Improving Wet and Dry Deposition of Aerosols in WRF‐Chem: Updates to Below‐Cloud Scavenging and Coarse‐Particle Dry Deposition

Abstract

Wet and dry depositions of aerosols in WRF‐Chem are revisited and updated based on recent observational findings. Traditionally, in‐cloud scavenging was thought to play a more dominant role in aerosol wet removal than below‐cloud scavenging. However, recent field measurements indicated a considerable contribution of below‐cloud scavenging of 50%–60% to total wet deposition. In contrast, the simulated contribution of in‐cloud scavenging in the previous version of WRF‐Chem was too large, exhibiting 88%–95%, likely due to the binary representation of cloud fraction. To reduce the model bias, this study adopts a continuous‐type cloud fraction and implements a semi‐empirical below‐cloud scavenging parameterization. Simulation results with the new scheme show that the contribution of below‐cloud (in‐cloud) scavenging is increased to 63%–66% (decreased to 34%–37%), well capturing the observational estimates. The magnitude of total wet deposition is increased by 18.2% for SO₄, 7.16% for NO₃, and 14.8% for NH₄, showing better agreements with observations particularly for SO₄ and NH₄ deposition. The increased wet removal with the new scheme reduces and so better reproduces surface PM_2.5 and PM₁₀ concentrations, which is also partly attributed to the increased contribution of below‐cloud scavenging. It is found that dry deposition velocity in the previous version was too high for coarse mode particles when friction velocity is large, which underestimates surface PM₁₀ concentration. The updated dry deposition scheme that is constrained by observations effectively improves PM₁₀ performance by reducing the dry deposition velocity for coarse mode particles.

Introduction

Wet and dry deposition are the ultimate removal processes of aerosols in the atmosphere. Wet deposition (or wet scavenging) is referred to two processes: in-cloud scavenging and below-cloud scavenging. Significant uncertainties, however, still remain in the representation of these removal processes in regional- and global-scale modeling (Croft et al., 2010). One of the reasons is that these processes are not well constrained by observations. Several recent observational studies unveiled that some of the previously known aspects of wet and dry deposition are not consistent with the observational findings. For wet deposition, recent sequential sampling measurements indicated that the contribution of below-cloud scavenging to total wet deposition ranges from 50% to 60% in north China (Ge et al., 2021; Xu, Ge, Wang et al., 2017) and can be up to 80% in India (Chatterjee et al., 2010). In previous air quality modeling, however, the importance of below-cloud scavenging has been overlooked and its processes were often simply treated or even ignored (Bae et al., 2012; Barth et al., 2000; Xu, Ge, Chen et al., 2019). Grythe et al. (2017) also shed light on below-cloud scavenging by stating that below-cloud scavenging accounts for more frequent removal of aerosols below 1,000 m and in-cloud scavenging dominates the total removal in the free troposphere.

A few regional modeling efforts have been made to improve below-cloud scavenging processes. Bae et al. (2012) implemented a new below-cloud scavenging scheme in the Community Multiscale Air Quality (CMAQ) model and showed large increases in inorganic aerosol wet deposition (SO₄²⁻, NO₃⁻, and NH₄⁺) by as much as a factor of 2, which leads to better agreements with observed wet deposition over East Asia. Lu and Fung (2018) examined the sensitivity of different below-cloud scavenging schemes to PM_2.5 simulation using the Comprehensive Air Quality Model with Extensions (CAMx) model. They reported a large sensitivity of PM_2.5 concentration by up to 50 μg m⁻³ depending on the schemes and showed that observation-based composition dependent below-cloud scavenging scheme improves PM_2.5 modeling performance. To the authors' best knowledge, however, wet deposition of aerosols particularly focusing on below-cloud scavenging has not yet been assessed for the WRF-Chem model, which is one of the most widely used community atmospheric chemistry transport models. For soluble gases, Bela et al. (2016) examined the wet scavenging and transport of soluble gases during the Deep Convective Clouds and Chemistry (DC3) field campaign using the WRF-Chem model, but they did not examine wet scavenging of aerosols. Yang et al. (2015) performed a case study of aerosol transport and wet scavenging in deep convective clouds by utilizing the DC3 campaign data. They pointed out that the wet scavenging efficiency of aerosols is underestimated and that in-cloud scavenging is the primary wet scavenging mechanism in WRF-Chem. Hence, they implemented secondary activation (activation of aerosols above cloud base) and showed that this increases wet scavenging efficiencies. However, including secondary activation is expected to increase the contribution of in-cloud scavenging, which further deviates from the observational findings. Thus, these previous studies urge that below-cloud scavenging in WRF-Chem should be reassessed and highly likely updated. The findings by Lu and Fung (2018) and Grythe et al. (2017) also suggest that below-cloud scavenging would play a more important role in surface and boundary-layer aerosol concentrations.

From the perspective of dry deposition, it has been reported that the most widely used scheme of Zhang et al. (2001) can produce reasonable dry deposition velocity (V_d) for bulk PM_2.5 by integrating the size-resolved V_d over the whole size distribution of PM_2.5 (Zhang & He, 2014). However, in this model the minimum V_d, which is likely around sizes of 0.1–0.5 μm, shifted a bit to larger particles (0.5–1 μm; Zhang & Vet, 2006), causing an overestimation of V_d for the accumulation and Aitken mode particles by up to one order of magnitude over smooth surfaces. This would not have a large impact on aerosol mass concentration, which is a major concern in air quality studies, but could have a significant impact on aerosol number concentration and thus matter in climate studies. A recent revision of Zhang et al. (2001)'s formula by Emerson et al. (2020) has corrected the location of the minimum V_d by adjusting the collection efficiencies by Brownian diffusion and interception, resulting in a better size-resolved V_d curve when compared to available field flux measurements. The performance of the updated V_d parameterization by Emerson et al. (2020) has yet to be validated in regional scale air quality models. Apart from the recent updates in V_d parameterization, Zeng et al. (2020) examined dry deposition schemes and dust emission schemes in WRF-Chem and reported that dust loading is very sensitive to the deposition schemes, and all dry deposition schemes underestimate dust loading especially in the downwind regions of deserts when the GOCART dust emission scheme is used. They showed that the dust emission scheme based on Shao et al. (2011) and the dry deposition scheme based on Zhang et al. (2001) performs the best for a severe dust storm over East Asia. In another study by Wu et al. (2018), the dry deposition scheme of Zhang et al. (2001) was replaced by Petroff and Zhang (2010) in CAM5, the latter produced much smaller V_d than the former for very small particles, in order to improve model simulation of black carbon concentration. These studies suggest that V_d representation in dry deposition schemes is critical in aerosol prediction and so needs to be well assessed in air quality modeling.

