Modeling Dust in East Asia by CESM and Sources of Biases

Abstract

East Asian dust has a significant impact on regional and global climate. In this study, we evaluate the spatial distributions and temporal variations of dust extinction profiles and dust optical depth (DOD) over East Asia simulated from the Community Earth System Model (CESM) with satellite retrievals from Luo et al. (2015a, 2015b) (L15), Yu et al. (2015) (Y15), and Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) level 3 (CAL-L3) products. Both L15 and Y15 are based on CALIPSO products but use different algorithms to separate dust from non-dust aerosols. We find high model biases of dust extinction in the upper troposphere over the Taklamakan Desert, Gobi Desert, and Tibetan Plateau, especially in the summer (June-July-August, JJA). CESM with dust emission scheme of Kok et al. (2014a, 2014b) has the best agreement with dust extinction profiles and DOD from L15 in the Taklamakan Desert and Tibetan Plateau. CESM with the default dust emission scheme of Zender et al. (2003a) underpredicts DOD in the Tibetan Plateau compared with observations from L15 due to the underestimation of local dust emission. Large uncertainties exist in observations from L15, Y15, and CAL-L3 and have significant impacts on the model evaluation of dust spatial distributions. We also assess dust surface concentrations and 10 m wind speed with meteorological records from weather stations in the Taklamakan and Gobi Deserts during dust events. CESM underestimates dust surface concentrations at most weather stations due to the inability of CESM to capture strong surface wind events.

1. Introduction

Mineral dust plays many important roles in Earth’s climate system. It can directly scatter and absorb solar and terrestrial radiation (e.g., Balkanski et al., 2007; Tegen et al., 1996). It can also change cloud properties by acting as cloud condensation nuclei (CCN) and ice nucleating particles (INPs) (e.g., DeMott et al., 2003; Rosenfeld et al., 2001). Dust aerosol impacts atmospheric chemistry by providing surface area for heterogeneous reactions and deposition of trace gases (e.g., Dentener et al., 1996). Dust also affects the biogeochemistry cycle by fertilizing land and ocean (e.g., Jickells et al., 2005). Although global climate models (GCMs) represent the main processes of dust cycle, large uncertainties exist in the simulated dust cycle (e.g., Boucher et al., 2013; Huneeus et al., 2011; Kim et al., 2014; Wu et al., 2018). It is still challenging to quantify the radiative forcing (RF) of dust due to aerosol-radiation (RFari) and aerosol-cloud interactions (RFaci) (Boucher et al., 2013).

East Asia is one of the main sources of atmospheric mineral dust, which accounts for ~20% of the global dust emission (Nagashima et al., 2016). Dust aerosol has a significant impact on regional climate of East Asia (Gu et al., 2016; Guo & Yin, 2015; Lou et al., 2017; Sun et al., 2012, 2017; Zhang et al., 2009); dust induced cooling can weaken the East Asia summer monsoon (EASM) and suppress precipitation (Guo & Yin, 2015; Sun et al., 2012, 2017). It is estimated that 26% of the East Asian dust is transported into the Pacific Ocean (Zhao et al., 2006). Thus the impacts of East Asian dust can go well beyond source regions, on hemispheric or even global scales (e.g., Creamean et al., 2013; Uno et al., 2009). However, modeling of dust in East Asia by GCMs still has large uncertainties and is not well validated by observations. Wu et al. (2018) found that dust emission from Coupled Model Intercomparison Project Phase 5 (CMIP5) models differs greatly in spatial distribution and intensity over East Asia. Pu and Ginoux (2018) showed that seasonal variations of dust optical depth (DOD) over northern China are not captured by most CMIP5 models.

In the past decades, modeling studies have been conducted to examine the trans-Pacific transport of Asian dust and physical processes affecting dust cycle in East Asia (e.g., Eguchi et al., 2009; Gong et al., 2003; Su & Toon, 2009; Uno et al., 2006, 2008; Wu et al., 2016; Yumimoto et al., 2009; Zhang et al., 2018). In these studies, modeled total aerosol extinction and dust extinction were compared with measurements from a few paths of Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) and surface lidar networks, which shows limited information on spatial distributions of dust in East Asia. Few studies have assessed the performance of GCMs in simulating dust in East Asia. Spatial distributions and temporal variations of dust from GCMs are still poorly constrained. Recently, long-term retrievals of dust extinction from CALIPSO have been developed to better understand the global dust transport (e.g., Luo et al., 2015a, 2015b; Proestakis et al., 2018; Yu et al., 2015). Proestakis et al. (2018) presented a nine-year dust spatial distribution over East Asia, showing strong seasonality of DOD in Taklamakan and Gobi Deserts (0.078 in winter and 0.193 in spring). Luo et al. (2015b) developed a new dust identification method by combining measurements from CloudSat and CALIPSO, which detects much higher dust occurrences than CALIPSO level 2 products (CAL-L2).

There have been many studies using visibility data and dust event records from weather stations to analyze the spatial and temporal distributions of dust storm frequencies in East Asia (e.g., An et al., 2018; Guan et al., 2017; Kurosaki & Mikami, 2003; Shao et al., 2003; Wu et al., 2018). Observed dust storm frequencies in northern China show a general decreasing trend from 1960s to 2000s, which can be mainly attributed to the reduction of surface winds (e.g., Ding et al., 2005; Guan et al., 2017; Qian et al., 2002). However, this approach provides weak constraints on dust emissions and dust surface concentrations, because it focuses on the frequency rather than the strength of dust storms, and visibility records are not purely indicative of dust particles (Mahowald et al., 2007; Zhao et al., 2006). Therefore, empirical relationships between visibility and TSP (total suspended particle) or PM₁₀ (particulate matter with an aerodynamic diameter < 10 μm) concentrations were derived (e.g., Shao et al., 2003; Song et al., 2007; Wang et al., 2008) and used for better comparisons of modeled dust surface concentrations (e.g., Li et al., 2017; Uno et al., 2006).

In this study, we compare dust extinction profiles and DOD simulated from the Community Atmosphere Model version 5 (CAM5) of CESM1.2 with satellite retrievals from Luo et al. (2015a, 2015b) (hereafter L15), Yu et al. (2015) (hereafter Y15), and CAL-L3. We pay attention not only to the physical processes causing the model biases of dust in East Asia but also to the uncertainties in dust extinction retrievals from CALIPSO and the impacts of these uncertainties on the model evaluation. We also compare the probability density functions (PDFs) of modeled dust surface concentrations and 10 m wind speed with meteorological records from surface weather stations. To better understand the model biases in East Asia, dust emission schemes of Kok et al. (2014a, 2014b) (hereafter K14) and Ginoux et al. (2001) (hereafter G01) were implemented in the Community Land Model version 4 (CLM4.0) of CESM. Treatment of sub-grid surface wind variability in calculating dust emission by Zhang et al. (2016) was also applied in CLM4.0. The goal of this study is to evaluate the performance of CESM over East Asia in the simulations of (1) dust extinction profiles and DOD, including their spatial distributions and temporal variations, (2) dust mass budgets in the Taklamakan Desert and Tibetan Plateau (see Figure 1), and (3) dust surface concentrations and 10 m wind speed with a special focus on their PDFs.

