The impact of the direct effect of aerosols on meteorology and air quality using aerosol optical depth assimilation during the KORUS-AQ campaign

Abstract

To quantify the impact of the direct aerosol effect accurately, this study incorporated the Geostationary Ocean Color Imager (GOCI) aerosol optical depth (AOD) into a coupled meteorology-chemistry model. We designed three model simulations to observe the impact of AOD assimilation and aerosol feedback during the KORUS-AQ campaign (May – June 2016). By assimilating the GOCI AOD with high temporal and spatial resolutions, we improve the statistics from the comparison AOD and AERONET data (RMSE: 0.12, R: 0.77, IOA: 0.69, MAE: 0.08). The inclusion of the direct effect of aerosols produces the best model performance (RMSE: 0.10, R: 0.86, IOA: 0.72, MAE: 0.07). AOD values were increased as much as 0.15, which is associated with an average reduction in solar radiation of −31.39 W/m², a planetary boundary layer height (−104.70 m), an air temperature (−0.58 °C), and a surface wind speed (−0.07 m/s) over land. In addition, concentrations of major gaseous and particulate pollutants at the surface (SO₂, NO₂, NH₃, ${\text{SO}}_4^{2-}$, ${\text{NO}}_3^{-}$, ${\text{NH}}_4^{+}$, PM_2.5) increase by 7.87 – 34% while OH concentration decreases by −4.58 %. Changes in meteorology and air quality appear to be more significant in high-aerosol loading areas. The integrated process rate analysis shows decelerated vertical transport, resulting in an accumulation of air pollutants near the surface and the amount of nitrate, which is higher than that of sulfate because of its response to reduced temperature. We conclude that constraining aerosol concentrations using geostationary satellite data is a prerequisite for quantifying the impact of aerosols on meteorology and air quality.

1. Introduction

East Asia often undergoes elevated aerosol concentrations originating from a wide spectrum of sources such as dust from the Taklamakan and Gobi Deserts [Kurosaki and Mikami, 2003; Takemura et al., 2002; Wang et al., 2008a], biomass burning primarily from Southeast Asia [Fu et al., 2012; Lin et al., 2013; Streets et al., 2003a; Uranishi et al., 2019], and anthropogenic emissions [Giorgi, 2002; Kang et al., 2019; Streets et al., 2003a]. An increases in anthropogenic emissions in Asia, accompanied by rapid growth of its economies, has been ongoing for many years; recently, however, many Asian countries have implemented clean air policies [Ohara et al., 2007; Souri et al., 2017; Streets et al., 2003b; Zheng et al., 2018]. Streets et al. [2003b] estimated an emission inventory in Asia based on TRACE-P experiments by the National Science Foundation (NSF) and the National Oceanic and Atmospheric Administration (NOAA) in 2000. Among the many countries in Asia, China was found to be the most dominant source of emissions. Since 2010, however, China has been implemented several clean air policies [Zheng et al., 2018]. Souri et al. [2017] showed that the trend of NO₂ in China changed to a decreasing trend in 2011 and 2012 by analyzing the long-term trend of tropospheric emission using the Ozone Monitoring Instrument (OMI) data from 2005 to 2014. In addition, the following study by Zheng et al. [2018] showed that Chinese anthropogenic emissions decreased by 17 – 62% from 2010 to 2017 and the emission reduction rate in China had accelerated since 2013 when China launched the Action Plan on the Prevention and Controls of Air Pollution.

Aerosols in the atmosphere can often degrade air quality, causing adverse effects on human health and the environment. Aerosols modify the radiation budget by the direct and indirect effects of aerosols. The direct effect is referring to the scattering or absorption of solar radiation by aerosols in the atmosphere. Reduction of solar radiation reaching the ground changes the surface air temperature and planetary boundary layer height (PBLH). The indirect effect is a result of aerosols acting as cloud condensation nuclei (CCN) and ice nuclei (IN), which alter cloud characteristics such as albedo and lifetime with smaller water droplets [Twomey et al., 1984; Giorgi, 2002; Lohmann and Feichter, 2005]. Meteorological changes such as solar radiation, air temperature, planetary boundary layer height, and wind speed by aerosol may affect air quality. To achieve a more thorough understanding of meteorology, air quality, and their interaction, we must identify the effects of aerosols on the radiative budget [Wong et al., 2012]. A number of studies have used satellite data and numerical models to quantify both the direct and indirect effects of aerosols and meteorological (climatological) responses [Haywood and Boucher, 2000; Takemura et al., 2005; Jacob and Winner, 2009; Wang et al., 2014; Xing et al., 2015a; Xing et al., 2015b; Wong et al., 2012; Xing et al., 2017; Yu et al., 2013]. Xing et al. [2015b] found that the direct effect of aerosols increases the concentration of surface air pollutants as a result of the stabilization of the atmosphere, which reduces ventilation. Wang et al. [2014] showed that reduced surface solar radiation of up to 53% in Beijing reduced the PBLH from 690 m to 590 m, which in turn it increased PM_2.5 concentrations during the haze period in 2013. In addition, because of changes in atmospheric dynamics and photolysis rate by direct aerosol effects, the O₃ formation over a surface can be profoundly affected [Xing et al., 2017; Wong et al., 2012].

