Download pdf - A numerical simulation of dust storms in China

Environmental Modelling & Software 19 (2004) 141–151www.elsevier.com/locate/envsoft

A numerical simulation of dust storms in China

Zhenxin Song∗

The Atmospheric Science Department of Lanzhou University, Lanzhou, 730000, People’s Republic of China

Received 30 September 2002; received in revised form 20 January 2003; accepted 18 February 2003

Abstract

Wind erosion occurs in many arid, semiarid and agricultural areas of the world. The desert areas of China, which occupyapproximately 13% of China’s total surface area, are major sources of Asian dust. The major wind-erosion areas are the sandylands in western and northwestern China together with the extensive regions of the Gobi desert in northern and northeastern China,especially along the basin of the Yellow River. In this paper, dust storms which occurred in China in the spring of 2002 weresimulated using an integrated numerical modeling system.

The purpose of the simulation is to produce quantitative predictions of wind erosion on regional scales. The integrated winderosion modeling system used in this study coupled the following three major components: (1) An atmospheric prediction model,together with a land-surface model; (2) a wind-erosion model and (3) a geographic information database. The atmospheric modelprovides the necessary input data for the wind erosion scheme, including wind speed and precipitation. It also provides input datafor the land-surface model that produces predictions for soil moisture. Dust transport and deposition are also considered in theatmospheric model. The wind-erosion model predicts streamwise saltation and dust emission rate for given atmospheric, soil andland surface conditions. The geographic information database provides spatially distributed parameters, such as soil type and veg-etation coverage, for the atmospheric, land surface and wind erosion models.

Dust storms in China occur mainly in spring and winter, but most frequently in April. In spring, surface soils frozen in theprevious winter become especially loose, creating a favorable condition for wind erosion. As an example, the severe dust stormsof 15–20 March were simulated. The results show the integrated modeling system can simulate the main characteristics of the duststorms. The system produced estimates of wind erosion intensity and patterns that are in agreement with observations. Such asystem offers the possibility of determining wind erosion patterns on broad scales with high spatial resolution, as well as dusttransport and deposition. 2003 Elsevier Ltd. All rights reserved.

Keywords: Wind erosion; Dust storm; Numerical simulation; Integrated modeling system

1. Introduction

Wind erosion is a serious environmental problem inarid and semi-arid regions of China and in many otherparts of the world. Strong wind erosion events, such assevere dust storms, may threaten human lives and causesubstantial economic damage. The northwestern Chinaregion is one part of the central Asia dust storm area.The desert areas of China, which occupy approximately13% of China’s total surface areas, are major sources ofAsian dust. These areas include the temperate arid land

∗ Present address: National Meteorological Centre, Zhong GuancunSouth Street 46, Beijing 100081, China. Tel.:+86-10-6840-7469; fax:+86-10-6840-8584.

E-mail address: [email protected] (Z. Song).

1364-8152/$ - see front matter 2003 Elsevier Ltd. All rights reserved.doi:10.1016/S1364-8152(03)00116-6

from 75°E to 125°E and from 35°N to 50°N (Liu, 1985).The major wind erosion areas are sandy lands in westernand northwestern China together with the extensiveregions of the Gobi desert in northern and northeasternChina, especially along the basin of the Yellow River(Liu, 1985; Walker, 1982). The dust storms occurring inthe north part of China and Mongolia are called EastAsian dust weather. Recently, dust storms occurred fre-quently in the spring in China, which caused the wideattention of the public and the government. Wind erosionis an environmental process influenced by geological andclimatic variations as well as human activities. It occurswhen a soil surface is unprotected by vegetation coverand sufficiently dry, under such conditions, wind is ableto pick up sand sized particles, which bounce along thesurface and eject more particles, including dust particles.

142 Z. Song / Environmental Modelling & Software 19 (2004) 141–151

Those dust particles, which usually contain most of theorganic matter and nutrients, may be carried a long dis-tance by the wind, notably as dust storm or dust hazes.It reduces soil productivity and leads to land degra-dation. Wind erosion causes loss to public utilities. Forinstance, dust suspension reduces visibility, sandblastingdestroys young crops, and dust related air pollutioncauses a health hazard, etc. Hence, the simulation andforecast of a dust storm is not only important for longterm sustainable agriculture but also has significanteconomic benefits.