Figure 1.

Illustration of the selected domains over East Asia with terrain height (m). The black dashed lines delineate the domains of longitude-altitude and latitude-altitude cross sections, respectively. Rectangles of black solid lines denote the Taklamakan Desert, Gobi Desert, and Tibetan Plateau for the analysis in this study. Blue dots represent 12 selected weather stations for the PDF analysis in this study.

The paper is organized as follows. Section 2 introduces CESM, dust emission schemes, treatment of sub-grid surface wind variability, and numerical experiments. Section 3 presents the observation data of dust extinction, DOD, and dust surface concentration used in this study. Section 4 evaluates modeled dust over East Asia with satellite retrievals and surface observations, and analyzes the dust mass budgets in the Taklamakan Desert and Tibetan Plateau. Discussion and conclusions are presented in section 5.

2. Methodology

2.1. Model Description

We use the CESM1.2 (Hurrell et al., 2013) with CAM5 (version 5.3.52) (Neale et al., 2010) as the atmospheric component. In this study, we use the four-mode version of MAM (MAM4) (Liu et al., 2016). Dust is carried in an Aitken mode, an accumulation mode, and a coarse mode with emission diameter bounds at 0.01–0.1 μm, 0.1–1.0 μm, and 1.0–10.0 μm, respectively. The particle size distribution for dust entrainment follows brittle fragmentation theory for vertical dust flux (Kok, 2011) with prescribed emission mass fraction of 1.1%, 0.00165%, and 98.9% for the three modes, respectively. The dust emission flux is calculated based on the Dust Entrainment and Deposition model of Zender et al. (2003a) (hereafter Z03), which is implemented in CLM4.0 (Oleson et al., 2010). Dust is emitted into four transport bins with diameter bounds at 0.1–1.0 μm, 1.0–2.5 μm, 2.5–5.0 μm, and 5.0–10.0 μm in CLM4.0, and then redistributed to MAM modes based on the above mass fractions.

2.2. Dust Emission Schemes

2.2.1. Z03

In the default Z03 scheme, the vertical mass flux of dust, F_j (kg m⁻² s⁻¹), in transport bin j is calculated as:

$$F_j=TS{f}_m\alpha Q_s{\sum }_{i=1}^l{M}_{i,j}$$ (1)

where T is a pre-factor that compensates for the sensitivity of dust emission to horizontal and temporal resolutions; S is the source erodibility factor (also called as source function); f_m is the grid cell fraction of exposed bare soil suitable for dust mobilization; α (m⁻¹) is the sandblasting mass efficiency; Q_s (kg m⁻¹ s⁻¹) is the total horizontal saltating mass flux, depending on the wind friction speed (u_*s, m s⁻¹); M_ij is the mass fraction of each source mode i carried in transport bin j. S is used to account for global variations in soil erodibility. S follows the geomorphic hypothesis and is characterized by the upstream area from which sediments may have accumulated locally through all climate regimes (Zender et al., 2003b).

2.2.2. K14

Kok et al. (2014a, 2014b) derived a new physically-based dust emission parameterization accounting for the increased sensitivity of dust emission fluxes to the soil state, which does not use the prescribed source function to shift dust emission towards the world’s most erodible regions. The expression for vertical dust flux, F_d (kg m⁻² s⁻¹) is given by:

$$F_d=C_d{f}_{\mathit{\text{bare}}}{f}_{\mathit{\text{clay}}}\frac{{\rho }_a(u_{*}^2-u_{*t}^2)}{u_{*st}}{\left(\frac{u_{*}}{u_{*t}}\right)}^{C_{\alpha }\frac{u_{*st}-u_{*st0}}{u_{*st0}}}(u_{*}>u_{*t})$$ (2)

where C_d is the dimensionless dust emission coefficient; f_bare is the bare soil fraction of the surface; f_clay is the soil clay fraction; ρ_a (kg m⁻³) is the air density; u_*t (m s⁻¹) is the soil the value of u_*t at standard atmospheric density at sea level; C_α is a dimensionless constant; and u_*st0 (m s⁻¹) is the minimum value of u_*st.

2.2.3. G01

In G01, the dust emission depends on surface wind speed at 10 m (u₁₀) instead of friction velocity. The vertical dust flux F_p of particle size class p is approximated by the expression:

$$F_p={\mathit{CSs}}_p{u}_{10}^2(u_{10}-u_t)(u_{10}>u_t)$$ (3)

where C is a dimensionless factor set to 1 μg s² m⁻⁵; S is the source function; u_t is the threshold velocity; and s_p is the fraction of each size class. S follows the topographic hypothesis and is defined as the fraction of alluvium (stream deposited sediments) available for wind erosion (Ginoux et al., 2001).

2.2.4. Sub-grid Surface Wind Variability

Zhang et al. (2016) introduced a sub-grid treatment of surface wind variability for dust and sea salt emissions using the grid-box mean physical quantities and their relationship to the sub-grid standard deviation of surface wind speeds derived from a global 15 km analysis dataset. In the treatment, emission fluxes are calculated multiple times using different wind speed samples of a Weibull probability distribution. The Weibull distribution depends on the standard deviation of sub-grid surface wind speed, which is approximated using four components: turbulence under neutral and stable conditions, dry convective eddies, moist convective eddies over the ocean, and air motions induced by mesoscale systems and fine-scale topography over land. Dust emission in East Asia increases by 25%, when global mean dust emission is tuned to the level of the default CESM1.2. The frequency distribution of dust emission changes, with more contribution from weaker (grid-mean) wind events and less contribution from stronger (grid-mean) wind events.

2.3. Experiments Design

We ran CAM5 with the finite-volume dynamical core at 0.9°×1.25° horizontal resolution with 56 vertical levels from 2006 to 2010, and the last 4-year results were used for analysis. The horizontal wind components u and v were nudged towards the Modern-Era Retrospective Analysis for Research and Applications version 2 (MERRA2) meteorology using relaxation time scale of 6 hours, and present-day monthly mean climatological SST and sea ice were used (Hurrell et al., 2008). Simulations were run with three different dust emission schemes (Z03, K14, G01) at three different horizontal resolutions (2°, 1°, 0.5°). As listed in Table 1, the first three experiments (Z03_f19, Z03_f09, Z03_f05) all use the default Z03 dust emission scheme and were conducted at 2°, 1°, and 0.5°, respectively. All tunable parameters (e.g., dust_emis_fact) are kept the same for these three experiments. Z03_f09 is the default experiment, and dust emission in Z03_f09 was tuned so that AOD in dusty region (DOD/AOD > 0.5) matches observations from the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard Terra and Aqua. SGV_f09, K14_f09, and G01_f09 were all run at 1°, and global annual mean dust emissions in the three experiments were tuned to be the same as Z03_f09 (within 1% difference). However, regional differences in dust emission can be large due to substantial differences in the dust emission schemes. SGV_f09 uses the default Z03 dust emission scheme but considers sub-grid surface wind variabilities in the dust emission calculation. K14_f09 and G01_f09 use the K14 and G01 dust emission scheme, respectively.