Such model studies, however, are limited by the high uncertainty of their aerosol simulations, especially in East Asia with high aerosol concentrations from various sources. This uncertainty may affect the accuracy of the quantification of aerosol effects. The accuracy of these models may be insufficient because of a lack of understanding of chemistry, dynamics and transport processes, problematic emissions inventories, and meteorological input data [Chai et al., 2017; Park et al., 2011]. For many years, numerous efforts have been devoted to assimilating observations into three-dimensional chemical transport models (CTMs) to improve the simulation of aerosol concentrations. Recently, well-characterized observations have become available from various sources (e.g., in-situ observations, ships, airplanes, and satellites), that can potentially be used for improving the accuracy of model predictions [Chai et al., 2017; Lee et al., 2016; Park et al., 2011; Saide et al., 2014]. Among the various sources of observational data, remote sensing satellites have broader spatial and temporal coverage than in-situ observations. Chai et al. [2017], using a combination of AOD data from Moderate Resolution Imaging Spectroradiometer (MODIS) and Airnow PM_2.5 measurements in North America, showed considerably enhanced model performance. Park et al. [2011] employed multiple methods to enhance the accuracy of CMAQ (Community Multiscale Air Quality) AOD, by assimilating QuikSCAT wind data and MODIS AOD.

The geostationary equatorial orbit satellite also provides continuous observational data over certain domains, so incorporating geostationary data could assist researchers in their efforts to close the knowledge gap and increase the reliability of CTMs over East Asia. Fortunately, the Geostationary Ocean Color Imager (GOCI), the first ocean color imager in geostationary orbit, has provided hourly aerosol optical properties eight times per day over East Asia since 2010 [Choi et al., 2018]. Since high temporal and spatial observational data from the GOCI has become available, many studies that focused on East Asia have benefited from the use of these data [Jeon et al., 2016; Lee et al., 2016; Pang et al., 2018; Park et al., 2014; Saide et al., 2014a; Xu et al., 2015]. Saide et al. [2014] assimilated both the GOCI and MODIS AOD data into the chemistry version of the Weather Research and Forecasting (WRF-Chem) model from April to May in 2012 in the Korean Peninsula. They showed that assimilating satellite AOD data considerably improves the accuracy of CTM by mitigating the underestimation of surface aerosol concentrations for periods with high aerosol concentrations from dust, biomass burning, and anthropogenic pollution. Moreover, improvement in model accuracy by GOCI AOD assimilation was five times as high as that of MODIS AOD assimilation, suggesting that geostationary data with higher temporal and spatial resolutions are more likely to improve model performance.

In this study, we used a data assimilation method to incorporate GOCI AOD into the WRF-CMAQ two-way coupled model that addresses direct aerosol radiative forcing and its effects on both meteorology and air quality during the KORUS-AQ campaign period. We performed three model simulations to determine the impact of the AOD assimilation and aerosol feedback and compared them to the conventional base run. The structure of this paper is as follows. Section 2 explains the WRF-CMAQ two-way modeling system, observations, and the assimilation method, Section 3 presents detailed results of the three model simulations from both the AOD assimilation and aerosol feedback, and Section 4 contains the summary and discussion.

2. Overview of modeling system, observational data and method

2.1. WRF-CMAQ two-way coupled model

The WRF-CMAQ two-way coupled model addresses the feedback between the chemistry and the meteorology [Wong et al., 2012]. The coupled model consists of two sub-models, the Weather Research and Forecasting (WRF) model v3.8 and the CMAQ model v5.2 [Byun and Schere, 2006], released by the U.S Environmental Protection Agency (EPA). A detailed description of the WRF-CMAQ two-way model can be found in Wong et al. [2012].

The two sub-models share a single domain with a 27 km horizontal resolution, covering the eastern part of China, Korea, and Japan, shown in Figure 1. The WRF model was configured to use the WRF Single-Moment 3-class (WSM3) scheme [Hong et al., 2004], the RRTMG scheme for longwave and shortwave radiation, the Pleim-Xiu land surface model [Pleim and Xiu, 1995; Xiu and Pleim, 2001] with the Pleim surface-layer scheme [Pleim, 2006], and the ACM2 planetary boundary layer (PBL) model [Pleim, 2007a, 2007b] and Kain-Fritsch (KF2) schemes [Kain, 2004] for cumulus parameterization. To improve the performance of the surface temperature simulation, we applied indirect soil moisture and temperature nudging techniques [Pleim and Gilliam, 2009; Pleim and Xiu, 2003]. With regard to the initial and boundary conditions, we used one-degree by one-degree NCEP FNL (final) operational global analysis data. For a reasonable sea surface temperature, we used a 0.5-degree real-time global sea surface temperature (RTG SST) [Thiébaux et al., 2003]. Additionally, we applied a four-dimensional data assimilation (FDDA) option every 6 hours above the PBL for the temperature, the water mixing ratio, and wind components with an order of magnitude of 10⁻⁵.

Figure 1.

Map of the model domain. The magenta dashed line denotes GOCI coverage. The AERONET sites are shown in red dots for validation.

We used carbon bond version 5 (CB05) [Sarwar et al., 2012] for the gas phase and AERO6 for the aerosol chemical mechanisms in the CMAQ and obtained anthropogenic emissions in Asia from the 2010 MIX emission inventory [Li et al., 2017] at a 0.25-degree spatial resolution. This inventory contains monthly averaged CB05 emission information that includes ten chemical species (sulfur dioxide (SO₂), nitrogen oxides (NO_X), carbon monoxide (CO), non-methane volatile organic compounds (NMVOC), ammonia (NH₃), particulate matter (PM₁₀ and PM_2.5), black carbon (BC), organic carbon (OC) and carbon dioxide (CO₂)) in five sectors (i.e., power, industrial, residential, transportation, and agricultural). For Korea, we used the 2011 Clean Air Policy Support System (CAPSS) emission high-resolution (1km) inventory from the National Institute of Environmental Research (NIER). Additionally, to estimate dynamic biogenic emissions, we used the Model of Emissions of Gases and Aerosols from Nature (MEGAN) version 2.1, a standalone version [Guenther et al., 2012] that requires various input data such as meteorological input data, plant functional types (PFTs), and the leaf area index (LAI). We provided WRF output for meteorological input data and a modified version of the MODIS LAI product [Yuan et al., 2011] for the LAI, which changes over time. For the PFTs, we used the recently developed 0.05° spatial resolution PFTs from Ke et al. [2012], and to generate chemical initial boundary conditions for the CMAQ model, we also used GEOS-Chem v10–01 [Bey et al., 2001].