In accordance with the recent observational findings of wet and dry deposition, we revisit the wet and dry deposition schemes used in the Model for Simulating Aerosol Interactions and Chemistry (MOSAIC) aerosol module (Zaveri et al., 2008) in WRF-Chem, propose new schemes, and evaluate the model performance against surface observations of wet deposition and PM_2.5 and PM₁₀ concentrations over East Asia. The previous in-cloud wet scavenging scheme is updated. For below-cloud scavenging, we implement the semi-empirical formula developed by Wang et al. (2014) with a modification to WRF-Chem. For dry deposition, the recently updated scheme by Emerson et al. (2020) is implemented to calculate dry deposition velocity.

Methodology and Observation Data WRF-Chem Modeling

The WRF-Chem modeling domain is shown in Figure 1. The version of the WRF-Chem model used in the study is 4.1.2. The horizontal resolution is 20 km and among the 40 vertical layers in total, 10 layers are placed below 1.5 km above ground level. For meteorology, the initial and boundary conditions are provided by the fifth generation ECMWF atmospheric reanalysis (ERA-5) data at 3-hr intervals. Analysis nudging is applied in all the simulations with a nudging coefficient of 3 × 10⁻⁴ s⁻¹ for wind and temperature and of 5 × 10⁻⁴ s⁻¹ for moisture. The physic options used in this study are given in Table S1 in Supporting Information S1. For gas chemistry and aerosol chemistry/dynamics, MOZART-4 mechanism coupled with MOSAIC (Knote et al., 2014, 2015) is used. The default in-cloud scavenging scheme with MOSAIC (Easter et al., 2004) is updated in the present study to take into account cloud fraction, and the details are given in Section 2.2. For below-cloud scavenging, we do not use the default scheme (Slinn, 1984) but newly implement the semi-empirical scheme proposed by Wang et al. (2014) into WRF-Chem (details in Section 2.2). For subgrid wet scavenging, the default scheme that is based on Grell and Devenyi (2002) is used. Dry deposition of gas species is calculated following Wesely (1989). For aerosol dry deposition, the original scheme by Binkowski and Shankar (1995) (aer_drydep_opt = 1) is updated in the present study following Emerson et al. (2020). The updates are described in Section 2.3. The anthropogenic emissions are based on the Regional Emission inventory in ASia (REAS) version 3.2. The full list of modeling configurations is given in Table S1 in Supporting Information S1, and further details of modeling setups can be found in Ryu et al. (2021). The study period is April 2015 and the meteorology in the simulations are re-initialized every 4 days starting from 15 UTC 28 March 2015. Note that gaseous and aerosol species as well as soil temperature and soil moisture are cycled from the previous runs. To examine the effects of the updated wet and dry deposition separately, four simulations are designed (Table 1). The simulation that uses the new wet and new dry deposition schemes is called the NEW simulation, and the simulation that use the old wet and old dry deposition schemes is the OLD simulation (i.e., the same as the previous version model). In the main article, the results of the NEW and OLD simulations are mostly presented and compared, and the results of the newwet_olddry and oldwet_newdry simulations are shown in Supporting Information S1.

Figure 1. WRF-Chem domain and 16 acid deposition monitoring network in East Asia (EANET) stations used in this study.

Table 1 WRF-Chem Simulations

Simulation name	Wet deposition	Dry deposition
newwet_newdry (NEW)	New	New
newwet_olddry	New	Old
oldwet_newdry	Old	New
oldwet_olddry (OLD)	Old	Old

We additionally conduct two sets of sensitivity experiments with horizontal resolutions of 10 and 40 km for the NEW and OLD simulations (i.e., four experiments in total) to examine the sensitivity of cloud properties, rainfall, and wet deposition to grid sizes. Other than the grid spacing, all the setups and options are the same as described above.

Updated Wet Deposition

The cloud fraction used in the MOSAIC aerosol module in WRF-Chem was a binary-type cloud fraction; that is, it is one if the sum of cloud water mixing ratio and ice mixing ratio is greater than a threshold of 1 × 10⁻³ g kg⁻¹ otherwise it is 0. Such binary-type cloud fraction leads to too large in-cloud scavenging (see Figure S1 in Supporting Information S1). In addition, this is not consistent with the cloud fraction used in other physics schemes such as radiation (e.g., the Rapid Radiative Transfer Model, RRTMG) and photolysis (e.g., updated tropospheric ultraviolet and visible, TUV, model). Hence, a continuous-type cloud fraction ranging from 0 to 1 is adopted in the MOSAIC wet deposition scheme, which is the same as the one used in the RRTMG and updated TUV modules. This cloud fraction parameterization was originally developed by Xu and Randall (1996), which utilizes the relative humidity, temperature, and mixing ratios of hydrometeors (e.g., cloud water, ice, and snow mixing ratio) in a grid box. Note that the cloud fraction used in the activation that is parameterized following Abdul-Razzak and Ghan (2002), is revised to use the continuous-type cloud fraction as well.

