A high-resolution numerical study of the Asian dust storms of April 2001

Authors


Abstract

[1] A comprehensive dust aerosol model is developed and fully coupled to the U. S. Navy's operational Coupled Ocean/Atmospheric Mesoscale Prediction System (COAMPS™). The model is used to simulate the Asian dust storms of 5–15 April 2001 at 27-km resolution with 46 vertical layers. Dust was primarily generated in the Gobi and Taklamakan Deserts between 6 and 9 April and then swept across vast areas of east Asia. The model performance is verified with satellite products and by observations of PM10 and lidar data from Lanzhou, Beijing, Hefei, Tsukuba, and Nagasaki. The model simulates the right timing and strength of dust events, predicting depths and magnitudes of the boundary layer and elevated layer of dust plumes that compare well with observed values. Numerical analysis shows that the first Mongolia cyclone on the 6 and 7 April and the cold front on 8 and 9 April, accompanied by a second Mongolia low, form the major dynamic forcing patterns that mobilize, transport, and vertically redistribute the dust. Both cyclones entrain the dust and transport dust to altitudes of 8–9 km, while at the top of the cyclone, transport is anticyclonic and to the northeast. The analysis of the individual dynamic and microphysical tendency terms in the mass continuity equation reveals that in the dust generation area, mechanical and convective turbulence plays the major role in mixing dust upward to the top of the planetary boundary layer. In the downstream cyclone area, vertical advection by the model-resolved upward motion in the cyclones is the dominant dynamic process that transports dust to high altitudes and into the westerlies, making it available for long-range transport. The mass budget calculation for the entire simulation period reveals that about 75% of the total dust production is redeposited to the Asian deserts, 20% falls onto nondesert areas through dry and wet deposition, and 1.6% falls into the China and Japan Seas.

1. Introduction

[2] In recent years, increased attention has been given to the large amounts of airborne dust derived from the deserts and desertified areas of the world and transported over scales ranging from local to global. This dust can have positive and negative impacts on human activities and the environment, including modifying cloud formation, fertilizing the ocean, degrading air quality, reducing visibility, transporting pathogens, and inducing respiratory problems. The direct and indirect atmospheric radiative forcing by the dust has implications for global climate change and presently is one of the largest unknowns in climate models. Asian dust storms are well-known spring season phenomena that occur when weather becomes dry, the surface starts to warm, and strong winds sweep over the deserts and surrounding areas. East Asia has two of Earth's major natural dust sources: the Taklamakan Desert in west China and the Gobi Desert in Mongolia and northwest China. Figure 1 shows the desert areas in east Asia, presented as the distribution of dust erodible fractions in 27-km grid resolution. Desertification has increased erodible area surrounding these deserts in recent years, and dust storm frequency in the region has increased in the last decade [S. Wang et al., 2000]. Severe dust storms not only impact east Asia, but also can reach far beyond the continent, as did the dust clouds of April 1998 and other years that drifted over the Pacific Ocean to the west coast of North America [Husar et al., 2001; Tratt et al., 2001; McKendry et al., 2001].

Figure 1.

Distribution of dust erodible fraction (0.1–1.0) in east Asia calculated from USGS land cover data set presented at 27-km grid resolution. The low values of erodibility at the edges of deserts are due to the nature of land cover and also to the graphics contouring interpolation. Places marked with stars are the locations used in model validation. The desert area east of Taklamakan and south of Gobi includes portions of Qinghai and Gansu Provinces of China.

[3] A tremendously strong Asian dust storm occurred from 6 to 9 April 2001. During this episode, dust was generated from Gobi and Taklamakan Deserts every day under large systematic winds and favorable surface thermodynamic conditions. From satellite images and surface weather station recordings, the huge dust plumes were observed covering vast areas of China and Mongolia. Blowing sand created zero visibility in some parts of Gansu Province in China near the dust sources. The dust front reached the Korean Peninsula on 8 April, and Japan on 9 April. Satellite images from SeaWiFs show the Asian dust clouds moving across the Pacific Ocean in the upper westerlies, and arriving in North America on 12 and 13 April. These same dust clouds traversed North America, reaching the Atlantic Ocean on 19 and 20 April.

[4] In this paper, we examine the dust episode of 5 through 15 April 2001 by utilizing a comprehensive dust aerosol model that is fully coupled to the U.S. Navy's operational mesoscale weather prediction system (COAMPS™) as an online module. We conduct a numerical simulation of this special dust storm episode at high resolution to investigate the meteorological mechanisms by which lifted dust is transported through the planetary boundary layer and into the free troposphere. Previous numerical studies of dust have focused on generation and long-range transport of Saharan, Asian, and Australian dust over downwind continents and oceans, the role of transport in dust life cycle, and the impacts of transport on global radiation budget [Westphal et al., 1987, 1988; Tegen and Fung, 1994, 1995; Z. Wang et al., 2000; Nickovic et al., 2001; Uno et al., 2001; Lu and Shao, 2001]. Studies of major dust events have ranged from regional scale and short period (weeks), to global scale and long period (years). All of the model simulations for these studies were made with coarse horizontal resolution with grid spacing greater than 100 km, and coarse vertical grid resolution. Therefore the model outputs revealed only the synoptic-scale features of dust transport and, because of the limited vertical resolution, shed little light on vertical processes.

[5] It has been assumed that rapid vertical mixing of dust from the surface layer to the free troposphere followed by rapid horizontal transport is the critical mechanisms for efficient long-range dust transport [Kang and Sang, 1991; Uno et al., 2001]. This hypothesis has never been closely investigated in numerical simulations due to limited grid resolutions. Hence quantitative understanding of individual dust events is still incomplete [Husar et al., 2001]. The April 2001 dust episode provides an opportunity to study an event having good local-scale data and was detected nearly three-fourths of the way around Earth. It is a “perfect” case on which to conduct numerical modeling to study the meteorological circulations of different scales and understand their combined roles in dust transport. Questions that are raised include: What is the transport pattern in this strong dust episode? What mechanisms redistribute dust mass in the vertical? What dynamic forcing supports dust long-range transport? These are some of the issues we plan to resolve in this case study. Other issues we want to understand involve the mass budgets: How much dust is mobilized throughout the intense period? How much dust leaves the deserts? How much deposits on the land and ocean? How much undergoes long range transport out of east Asia. Since this case is a rarely seen strong dust episode, these mass budgets will yield an estimation of the upper limit of Asian dust storms on regional and global environments.