We used the AOD at a 550nm wavelength computed within the coupled model based on the IMPROVE (Interagency Monitoring of Protected Visual Environments) reconstruction method suggested by Pitchford et al. [2007] that calculates the AOD by integrating the aerosol mass extinction coefficient (MEE, σ_ext(z)) regarding the altitude and the aerosol extinction coefficient. The light extinction is derived from six major components—ammonium sulfate, ammonium nitrate, organic mass, elemental carbon, fine soil, and sea salt—and applied hygroscopic enhancement factors to the MEE of ammonium sulfate, ammonium nitrate, and sea salt aerosols.

We ran the simulation between April 10 and June 19, 2016, and spent 22 days in April for model spin-up. We set the time step for the WRF at 120 seconds and the call ratio between the WRF and the CMAQ as 3:1, which indicates that the coupled model exchange data, that is, CMAQ, receives meteorological information, and WRF receives aerosol feedback data from CMAQ every 360 seconds. We conducted three experiments: with aerosol feedback and AOD assimilation (hereafter, YF Assim), without aerosol feedback and with AOD assimilation (hereafter, NF Assim), and a base run without either aerosol feedback or AOD assimilation (hereafter, NF). These three simulations are briefly explained in Table 1.

Table 1.

Description of the three experiments in this study.

Experiments	Aerosol Feedback	AOD Assimilation
NF	off	Off
NF Assim	off	On
YF Assim	on	On

2.2. GOCI/MODIS AOD

To obtain an optimal estimation of our regional model with regard to the AOD, we used the AOD from the GOCI sensor from the geostationary orbit onboard the Communication Ocean and Meteorological Satellite (COMS). GOCI level 1B (L1B) data provided hourly daylight spectral images eight times a day (9:30 – 16:30 LST) over East Asia, extending 2,500 km by 2,500 km centered at 36°N, 130°E in eight spectral channels (412, 443, 490, 555, 660, 680, 745, and 865 nm) [Choi et al., 2016, 2018]. We acquired AOD data at 550 nm at a 6 km spatial resolution from the improved GOCI Yonsei aerosol retrieval (YAER) algorithm version 2 (V2), which generates advanced cloud-masking procedures and surface reflectance calculations [Choi et al., 2018]. Choi et al. [2018] compared GOCI AOD to the Aerosol Robotic Network (AERONET) in East Asia and the Sun-Sky Radiometer Observation network (SONET) in China observation from 2011 to 2016, the GOCI AOD YAER V2 data showed higher accuracy with reduced bias (0.01 for land and ocean) and a greater f within EE_MDT (i.e., the fraction of data within EE_MDT (the expected error of the MODIS dark spot)) (60% for land and 71% for the ocean) than the V1 data (−0.07 (42%) for land and 0.04 (62%) for the ocean).

Figure 1 shows that as the GOCI domain covers only a portion of the model domain, it may introduce two main problems: i) As GOCI images do not cover known dust sources located in northern China, Mongolia, and Kazakhstan, considering only GOCI AOD is only capable of capturing outflow of dust and it may not be sufficient. ii) GOCI AOD assimilation into the CMAQ model causes significant changes in the total number of aerosols, but in regions smaller than the model domain. Hence, such changes may lead to unexpected problems related to the consistency of transport and chemistry. To address these problems, we assimilated MODIS AOD only outside of the GOCI domain to enlarge the coverage of the AOD data within the domain of this study and to minimize the difference inside and outside of the GOCI domain.

MODIS (Collection 6.1) Level 2 AOD data at 550 nm provides global ambient aerosol optical properties over ocean and land with 10 km spatial resolution from Terra and Aqua platforms (MOD04 and MYD04). For the data quality, we used AOD values having quality flag greater than zero. Recently, Wei et al. [2019] evaluated MODIS Collection 6.1 AOD product with AERONET data at 384 sites from 2013 to 2017, the Collection 6.1 AOD data showed better accuracy globally than the previous product.

2.3. DC-8 aircraft measurements during the KORUS-AQ

To compare the model simulations, we used NASA DC-8 aircraft measurement data from the columnar AOD at 550 nm by 4STAR (Spectrometer for Sky-Scanning, Sun-Tracking Atmospheric Research), the air temperature, the wind speed, SO₂, NO₂, O₃, ${\text{SO}}_4^{-}$, ${\text{NO}}_3^{-}$, and ${\text{NH}}_4^{+}$ during the KORUS-AQ campaign. We used one-minute averaged DC-8 aircraft measurement data and excluded data measured at an altitude of 500 m or less. Most flight measurements were taken over the Korean Peninsula and the Yellow Sea during the KORUS-AQ campaign, which identified the controlling factors of air quality in the Korean Peninsula. The modeled data were spatially and temporally model grid points matched to the observations. In particular, as observed AOD values represent columnar AOD from the altitude of the flights to the top of the atmosphere, we used modeled AOD calculated by the summation of the partial AOD from a certain level of the flights to the top.