Previously, the in-cloud scavenging is parameterized by a first-order loss rate (Equation 1), assuming that the removal rate of cloud-borne aerosols due to collection and conversion is the same as the conversion rate of cloud water droplets to rain droplets. [Image Omitted. See PDF]where F_rainout is the fraction of cloud-borne aerosols scavenged by rain droplets (rainout), r_c (s⁻¹) is the rate of conversion of cloud droplets to rain droplets that is obtained from microphysics parameterization, and Δt (s) is the time step. In general, r_c ranges from 10⁻⁵ to 10⁻³ s⁻¹ and the time step used in the present study is 60 s, so it is acceptable to approximate the removal rate as in Equation 1. No in-cloud scavenging by other hydrometeors (e.g., ice) is considered in the present study. In the new scheme, r_c is divided by cloud fraction (f_cld) to represent the in-cloud conversion rate. Then, the cloud fraction is applied to the rainout fraction (Equation 2). [Image Omitted. See PDF]

The scavenged cloud-borne aerosols are assumed to be all removed from the model domain; that is, all rainout aerosols are assumed to fall to the surface. So, no evaporation of scavenged aerosols by rainout is considered during precipitating and this needs to be updated in a future study. However, the resuspension of cloud-borne aerosols is considered when clouds dissipate (when cloud fraction decreases over time). The product of F_rainout and concentration of cloud-borne aerosols is vertically integrated, and this is assigned as wet deposition by in-cloud scavenging (Q_in-cloud). [Image Omitted. See PDF]where ρ (kg m⁻³) is the air density, C_i (μg kg⁻¹) is the mass mixing ratio of cloud-borne aerosols at ith bin, and z_top is the top height of model domain.

In the previous version of WRF-Chem, below-cloud scavenging by Brownian diffusion, interception and impaction was computed based on Slinn (1984). However, it turns out that the below-cloud scavenging in the original method is too small (see Figures S1 in Supporting Information S1 and 4). Therefore, below-cloud scavenging is parameterized based on the semi-empirical formula by Wang et al. (2014). The formula computes size-resolved below-cloud scavenging coefficients (Λ_rain for rain and Λ_snow for snow) at a given rain/snow rate. It is found that the scavenging coefficients obtained by the original formula of Wang et al. (2014) are too small; for example, Λ_rain for accumulation mode aerosols ranges from 10⁻⁷ to 10⁻⁶ s⁻¹ at a rain rate of 1 mm hr⁻¹. On the other hand, the field measurements by Xu, Ge, Chen et al. (2019) show a range of 10⁻⁵–10⁻⁴ s⁻¹ at the same rain rate. It is well know that theoretical estimates of below-cloud scavenging coefficients are lower by 1–2 order of magnitudes than measured ones (Xu, Ge, Chen et al., 2019). Wang et al. (2011) investigated the impact of vertical turbulence on below-cloud scavenging coefficients and highlighted that a consideration of vertical turbulence can explain almost all discrepancies between theoretical and field measured scavenging coefficients for particles in 0.01–3 μm diameter ranges. As we do not consider vertical turbulence in the present study, a factor (κ) is simply considered and κ is set to 200 by performing some sensitivity simulations. [Image Omitted. See PDF]where A and B are the empirical coefficients that are a function of aerosol diameter, and P (mm h⁻¹) is the precipitation rate. The coefficients A and B are different for rain and snow, and the formulas for A and B can be found in Wang et al. (2014). Note that the same formulas by Wang et al. (2014) were implemented in WRF-Chem by Ryu et al. (2021), but a different κ (500) was used and below-cloud scavenging by snow and in-cloud scavenging were not considered in their study. We do not consider chemical-composition dependent Λ_rain and Λ_snow, and this will deserve a future research. The fraction of washout is similarly represented by the first-order loss rate, so the wet deposition by below-cloud scavenging (Q_below-cloud) is computed as follows. [Image Omitted. See PDF]

Here, C_i is the mass mixing ratio of interstitial aerosols at ith bin. The Q_in-cloud and Q_below-cloud represent the wet scavenging by resolved clouds and precipitation, respectively, in the present study. The wet scavenging of precursor gases (SO₂ for SO₄²⁻, HNO₃ for NO₃⁻, and NH₃ for NH₄⁺) and wet scavenging of gases and aerosols by convective clouds/rain are included in the wet deposition computation, and the sum of all scavenging is called the total wet deposition. It is assumed that all of the precursor gases in rainwater are present as the soluble ions in the observations (Appel et al., 2011). The wet deposition of SO_X (SO₂ and particulate SO₄²⁻), NO_X (HNO₃ and particulate NO₃⁻), and NH_X (NH₃ and particulate NH₄⁺) are simply referred to as SO₄, NO₃, and NH₄ wet deposition, respectively in this study. It should be noted that our goal is to update and improve grid-scale wet deposition and that the wet scavenging of gases and aerosols by convective clouds/rain are not therefore updated in the present study.