2. Numerical Model Description

[6] We use the atmospheric component of the U.S. Navy's operational Coupled Ocean/Atmospheric Mesoscale Prediction System (COAMPS™) for this case study. It is a nonhydrostatic and compressible dynamic model applied in a terrain-following sigma vertical coordinate σ = H(z − zs)/(H − zs), where H is the depth of the model domain and zs the terrain height. It predicts turbulent kinetic energy (TKE) for sub-grid-scale diffusion, uses a force-restore method in the surface energy budget, and contains explicit cloud microphysics. Friction velocity as the surface momentum flux is calculated from Monin-Obukhov surface layer similarity theory. The surface wind is then obtained from the integration of surface layer flux profiles. Ground wetness is calculated following the algorithm of Louis [1979], using precipitation, latent heat flux, and moisture capacity of the ground. The complete details of the model structure, dynamics, and physics can be found in the work of Hodur [1997]. Mesoscale data assimilation is performed at 12-hour incremental update cycles using the meteorological data from the world weather station network. There are a total of 1370 surface stations and 216 radio soundings located in our model domain, shown in Figure 1. This assimilation of conventional meteorological observations, as well as satellite-derived quantities, insures initial conditions with mesoscale features. The analyses and forecast fields of the Navy's operational global atmospheric prediction system (NOGAPS) [Hogan and Rosmond, 1991] are used for updating the lateral boundary conditions every 6 hours during forecasts.

[7] A dust microphysical aerosol model is developed and fully embedded in COAMPS™, i.e., an online module of the prediction system, using the exact model's meteorological fields at each time step and at each grid point. Therefore the dust module has the same grid structure and feature as the dynamics model, for example, having multiple nested grids and interaction between dust fields on the high- and low-resolution grids. The mass conservation equation for dust concentration of a particle concentration C in generalized form is

equation image

where u, v, w are the components of wind vector in x, y and z directions; vf is particle settling velocity; Dx, Dy and Dz are turbulent mixing terms in x, y and z; Csrc is dust source term, i.e., dust mobilization from erodible lands; Csnk is the dust sink term which includes dry deposition at the surface and wet removal by precipitation.

[8] There are basically two types of dust emission formulae used in dust aerosol modeling. The first type is a u*-dependent flux [e.g., Westphal et al., 1987] and the other type is wind speed-dependent flux developed by Gillette [1978]. Liu and Westphal [2001] conducted sensitivity studies of dust emission on grid resolution by comparing these two types of dust flux formulae and with observations. They found that the u*-driven flux is more realistic and preferable than the wind speed-driven flux scheme. The latter method failed to predict the smaller dust events and diurnal variations during a two-week period of April 1998 in east Asia due to the lack of a dependence on thermal stability and wind shear.

[9] The dust emission takes the formula of Nickling and Gillies [1993], which describes the vertical dust flux F (kg m−2 s−1) as proportional to friction velocity (u*) raised to the forth power (or the square of surface wind stress):

equation image

Term A in (2) is the fraction of model grid box that is dust erodible, ranging from 0.0 to 1.0., and u*t (m s−1) is the threshold friction velocity. The fractional erodibility A is derived using a 1-km-resolution land cover data set produced by the U. S. Geological Survey (USGS). As a further refinement, the land survey by Clements et al. [1957] is used to estimate that an average of 13% of each 1-km pixel in the USGS database is erodible. Therefore the flux calculated by (2) must be further scaled by a factor of 0.13 to yield total dust production. The USGS land cover data set shows that two land cover types dominate the deserts of east Asia: desert and “semi-desert sage or shrub.” Because of such a high coverage of only two land surface types, using a single threshold is appropriate in this case study. We choose a value for image of 0.65 m s−1 that is adapted from previous modeling work [Westphal et al., 1987, 1988] and field experiments [Gillette and Passi, 1988]. Dust emission is further restricted to erodible areas where the soil is relatively dry. The COAMPS model predicts a surface wetness variable (ground wetness) with a soil model that depends on precipitation and evaporation. Dust lifting is allowed when ground wetness is less than 0.3, a value derived from the average climatological data provided by Navy's NOGAPS and the USGS land cover data set in springtime over the Asia dust source areas.

[10] In sensitivity tests, we have found that realistic dust simulations are achieved, while maintaining computational practicality, by using 10 particle size bins, ranging from 0.04 to 36 μm in diameter. The particle size increases in geometrical volume at a constant ratio of 6.5, which is calculated following an algorithm described in the work of Westphal et al. [1987] and Toon et al. [1988]. Particles beyond this range are considered trivial in our transport study, with our focus on mass distributions, because of the quick gravitational settling of giant particles near dust sources, and the very small amount of dust mass having submicron particle sizes. Dust is modeled as a generalized species with density of 2650 kg m−3, and assuming a bimodal lognormal size distribution in mass. The mass median radius and geometric standard deviation in the size distribution are predicted by the sandblasting model of Alfaro et al. [1997] at each time step, based on the particle's saltating kinetic energy in a dynamical environment using the wind speed and particle size.

[11] Dust advection in both horizontal and vertical directions uses a fifth-order accurate flux-form scheme developed by Bott [1989a, 1989b] and briefly described as follows. The algorithm performs a polynomial fitting to the advected field in each grid box to make the fitting curves approach the real distributions. The fluxes at grid-box boundaries are integrated upstream from the fitting curves. The coefficients of the polynomial interpolation for each grid box are obtained by assuming the area under the fitting curve is preserved, and a weighting flux treatment is conducted to minimize both amplitude and phase errors, so that mass conservation and positive definition are achieved. Sub grid-scale turbulent mixing, e.g., Dx, Dy and Dz in equation (1), is solved implicitly with the eddy diffusivities for scalar generated by TKE closure, as are the other meteorological variables.

