Numerical Simulations of Asian Dust-Aerosols and Regional Impact on Weather and Climate-Part II : PRCM-Dust Model Simulation

The Purdue Regional Climate Model (PRCM)-Dust, an on-line coupled regional climate-dust model, has been developed to study the dust life cycle and radiative effect of dust on regional weather and climate in Asia in April 1998. The dust model is built on the PRCM discussed in Part I. The dust module includes the major phases of the mineral dust cycle, such as emission, advection, diffusion, and deposition. Furthermore, the optical properties of the dust aerosols are included in the radiation calculation of the PRCM, so that the feedback of radiative forcing of dust-aerosols can be considered in the PRCM modeling. The PRCM-Dust successfully simulated the uplifted dusts to reached around 500–800 hPa over the source regions and remained at 3–5 km or higher to be transported further eastward. The spatial and temporal distributions of the dust aerosols were consistent with the satellite images, the TOMS Aerosol Index maps, lidar observations, and the surface network reports. The regional climatic impacts due to radiative effect of dust aerosols include warming over the high concentration regions in North Asia and cooling in dust-free South Asia. Since there is no nudging, restart, or other artificial forcing in the PRCM-Dust model, the simulated dust concentrations, size distribution, as well as meteorological and soil environments follow the conservation laws of physics. They are suitable to study the interactions among dusts, soil, and regional weather/climate.


INTRODUCTION
Asian dusts occur frequently during the spring in East Asia.The transport mechanism and the optical properties of Asian dust have been studied for years using ground-based instruments such as lidar, Sun photometer, sky radiometer, and satellite (Iwasaka et al., 1983;Kai et al., 1988;Nakajima et al., 1989;Tanaka et al., 1989;Qiu and Sun, 1994;Husar et al., 2001;Wang et al., 2010Wang et al., , 2011)).Since the dust particles have distinguishable physical and optical properties, including a yellow color (UV absorbing), large size, and nonsphericity, it is relatively easy to determine the signature of Asian dust using ground-based instruments.The source regions are mainly in the Gobi Desert and the Taklimakan Desert in Central Asia (Bergametti, 1992).Those dusts can be easily delivered into the free troposphere and travel a long distance, sometimes reaches North America as in April 1998.Satellite remote sensors such as the Sea-viewing Wide Field-ofview Sensor (SeaWiFS) and the Total Ozone Mapping Spectrometer (TOMS) revealed daily images of this traveling Asian dust plume, as shown in Fig. 1 of Part I (Husar et al. 2001;Sun et al., 2013).Mineral dust plays an important role in altering the earth's radiative budget through the scattering and absorption of the solar and longwave radiations.The uplifted dust also influences cloud microphysical processes, atmospheric chemistry, and visibility, and even large-scale circulation.Numerical models have also been applied to study of dust storms.Nickovic et al. (2001) developed the SKIRON/Eta dust modules based on the NCEP regional atmospheric model output, which was reinitialized every 24 hours.Their off-line simulated dusts reached California on April 26 (Fig. 13 of Nickovic et al., 2001), and compared favorably with observations, although lots of simulated dusts remained in the source regions, while majority of the observed dusts were transported eastward by the prevailing westerly.In and Park (2002) applied a 3-D air quality model, which included dust emission, eddy diffusion, transport, dry and wet deposition, using the MM5 meteorological outputs to study the dust storms that occurred on April 14-19, 1998 in China.Their results showed that the dusts were transported southeastward by the wind at 1500 m in height.On April 15-18, the dust-laden air in China moved southward, then turned counterclockwise in association with a low-pressure vortex, and then moved to South Korea and southern Japan.Meanwhile, on April 19 a dust storm developed in the northern China which moved eastward along the 45°N latitude line until April 22.Overall, these simulations are reasonable, although the simulated concentration on April 16-18 reached Taiwan and even further south, beyond the observed boundary, as shown in Fig. 1 (In and Park, 2002).This extensive southward transport of dusts may be due to the simulated dust mainly stayed below 2 km -within the prevailing northerly wind.However, the mean height of the dust plumes was observed at about 4 to 5 km in Seoul and Hefei with a prevailing westerly wind (Murayama et al., 2001).Murayama et al. (2001) applied an on-line tracer model to study the dust and aerosols transport