Updated Dry Deposition

The V_d is calculated using the updated scheme of Emerson et al. (2020) as follows. [Image Omitted. See PDF]where V_g is the gravitational settling velocity, R_a is the aerodynamic resistance, and R_s is the surface resistance. The default MOSAIC dry deposition module follows the method developed by Binkowski and Shankar (1995). In their study, however, the Cunningham correction factor (C_c) was not considered in the gravitational settling velocity. So, V_g is revised to include C_c. [Image Omitted. See PDF]where g is the gravitational acceleration, ν is the kinematic viscosity of air, ρ_p and ρ_air is the density of particle and air, respectively, and d_p is the particle diameter. The C_c formula given in Tsuda et al. (2013) is adopted. The surface resistance R_s is computed as [Image Omitted. See PDF]

E_B, E_Im, and E_In is the collection efficiency by Brownian diffusion, impaction and interception, respectively, u_* is the friction velocity, and ε₀ is an empirical coefficient. In the previous version, the influence of interception was not considered. Each collection efficiency is computed as follows. [Image Omitted. See PDF] [Image Omitted. See PDF] [Image Omitted. See PDF]where Sc is the Schmidt number, St is the Stokes number, the coefficients C_b, C_Im, C_In, H, α, β, and γ are empirical coefficients. Note that H is set to be land-use dependent as E_In depends on surface properties (Emerson et al., 2020). In the present study, the empirical coefficient ε₀ is also set to be dependent on land-use type. Three types of land use are simply considered in the present study, which are forest, grass, and water. Twenty Moderate Resolution Imaging Spectroradiometer (MODIS) land-use/land-cover (LULC) categories that are used in the present study are classified into the three types: forest and urban MODIS LULCs are set to forest type, water MODIS LULC is set to water type, and the other remaining LULCs are set to grass type. It should be noted that not all coefficients are given in Emerson et al. (2020), and so some of them (H_forest, H_grass, H_water, ε_0,forest, ε_0,grass, and ε_0,water) are determined to follow the dry deposition velocities shown in Emerson et al. (2020). Table 2 provides the coefficients used in the present study. As Emerson et al. (2020) demonstrated, the particle size exhibiting minimum V_d values is found around 0.1 μm in the new scheme whereas it is around 1 μm in the old scheme (Figure 2). It should be noted that we set the dry deposition velocity for the third bin in the MOSAIC to be similar to that in the previous version. As a result, the revised dry deposition velocities for the first and second bins are lower by a factor of 1–2 than the old ones. However, the differences in V_d between the new and old schemes result in only small differences in PM_2.5 concentration (i.e., 0.7 μg m⁻³ for monthly mean PM_2.5 averaged over observation stations, see Table 3). This is because the V_d for the first and second bins are already small in the old scheme. Once comprehensive direct measurements of dry deposition velocities and dry deposition fluxes are available, some coefficients might need to be modified/optimized in the future.

Table 2 Coefficients Used in the Dry Deposition Velocity Computation for Equations 8–11

Cb	CIm	CIn	α	β	γ	Hforest	Hgrass	Hwater	ε0,forest	ε0,grass	ε0,water
2	0.4	2.5	0.8	1.7	0.8	0.1	0.2	100	0.008	0.004	0.006

Table 3 Performance Statistics for PM_2.5 and PM₁₀ in April 2015

	PM2.5 (n station = 555)				PM10 (n station = 531)
	NEW	newwet_olddry	oldwet_newdry	OLD	NEW	newwet_olddry	oldwet_newdry	OLD
mean	48.5	47.8	53.6	51.5	80.9	62.3	82.8	64.9
MB	1.55	0.817	6.66	4.50	−3.66	−22.2	−1.77	−19.7
RMSE	24.2	24.1	26.8	26.3	44.0	44.1	44.1	44.0
NMB	4.08	2.59	15.1	10.3	−3.63	−25.2	−1.43	−22.4
NME	41.3	41.2	46.8	45.9	38.8	40.2	39.2	40.5
r	0.583	0.581	0.560	0.555	0.545	0.524	0.537	0.506

Note. The MB, RMSE, NMB, NME, and r indicate the mean bias, root-mean-square-error, normalized mean bias, normalized mean error, and correlation coefficient, respectively. The values are averaged over stations. The unit of mean, MB, and RMSE is μg m⁻³, and that of NMB and NME is %. r is unitless.

In addition, V_d for activated aerosols is reduced from 10 cm s⁻¹ to 0.22 cm s⁻¹. It turned out that the previous value of 10 cm s⁻¹ yields too large removal of activated (cloud-borne) aerosols particularly over water. Tav et al. (2018) measured fog droplet deposition velocities and showed that V_d for fog droplet over bare soil is less than 2.2 cm s⁻¹ when wind speed is below 4 m s⁻¹. As our concern is water surface, V_d is reduced by a factor of 10 from the measured V_d over bare soil. This might seem to be a rough estimate, but it is similar to V_d for 5 μm-size particle obtained using the revised dry deposition scheme (e.g., 0.246 cm s⁻¹ see the below). The results with the updated V_d for activated aerosols show increases in PM_2.5 concentration over ocean, along coastlines and over southwestern China inland where clouds are frequently present (see Figures 5c and S4 in Supporting Information S1), showing reasonably good agreements with observed PM_2.5.

Two examples of V_d with different friction velocity (u_*) in Figure 2 show that the differences in V_d for coarse mode particles between the two schemes are large when u_* is large. When u_* is 0.2 m s⁻¹, V_d for particles larger than ∼1 μm in the new scheme is very similar to that in the old scheme. On the other hand, when u_* is 0.8 m s⁻¹, V_d for a large particle is much larger in the old scheme by an order of magnitude (e.g., 7.2 cm s⁻¹ for a particle with dry diameter of 5 μm) than in the new scheme (0.246 cm s⁻¹ for the same size particle over grass type surface). This large V_d for coarse mode particles in the old scheme leads to too large dry deposition and eventually low concentration of PM₁₀ (see Section 3.4 for details).

Figure 2. Dry deposition velocity (Vd) with particle diameter when the friction velocity (u*) is (a) 0.2 m s−1 and (b) 0.8 m s−1. The dashed-gray line indicates Vd in the old version. Open (filled) markers, corresponding to the particle sizes of the model for simulating aerosol interactions and chemistry (MOSAIC) 4 bins (geometric mean diameters of dry particles), represent the results with old (new) schemes.

