over Australia. Relatively large O₃ decreases also occur over south Asia ($-\mathrm{1.55}\pm \mathrm{0.93}$ ppb decade⁻¹), as well as north Africa ($-\mathrm{1.7}\pm \mathrm{0.25}$ ppb decade⁻¹). Notably, a weak O₃ decrease occurs in east Asia ($-\mathrm{0.45}\pm \mathrm{0.51}$ ppb decade⁻¹), which may be related to relatively weak VOC reductions (Fig. 2). In addition to significant reduction in the Arctic, the other latitudinal bands also exhibit significant reductions in O₃.

Over all 12 world regions, significant surface warming occurs in both the weak and strong air quality control scenarios, due to continued increases in CO₂ (and CH₄). More importantly, NTCF mitigation – due to reduced cooling from reductions in non-absorbing aerosol (e.g., sulfate) – also yields significant warming, with a significant increase in land-only surface temperature of 0.09±0.02 K decade⁻¹. Significant warming also occurs in all but one world region (Australia is the lone exception) due to NTCF mitigation, ranging from 0.05±0.02 K decade⁻¹ over southeast Asia to 0.16±0.05 K decade⁻¹ over central and north Asia. Relatively large warming also occurs over east Asia (0.11±0.05 K decade⁻¹) and south Asia (0.12±0.04 K decade⁻¹; see also Fig. S4). Furthermore, large warming of the Arctic (60–90^∘ N) occurs (0.15±0.09 K decade⁻¹), particularly in the East Siberian and Beaufort seas, north of western Canada/Alaska and around the Canadian Arctic Archipelago (Fig. 6). This result is consistent with recent studies showing high Arctic sensitivity to aerosol reductions (Acosta Navarro et al., 2016; Lewinschal et al., 2019; Westervelt et al., 2020). Other latitudinal bands also significantly warm, including the NH midlatitudes (30–60^∘ N), tropics (30^∘ S–30^∘ N), and Southern Hemisphere (SH) midlatitudes (60–30^∘ S) at 0.10±0.03, 0.05±0.01, and 0.03±0.02 K decade⁻¹, respectively.

Figure 6The 2015–2055 annual mean surface temperature trends and model trend realization agreement over the Arctic. Surface temperature (a–c) trends (K decade⁻¹) and (d–f) model trend realization agreement (%) for (a, d) weak air quality control, (b, e) strong air quality control, and (c, f) NTCF mitigation. Stippling denotes trend significance at the 95 % confidence level based on a standard t test. Trend realization agreement represents the percentage of models that agree on the sign of the trend. Red colors indicate model agreement on a positive trend; blue colors indicate model agreement on a negative trend. White areas indicate lack of agreement on the sign of the trend.

Warming is consistent with the increase in ERF, with most world regions yielding significant positive ERF trends. Little correspondence exists between regions that warm the most and their ERF trend. This is not necessarily surprising, as forcing and response do not need to occur in the same regions, due to climate feedbacks, remote teleconnection and other processes. For example, central and north Asia and the Arctic warm the most, but there is not a particularly large increase in their regional ERF. Similarly, southeast Asia warms the least, but this region features a relatively large ERF increase.

Significant increases in the hottest day also occur, with larger increases under strong relative to weak air quality controls. NTCF mitigation yields significant increases in the hottest day for all but two world regions (Australia and south America are the exceptions). A significant increase in the hottest day also occurs over all land regions (0.09±0.03 K decade⁻¹), with five of six models yielding a significant increase (Table 1; MIROC6 is the exception). Thus, NTCF mitigation unmasks the warming due to GHG increases resulting in robust increases in both surface air temperature and the hottest day over nearly all world regions. We note that the lone area with cooling is the north Atlantic (around Iceland and southwest of Svalbard; Fig. 6), which may be associated with a weakening of the Atlantic Meridional Overturning Circulation (AMOC) (Delworth and Dixon, 2006; Cai et al., 2006; Menary et al., 2013). Figure 6d–f show that this cooling is a robust feature, with ∼70 % of the models yielding cooling here.