Considerable insight has been gained into the physicsof wind erosion since Bagnold published his pioneerwork The Physics of Blown Sand and Desert Dunes in1941 (Bagnold, 1941). Suspension, saltation, and creepare the three distinct modes which occur during winderosion (Bagnold, 1941). Shao (2000) treats the physicsof wind erosion rigorously from the viewpoint of fluiddynamics and soil physics. The purpose of developing awind erosion modeling system is to produce a quantitat-ive prediction of wind erosion on scales from paddockto global. The system should have the capacity of mode-ling the complete wind erosion process, from particleentrainment through transport to deposition. It is a formi-dable task because wind erosion is governed by a widerange of factors involving atmospheric conditions, soilstates and surface properities. A lot of progress on thesimulation of dust weather has made. The first attempt tocombine the information of atmospheric data with land-surface data for wind erosion assessment was made byGillette and Hanson (1989) in their investigation of thespatial and temporal variations of dust production in theUnited States. In atmospheric studies, dust emission andtransport have been under research since the late eighties(e.g. Westphal et al., 1988; Tegen and Fung, 1994,1995). However, in most of these studies, crude winderosion schemes and coarse land surface data were used,which limited the reliability of the modeling results.Marticorena and Bergametti (1995); Shao et al. (1996)and Marticorena et al. (1997) have developed betterwind erosion schemes which account for the impact ofsurface properties on sand drift and dust emission. Shaoand Leslie (1997) and Lu and Shao (2001) havedeveloped and implemented an almost fully integratedwind erosion modeling and prediction system.

In the spring of 2002, a research group was estab-lished in CMA (Chinese MeteorologicalAdministration). Members of the group come from NMC(National Meteorological Centre), NSMC (National Sat-ellite Meteorological Centre), IAP CAS (Institute ofAtmospheric Physics, Chinese Academy of Sciences)and IGE CAS (Institute of Geography, Chinese Acad-emy of Sciences). The Group used an integrated winderosion modeling system developed by Shao and Li(1999 and Shao and Lu, 2000), land surface data andGIS data to make real time forecast of dust storms that

occurred in China from March to May in 2002. Duringthese periods, NMC provided numerical forecasts pro-ducts on dust weather every day. It is the first real timeforecast of dust weather in China. In this paper, wereport the basic facts on dust simulation and predictionin China in the spring of 2002. At the same time themodel results are compared with observation images.

2. An integrated wind erosion prediction system

2.1. System structure

The framework of an integrated wind erosion mode-ling system is as illustrated in Fig. 1. It is composed ofan atmospheric model, a land surface scheme, a winderosion scheme, a transport and deposition scheme anda geographic information database. The atmosphericmodel provides input data for other three model compo-nents. The land surface scheme simulates energy,momentum and mass exchanges between the atmos-phere, soil and vegetation, but more important in thecontext of wind erosion modeling, it produces the soilmoisture as an output. The wind erosion scheme obtainsfriction velocity from the atmospheric model, soil moist-ure from the land surface scheme and other spatially dis-tributed parameters from the GIS database. The winderosion scheme predicts streamwise saltation flux anddust emission rate for different particle-size groups. Thetransport and deposition model obtains flow velocity,turbulence data and precipitation from the atmosphericmodel and dust emission rate and particle-size infor-mation from the wind erosion scheme. Fig. 1 also illus-trates a possible computational procedure, the atmos-pheric model is first run after initialization for

Fig. 1. The structure of integrated wind erosion modeling systemconsisting of an atmospheric prediction model, a land surface model,wind erosion model, a transport and deposition scheme and a GIS data-base.

143Z. Song / Environmental Modelling & Software 19 (2004) 141–151

atmospheric dynamics and atmospheric physics. This isfollowed by running the land surface scheme and winderosion scheme. Finally, the calculation of dust transportand deposition is carried out.