2.4. Optimal Interpolation

To more accurately represent aerosol concentrations over East Asia, we used the GOCI AOD with the optimal interpolation (OI) method. The OI method, which generates analyzed values from modeled values and observations, is an effective tool for minimizing model uncertainty [Candiani et al., 2013; Chai et al., 2017; Tang et al., 2015; Wang et al., 2013]. The duration of adjustment by the OI method varied, depending on the lifetime of the target species and the frequency of the adjustment [Tang et al., 2015].

In this study, we adopted the OI method carried out by Tang et al. [2015] and assumed no correlation between background and observation errors, the error covariances of the background and observations with zero bias Gaussian probability density functions, and a linear relationship between the background and observations. To define the weight of the observations, determined by the distance between the observations and the model grids within the influence of radiance, we added the Cressman function (W). The analyzed the values from the OI method as follows:

$${\text{X}}_{\text{a}}={\text{X}}_{\text{b}}+{\text{BH}}^{\text{T}}{\left({\text{HBH}}^{\text{T}}+{\text{W}}^{-1}\text{O}\right)}^{-1}\left(\text{Y}-{\text{HX}}_{\text{b}}\right)$$ (1)

$$\text{W}=\left({\text{R}}^2-{\text{r}}^2\right)/\left({\text{R}}^2+{\text{r}}^2\right)$$ (2)

where X_a and X_b are the analyzed (posteriori) and background (priori) values, respectively, and B and O are the error covariance matrices of the background (priori) and the observations, respectively. To set the observation error covariance, we followed Choi et al. [2018]. H is the observational operator, R (60 km in this study) the influence radius, and r the Euclidian distance between the observations and the model grid cell. Given the spatial irregularities and random noise associated with the GOCI AOD, we determined a geometric weighting function with a constant radius of influence that incorporates a relatively larger number of observations beyond the target model grid cell into the assimilation framework. Using all qualified GOCI pixels from May to June 2016 (Figure S1(a)), we compared the normalized cumulative semivariogram to the normalized distance to identify the underlying function and found that the Cressman function accurately (r² = 0.98) represented the spatial autocorrelation of the data. We further determined the radius of influence based on the distance at which the variance between AOD observations was equivalent to that of the retrieval reported in Choi et al. [2018], shown in Figure S1(b). When satellite AOD data were available (8 times per day), we used the ratio X_a/X_b of each grid point to adjust all aerosol components of the model initial condition throughout the entire column.

3. Results

3.1. The impact of the satellite AOD assimilation and the aerosol feedback on the AOD

Figure 2 shows the time series of the 24-hour average simulated AOD from the three experiments and compare them to AERONET observations located in the GOCI domain from May to June 2016. Using Ångström exponent information provided for each AERONET site, AOD at 550 nm from AERONET was calculated. NF without either assimilation or aerosol feedback could capture the general trend of AOD compared to observation during the study period; this scenario, however, showed some biases: overestimations on May 10 and May 26, and June 4 to 6, as well as underpredicted AOD values, particularly between May 19 and May 24. The results of NF Assim tended to yield more accurate AOD values than those of NF after incorporating the satellite observations through the OI method. For instance, highly overestimated peaks on May 10 and June 4 were mitigated. Thus, the severe underestimation between May 19 and May 24 had significantly improved. The benefits of AOD assimilation were reflected in the comparison of back-scattering coefficients from CALIPSO Lidar L1B profile data at 533 nm and the corresponding simulated coefficients (NF and NF Assim) at 550 nm on May 22 when NF produced a severe underestimation (Figure 3). Compared to CALIPSO data (Figure 3 (a)), NF (Figure 3 (b)) was able to simulate a plume observed at an altitude of 3–4 km, but the magnitude and the values over the surface were significantly underestimated. With the GOCI AOD assimilation, the results of NF Assim (Figure 3 (c)) showed well-adjusted values in magnitude. Aerosol plumes appearing at 3 km, around 34–40°E, and aerosol flumes in the upper atmosphere were more realistically captured.

Figure 2.

Comparison of the time series of the AOD and AERONET observations for the entire GOCI domain.

Figure 3.

Comparison of the back-scattering coefficient with (a) CALIPSO Lidar L1B profile data (532 nm) and modeled back-scattering coefficients from (b) NF and (c) NF ASSIM at 550 nm.

The results of YF Assim, shown in Figure 2, indicate that the direct effect of aerosols can improve AOD. The direct effect of aerosols both increased and decreased AOD. More specifically, on June 4, the AOD value (0.95) of NF Assim had significantly improved compared to observation (0.53) than that of NF (1.70), but it was still overestimated. Encouragingly, YF Assim with the direct effect of aerosols reduced the AOD value by 25%.

Table 2 shows the statistical results for simulated AOD averaged over 24 hours at all AERONET observations in the GOCI domain throughout the study period. Overall, NF produced the largest error (RMSE: 0.23, MAE: 0.13) and the lowest correlation (CORR: 0.64, IOA: 0.49) among the three experiments. After the AOD assimilation, NF Assim, with a much smaller error (RMSE: 0.12, MAE: 0.08) and an increased correlation (CORR: 0.77, IOA: 0.69), improved accuracy, shown in Figure 2. YF Assim, with the direct effect of aerosols, generated the best results for all statistical values (RMSE: 0.10, MAE: 0.07, CORR: 0.86, IOA: 0.72). These results indicate that the aerosol adjustment by OI method with the GOCI AOD produced relatively accurate aerosol estimation. Thus, it also enabled us to ascertain the impact of the direct effects of aerosols more precisely.

Table 2.

Statistical results of the comparison of the averaged AOD the AERONET observations from May to June 2016 (RMSE: root-mean-square error; CORR: correlation, MAE: mean-absolute error; IOA: index of agreement).