Table 1.

Experiment design

Experiments	Dust emission scheme	Sub-grid wind variability	Horizontal Resolution
Z03_f19	Zender et al. (2003a)	Not applied	2 degree 1.9°x2.5°
Z03_f09	Zender et al. (2003a)	Not applied	1 degree 0.9°x1.25°
Z03_f05	Zender et al. (2003a)	Not applied	0.5 degree 0.5°x0.63°
SGV_f09	Zender et al. (2003a)	Zhang et al. (2016)	1 degree 0.9°x1.25°
K14_f09	Kok et al. (2014a, b)	Not applied	1 degree 0.9°x1.25°
G01_f09	Ginoux et al. (2001)	Not applied	1 degree 0.9°x1.25°

To make an apple-to-apple comparison of modeled dust extinction with satellite observations, two treatments were applied to collocate model results and the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) measurements: (1) Dust extinction retrievals from L15 and Y15 were averaged into CAM5 0.9°×1.25° grid boxes and 38 pressure levels (25 hPa), and the nighttime CAM5-simulated dust extinctions were sampled every 10 seconds along the CALIPSO satellite tracks; (2) The CAM5-simulated dust extinctions were set to missing values in and below the vertical layer where cloud fraction is 100%, accounting for the fact that dust inside clouds, adjacent to the cloud bottom, and bellow optical thick clouds cannot be retrieved from CALIOP. Collocated dust extinction from model experiments is then integrated vertically to get the DOD value.

3. Observations

3.1. Dust Extinction and Dust Optical Depth

In this study, we use the dust extinction dataset developed by Luo et al. (2015a, 2015b) for the period of 2007 to 2009. In CAL-L2, aerosol is classified into six subtypes: clean marine, dust, polluted dust, clean continental, polluted continental, and smoke (Omar et al., 2009). CAL-L2 does not resolve dust extinction from polluted dust and has difficulties in separating optically thin ice clouds from dust. The polluted dust in CAL-L2 is treated as dust mixing with smoke (Omar et al., 2009), which is not appropriate for dust mixing with other types of aerosols (e.g., sea salt). Luo et al. (2015a) developed a new dust separation method to derive the dust backscatter coefficient (β_d, m⁻¹ sr⁻¹) in the lidar equation inversion stage using the CAL-L1B data, which has less uncertainties than doing the partition based on lidar inversion products (i.e., CAL-L2) in previous studies (e.g., Amiridis et al., 2013; Proestakis et al., 2018; Yu et al., 2015). The dust extinction can be easily obtained by assuming a dust extinction-to-backscatter ratio (lidar ratio, S_d) of 55 sr for β_d. Luo et al. (2015b) developed a new dust identification method for the thin dust layer detection by using combined lidar-radar cloud masks from the Cloud Profiling Radar (CPR) onboard CloudSat and the CALIOP onboard CALIPSO.

Yu et al. (2015) derived β_d from CAL-L2 total aerosol backscatter coefficient (β_a) and the ratio of dust to total backscatter (f_d). f_d is calculated based on observed particulate depolarization ratio as follows:

$$f_d=\frac{\left(\delta -{\delta }_{nd}\right)\left(1+{\delta }_d\right)}{(1+\delta )\left({\delta }_d-{\delta }_{nd}\right)}$$ (4)

where δ is CALIOP observed particulate depolarization ratio; δ_d and δ_nd are priori knowledge of depolarization ratios of dust and non-dust, respectively. We calculate f_d in the “lower-bound dust fraction” scenario (δ_d=0.30, δ_nd=0.07) and the “upper-bound dust fraction” scenario (δ_d=0.20, δ_nd=0.02). S_d is given a value of 45 sr. In this study, we apply the dust separation method to retrieve dust extinction under the cloud free condition based on CAL-L2 version 4 lidar products. Several quality control procedures are performed to ensure the retrieval quality. Dust extinction is only retrieved during nighttime, because sunlight interference reduces the accuracy of lidar observations (Winker et al., 2009). Only layers with cloud-aerosol discrimination (CAD) scores between −100 and −70 and extinction quality control flag values of 0, 1, 16, and 18 are selected (Yu et al., 2010; Yu et al., 2015). We include the aerosol free condition (dust extinction is zero) when we calculate the dust extinction profiles. DOD for L15 and Y15 is integrated vertically from dust extinction retrievals. We also use CAL-L3 version 3 nighttime cloud-free lidar data (Winker et al., 2013) of dust extinction profiles and DOD from 2007 to 2009. Aerosol extinction profiles from CAL-L2 are aggregated onto a global 2°x5° grid in CAL-L3. We only use extinction profiles and DOD of dust aerosol subtype. The original L15 and Y15 are ungridded geo-located profiles. The spatial resolution of L15 is 60 m vertically and 1 km horizontally. Y15 is based on CAL-L2 version 4.10 5 km aerosol profile product. The spatial resolution of Y15 is 60 m vertically and 5 km horizontally.

One of the main factors contributing to the uncertainties of dust extinction retrievals is the determination of lidar ratio S_d. In L15 and Y15, β_d is converted to dust extinction with S_d of 55 and 45 sr, respectively. In CAL-L3, dust aerosol type is identified with an aerosol lidar ratio S_a of 40 sr. The derivation of β_d from L15 and Y15 also constitutes an important source of uncertainties. Luo et al. (2015a) evaluated the new dust separation method with airborne High Spectral Resolution Lidar (HSRL) measurements of a dust mixed with sea salt case. They found that β_d is underestimated compared to HSRL with a mean difference of −1×10⁻⁴ km⁻¹ sr⁻¹ and a mean relative difference of −28%. The assumptions of δ_d and δ_nd in L15 and Y15 strongly affect β_d. Yu et al. (2015) estimated that such uncertainty is about 15–34%. The dust extinction in and below optically thick clouds cannot be retrieved in L15 due to the strong attenuation of cloud scattering. Because of the CALIOP’s relative low sensitivity, some tenuous dust layers may not be detected.

3.2. Dust Surface Concentrations

In this study, we use the surface meteorological records at 3 hour intervals by 449 weather stations in China and Mongolia, including surface wind at 10 m, dust events, and visibility. In these records, four categories of dust events are reported: suspended dust, blowing dust, dust storm, and severe dust storm. Shao et al. (2003) developed an empirical relationship between TSP concentration and visibility for East Asia as follows:

$$C_{dust}=\{\begin{array}{ll}3802.29D_v^{-0.84}\hfill & D_v<3.5\hfill \\ \exp (-0.11D_v+7.62)\hfill & D_v\ge 3.5\hfill \end{array}$$ (5)

where C_dust is the dust concentration (μg m⁻³), D_v is the observed visibility (km). Note that the size range of TSP (aerodynamic diameter less than 20–50 μm) is much larger than PM10 (aerodynamic diameter less than 10 μm), which is not appropriate for the evaluation of dust surface concentrations from CAM5. Song et al. (2007) presented an empirical relationship between PM₁₀ concentration and visibility for East Asia as follows:

$$C_{\mathit{\text{dust}}}=\{\begin{array}{ll}\exp \left(8.75-0.58D_v\right)\hfill & D_v<3.5\hfill \\ 4589.80D_v^{-1.35}\hfill & D_v\ge 3.5\hfill \end{array}$$ (6)

In this study, we use the relationship between PM₁₀ concentration and visibility estimated by Wang et al. (2008) due to better correlation and longer sampling period. We also test the relationships from Shao et al. (2003) and Song et al. (2007) and compare the derived dust concentrations with the estimation using Wang et al. (2008). The dust surface concentrations, C_dust, is approximated as follows:

$$C_{\mathit{\text{dust}}}={10}^8\times {({10}^3\times D_v)}^{-1.418}$$ (7)

We calculate dust surface concentrations derived from visibility records using equation (7) when dust events occur.

