This study presents a detailed examination of east Asian dust events during March–April of 2001, by combining satellite multisensor observation (Total Ozone Mapping Spectrometer (TOMS), Moderate-Resolution Imaging Spectroradiometer (MODIS), and Sea-Viewing Wide Field-of-View Sensor (SeaWiFS)) meteorological data from weather stations in China and Mongolia and the Pennsylania State University/National Center for Atmospheric Research Mesoscale Modeling System (MM5) driven by the National Centers for Environmental Prediction Reanalysis data. The main goal is to determine the extent to which the routine surface meteorological observations (including visibility) and satellite data can be used to characterize the spatiotemporal distribution of dust plumes at a range of scales. We also examine the potential of meteorological time series for constraining the dust emission schemes used in aerosol transport models. Thirty-five dust events were identified in the source region during March and April of 2001 and characterized on a case-by-case basis. The midrange transport routes were reconstructed on the basis of visibility observations and observed and MM5-predicted winds with further validation against satellite data. We demonstrate that the combination of visibility data, TOMS aerosol index, MODIS aerosol optical depth over the land, and a qualitative analysis of MODIS and SeaWiFS imagery enables us to constrain the regions of origin of dust outbreaks and midrange transport, though various limitations of individual data sets were revealed in detecting dust over the land. Only two long-range transport episodes were found. The transport routes and coverage of these dust episodes were reconstructed by using MODIS aerosol optical depth and TOMS aerosol index. Our analysis reveals that over the oceans the presence of persistent clouds poses a main problem in identifying the regions affected by dust transport, so only partial reconstruction of dust transport routes reaching the west coast of the United States was possible.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 It has been recognized that windblown mineral dust has significant effects on climate and the environment [Intergovernmental Panel on Climate Change, 2001]. Despite the recent advances in dust studies, quantification of adverse dust impacts remains a challenging, unresolved problem [Sokolik et al., 2001]. One of the outstanding issues is the quantification of spatial and temporal variability of the burden and properties of atmospheric dust at all relevant scales. The goal of this paper is to examine the extent to which the routine surface meteorological observations (including visibility) and satellite data help to characterize the spatiotemporal distribution of dust plumes. A case study presented here focuses on Asian dust in spring of 2001, addressing the active source regions of mineral dust in China and Mongolia, midrange transport and transpacific, long-range transport of dust outbreaks on a case-by-case basis. Adequate and consistent characterization of an individual dust event is central to establishing a reliable climatology, ultimately leading to improved assessments of dust impacts on the environment and climate.
 The data sets used in this study include meteorological characteristics and visibility records from 750 weather stations located in China and Mongolia, as well as the images and products provided by several satellite sensors: Total Ozone Mapping Spectrometer (TOMS), Moderate-Resolution Imaging Spectroradiometer (MODIS) and Sea-Viewing Wide Field-of-View Sensor (SeaWiFS). The National Center for Atmospheric Research Mesoscale Modeling System (MM5) atmospheric dynamic model driven by National Centers for Environmental Prediction (NCEP) Reanalysis data is also employed here in conjunction with observations.