Experiments	RMSE	CORR	MAE	IOA
NF	0.23	0.64	0.13	0.49
NF Assim	0.12	0.77	0.08	0.69
YF Assim	0.10	0.86	0.07	0.72

Figure 4 displays the contour maps of the averaged GOCI AOD and simulated AOD from the three experiments during the study period and the GOCI time period. Also, Figure 4 (e) shows the effects of the GOCI AOD assimilation, and Figure 4 (f) shows the direct effect of aerosols. According to the GOCI AOD, eastern China, with its high anthropogenic emissions, has elevated AODs, mainly the results of a large number of industries and vehicles [Che et al., 2015; Wang et al., 2008]. A comparison between the simulated AOD from the NF experiments and the GOCI AOD data suggests that the NF significantly overestimates aerosol concentrations. Fortunately, the data assimilation, in conjunction with the GOCI AOD, enabled us to reduce the discrepancy between the simulated values and the observations. The AOD assimilation decreased AOD values up to 0.32 over East China and the Yellow Sea. By contrast, the AOD values of northern China and Japan slightly increased. To investigate the impact of AOD assimilation, we represented the comparison between the time series of the simulated AOD and observations from AERONET for sites with separate positive/negative impacts of data assimilation in Figure S2. The capability of the AOD assimilation was clearly seen at both overestimated and underestimated sites. Regarding the direct effect of aerosols, we found that AOD enhancement (in Figure 4 (f), YF Assim – NF Assim) appeared over regions in eastern China and Korea. The high values of AOD enhancement were 0.15 in Hangzhou, the Shandong Peninsula, and Tianjin, all areas of high industrial activity and emissions [Liu et al., 2019; Zhang et al., 2008].

Figure 4.

Spatial distribution of averaged (a) GOCI AOD, (b) NF, (c) NF Assim, (d) YF Assim and differences (NF Assim – NF, YF Assim – NF Assim) between the effects of AOD assimilation and aerosol direct forcing.

To gain more insight into AOD enhancement with the direct effect of aerosols, we plotted the spatial distribution of particulate speciated AOD from NF Assim and YF Assim, depicted in Figure 5. We selected only the main contributors to the total AOD for comparison. During the study period, the sulfate AOD was the most dominant component, accounting for 49.81 – 54.02% of the total AOD, followed by nitrate AOD and organic mass AOD (16.29 – 17% and 20.23 – 23.90%, respectively). Interestingly, the responses to aerosol feedback of particulate speciated AOD varied. The AOD of sulfate increased by 5.39% and nitrate by about 5.02% in the domain average.

Figure 5.

Spatial distribution of averaged major speciated AOD for the entire period of NF Assim, YF Assim and their differences (YF Assim – NF Assim)

Conversely, the AOD of organic mass decreased by about 1%. This conflicting tendency might be associated with the diverse physical and chemical processes involved in their formation and loss. High enhancement of the AOD (in sulfate and nitrate) to the direct effect of aerosols also occurred in the highly polluted area of eastern China, suggesting that the impact of the direct effect of aerosols depends on the air quality associated with feedback impact on meteorology [Xing et al., 2015b]. This will be further discussed in a later section.

Figure S3–S6 presents comparisons of the DC-8 aircraft measurement data and modeled values from three scenarios of aircraft with their flight paths. The gray shadow represents times when GOCI AOD data were available and the data assimilation was done. The columnar AODs, meteorological factors (i.e., air temperature and wind speed), and air pollutants (i.e., SO₂, NO₂, O₃, ${\text{SO}}_4^{-}$, ${\text{NO}}_3^{-}$, and ${\text{NH}}_4^{+}$) of the four flights (on May 4, May 10, May 19, and May 30) were compared. We note that most of the DC-8 aircraft measurements during the KORUS-AQ campaign were observed primarily over the Korean Peninsula and the Yellow Sea, so the impact of aerosol feedback was not clearly observed compared to the impact of AOD data assimilation because of the relatively lower AOD values over the Korean Peninsula. Nevertheless, the model performance was relatively comparable to aircraft observations.

3.2. The impact of aerosol feedback on meteorology

The direct effect of aerosols perturbs the total amount of short-wave radiation by absorbing or scattering of incoming radiation from the sun to the earth’s surface, thereby modulating several meteorological factors such as the PBLH, the air temperature, and the wind speed [Wong et al., 2012; Xing et al., 2015b]. Figure 6 shows the average spatial distributions of daily total downward short-wave radiation, the PBLH, the 2 m air temperature, the 10 m wind speed, and the cloud fraction over the ocean from NF Assim and YF Assim simulations and their differences for the entire study period at the time of GOCI observations. The reduction of simulated daily total downward short-wave radiation was −404.93 W/m² domain-wide, which is 7.8% of the total short-wave radiation because of the scattering or absorption of solar radiation by aerosols. This reduction of incoming solar radiation can suppress surface heating, which is closely associated with the production of convective eddies and upward transport [Wang et al., 2014]. Therefore, the PBLH decreased by −46.39 m over land, resulting in constrained vertical mixing of air pollutants primarily emitted from the surface, which can increase surface concentrations of air pollutants [Wang et al., 2014; Xing et al., 2011]. In addition, the 2m air temperature also decreased (by −0.52 °C over land), as did the 10 m wind speed (by −0.08 m/s). Normally, the largest reductions occurred over the regions undergoing high aerosol loadings.

Figure 6.

The direct effects of aerosols on meteorology for daily total downward short-wave radiation (SWDOWN), the PBLH, the 2 m air temperature, the 10 m wind speed, and the cloud fraction over the ocean.

