Measurement report: Long-term changes in black carbon and aerosol optical properties from 2012 to 2020 in Beijing, China

Atmospheric aerosols play an important role in radiation balance of the earth-atmosphere system. However, our knowledge of the long-term changes in black carbon (BC) and aerosol optical properties in China are very limited. Here 20 we analyze the nine-year measurements of BC and aerosol optical properties from 2012 to 2020 in Beijing, China. Our results showed large reductions in eBC by 67 % from 5.54 ± 5.25 μg m-3 in 2012 to 1.80 ± 1.54 μg m-3 in 2020, and 47 % decreases in light extinction coefficient (bext, λ = 630 nm) of fine particles due to clean air action plan since 2013. The seasonal and diurnal variations of eBC illustrated the most significant reductions in the fall and night time, respectively. ΔeBC/ΔCO also showed an annual decrease from ~ 6 to 4 ng m-3 ppbv-1 and presented strong seasonal variations with high 25 values in spring and fall, indicating that primary emissions in Beijing have changed significantly. As a response to clean air action, single scattering albedo (SSA) showed a considerable increase from 0.79 ± 0.11 to 0.88 ± 0.06, and mass extinction efficiency (MEE) increased from 3.2 to 3.8 m2 g-1. These results highlight an increasing importance of scattering aerosols in radiative forcing, and a future challenge in visibility improvement due to enhanced MEE. Brown carbon (BrC) showed similar changes and seasonal variations to eBC during 2018 – 2020. However, we found a large increase of 30 secondary BrC in the total BrC in most seasons, particularly in summer with the contribution up to 50 %, demonstrating an enhanced role of secondary formation in BrC in recent years. The long-term changes in eBC and BrC have also affected https://doi.org/10.5194/acp-2021-637 Preprint. Discussion started: 18 August 2021 c © Author(s) 2021. CC BY 4.0 License.

Beijing as one of the largest megacities in the world has been a great success in decreasing PM 2.5 during the last decade by implementing clean air action plan . Many previous studies focused on the changes in aerosol chemical components and the influences of emissions and meteorological conditions (Lei et al., 2020;Sun et al., 2020b;Sun et al., 2018). The mass concentration, mixing state, optical property and coating chemical composition of BC in Beijing were 70 also widely investigated in field campaigns in specific seasons (Din et al., 2019;Sun et al., 2021;Xie et al., 2020;Han et al., 2017). However, our understanding of the long-term changes of black carbon, aerosol optical properties and radiative effects as a consequence of clean air action are very limited.
In this study, we conducted nine-year measurements of eBC and light extinction coefficient (λ = 630 nm) by using Aethalometers along with cavity attenuated phase shift (CAPS) extinction monitor in Beijing. The long-term changes in 75 eBC, b ext , SSA and mass extinction efficient (MEE) are investigated, and their annual, seasonal and diurnal variations are elucidated. Moreover, we illustrate the changes in BrC absorption and absorption Ångström exponent (AAE) by using three-year measurements from 2018 to 2020. Particularly, the contributions of primary emissions and secondary formation to BrC absorptions are quantified and their changes during the past three years are demonstrated. Finally, the impact of the changes in BC and BrC on direct radiative forcing is estimated and discussed. 80 where b ext is the light extinction coefficient at 630 nm. The mass concentration of eBC was converted to b abs at 630 nm using MAE of 7.9 in spring and summer and 7.4 in fall and winter, respectively (Han et al., 2017).
The mass extinction efficiency (MEE) of PM 2.5 was derived as the ratio of b ext to the mass concentration of PM 2.5 , 105 = . (3) The absorption Ångström exponent (AAE) can be determined using Eq. (4) (Moosmüller et al., 2011), and the b abs,BC at each wavelength was estimated assuming an AAE = 1 for pure BC (Bond and Bergstrom, 2006). After subtracting b abs, BC from the total absorption coefficient b abs, total , the BrC absorption coefficient (b abs, BrC ) can be estimated with Eq. (5). In this study, we calculated BrC absorption at 370 nm that referred to b abs, BrC . Note that we may slightly overestimate the 110 absorption of BrC due to the influence of dust though the MAE of dust was much lower than BC and BrC (Yang et al., 2008),.

