Aerosol radiative effects and feedbacks on boundary layer meteorology and 1 PM 2 . 5 chemical components during winter haze events over the 2 Beijing-Tianjin-Hebei region

16 An online-coupled regional chemistry/aerosol-climate model (RIEMS-Chem) was developed 17 and utilized to investigate the mechanisms of haze formation and evolution and aerosol 18 radiative feedback during winter haze episodes in February-March 2014 over the 19 Beijing-Tianjin-Hebei (BTH) region in China. Model comparison against a variety of 20 observations demonstrated a good ability of RIEMS-Chem in reproducing meteorological 21 variables, PBL heights, PM2.5 and its chemical components, as well as aerosol optical 22 properties. The model performances were remarkably improved for both meteorology and 23 chemistry by taking aerosol radiative feedback into account. The domain average aerosol 24 radiative effects (AREs) were estimated to be -57 W m at the surface, 25 W m in the 25 atmosphere and -32 W m at the top of atmosphere (TOA), respectively, during a severe haze 26 episode (20–26 February), with the maximum hourly surface ARE reaching -384 W m in 27 southern Hebei province. The average feedback-induced changes in 2-m air temperature (T2), 28 10-m wind speed (WS10), 2-m relative humidity (RH2) and planetary boundary layer (PBL) 29 height over the BTH region during the haze episode were -1.8 C, -0.5 m s, 10.0% and -184 30

7 temperature and precipitation over east Asia (Fu et al., 2005). 179 The online-coupled model RIEMS-Chem has been developed in recent years by 180 incorporating major atmospheric chemistry/aerosol processes into the host model. Transport  formation is parameterized by a two-product model (Odum et al., 1997). 197 Current atmospheric chemistry models generally tend to underpredict sulfate 198 concentrations, especially in source regions during wintertime, such as north China,which 199 could be due to uncertainties in the treatment of chemical formation mechanism. Recent Aerosols and Clouds) (Hess et al., 1998). In this study, measurements in Beijing are used to 214 represent aerosol size distribution more realistically and to constrain the model. During the 215 study period, a scanning mobility particle sizer (SMPS; TSI, Inc., Shoreview, MN, USA) was 216 used to measure aerosol size distribution (Ma et al., 2017)  days, whereas about 70% of aerosols were externally mixed with BC in clean days, so an 225 internal mixing assumption was adopted for model simulation because this study focuses on 226 haze events. Recent measurements also exhibited that the geometric mean radius of dry 227 aerosol internal mixture during haze evolution from light-moderate to severe pollution stages 228 just increased slightly from 0.10 μm to 0.12 μm (Ma et al., 2017), so an average of 0.11 μm is 229 chosen for the geometric mean radius of internal mixture, with standard deviation of 1.65. 230 Aerosol optical parameters including extinction coefficient, single scattering albedo and 231 asymmetry factor were calculated by a Mie-theory based method developed by Ghan and 232 Zaveri (2007). In this method, the optical properties of different types of aerosols are  where Va is the total volume of dry aerosols, Vj is the volume of each aerosol component j. 253 The refractive index of internally mixed aerosols is calculated using the Maxwell-Garnett 254 mixing rule: where Rw is the refractive index of the internal mixture, Ri and Rs are the refractive index 258 of insoluble components (BC and POA) and soluble components (inorganic aerosols, SOA 259 and water), respectively. Vi represents the volume of insoluble components, V represents the 260 total volume of wetted aerosols.

261
After obtaining the wet diameter (Dp) and refractive index of the internally mixed 262 10 aerosols (Rw), the aerosol optical properties (Q) can be derived from formula (1)  effective radius re is calculated based on Nc, WL and the cube of the ratio of the mean volume 275 radius and the effective radius of the cloud-droplet spectrum following Martin et al. (1994).

276
The effect of aerosols on ice nuclei and convective cloud is not treated yet in this model 277 because of the complexity and limitation in knowledge.

278
The aerosol optical parameters and Nc due to aerosol activation calculated above are 279 transferred into radiation module to account for the perturbation of radiation and atmospheric 280 heating rate due to aerosol direct and indirect effects.         reasonably well in the FULL case, although the peaks were somewhat underpredicted in 478 some days, which could be partly due to the overprediction of wind speed (Table 1)  with R of 0.8 and NMB of -7% (Table 2), which demonstrates a good model performance for  Table 2). Another important finding is that the duration of haze episode was prolonged by 523 about 2-3 hours by the aerosol radiative feedback compared with that without aerosol 524 feedback ( Figure 2c).

525
To evaluate the overall model performance on PM2.5 and its gas precursors in the BTH 526 region, we also collected observations at 80 surface stations in 13 cities of the BTH from the   and 0.88, with Rs of 0.8, 0.7 and 0.7 and NMBs of 4%, 10% and 5%, respectively (Table 2).

557
The above comparison demonstrates a good ability of the model in estimating aerosol optical 558 properties during the study period, which could be attributed to both the good performance  Table 3 summaries the performance statistics for daily mean AOD. In general, the model   study period, the regional mean AOD in the BTH region was 0.78 (Table 4) 19 W m -2 and -18 W m -2 at the surface, in the atmosphere and at the TOA, respectively (Table   652 4). The indirect radiative effect was also estimated to be about -2 W m 2 at the surface and the 653 TOA on average, much smaller than the direct radiative effect; therefore, the total aerosol 654 radiative feedback is predominated by direct radiative effect during the study period.  The aerosol feedback during the first haze episode was further explored due to the much 701 higher PM2.5 level than the period average.  It is worthwhile to further explore the effect of aerosol feedback during haze evolution. 718 We divided haze episode into three stages, the growth stage is defined as the time period of The impact of aerosol radiative feedback in Beijing (Table 6) was stronger than the February due to the arrival of a cold front.