2.2. Weather prediction model

The atmospheric model of the integrated system is ahigh resolution limited area weather prediction modeldeveloped at The University of New South Wales byLeslie and his colleagues (Leslie and Purser, 1991),referred to as HIRES (High Resolution Limited AreaModel). It is a primitive equation model on a Lambert-Conformal projection and utilizes the s coordinate withthe Arakawa C grid. The equation system used fornumerical weather prediction consists of seven basicequations for velocity components, the continuity equ-ation, the thermodynamic equation, the moisture equ-ation and the equation of state. As dust transport is alsoof concern, the dust concentration equation has beenadded to the equation system. The simulation area is30°E, 5°N to 180°E, 65°N with spatial resolution of 50km. The area of data analysis is 72°E, 5°N to 148°E,53°N. The atmospheric data required for HIRES initialis-ation and boundary conditions are derived from theT213-GCM of the China Meteorological Administration.In the vertical, the atmosphere is divided into 16 layers.An advanced soil moisture parameterization scheme hasbeen linked (Irannejad and Shao, 1998).

2.3. Wind erosion model

The wind erosion model comprises three key para-meterizations representing: (i) the erosion threshold fric-tion velocity u∗t , (ii) the streamwise sand flux Q, (iii)the dust emission flux F(i) for N size classes of dustparticles. The modeling of these processes is based,respectively, on a model of the wind erosion attenuationby roughness elements, the saltation model of Owen(1964). The main outputs from the wind erosion modelare threshold velocity u∗t (m/s), horizontal sand flux Q(of dimensions M L�2 T�1), and vertical dust flux F(g/m2s). The vertical dust flux F then become an inputas the dust source term in the dust transport model. Inour simulation and forecast six particle bins are used inthe model. A division of dust particles into different sizegroups has been proposed to be d�2 µm, 2 � d�11µm, 11 � d�22 µm, 22 � d�40 µm, 40 � d�80µm and d � 80 µm. It is assumed that particles sus-pended in the atmosphere are composed of N particlesize , each with a size di (i=1,…N). The discussion islimited to a single particle size, where the multi-particlecase can be reproduced by superimposing the single par-ticle situations. If the concentration of the ith particle isci = c(di), then the total concentration is

ctotal � �N

i=1

ci

. The transport model predicts atmospheric dust concen-tration by solving a continuity equation of dust writtenin the form of Eqs. (1) and (2). Writing the equations toa s coordinate, produces

∂psc(d)∂t

�∂psuc(d)

∂x�

∂psvc(d)∂y

�∂

∂sc(d)(pss

�

� grwt) � ps

∂∂x

Kphr∂c(d) /r

∂x� ps

∂∂y

Kphr∂c(d) /r

∂y(1)

�g2

ps

∂∂sKphr3

∂c(d) /r∂s

with boundary conditions

c(d)(pss�

� grwt)�g2

ps

Kpzr3∂c(d)∂s

� grF(d) at the surface

∂c(d) /r∂z

� 0 at the top

(2)

where c(d) is the concentration of dust particles of diam-eter d, wt is the settling velocity of particles (which is afunction of d), and F is the vertical dust flux; u, v, s�

and ps are wind velocity and surface pressure, respect-ively, and r is air density. The horizontal dust particlediffusivity Kph is assumed to be equal in the x and ydirections. The vertical dust particle diffusivity Kpz isassumed to be a function of the particle diameter d.

2.4. Dust emission model

Lu and Shao (1999) have proposed a dust-emissionmodel which, in contrast to energy-based models, esti-mates dust emission on the basis of the volume removedby impacting sand grains as they plough into the soilsurface. Also, in this model, saltation bombardment isconsidered to be the main mechanism for dust emission.In our simulation, dust emission model developed byShao (2001) was used. Three mechanism responsible fordust emission can be identified: (1) direct liftoff of dustparticles by aerodynamic forces; (2) release of dust par-ticles as saltating particles strike the surface causingabrasion; and (3) disintegration of dust coats on sandgrains and clay aggregates during saltation. The dustemission rate related to these three mechanisms can beformally expressed as

F � Fa � Fb � Fc (3)

where Fa is aerodynamic lift, which is insignificant ingeneral, because particles lifted by fa (aerodynamicforces) are weak in normal wind erosion conditions. Fb issaltation bombardment, which refers to striking particlesovercome fi (inter-particle binding forces) and result instrong emission. Fc is aggregates disintegration, whichmeans fine particles exist as aggregates. In weak events,


they behave as grains. While in strong events, they disin-tegrate.