between April 13 and 30, 1998.The tracer model was coupled to the Regional Atmospheric Modeling System (RAMS) (Pielke, 2002).They also applied the four-dimensional data assimilation (FDDA) based on ECMWF global analysis data to their meteorological fields to reduce the errors in the meteorological model simulations.The height of the simulated dust layer was about 3 km over Japan due to a descending motion there, but was higher (4 to 5 km) in Seoul and Hefei as observed by lidar.Using the polarization lidar, they also commonly observed the coexistence of cirrus and a thin dust layer in the upper troposphere in Hefei and Japan.The trajectories derived from ECMWF showed the apparent effects of the cutoff Fig. 1.The vertically integrated daily mean TSP concentration expressed in common logarithm (µg/m 2 ) with the wind vectors at a height of 1500 m (In and Park, 2002).low at Hefei and Nagasaki when it passed over those regions, e.g., a strong downward motion and a counterclockwise curved trajectory.Shao et al. (2003) applied an integrated wind erosion modeling system (IWEMS) for 24-, 48-, and 72-hour forecasts of northeast Asian dust events in March and April 2002.Their short-term predictions are comparable with synoptic records from the meteorological network and dust concentration measurements at 12 stations in China, Japan, and Korea.Hai et al. (2011) applied a regional dynamical model (Weather Research Forecasting) coupled with a dust model (WRF-Dust) to study a dust storm from March 19 to March 22, 2010.On March 19, their calculated dust concentrations in the dust region (Taklimakan and Gobi Deserts) were around 500-700 µg/m 3 .On March 20, this calculated dust was transported from the source region to central-eastern China with concentrations of 100-150 µg/m 3 , which was much lower than the measured high dust concentrations (500-700 µg/m 3 or more).Using the coupled WRF model and the Community Multi-scale Air Quality model (CMAQ), Chatani et al. (2011) obtained Suspended Particulate Matter (SPM) that was significantly underestimated compared with observations in Japan.Chatani et al. (2011) applied the FDDA grid nudging to temperature, specific humidity and wind components with a nudging coefficient of 1.0 × 10 -4 .The WRF-CMAQ model also reveals a significantly underestimated 24-hr average PM 2.5 concentration in Bangladesh in January 2004 (Muntaseer Billah Ibn Azkar et al., 2012).
There is an increasing interest in mineral dust in recent years due to its significant impact on global climate change as well as on global air quality.Mineral dust also plays an important role in altering the earth's radiative budget through the scattering and absorption of the solar radiation (Haywood and Boucher, 2000;Harrison et al., 2001).In addition, the uplifted dust influences the microphysical processes and atmospheric chemistry of clouds, and impact atmospheric visibility.
Because it is difficult to accurately predict emission of aerosols, soil and meteorological conditions using a model, nudging or restart is commonly applied to meteorological forecasting in many weather-air quality coupled models as discussed before.Some dust models used the outputs and/or reanalysis of meteorological models (Uno et al., 2006).In the present study we present the PRCM-Dust model and the simulations of dusts integrated for 17-days continuously without nudging or restart.It demonstrates that the comprehensive PRCM-Dust model is capable of reproducing the observed front, cutoff low, onset of dust storms, and dust distributions.The model also provided a unique data set for the further study of interactions among dust, radiative forcing, soil, and weather/climate without artificial nudging or restart.

DUST MODEL
Dust emission depends on land surface properties, wind speed, and particle size.An on-line emission module of Ginoux et al. (2001) was incorporated into the PRCM-Dust.Dust production at the source regions is based on the dust source function which describes the potential production ability in different areas; the threshold velocity that allows dust particles to be blown into the atmosphere; and the surface wind at the source regions.Ginoux et al. (2001) assumed that a basin with pronounced topographical variations contains large amount of sediments that are accumulated essentially in the valleys and depressions, while in a relatively flat basin the alluvium is distributed homogeneously.A source function S has been introduced that is the fraction of alluvium available for wind erosions, and also is the probability to accumulate sediments.Also, only a land surface with bare soil is considered a possible dust source.The domain of the model and the source function based on a 1° by 1° dataset are shown in Fig. 2 (from Fig. 1 of Ginoux et al. (2001)).