Over all land surfaces, a significant precipitation increase also occurs in both scenarios at 0.012±0.005 and 0.022±0.006 mm d⁻¹ decade⁻¹ under weak and strong air quality control, respectively. Thus, NTCF mitigation – by unmasking GHG-induced warming – also yields a significant increase in land precipitation at 0.011±0.003 mm d⁻¹ decade⁻¹. The effect of NTCF mitigation on precipitation over individual world regions, however, has mixed significance and ranges from 0.003±0.035 mm d⁻¹ decade⁻¹ over Australia to 0.044±0.022 mm d⁻¹ decade⁻¹ over south Asia. Note that some world regions exhibit decreases in precipitation under both weak and strong air quality control (e.g., Central America), such that NTCF mitigation yields a weaker decrease (as opposed to an absolute increase). In addition to south Asia, a significant precipitation increase also occurs over central and north Asia (0.008±0.005 mm d⁻¹ decade⁻¹), east Asia (0.038±0.014 mm d⁻¹ decade⁻¹), and the Arctic (0.010±0.005 mm d⁻¹ decade⁻¹). Although southeast Asia also exhibits a relatively large increase in precipitation, it is not significant (0.019±0.041 mm d⁻¹ decade⁻¹). Both south and north Africa yield precipitation increases, but the bulk of the African precipitation increase occurs over east Africa (Fig. S5). From a latitudinal perspective, in addition to the Arctic, the NH midlatitudes, tropics, and SH midlatitudes all experience a significant increase in precipitation at 0.010±0.005, 0.011±0.001, and 0.004±0.001 mm d⁻¹ decade⁻¹, respectively. Thus, NTCF mitigation generally increases precipitation in most world regions (although, in some regions, this is a smaller decrease) but the signal is less robust than that for surface temperature. Furthermore, in agreement with prior studies (Levy et al., 2013; Westervelt et al., 2017; Zhao et al., 2018; Westervelt et al., 2018; Scannell et al., 2019), precipitation increases in several monsoon regions, including east Africa, south Asia, and east Asia.

Precipitation extremes, including the wettest day and in particular CDD, also exhibit regional uncertainty under NTCF mitigation, with most regions lacking a robust response. Similar to the significant increases in mean precipitation, significant increases in the wettest day also occur in central and north Asia, east Asia, south Asia, and the Arctic. The NH midlatitudes and tropics (but not the SH midlatitudes) also experience a robust increase in the wettest day. NTCF mitigation also yields robust CDD increases in south America and south Africa, and robust CDD decreases in Canada and the Arctic. Outside of the Arctic, no other latitudinal bands yield a robust CDD response under NTCF mitigation.

3.3 Seasonal climate and air quality trends

Figure 7 shows the regional surface temperature, precipitation and air quality responses during June–July–August (JJA) and December-January-February (DJF). Seasonal air pollution, including both O₃ and PM_2.5, exhibits robust decreases in nearly all world regions under NTCF mitigation. Over land regions, slightly larger O₃ decreases occur during JJA relative to DJF, at $-\mathrm{1.41}\pm \mathrm{0.16}$ and $-\mathrm{0.86}\pm \mathrm{0.34}$ ppb decade⁻¹, respectively. This seasonal contrast is more pronounced over the NH midlatitudes, where the JJA (DJF) decrease is $-\mathrm{1.87}\pm \mathrm{0.17}$ ($-\mathrm{0.72}\pm \mathrm{0.52}$) ppb decade⁻¹. In contrast, slightly larger PM_2.5 decreases occur during DJF relative to JJA, at $-\mathrm{0.67}\pm \mathrm{0.12}$ µg m⁻³ decade⁻¹ and $-\mathrm{0.48}\pm \mathrm{0.08}$ µg m⁻³ decade⁻¹, respectively. As with the annual mean, the largest JJA and DJF O₃ reductions occur over Central America and southeast Asia (and north Africa during DJF). The lone regional increase in O₃ occurs during DJF in east Asia at 0.95±0.52 ppb decade⁻¹. The largest JJA decreases in PM_2.5 occur in east Asia ($-\mathrm{1.64}\pm \mathrm{0.31}$ µg m⁻³ decade⁻¹) and south Asia ($-\mathrm{1.67}\pm \mathrm{0.32}$ µg m⁻³ decade⁻¹). These regions also exhibit large DJF decreases in PM_2.5, particularly south Asia at $-\mathrm{5.55}\pm \mathrm{1.2}$ µg m⁻³ decade⁻¹.