2.5. Dry and wet deposition

Dust particles are delivered back to the surface byboth dry and wet deposition. Dry deposition is the dustflux from the atmosphere to the surface through molecu-lar and turbulent diffusion and gravitational settling,while wet deposition is the dust transfer to the surfacethrough precipitation. Dry deposition dust flux, Fd , canbe expressed as

Fd � �rwd[c(z)�c(0)] (4)

where c(0) and c(z) are, respectively, dust concentrationat the surface and at the reference level z and wd is thedry-deposition velocity. Raupach (1991) have proposeda single-layer dry-deposition model which is lessdemanding on data and parameterizations. In this model,the dry-deposition velocity is treated as a bulk single-layer conductance made up of three components actingin parallel

wd � �wt � gbb � gbm (5)

where gbb is molecular conductance and gbm is impactionconductance. Wet deposition is not considered in thiswork.

2.6. Input GIS (Geographic Information System) dataand model output

The stationary land surface parameters required forthe model are: the soil type index, vegetation type index,vegetation height, leaf area index (LAI) and land usestatus index. We needed to handle the GIS data beforethese data were used in the model system. There arethree steps in the pre-processor. The first step is to ana-lyze original GIS data. Most of the GIS parameters areregrouped into three categories: water, erodible and non-erodible. For a water surface, the parameter values willbe set as 0; if the soil index is equal to those non-erodiblesoil indices, the soil index will be re-set to 999. In othercases, the parameter value is kept as the original. Thesecond step is to calculate mapping enlargement factorR. Without considering the effect of soil moisture, thres-hold friction velocity can be treated as a stationary para-meter and pre-calculated if monthly or even short timeprediction is of interest. For each GIS grid, the enlarge-ment factor R is calculated. Finally the soil type index,enlargement factor, and the fraction of uncovered surfacearea need to be input into HIRES. Each Hires grid isdivided into several fractions according to the soil typeindex. The sub-areas (GIS grids) with same soil indexare added together regardless of their location within theHIRES grid.

GIS data is important in wind erosion simulation.

Wind erosion modelling requires spatial-distributed datafor soil and vegetation. The resolution of the GIS datais 5 km. Soils are normally divided into a number ofsoil classes. In our simulation, soils are divided into 30primary classes and many secondary classes. Althoughthe classification may not be directly useful for winderosion modelling, it provides the basic for furthermanipulation. Among the 30 soil classes, 10 are non-erodible soil (7 stabilized soils plus rocks, peats or salinelakes). The rest of the soil classes can be regrouped into11 USDA soil-texture classes, according to the descrip-tive information or to the particle-size analysis for eachprimary class. A particle-size distribution, both mini-mally-dispersed and fully-dispersed can be assigned foreach USDA soil texture class. Vegetation data provide arange of parameters such as vegetation height, fractionalvegetation cover and leaf-area index. From the veg-etation database, a reasonable estimate can be made ofquantities such as vegetation height and vegetation-coverfraction. The estimate of leaf-area index can draw on theremotely-sensed NDVI (Normalized Difference Veg-etation Index) data. We used the remotely-sensed NDVIdata from March to May in 2002 in our simulationexperiments.

All the variables including 3D variables (such aswind, pressure, and dust concentration fields) and 2Dsurface variables (such as vertical dust flux F, soil moist-ure w) are stored as binary access format. The basic dataunit is a 2D horizontal plane which is ‘sliced’ for everyvertical level. For a 2D variable, only one unit is used.For a 3D variable, sixteen (corresponding to the numberof vertical levels used in this study) units are necessary.For short time prediction, output is written in an hourlyinterval; for long time simulation, output is written in 6-hourly interval. In this study, we make a 72 h predictionand 3-hourly interval is used.

3. Results of the simulation

There are obvious difficulties in quantitative wind-erosion modeling, as both dust-emission rate and stream-wise saltation flux are sensitive to input data, such assoil moisture and frontal-area index, which are difficultto determine accurately. Nevertheless, wind erosionmodels developed recently have produced estimates ofwind erosion intensity and patterns which are reasonableagreement with observations (Marticorena and Berga-metti, 1995; Shao et al., 1996; Shao and Leslie, 1997and Lu and Shao, 2001). The integrated wind erosionmodeling system is nested with T213 global model inNMC/CMA, which provides the initial and boundarydata for the integrated wind erosion system. In theexperiment in spring of 2002, we developed the pre-pro-cess program to read T213 forecast database and trans-form them from isobar level to the atmospheric vertical


level. In dry climatic conditions, wind erosion risk inspring in China is high. As a consequence, a consider-able proportion of land surface had little protectiveannual vegetation and the land surface was susceptibleto wind erosion.

