Aerosol reductions outweigh circulation changes for future improvements in Beijing haze

. Despite local emission reductions, severe haze events remain a serious issue in Beijing. Previous studies have sug-gested that both greenhouse gas increases and aerosol decreases are likely to increase the frequency of weather patterns conducive to haze events. However, the combined effect of atmospheric circulation changes and aerosol and precursor emission changes on Beijing haze remains unclear. We use the Shared Socioeconomic Pathways (SSPs) to explore the effects of aerosol and greenhouse gas emission changes on both haze weather and Beijing haze itself. We conﬁrm that the occurrence of haze 5 weather patterns is likely to increase in future under all SSPs, and show that even though aerosol reductions play a small role, greenhouse gas increases are the main driver, especially during the second half of the 21st century. However, the severity of the haze events decreases on decadal timescales by as much as 70% by 2100. The main inﬂuence on the haze itself is the reductions in local aerosol emissions, which outweigh the effects of changes in atmospheric circulation patterns. This demonstrates that aerosol reductions are beneﬁcial, despite their inﬂuence on the circulation.


Introduction
Over a million premature deaths in China were attributed to poor air quality in 2010, accounting for over 30% of the mortality due to air pollution worldwide (Lelieveld et al., 2015;Zhang et al., 2017). Ambitious clean air policies, designed to address this serious issue, have resulted in dramatic reductions in Chinese emissions of the particulate matter and gases that contribute to poor air quality since 2008 (Zheng et al., 2018;Li et al., 2017). Sulphur dioxide (SO 2 ) and black carbon (BC) emissions occurrence and persistence of haze events. In fact, atmospheric circulation patterns conducive to haze events may increase in the future as the climate warms (Cai et al., 2017;Pei and Yan, 2018). Continued haze events could also be due to a lack of mitigation of secondary aerosols (Huang et al., 2014;Zheng et al., 2018;An et al., 2019;Le et al., 2020). In 2020, severe haze was formed during the COVID-19 lockdown despite emission reductions of up to 90% (Le et al., 2020).
A metric that has been recently introduced to characterise the meteorological conditions associated with haze over north 25 China, and Beijing in particular, is the haze weather index (HWI) (Cai et al., 2017). The HWI accounts for the role of circulation and vertical stratification, the two main meteorological factors involved in haze occurrence. By analysing 15 climate models under the Representative Concentration Pathway 8.5 (RCP8.5 , frequently referred to as the "business as usual" scenario and featuring a large future increase in greenhouse gas emissions), Cai et al. (2017) predicted a 50% increase in the frequency and a 80% increase in the persistence of these meteorological conditions in the second half of the 21st century compared to the 30 20th century as a consequence of warming and circulation changes induced by greenhouse gases.
Aerosol reductions may also modulate meteorological conditions and, subsequently, lead to increases in HWI (Jiang et al., 2017;Liu et al., 2019;Zhang et al., 2020a). Affected by aerosols, and also affecting their accumulation in northeast China, the HWI is intrinsically intertwined with aerosol emissions and haze formation. Yet, meteorological conditions represent a key uncertainty in future climate projections and a large source of discrepancy among models (Callahan and Mankin, 2020).

35
New future scenarios used in the Sixth Coupled Model Intercomparison Project (CMIP6), the Shared Socioeconomic Pathways (SSPs), are designed to cover a wide range of narratives of socioeconomic development and energy consumption in the 21st century, and capture a much wider range of uncertainty in aerosol emission pathways (Figure 1) than the Representative Concentration Pathways used in CMIP5 (Riahi et al., 2017). This presents an opportunity to explore the relative roles of aerosol and greenhouse gases in driving changes in the HWI, and the interplay between the effects of aerosol changes on the 40 atmospheric circulation and on the composition of the haze itself. In this study we examine whether, in scenarios with rapid aerosol reductions, increases in haze frequency or severity may not be as large as trends in circulation-based metrics alone suggest.

Future emission pathways 45
Future aerosol and greenhouse gas pathways in CMIP6 are described by the Shared Socioeconomic Pathways (SSPs), which represent a range of socioeconomic narratives (Riahi et al., 2017), designated by the first digit of the three numbers in the scenario names: 1 describes a sustainable development; 2 is a medium change narrative; 3 describes a regional rivalry situation; and 5 describes a fossil-fuel development. The following two digits separated by a decimal point show the global mean radiative forcing by 2100 in W/m 2 . The air pollution pathways have been designed to be consistent with the SSPs, but also make specific 50 assumptions about the stringency of air quality policy. We use four SSPs in this work: 1-2.6, 2-4.5, 3-7.0 and 5-8.5. Figure 1 shows the time series of area-averaged SO 2 , BC and CO 2 emissions from East Asia, South Asia, and the globe in four SSPs, during 2015-2100. Over East Asia, both SO 2 and black carbon (BC) emissions will decrease from present-day to the end of the century in most SSPs, except in SSP 3-7.0, wherein emissions continue to increase until 2040. The change in aerosol emissions over East Asia shows a rapid reduction in SSP 1-2.6, and comparable moderate reductions in SSPs 2-4.5 and 5-8.5.