The threshold velocity in the dust emission module of the PRCM-Dust takes into account the particle size and soil moisture content as suggested by Belly (1964).In the Goddard Chemistry Aerosol Radiation and Transport (GOCART) model (Ginoux et al., 2001), the relationship  Ginoux et al., 2001).
has been modified to use the first layer surface wetness (w = w g (soil moisture content)/w sat (saturated soil moisture)) to determine the threshold velocity.The threshold velocity (u t ) is given by ,0 10 According to Iversen and White (1982) and Greeley and Iversen (1985), the size dependence of the threshold wind friction velocity in the PRCM-dust was parameterized as: where a = 1331, b = 0.38, and x = 1.56,D p is the effective diameter (m) of a given size class, ρ p is the particle density (g/m 3 ), g is the gravitational acceleration, ρ a is density of air, and M is the threshold parameter normalized by Reynolds number function.The relationship between M and D p can be found in Fig. 3 of Iversen and White (1982).
Finally, the flux F p of particle size p can be approximated by , otherwise where C is a dimensional factor equal to 1 µgs 2 /m 2 , S is the source function (the function of alluvium available for wind erosion), u 10m is the horizontal wind speed at 10 m, u t is the threshold velocity, and s P is the fraction of each size class.The threshold friction velocity is a function of particle diameter (shown in Fig. 3).The fraction of clay is based on the erodible clay representing 1/10 of the total mass of emitted silt.
The transport of dusts in the PRCM-Dust is performed simultaneously (on-line) as passive scalar quantities.The dry deposition of dust aerosols is considered by the gravitational settling for each model's vertical layer and surface deposition velocity.The gravitational settling velocity V s depends on dust particle size and density: where υ is the viscosity of air.C is Cunningham's correction for small particles, and is given by 0.55 2 1 (1.257 0.4 ) where λ is the mean free path of air.The removal of dust aerosols by wet deposition in PRCM-dust is calculated using the model precipitation water at each model level, where C is dust concentration, is the washout parameter (= 5 × 10 5 , Nickovic et al., 2001), and ∂P/∂t is the modeled precipitation rate.The precipitation rate at each model layer is determined from the surface precipitation rate according to the liquid water mixing ratios at each layer of the model.Currently, the atmospheric radiation parameterization scheme in PRCM, following Chou and Suarez (1999) and Chou et al. (2001), consists of 11 wavelength spectral bands.The radiation module includes the absorption due to water vapor, O 3 , O 2 , CO 2 , clouds and aerosols.Interactions among the absorption and scattering by clouds, aerosols, molecules (Rayleigh scattering), and the surface are included.There are eight spectral bands in the ultraviolet (UV) and photosynthetically active (PAR) region that use single O 3 absorption coefficients and Rayleigh scattering coefficients.Another three bands in the infrared region apply the Kdistribution method to treat absorption (Chou and Suarez, 1999).In order to consider the radiative impact on the regional climate caused by dust aerosols, the radiation module in the PRCM was enhanced with online dust optical thickness calculations.The optical thickness at each vertical layer was calculated for each wavelength of the spectral band using the dust size-specific concentrations.The optical thickness  can be calculated by ( , ) 3 ( ) 4 where Q ext (λ, r i ) = the extinction efficiency factor at wavelength λ and effective radius r i ; (Q ext no unit) M i = column mass loading of the size class i (kg/m 2 ) ρ i = mass density of the size class i (kg/m 3 ) The values of Q ext (λ, r i ) were calculated using the Mie theory assuming a lognormal size distribution (Chin et al., 2002).In the radiation module of the PRCM, the optical thickness was calculated at each vertical layer instead of one for the whole column.Thus, the column mass loading used in the above equation was substituted with the layer mass loading.The values of the extinction efficiency factor at 500 nm, as well as other physical properties for each size class, can be found in Table 1.Because we lacked the complete set of optical properties at all wavelengths, the aerosol optical thickness of the wavelength bins other than 500 nm were linearly interpolated in the PRCM-Dust.Linear relationships have been observed in field experiments (Eck et al., 1999).The values for mineral transported aerosols at wavelength λ = 550 nm given by Hess et al. (1998) and Koepke et al. (1997) were applied for each wavelength of the spectral band at each vertical level in the radiation module.We also set the single scattering albedo ω = 0.95 and asymmetry factor g = 0.80.Fig. 4 shows the schematic diagram of components in the integrated climate-dust model, the PRCM-Dust (Yang and Sun, 2003;Yang, 2004;Sun et al., 2009).