4. Results

Table 2 gives the annual mean global dust mass budgets and DOD for model experiments. As the model horizontal resolution changes from 2° to 0.5°, dust emission increases from 2692 (Z03_f19) to 5023 Tg yr⁻¹ (Z03_f05) resulting in the increase of dust burden from 17.3 (Z03_f19) to 31.0 Tg (Z03_f05), which further leads to the increase of DOD from 0.0142 (Z03_f19) to 0.0259 (Z03_f05). Finer model resolution can increase surface wind and result in higher dust emission and column burden (e.g., Gläser et al., 2012). Dry deposition is the dominant removal process of dust compared with wet deposition, which is approximately 2400 Tg yr⁻¹ for Z03_f09, SGV_f09, and K14_f09. G01_f09 has notable higher dry deposition (2670 Tg yr⁻¹) and lower wet deposition (1159 Tg yr⁻¹) than other 1° experiments. Although global dust emission from SGV_f09, K14_f09, and G01_f09 is tuned to be the same as Z03_f09 (~3820 Tg yr⁻¹), DOD from SGV_f09, K14_f09, and G01_f09 is slightly higher than Z03_f09 due to longer dust lifetime, which can be further attributed to the decrease of removal rate. The size distribution of dust particles can also affect DOD in addition to dust lifetime.

Table 2.

Annual mean global dust mass budgets and DOD for different experiments

	Z03 f19	Z03 f09	Z03 f05	SGV f09	K14 f09	G01 f09
Emission (Tg yr−1)	2692	3817	5023	3824	3823	3824
Dry deposition (Tg yr−1)	1704 (0.270)	2447 (0.276)	3260 (0.288)	2435 (0.269)	2411 (0.257)	2670 (0.290)
Wet deposition (Tg yr−1)	991 (0.157)	1375 (0.155)	1768 (0.156)	1393 (0.154)	1416 (0.151)	1159 (0.126)
Lifetime (days)	2.34	2.32	2.25	2.36	2.45	2.40
Burden (Tg)	17.3	24.3	31.0	24.8	25.7	25.1
DOD	0.0151	0.0215	0.0278	0.0219	0.0226	0.0233

Note. The values in parentheses are mean removal rates (deposition/burden, d⁻¹).

Figure 1 provides a map of the selected domains for analysis in this study. Area 1 (60°E-150°E, 35°N-50°N) is for the longitude-altitude cross sections of dust extinction in Figure 2. Area 2 (80°E-90°E, 25°N-55°N) is for the latitude-altitude cross sections of dust extinction in Figure 3. Areas 3–5 denote the Taklamakan Desert (75°E-95°E, 35°N-45°N), Gobi Desert (100°E-115°E, 37°N-47°N), and Tibetan Plateau (80°E-95°E, 30°N-35°N) for the analysis of dust extinction profiles, DOD, dust mass budgets, and dust surface concentrations in Table 3 and Figures 4, 7, 8, 9, and 10. Blue dots represent 12 selected weather stations for the PDF analysis of dust surface concentrations and 10m wind speed in Figures 11, 12, 13, and 14.

Figure 2.

Longitude-altitude cross sections (60°E-150°E, 35°N-50°N) of seasonal mean dust extinction (km⁻¹) from the default experiment (Z03_f09), L15, and Y15 during 2007–2009.

Figure 3.

Latitude-altitude cross sections (80°E-90°E, 25°N-55°N) of seasonal mean dust extinction (km⁻¹) from the default experiment (Z03_f09), L15, and Y15 during 2007–2009. The white space (25°N-45°N, >500 hPa) represents the Tibetan Plateau.

Table 3.

Annual and seasonal mean DOD in the Taklamakan Desert, Gobi Desert, and Tibetan Plateau

Taklamakan	Luo 2015	Yu 2015	CAL-L3	Z03_f09	SGV_f09	K14_f09	G01_f09
ANN	0.103	0.096	0.228	0.168 (0.190)	0.209 (0.231)	0.070 (0.077)	0.069 (0.077)
MAM	0.138	0.165	0.425	0.284 (0.281)	0.329 (0.328)	0.112 (0.111)	0.099 (0.102)
JJA	0.106	0.115	0.229	0.228 (0.290)	0.292 (0.358)	0.097 (0.114)	0.091 (0.109)
SON	0.092	0.070	0.141	0.119 (0.141)	0.157 (0.179)	0.051 (0.057)	0.061 (0.068)
DJF	0.072	0.055	0.113	0.043 (0.043)	0.055 (0.057)	0.022 (0.024)	0.024 (0.027)
Gobi
ANN	0.049	0.055	0.066	0.104 (0.111)	0.118 (0.126)	0.105 (0.110)	0.072 (0.077)
MAM	0.074	0.113	0.119	0.193 (0.200)	0.207 (0.213)	0.189 (0.195)	0.111 (0.114)
JJA	0.034	0.053	0.053	0.126 (0.137)	0.155 (0.165)	0.136 (0.145)	0.096 (0.103)
SON	0.040	0.032	0.034	0.083 (0.085)	0.095 (0.100)	0.082 (0.078)	0.067 (0.067)
DJF	0.043	0.039	0.058	0.021 (0.021)	0.023 (0.024)	0.019 (0.021)	0.019 (0.022)
Tibet
ANN	0.024	0.012	0.011	0.011 (0.011)	0.014 (0.013)	0.020 (0.020)	0.006 (0.006)
MAM	0.036	0.034	0.026	0.017 (0.017)	0.021 (0.019)	0.035 (0.039)	0.010 (0.011)
JJA	0.022	0.017	0.009	0.022 (0.021)	0.028 (0.026)	0.026 (0.024)	0.009 (0.008)
SON	0.016	0.004	0.002	0.004 (0.004)	0.005 (0.005)	0.011 (0.010)	0.004 (0.004)
DJF	0.023	0.004	0.006	0.002 (0.002)	0.002 (0.002)	0.008 (0.008)	0.001 (0.001)

Note. The values in parentheses are DOD from direct model output.

Figure 4.

Vertical profiles of seasonal mean dust extinction (km⁻¹) from 1° model simulations (Z03_f09, SGV_f09, K14_f09, G01_f09), L15, Y15, and CAL-L3 over the Taklamakan Desert (75°E-95°E, 35°N-45°N), Gobi Desert (100°E-115°E, 37°N-47°N), and Tibetan Plateau (80°E-95°E, 30°N-35°N) during 2007–2009.

