Aerosol‐Radiation Interactions in China in Winter: Competing Effects of Reduced Shortwave Radiation and Cloud‐Snowfall‐Albedo Feedbacks Under Rapidly Changing Emissions

Abstract Since 2013, Chinese policies have dramatically reduced emissions of particulates and their gas‐phase precursors, but the implications of these reductions for aerosol‐radiation interactions are unknown. Using a global, coupled chemistry‐climate model, we examine how the radiative impacts of Chinese air pollution in the winter months of 2012 and 2013 affect local meteorology and how these changes may, in turn, influence surface concentrations of PM2.5, particulate matter with diameter <2.5 μm. We then investigate how decreasing emissions through 2016 and 2017 alter this impact. We find that absorbing aerosols aloft in winter 2012 and 2013 heat the middle‐ and lower troposphere by ∼0.5–1 K, reducing cloud liquid water, snowfall, and snow cover. The subsequent decline in surface albedo appears to counteract the ∼15–20 W m−2 decrease in shortwave radiation reaching the surface due to attenuation by aerosols overhead. The net result of this novel cloud‐snowfall‐albedo feedback in winters 2012–2013 is a slight increase in surface temperature of ∼0.5–1 K in some regions and little change elsewhere. The aerosol heating aloft, however, stabilizes the atmosphere and decreases the seasonal mean planetary boundary layer (PBL) height by ∼50 m. In winter 2016 and 2017, the ∼20% decrease in mean PM2.5 weakens the cloud‐snowfall‐albedo feedback, though it is still evident in western China, where the feedback again warms the surface by ∼0.5–1 K. Regardless of emissions, we find that aerosol‐radiation interactions enhance mean surface PM2.5 pollution by 10%–20% across much of China during all four winters examined, mainly though suppression of PBL heights.

. Aerosol mask used for eliminating the effect of aerosols on radiation over China Figure S2. Comparison between GEOS-GC and MERRA-2 for surface air temperature and PBL height for DJF (2012-13 & 2016-17) Figure S3. Comparison between GEOS-GC and GC-Offline PM2.5 for DJF (2013 and 2016-17) Figure S4. Impact of aerosol-radiation interactions on radiative flux and heating for 2012-13 (DJF) Figure S5. Standard deviation of the impact of aerosol-radiation interactions on air temperature at 850 hPa for 2012-2013 (DJF) and 2016-17 (DJF) Figure S6. Climatology and impact of aerosol-radiation interactions on selected meteorological variables for 2012-13 (DJF) Figure S7. Mean wind speed and direction for 2012-13 (DJF) Figure S8. Reconstructed AAOD at 500 nm from simulated mean aerosol optical depth (AOD) for December-January-February (DJF) as simulated in GEOS-GC Figure S9. Impact of aerosol-radiation interactions at 700 hPa on short wave heating and air temperature for 2012-13 (DJF) Figure S10. Change between 2012-13 and 2016-17 in December-January-February (DJF) in the clear sky impacts of aerosols on shortwave radiation absorbed by the atmosphere and shortwave radiation reaching the surface Figure S11. Change in impact of aerosol-radiation interactions between 2012-13 and 2016-17 (DJF) on shortwave radiation absorbed by the atmosphere, shortwave heating rate at 850 hPa, cloud liquid water path, falling snow at the surface, and surface albedo Figure S12. Changes between 2012-13 and 2016-17 for December-January-February (DJF) column mean single scattering albedo (SSA) Figure S13. Impact of aerosol-radiation interaction for 2016-17 (DJF) for feedbacks involving aerosol, clouds, snow, and surface albedo Figure S14. Effect of aerosol-radiation interactions on atmospheric stability and PM2.5 for 2016-17 (DJF) Figure S15. Change in impact of aerosol-radiation interactions between 2012-13 and 2016-17 (DJF) on geopotential height and windspeed at 700 hPa Figure S16. Impact of aerosol-radiation interactions on radiative flux and heating for 2016-17 (DJF) Figure S17. Climatology and impact of aerosol-radiation interactions on selected meteorological variables for 2016-17 (DJF) Figure S18. Mean air temperature at 850 hPa for 2012-2013 (DJF) and 2016-17 (DJF) Figure S19. Mean wind speed and direction for 2016-17 (DJF) Figure S20. Impact of aerosol-radiation interactions at 700 hPa for 2016-17 (DJF) on shortwave heating and air temperature Figure S21. Impact of aerosol-radiation interactions on atmospheric circulation patterns for 2016-2017 December-January-February (DJF)

S1: GEOS-ESM and GOCART model details