In this paper, the integrated wind erosion modelingsystem described above was applied to simulate the duststorms during March, April and May, 2002 over China.The integrated wind erosion prediction system simulatedthe total process of the dust storm and forecasted manyvariables of the atmosphere, soil and wind erosion,which can be used to describe the synoptic system, thedistribution and strength of the dust sources, the concen-tration, transport and deposition of dust storms.Throughout the simulation, Beijing local time is used. Itis 8 hours before Coordinated Universal Time (UTC)(Greenwich Mean Time).

3.1. Dust sources

The physical variable describing the sources of dustweather was written as F, which is the dust vertical fluxin the land surface, and the dimension of F is[ML�2T�1]. The variable F denotes the mass flux of dustper time and per area. For example, F also can be usedto illustrate how much dust was emitted during one dayand per square kilometer area. The temporal and spatialvariation of F also represents the temporal and spatialchange of dust sources. Fig. 2 gives such an example ofthe distribution of dust sources which arouse the duststorm weather of 15 March 2002. Fig. 2 shows thelocations of dust sources in Mongolia and the deserts ofthe north part of China. The integrated prediction systemforecast successfully the dust sources, which mainly cor-responded with the distribution of deserts. The BadainJuran desert and the Loess Plateau region to the west ofBeijing also are major dust sources.

Fig. 2. Dust sources distribution of the strong dust storm weather on15 March, 2002.

3.2. The content of dust in the atmosphere

The physical variable used to describe the content ofdust-sized particles in the atmosphere is written as C,which is the dust mass content per volume. The dimen-sion of C is [ML�3]. The variation of C is determinedby mass conservation equation, and is influenced byadvection, diffusion, wind erosion and deposition pro-cesses. All the physical process are associated with thesize of the particles, which were also divided into severalgroups, such as d � 2 µm, 2�d � 11 µm and 11�d� 22 µm etc. For each particle type, the content of dustin the atmosphere is calculated. Fig. 3 shows the contentof four different particle types in the near surface layer.As a example of the strong dust weather of 15 March2002, it is found that particles with a diameter smallerthan 11 µm have a similar distribution pattern. Themaximum content of particles with 11�d � 22 µm inthe atmosphere exceeds 300 µg/m3.

3.3. The deposition of dust

The physical variable of describing the dust depositionis written as D, which is the vertical flux of dust to theland surface. D has the same dimension as F, that is,[ML�2 T�1]. At the same time D represents the massflux of dust particle pre time and area. Fig. 4 shows thedry deposition from the strong dust events of 15 March2002. The location and pattern of dry deposition isfocused on the NNW part of China and Mongolia. Fromthe predicted dry deposition, we find the dust swept

Fig. 3. The dust concentration distribution of strong dust weather on15 March 2002, C1 represents concentration of particles with diameterd � 2 µm; but C2 and C3 represent particles of 2�d � 11 µm and11�d � 22 µm.


Fig. 4. Dry deposition distribution of the strong dust weather of 15March 2002.

through the desert regions of the northwestern and thenorthern China and were carried to the region of Korea.The modeling system successfully forecast the dry depo-sition of the dust.

3.4. The movement and the path of dust

An integrated wind erosion modeling system offersthe possibility of determining wind erosion patterns onbroad scales with high spatial resolution, as well as dusttransport. At the same time, the movement and the pathof dust under all kinds of weather conditions were alsocalculated and analyzed by using the integrated predic-tion system. In Fig. 5, we present an example of applyinga wind erosion modeling system to the prediction of the

Fig. 5. The development and movement of the strong windy weather.

dust storm events of 14–18 March 2002 over China.From the figure of s = 0.6 (about 600 hpa), on 14 Marchwe found that the dust began to occur in the boundaryregions between inner Mongolia and Gansu province andmoved towards the southeast with time. After 11 BSTof 15 March, most parts of the dust storm moved out ofChina and influenced Korea. In the afternoon of 15March, the dust storm strengthened in the original sourcearea. From the figure of s = 0.998 (about 1000 hpa), thesame situation can also be found. In the near surface,the dust followed the same process of developing, mov-ing and finally reaching Korea. Fig. 5 clearly shows thewhole pathway of the dust storm. We also believe theintegrated prediction system has the ability to forecastthe path of motion of the dust storm.