55
In SSP 1-2.6, aerosol emissions reduce rapidly until 2050, and then flatten until 2100. In SSPs 2-4.5 and 5-8.5, emissions reduce gradually and reach the minima by 2100. Radiative forcing at the end of the 21st century varies consistently with the amount of global CO 2 emissions.

Air quality and circulation indices
The Haze Weather Index (HWI) is defined following Cai et al. (2017), using a combination of meteorological variables that 60 capture the key characteristics of weather patterns conducive to severe haze in and around Beijing.
where ∆T is the temperature difference between 850 hPa (averaged over 32. The HWI is then calculated as the standardised anomaly of the sum of the three terms. HW I > 0 designates a meteorological 70 pattern conducive to a haze event. The strength of the East Asian Winter Monsoon is measured by the WCI index (Wang and Chen, 2014), which considers the sea level pressure (SLP) differences between Siberia and the northern Pacific, as well as between Siberia and the Maritime Continent:

75
where SLP xx is the standardised SLP anomaly over Siberia (sib; 40 The first term represents a zonal pressure difference, while the second represents a meridional difference. WCI is the standardised anomaly of the sum of these differences.
W CI > 0 indicates a strong winter monsoon, which makes Beijing haze events less likely.
Both indices quantify the strength of the East Asian Winter Monsoon, although HWI is designed to capture specific patterns 80 over Beijing. By using two indices based on different variables, any results will be considered to be more robust if they are consistent across the indices, as the different variables involved should be differently affected by model biases. HWI is negatively related to WCI, as a positive HWI indicates a weak winter monsoon (i.e., a negative WCI). To make the comparison straightforward, the sign of WCI will be reversed and denoted as WCI* throughout, so that positive values of both indices indicate an increased likelihood of haze.

Data
Monthly pressure level variables are used to calculate seasonal-mean HWI and WCI* during December-January-February (DJF). In its original definition (Cai et al., 2017), HWI was computed from daily data. Yet, despite a slightly reduced magnitude, values of the HWI using monthly data are consistent with those based on daily data (Zhang et al., 2020a). Since the use of monthly data allows us to use a greater number of CMIP6 models, monthly HWI is computed and analysed here. Any model 90 that has data available in both the CMIP6 historical experiment and one of the SSPs is included. As a result, 17 models are used (

100
where P M 2.5 is the sum of the dry aerosol mass mixing ratio of BC, total organic carbon (OA -both primary and secondary sources), sulphate (SO 4 ), sea salt (SS) and dust (DU ) from the lowest model level. A scaling factor of 0.25 for SS and 0.1 for DU has been used to calculate the approximate contribution from these components to the fine size fraction (< 2.5µm), and is applied to data from all models. Unfortunately, the availability of aerosol variables from CMIP6 models is limited compared to the atmospheric variables. Models and experiments that include AOD and/or PM 2.5 are highlighted in Table 1.

105
The fifth generation European Center for Medium-Range Weather Forecasts atmospheric reanalysis (ERA5, Hersbach et al., 2020) is used to evaluate the present-day distribution of HWI and WCI*, and the constituent variables, in CMIP6 models.
Compared to previous reanalysis versions, ERA5 has improvements in model physics, core dynamics and data assimilation.

Present-day East Asian Winter climate in CMIP6
The CMIP6 multi-model mean captures the pattern and magnitude of the key features of the East Asian Winter climate rea- Kolmogorov-Smirnov test is applied, which indicates that indices from ERA5 and CMIP6 are drawn from the same distribution. A closer inspection of Figure 3 shows that the CMIP6 models tend to simulate a stronger winter circulation than in

125
In all future periods, HWI is larger than in the present day (Figure 4a), indicating that weather patterns conducive to haze will occur more frequently, be more severe, or both. WCI* is also greater, indicating a weaker winter monsoon circulation, and more favourable conditions for haze.
The positive HWI anomaly is larger in SSPs 3-7.0 and 5-8.5 relative to other SSPs by 2045-2054 and reaches its maximum