[12] Particle terminal velocity (vf) in equation (1) for sedimentation is calculated with Stokes Law using dust particles properties and the kinetic characteristics of air:

equation image

where Rp is dust particle radius (m), ρp dust density (= 2650 kg m−3), g gravity (m s−2), Cc the Cunningham correction factor, ν kinetic viscosity of air (m2 s−1), and ρa air density (kg m−3). Cc is defined in terms of the mean free path of air molecules (λ in units of m) as follows:

equation image

The values of λ and ν are obtained from the list of various air conditions in the work of Seinfeld [1986]. Sedimentation is solved in the vertical advection by combining the terminal velocity with the dynamical vertical velocity.

[13] Dry deposition is modeled for dust in the bottom grid layer. The time tendency at the surface is defined as the flux divergence over the depth of bottom grid layer (Δz):

equation image

where vd is the deposition velocity. In general, the deposition velocity has considerable uncertainty due to variations in meteorology and surface characteristics. In this case, we take a generalized approach by calculating vd based on surface wind stress and wind speed at 10m height (Um10) [Stull, 1988]

equation image

[14] Wet deposition by convective and stable precipitation is calculated at all levels in the vertical based on the scavenging rate Λ (s−1):

equation image

The parameterization of scavenging rate for both dust washout and rainout processes is obtained from Pruppacher and Klett [1978], by assuming wet removal is independent of particle sizes, such that

equation image

where R is the precipitation rate, H is either the cloud depth (m) for convective precipitation, or the model layer depth for stable precipitation. Both the removal processes of dry and wet deposition are solved implicitly to guarantee positive dust concentrations all the times.

[15] The model domain extends vertically to 36 km with 46 grid layers ranging from 10 m thick at the surface to 6 km thick at the top, with 30 grid layers below 8 km. In the horizontal, the model uses a 301 × 221 grid point mesh with 27 km grid spacing to cover the domain shown in Figure 1. The 27-km grid resolution is about one third of resolution of NOGAPS, the Navy's global circulation model. The combination of high resolution in both the vertical and horizontal dimensions allows the model to capture both synoptic and mesoscale features of dynamic systems and detailed distributions of dust mass. The model was used to continuously simulate dust for the period of 1–16 April 2001. While the meteorological fields were updated via data assimilation every 12 hours at 0000 and 1200 UTC, there was no dust assimilation; the model began each new forecast cycle with the previous forecasted dust distribution as the initial condition. The first few days of the simulation allow the model to develop the mesoscale weather features necessary for realistic dust predictions.

3. Meteorological Condition and Dust Production

[16] Figure 2 shows the COAMPS 12-hour forecasts of geopotential height and air temperature on the 500 mb surface at 0000 UTC from 6 to 9 April. The 12-hour forecasts of sea level pressure for the same time are shown in Figure 3. On 6 April, a short 500 mb trough west of Lake Baikal of Russia stretches southward to the west of China (Figure 2a). A strong temperature gradient exists along the trough axis with warm advection in front of the trough and cold advection behind. This upper air disturbance has caused a significant pressure decrease in the lower atmosphere as the result of cyclogenesis, and a deep low-pressure system has developed in northern Mongolia with a central sea level pressure of 988 mb (Figure 3a). The Mongolia low continues deepening as it moves eastward, with the strong pressure gradient to the south of the cyclone passing across Gobi Desert and causing dust mobilization. Figure 4a shows the modeled dust flux at 0600 UTC (local afternoon) with a maximum of 5.0 mg m−2 s−1. The surface station observations within and near the dust source areas confirm the occurrence of dust production at this time and in the areas.

Figure 2.

COAMPS 12-hour forecast of 500 mb geopotential height (m) and temperature (°C) valid at 00Z for (a) 6, (b) 7, (c) 8, and (d) 9 April.

Figure 3.

COAMPS 12-hour forecast of sea level pressure (mb) with H and L markers to indicate low and high systems valid at 00Z for (a) 6, (b) 7, (c) 8, and (d) 9 April.

Figure 4.

Modeled surface dust fluxes (mg m−2 s−1, shaded contours), and surface dust storm observations (dollar signs) at times indicated in the plots for (a) 6, (b) 7, (c) 8, and (d) 9 April.

[17] On 7 April, the midlatitude cyclone moves to the east of Mongolia and northeast China. The central sea level pressure drops to 984 mb (Figure 3b), and the cyclone becomes cutoff (Figure 2b). On the same day, another strong cold air outbreak from the polar region, with a cold center of −40°C, moves along the deep 500-mb trough east of Ural Mountains of Russia (Figure 2b). The corresponding surface high arrives in Kazakhstan, and the cold front runs across western China and Mongolia (Figure 3b). Strong winds and dust mobilization occur as the front crosses the Taklamakan and western Gobi Deserts. Figure 4b shows the dust flux at 1200 UTC, in shaded contours, is generally confined within the areas defined by observed surface dust. The peak modeled dust flux reaches 12.0 mg m−2 s−1 at this time.

[18] By 8 April, the 500-mb trough has moved down from the Ural Mountains to western China and Mongolia. Part of the trough deepens at 45°N and becomes another cutoff low at 500 mb (Figure 2c) that is followed by a cold center of −40°C with strong cold advection. At the surface, another Mongolia low has developed on the front (Figure 3c) through frontogenesis. The presence of this Mongolia low in combination with the high behind the cold front in Kazakhstan enhances the pressure gradient, increasing dust production to the highest levels seen during the case study period of 5 to 15 April. Figure 4c shows large areas of dust lifting in the Gobi and Taklamakan Deserts. The modeled dust flux reaches a maximum of 24.0 mg m−2 s−1 at 0600 UTC.

[19] On 9 April, both the 500-mb cutoff low and the surface Mongolia low move to eastern Mongolia, while the cold front moves further south and east (Figures 2d and 3d) as the cold air mass marches southeastward driven by the upper level northeasterly flows. Strong winds in the west side of the cyclone and behind the cold front continue to mobilize dust in the Gobi and Taklamakan Deserts, as seen on Figure 4d, with the maximum dust flux reaching 5.0 mg m−2 s−1 at 0600 UTC and with many observation stations reporting dust at the same time. In section 6 of mass budget calculation, we will show that the daily integrals of dust production in the whole source area from 6 to 9 April are much larger than the other days of the study period.