803
PA was used to provide insights into the evolution mechanism of the haze episode, which 804 was divided into the clean, growth, persistence and dissipation stages in this study. Figure 8   805 shows the average process budgets for changes in PM2.5 (which is the sum of sulfate, nitrate,  (Figure 8b). It was also noticed that HADV 881 contributed to PM2.5 production in this stage, which was due to mass import to Beijing from 882 upwind areas by northwesterlies.

883
It should be mentioned that the contribution of emission was unchanged because the 884 monthly based emission inventory from MEIC was used, and the contribution of cloud 885 process was generally negligible throughout the period because there was little cloud and 886 precipitation during the study period. 887 We further use PA to interpret evolution processes of primary (BC) and secondary 888 (sulfate and nitrate) aerosols.

31
Black carbon is considered to be inert and chemical inactive, so it is governed solely by 890 physical processes. In the clean stage, BC production was contributed solely by emission (5.7 891 g m -3 h -1 ), whereas vertical diffusion and dry deposition contributed equally to BC loss (-2.7 892 g m -3 h -1 ), and other processes were negligible (Figure 8c). In the growth stage, the  (Figure 8c).

905
As for secondary aerosols, like sulfate, contribution from direct emission was near zero.

906
In the clean stage, gas chemistry (5.9 g m -3 h -1 ) was the predominant process for sulfate 907 production, and vertical diffusion contributed most to the loss (-5.2 g m -3 h -1 ) (Figure 8d). In 908 the growth stage, contribution from vertical diffusion was reduced to -3.9 g m -3 h -1 mainly 909 due to the decreased vertical diffusivity (Figure 8f), whereas positive contribution from gas 910 chemistry increased to 6.6 g m -3 h -1 , which was resulted from competitive processes. For 911 sulfate formation from gas chemistry (SO2+OH→H2SO4, followed by nucleation or 912 condensation into particulate phase), the oxidation of SO2 to sulfate was weakened because of 913 decreasing OH radical due to increasing aerosol attenuation of solar radiation, however, SO2 914 increased due to weakened vertical diffusivity, leading to a slight net increase of sulfate 915 concentration compared with the clean stage. It is noteworthy that the sulfate production rate 916 from heterogeneous reactions increased to 2.7 g m -3 h -1 , mainly due to the increases in SO2, 917 aerosol surfaces and RH (as well as aerosol water content). All the processes led to a net 918 sulfate production rate of 2.7 g m -3 h -1 , in which chemistry played a predominant role (IPR concentrations. An observational study for the same haze period in Beijing (Ma et al., 2017) 970 also suggested the important role of regional transport from the south of Beijing in haze 971 formation.

972
For sulfate (Figure 9d), although chemical processes still contributed most to sulfate 973 production in the growth stage (6.0 g m -3 h -1 ), it is noticed that gas chemistry (5.9 g m -3 h -1 ) 974 accounted for most of the sulfate production, whereas contribution from heterogeneous 975 reactions was smaller than that in the first haze episode mainly due to lower relative humidity.

976
In the growth stage, the net IPR was 1.9 g m -3 h -1 , 30% smaller than that for the first haze, 977 indicating a weaker secondary aerosol formation during this haze episode. In the persistence 978 stage, sulfate production from gas phase oxidation was almost balanced by the loss from dry  for VADV, respectively, which were obviously smaller than those in the first haze episode. In 1000 the dissipation stage, physical processes except HADV all contributed to the loss of PM2.5.

1001
Compared with the first haze episode, the negative IPR of VADV decreased mainly due to the 1002 larger wind speeds in this episode, as more PM2.5 was removed by VADV, the remaining 1003 PM2.5 loss by vertical diffusion decreased, consequently a weakened VDIF. The positive IPR 1004 of HADV increased as well due to larger wind speed than that in the first episode in this 1005 stage.

1006
The above process analyses reveal that for the first haze episode (20-26 February) in 1007 Beijing, local emissions and chemical processes were the main contributors to the formation 1008 and persistence of the haze event. However, for the second haze (1-4 March), regional

1023
The definition of the four stages during haze evolution is the same as that in section 5.1.1. h -1 from HADV+HDIF) and led to a net ΔIPR of 0.04 g m -3 h -1 (Figure 10b)   For nitrate (Figure 10d), the feedback-induced IPR changes in the clean stage were 1067 similar to those for sulfate. In the growth stage, remarkable increases in nitrate formation 1068 from Thermo and HET processes occurred, with the ΔIPRs of 3.30 g m -3 h -1 and 0.50 g m -3 1069 37 h -1 , respectively (Figure 10d). The increased gas precursors and RH due to the aerosol 1070 feedback reinforced chemical formation processes. In this stage, the overall ΔIPR was 3.90 1071 g m -3 h -1 , suggesting a faster increasing rate in nitrate concentration in consideration of 1072 aerosol feedback. In the persistence stage, the ΔIPR by Thermo was smaller than that in the 1073 growth stage (Figure 10d). This could be explained that the apparent increase in sulfate h -1 from physical processes (HADV+VADV+HDIF+VDIF+DDEP) (Figure 10a). In the 1087 growth stage, the net ΔIPR was 9.50 g m -3 h -1 , which meant in every hour, approximate 9.50 1088 g m -3 of PM2.5 mass was elevated in Beijing due to the feedback effect. The above 1089 feedback-induced difference in the change rate of PM2.5 (ΔIPR) resulted from a combined 1090 effect from chemical processes (7.27 g m -3 h -1 ) and physical processes (2.23 g m -3 h -1 ), 1091 which suggested that chemical processes contributed more to the PM2.5 increase than physical 1092 processes. However, it was noted that the increased contribution from chemical processes 1093 was related to increasing gas precursors, which was partly associated with physical processes.