The PBLH decreased by −46.39 m over land but increased by 12.70 m over the ocean. One possible explanation for this finding is that the difference between the thermal inertia of land and the ocean can result in diverse responses of the PBLH to aerosol feedback. The surface temperature over land decreases rapidly because of its relatively smaller thermal inertia, whereas changes in the temperature of the ocean occur more slowly. Because of the lower aerosol concentration, shown in Figure 6, the difference between the 2 m air temperature was relatively smaller over the ocean than it was over land. As the relatively colder air was transported from land more efficiently, the result of the increased thermal gradient between the land and the ocean, the atmospheric conditions became less stable, increasing the heat and moisture flux at the surface (shown in Figure S7), which in turn resulted in a higher PBLH over the ocean. The cloud fraction in the lowest layer also increased over the ocean. As the lower PBLH over land prevented the ventilation of pollutants over the surface, air pollution became worse, especially near sources of emissions [Wang et al., 2014; Xing et al., 2017].

Figure 7 shows the diurnal variation of meteorological variables and difference caused by aerosol feedback with regard to AOD over land. The diurnal variation over the ocean is shown in Figure S8. In Figure 7, the diurnal variation of meteorological variables shows that the impact of AOD assimilation was evident during the time when the GOCI data were available (9:30 – 16:30 LST). The maximum reduction in the PBLH was at 15 LST. Meanwhile, the strong aerosol feedback of the 2 m air temperature and the 10 m wind speed occurred slowly. Domain-wide, the largest reduction in the PBLH was −31.59 m, which corresponded to the largest reduction in short-wave radiation of −56.28 W/m², and the largest reduction increased as the AOD increased. In the model grids with an AOD higher than 0.8, which indicates a highly polluted region, the PBLH decreased even more, −89.09 m (−104.70 W/m²), the largest decrease. Consequently, the 2m air temperature and the 10 m wind speed also responded to aerosol feedback primarily caused by a reduction of short-wave radiation. The largest reduction was −0.58 °C and −0.07 m/s domain-wide for the 2 m air temperature and the 10 m wind speed, respectively. Table 3 summarizes the aerosol feedback on meteorological variables per unit AOD. All values were calculated on days with less cloud fraction (< 0.2) and low relative humidity (< 25%), which can show the relationship between the aerosol feedback and the AOD clearly. Most of the meteorological variables of East China exhibited greater aerosol feedback (16.36 – 22.36%).

Figure 7.

Diurnal averaged (a) short-wave radiation, (b) PBL height, (c) 2 m air temperature and (d) 10 m wind speed over land. The black line represents the diurnal variation of NF Assim. The black and blue dotted lines show the direct effect of aerosol (the difference between NF Assim and YF Assim) in the entire domain and grids with an AOD of more than 0.8 (polluted region).

Table 3.

Aerosol feedback on meteorological variables per unit AOD. All values were carried out during days with a small cloud fraction (< 0.2) and low relative humidity (< 25 %).

Variables	Domain-wide (land)	East China
Daily total SWDOWN	−1164.2 W/m2	−1353.5 W/m2
PBLH	−157.34 m	−177.43 m
2 m air temperature	−1.61 °C	−1.97 °C
10 m wind speed	−0.26 m/s	−0.23 m/s

3.3. The impact of aerosol feedback on air quality

As shown in 3.1, the direct effect of aerosol modifies the amount of incoming solar radiation, resulting in reduced PBLH, surface air temperature, and wind speed. This meteorological change can affect air quality because reduced incoming solar radiation reaching the surface lowers the rate of photolysis, which is relevant to the generation of radicals, and because a lower PBLH can reduce the ventilation of chemical species on the surface. Figure 8 shows the impact of aerosol feedback on concentrations of major gaseous and particulate pollutants, including PM_2.5 and ozone at the surface. We averaged the values over the KORUS-AQ period and during the time of GOCI observations. We found that several chemical compounds exhibited non-linear responses to the direct aerosol impact. For one, most pollutants responded differently over the ocean and land, likely the result of the different response of the PBLH, as discussed in the previous section. Because of the lower PBLH, concentrations of gaseous species such as SO₂, NO₂, and NH₃ increased domain-wide over land by 0.34 ppbV (9.39%), 0.23 ppbV (6.75%) and 0.14 ppbV (5.01%), respectively. Not surprisingly, the enhancement was concentrated in East China, with its abundant aerosol loadings. Similarly, concentrations of aerosol species such as $S{O}_4^{2-}$, $N{O}_3^{-}$ and $N{H}_4^{+}$ increased. Average increases over land were 0.34 μg/m³ (6.19%), 0.63 μg/m³ (17.95%), and 0.27 μg/m³ (11.33%). In addition, the concentration of PM_2.5 increased by 1.75 μg/m³ (7.87%). At the same time, however, the concentration of OH decreased by − 7.01 E-6 ppbV (− 4.58%), primarily caused by the reduced amount of solar radiation reaching the ground because of the increased amount of scattered solar radiation due to the stabilization of the atmosphere near the surface, which resulted in the lower temperature. Ozone concentrations increased by 0.14 ppbV domain-wide, but those in East China decreased −0.48 ppbV (−0.89%). Although we assumed that the stabilization of the atmosphere was generally responsible for the increase in the surface ozone concentration, reductions stemming from reduced solar radiation and the enhanced reaction between OH and NO_x rather than VOC/CO in the NO_x-saturated regions. Aerosol feedback on both chemical species and the AOD is listed in Table 4, which shows that most of the chemical species, except for ozone, exhibited greater increases in East China (13.25 – 236.96%). The increased concentration of PM_2.5 was mainly followed by enhanced concentration of sulfate and nitrate. The formation of sulfate ($S{O}_4^{2-}$) begins with the oxidation of SO₂ mostly via OH radicals. Then, sulfate can be formed by the bisulfate dissociation reaction ($\text{HSO}{4}^{-}\leftrightarrow {\text{H}}^{+}+{\text{SO}}_4^{2-}$) and transformation from sulfuric acid (H₂SO₄) into ${\text{SO}}_4^{2-}$ ((NH4)₃H(SO₄)₂, (NH₄)₂SO₄ and NH₄HSO₄) as the presence of NH₃ [Seinfeld and Pandis, 2006]. As a result, high temperatures together with high OH concentrations favor the formation of sulfate [Park et al., 2011; Song et al., 2008]. During the daytime, $N{O}_3^{-}$ (NH₄NO₃) forms with NH₃ and nitric acid (HNO₃), the result of the oxidation of NO₂ with OH radicals [Seinfeld and Pandis, 2006]. Unlike the formation of sulfate, the formation of nitrate is thermodynamically more unstable to stay in particulate phase [Park et al., 2011; Song et al., 2008]. Despite the reduced temperatures and OH concentrations from aerosol feedback, enhanced concentrations occurred in the concentrations of both $N{O}_3^{-}$ and $S{O}_4^{2-}$. Figure 9 shows vertical distributions of sulfate and nitrate concentrations over East China and South Korea (Figure 8, (d) and (e), brown boxes) during study period. The concentrations of the sulfate and nitrate increased dominantly near the surface, and the increase in East China (0.84 μg/m³ and 1.09 μg/m³) was larger than it was in South Korea (0.16 μg/m³ and 0.44 μg/m³) because of higher aerosol loading in East China. We assume that reduced ventilation with aerosol forcing held the chemical species near the source regions, resulting in enlargements of the concentrations near the surface. The increase caused by the enlargements outweighed the decrease by the reduced chemical formation resulting from the reduced OH addressed above; thus, sulfate and nitrate concentrations near the surface increased. The enhancement of the nitrate concentration was much greater than that of the sulfate concentration because of the more favorable formation of nitrate at lower temperatures.