MODEL RESULTS
The terrain, vegetation, soil property, model setup, initial and boundary conditions of meteorological simulations were discussed in Part I (Sun et al., 2013).Fig. 4 shows the diagram of the on-line integration of the coupled PRCM and Dust model, and the interfaces between the two models such that dust and weather can have active interactions.As mentioned previously, two major dust events occurred in   and Park, 2002) joined the dusts lifted by the second dust storm in the Gobi Desert (see Fig. 9) and were transported further east.On April 22, the dust storms continued in the Gobi and Taklimakan Deserts, which transported more dusts to the east and southeast.Fig. 9 shows the distributions of the dust mixing ratios (mg/kg) at σ-pressure level = 0.498 (~550 hPa) and TOM AI map on April 22, 1998, when a large dust plume covered northeast China, North Korea, Japan Sea, north Japan, and beyond.A high concentration-plume, which restarted on April 19, also formed in the Taklimakan Desert as shown in both simulation and TOM map.On April 22, the height of the major dust layer was about 3 km over Japan but was higher (4 to 5 km) in Seoul and Hefei.The average height of the dust plume was about 5000 m in northeast China according to the lidar network observation (Murayama et al., 2001).A thin dust layer in the upper troposphere was also observed in Hefei and Japan.Evidence of the coexistence of dust and cirrus was shown by the polarization lidar (Murayama et al., 2001)   (mg/kg) Figs. 11-13 show the vertical distributions of dust aerosols at cross-sections A-A and B-B (in Fig. 2) on April 20-22, 1998, respectively.On April 20, dust lifted in the Gobi Desert and was transported to high altitudes near the Gobi Desert between 35-50°N (A-A cross-section).Dust mass concentrations over 6 mg/kg were produced at the source region on April 20.Similar amounts of dust concentration were also predicted on April 15-16.In and Park (2002) predicted the daily mean dust concentration to be more than 6 mg/m 3 at the Gobi site on April 15 (Fig. 9 of In and Park).On April 20, the dusts spread and propagated southeastward with the dust aerosols concentrated at around 800 hPa in the 1200 km downwind region along the B-B cross section.Then the plumes continued moving to the upper levels.On April 21, very high dust mass concentrations were simulated at a high altitude (~5 km).These calculated high dust concentrations in the 1200 km-downwind region reached about 0.8-1.0mg/kg (with a local extreme of 2.5 mg/kg) in the upper layers, with the surface concentration reaching 0.3-0.4mg/kg (Fig. 12).The observed total suspended particle (TSP) reached 238 µg/m 3 (with maximum of 265 µg/m 3 ) and the PM 10 reached 150 µg/m 3 (with maximum of 192 µg/m 3 ) during heavy dust periods in Seoul (Choi et al., 2001).Meanwhile, Chung et al. (2003)   of 861-996 µg/m 3 with a visibility as low as 1.4-1.6 km from heavy dust storms that originated from China and Mongolia.Lidar observations confirmed that dust was uplifted to a higher altitude (~5 km) and then transported further east to Korea, Japan, and even across the Pacific Ocean with the cutoff low.This was also validated by satellite images that showed that the dust layer actually resided on top of low altitude clouds, significantly reducing the cloud albedo particularly near the ultraviolet (Husar et al., 2001).