[20] During the four-day period of strong dust storms, dust production reaches a maximum in the afternoon (around 0600 UTC) on 6, 8, and 9 April and later afternoon (1200 UTC) on 7 April, under widespread strong winds typical of severe weather systems. This timing of daily dust generation has previously been described in a study of the April 1998 Asian dust storms by Liu and Westphal [2001] and is consistent with the daily timing of maximum-flux observed over a two-week period of April 1994 by Wang et al. [1995]. It indicates that diurnal variation of thermal static stability plays an important role in dust mobilization. This factor is included in the calculation of friction velocity and taken into account in the u*-dependent dust flux formula (2). Since u* cannot be verified directly by observation, we compare the modeled surface high wind areas with the observed high winds (Figure 5) at the time when dust flux reaches its maximum value. The choice of 8 m s−1 as the minimum strong wind speed in this figure is based on the wind thresholds for the wind-driven dust flux formulas used by Uno et al. [2001] and Tegen and Fung [1994, 1995] in their dust modeling. The modeled and observed high wind speed areas are quite consistent at most times and locations suggesting that our model has correctly predicted dust mobilization.

Figure 5.

COAMPS 12-hour forecast of surface wind speed (m s−1, line contours) and observed surface wind speed (shaded contours; winds greater or equal to 8 m s−1) for (a) 6, (b) 7, (c) 8, and (d) 9 April.

[21] In the remainder of the simulation, i.e., from 10 to 15 April, no large-area systematic dust generation is found in our modeling or in the station observations. This is due to the presence of high-pressure systems over Mongolia and most of China for the days following the passage of cyclones and fronts, thus reducing strength of winds in the dust source areas. Nevertheless, some small-scale dust mobilization is seen every afternoon (around 0600 UTC) in the southern Gobi Desert and the eastern Taklamakan Desert, and on 10 April, in a small region of Liaoning Province in northeast China (not shown). The daily maximum fluxes are less than 2.0 (mg m−1 s−1), with the exception of 3.5 seen on 12 April. This dust production is much less than that of 6 to 9 April.

4. Dust Transport Analysis

[22] In the previous section, we found that the first Mongolian cyclone on 6 and 7 April, and the cold front on 8 and 9 April accompanied by a second Mongolian low system, were the major dynamic features that mobilized dust during this study period. The first Mongolia low is readily apparent in the SeaWiFS image (Figure 6b) as a spiral band of clouds over eastern Mongolia and China. Heavy amounts of dust are present between the cloud bands as well as over the Taklamakan, and Gobi Deserts. For comparison we show the distribution of dust mass load (vertical integral) at 0000 UTC on 7 April, as well as the forecasted 700-mb geopotential heights, in Figure 6a. The observed and simulated dust features are generally aligned with the 700-mb flow except east and north of the low. The vertical structure of dust mass distribution is illustrated in Figure 7 in four horizontal slices at the surface and heights of 2-km, 4-km, and 6-km, together with the streamlines at the corresponding heights. Nearest the cyclone, dust mass is transported cyclonically around its center, with the plume nearer the center at higher altitudes. At middle to high altitudes and east and north of the low, dust is transported anticyclonically to the northeast and then east. Figure 8 shows a vertical cross section of dust mass and streamlines at 0000 UTC of 7 April along longitude 122.5°E, or east of the low and intersecting the dust plume at 42°N and 50°N. It shows two distinct elevated maximums where it intersects the plume, at 42°N and 50°N, the first between 1 and 3 km in the region of expected strong upward motion. The maximum at 50°N is elevated about 2 km higher than the one at 42°N and vertical motions are weaker. This height difference results from the vertical transport of dust as it passed through the cyclonic system.

Figure 6.

(a) COAMPS modeled dust mass load (vertical integral; mg m−2) and the 700-mb geopotential height (m) valid at 00Z, 7 April. (b) NASA GSFC SeaWiFs satellite image for 7 April.

Figure 7.

Modeled dust mass concentration (mg m−3) and the streamlines at (a) the surface, (b) 2 km, (c) 4 km, and (d) 6 km constant heights, valid at 00Z 7 April.

Figure 8.

Vertical cross section of modeled dust mass concentration (mg m−3) and streamlines along 122.5°E from 35°N to 55°N valid at 00Z 7 April.

[23] The second Mongolia cyclone on 9 April plays a similar role in dust transport. Figure 9a shows the simulated optical depth and 700-mb geopotential height at 1200 UTC for 9 April. In this case, the dust plume is entrained completely around the cyclone leaving a clear eye in the center. The dust plume distribution generally resembles TOMS aerosol index on 9 April (Figure 9b) Both the modeled and observed optical fields indicate high dust concentrations in the plumes over and near the desert areas and around the cyclone. Four horizontal slices of dust concentration at surface, and at 2-km, 4-km and 6-km heights, with corresponding streamlines, are presented in Figure 10 showing the vertical structure of this dust plume. Near the surface, the strong westerly winds behind the cold front over the central China mobilize and transport massive dust plumes eastward, away from the deserts. At higher altitudes, a complicated pattern emerges. West of the low, a cyclonic structure of the plume appears with the dust returning to the source region in the northeasterly flow. East of the cyclone, there is anticyclonic transport to the east toward Japan. Figure 11 shows the vertical cross section of dust mass and streamlines from 1200 UTC on 9 April along 48°N. It intersects the major dust features in three places. Through comparison with Figure 9a, it is obvious that the plume between 105°E and 110°E in Figure 11 is the returning branch the west of the cyclone. The plume between 115°E and 120°E is the northbound branch east of the low. Both plumes are cyclonically curved (Figures 10c–10d) and still subject to model-resolved upward motion. Peak concentrations are at about 5 or 6 km. The plume located between 125E and 135E is within the anticyclonic eastward transport, which, as indicated by the streamlines, is dynamically associated with the prevailing sinking motion. The maximum center of the plume was elevated to 7.5 km while passing through the cyclone.

Figure 9.