Figure 7.

Seasonal variations of (a) dust emission, (b) dust dry deposition, (c) removal rate of dust dry deposition, (d) dust transport, (e) dust wet deposition, (f) removal rate of dust wet deposition, (g) dust burden, and (h) dust optical depth in the Taklamakan Desert (75°E-95°E, 35°N-45°N).

Figure 8.

Similar to Figure 7, but for the Tibetan Plateau (80°E-95°E, 30°N-35°N).

Figure 9.

Simulated annual mean dust surface concentrations (μg m⁻³) and dust concentrations derived from visibility records at 58 weather stations in the Taklamakan Desert during dust events. The color indicates the annual mean number of observed dust events.

Figure 10.

Same as Figure 9, but for 125 weather stations in the Gobi Desert.

Figure 11.

Probability density functions of dust surface concentrations sampled during dust events from 1° experiments (Z03_f09, SGV_f09, K14_f09, G01_f09) and visibility records at 6 weather stations in the Taklamakan Desert. The values indicate the difference of mean dust surface concentrations (Z03_f09 - Obs).

Figure 12.

Same as Figure 11, but for selected observation sites in the Gobi Desert.

Figure 13.

Probability density functions of 10 m wind speed sampled during dust events from experiments at different resolution (Z03_fl9, Z03_f09, Z03_f05), MERRA2, and observations at the same weather stations as Figure 11 in the Taklamakan Desert. The values indicate the difference of mean 10 m wind speed (Z03_f09 - Obs).

Figure 14.

Same as Figure 13, but at the same weather stations as Figure 12 in the Gobi Desert.

4.1. Dust Extinction and Dust Optical Depth

Figure 2 shows the longitude-altitude cross sections (60°E-150°E) averaged over 35°N-50°N of modeled seasonal mean dust extinction from Z03_f09 in comparison with satellite retrievals from L15 and Y15. High dust extinction is found in the lower troposphere (>600 hPa) from 60°E to 120°E, indicating two major source regions in East Asia: the Taklamakan and Gobi Deserts. In general, CAM5 overestimates dust extinction over the Taklamakan and Gobi Deserts in MAM (March-April-May), JJA (June-July-August), and SON (September-October-November) but underestimates dust extinction in DJF (December-January-February) compared with observations from L15 and Y15. Modeled dust extinction has its maximum in MAM and its minimum in DJF, which is similar to L15 and Y15. The high model biases of dust extinction in the lower troposphere (>600 hPa), especially in MAM, can be attributed to the overestimation of dust emission over the Taklamakan and Gobi Deserts. The high model biases of dust extinction in the upper troposphere (<300 hPa), especially in JJA, can result from excessive convective transport of dust (e.g., Allen & Landuyt, 2014) and lack of secondary activation of aerosols entrained into convective updrafts (e.g., Wang et al., 2013; Yu et al., 2019). It can also be attributed to the transport in the upper troposphere of dust from Africa and Middle East (e.g., Tanaka et al., 2005). Note that dust extinction from Y15 is larger than L15 in the lower troposphere but lower than L15 in the upper troposphere, which is due to the improved dust identification and separation method by Luo et al. (2015a, 2015b). Luo et al. (2015a) showed that the improved dust separation method produces lower β_d than the method using lidar inversion products (i.e., CAL-L2 total aerosol extinction), which leads to lower dust extinction in the lower troposphere. Luo et al. (2015b) showed that the improved dust identification detects higher dust occurrences than CAL-L2, which leads to higher dust extinction in the upper troposphere.

Figure 3 shows the latitude-altitude cross sections (25°N-55°N) averaged over 80°E-90°E of modeled seasonal mean dust extinction from Z03_f09 in comparison with satellite retrievals from L15 and Y15. High dust extinction is found to the south of the Tibetan Plateau due to dust sources in India and Pakistan and desert dust transported from North Africa and the Middle East. High dust extinction to the north of the Tibetan Plateau corresponds to dust emitted from the Taklamakan Desert. Modeled dust extinction is lower than observations from L15 and Y15 in the lower and middle troposphere to the south of the Tibetan Plateau, which may be due to the underestimation of local dust emission or dust transport from North Africa and the Middle East. Similar to Figure 2, CAM5 overestimates dust extinction in the upper troposphere over the Tibetan Plateau compared with observations. However, CAM5 underestimates dust extinction in the middle troposphere (400–600 hPa) over the Tibetan Plateau in MAM, SON, and DJF. This suggests that CAM5 may underestimate dust emission over the Tibetan Plateau and produce the wrong seasonality of dust burden. Ge et al. (2014) found that dust occurrences from CAL-L2 are high in MAM and JJA but low in SON and DJF over the Tibetan Plateau. Sun et al. (2017) found a significant negative correlation between the dust burden in the Tibetan Plateau and the anomaly in the EASM index, highlighting the importance of local dust emission in the Tibetan Plateau. We also compare latitude-altitude cross sections (110°E-120°E, 20°N-60°N) of seasonal mean dust extinction from Z03_f09 with observations from L15 and Y15 (see supporting information Figure S1).

Figure 4 compares vertical profiles of modeled dust extinction from 1° experiments (Z03_f09, SGV_f09, K14_f09, G01_f09) with satellite retrievals from L15, Y15, and CAL-L3 in the Taklamakan Desert (75°E-95°E, 35°N-45°N), Gobi Desert (100°E-115°E, 37°N-47°N), and Tibetan Plateau (80°E-95°E, 30°N-35°N). Overall, K14_f09 tends to have the best agreement with dust extinction from L15 in the Taklamakan Desert and Tibetan Plateau. G01_f09 tends to have the best agreement with dust extinction from L15 in the Gobi Desert. Dust extinction from the four experiments is close to each other in the upper troposphere (<300 hPa). CAM5 overestimates dust extinction in the upper troposphere over the Taklamakan Desert, Gobi Desert, and Tibetan Plateau in MAM, JJA, and SON. In the Taklamakan Desert (Figures 4a-4d), dust extinction from Z03_f09 and SGV_f09 is higher than dust extinction from K14_f09 and G01_f09 in the lower and middle troposphere. Dust extinction from K14_f09 and G01_f09 agrees well with observations from L15 and Y15 in the lower troposphere (>600 hPa) in MAM, JJA, and SON, while dust extinction from Z03_f09 and SGV_f09 is much higher than observations L15 and Y15 in the lower troposphere during the three seasons, especially in MAM can be one order of magnitude. This indicates that CAM5 with Z03 overestimates the dust emission over the Taklamakan Desert (see Figures 6 and 7). Dust extinction from Z03_f09 and SGV_f09 is in better agreement with observations from L15 and Y15 in the lower troposphere in DJF than dust extinction from K14_f09 and G01_f09. In the Gobi Desert (Figures 4e-4h), dust extinction from G01_f09 is lower than dust extinction from Z03_f09, SGV_f09, and K14_f09, especially in MAM. All four experiments overestimate dust extinction in MAM, JJA, and SON but underestimate dust extinction in DJF in the lower troposphere. In the Tibetan Plateau (Figures 4i-4l), dust extinction from K14_f09 agrees better with observations from L15 than dust extinction from Z03_f09 and SGV_f09 in the middle troposphere, especially in MAM and DJF, while G01_f09 gives the worst performance. This indicates that CAM5 underestimates the dust emission over the Tibetan Plateau and produces a wrong seasonality of dust mass burden (see Figures 6 and 8).