Quantification of primary and secondary BrC absorption 115
The absorption of BrC at 370 nm can be segregated into primary (b abs, Primary BrC ) and secondary (b abs, Secondary BrC ) using Eq.
where b abs, BC was the absorption at 880 nm, (b abs, BrC /b abs, BC ) pri is the ratio of primary BrC absorption to BC absorption.
Considering b abs, BC = [eBC]*MAE, we simplify the Eq. (6) to Eq. (7). 120 In this study, (b abs, BrC /eBC) pri was determined by the newly developed MRS method (Wu and Yu, 2016) using mass concentration of BC as a tracer (Wang et al., 2019a). In MRS calculation, the correlation (R 2 ) between measured eBC and estimated b abs, Secondary BrC was examined as a function of a series of hypothetical (b abs, BrC /eBC) pri . The (b abs, BrC /eBC) pri with the minimum correlation coefficients (R 2 ) between BC and b abs, Secondary BrC was assumed as the most statistically probable 125 (b abs, BrC /eBC) pri considering the independent variations between BC and b abs, Secondary BrC .

Estimation of radiative forcing of BC and BrC
We estimated the direct radiative forcing (ΔF R ) caused by BC and BrC at the top-of-atmosphere (TOA) based on forcing equations suggested by pervious study (Chylek and Wong, 1995), the modified wavelength-dependent version of the equation is given as below (Chen and Bond, 2010): 130 where S is the solar irradiance (W m -2 ), is the atmospheric transmission (unitless), is the fractional cloud amount (unitless), is the surface reflectance (unitless), and is the backscatter fraction (unitless), and and are the aerosol scattering and absorption optical depths (unitless), respectively. Wavelength-dependent S(λ) and ( ) are derived from the ASTM G173-03 reference spectra (Chen and Bond, 2010). and are 0.6 and 0.19, respectively based 135 on previous studies (Bond and Bergstrom, 2006;Wang et al., 2019b). is 0.29 (Charlson et al., 1992). Based on method in previous study (Wang et al., 2019b), τ scat and τ abs can be estimated as τ scat (λ) = b sca (λ) × H eff and τ abs (λ) = b abs (λ) × H eff , respectively, where b sca (λ) and b abs (λ) are scattering and absorption coefficients, respectively, and H eff is effective height.
The effective heights can be derived from the relationship between aerosol optical depth τ (=τ scat + τ abs , available from the Aerosol Robotic Network data archive) and light extinction coefficient by CAPS. The detail results of H eff in four seasons 140 are shown in Table S2.
The uncertainties of BC and BrC absorption ΔF R for (including primary and secondary ones) were quantitatively determined through the use of Monte Carlo simulations. Note that the uncertainty was represented as one standard deviation (±1σ) or the coefficient of variation (CV, σ divided by the mean) expressed as a percentage. According to uncertainty propagation, the CV for b abs, BC (λ) as follows, CV babs,BC(λ) ≈{ [(CV babs,BC, 880 ) 2 +[CV α *α*ln(880/λ)] 2 } 1/2 , where CV babs,BC, 880 145 and CV α represent the uncertainty of measured absorption coefficient at 880nm (~ 25 %) and absorption Ångström exponent of pure BC (~ 10 %) (Lu et al., 2015;Bond et al., 2013;Lack and Langridge, 2013;Gyawali et al., 2009), respectively. The CV for b abs, BrC (λ) could be quantified as follows, CV babs, BrC(λ) ≈ { [(CV babs,BrC, 370 ) 2 +[CV β *β*ln(370/λ)] 2 } 1/2 where CV babs,BrC, 370 and CV β represent the uncertainties of BrC absorption coefficient at 370 nm (CV babs,BrC,370 ≈ { [(CV babs,total,370 ) 2 +[ CV α *α*ln(880/370)] 2 } 1/2 ≈ 26 %) and absorption Ångström exponent of BrC (fitting uncertainty ~ 10 150 %), respectively. Similarly, CV babs, PriBrC, (λ) and CV babs, SecBrC(λ) also could be quantified. The CVs were as the parameters for the Monte Carlo analysis, and 100000 simulations were conducted to evaluate uncertainties.  (Table   S1). The annual mean concentration in 2020 was similar to that in Milan (Mousavi et al., 2019), lower than that in Xiamen (Deng et al., 2020), Shanghai (Wei et al., 2020), and Hefei (Zhang et al., 2015), yet higher than that in Nanjing (Jing et al., 2019) and New York City (Rattigan et al., 2013). A similar reduction in CO by 68 % from 2012 to 2020 (Fig. S2) also 160 indicated that the primary emissions from incomplete combustion reduced significantly in the past decade. Considering that different primary sources showed different emissions of BC and CO (Spackman et al., 2008;Derwent et al., 2001), we calculated △eBC/△CO as the ratio of (eBC-eBC 0 ) and (CO-CO 0 ) which was widely used to identify the variations of BC sources (Kondo et al., 2006;Subramanian et al., 2010). CO 0 was determined as values of the 1.25 percentile of each year and eBC 0 concentration was assumed as zero (Pan et al., 2011;Han et al., 2009). As shown in Fig. 1a, the annual mean 165 values of △eBC/△CO and eBC/PM 2.5 presented similarly decreasing trends, indicating a significant change in the structure of primary emission sources. Fig. 2 presents the monthly variations of eBC, △eBC/△CO and eBC/PM 2.5 . The eBC showed consistently seasonal patterns across different years with wintertime eBC almost twice that in summer mainly due to largely enhanced coal combustion emissions in heating season (Sun et al., 2018), consistent with the higher eBC/PM 2.5 in wintertime. △eBC/△CO presented pronounced seasonal variations. The highest values up to 12.0 ng m -3 170 ppbv -1 occurred in spring and fall likely due to the influences of biomass burning emissions (Pan et al., 2011;Han et al., 2009;Streets et al., 2003;Westerdahl et al., 2009;Spackman et al., 2008). However, the monthly average △eBC/△CO became relatively constant after 2018, suggesting that the primary emission sources of eBC and CO were relatively stable after the five-year clean air action (2013 -2017) in Beijing (Spackman et al., 2008).  (Table S1) also supported the long-term decreases in eBC and eBC/PM 2.5 . The eBC in spring decreased by 53 % from 2013 to 2020 with the similarly significant decrement in △eBC/△CO. The average 180 △eBC/△CO decreased from over 10 to below 5 in spring and fall, which suggested that the decreases of BC were mainly due to the reduced biomass burning emissions in the past eight years (Pan et al., 2011;Han et al., 2009;Streets et al., 2003;Westerdahl et al., 2009;Spackman et al., 2008). Different from the fall when BC emissions reduced more than other scattering pollutants, the values of eBC/PM 2.5 were relatively stable in spring. In comparison, the eBC and△eBC/△CO in summer were lower than those in other seasons except 2012, and they did not change substantially in recent years. This is 185 likely due to relatively stable sources in summer, i.e., vehicle exhausts. Although the wintertime eBC decreased by more than 60 % from 2012 to 2019, the different source contributions to BC were relatively constant in winter over eight years as indicated by flat △eBC/△CO (3 ~ 6 ng m -3 ppbv -1 ). One explanation is that coal combustion and biomass burning emissions in winter were reduced similarly in Beijing due to the promotion of clean fuels. Overall, the results above the decreases in eBC during the last eight years in Beijing were mainly due to the changes in spring, fall and winter, and the 190 reasons for the changes were different between winter and the other two seasons. In addition, more attention should be paid to the BC reduction in winter in the future based on the analysis of the four seasons.

Temporal variations of BC
Before 2015, the diurnal variation of eBC ( Fig. 4) showed morning peaks which were mainly due to the traffic rush hours in four seasons. After the "China 5" standard applying nationally and eliminating 5 million old vehicles in China , the morning peaks of eBC disappeared. Instead, they eBC presented similar and pronounced diurnal variations during four seasons characterized by the lowest mass concentrations in the afternoon due to high mixing layer height (MH) and low emissions, consistent with previous studies in Beijing . Due to deeper developments of boundary layer in spring and summer, the lowest values occurred during 15:00 ~ 17:00 which were later than those in fall and winter.