(a) COAMPS modeled optical depth and the 700-mb geopotential height valid at 12Z 9 April. (b) TOMS aerosol index of east Asia dust on 9 April (provided by N. C. Hsu). The apparent gap at 120°E is due to a data void.

Figure 10.

Same as Figure 7 but valid at 12Z 9 April.

Figure 11.

Vertical cross section of modeled dust mass concentration (mg m−3) and the streamlines along 48°N from 100°E to 145°E valid at 12Z 9 April.

[24] In order to give a more detailed quantitative analysis of transport dynamics, the dust mass continuity equation of the model is examined for the contributions to the mass tendency from each individual dynamical and physical process. The mass tendency can be described from equation (1) in a generalized form such that:

equation image

The prognostics of this equation are accumulated over each hour and at each model level during the model run and used to reveal the hourly averaged mass tendency.

[25] For the first cyclone on 6 and 7 April, we apply (6) at a point in the source area at 105°E and 42°N in the southern Gobi Desert at the time of dust-flux peak 0600 UTC on 6 April (see Figures 3a and 4a). The dust mobilization conditions around this location are quite uniform in topography, so no horizontal averaging over the area is necessary. The contributions to the time tendency from all the terms above (except dry deposition) are shown in Figure 12a. It is obvious that in the source area, the turbulence in the convective boundary layer contributes the most to the time tendency. The second most important factor is the horizontal advection (negative values) that moves newly generated dust out of the source area. The vertical advection by the model-resolved motion is small. Sedimentation reaches a maximum near the surface as large particles quickly fall out soon after uplifting. The turbulent mixing reaches its negative maximum near the surface, mainly balancing the sedimentation. There is no scavenging in the source area at this time. Figure 12b shows the vertical profile of hourly averaged concentration at the same location, showing that the dust mass is confined within the boundary layer and it decreases with height. There is a maximum above the surface that results from the net effect of turbulent mixing and sedimentation.

Figure 12.

(a) Model calculated dust tendency (mg m−3 hour−1) of horizontal advection, vertical advection, turbulent mixing, scavenging, and sedimentation at a source location (105°E, 42°N) at 06Z 6 April. (b) Modeled vertical profile of hourly averaged dust concentration (mg m−3) at a source location (105°E, 42°N) at 06Z 6 April.

[26] An area downstream of the cyclone, between 120–130°E and 40–50°N (Figure 6a), is chosen to be the area for a quantitative analysis of transport at 0000 UTC on 7 April. The area-averaged values of time tendency are plotted in Figure 13a. The vertical transport by the model-resolved motion, i.e., the cyclone's vertical component, becomes a dominant contributor to the time tendency in addition to the horizontal advection. The distribution, with negative values below 3 km and positive values between 3 and 7 km in the vertical advection term, indicates the systematic motion of the cyclone is moving the dust from lower layers to upper layers. The vertical profile of mass concentration for the same averaging area at the same time (Figure 13b) shows an elevated dust layer with a maximum located around 3-km elevation as a result of this vertical advection. Turbulent mixing in Figure 13a is relatively small in the planetary boundary layer, but reaches a maximum near the surface due to downward diffusion. Sedimentation decreases with height as fewer larger particles remain. There is a slight maximum in the scavenging at 3-km height, a negative in time tendency that removes dust mass directly from the air.

Figure 13.

(a) Same as Figure 12a but averaged in a downstream area of 120–130°E and 40–50°N at 00Z 7 April. (b) Same as Figure 12b but averaged in a downstream area of 120–130°E and 40–50°N at 00Z 7 April. Note the change in the x axis scale.

[27] To investigate the dust transport by the second cyclone system on 9 April, the same location used in Figure 12a is selected to examine the time tendency for the upstream source area. The output time is also at 0600 UTC for 9 April, the maximum dust flux period of the day due to the strong pressure gradient behind the cold front (Figure 3d). Figure 14a shows the time tendency of all the terms at this source point. Similar to Figure 12a, nearly all the transport occurs below 3 km, where the local convective boundary layer is fully developed. Turbulence is the dominant forcing in the mass vertical redistribution. It reaches a negative maximum near the surface to balance the sedimentation. The negative horizontal advection means dust moving downstream out of the source area. The noticeable vertical advection, with a distribution of negative tendencies at lower elevation and positive tendencies at higher elevation, indicates the model-resolved vertical motion associated with the cold front system also transport dust upward, contributing to vertical mass redistribution, a feature not seen on 6 April. Figure 14b shows the vertical dust profile at the same location is well mixed in the boundary layer, which is due to both turbulent convection and the systematic lifting by the cold front. The profile on 9 April looks quite different from the one on 6 April (Figure 12b) when vertical advection was not significant.

Figure 14.

(a) Model calculated dust tendency (mg m−3 hour−1) of horizontal advection, vertical advection, turbulent mixing, scavenging, and sedimentation at a source location (105°E, 42°N) at 06Z 9 April. (b) Modeled vertical profile of hourly averaged dust concentration (mg m−3) at a source location (105°E, 42°N) at 06Z 9 April. Note the change in the x axis scale.

[28] The downstream transport area for the investigation is chosen between 112–122°E and 40–50°N, near the cyclone center in Figure 9a. The area-averaged time tendency is presented in Figure 15a. The vertical and horizontal advections dominate, except near the surface where turbulent mixing reaches a maximum, and both advection processes are active from 1.5 km up to an altitude of 9 km. Large vertical transport by the cyclone vertical component results in a deep elevated dust plume, as shown in Figure 15b: the dust plume extends upward to 9 km elevation. This is also consistent with the upward streamlines displayed in Figure 11 between 110°E and 125°E. The vertical profile also displays two elevated maximums at 2 km and 4.5 km. These two heights are where the maximum horizontal advection occur in Figure 15a, meaning upstream mass is being transported into the area. The scavenging in the second cyclone is slightly larger than the first one (Figure 13a).

Figure 15.

(a) Same as Figure 14a but averaged in a downstream area of 112–122°E and 40–50°N at 12Z 9 April. (b) Same as Figure 14b but averaged in a downstream area of 112–122°E and 40–50°N at 12Z 9 April. Note the change in the x axis scale.