GEOS-ESM uses a modular Earth System Modeling Framework (ESMF, Hill et al., 2004). The model features a cubed sphere horizontal grid (Putman and Lin, 2007), with a hybrid sigma-pressure discretization for vertical coordinates (Simmons and Burridge, 1981). Largescale dynamics and transport are computed with a flux-form semi-Lagrangian finite-volume dynamics scheme (Lin, 2004;Putman and Lin, 2007). Convection is parameterized with a Relaxed Arakawa-Schubert scheme (Moorthi and Suarez, 1992), and deep convection pairs this scheme with a stochastic Tokioka-type trigger function (Tokioka et al., 1988). Cloud cover and cloud water and ice volume are calculated with a single moment parametrization based on water vapor, condensate mass flux, and relative humidity, but not depending on aerosol concentration (Bacmeister et al., 2006;Rienecker et al., 2008;Molod et al., 2012). Snow cover area and albedo are derived via a three-layer snow model coupled to a catchment based hydrology model that takes into account vegetation masking by land cover type. The model yields an area-weighted composite albedo. Effects of aerosol deposition on snow are not taken into account (Koster et al., 2000;Stieglitz et al., 2001). Albedo in snow-free areas is determined using MODIS data (Mood et al., 2005;Rienecker et al., 2008). Boundary layer mixing in the model is based on Lock et al. (2000), which depends on the Richardson number (Louis et al, 1982). Over land, PBL height is calculated as a function of the bulk Richardson number (McGrath-Spangler and Molod, 2014). The radiation scheme considers absorption by aerosols in the longwave and both scattering and absorption of aerosols in the shortwave (Chou et al., 1992;Chou and Suarez, 1994).
Aerosol species in GOCART include sulfate, nitrate, ammonium, sea salt, dust, organic carbon (OC), and black carbon (BC). Sulfate is formed via gas-phase and aqueous-phase oxidation of SO2, using prescribed oxidant fields (Chin et al., 2000;Chin et al., 2002;Colarco et al., 2010). Nitrate is divided into three size bins, while dust and sea salt each have five size bins (Bian et al., 2017;Colarco et al., 2017). Particulate nitrate formed through thermodynamic partitioning is assigned the smallest size bin, while nitrate formed from heterogeneous reactions on dust or sea salt is distributed among size bins, depending on the size of the particles on which the heterogeneous reactions occur (Bian et al., 2017;Colarco et al., 2017). BC and OC are divided into hydrophilic and hydrophobic fractions, and only primary OC is considered (Chin et al., 2002;Colarco et al., 2010).
Aerosol optical properties are calculated for each species and size bin assuming constant particle densities as well as lognormal sub-bin particle size distributions for all aerosol types other than dust. For dust, we apply a power-law distribution for the sub-bin particle number size distributions (Chin et al., 2002;Colarco et al., 2010). Calculation of aerosol optical properties accounts for the effects of relative humidity on hygroscopic growth, for most species. For dust aerosol, the optical properties are independent of relative humidity (Chin et al., 2002;Colarco et al., 2010).

S2: Model evaluation
In our model setup, GEOS-ESM is freely running and not tied to observations, except for the initialization, when one month is nudged to MERRA-2 assimilated meteorology (Section 2.4). Here we compare the 4-year DJF mean simulated surface temperature and PBL heights from GEOS-GC with those from MERRA-2. To calculate statistical significance, we conduct a pooled t-test, with one sample comprising the MERRA-2 DJF monthly means across 2012-2013 and 2016-2017 and the other sample comprising the DJF monthly means for those years across all five ensemble members. We find that GEOS-GC has a positive bias of ~4 K for much of northwestern and north-central China ( Figure S1). GEOS-GC also underestimates surface air temperature compared to MERRA-2 for part of south-central China, near the eastern edge of the Tibetan Plateau. GEOS-GC underestimates PBL height compared to MERRA-2 by 25% or more for most of southern China, with the largest bias seen on the edges of the Tibetan Plateau ( Figure  S2). In contrast, GEOS-GC overestimates PBL height compared to MERRA-2 by ~25% for northern China along the border of Mongolia ( Figure S1). These differences arise from the using the model radiation scheme with GEOS-Chem aerosols rather than GOCART aerosols and from the absence of observational data assimilation in the GEOS-GC simulations.