Comparatively, ubiquitously higher concentrations of eBC in early morning resulted from a synergetic effect of shallow boundary layer height and high emissions from heavy-duty vehicles and diesel trucks that are allowed to enter the city only 200 between 23:00 and 6:00. Consistently, the diurnal cycles of △eBC/△CO presented the highest values at 2:00 ~ 6:00 before 2019 due to the differences in vehicle emissions throughout the day. For example, previous studies found that CO is emitted primarily from gasoline vehicles while BC is dominated from diesel trucks and heavy-duty vehicles (Kondo et al., 2006).
We also observed the decreases in △eBC/△CO and eBC from 2013 to 2019 were more significant at night, highlighting that the reductions of diesel truck and heavy-duty vehicle emissions at night contributed significantly to the decreases of 205 BC in Beijing. Differently, the diurnal cycles of △eBC/△CO in winter were less pronounced than other seasons (Fig. 4h), indicating that the sources of BC were relatively stable during heating period although the mass concentrations decreased.
Particularly, we found that the diurnal variations of both △eBC/△CO and eBC in 2020 were less pronounced during four seasons. One reason was likely due to the fact that primary emissions e.g., vehicle emissions, were significantly reduced due to the influences of COVID-19. According to the diurnal variations, we found even though diesel vehicle emission 210 reduction contributed much to the whole BC decrease, its emission should also be controlled especially in summer and fall. regional transport. Different from the summer, high concentrations of eBC occurred dominantly in a small region close to the sampling site during other three seasons, suggesting the dominant source contributions from local emissions. However, the regional transport from the south and southeast was also found to play an important role, e.g., spring 2015 and 2020, and winter 2017. It's interesting to note that the eBC from the southwest with low WS decreased significantly over eight years in fall while it still exceeded 3 μg m -3 from the southeast in the fall of 2019. These results indicate that the source 220 regions of eBC can be substantially different in different years depending on meteorology. By comparing with the seasonal variation of △eBC/△CO, we inferred that biomass burning emissions from the southwest and regional transport from the southeast are two important non-local sources of eBC in Beijing. In conclusion, in the process of implementing air pollutant reduction actions, the synergistic control of the surrounding areas of Beijing should not be neglected as well. to 2020, while that of SSA was increased from 0.79 ± 0.11 to 0.88 ± 0.06. Such an increase in SSA could likely shift the radiative forcing from positive to negative (Hansen et al., 1997;Lee et al., 2007). Besides SSA, MEE is also a key factor reflecting the responses of atmospheric light properties to aerosol composition changes. In particular, the annual mean played more important roles than absorbing aerosol species in radiative forcing. This change is consistent with the findings of previous studies showing increased contributions of high scattering ammonium nitrate in fine particles (Y. Huang et al., 2013;Lei et al., 2020). The seasonal variations of SSA and MEE showed generally higher values in winter and lower values in summer (Fig. S4). Such seasonal trends are overall similar to those of eBC/PM 2.5 and eBC (Fig. 2), indicating that non-BC aerosol species in winter appeared to have higher scattering efficient than those in summer. 235 Fig. 6 shows that the seasonal average SSA presented similar increasing trends during all seasons indicating a more effective controls of absorbing aerosol (i.e., eBC) than scattering components during the last decade. This is consistent with recent studies showing larger reductions in primary aerosol species than secondary species as responses to emission controls (Sun et al., 2020b). The most significant increase of SSA was observed in fall from 0.75 ± 0.12 to 0.87 ± 0.07 during 2012-2020, followed by summer from 0.77 ± 0.12 in 2012 to 0.88 ± 0.06 in 2020. The highest seasonal average 240 SSA (0.88 ± 0.06) was observed in summer 2020, which is close to that during the COVID-19 outbreak in spring 2020, yet was much higher than 0.82 ± 0.05 observed in North China Plain in 2009 (Ma et al., 2011). We also noticed that the increase in SSA was becoming smaller over past eight years indicating that the relative contributions of light absorbing and scattering components became relatively stable as the progress of clean air action. Fig. 7 also shows the seasonal average of b ext and MEE over past nine years. The MEE increased mostly by more than 43 % in summer from 2012 to 2020 although 245 b ext decreased ubiquitously during all seasons, most notably in the fall from 432 Mm -1 in 2014 to less than 140 Mm -1 in 2020 (~ 68 %). Comparatively, b ext was relatively stable