 Meteorological parameters (wind speed, precipitation, temperature, etc.) and visibility seem to be the logical candidates for an analysis of the spatiotemporal distribution of dust. These characteristics are routinely reported from numerous weather stations located in China and Mongolia. In addition to land surface properties, meteorological parameters (especially, winds) are the key factors governing the emission and transport of dust, whereas visibility degradation is controlled by the dust loading. A number of previous studies have used meteorological observations to study the dust events in China [e.g., Zhang et al., 2001; Qian et al., 2002; Xuan and Sokolik, 2002], but just a few studies have considered both China and Mongolia [Chen and Chen, 1987; Sun et al., 2001; Shao and Wang, 2003; Shao et al., 2003]. These studies differ in the number of stations included in the analysis, meteorological characteristics considered, as well as in time periods and time-averaging scales. For instance, Sun et al.  analyzed wind speed, precipitation and dust storms (defined as dust events with visibility <1 km) from 174 Chinese weather stations from 1960 to 1999. Their study reported the spatial distribution of the frequency of dust storms, as well as preferential midrange transport routes. Qian et al.  analyzed dust weather types (including dust haze, blowing sand and dust storms) for the 1954–1998 time period using the data from 338 Chinese stations. They also have used monthly air mean temperature and precipitation from 160 stations and National Center for Atmospheric Research (NCAR)-NCEP Reanalysis data of 850-hPa geopotential height and 1000-hPa air temperature with a horizontal resolution of 2.5 × 2.5 degrees. Despite the fact that overlapping meteorological data sets were used in the above studies, there are various discrepancies in the reported results. Sun et al. and Qian et al. both agreed that the Taklamakan and Gobi deserts in northern China and southern Mongolia are the main source regions of dust in east Asia. However, the former study estimated a maximum annual mean number of dust storms of 20 and 40 for the Taklamakan and Inner Mongolian Gobi, respectively, whereas the latter study reported 35 and 20 dust storms for those regions. There are also noticeable differences in the reported spatial distribution of dust storms (see Figure 2a of Qian et al.  and Figure 5 of Sun et al. ). Furthermore, there are some discrepancies between the studies that characterized dust outbreaks in spring of 2001. Gong et al.  identified four dust events in spring of 2001 associated with low-pressure systems, whereas Gao et al.  reported 20 dust storms. Here we explore the potential sources of these discrepancies by performing a detailed analysis of the meteorological data for spring of 2001. Several specific questions are of interest. What is an appropriate definition of a dust event? How much can we actually learn from surface visibility measurements about dust events (e.g., start time and duration)? What other meteorological observations can be used in addition to visibility? Can meteorological data provide any constraints for the dust emission schemes? To what extent can ground-based meteorological data be used to identify the active source regions and midrange transport?
 The linkage between individual dust events in the source region and their midrange transport routes is important because the properties of atmospheric dust and thus its impacts are controlled to the large extent by dust sources. Midrange transport of east Asian dust has been the focus of a number of previous studies, and several preferential midrange transport routes of dust in east Asia have been suggested. On the basis of the 40-year meteorological data, Sun et al.  suggested four types of transport routes (called A, B, C and D; see Figure 3 of Sun et al.). In contrast, Zhang et al.  defined five general pathways (see Figure 5 of Zhang et al.): northeast, northerly Mongolian, northern desert path, western desert path and turning path (north to Beijing). The routes B and C in Sun et al. are similar to the western desert path and northerly Mongolian path of Zhang et al. Another study by Chen and Chen  identified four types of dust storm systems on the basis of the synoptic charts for the 10 years: stationary, moving (corresponding to type B of Sun et al.), stationary-moving, and high-level transport. However, the latter classification was focused more on the synoptic conditions favorable for dust storm formation rather than on transport routes. Whereas above studies relied on the ground-based observations only, here we examine the midrange transport routes of east Asian dust during spring of 2001 by utilizing both surface and satellite observations.
 It has been demonstrated that east Asian dust could be transported over the North Pacific reaching the United States and Canada. For instance, a detailed analysis of a single dust outbreak (19–25 April 1998) was conducted by Husar et al.  by utilizing several data sets (such as AVHRR aerosol optical depth, IMPROVE network concentration measurements, AERONET sun photometer data, TOMS aerosol index, etc.). Recently, Jaffe et al.  discussed several other cases of transpacific episodes of dust transport occurring during 1993–2001. In addition to observational evidence, several modeling studies simulated transpacific transport of individual Asian dust outbreaks [Uno et al., 2001; Takemura et al., 2002; Chin et al., 2003]. It would be desirable to relate the number of long-range dust episodes to the number of dust events in the source regions, as well as to understand the conditions favorable for long-range transport of dust. Here we examine the capability of TOMS, MODIS and SeaWiFS observations in tracking the dust plumes and constraining the areas affected by dust transport over the land and oceans. The latter is of particular importance to the number of problems ranging from regional climate radiative forcing of dust to dust iron deposition in the oceans. The important question is: What spatiotemporal resolutions of dust coverage can be achieved with MODIS and TOMS, considering sensor footprints, the time of data acquisition, and cloud contamination? Do dust spatial distributions inferred from independent satellite products agree well enough to provide a reliable constraint for aerosol transport model validations?