Observation Data Wet Deposition Observations

The Acid Deposition Monitoring Network in East Asia (EANET) data (https://www.eanet.asia/, last access: 9 November 2020) are utilized to evaluate the model performance. Some of the EANET stations are not used in this study because some of them reveal large discrepancies in daily rainfall occurrences between the observations and simulations during the study period. As such, 16 EANET stations are used in this study and these are marked in Figure 1. The daily measurements of non-sea-salt SO₄ (hereafter, simply SO₄), NO₃, NH₄ wet deposition and rainfall at the EANET stations are compared to the simulated ones.

Surface PM_2.5 and PM₁₀ Observations

The surface observations of PM_2.5 and PM₁₀ over South Korea are obtained from https://www.airkorea.or.kr/ (last access: 31 January 2020). For China, surface PM_2.5 and PM₁₀ observation data are downloaded from https://quotsoft.net/air/ (last access: 26 July 2021). For all surface observations, the stations in which at least 80% of data are available during the study period are selected. When multiple stations belong to a single WRF-Chem grid box, because some of them are located quite close to each other, the data within ±30% of the median value are selected and then averaged. The number of stations used in the model evaluation is 555 for PM_2.5 and 531 for PM₁₀. Several statistic metrics evaluating the model performance are given in Table 3, and the metrics are chosen following the recommendations by Emery et al. (2017).

Results and Discussion Wet Deposition

The spatial distributions of wet deposition and rainfall in April 2015 in the NEW and OLD simulations are compared in Figure 3. The spatial distributions of wet deposition as well as rainfall agree well with the observations at the EANET stations. Although the spatial distributions of wet deposition between the NEW and OLD simulations are found to be similar to each other, the magnitude of wet deposition is overall increased in the NEW simulation (Figures 3i–3k). Note that the monthly rainfall in the NEW and OLD simulations is generally very similar and so the difference is small (Figure 3l). Thus, the differences in wet deposition between the two simulations can be interpreted as due to the difference in wet deposition schemes. The monthly SO₄, NO₃, and NH₄ wet deposition averaged over the EANET stations are increased by 18.2%, 7.16%, and 14.8%, respectively, resulting in better agreements with the observations in terms of mean values for SO₄ and NH₄ (Figure 4); for the domain-wide averages over land, they are increased by 23.1%, 8.90%, and 21.0%, respectively. The smaller increase in NO₃ wet deposition than those in the others is due to the significant contribution of HNO₃ wet deposition to the total NO₃ wet deposition (Figure 4b). The significant contribution of dissolved HNO₃–NO₃ wet deposition is consistent with the result of Bae et al. (2012). The NO₃ wet deposition is slightly overestimated on average both in the NEW and OLD simulations, and currently it is not clear whether it is due to the overestimated wet deposition of HNO₃ or particulate NO₃.

Figure 3. (Top, a–d) Monthly wet deposition of sulphate (SO4), nitrate (NO3), ammonium (NH4), and rainfall in the OLD simulation, respectively. (Middle, e–h) Same as (top) but in the NEW simulation. (Bottom, i–l) Difference between the NEW and OLD simulations.

Figure 4. Monthly wet deposition of (a) SO4, (b) NO3, and (c) NH4 and (d) monthly rainfall at the EANET observation stations in the OLD (vertical bars with small dots) and NEW (vertical bars with slant lines) simulations. The observed deposition and rainfall are denoted by red circles. Wet deposition by in-cloud scavenging is represented by yellow, below-cloud scavenging by light blue, dissolved precursor gases by resolved clouds/precipitation by pink, wet deposition of aerosols by convective clouds/precipitation by green, and dissolved precursor gases by convective clouds/precipitation by navy. The gray line at the center of each vertical bar indicates the standard deviation of wet deposition/rainfall at nine model grids centered at an observation station.

The most noticeable result is the increased (decreased) contribution of below-cloud (in-cloud) scavenging in the NEW simulation as compared to that in the OLD simulation (Figures 4, S1, and S2 in Supporting Information S1). In the old scheme, only small amount of aerosols are removed by below-cloud scavenging and majority of the aerosols are removed by in-cloud scavenging. The fraction of Q_below-cloud to Q_below-cloud + Q_in-cloud in the OLD simulation is 5.4% for SO₄, 11.8% for NO₃, and 4.9% for NH₄ wet deposition on average at the EANET stations. On the other hand, the fraction, Q_below-cloud/(Q_below-cloud + Q_in-cloud), is increased to 63.5% for SO₄, 66.2% for NO₃, and 64.1% for NH₄ wet deposition in the NEW simulation. For domain-wide averages over land, the fraction, Q_below-cloud/(Q_below-cloud + Q_in-cloud), is increased to 69.0% from 5.15% for SO₄ and 66.6% from 7.67% for NO₃ and 66.6% from 6.87% from NH₄ wet deposition when comparing the NEW and OLD simulations (compare Figures S1, S2 in Supporting Information S1). The fraction of below-cloud scavenging obtained in the present study are comparable to the observational findings of Xu, Ge, Chen et al. (2019) and Ge et al. (2021), namely by 50%–60%, whereas that by the old wet deposition scheme is apparently underestimated.