at 230 Mm -1 in spring before 2019 and decreased substantially by 40 % in 2019 due to significant reductions in fine particles. Although b ext was comparable in springs of 2019 and 2020, SSA was increased by 8 %. These results suggest the aerosol composition has also played an important role in changing aerosol optical properties. For example, higher SSA in spring 2020 resulted from larger reductions in primary emissions 250 e.g., absorbing eBC, than scattering secondary aerosol due to the decreases in anthropogenic emissions during the COVID-19 lockdown. The increase of MEE from 2.6 to 3.6 m 2 g -1 also suggested a significant change in scattering aerosol composition from 2019 to 2020, consistent with the results in a previous study (Lei et al., 2020). Compared with spring, b ext decreased by 60 % in summer from 2012 to 2015 and then gradually increased afterwards. Similarly, b ext also showed a sharp decrease of 60 % in winter from 2014 to 2017 and after that it continuously increased to > 200 Mm ¬1 in 2019. 255

Temporal variations of aerosol optical properties 225
Considering the increased SSA yet relatively constant mass concentrations of eBC, we inferred that the increased light extinction in winter was mainly caused by scattering aerosols that can vary substantially in different years due to the changes in meteorological conditions . Overall, the results in this study clearly demonstrate the responses of aerosol optical properties to the changes in aerosol composition since clean air action in 2013. In summary, seasonal variations of SSA and MEE indicates that scattering aerosols will become a new challenge affecting atmospheric visibility 260 in all seasons.
The diurnal cycles of b ext and SSA in four seasons are shown in Fig. 7. The diurnal variations of SSA were similar and pronounced during all seasons that were characterized by the peaks at afternoon, consistent with previous studies in Beijing Han et al., 2017). Before 2015, SSA presented an obvious valley during 7:00 ~ 9:00 mainly due to the highest values during 12:00 ~ 13:00. The major reason is the reduced eBC emissions during daytime and enhanced emission and photochemical production of secondary scattering aerosols (Han et al., 2017). b ext also presented similar diurnal variations over nine years which were characterized by higher values at night and lower values during daytime. One of the major factors driving the diurnal variations is the evolution of boundary layer height 270 (Han et al., 2017;Xie et al., 2019;Han et al., 2015). As a response, b ext reached the minimum at 12:00 ~ 14:00 in fall and winter whereas it occurred during 16:00 ~ 18:00 in spring and summer due to a deeper mixing convection in late afternoon.
In fall, the value of b ext decreased while SSA increased at nighttime from 2012 to 2019, indicating that the reduction of BC at night had a significant impact on the decrease of b ext . Note that b ext increased by more than 62 % in winter from 2018 to 2020 as discussed above, yet the reasons causing the increased b ext were different according to the diurnal variations.  BrC was the highest during winter which was approximately twice that in spring and fall, and five times higher than that in summer. Consistent with the seasonal variations of eBC, the absorption of BrC in 2018 was generally higher than other 285 years mainly due to the increased biomass burning emissions. The lowest AAE ubiquitously occurred in summer while the highest value up to 1.5 occurred in winter, consistent with previous studies (Xie et al., 2020;Xie et al., 2019). Despite the stronger absorption in fall than spring in 2018 and 2019, the AAE was similar indicating the similar emission sources of BrC in the two seasons (Ran et al., 2016). Note that AAE was up to 1.39 in 2020 and showed a higher frequency at AAE > 1.3 in spring (Fig. 9), suggesting that the emissions with high combustion efficiency (e.g., traffic) decreased much more 290 than the low efficiency sources (e.g., biomass burning) during the COVID-19 lockdown in Beijing. As illustrated in Fig.   9, the distribution of AAE in summer mainly concentrated in the range of 1.0 -1.3 due to the low source emissions BrC than other seasons, particularly primary coal combustion and biomass burning emissions. However, we found a change in AAE distribution in summer 2020, which was characterized by a higher frequency at AAE > 1.3 suggesting a stronger BrC absorption. Further analysis showed that such a change was mainly due to the enhanced contribution from secondary BrC. 295

Temporal variations of light absorption of BrC