 The paper is organized as follows. In section 2 we introduce the data used in this study and procedures utilized for the data analysis. Section 3 deals with the sources and midrange transport of east Asian dust, whereas section 4 presents the reconstruction of long-range transport of dust over the Pacific Ocean. Section 5 summarizes our main results and discusses the implications.
2. Data Description and Approach
 Several data sets in conjunction with the NCAR MM5 meteorological fields were included in our analysis. Figure 1 shows ground-based meteorological data and satellite multisensor (MODIS, TOMS and SeaWiFS) products used in this study. All data sets were incorporated into the Asian Dust Databank, which we have been developing over the past years.
 Meteorological data for China and Mongolia were provided by the National Meteorological Center of the China Meteorological Administration and the NOAA National Climatic Data Center. The ground-based meteorological stations systematically report several meteorological variables such as visibility, temperature, relative humidity, wind, precipitation, etc., as well as several types of weather occurrences. Overall, the data from 750 stations for the spring of 2001 were analyzed.
 Several important issues must be addressed in analyzing the meteorological data that may affect the characterization of dust events. The first issue is that the stations have different frequencies of observations. In addition, not all meteorological parameters are reported sometimes. In March of 2001, 388 Chinese stations reported visibility. Of those, 240 stations measured visibility 4 times a day, 92 stations reported visibility each hour and 56 stations every 3 hours. In April, 356 Chinese stations measured visibility every 6 hours, 56 stations every 3 hours and 95 stations hourly. In Mongolia, 38 stations reported visibility each 3 hours during both March and April. Figure 2 shows the location of weather stations whose data were used in our analysis. The stations located in the dust source regions (according to the classification of Xuan and Sokolik ) are denoted by triangles and squares, showing different frequencies of visibility observations. The open squares and solid triangles denote the Chinese stations having a frequency of visibility observations of 1 hour and 6 hours, respectively, whereas the solid squares denote the Mongolian stations in the source region measuring visibility every 3 hours. The plusses show the stations measuring visibility in China and Mongolia, which are located outside the source regions. Overall, we analyzed the 1464 visibility fields for March and April of 2001. Because of the differences in the frequency of observations and time zones, only visibility fields for 0000 UTC, 0600 UTC, 1200 UTC and 1800 UTC include all 750 stations.
 Examining Figure 2, one can notice the heterogeneous spatial distribution of the stations. Thus an interpolation procedure would be required to characterize the spatial pattern of dust events, as well as to provide continuous coverage to facilitate comparisons of ground-based data with transport models and satellite observations. The choice of the interpolation technique does affect interpolated fields in terms of both the spatial distribution and the values of interpolated characteristics. Unfortunately, the previous studies, dealing with the analysis of Chinese and Mongolian weather data, rarely discussed interpolation procedures used, with a few exceptions [e.g., Shao and Wang, 2003]. This might introduce various biases in relating the spatial distribution of dust reported by different studies, especially in the dust source regions where the number of weather stations is limited. We tested several interpolation techniques (such as the minimum curvature, nearest neighbor, polynomial regression and kriging method) and found various differences in interpolated fields produced by different techniques. On the basis of our analysis, we selected the kriging method to perform the spatial interpolation of ground-based data (visibility and winds) in this study. The kriging method is described elsewhere [e.g., Matheron, 1976]. We believe that differences in the interpolation procedures and in the number of considered stations could be responsible for some discrepancies in the spatial patterns of dust events in China reported by previous studies as discussed above.