[29] Through above the diagnostic and prognostic analyses of dust transport, we can see the major factors controlling the transport in the source and downstream areas. In the upstream source area, turbulence in the convective boundary layer plays the major role in the vertical mass redistribution, mixing dust upward to the top of the planetary boundary layer, though the cold front also contributes to the vertical transport. In the downstream cyclone area, dust is entrained into the cyclonic structure and transported to altitudes as high as 6 to 9 km. At the edge of cyclone, dust is transported anticyclonically, and sinking, to the east. Therefore the large-scale dynamic forcing that uplifts dust from the boundary layer to the free atmosphere is provided by the two cyclone systems during their transport across the east Asian continent from 6 to 10 April.

5. Dust Transport Validation

[30] Modeling validation is critical to the application of a numerical model, and to the credibility of numerical simulation studies. In the above transport analyses, the comparison of the modeled dust plume in Figure 6a with the satellite image of Figure 6b, and the comparison of modeled optical depth in Figure 9a with the TOMS aerosol index of Figure 9b show the close resemblances of dust plume patterns at two separate times. The observations and measurements at the five Chinese and Japanese cities marked in Figure 1 will be used in this section for further quantitative model validation.

[31] Lanzhou is located at 103.5°E and 36.0°N, in the northwest of China. It is very close to the source area, being near the southern edge of the Gobi Desert and to the west of the Taklamakan Desert (see Figure 1). The first dust storm arrived in Lanzhou on 6 April, and the second one arrived on 8 April and continues to the early hours of 9 April. Figure 16 shows both the observed surface PM10 concentration provided by Lanzhou Environmental Monitoring Station and the model output of surface PM10 at Lanzhou from 6 to 9 April. The model catches the two major dust events, showing the storm on 8 April being stronger than the one on 6 April, and predicts the minor peak following the main peak in PM10 on 8 April. The observed elevated PM10 concentration between the two dust events may be caused by local air pollution, since Lanzhou has poor air quality due to air pollution the model is not yet able to predict. The model simulates the correct timing of dust events, and predicts comparable event strengths (e.g., the maxima of PM10) at Lanzhou.

Figure 16.

COAMPS modeled surface PM10 concentration (mg m−3) at Lanzhou from 6 to 9 April compared with the observed PM10 (provided by Lanzhou Environmental Monitoring Station).

[32] During the ACE-Asia international experiment in the Spring 2001, the Asian Lidar Observation Network conducted a series of daily observations of dust, with measurements of lidar backscattering intensity, depolarization ratios, extinction coefficients and other parameters. Beijing, one of the lidar sites, is located at 116.3°E and 39.9°N, some distance downwind from the deserts in the east (see Figure 1). The city was affected by the dust storms from 6 to 11 April. Figure 17a shows the time-height cross section of range-corrected backscattering intensity at a wavelength 532 nm, which detects most types of aerosol particles in the air. During the dust events, dust is the dominant aerosol in Beijing. The modeled dust passing through Beijing is plotted in Figure 17b as a time-height cross section of mass concentration. Both the modeled and measured cross sections have very similar features in dust distribution. The first dust storm hit the city from 6 to 8 April, with the dust plume arriving in an elevated layer having a maximum center at 1.5 km and then descending over time until the lower edge of the plume touched the ground. The second dust storm arrived on 9 and 10 April, with the plume appearing in a deep layer and its top reaching 5–6 km altitude. A small episode immediately followed the second strong storm, arriving around 11 April. The lidar data in Figure 17a shows strong backscattering in the boundary layer on 10 and 11 April, which is due to the occurrence of local air pollution that enhanced the lidar intensity. The model predicts both the right timing and comparable plume depth for dust events at Beijing.

Figure 17.

Time-height cross sections of (a) lidar range-corrected backscattering intensity at 532 nm (provided by Asian Lidar Observation Network) and (b) COAMPS modeled dust mass concentration (mg m−3) at Beijing from 6 to 11 April.

[33] Tsukuba and Nagasaki also experienced dusty air pollution when Asian dust plumes crossed the China Yellow Sea and the Japan Sea to the Far East. Tsukuba is at 140.1°E and 36.1°N in eastern Japan, and Nagasaki at 129.9°E and 32.8°E in southern Japan (Figure 1). Figure 18a shows the lidar depolarization ratio at Tsukuba from 9 to 13 April, and Figure 18b shows the model output of dust mass concentration over the same period. Depolarization ratio indicates the presence of nonspherical aerosol particles that mainly are dust, soot, sea salt and ice crystals in the air. Both plots in Figure 18 show an elevated dust plume arriving at Tsukuba on 10 April and continuing until 11 April. The plume appears in a vertical range of 2 km to 7 km elevation. A second small plume arrives on 12 April, appearing in both the boundary layer and free atmosphere for a short period. Figure 19 shows the depolarization ratio (a) and the modeled dust concentration (b) from 10 to 14 April at Nagasaki. It can be seen from the two figures that the model predicts the major dust events affecting the city: an elevated dust plume passing through from 10 to 11 April having an approximately 4-km deep layer between 3 km and 8 km altitudes, and a low-level dust plume appearing in the planetary boundary layer on 12 and 13 April. Some sparse elevated dust plumes also show up on 13 and 14 April in the model simulation, while the lidar shows more continuous elevated dust plume. The lidar detects the boundary layer of aerosol from 12 to 14 April, about 24 hours longer than the model predicts. The discrepancies are either due to the appearance of nondust aerosol particles (e.g., smog) in the lidar measurement or due to the modeling deviations in dust-plume path or in dust concentration at Nagasaki.

Figure 18.

Time-height cross sections of (a) lidar depolarization ratio at 532 nm (provided by Asian Lidar Observation Network) and (b) COAMPS modeled dust mass concentration (mg m−3) at Tsukuba from 9 to 14 April.

Figure 19.

Time-height cross sections of (a) lidar depolarization ratio at 532 nm (provided by Asian Lidar Observation Network) and (b) COAMPS modeled dust mass concentration (mg m−3) at Nagasaki from 10 to 15 April.