As the total wet deposition includes the dissolved gases and removal by convective clouds/rain in the present study, the contribution of below-cloud scavenging relative to the total wet deposition is 58.5% (4.97%) for SO₄, 32.6% (6.19%) for NO₃, and 51.6% (3.89%) for NH₄ wet deposition in the NEW (OLD) simulation. The average fractions of below-cloud scavenging relative to the total SO₄ and NH₄ wet deposition obtained using our revised scheme are indeed in line with those found in the observations. As mentioned above, the low contribution of below-cloud scavenging to the total NO₃ wet deposition is due to the contribution of dissolved HNO₃ (48.8% on average at the EANET stations). For an observational study, therefore separating a fraction of dissolved HNO₃ from total NO₃ wet deposition might be necessary. It is found that HNO₃ wet deposition tends to be increased but particulate NO₃ wet deposition decreased in the NEW simulation as compared to those in the OLD simulation (not shown). These are presumably due to the semi-volatile characteristic of NH₄NO₃ aerosol and high solubility of HNO₃, but a further investigation is required. Note that the new wet deposition scheme barely influences SO₂ wet deposition and this is also interpreted as due to the nonvolatile SO₄ aerosol and relatively low SO₂ solubility. It is noteworthy that the contribution of convective clouds/rain (subgrid scale wet scavenging) to the total wet scavenging is generally small in the present study (Figure S3 in Supporting Information S1). For example, the fraction of the subgrid scale wet scavenging to the total wet scavenging is 9.5% for SO₄, 3.3% for NO₃ and 7.9% for NH₄ in terms of domain-averaged monthly values.

Although the new wet deposition scheme generally increases the magnitudes of wet deposition, it is reduced in some regions with the new wet deposition scheme. Such decreases in wet deposition with the new wet deposition scheme are mostly found over southwestern China (Figures 3i–3k). This is interpreted as the smaller role of below-cloud scavenging relative to in-cloud scavenging over this region (Figures S2g, S2h, and S2i in Supporting Information S1). The fraction of below-cloud scavenging, Q_below-cloud/(Q_below-cloud + Q_in-cloud), ranges from ∼0.1 to 0.4 over southwestern China, which is lower than the domain-wide averages over land. This region is characterized by small amount of rainfall relative to liquid water path (Figure S4 in Supporting Information S1). The ratio of rainfall to liquid water path over this region is generally smaller than 5% (Figure S4c in Supporting Information S1). The smaller ratio indicates the smaller potential role of below-cloud scavenging; in other words, the larger ratio of liquid water path indicates the larger potential role of in-cloud scavenging. Because the wet deposition by in-cloud scavenging was largely overestimated in the old wet deposition scheme, the difference in wet deposition between the NEW and OLD simulations (NEW minus OLD) can be negative over southwestern China where the contribution of in-cloud scavenging is potentially large.

It is found that the difference in wet deposition between the NEW and OLD simulations is mostly attributed to the difference in wet deposition scheme (middle row in Figure S5 in Supporting Information S1). In other words, the new dry deposition scheme has minimal influences on the difference in wet deposition between the two simulations (bottom row in Figure S5 in Supporting Information S1).

Influence of New Wet Deposition Scheme on Surface PM_2.5 and PM₁₀

The revised wet deposition scheme considerably influences the surface PM_2.5 concentration (Figures 5 and 6). In general, PM_2.5 concentration is lower in the NEW simulation than in the OLD simulation, exhibiting better agreements with surface PM_2.5 observations (Figures 5, 6 and Table 3). The metrics (NMB, NME, and r) in the NEW simulation show all improvements as compared to those in the OLD simulation, and all of them fall into the criteria recommended by Emery et al. (2017). In particular, it is remarkable that NMB is reduced from 10.3% in the OLD simulation to 4.08% in the NEW simulation and this satisfies the performance goal for daily PM_2.5 (±10%) recommended by Emery et al. (2017). Even though the lower dry deposition velocities and correspondingly higher aerosol concentrations are obtained with the new dry deposition scheme, the lower PM_2.5 concentrations found in the NEW simulation are interpreted as a larger role of wet deposition in decreasing PM_2.5 concentration. For example, with the same new dry deposition scheme, the mean PM_2.5 concentration is lower by 5.1 μg m⁻³ in the NEW simulation than in the oldwet_newdry simulation (Table 3). On the other hand, the influence of the new dry deposition scheme on PM_2.5 is smaller than that of new wet deposition scheme: the difference in mean PM_2.5 between NEW and newwet_olddry simulations is 0.7 μg m⁻³.

Figure 5. Monthly-mean surface PM2.5 in the (a) OLD and (b) NEW simulations. (c) Difference in PM2.5 between the NEW and OLD simulations. Observed PM2.5 are marked with circles with the same colors. (d–f) are the same as (a–c) but for PM10.

Figure 6. Probability density function (PDF) of daily-mean (a) PM2.5 in the observations and in the four simulations. The numbers in the parenthesis are the monthly means over observation stations. (b) Same as in (a) but for PM10.

The probability density functions (PDFs) of PM_2.5 are better captured when the new wet deposition scheme is implemented (newwet_olddry and NEW simulations) than those when the old wet deposition scheme is used (Figure 6a). The overestimated occurrence frequency for high levels of PM_2.5 (higher than 80 μg m⁻³) is reduced, and at the same time the underestimated occurrence frequency of medium levels of PM_2.5 (25–50 μg m⁻³) is enhanced. Gong et al. (2011) tested two different below-cloud scavenging schemes, one of which computes upper bounds of below-cloud scavenging coefficients and the other computes lower bounds. They showed that surface PM_2.5 concentration can be different by up to 10% due to the different schemes. Although both in-cloud and below-cloud scavenging treatments are updated in the present study, a similar degree of difference in terms of mean PM_2.5 concentration is found in our study (9.5%) due to the different wet deposition schemes (Table 3). It is argued that the enhanced contribution of below-cloud scavenging with the new wet deposition scheme more effectively decreases the surface PM_2.5 concentration. When cloud base heights are above boundary layer height and the clouds are precipitating clouds, for example, the in-cloud scavenging removes aerosols mostly above boundary layer height whereas the below-cloud scavenging removes aerosols within the boundary layer. In addition, in-cloud scavenging does not take into account the removal by precipitation originating from melting snow and/or ice. So, the under-representation of below-cloud scavenging in the old wet deposition scheme likely results in high aerosol concentrations near the surface, which is consistent with the previous studies of Grythe et al. (2017) and Lu and Fung (2018).