Figure 8.

Spatial distribution of the differences among the chemical species caused by the direct effects of aerosol forcing (YF Assim – NF Assim)

Table 4.

Aerosol feedback on chemical species per unit AOD. All values were carried out on days with a small cloud fraction (< 0.2) and low relative humidity (< 25 %).

Variables	Domain-wide (land)	East China
SO2	1.06 ppbV	3.08 ppbV
NO2	0.46 ppbV	1.55 ppbV
NH3	0.43 ppbV	1.07 ppbV
${\text{SO}}_4^{2-}$	1.05 μg/m3	1.34 μg/m3
${\text{NO}}_3^{-}$	1.92 μg/m3	2.35 μg/m3
${\text{NH}}_4^{+}$	0.83 μg/m3	0.96 μg/m3
OH	−1.61 E-05 ppbV	−2.60 E-05 ppbV
O3	0.43 ppbV	−1.68 ppbV
PM25	6.67 μg/m3	8.33 μg/m3

Figure 9.

Vertical distributions of (a) sulfate and (b) nitrate concentrations over East China and South Korea (refer to the dashed boxes in Figures 8(d) and 8(e)).

To be more concise, we referred to the Integrated Process Rate (IPR), a process analysis (PA) module in the CMAQ, to determine the contribution of each process (HADV: horizontal advection, HDIF: horizontal diffusion, ZADV: vertical advection, VDIF: vertical diffusion, DDEP: dry deposition, AERO: aerosol chemistry, CLDS: the cloud process and aqueous chemistry, EMIS: emissions) in the formation of sulfate and nitrate formation, shown in Figure 10. We conducted the analysis within the PBLH, which is varying over time and simulations, in an area with high aerosol loading (latitude 31.5 ~ 38.5°N, longitude 118 ~ 123°E, East China) during the GOCI time. With regard to sulfate, vertical advection and dry deposition contributed to increasing the concentration during the entire period (Figure 10 (a)). The dry deposition was a major sink of sulfate during the daytime, but the stabilization of the atmosphere delayed the velocity of the deposition, increasing the sulfate concentration by 0.17μg/m³. In addition, vertical advection also contributed to increasing sulfate concentration in PBL about 0.66 μg/m³. The contribution of aerosol chemistry with aerosol forcing decreased (−0.12 μg/m³) because of the lower temperature and OH concentration; the increased sulfate concentration by vertical advection and dry deposition, however, outweighed the other negative contributions. With regard to nitrate, the positive process contributions were vertical advection, aerosol chemistry and cloud process/aqueous chemistry. The contributions of vertical advection and cloud process/aqueous chemistry increased with the aerosol feedback by 0.47μg/m³ and 0.03μg/m³, respectively. Their contributions, however, were not as strong as the contribution of aerosol chemistry (6.43μg/m³), in which lower temperature with aerosol forcing shifted the equilibrium (NH₃(g) + HNO₃(g) ↔ NH₄NO₃(aq)) toward the particulate NH₄NO₃. Figure 10 (c) shows that the concentrations of sulfate and nitrate corresponded to AOD values because high aerosol loading strongly impacts aerosol direct forcing (R = 0.56, 0.48). Hence, the positive process contributions of sulfate and nitrate were consistent with the AOD (Figures 10 (d) and (e)). High aerosol loading was present on May 24, 31 and June 6 – 9; likewise, the dominant positive process contributions of sulfate and nitrate (vertical advection and aerosol chemistry, respectively) exhibited high peaks on the same days.

Figure 10.