 Using measured winds and winds modeled with the MM5 model, we attempt to identify the active dust sources and the duration of dust emission (see section 3). The weather stations report winds at the 10 m height. Winds from the Chinese stations are available 4 times a day, whereas the Mongolian stations report winds 8 times a day.
 The wind fields were simulated with the NCAR MM5 model over a domain centered at 44°N and 96°E with a spatial resolution of 10 km. The domain (34.5°–69.9°N) × (51.3°–122.1°E) was selected to cover the dust sources in both China and Mongolia. MM5 runs were initialized by the NCEP Reanalysis data with a resolution of 2.5° × 2.5° each 6 hours. The integration time step was 30 s and the output was saved every 20 min for a 120-hour run.
 In addition to surface observations and MM5 simulations, we analyzed the data from TOMS, MODIS and SeaWiFS. The aerosol index (AI) retrieved from TOMS on board the Earth Probe (EP) satellite was used to detect dust over both the land and oceans. The EP TOMS AI is retrieved using two channels centered at 0.331 μm and 0.36 μm. These channels are sensitive to aerosols but less sensitive to absorption by gases. A spatial resolution of the EP TOMS sensor is about 40 km at nadir and it decreases with the increasing of the satellite viewing angle (up to 200 km). The current AI product (available at http://toms.gsfc.nasa.gov/aerosols/aerosols.html) represents 3-day composite images: Images use averaged data for 1 day and data from the 2 previous days to fill the data gaps (N. C. Hsu, personal communication, 2004).
 We also analyzed the red-green-blue channel (RGB) images and aerosol optical depth (AOD) at 0.55 μm from the MODIS sensor on board the Terra satellite. MODIS RGB images are Level-1 data reported as 5-min granules with the 1 km resolution. The MODIS aerosol product (MOD04_L2) contains data that have a spatial resolution (pixel size) of 10 km (at nadir). Each MOD04_L2 product file covers a 5-min time interval (based on the start time of each MODIS Level-1B granule). This means that the MOD04_L2 output grid is 10 km (at nadir) in width and 10 km in length for the nine consecutive granules. Each granule has an output grid size of 135 by 204 pixels. To illustrate the MODIS spatial coverage over east Asia, Figure 3 shows the Terra MODIS orbit track as the 5-min granules with the retrieved aerosol optical depth for 1 April 2001. One can notice that AOD are missing over the large areas that renders the characterization of dust spatial distribution over the land with MODIS aerosol optical depth data alone highly problematic (see further discussion in section 3.3).
 The daily mean AOD retrieved from MODIS (Level-3 global product MOD08_D3) were also used in our analysis. The daily mean AOD is derived from Level-2 data and averaged in space to a 1° × 1° equal-angle grid that spans a 24-hour (0000 to 2400 Greenwich Mean Time) interval.
3. Sources and Midrange Transport of Dust in East Asia
3.1. Definitions of a Dust Event and Its Duration
 Our examination of the previous studies has revealed that existing discrepancies in the reported frequency and duration of dust events could be explained to a large extent by the differences in used definitions. This problem arises because various criteria can be employed to define a dust event. For instance, several different definitions were proposed on the basis of the range of visibility values. Goudie and Middleton  define a dust storm as occurring when the visibility is decreased below 1 km, whereas the China Meteorological Administration uses the following operational classification: dust haze (visibility < 10 km), blowing sand (visibility between 1 and 10 km) and sand storm (visibility < 1 km). Shao and Wang  used four dust weather types reported by meteorological stations following the World Meteorological Organization protocol. However, the dust weather type is determined by different observers and thus prone to subjective errors [Shao and Wang, 2003]. In turn, Prospero et al.  determined the frequency and sources of dust storms relying on the TOMS aerosol index. Gong et al.  identified the dust events on the basis of prevailing meteorological conditions associated with dust outbreaks.