Figure 6.

Spatial distributions of annual mean dust emission (μg m⁻² s⁻¹) from model experiments over East Asia. The mean values are dust emission averaged over the black rectangles (70°-130°E, 25°-55°N). The boxes of black dashed lines indicate the Taklamakan Desert, Gobi Desert, and Tibetan Plateau.

Note that dust extinction from CAL-L3 is higher than L15 and Y15 in the lower troposphere but lower than L15 and Y15 in the middle and upper troposphere. Amiridis et al. (2013) calculated f_d to derive the dust extinction using the similar method (their equation 5) as Yu et al. (2015). They found that the new separation method reduces the high biases of dust extinction in the lower troposphere (<1000 m) and increases the dust extinction in the middle and upper troposphere compared with the original CAL-L3 dust extinction. The uncertainties in dust extinction profiles from L15, Y15, and CAL-L3 can have significant impacts on the model evaluation of dust spatial distributions. CAM5 tends to have less high model biases of dust extinction in the upper troposphere over the Taklamakan Desert, Gobi Desert, and Tibetan Plateau when compared with observations from L15. High model biases of dust extinction from Z03_f09 and SGV_f09 in the lower troposphere over the Taklamakan and Gobi Deserts become larger when comparing modeled dust extinction with observations from L15. We also compare vertical profiles of modeled dust extinction from experiments with Z03 at different resolutions (Z03_f19, Z03_f09, Z03_f05) with observations from L15, Y15, and CAL-L3 (supporting information Figure S2). The high-resolution experiment (Z03_f05) produces higher dust extinction than the low-resolution experiment (Z03_f19), which is most prominent in the Taklamakan Desert.

Figure 5 shows the spatial distributions of modeled DOD in comparison with satellite retrievals from L15, Y15 and CAL-L3. DOD from Z03_f09 and SGV_f09 is much higher than DOD from L15 and Y15 in the Taklamakan Desert due to the overestimation of dust emission (see Figures 6 and 7), while K14_f09 and G01_f09 show slightly lower DOD than L15 and Y15. All model experiments overestimate DOD in the Gobi Desert compared with observations from L15 and Y15. As model resolution increases from 2° to 0.5°, Z03_f05 produces larger DOD than Z03_f19 due to the increase of dust emission. DOD from CAL-L3 is much higher than DOD from L15 and Y15 in the Taklamakan Desert mainly due to the aforementioned high biases of dust extinction below 1000 m.

Figure 5.

Spatial distributions of global annual mean DOD from model experiments, L15, Y15, and CAL-L3. The black boxes indicate the Taklamakan Desert, Gobi Desert, and Tibetan Plateau. The stripe pattern of white space in (c) is due to the date collocation. Some grid boxes don’t have overpasses of CALIPSO satellite track due to the fine horizontal resolution.

Table 3 gives the annual and seasonal mean DOD from 1° experiments (Z03_f09, SGV_f09, K14_f09, G01_f09) in comparison with satellite retrievals from L15, Y15, and CAL-L3 in the Taklamakan Desert, Gobi Desert, and Tibetan Plateau. In the Taklamakan Desert, annual mean DOD from Z03_f09 (0.168) and SGV_f09 (0.209) is much higher than observations from L15 (0.103) and Y15 (0.096) due to the overestimation of dust emission, while DOD from K14_f09 (0.070) and G01_f09 (0.069) is lower than DOD from L15 and Y15. K14_f09 and G01_f09 has less model biases of DOD than Z03_f09 and SGV_f09. DOD from Z03_f09 reaches its maximum in MAM (0.284) and its minimum in DJF (0.043). All four experiments tend to underestimate DOD in DJF. In the Gobi Desert, all model experiments overestimate DOD compared with observations from L15 and Y15. G01_f09 has less model biases of DOD than other experiments. In the Tibetan Plateau, annual mean DOD from Z03_f09 (0.011) is slightly lower than DOD from Y15 (0.012) but much lower than DOD from L15 (0.024). DOD from Z03_f09 and SGV_f09 peaks in JJA, while DOD from L15 peaks in MAM. This indicates that CAM5 with Z03 produces the wrong seasonality of dust mass burden in the Tibetan Plateau. K14_f09 tends to have the best agreement with DOD from L15, while G01_f09 tends to give the worst agreement with DOD from L15. K14_f09 produces larger dust emission than other experiments over the Tibetan Plateau, which significantly increases DOD in MAM (0.035), SON (0.011), and DJF (0.008) and improves the seasonality of DOD. Note that DOD from CAL-L3 is two times of DOD from L15 and Y15 in the Taklamakan Desert, but is lower than L15 in the Tibetan Plateau.

We also show mean DOD from direct model output, not from the integration of collocated dust extinction profiles, in Table 3. DOD from direct model output is close (within 10% difference) to the DOD from collocated dust extinction profiles. Since we set dust extinction to missing values in and below the vertical layer where cloud fraction is 100%, DOD from the integration of collocated dust extinction profiles is generally lower than DOD from direct model output. Larger differences (~20%) of DOD are found in JJA. In the Taklamakan Desert, DOD from direct model output peaks in JJA, while DOD from collocated dust extinction profiles peaks in MAM. The collocation method improves the seasonality of modeled DOD. The observational uncertainties and limitations can have considerable influence on model evaluation of aerosols. Ma et al. (2018) found that using instrument simulators can bring model estimates of cloud susceptibility to aerosols close to satellite estimates. They found that the AOD is slightly lower than direct model output (see AOD from CAM5_cld and CAM5_clim in their supplementary information) when accounting for the orbital sampling issue and the fact that lidar measurements cannot detect below-cloud aerosols.

4.2. Dust Budget

In this section, we examine the dust mass budgets in the Taklamakan Desert and Tibetan Plateau to understand the reasons behind the seasonality of modeled dust extinction and DOD. Figure 6 shows the spatial distributions of annual mean dust emission over East Asia. CAM5 with Z03 produces larger dust emission than CAM5 with K14 and G01 in the Taklamakan and Gobi Deserts. Because the sub-grid treatment of surface wind variability changes the frequency distribution of dust emission with more contribution from weaker wind events and less contribution from stronger wind events, dust emission increases from 540 (Z03_f09) to 606 Tg yr⁻¹ (SGV_f09), which is consistent with Zhang et al. (2016). As model resolution increases from 2° to 0.5°, dust emission increases from 458 (Z03_f19) to 715 Tg yr⁻¹ (Z03_f05) due to better resolved sub-grid variability of surface wind. In the Tibetan Plateau, CAM5 with K14 produces considerable dust emission (18 Tg yr⁻¹), whereas CAM5 with Z03 and G01 produces little or even none dust emission (2 and 0 Tg yr⁻¹). Since K14 does not use source function but uses the soil’s threshold friction velocity to account for the soil’s susceptibility to wind erosion, CAM5 with K14 produces more spreads in dust source region than CAM5 with Z03 and G01.