3.5. The spatial distribution of dust

The integrated prediction system can be used to ana-lyze the spatial structure of dust concentration. Fig. 6shows the distribution of dust concentration at differentvertical levels, but at the same time period. Fig. 6(a) isthe distribution of dust concentration at 400 hpa, on 14BST of 14 March. We can find at 400 hpa, the locationand structure of the concentration of dust storm matchedthat of 500 hpa. The values of the dust concentration athigher levels are lower than the concentrations at lowerlevels. Most of the dust lies in the near surface level.The concentrations of dust decrease as the heightsincrease. As heights increase, the location of highest dustconcentration moves towards the east. We also image thethree dimensional distribution of the dust concentration.


Fig. 6. The dust concentration distribution at different heights on 14 March 2002.

3.6. Columnar mass of dust

The columnar dust mass in the atmosphere can beexpressed as Mt = �Ctdz, and the dimension is [ML�2].Fig. 7 gives us an example of dust load on 15 March2002 in the vertical direction. The main parts of the dustload are located in the north and northwest of China.

Fig. 7. The columnar dust mass in the atmosphere.

Furthermore, the location of dust concentration movesout of China and extends to South Korea, Korea andJapan. The dust load is important to determine the con-tent of dust in the atmosphere.

4. Comparison with observation

4.1. The improvement of numerical simulation andprediction results

Northeast Asian dust storms were active betweenMarch and May 2002 and a severe event occurred on 19and 20 March 2002. We carried out intensive numericalexperiments using the integrated wind erosion modellingsystem and were able to successfully predict all majordust storm events during the period between March andMay. In our paper, we give an example of prediction ofa dust storm. These predictions are in excellent agree-ment with the surface station observations, demonstrat-ing the capacity of the integrated modelling system. Itcan also show that the spatial and temporal evolution ofentire dust storm episodes are well predicted.


In the spring of 2002, we made a 72-hour forecast onEast Asian weather every day. During these periods, wefound and resolved, step by step, some problems includ-ing the modeling system itself and GIS data. The forecastresults of the integrated modeling system has beenimproved dramatically. The top part of Fig. 8 gives usthe results of the dust concentration simulated at earlystage. The bottom part of Fig. 8 shows us the results ofthe dust concentration simulated after the GIS data andmodel system were revised. The results before revisionhave some errors, for example, in the southwest part ofChina occurred virtual dust regions. Furthermore, thearea of dust region is larger and displaced towards thesouth compared with the revised results. At the earlystage of the experiment, forecasters and the authorsfound these problems and we tried to resolve them.

We corrected and revised GIS data because we foundthe results of friction velocity and threshold frictionvelocity are not correct in some regions. A key variableto determine in a wind erosion scheme is the thresholdfriction velocity, u∗t. Several surface and soil-related fac-tors strongly affect the magnitude of u∗t, including soiltexture, soil moisture, salt concentration, surface crustingand presence of surface roughness elements, such asvegetation and pebbles. Some of them may be modifiedduring a wind erosion event. For example, the particle-size distribution of the topsoil may become coarser assmall particles are transported away from the source andaerodynamic roughness length may increase as large soil

Fig. 8. The forecast results before revision compared with simulatedresults after revision for the strong dust weather of 20 March 2002.

aggregates emerge from the surface. Consequently, u∗t

may also change during the wind erosion process. In oldsimulation cases, GIS data is not accurate in some areasand the simulation results are also affected by GIS data.