Using the NARCM (Northern Aerosol Regional Climate Model), Zhao et al. (2006) obtained a maximum column loading of 1000 kg/km 2 or more near the source region on March 19, 2002.An intercomparison study of dust emission/ transport models over Asia (DMIP) for two dust episodes in 2002 were performed utilizing 8 dust models based on the output of global or regional meteorological models with objective analysis results (including NCEP, ECMWF, JMA, and NOGAPS) (Uno et al., 2006).The observed maximum TSP concentration during March 15-25, 2002 andApril 4-14, 2002 exceeded 12 mg/m 3 at Beijing, and 2 mg/m 3 at Seoul.They found that the results of the model correctly captured the onset and cessation timing of the major dust event at the observation site.However, the maximum concentration of each model differed by a factor of 2-4 times.The modeled wind over the Taklimakan area and Tibetan Plateau differed considerably between meteorological models.Some models indicate very calm conditions in the Taklimakan Desert, whereas other models showed a systematic easterly wind.The dust emission flux is fundamentally proportional to the third or fourth power of the surface friction velocity (u*).Thus, even small differences (say, 2-3 m/s) will result in a difference in dust emission flux of a factor of 2 or 3 times.Their results indicated that the difference in model results may be due to the meteorological parameters.Improvement of the meteorological model is a key to reducing the differences among the dust models.However, each dust model is strongly connected to its own meteorological driver (model) and the participating groups considered using unified meteorological conditions impractical (Uno et al., 2006).
Our total columnar dust mass loading reached 3000 mg/m 2 on April 15-16, 1998 (Fig. 7), which may be comparable with the dust loading of 2967 mg/m 2 with concentration of 989 µg/m 3 (with a depth of 3.0 km) observed in Korea on January 25, 1999 (Chung et al., 2003), and the mean column loading around 1500 mg/m 2 in Beijng simulated from the models in April of 2002 (Uno et al., 2006).But much less than the 12 mg/m 3 observed at Beijing on March 22, 2010.It should be noted that the simulation on a grid of 60 × 60 km 2 can be different from a single station measurement.
The dusts detected by SeaWIFS and the PRCM-Dust simulations were transported near but not beyond Taiwan in April 1998.On the other hand, the dusts simulated by In and Park (2002) spread further south following the wind at a height of about 1500 m, as shown in Fig. 1.This might have been caused by their low simulated mean height of dusts (< 2 km), as discussed before.Meanwhile, the discrepancy of the dusts detected by TOM (in green contour) and by SeaWiFS (yellow-reddish color) in Fig. 8 was also mentioned in Huser et al. (2001).
Fig. 14 shows the size distributions of the dust particles at 1200 km downwind (B-B).At the source region, the majority of the suspended dust particles had a radius of 1.8-3 µm and is classified as large silts (not shown).The production of sub-micron dust particles (clay) is very limited due to the high threshold wind velocity required to lift these particles.At the region 1200 km downwind, the majority of the dust population was also large silt (with a radius of 1.8-6 µm, i.e., (diameter is 3.6-12 µm)).Most of the dust particles over region were distributed between 500 and 800 hPa.According to Chun et al. (2001), size-separated number concentrations of aerosols ranging from 0.3 to 25 µm were observed in Seoul and Anmyon Island on the west coast of Korea during the Asian dust period in the spring of 1998.During heavy dust period, the number size distributions of aerosols observed in both places were characterized by decreases in the small sizes (≤ 0.5 µm) and an increase in the larger sizes (1.35 to 10 µm), which possibly originated from the dust source regions.
Overall, the PRCM-Dust model results demonstrated that the coarse particles in the source region could be uplifted quickly to reach high altitudes and formed a dust layer near 800-500 hPa.As discussed in Part I, at 700-500 hPa, the wind speed could reach as high as 50 m/s.The simulations also revealed that dusts could reach 200 hPa (Figs.12(b), 13(b), and 14) and be situated near the circus clouds, as observed (Huser et al., 2001, Murayama et al., 2001).