Figure 7 shows seasonal variations of dust mass budgets and DOD in the Taklamakan Desert. In Figure 7a, dust emission from Z03_f09 and SGV_f09 tends to have much stronger seasonal variations and is much higher in MAM and JJA than dust emission from K14_f09 and G01_f09. The maximum of dust emission from Z03_f09 in May (~1600 kt d⁻¹) is much larger than that of K14_f09 (~600 kt d⁻¹), whereas the minimum of Z03_f09 in January is only slightly higher than that of K14_f09. Although dust emission from Z03_f09 strongly decreases after May, dust burden from Z03_f09 tends to be close in MAM and JJA (Figure 7g). The removal rate of wet deposition decreases from ~0.4 in May to ~0.2 d⁻¹ in August (Figure 7f). Together with low removal rate of dry deposition in JJA (Figure 7c), this leads to the second peak of dust burden in August. In Figure 7h, CAM5 with Z03 greatly overestimates DOD from April to September compared with observations from L15 and Y15 mainly resulting from the overestimation of dust emission. DOD from K14_f09 and G01_f09 is slightly lower than observations from L15 and Y15 in MAM and JJA due to the underestimation of dust emission. DOD from CAL-L3 is much larger than DOD from L15 and Y15, especially in April (>0.5).

Similar to Figure 7, Figure 8 shows seasonal variations of dust mass budgets and DOD in the Tibetan Plateau. In Figure 8a, dust emission from K14_f09 tends to have much stronger seasonal variations and is much higher than dust emission from Z03_f09 and SGV_f09. G01_f09 does not have dust emission in the Tibetan Plateau, which indicates dust burden from G01_f09 is all from transport. The maximum of dust emission from K14_f09 in April (~115 kt d⁻¹) is much larger than that of Z03_f09 in January (~10 kt d⁻¹), whereas the minimum of K14_f09 in August is slightly higher than that of Z03_f09 in October. Due to the strong dust transport to the Tibetan Plateau during April to August (Figure 8d), dust burden from Z03_f09 peaks in June and is only slightly lower than that from K14_f19 in JJA. In Figure 8h, CAM5 with K14 increases the DOD and improves the seasonality of DOD compared with other experiments. DOD from Y15 and CAL-L3 is much lower than DOD from L15 in SON and DJF. Seasonal variations of dust mass budgets and DOD in the Gobi Desert and southern China are shown in Figure S3 and S4, respectively.

4.3. Dust Surface Concentration

Figures 9 and 10 compare simulated annual mean dust surface concentrations with observations derived from visibility records at 58 and 125 weather stations in the Taklamakan and Gobi Deserts, respectively. We convert visibility to dust concentrations using equation (7) following Wang et al. (2008) and sample modeled dust surface concentrations when dust events are reported. Dust events are more frequently observed in the Taklamakan Desert than in the Gobi Desert. In general, CAM5 underestimates dust surface concentrations compared with observations, and modeled dust surface concentrations have a poor correlation (<0.3) with observations. K14_f09 and G01_f09 have larger low biases of modeled dust surface concentrations than Z03_f09. As model resolution increases from 2° to 0.5°, the low biases of modeled dust surface concentrations are reduced from Z03_f19 to Z03_f05. Note that the robustness of the conclusions is subject to the uncertainties of visibility-derived dust surface concentrations. Wang et al. (2008) derived power-law functions between PM₁₀ and visibility based on observations from 2001 to 2006 with a good correlation (~0.9). Relationships derived by Song et al. (2007) are only based on two-year observations with a relative poor correlation (<0.5). Note that dust surface concentrations from Z03_f09 can be close to or even higher than those derived from observed visibility when the relationships by Shao et al. (2003) and Song et al. (2007) are used.

Figures 11 and 12 further compare PDFs of modeled dust surface concentrations from 1° experiments (Z03_f09, SGV_f09, K14_f09, G01_f09) with PDFs of dust surface concentrations derived from visibility records at 12 weather stations in the Taklamakan and Gobi Deserts, respectively. In Figures 11a-11c, we show PDFs at three weather stations where mean dust concentrations from Z03_f09 are higher than observations. In Figures 11d-11f, we show PDFs at three weather stations where mean dust concentrations from Z03_f09 are lower than observations. Compared to observations in Figures 11a-11e, PDFs from Z03_f09 and SGV_f09 are shifted towards higher dust concentrations (1000–5000 μg m⁻³), while PDFs of dust concentrations from K14_f09 and G01_f09 mostly peak in the same range as observations (100–500 μg m⁻³). CAM5 overestimates the frequencies of low dust concentrations (<10, 10–50, and 50–100 μg m⁻³) compared to observations due to the limitations of observations. According to equation (7), dust surface concentrations can be lower than 100 μg m⁻³, if visibility is higher than 17 km. Dust events, especially suspended dust, are rarely reported for this visibility range. CAM5 with Z03 overestimates the mean dust surface concentrations at Tazhong, Aksu, and Alar due to the high biases of frequencies for dust concentrations above 1000 μg m⁻³ (1000–5000 μg m⁻³, 5000–10000 μg m⁻³, >10000 μg m⁻³). CAM5 with Z03 underestimates the mean dust surface concentrations at Golmud, Ruoqiang, and Lenghu due to the low biases of frequencies for dust concentrations above 5000 μg m⁻³ (5000–10000 μg m⁻³, >10000 μg m⁻³). Similar conclusions can be drawn from Figure 12. PDFs of dust concentrations from G01_f09 mostly peak in the same range as observations. CAM5 with Z03 overestimates the mean dust surface concentrations at Linhe, Datong, and Erenhot due to the high biases of frequencies for dust concentrations above 1000 μg m⁻³. CAM5 with Z03 underestimates the mean dust surface concentrations at Zamyn-Uud, Dalanzadgad, and Galuut due to the low biases of frequencies for dust concentrations above 5000 μg m⁻³. The low biases of frequencies for high dust concentrations indicates the issues in PDFs of surface wind speed, which will be discussed in the next session.

4.4. Surface Wind

The dust emission flux is highly sensitive to the surface wind speed (Luo et al., 2003; Schmechtig et al., 2011). In Z03 and K14, the dust emission varies with the cube of friction velocity; in G01, the dust emission is proportional to the cube of 10 m wind speed. Menut (2008) found that the dust emission can vary by a factor of 3 when using wind fields from two reanalysis (NCEP and ECMWF). Largeron et al. (2015) compared PDFs of surface wind speed from three global reanalysis (ERA-interim, NCEP-CFSR, and MERRA) with high-frequency observations in the Sahel and found that the three global reanalysis systematically underestimates the frequencies of strong surface events. This suggests that using the surface wind from the three global reanalysis to calculate dust emission may fail to capture strong dust events.