[34] The last city used in this model validation is Hefei, located at 117.2°E and 31.9°N, southeast of the deserts (Figure 1). In the model simulation period, the dust plumes first are transported eastward, following the cyclones' movements from 6 to 9 April (Figures 7 and 11), then are pushed southeastward by the northwesterly winds after a deep and wide trough is established over East China and Japan from 10 to 14 April (not shown). Thus Hefei was affected by dust plumes on 11 and 14 April, even though the city is much closer to the dust source area than Tsukuba and Nagasaki. Figures 20 and 21 demonstrate two comparisons of the model calculated extinction coefficient of dust and the lidar measured extinction coefficient for 11 and 14 April, respectively. (The lidar data is provided by China Anhui Institute of Optics and Fine Mechanics.) On the 11 April of Figure 20, the model predicts a boundary layer of dust plume, whose depth and magnitude of extinction coefficient are equivalent to the observation. Above the boundary layer, the model predicts two shallow elevated dust layers that can also be seen in the observation, though the modeled ones are slightly larger than the observed. The observations also show a layer of large extinction at high altitudes between 8.5 km and 13 km in Figure 20b, resulting from cirrus clouds. On 14 April of Figure 21, the model predicts an elevated dust plume layer in the middle of the troposphere between 3 km and 8 km, which matches the observation very well in height and magnitude for each output time. The lidar also detected some other air pollution in the lower boundary layer in Figure 21b, which this model currently does not model.

Figure 20.

(a) COAMPS modeled extinction coefficient of dust (km−1) and (b) lidar-observed extinction coefficient at Hefei at marked times of 11 April.

Figure 21.

Same as Figure 20 but for 14 April.

[35] Through the validations of transport modeling at these five cities and with satellite remote sensing, it is shown that this model simulates the right timing and generates comparable strengths of dust events. It predicts the boundary layer and elevated layers of dust plumes in depths and magnitudes comparable to the observed values. It is noted that the closer to the dust source area, the better the model validation proves in the above comparisons. The good transport validation justifies the transport analyses of diagnostics and prognostics in the previous section, and will support the reliability of the mass budget calculation in the next section.

6. Mass Budget Calculation

[36] In order to estimate the impact of strong dust storms on the regional and global environment from a period of continuous and intense dust production, the dust mass budget is calculated throughout the simulation period from 5 to 15 April. Two groups of particles, PM10 and PM2.5, are considered air pollution problems for the environment and human health. The model results reveal that particles larger than PM10 mostly drop off in and near the dust source area, and that the mass concentration for all particles in long-range transport out of the deserts is very close to the PM10 concentration. Therefore we primarily consider PM10 and PM2.5 in this mass budget calculation. We calculate daily dust production, wet deposition and dry deposition (including sedimentation) onto the land and into the ocean, and the mass outflow at the northeast and east boundaries which is subject to transport to the Pacific and beyond.

[37] Figure 22 shows the daily dust production of PM10 and PM2.5 from 6 to 15 April. It is obvious that large dust mobilization occurs from 6 to 9 April, with the largest mobilization on 8 April, which is why we focus on these days in the discussion of dust production and transport analyses in sections 3 and 4. As expected, the production of PM2.5 is a small fraction of PM10, but PM2.5 particles undergo long-range transport and are more often found in the Pacific and North America [Tratt et al., 2001; Husar et al., 2001; McKendry et al., 2001]. The total dust production throughout the simulation period is listed in Table 1, showing that total mass mobilized for all particles (less than or equal to 36 μm) is 643 megatons, while PM10 is 552 megatons and PM2.5 is 113 megatons.

Figure 22.

Modeled daily dust production of PM10 and PM2.5 particles (megatons) integrated over all dust source areas from 6 to 15 April.

Table 1. Mass Budget Calculation (From 5 to 15 April) of Dust Production and Removal Processes (in Megatons) and Percentage of Removal Processes to the Production of Each Particle Group, Integrated to the End of the Study Period for All Particles (Less Than 36 μm) and Particles of PM10 and PM2.5
Particle GroupsAll Particles, <36 μmPM10PM2.5
Total mass production643552113
Dry deposition on deserts only378 (58.8%)314 (56.8%)61 (53.7%)
Wet removal on deserts only102 (15.9%)92 (16.6%)20 (17.3%)
Dry deposition on land outside of deserts72 (11.2%)58 (10.5%)72 (9.8%)
Wet removal on land outside of deserts53 (8.3%)50 (9.1%)11 (9.9%)
Dry deposition on ocean only1.9 (0.3%)1.7 (0.3%)0.2 (0.2%)
Wet removal on ocean only8.4 (1.3%)8.3 (1.5%)1.8 (1.6%)
East boundary outflow23 (3.6%)23 (4.2%)7.5 (6.6%)

[38] There are no individual dust cases in the literature that provide dust production values, so we have no references that we could directly compare with our case. However, there are quite a few estimates of global-scale annual and seasonal dust production rates through modeling and diagnostic analyses [Andreae, 1995; Duce, 1995; Tegen and Fung, 1995; Dentener et al., 1996; Mahowald et al., 1999; Ginoux et al., 2001]. There are also several annual dust emission rates specifically estimated for China deserts only. By calculating the atmospheric dust deposition to five Asian/Pacific regions at some sampling sites, and assuming homogeneous deposition in each region as well as total deposition equal to production, Zhang et al. [1997] derived an annual dust production of 800 megatons (ranging from 500–1100 megatons) in China deserts, which include Taklamakan Desert and Gobi Desert in Inner Mongolia. On the basis of climatological data, land surface characteristics, annual mean winds and precipitation rates, and some other dust emission associated factors, Xuan [1999] and Xuan et al. [2000] used U.S. EPA dust emission formulas to derive annual dust production in China deserts of 43 megatons of TSP50, and 25 megaton of TSP30. Both studies found that springtime is the worst dust-emitting season, producing more than half of the annual emission. Our modeled dust production for the case study period is smaller than, but of the same order, as the annual rate from Zhang et al., but, is an order of magnitude larger than the rate given by Xuan [1999] and Xuan et al. [2000]. The case in our study is an 11-day period (5–15 April) of intense dust storms. Viewing the meteorological data records in East Asia, we found this period to be the worst in the season and the worst of the year so it could be an extreme event in dust history.