Compared with summer, the AAE distribution was relatively stable in fall and winter, and the distribution range of ~ 1.2 -1.9 in winter was overall higher than that in other seasons As shown in Fig. S5, the diurnal variations of BrC absorption were similar to eBC with generally higher values at nighttime except in summer during 2018 -2020. This result indicated that the primary emissions related BC were also the main sources of BrC in spring, fall and winter. In comparison, the diurnal variations of BrC absorption were largely different 300 from eBC in summer, possibly due to the fact that BrC was significantly influenced by secondary organic aerosols. The diurnal variation of BrC absorption in summer 2020 was different from previous years while the variations of eBC did not change significantly, supporting the increased contribution of secondary aerosol to BrC. Generally, the AAE showed a minimum at night followed by a daytime increase from 8:00 to 12:00 during four seasons (Fig. S5), suggesting that photochemical production contributed dominantly to the BrC formation during daytime. 305 By using the minimum R squared method (MRS) (Wu and Yu, 2016), we estimated the primary and secondary BrC absorptions in each month during 2017-2020. As shown in Fig. S6, the monthly variations of BrC absorption were similar and pronounced which were characterized by high values in January and low values in July. Despite this, the primary BrC absorption decreased gradually from 2017 to 2020, mainly due to the decreased emissions of biomass burning and coal combustion. We further explored the seasonal variations of primary and secondary BrC. As shown in Fig. 8b, BC dominated 310 ultraviolet light absorption at 370 nm during four seasons with the highest contribution being in summer (~ 85 %) and the lowest in winter (~ 60 %). One reason is because BrC from biomass burning and coal combustion in summer was small, consistent with lower AAE in summer than other seasons (Fig. 9). Note that the average contribution of BrC to the total absorption in summer increased to 16 % from 2018 to 2020 due to enhanced photochemical production associated with stronger atmospheric oxidation capacity. Comparatively, the contributions of primary BrC to the total absorption decreased 315 from 75 % to 50 %. The contributions of BrC absorption were comparable in spring and fall, accounting for 25-30 %. Due to the decreased primary emissions and enhanced secondary production during the COVID-19, we found that the contribution of BrC absorption was increased by more than 7 % in spring from 2018 to 2020. In comparison, the contributions of BrC were larger than 40 % in winter with slightly downward trends in past three years. The declines of primary emissions might be an explanation, mainly due to the replacement of coal to natural gas for residential heating in 320 recent years.
Although the contributions of BrC absorption to the total absorption were relatively stable during four seasons from 2018 to 2020, the relative contributions of primary and secondary BrC changed significantly (Fig. 8c). Overall, the primary BrC was much higher than secondary BrC, yet showing decreasing trends from 2017 to 2020 except in spring. While the primary BrC contributed more than 75 % to the total BrC in fall and winter, they reached the minimum in summer (50-75 %). The 325 contribution of summertime primary BrC decreased by more than 25 % from 2018 to 2020, and that of secondary BrC increased up to 50 % to the total BrC in summer 2020 with an increase in AAE to 1.2. Given that eBC was continuously decreased in summer in past three years, the increases in AAE were mainly due to larger secondary BrC production from photochemical reaction. We also observed a large increase in secondary BrC in winter from 2018 to 2020. While the secondary BrC was negligible in 2018, the contribution increased to ~ 25 % in 2020, suggesting that secondary production 330 of BrC became more important in winter, which is consistent with the continuous increase SOA in winter in recent years (Lei et al., 2020). Similar increases in secondary BrC were also observed in fall.

Direct radiative forcing of BC and BrC
As shown in Fig. 10, the annual mean ΔF R caused by BC was about +3.00 W m −2 in 2012 (from August in 2012 to June in 2013), close to that previously reported in north China . However, ΔF R decreased substantially by 64 % 335 (+1.09) in 2020, suggesting that the BC radiative forcing was largely reduced during the last decade. Previous studies (Ding et al., 2016) had shown that the aerosol-boundary layer feedback to unit quantity of BC will be lower in higher aerosol loading case as solar radiation weakened. In our study, such decrease of BC radiative effect likely contributed much to quick improve the air quality. And the relatively lower ΔF R caused by BC in recent year could facilitate the dispersion of pollutants in the boundary layer, which in turn will maintain air pollution at a low level. The seasonal variation of BC ΔF R 340 ( Fig. S7) suggested the largest decrease in summer and fall. In addition, we noticed that the BC ΔF R was relatively stable in each season from 2019 and 2020, consistent with the small changes in eBC concentrations. We also estimated the radiative effects of BrC. As shown in Fig. 10, BrC ΔF R decreased by 43 % from +0.30 in 2018 to +0.17 W m −2 in 2020.