The larger wet removal by the new wet deposition scheme also reduces the surface PM₁₀ concentration. The mean PM₁₀ concentration is reduced by 1.9 μg m⁻³ due to the enhanced wet deposition in the NEW simulation as compared to that in the oldwet_newdry simulation that uses the same new dry deposition scheme (Table 3). This increases the negative bias errors measured by MB and NMB. However, the lower RMSE and NME and the higher correlation coefficient in the NEW simulation indicate a better performance with the new wet deposition scheme for PM₁₀.

Sensitivity to Horizontal Grid Spacing

It is found that the monthly mean of cloud-coverage fraction increases as the horizontal grid size increases both for the NEW and OLD simulations (Table S2 in Supporting Information S1) as one can expect. The rate of increase in cloud-coverage fraction from 10-km grid spacing to 40-km grid spacing is about 7% (e.g., from 0.202 with 10-km grid spacing to 0.2163 with 40-km grid spacing for the OLD simulations). On the other hand, the area-averaged monthly rainfall decreases as the horizontal grid size increases. The reduction in rainfall amount with increasing horizontal grid spacing is commonly reported in many previous studies (e.g., Chen & Dai, 2019; Rauscher et al., 2016). The area-averaged below-cloud scavenging ratio to total scavenging slightly increases with horizontal grid spacing both for the NEW and OLD simulations (the right-most columns in Table S2 in Supporting Information S1). This is mostly due to the decreases in in-cloud scavenging amount (Q_in-cloud) with increasing grid spacing. Even though cloud-coverage fraction increases with grid spacing, the area-averaged r_c considerably decreases as the horizontal resolution becomes coarser (i.e., from 3.57 × 10⁻⁴ s⁻¹ with 10-km grid spacing to 2.83 × 10⁻⁴ s⁻¹ with 40-km grid spacing). The decrease in r_c is interpreted as a result of reduced vertical mass flux and convergence with increasing grid spacing (Rauscher et al., 2016), analogous to decrease in rainfall. Thus, it can be concluded that the model's representation of in-cloud scavenging amount decreases as horizontal resolution becomes coarser due to the reduced conversion rates of cloud droplets to rain droplets with increasing grid spacing, and this effect is larger than the increase in cloud-coverage fraction. Nevertheless, the fraction of below-cloud scavenging to total scavenging marginally varies depending on the horizontal grid size (64%–67% for the NEW simulations). In addition, the contribution of grid-scale wet deposition by resolved clouds and precipitation increases as the horizontal grid size decreases (compare Figures S3 and S6 in Supporting Information S1). On the other hand, the contribution of subgrid scale wet deposition decreases as the horizontal resolution becomes higher, which is consistent with the decreased role of parameterized convective transport of trace gases with increasing horizontal resolution (Li et al., 2018). These sensitivity simulation results support that our conclusions are robust across horizontal resolutions.

Dry Deposition

As seen in Figure 2, the largest difference in V_d between the new and old dry deposition scheme is found for coarse mode particles in the daytime during which the friction velocity is large and the aerodynamic resistance is small. The monthly dry deposition fluxes for the fourth bin (aerosols with diameters ranging 2.5–10 μm) clearly show the large differences between the old and new dry deposition schemes (Figure 7). Note that the NEW and newwet_olddry simulations are compared here in order to exclude the influences of different wet deposition schemes. The differences in dry deposition fluxes for smaller aerosols (diameters less than 2.5 μm) are relatively small as compared to those for the coarse aerosols (see Figure S7 for the comparison in Supporting Information S1). The most distinctive differences are found for other inorganic (OIN) aerosols. In the MOSAIC, the aerosols that are emitted from deserts are assigned as OIN aerosols. The region exhibiting very large dry deposition for OIN aerosols is a part of the Gobi Desert. The smaller V_d for coarse mode particles in the new scheme tremendously reduces dry deposition fluxes, and it is reduced by 76.5% for OIN aerosols when averaged over the desert area (2.3 g m⁻² month⁻¹ with the new scheme and 9.8 g m⁻² month⁻¹ with the old scheme). The NO₃ dry deposition also shows noticeable differences between the two schemes. In the old scheme, the dry deposition for coarse mode NO₃ aerosols is considerably high along coastlines. The large dry deposition along coastline is due to the high concentration of coarse mode NO₃ aerosols that originate from cloud-borne NO₃ aerosols there (not shown). Despite nonnegligible reductions in dry deposition are found for NO₃ (6.95 mg m⁻² month⁻¹ for domain-wide average over land) as well as SO₄ (3.08 mg m⁻² month⁻¹) and NH₄ (1.13 mg m⁻² month⁻¹) aerosols, the reduction amounts are largest for OIN aerosols (802.9 mg m⁻² month⁻¹).

Figure 7. (Top, a–d) Monthly dry deposition of sulphate (SO4), nitrate (NO3), ammonium (NH4), and other inorganic (OIN) aerosols with diameters of 2.5–10 μm (the fourth bin) in the newwet_olddry simulation, respectively. (Middle, e–h) Same as (top) but in the NEW simulation. (Bottom, i–l) Difference between the NEW and newwet_olddry simulations. The legend values for the differential OIN (l) are −5 × 104, −1 × 104, −5 × 103, −1 × 103, −5 × 102, −1 × 102, −5 × 101, −1 × 101, −5, −1, 0, 1, 5, 1 × 101, 5 × 101, 1 × 102, 5 × 102, 1 × 103, 5 × 103, 1 × 104, and 5 × 104 mg m−2 month−1.