 For the sake of consistency, here we adopt the following definition: a dust event occurs when two or more stations located in a particular source region report visibility less or equal than 5 km. Dust events are further subdivided into two types: dust storm (visibility < 1 km) and blowing dust (1 km < visibility <5 km). On the basis of this definition, we use the visibility records from meteorological stations located in potential dust source regions to identify and classify the individual dust events. The visibility data from stations outside of the dust sources were used to constrain midrange transport (see section 3.2.2). Our reasoning for using the visibility data for the classification of dust events is that visibility is the only systematic ground-based observation, which is readily available and suitable for at least qualitative assessments of the spatial and temporal distribution of dust outbreaks, though various issues must be addressed in the analysis of visibility observations as discussed below. Another advantage is that visibility observations have been carried out by weather stations in China and Mongolia since about 1950 and thus they seem to be appropriate for developing a climatology of the dust events in this region.
 Another important issue is how to define the start and the end of a dust event. Several different criteria were used in the literature. Chen and Chen  define a dust storm start when at least three adjacent stations continuously report lithometeors of any intensity for at least three synoptic times (i.e., >24 hours). The China Meteorological Administration reports blowing dust when five or more stations measure visibility less than 10 km at the same time and dust storm when three or more stations report visibility less than 1 km. Our classification requires two or more stations to report visibility less than 5 km for blowing dust and 1 km for a dust storm and is based on statistical analysis of the frequency of visibility values observed by the meteorological stations in spring of 2001.
 It would require some time, from the start of the dust emission process, for dust particles to accumulate in the atmospheric boundary layer resulting in the visibility degradation. This implies that the use of visibility in determining the start time of the dust event might lead to a time delay relative to the beginning of the entrainment of dust. We will further address this issue in section 3.2.3 by analyzing the time series of visibility and winds at selected weather stations and discuss the implications for validation of dust emission schemes.
 Some ambiguity exists in defining the duration of a dust event. Following the visibility-based classification, it seems reasonable to define the duration as a time needed for visibility to recover to its background value. However, this time can differ between the individual stations because of their location and the frequency of visibility observations. Moreover, visibility records cannot be used in the case of long-range, transpacific transport because no regular visibility observations are carried out over the oceans. It becomes clear that this term might have a different meaning depending on the application ranging from the duration of a dust event in the source region to the duration of long-range transported dust episodes. Here we use the term “duration” to characterize the interval of time required for the visibility to decrease from 5 km to its minimum value and recover back to 5 km (see further discussion in section 3.2.1).
 One of the key issues in characterizing the dust events in the active dust source regions is a limited number of the weather stations. Most of the stations in China are located near populated, industrial regions and far from dust production areas. Visibility degradation is often observed over the industrial regions of China (mainly, eastern and southeastern parts of the country). A reasonable question arises on how to discriminate dust from local pollution episodes. Dust storms in east Asia originate in desert and semidesert areas having low population density. In those regions a pollution plume hardly can be mistaken with the dust plume since urban and industrial pollution never degrades the visibility lower than 5 km, whereas in all dust events identified in this study visibility was lower than 5 km, and for 60% of the cases below 1 km (see section 3.2.1). However, the reconstruction of midrange transport of dust especially over the polluted east and southeast parts of China must be performed with care. To illustrate, Figure 4 shows interpolated visibility fields for 6, 7, 8, and 9 April 2001 at 0600 UTC (the time when all stations report visibility). The plots clearly show the degradation of visibility over the Gobi and Taklamakan during the so-called “perfect dust storm,” which represents two separate dust episodes originating on 6 and 8 April and associated with low-pressure systems [Huebert et al., 2003]. Figure 4 also shows that some urban areas in southeast China could have low visibility (between 8 and 15 km). However, the presence of dense dust plumes can be unmistakably identified because of their large area of coverage and associated lower visibility (0–5 km).