Figure 7Regional climate and air pollution seasonal responses to NTCF mitigation. Bar plots show regional 2015–2055 June–July–August (JJA; left panels) and December–January–February (DJF; right panels) trends in (a–b) surface temperature (T_s); (c–d) precipitation (Precip); (e–f) surface ozone (O₃); and (g–h) surface particulate matter (PM_2.5) for weak (red) and strong (blue) air quality control, and NTCF mitigation (black). Bar center (gray horizontal line) shows the multi-model mean trend, estimated as the average of each model's mean trend. Bar length represents the 95 % confidence interval, estimated as $\mathrm{2}\mathit{\sigma }/\sqrt{n}$, where σ is the standard deviation of the individual model mean trends and n is the number of models. World regions are identical to those in Figure 5. Trend units are K decade⁻¹ for T_s; mm d⁻¹ decade⁻¹ for Precip; µg m⁻³ decade⁻¹ for PM_2.5; and ppb decade⁻¹ for O₃.

Download

NTCF mitigation yields similar warming in both seasons (see also Figs. S6–S7). Over all land surfaces, JJA warming is 0.09±0.02 K decade⁻¹; DJF warming is 0.09±0.04 K decade⁻¹. Consistent with the annual mean warming, relatively large JJA warming also occurs in central and north Asia (0.16±0.06 K decade⁻¹), south Asia (0.10±0.05 K decade⁻¹), and east Asia (0.10±0.06 K decade⁻¹), as well as Canada (0.14±0.05 K decade⁻¹). DJF warming is largest in similar regions as JJA, including central and north Asia and south Asia (0.20±0.12 and 0.13±0.04 K decade⁻¹) and east Asia (0.13±0.12 K decade⁻¹). Arctic warming is most pronounced during DJF, where the rate of warming is about double that during JJA (0.23±0.16 versus 0.12±0.05 K decade⁻¹). Similar JJA and DJF warming occurs for the NH midlatitudes (0.11 versus 0.10±0.03 K decade⁻¹), tropics (0.05 versus 0.04±0.02 K decade⁻¹), and SH midlatitudes (0.03±0.02 K decade⁻¹ for both seasons). As with the annual mean warming, central and north Asia, east Asia, and south Asia generally warm the most during JJA and DJF, with large Arctic warming during DJF.

Regional seasonal precipitation responses continue to exhibit relatively large uncertainty, as most world regions lack a robust response (see also Figs. S8–S9). Central and north Asia, east Asia, and south Asia yield robust JJA increases in precipitation under NTCF mitigation at 0.015±0.008, 0.053±0.034, and 0.089±0.047 mm d⁻¹ decade⁻¹, respectively. The increase in south and east Asia precipitation is consistent with aerosol reductions driving enhanced monsoonal flow. Interestingly, there is also a significant increase in south Asian precipitation during DJF. Canada and north Africa in particular also exhibit robust increases in DJF precipitation. As with the annual mean, most of the increase in DJF precipitation over Africa occurs in east Africa (Fig. S9).

Similar results generally exist for the other seasons, March–April–May (MAM) and September–October–November (SON) (Fig. S10). The largest decrease in O₃ occurs in Central America, south Asia, and southeast Asia, as well as north Africa. The largest PM_2.5 decreases occur in east Asia, south Asia, and southeast Asia. Over all land surfaces, MAM and SON surface warming are both 0.09±0.02 K decade⁻¹. Maximum MAM (SON) regional warming occurs in central and north Asia (Arctic) at 0.18±0.05 (0.17±0.09) K decade⁻¹. Relatively large MAM warming also occurs in east Asia (0.13±0.07 K decade⁻¹) and south Asia (0.11±0.06 K decade⁻¹); relatively large SON warming occurs in Canada (0.15±0.06 K decade⁻¹) and Europe (0.13±0.05 K decade⁻¹). Precipitation responses are again less robust, although east Asia experiences robust increases in both seasons (0.03±0.02 in MAM and 0.06±0.03 mm d⁻¹ decade⁻¹ in SON). Relatively large SON precipitation increases also occur for Central America and south Asia.