135
In the first half of the 21st century, the competition between the response to aerosol and CO 2 changes are visible. During 2025-2034, WCI* has the smallest increase in SSP 3-7.0, where aerosol emissions continue to increase. In contrast, WCI* has the largest increase in SSP 1-2.6, where aerosol emissions decrease sharply during the same period. This is consistent with aerosol reductions driving increases in haze indices identified by Zhang et al. (2020a). Although changes in CO 2 emissions are different between SSPs 2-4.5 and 5-8.5, reductions in aerosol emissions are similar in this period (2025)(2026)(2027)(2028)(2029)(2030)(2031)(2032)(2033)(2034). This similarity is 140 reflected in the median WCI* anomalies, further suggesting that the aerosol emission pathway influences the relative magnitude of WCI* anomalies during the early 21st century. However, compared to responses to CO 2 in the second half of the 21st century, these differences in HWI and WCI* across the SSPs are small and insignificant ( Figure A3). HWI changes are similar to but weaker than WCI*, which can be attributed to the larger internal variability in HWI due to the smaller spatial scales of the component terms.

145
Both HWI and WCI* show that patterns conducive to haze events become more likely with increases in CO 2 emissions in the long term. However, without sources of aerosol, the formation of haze is unlikely. The actual change in haze itself depends on changes in both the atmospheric circulation and aerosol concentrations. AOD at 550nm and PM 2.5 are used as indicators of haze severity. Anomalies relative to the present day for HW I > 1 (the threshold for 'haze days' in the present climate) are shown in Figure 5. Changes in future AOD and PM 2.5 over Beijing correspond to those in aerosol precursor emissions over East Asia (Figure 1). The mean AOD and PM 2.5 rapidly reduce to values below the present-day level (by more than 25%) in all scenarios apart from SSP 3-7.0, where the mean AOD and PM 2.5 are greater than the present-day level until 2050 (up to 35%), consistent with continued increases in local emissions. As indicated by changes in AOD and PM 2.5 , haze events in all scenarios except SSP 3-7.0 become less severe ( Figure 5 a and c), despite the concomitant increases in HWI and WCI*. Future changes in AOD and PM 2.5 follow the reductions in future aerosol emissions, rather than the increase seen in the circulation metrics, 155 indicating that the reduction in aerosol emissions outweighs the increase in haze weather patterns and dominates changes of future haze events.
HWI, however, still can be useful in predicting future haze events. Figure 5 b and d show the percentage changes in AOD and PM 2.5 between haze (HW I > 1) and contemporary non-haze days (HW I < 0). In all periods, AOD and PM 2.5 have similar relative anomalies. Except SSP 1-2.6, the anomalies range between 20 ∼ 40%. This indicates that by using a certain value of 160 HWI to define haze events (for example, HW I = 1), air pollutant increases by a certain amount regardless of the value of the baseline. However, as future aerosol emissions will reduce, a larger HWI threshold is likely to be needed to identify circulation patterns likely to cause PM 2.5 concentrations to exceed dangerous levels.

Conclusion and Discussion
This study investigated 21st century changes in Beijing haze events using CMIP6 models. Circulation patterns conducive to the 165 formation of haze increase in all future scenarios due to the weakening of East Asian winter monsoon, with a clear relationship with increases in CO 2 at the end of the century. Scenarios with the largest CO 2 emission have significantly larger increases in two haze weather indices, HWI and WCI*, by 2100. The opposing impacts of aerosols on these patterns can be seen in the near future (2025)(2026)(2027)(2028)(2029)(2030)(2031)(2032)(2033)(2034) in SSP 3-7.0 where aerosol emissions continue to increase, moderating increases in the haze indices.
However, although near-future changes in HWI and WCI* are consistent with differences in aerosol emission across the SSPs, 170 the differences in aerosol pathways are not large enough to result in significant differences between the SSPs.
Future changes in the severity of haze events themselves were evaluated using anomalies in AOD at 550nm and surface PM 2.5 . Despite increases in HWI and WCI*, the severity of Beijing haze decreases with reductions in aerosol and precursor emissions. This shows that the decrease in aerosol emissions under strong air pollution mitigation scenarios outweighs the continued increase in haze weather patterns and weakening of the winter monsoon. The above findings indicate that using 175 meteorological indices alone to investigate future changes in haze can be misleading, and should be complemented by the analysis of changes in air quality metrics.
We show that reducing aerosol emissions is beneficial for Beijing air quality in the long term, despite their reductions making the atmospheric circulation patterns associated with haze more likely. The severity of haze events reduces most in SSP 1-2.6, which has the fastest and largest emission reduction, while it reduces the least in SSP 3-7.0, which has the slowest and smallest abb7431, 2020.