[39] Figures 23a and 23b shows time series of PM10 dry deposition and wet removal on land and ocean, and the boundary outflow in the east and the northeast, with values representing accumulations in time and in the percentage of PM10 dust production. It is obvious that most of the dust production redeposits directly to the deserts, with dry deposition taking more than 50% and wet removal taking nearly 20%. Dry deposition on land, other than the deserts, removes the third largest portion of the production, totaling more than 8% through the whole period. Wet deposition on land is the fourth largest removal process, after dry deposition, totaling more than 5% over the most period. The other terms, dry and wet deposition on ocean and the east boundary outflow, are all smaller than the deposition on land, all being less than 5%, with the dry deposition on ocean being the least, because dust plumes are mostly elevated over the ocean which is far away from the deserts.

Figure 23.

(a) Modeled dry and wet deposition on desert and nondesert land as percentage of total PM10 production and accumulated in time. (b) Percentage of modeled dry and wet deposition on ocean and the east boundary outflow from 5 to 15 April. (Different vertical scales used in Figures 23a and 23b.)

[40] Figures 24a and 24b shows the time series of PM2.5's removal processes. PM2.5 has proportional removal allocations that are similar to those of PM10 in Figure 23. However, there are two major differences: PM2.5 has a larger percentage of east boundary outflow, but a smaller percentage in dry deposition on land and ocean than the PM10. The differences of percentage increase in time, so that the fraction of PM2.5 within PM10 increases in time, especially in the east boundary outflow. This is not surprising: the smaller particles have longer lifetime cycles than the larger ones and are more subject to long-range transport than the large ones. Another difference is that PM2.5 wet removal is equivalent to or even slightly larger than the dry deposition on land, whereas PM10 wet removal is consistently smaller than the dry deposition on land. This means that small particles are less subject to dry deposition because of less gravitational settling than large particles. These differences can also be seen in Table 1, which lists the total budget values of these removal processes at the end of the simulation period for all particle groups.

Figure 24.

Same as Figure 23 but for PM2.5 particles.

[41] The results of high percentages of dry and wet removal processes over land in Table 1 and Figures 23 and 24 are consistent with the other mass budget studies for global and regional ranges in annual and seasonal scales by Tegen and Fung [1995], Zhang et al. [1997] and Ginoux et al. [2001]. About 75% of all mobilized dust particles is redeposited at the source area, a value that is much higher than the estimate of 30% quoted by Zhang et al. [1997] for their annual dust budget rates. One explanation for the difference arises from two different approaches. In this study, we conduct a numerical case simulation having explicit dynamics and microphysics and predicting complete dust life cycles for an 11-day period of strong dust episode. In Zhang's study, the value 30% was obtained from a climatological estimation in which dust production was made equal to deposition, and the study covered a 4-year period from 1991 to 1994. Therefore it is the mean annual representation of dust.

7. Summary

[42] In this dust case study, we describe the U.S. Navy's COAMPS™ mesoscale dust aerosol model and use the model to conduct dust simulations for the period of 5–15 April 2001. The dust model is an online module of COAMPS™ and makes full use of the high-resolution meteorological dynamics at each time step. The model performance is well verified for dust transport at different locations, from near the dust sources to long-range downwind areas, through the observations of PM10, lidar and satellite products. It is also verified with meteorological observations for winds and dust mobilization. The model is proven to simulate the right timing and comparable strength of dust events. It predicts the boundary layer and elevated layer of dust plumes with depths and magnitudes equivalent to the observed values. The good modeling validation justifies the transport analyses and mass budget calculations presented in this dust case study.

[43] The dynamics of dust transport is analyzed for this strong dust episode using the diagnostics of the three-dimensional distribution of mass concentration and winds, and the prognostics of the dust mass continuity equation with details of individual dynamical and microphysical time-tendency terms. We find that in the upstream dust source area, mechanical and convective turbulence in the planetary boundary layer plays the major role in vertical dust mass redistribution, mixing dust upward to the top of the boundary layer, and that the cold front (on 8 and 9 April) also contributes to the vertical transport when it passes over the deserts. In the area downstream of cyclones, dust is entrained into the cyclone and transported to altitudes as high as 8–9 km and into the westerlies, becoming subject to long-range transport. At the edge of cyclones, dust is transported anticyclonically to the east and northeast. It is the cyclone systems (the first on 6 and 7 April, the second on 8 and 9 April) that provide the large-scale dynamic forcing to lift the dust from the planetary boundary layer to the high-up free atmosphere during their movement across the east Asian continent. It appears likely that in the absence of strong cyclone systems, little dust would be lifted high in the atmosphere and available for long-range transport.

[44] The dust mass budget is calculated to estimate the impact of such an extremely strong dust episode on the regional and global environment. The budget reveals that the total dust production for all particles (diameters less than 36 μm) is about 643 megatons over the period of 5 to 15 April, most of which is generated from 6 to 9 April. This total is smaller than, but in the same order as the annual dust production rate, 800 megatons, estimated by Zhang et al. [1997]. Of the total dust production, about 75% is redeposited on the deserts, 20% deposited on land other than deserts, 1.6% falls into the ocean, and 3.6% (mostly suspended particles) crosses the east and northeast model boundaries. The boundary dust outflow is dominated by PM10, and is subject to the transport over the Pacific and beyond. Since fine particles of PM2.5 have longer lifetime than the large PM10 particles, proportionally less PM2.5 is removed than PM10 during the long-range transport, and the fraction of PM2.5 in the boundary outflow increases with time.

Acknowledgments

[45] We are very grateful to Christina N. C. Hsu for her generosity of providing the TOMS aerosol index plot, and to Krzysztof M. Markowicz for his work on converting lidar data formats. We are very appreciative of the valuable discussion with Piotr J. Flatau, Betsy E. Raid and Annette L. Walker. The support of sponsors, the U. S. Office of Naval Research, and the Naval Research Laboratory through program PE-0602435N, as well as the support from China National Basic Research Project G2000048703 are gratefully acknowledged. This work is also supported in part by a grant of HPC time from the U.S. Department of Defense Shared Resource Center.

Ancillary

Advertisement