IPR results of the AOD and sulfate and nitrate concentrations (a) and (b) averaged contribution of sulfate and nitrate concentrations from various processes, respectively, (c) changes in sulfate nitrate concentrations with an AOD anomaly, (c) the time-series of the positive contribution to sulfate concentrations, and (d) the time-series of the positive contribution to nitrate concentrations

4. Summary and discussion

To evaluate the impact of the direct aerosol effect on meteorology and air quality, we employed the WRF-CMAQ two-way model, which exchanged frequent meteorological input and aerosol feedback during the KORUS-AQ campaign period over East Asia. We also used the optimal interpolation method to incorporate the GOCI AOD so that the model would be able to generate accurate estimates of aerosol concentrations in the atmosphere that are close to those that quantified aerosol feedback. Through validation with AERONET sites located within the GOCI coverage area and a comparison to CALIPSO Lidar, we found that the inclusion of the GOCI AOD assimilation increased the accuracy of model performance by mitigating underpredictions/overpredictions of the current model resulting from high temporal and spatial resolution, a benefit of geostationary data (6 km, 8 times per day).

Both GOCI AOD assimilation and aerosol forcing positively affected model performance (see Table 2). The AOD enhancement was associated with aerosol feedback in both meteorology and chemistry. The impact of the direct aerosol effect was also clearly shown. Regarding meteorology, a significant reduction in the amount of short-wave radiation caused by the absorption and scattering of solar radiation by aerosols resulted in changes in several meteorological variables such as the PBLH, the air temperature, and the wind speed over land. The feedback on the meteorology of the land and the ocean behaved in opposite ways because their thermal inertia differs. The increased PBLH over the ocean carried out the different tendency of aerosol feedback.

In response to direct aerosol impact, the concentrations of surface air pollutants were modified over land not only because of stabilization of the atmosphere with a lower PBLH and less ventilation but also because of the reduced amount of solar radiation. Because of the reduced ventilation, most of the gaseous and particulate species such as SO₂, NO₂, NH₃, $S{O}_4^{2-}$, $N{O}_3^{-}$, $N{H}_4^{+}$, and PM_2.5 increased over land due to the lower temperature and reduced amount of solar radiation; on the other hand, OH concentrations decreased. Aerosol feedback to ozone concentration both increased (domain-wide) and decreased (in highly polluted regions) by reduced ventilation. The results of the integrated process rate showed that a vertical advection process was the dominant contributor to the formation of sulfate, exceeding the reduction by aerosol chemistry (lower temperature and OH concentrations). Even though aerosol chemistry, in conjunction with lower temperature, was the main contributor to nitrate formation, vertical advection was also the dominant contributor to its formation on average. This surface enhancement was observed in the vertical distributions of sulfate and nitrate concentrations. Interestingly, the change in nitrate concentrations near the surface was greater than that in sulfate concentrations mainly the results of its favorable formation at lower temperatures.

One of the limitations of this study is that we adjusted the aerosol concentrations in all troposphere layers of the CMAQ model with a certain ratio that could have missed vertical properties of the aerosol plume that did not exist in the base simulation. Addressing this limitation would require additional information such as Lidar observations. In such a case, we could alleviate the likelihood of excessive overestimation of concentration on the surface, a frequent occurrence of dust and biomass burning events. Nevertheless, one of the advantages of using GOCI data, which have high temporal coverage (8 times per a day), is that they significantly improve model performance, which leads to the more accurate estimation of the aerosol direct forcing effects on meteorology and air quality. In addition, the uncertainty of emission data is not negligible. This problem can be minimized in several ways (i.e., emission inventory adjustment with inverse modeling). In this study, we used a data assimilation technique to improve the accuracy of the model before estimating the impact of direct aerosol feedback.

This study found that aerosols in the atmosphere increase surface aerosol concentrations through changing the meteorology. It also showed that AOD assimilation using the GOCI AOD as geostationary data enhance the performance of the model and increase the reliability of the quantification of aerosol feedback. Thus, the approach used here would be useful in East Asia, with its consistently high levels of aerosols concentrations. Without a concerted effort to reduce aerosol concentrations (e.g., emission control), this situation, under a stable atmosphere, will persist with considerably enhanced feedback. Thus, a rigorous study that examines the effects of aerosol forcing on climate change would be a worthy direction of research.

Supplementary Material

Supplement1

Figure S1. (a) Normalized cumulative semivariogram with respect to the normalized distance using all available GOCI data pixels during the study period; (b) the variance between the AOD and the distance

Figure S2. Comparison of the time series of the AOD and AERONET observations for sites exhibiting the positive/negative impacts of data assimilation.

Figure S3. Comparison of the DC-8 aircraft measurement data and modeled values on May 4, 2016. The gray shadow represents the times when GOCI data were available.

Figure S4. Comparison of the DC-8 aircraft measurement data and modeled values on May 10, 2016. The gray shadow represents the times when the GOCI data were available.

Figure S5. Comparison of the DC-8 aircraft measurement data and modeled values on May 19, 2016. The gray shadow represents the times when the GOCI data were available.

Figure S6. Comparison of the DC-8 aircraft measurement data and modeled values on May 30, 2016. The gray shadow represents the times when the GOCI data were available.

Figure S7. The direct effect of aerosols on meteorology for heat flux at the surface (HFX), moisture flux at the surface (QFX), and the difference between the 2 m air temperature and skin temperature.

Figure S8. Diurnal averaged (a) short-wave radiation, (b) PBL height, (c) 2 m air temperature and (d) 10 m wind speed over the ocean. The black line represents the diurnal variation of NF Assim. The black dotted lines show the direct effect of aerosols (the difference between NF Assim and YF Assim) in the entire domain.