EFFECT OF DUST ON REGIONAL WEATHER/CLIMATE
The radiative forcing (see IPCC, 1994;Hansen et al., 1997;Shine and Forster, 1999) due to mineral aerosols originating from anthropogenic activity has been estimated at +0.09 W/m 2 for the global and annual mean, with a solar and longwave contribution of -0.25 W/m 2 and +0.34 W/m 2 , respectively (Tegen and Lacis, 1996).The net radiative forcing due to mineral aerosols is uncertain, as the solar and longwave forcing depend largely on the albedo (surface albedo or cloud albedo), refractive indexes, size, and altitude of the aerosols (Myhre and Stordal, 2001).Combining the observed data and GCM simulations, Myhre and Stordal found that an increase of 1 km in the altitude of homogeneous mineral aerosols, resulted in an increase in longwave forcing by about 30%.Solar forcing is more positive because a higher fraction of aerosols are above the clouds (with high albedo) than in the reference case.The net forcing based on the KHSS97 and the TF95 data sets is clearly showing a positive forcing over regions with high albedo (either surface albedo or clouds present) total forcing.Both KHSS97 and TF95 data sets yield the largest forcing in the regions around the Sahara.Positive forcing was also found in Mongolia, north China, Korea, and Japan (with the magnitude of 1 W/m 2 over northeast China for annual mean radiative forcing due to mineral content (≥ 1.5 g/m 2 ) ) from KHSS97, as shown in Plates 1 and 2 of Myhre and Stordal (2001).The global and annual mean clear-sky solar forcing is -0.51 W/m 2 and -1.16 W/m 2 based on the KHSS97 and TF95 data sets (Myhre and Stordal, 2001), respectively.On the other hand, Ahn et al. (2007) applied MM5 to study the effect of direct radiative forcing of Asian dust aerosol on meteorological fields of the intense Asian dust event on 18-23 March 2002.Their simulations showed a cooling on the surface over the high concentration regions and warming near the top of the dust layer.
Fig. 15(a) shows the difference of the simulated surface temperature with dust compared to that without dust (i.e., Control Case from Part I).The difference is +1.81K in northeast Asia and -3.03K in SE Asia due to the radiative effects induced by the large amount of dust aerosols in North and Northeast Asia.This is consistent with the results of Myhre and Stordal (2001) as well as the bias of April 1998 from the 10-year mean of the ECMWF reanalysis (Fig. 16).The results of the model indicated that the influence of dust was far beyond the polluted areas, and reached all of South Asia and India.The decrease of (-3.03 K) in Southeast Asia was larger than the increase of (+1.81) in NE Asia.
The warming over the areas with a high concentration of dust in April 1998 was mainly because most of the dust was located above the low clouds or even coexisted with the cirrus clouds.This was also validated by satellite images that showed that the dust layer actually resided on top of low-lying clouds and significantly reduced the cloud albedo, particularly near the ultraviolet (Husar et al., 2001).The dust layer also trapped longwave radiation from below similar to high clouds.The entire budget of course depends on concentration, size distribution, the mean height of the plume, clouds, etc.Our results are different from Ahn et al, which may be due to the differences in dust property, since their study was based on an event in March 2002, while our study is based on an event in April 1998.It is also possible that the simulated mean height of the dusts from MM5 was lower than observed, according to the simulations of In and Park (2002) and as discussed in Fig. 1, which could result in a cooling at the ground over the high concentration area.