However, such a value was much higher than the global mean (+0.04 ~ 0.11 W m -2 ) (Feng et al., 2013). The scattering radiative forcing of BrC was estimated at -1.00 ~ -1.65 W m −2 . The absorbing radiative forcing of BrC led to ∼18 % 345 reduction in the amount of negative radiative forcing caused by BrC scattering compared to results from the non-absorbing assumption. The seasonal variation of BrC ΔF R (in Fig. S7) showed a large decrease during all seasons from 2018 to 2019.
However, compared with 2019, the BrC ΔF R became stable in summer 2020 which was different from the decreases in spring and fall. We also estimated the primary and secondary BrC ΔF R . Primary BrC ΔF R was approximately +0.16 W m −2 in 2020 decreasing by 41 % compared with 2018 (+0.27 W m −2 ). Such a value was higher than the global average of 350 radiative forcing (+0.11 W m −2 ) from POA (Lu et al., 2015). Compared with primary BrC, the secondary BrC ΔF R was generally small yet showing an increase from +0.005 W m −2 in 2019 to +0.016 W m −2 in 2020. The probability distributions of ΔF R for BC and different types of BrC are shown in Fig. S8. The uncertainties of BC and BrC absorption ΔF R are comparably about 27 ~ 28 %. And the uncertainties for primary and secondary BrC absorption ΔF R are about 32 % and 43 %, respectively. 355

Conclusions
Nine-year measurements of eBC and light extinction coefficient in Beijing were analyzed in this study. Our results showed that the annual mean eBC concentration decreased by 67 % from 5.54 µg m -3 in 2012 to 1.80 µg m -3 in 2020, and the decreases dominantly occurred at nighttime suggesting an effective control of primary emissions due to clean air action since 2013. b ext showed similar reductions by 47 % from 2012 to 2020. We also observed a pronounced seasonal variation 360 in △eBC/△CO with high values in spring and fall, and a gradual decrease in recent years, indicating a significant change in primary sources. As a response of the changes in primary and secondary aerosols, SSA increased substantially from 0.79 ± 0.11 in 2012 to 0.88 ± 0.06 in 2020, and it presented similar increasing trends during all seasons. These results highlight increasingly important role of scattering aerosol in radiative forcing. Similarly, the seasonal average MEE increased gradually from 2012 to 2020, and the increase was most significant in summer by more than 43 %. The increased MEE 365 explained the fact that PM 2.5 decreased substantially after clean air action, while the visibility did not show similar improvements as PM 2.5 .
We further analyzed the changes in BrC during 2018 -2020. The BrC absorption presented the pronounced seasonal variation with the highest value in winter, after quantifying the primary and secondary BrC, we found that the primary emissions co-emitted with BC were the main sources of BrC during most seasons while the secondary BrC was also 370 important in summer. In particular, the contribution of secondary BrC to the total BrC showed a large increase in summer, and it was up to 50 % in summer 2020. These results indicated the BrC from secondary formation played an increasing role in the absorption at 370 nm during 2018 -2020 in Beijing. By estimating the direct radiative forcing caused by absorbing aerosols, we found that the annual mean BC ΔF R decreased by 64 % from +3.00 W m −2 in 2012 to +1.09 W m −2 in 2020, and that of BrC decreased from +0.30 to +0.17 W m −2 during 2018 -2020. Considering that the BC-induced 375 aerosol and boundary layer feedback plays an important role in severe haze formation, the decreases in BC and radiative forcing would weak the interaction between aerosol and boundary layer, and help mitigate air pollution.
Data availability. The data in this study are available from the authors upon request (sunyele@mail.iap.ac.cn).
Author contributions. YS and JS designed the research. JS, WZ, CX, CC, and TH conducted the measurements. JS, ZW, 380 WZ and CC analyzed the data. CW, QW, ZL, JL, PF and ZiW reviewed and commented on the paper. JS and YS wrote the paper.
Competing interests. The authors declare that they have no conflict of interest.