4.2. The simulation of strong dust weather—20 March2002

We forecast all the dust weather occurring fromMarch to April in 2002 by using the integrated predic-tion modeling system. The revised system predicted thestrong dust weather processes. In the following is anexample of strong dust weather. From 19–20 March2002, severe dust storms occurred in Beijing, whichaffected people’s lives, traffic and caused substantialeconomic damage. Fig. 9(a) shows us the observationresults. The meteorological station observed the dustweather and recorded the observations. We find fromFig. 9(a) that at 08 BST on 20 March, the dust weatherhad occurred in the south part of Mongolia, inner Mong-olia, Xinjiang, Gansu provinces and Beijing. From Fig.

Fig. 9. The observation of ground station (a) and the simulated dustconcentration (b) at 08 BST on 20 March 2002.


10(a) we also found the dust weather system movingtowards the east and the areas of dust concentrationbecame larger than the area of dust concentration at thetime of 20 BST.

Given the uncertainties involved in the model, it isimportant to verify the simulated results with obser-vations. The predictions can be compared with obser-vation image. Compare Fig. 9(a) with Fig. 9(b), also atthe same time, comparing Fig. 10(a) with Fig. 10(b), wefind the forecast regions of dust concentration are ingood agreement with the observations. It is the first realtime forecast of dust storms in China.

Fig. 11 shows the simulated surface wind field. It canbe seen from Fig. 11(a) that strong wind regionsoccurred in Mongolia and in inner Mongolia, further-more, strong cyclonic circulation was developed at thattime. The center locations of the strong wind areas arein agreement with the observation. At the same time, agood agreement is also found between the location andregions of strong wind and the simulated dust concen-tration. On 21 March, in inner Mongolar and Korea, pen-insula strong winds occurred and also formed a cyclonesystem which is in good agreement with the simulateddust concentration in these two regions. The experimentsfrom March to April have proven that the integrated sys-tem is a successful dust weather prediction system.

Many issues in dust modelling are unsolved, but the

Fig. 10. The observation of ground station (a) and the simulated dustconcentration (b) at 20 BST on 20 March 2002.

Fig. 11. Simulated wind field in the near surface, (a) 08 BST 20March 2002 (b) 08 BST 21 March 2002.

accurate estimation of dust emission seems to be the key.Even with the best model currently available, the uncer-tainties in the modelling of dust events are very large.Observation data are often unavailable or insufficient formodel validation. Under such circumstances, a synopticrecorder of dust activities is of particular importancebecause it provides a general guidance to numericalsimulation. It is a pity that we did not get any data ofdust concentration and I think we will compare our fore-casts of airborne dust concentrations with measured datain the future. Yet, there exist few observations that are


dedicated to the quantification of dust events, althoughefforts are being made by the China MeteorologicalAdministration for specialized measurements. Hence,traditional weather records are to be the best data.

Fig. 12 shows the simulated temporal evolution ofdust concentration section across Beijing occurring on20 March. It can be found that on 20 March Beijingsuffered a strong dust storm. The value of the dust con-centration reached 1 mg/m3 . On 21 March, Beijing alsoexperienced dust weather, but compared with the duststorm on 20 March, the intensity was weak. The forecastresults are in agreement with observations.

5. Summary and conclusions

An integrated wind erosion prediction system hasbeen briefly described with emphasis on the physicallybased wind erosion model and its linkage to a detailedGIS database. A systematic approach has been taken intomodeling wind erosion by using a wind erosion modeland a dust transport model and the coupling thesecomponents with a high-resolution weather predictionmodel. The integrated prediction system not only fore-casts the emission sources, temporal and spatial structuredistribution of dust, but can also be used for operationaldust weather forecasting every day. The simulated evol-ution of dust storms is in qualitative agreement with theobservations. Quantitative agreement is not verified forlack of observational data. The total simulation of the

Fig. 12. Simulated temporal evolution of dust concentration sectionacross Beijing.

15–20 March 2002 dust storm events in China comparedreasonably well with observations. In CMA the first realtime forecast of dust weather has been carried out andalso a great success.

In the simulation of dust storm events, the reliabilityof the atmospheric forcing data and the availability ofland-surface parameters are two additional constrainsimposed upon wind erosion modeling. Wind erosionevents are often associated with the development of cer-tain synoptic and sub-synoptic severe weather events,and these types of weather events are often the mostdifficult to describe and predict using atmospheric mod-els. We require high-resolution land–surface parametersfor soil texture, soil hydraulic properties, vegetationcharacteristics and surface aerodynamic properties. Theresolution of the data used in the modeling system isstill too coarse. Although we have some problems toovercome, our experiments in the spring of 2002 havedemonstrated that it is not impossible to predict theoccurrence of individual wind erosion events withreasonable confidence and estimate their intensity to thecorrect order of magnitude. It is therefore hopeful thatas various aspects of the modeling system improve, thesimulation and prediction of dust weather will becomesatisfactory.