The simulated dusts modified the radiative fluxes in the deep atmosphere over the polluted areas.This resulted in a change of temperature, pressure, wind, cloud, and precipitation over that area.The disturbances caused by the dusts introduced a geostrophic adjustment and secondary circulations (Sun, 2007).Hence, the influence of the dust spreads from the polluted area.At the same time however, it took time for this geostrophic adjustment to take place and be transported over a larger area.Hence, the impact of dusts on regional weather cannot be fully simulated in a model using short-term integration, or with a restart every day (or a few days), as is done in most models.The effect of dust depends strongly on its detailed distribution, and is controlled by both horizontal and vertical component winds, stratification, and vertical mixing, etc. of the meteorological model.Hence, an accurate meteorological model is necessary for a coupled meteorology-pollution modeling.This study also implies that the weather/climate of South Asia and Southeast Asia can be affected by events in the mid-or high-latitude.The relative humidity bias at 850 hPa from the 10-year mean analysis (Fig. 16(b)) shows a wet bias in northeast China, Southwest Asia, and India, but a dry bias in Southeast Asia and North Asia near 50°N, which does not match the pattern of surface temperature bias (Fig. 16

SUMMARY
An on-line coupled regional climate-dust model, the PRCM-Dust, was developed and applied to study the April 1998 dust storms in East Asia.The PRCM-Dust successfully simulated the uplifted dusts to reach around 800 hPh or higher over the source region and remained at 3-5 km or higher in the downwind regions.Dusts were transported southeastward with a cutoff low, then moved northeastward before April 18, and then were mainly transported eastward after April 18.They are consistent with the observed trajectory shown in Fig. 9.The horizontal distributions of the dust aerosols were consistent with the satellite images, the TOMS Aerosol Index maps, lidars, surface network reports, and the isentropic trajectory derived from the weather map.The radiative forcing of dust aerosols induced a regional warming in N Asia and cooling in S Asia.They are consistent with KHSS97 and TF95 data.This on-line coupled PRCM-Dust model reproduced the meteorological fields and dust cycles during a 17-day integration without artificial nudging or restart.The detailed results can be useful to study the onset of dust storms, the transport, dispersion, and deposition of dusts, as well as the interactions among aerosols, clouds, meteorological and soil environments, without contamination from artificial nudging or restart.This study also confirms that an accurate meteorology model is necessary before we can work on coupled meteorology-pollution modeling.More theoretical researches, observational studies, and improvements of numerical models are required in order to have a better understanding and prediction of dust storms.

Fig. 4 .
Fig. 4. The schematic illustration of the components in the integrated climate-dust model, the PRCM-Dust.
and was detected by satellite observations (Huser et al., 2001).The dust storms in the Gobi Desert subdued by April 23, as shown in both the simulations and the observations in Fig. 5.The simulated strong cyclonic wind pattern at 700 hPa at 00Z16 and 00Z17 shown in Figs.7-8 of Part I and the U-shape high concentration over the central-eastern China located at 3-5 km height are in good agreement with the observed back trajectory derived from Seoul at 4 km altitude (Choi et al., 2001) (Fig. 10(a)).The wind at 500-700 hPa as well as the orientation of dust concentration shown in Fig. 9(a) and TOM map in Fig. 9(b) became more west-east oriented after 00Z19 (Figs.10-11 of Part I).They are in good agreement with the observed back trajectory derived from Seoul during April 19-23, as shown in Fig. 10(b)(Choi et al., 2001).
Figs. 11-13 show the vertical distributions of dust aerosols at cross-sections A-A and B-B (in Fig.2) on April 20-22, 1998, respectively.On April 20, dust lifted in the Gobi Desert and was transported to high altitudes near the Gobi Desert between 35-50°N (A-A cross-section).Dust mass concentrations over 6 mg/kg were produced at the source region on April 20.Similar amounts of dust concentration were also predicted on April 15-16.In and Park (2002) predicted the daily mean dust concentration to be more than 6 mg/m 3 at the Gobi site on April 15 (Fig.9ofIn and  Park).On April 20, the dusts spread and propagated southeastward with the dust aerosols concentrated at around 800 hPa in the 1200 km downwind region along the B-B cross section.Then the plumes continued moving to the upper levels.On April 21, very high dust mass concentrations were simulated at a high altitude (~5 km).These calculated high dust concentrations in the 1200 km-downwind region reached about 0.8-1.0mg/kg (with a local extreme of 2.5 mg/kg) in the upper layers, with the surface concentration reaching 0.3-0.4mg/kg (Fig.12).The observed total suspended particle (TSP) reached 238 µg/m 3 (with maximum of 265 µg/m 3 ) and the PM 10 reached 150 µg/m 3 (with maximum of 192 µg/m 3 ) during heavy dust periods in Seoul(Choi et al., 2001).Meanwhile,Chung et al. (2003) found that in Korea from 1997 to 2000 the maximum concentrations of TSP recorded were 989-1396 µg/m 3 with PM 10 values
Fig. 16.ECMWF April 1998 monthly mean (a) surface temperature and (b) RH bias from 10-year mean.

Table 1 .
Summary of physical and optical properties of dust particle size class.
1The fraction of each size classes source function.2Extinctionefficiency factor (no unit) at wavelength λ = 500 nm.