These differences in meteorology lead inevitably to differences in the PM2.5 generated by GEOS-GC compared to that in GC-Offlinei.e., the offline version of GEOS-Chem driven by MERRA-2 assimilated meteorological fields. Here we compare the 2013, 2016, and 2017 ensemble mean of wintertime (DJF) surface PM2.5 concentrations from GEOS-GC with those from the GC-Offline simulation described by Zhai et al. (2021). As the Zhai et al. (2021) simulation has very low biases in PM2.5 compared to surface observations, the close match between GEOS-GC and GC-Offline suggests relatively small errors in the GEOS-GC PM2.5 compared to observations To calculate statistical significance in the differences between the two sets of simulations, we conduct a pooled t-test, with monthly mean PM2.5 concentrations from GC-Offline as one sample and monthly means of PM2.5 across all five GEOS-GC ensemble members as the other sample. We find that the ensemble mean of our simulations has a maximum bias for PM2.5 in DJF of ~40 μg m -3 in a few small areas in eastern China, but these differences are generally not statistically significant across most of the region ( Figure S2). Although not statistically significant, the local maxima in PM2.5 in GEOS-GC has a larger difference compared to GC-Offline than the local maxima across much of eastern China. The higher bias over the Sichuan basin is due to the low bias in PBL heights in the region compared to MERRA-2. The PBL height bias is smaller over the east coast of China, meaning the local maxima in PM2.5 across eastern China is less affected by this issue.    (Zhai et al., 2021). Panel (b) shows results from the five-member ensemble mean of GEOS-GC run in freely running mode. Panels (c) shows the differences between GEOS-GC and GC-Offline. In panel (c) only statistically significant differences (p < 0.05) are shown. Figure S4: Impact of aerosol radiation interactions on shortwave radiative flux and heating for 2012-2013 December-January-February (DJF). Lefthand panels show results for all-sky conditions while righthand panels show clear-sky conditions. The impacts are calculated as the difference between the ensemble means for GEOS-GC and GEOS-GC-China0. Only statistically significant changes (p < 0.05) are shown. Panels (a) and (b) present the change in shortwave radiation reaching the surface, panels (c) and (d) the change in shortwave radiation absorbed by the entire atmospheric column, and panels (e) and (f) the change in radiative heating rate due to shortwave (SW) radiation at 850 hPa.     (Colarco et al., 2010), and 0.9 for organic aerosol and brown carbon. Panels (c) and (d) use an assumed mean SSA of 0.88 for East Asian aerosol, derived from AERONET retrievals of mixed East Asian aerosol (Li et al., 2015).     (Colarco et al., 2010), and 0.9 for organic aerosol and brown carbon.   Figure S16: Impact of aerosol radiation interactions on shortwave radiative flux and heating for 2016-2017 December-January-February (DJF). Lefthand panels show results for all-sky conditions while righthand panels show clear-sky conditions. The impacts are calculated as the difference between the ensemble means for GEOS-GC and GEOS-GC-China0. Only statistically significant changes (p < 0.05) are shown. Panels (a) and (b) present the change in shortwave radiation reaching the surface, panels (c) and (d) the change in shortwave radiation absorbed by the entire atmospheric column, and panels (e) and (f) the change in radiative heating rate due to shortwave (SW) radiation at 850 hPa. Figure S17: Impact of aerosol-radiation interactions for 2016-2017 December-January-February (DJF) for selected variables. The left column shows the mean GEOS-GC values of (a) total precipitation rate and (c) cloud liquid water path, and (e) snowfall at the surface. The right column shows the impacts of aerosol-radiation interactions on (b) precipitation rate, (d) cloud liquid water path, and (f) snowfall at the surface. Only statistically significant changes (p < 0.05) are shown.  Figure S120: Impact of aerosol radiation interactions at 700 hPa on (a) all-sky shortwave radiative heating and (b) air temperature for 2016-2017 December-January-February (DJF). The impacts are calculated as the difference between the ensemble means for GEOS-GC and GEOS-GC-China0. Only statistically significant changes (p < 0.05) are shown. Figure S21: Impact of aerosol-radiation interactions on atmospheric circulation patterns for 2016-2017 December-January-February (DJF). The panels show the changes in geopotential height (a) and in wind speed and direction at (b) 700 hPa, (c) 500 hPa, and (d) 850 hPa. The impacts are calculated as the difference between the ensemble mean for GEOS-GC and GEOS-GC-China0. Only statistically significant changes (p<0.05) are shown for geopotential height and wind speed. Arrows for (b-c) show the net direction of the wind changes. Blue-shaded regions thus indicate a decrease in wind speed in the opposite direction of the overlying arrows.