Figures 13 and 14 compare PDFs of 10 m wind speed sampled during dust events from Z03_f19, Z03_f09, and Z03_f05 with PDFs of MERRA2 and ground observations at the same 12 weather stations in the Taklamakan and Gobi Deserts, respectively. As the model resolution increases from 2° to 0.5°, PDFs tend to be shifted toward higher wind speed from Z03_f19 to Z03_f05. High model resolution can better capture sub-grid surface wind variability and increase the mean surface wind speed leading to high dust emission. In Figures 13a and 14a, CAM5 overestimates both the mean 10 m wind speed and frequencies of high wind speed (>8.5 m s⁻¹) at the stations where high model biases of dust surface concentrations are found. In Figures 13d-13f and 14d-14f, CAM5 underestimates both the mean 10 m wind speed and frequencies of high wind speed at the stations where low model biases of dust surface concentrations are found. The overall low model biases of dust surface concentrations during dust events (see Figures 9 and 10) can be attributed to the inability of CAM5 to capture strong surface wind events at most weather stations in the Taklamakan and Gobi Deserts. In Figures 13b-13c and 14b-14c, CAM5 slightly underestimates the mean 10 m wind speed and does not overestimate the frequencies of high wind speed at the stations where high model biases of dust surface concentrations are found. PDFs of MERRA2 are shifted toward higher wind speed in Figures 13a-13c but toward lower wind speed in Figures 13d-13f. CAM5 underestimates the frequencies of high wind speed in all six sites in Figure 13 compared with MERRA2. Similar conclusions can be drawn from Figure 14 for PDFs of MERRA2. We should note that dust emission schemes of Z03 and K14 use friction velocity instead of 10 m wind speed and the relationship between dust emission and surface wind is nonlinear, which can cause the inconsistency between model biases of dust surface concentrations and PDFs of 10 m wind speed.

5. Discussion and Conclusions

In this study, we implement treatment of sub-grid surface wind variability by Zhang et al. (2016) and dust emission schemes of Kok et al. (2014a, 2014b) and Ginoux et al. (2001) into CAM5 to better understand the model biases of dust in East Asia. We evaluate the spatial distributions and temporal variations of dust extinction and DOD by comparing model results from CAM5 with satellite retrievals from Luo et al. (2015a, 2015b), Yu et al. (2015), and CALIPSO level 3 products. We analyze the dust mass budgets in the Taklamakan Desert and Tibetan Plateau to examine the physical processes causing the model biases of dust extinction and DOD. CAM5 overestimates dust extinction in the upper troposphere (<300 hPa) over the Taklamakan Desert, Gobi Desert, and Tibetan Plateau in MAM, JJA, and SON compared with observations from L15 and Y15, especially in JJA can be one order of magnitude. The high model biases of dust extinction in the upper troposphere can be attributed to excessive convective transport of dust (e.g., Allen & Landuyt, 2014), lack of secondary activation of aerosols entrained into updrafts (e.g., Wang et al., 2013; Yu et al., 2019), and strong dust transport in the upper troposphere from Africa and Middle East (e.g., Tanaka et al., 2005).

Overall, CAM5 with K14 has the best agreement with dust extinction profiles and DOD from L15 in the Taklamakan Desert and Tibetan Plateau. CAM5 with G01 has the best agreement with dust extinction profiles and DOD from L15 in the Gobi Desert. In the Taklamakan Desert, annual mean DOD from K14_f09 and G01_f09 is lower than DOD from L15 and Y15, while annual mean DOD from Z03_f09 and SGV_f09 is much higher than observations due to the overestimation of dust emission. Dust extinction from K14_f09 and G01_f09 agrees well with observations from L15 and Y15 in the lower troposphere (>600 hPa) in MAM, JJA, and SON, while dust extinction from Z03_f09 and SGV_f09 is much higher than observations during the three seasons, especially in MAM can be one order of magnitude. In the Gobi Desert, annual mean DOD from all experiments is higher than observations from L15 and Y15. Dust extinction from G01_f09 is lower and agrees better with observations from L15 and Y15 than dust extinction from Z03_f09, SGV_f09, and K14_f09, especially in MAM. In the Tibetan Plateau, K14_f09 produces much larger dust emission than other experiments, which improves DOD and its seasonality compared with observations from L15. Dust extinction from K14_f09 agrees better with observations from L15 than dust extinction from Z03_f09 and SGV_f09 in the middle troposphere, especially in MAM and DJF, while G01_f09 gives the worst performance.

Large uncertainties exist in satellite retrievals of dust extinction profiles and DOD from CALIPSO and can have significant impacts on the model evaluation of dust spatial distributions. In the lower troposphere, dust extinction from L15 is the lowest, while dust extinction from CAL-L3 is the highest due to different dust separation methods. L15 and Y15 reduce the high biases of dust extinction compared with CAL-L3 (Amiridis et al., 2013). In the upper troposphere, dust extinction from L15 is the highest, while dust extinction from CAL-L3 is the lowest. The improved dust identification method by Luo et al. (2015b) detects much higher dust occurrences than CAL-L3, which leads to higher dust extinction in the upper troposphere. High model biases of dust extinction in the upper troposphere over the Taklamakan Desert, Gobi Desert, and Tibetan Plateau become less when comparing modeled dust extinction with observations from L15. Larger high model biases of dust extinction are found in the lower troposphere over the Taklamakan and Gobi Deserts when comparing modeled dust extinction with observations from L15. The observational uncertainties and limitations can also have considerable influence on the model evaluation. Although DOD from collocated dust extinction profiles are close to DOD from direct model output, the collocation method used in this study improves the seasonality of DOD in the Taklamakan Desert. Ma et al. (2018) implemented an aerosol lidar simulator in CAM5, which makes use of sampling and retrieval procedures similar to CALIOP. The aerosol lidar simulator outputs attenuated total backscatter and total aerosol extinction. It would be better to use the aerosol lidar simulator in the future.

We also assess dust surface concentrations and 10 m wind speed with meteorological records from weather stations during dust events. CAM5 underestimates dust surface concentrations during dust events in Taklamakan and Gobi Deserts. CAM5 fails to capture strong surface wind events at most weather stations, which causes the low model biases of frequencies for high dust concentrations and further leads to the low model biases of dust surface concentrations during dust events. As model resolution increases from 2° to 0.5°, PDFs are shifted towards higher wind speed from Z03_f19 to Z03_f05, because high model resolution can better capture sub-grid surface wind variability and increase the mean surface wind speed. We derive dust surface concentrations from visibility records during dust events due to a lack of high-frequency PM₁₀ measurements and direct measurements of dust. The robustness of the conclusions is limited by the observer biases in visibility records, uncertainties in the empirical relationships between visibility and PM₁₀, differences between dust concentrations and PM₁₀ during dust events, and instrument accuracy in measurements of surface wind speed.

Key Points:

• CESM with K14 dust emission scheme has the best agreement with dust extinction profiles and DOD from CALIPSO in East Asia.

• Large uncertainties exist in observations from CALIPSO and have significant impacts on the model evaluation of dust spatial distributions.

• CESM underestimates dust surface concentrations in East Asia due to the its inability to capture strong surface wind events.ssss