Acknowledgements

Professor Wang Jianjie of the Numerical Meteorologi-cal Centre in CMA has been very helpful in providingcomputers and work environments. The author acknowl-edges that the Computational Modeling System(CEMSYS4) used in this study is provided by Dr. Yap-ing Shao through a collaborative project between Cityuniversity of Hong Kong and the China MeteorologicalAdministration, and the atmospheric prediction model,as part of CEMSYS4, was originally developed by Prof.Lance M. Leslie. Mr. Zhang Shihuang assisted in thepreparation of GIS data.

References

Bagnold, R.A., 1941. The Physics of Blown Sand and Desert Dunes.Methuen, London.

Gillette, D.A., Hanson, K.J., 1989. Spatial and temporal variability ofdust production caused by wind erosion in the United States. J.Geophys. Res. 94D, 2197–2206.

Irannejad, P., Shao, Y., 1998. Description and validation of the atmos-phere land surface interaction scheme (ALSIS) with HAPEX andCabauw data. Global Planet. Change 19, 87–114.

Leslie, L., Purser, R., 1991. High-order numerics in a three-dimen-sional time-split semi-Lagrangian forecast model. Mon. Wea. Rev.119, 1612–1632.

Liu, T.S., 1985. Loess and Environment. China Ocean Press, Beijingpp 215.


Lu, H., Shao, Y., 1999. A new model for dust emission by saltationbombardment. J. Geophys. Res. 104, 16827–16842.

Lu, H., Shao, Y., 2001. Toward quantitative prediction of dust storms:An integrated wind erosion modelling system and its application.Env. Modelling & Software 16, 233–249.

Marticorena, B., Bergametti, G., 1995. Modeling the atmospheric dustcycle: 1. Design of a soil-derived dust emission scheme. J. Geo-phys. Res. 100, 16415–16430.

Marticorena, B., Bergametti, G., Aumont, B., Callot, Y., N’Doume,C., Legrand, M., 1997. Modeling the atmospheric dust cycle: 2.Simulation of Saharan dust sources. J.Geophys. Res. 102D (40),4387–4404.

Owen, R., 1964. Saltation of uniform grains in air. J. Fluid Mech. 20,225–242.

Raupach, M.R., 1991. Saltation layer vegetation canopies and rough-ness lengths. Acta Mechanica 1 (Suppl.), 135–144.

Shao, Y., 2001. A model for mineral dust emission. J. Geophys. Res.106, 20239–20254.

Shao, Y., Lu, H., 2000. A simple expression for wind erosion thresholdfriction velocity. J. Geophys. Res. 105, 22437–22443.

Shao, Y., Li, A., 1999. Numerical modeling of saltation in atmosphericsurface layer. Boundary-Layer-Meteorol. 91, 199–225.

Shao, Y., Raupach, M.R., Leys, J.F., 1996. A model for predictingAeolian sand drift and dust entrainment on scales from paddock toregion. Aust. J. Soil Res. 34, 309–342.

Shao, Y., Leslie, L.M., 1997. Wind erosion prediction over Australiancontinent. J. Geophys. Res. 102, 30091–30105.

Shao, Y., 2000. Physics and Modeling of Wind Erosion. Kluwer Aca-demic Publishers, USA.

Tegen, I., Fung, I., 1994. Modeling of mineral dust in the atmosphere:sources, transport, and optical thickness. J. Geophys. Res. 99,22897–22914.

Tegen, I., Fung, I., 1995. Contribution to the atmospheric mineral aero-sol load from land surface modification. J. Geophys. Res. 100,18707–18726.

Walker, A.S., 1982. Deserts of China. Am. Sci. 70, 366–376.Westphal, D.L., Toon, O.B., Carson, T.N., 1988. A case study of mobi-

lization and transport of Saharan dust. J. Atmos. Sci. 45, 2145–2175.