The smaller V_d with the new dry deposition scheme leads to increasing PM₁₀ concentration greatly (Figure 5), which agrees better with surface PM₁₀ observations (Table 3; Figures 5 and 6). When the old dry deposition scheme is used with the same new wet deposition scheme, PM₁₀ concentration is largely underestimated by 22.2 μg m⁻³ in the newwet_olddry simulation. On the other hand, MB is significantly reduced to −3.66 μg m⁻³ when the new dry deposition is used in the NEW simulation. The largest differences in PM₁₀ concentration are found over the Gobi desert and nearby areas, indicating that these differences predominantly originate from dust aerosols. The spatial patterns of differential PM₁₀ concentration (Figure 5f) are consistent with those of differential dry deposition fluxes for OIN aerosols (Figure 7l). The PDFs of PM₁₀ are also much better captured in the simulations that use the new dry deposition scheme than those using the old dry deposition scheme (Figure 6b). Our result is consistent with the result of Zeng et al. (2020). They showed that the original MOSAIC dry deposition scheme based on Binkowski and Shankar (1995) greatly underestimates PM₁₀ concentration by 146 μg m⁻³ for a severe dust storm over East Asia (the observed mean PM₁₀ was 172.7 μg m⁻³ but simulated one was only 26.45 μg m⁻³). As they demonstrated that the dust emission scheme by Shao et al. (2011) performs better and produces larger dust loading than that by the GOCART scheme (which is used in the present study), a future study that uses different dust emission schemes would be required. A better dust emission scheme is expected to resolve the negative bias error still found with the new dry deposition scheme. In short, it can be concluded that V_d for coarse mode particles in the previous version was too large especially in the daytime and that the new dry deposition scheme effectively resolves the large negative bias of PM₁₀ that was found in the previous version.

It is noteworthy that the updated dry deposition scheme plays a much larger role in PM₁₀ than the updated wet deposition scheme in the present study (also see Figure S8 in Supporting Information S1). With the same new (old) dry deposition scheme, the difference in PM₁₀ concentration between the new and old wet deposition scheme is 1.9 μg m⁻³ (2.6 μg m⁻³) that is computed as the oldwet_newdry minus NEW simulations (OLD minus newwet_olddry simulations), which is much smaller than that due to the difference in dry deposition scheme, that is, ∼18 μg m⁻³.

Summary and Conclusions

The updated wet and dry deposition schemes improve the model performance in reproducing soluble inorganic ion wet deposition and surface PM_2.5 and PM₁₀ concentrations over East Asia. The updated wet deposition scheme incorporating semi-empirical below-cloud scavenging parameterization increases the total wet deposition by 18.2% for SO₄, by 7.16% for NO₃, and 14.8% for NH₄ on average at the EANET observation stations as compared to the previous wet deposition scheme. In the previous scheme, the contribution of in-cloud scavenging was too large likely due to the usage of binary-type cloud fraction (0 or 1), whereas that of below-cloud scavenging was too small. For SO₄ wet deposition, for example, the contribution of below-cloud scavenging is greatly increased to 63.5% with the new scheme from 5.4% with the old scheme. The increased contribution of below-cloud scavenging is in line with the recent observational evidence. As a result of the increase in total wet deposition together with the increased contribution of below-cloud scavenging, the simulated surface PM_2.5 concentration decreases overall and shows a better agreement with the observed one.

The previous dry deposition scheme in WRF-Chem is found to produce too large dry deposition velocities for coarse-mode particles by an order of magnitude when friction velocity is large, resulting in large underestimations of surface PM₁₀. Such difference in dry deposition velocity between the new and old scheme leads to large differences in surface PM₁₀ but small difference in surface PM_2.5. The updated scheme greatly increases surface PM₁₀ concentrations by ∼18 μg m⁻³, and so reproduces observed PM₁₀ much better. It is noteworthy that model performance improvements, when compared to ground measurements, could include compensation errors between schemes chosen for various chemical and physical processes in the mass continuity equation, and so closer agreements with observations do not always imply that these are solely due to the updated schemes. Thus, more systematic and extensive model evaluations would be required in the future.

In addition to air quality studies, the updated WRF-Chem model is expected to be utilized in eco-environmental studies because wet deposition from the atmosphere significantly affect soil acidification, eutrophication, and net primary productivity. Although our updates show satisfactory improvements overall, there are still several features that need to be further developed in future studies, such as in-cloud scavenging by ice-phase hydrometeors, partial/total resuspension of falling cloud-borne aerosols before reaching the surface, chemical-composition dependent scavenging coefficients, and ice retention of soluble gases. For example, wet scavenging by ice-phase hydrometeors can play a key role in wintertime and/or in high latitude regions. In this context, our model results are expected to vary widely if the updates are applied to different regions (e.g., desert, tropical, and artic regions) or different seasons. In tropical regions or the regions where deep convection prevails, subgrid scale wet scavenging by convection would play a more important role. Hence, extensive model evaluations over different regions in the globe and in different seasons are worthy and deserve future studies. For resuspension of aerosols, a global model study demonstrated that aerosol resuspension produces modest reductions in aerosol mass and number concentration in a global mean sense because the re-suspended aerosols belong to coarse-mode particles that have short lifetime (H. Wang et al., 2020). It is, however, expected that resuspended aerosols can increase aerosol mass concentration at local and short-term time scales, thus calling attention to a future relevant study. Retention of trace gases on ice is not considered in the present study. As it can alter gas concentrations and their vertical distribution, wet deposition of gases can be influenced to some extent, which requires a further study.

Acknowledgments

We thank three anonymous reviewers for providing their helpful and valuable comments and suggestions. This study is supported by the Korea Meteorological Administration Research and Development Program under Grant KMI2020-01413.

Data Availability Statement

The Acid Deposition Monitoring Network in East Asia (EANET) observations are available at https://www.eanet.asia/. The surface PM_2.5 and PM₁₀ data over South Korea are available at https://www.airkorea.or.kr/, and those over China are at https://quotsoft.net/air/. The updated WRF-Chem codes are available at https://doi.org/10.5281/zenodo.5895233.