3.2. Analysis of Ground-Based Data and MM5 Simulations
3.2.1. Dust Events in the Source Regions
 On the basis of the visibility analysis for the spring of 2001, we concluded that there were five main dust production regions: (A) Taklamakan desert, (B) central Chinese Gobi, (C) central Mongolian Gobi, (D) eastern Chinese Gobi, and (E) eastern Mongolian Gobi. Figure 5 shows the number of stations that reported visibility <1 km (black lines) and <5 km (gray lines) as a function of time in each region. The location of source regions is shown in Figure 6. These regions support the classification proposed by Xuan and Sokolik : Type 1 sources include the deserts in dry-agricultural areas, type 2 are Gobi deserts and deserts located on the plateaus, and type 3 sources are deserts and Gobi deserts located in topographic lows (see Figure 2).
Figure 5 shows that some dust outbreaks were reported only during one observational time. Since our visibility data have different temporal resolution (1 hour, 3 hours and 6 hours) (see Table 1), we selected the lowest frequency to be an indicator of an occurrence of a dust event (i.e., we selected only these cases that lasted more than 6 hours). Thus for a dust event to be identified, we required at least two consecutive observations for stations reporting each 6 hours, and 3 or 6 reports for the stations measuring visibility each 3 hours and 1 hour, respectively. Thus our complete definition of a dust event in this study is the following: a dust event occurs when two or more stations located in a particular source region report visibility less or equal than 5 km during at least 6 hours.
Table 1. Frequency of Visibility Observations in the Source Regions
 Out of all outbreaks shown in Figure 5, Table 2 summarizes only those that lasted more than 6 consecutive hours. Also shown in Table 2 are the starting date and time, event type (DS, dust storm; BD, blowing dust), duration by the event type, event duration, and midrange transport routes. In total, 35 dust events were identified for March and April 2001: 8, 10, 7, 7 and 3 in A, B, C, D and E, respectively. Some events consisted of one dust event type (e.g., blowing dust), whereas other consisted of several dust event types (e.g., blowing dust, dust storm and blowing dust). We found 40 cases of blowing dust and 25 cases of dust storm for the considered regions. Regions A and B showed the highest dust activity. For each dust event, midrange transport routes were reconstructed by combining visibility and wind fields and constrained against satellite observations as discussed below.
Table 2. Classification of Dust Events for Spring of 2001
 One problem that we encountered while performing the classification of dust events in the source region was the low frequency of observations. For instance, we had just a few stations in the source regions with the frequency of observations of 1 hour. This renders the determination of the exact start of the dust event problematic. As a result, starting times reported in Table 2 are within an error of at least ±1 hour. Unfortunately, the measured surface winds also cannot be used for constraining the starting time of a dust event since the frequency of wind measurements is even lower (between 3 and 6 hours). The low observational frequency of meteorological data limits their use in validating the dust emission schemes since dust production commonly occurs at shorter timescales.
 It is important to point out that the number of the dust events in spring of 2001 identified in this study (see Table 2) differs from those reported by Shao and Wang  and Gong et al. , mainly because of different definitions used. Gong et al.  reported only four major dust storm episodes in spring of 2001 based on synoptic considerations: DS1 (2–6 March), DS2 (21–27 March), DS3 (4–14 April) and DS4 (29 April to 1 May). The problem in using the approach of Gong et al. for dust event classification is that it requires an analysis of synoptical conditions which can be subjective depending on the information used (e.g., meteorological fields predicted by a dynamical model versus weather forecast based on synoptic observations), as well as the assumptions on how to relate a particular synoptic condition and dust storm occurrence. In turn, the frequency-based approach of Shao and Wang  does not allow linking an individual dust event in the source region to its midrange and long-range transport. We believe that the classification of the dust event and its duration proposed in this study overcomes the above limitations, though our approach depends on the number of stations and source regions considered. It becomes clear that a consistent approach for characterizing the dust events that would be widely adopted by the scientific community is urgently needed.
3.2.2. Midrange Transport of Dust
 First, we used the visibility data to reconstruct midrange transport of the dust outbreaks on the case-by-case basis. Then the satellite data were examined as to whether they could provide an additional constraint. To illustrate the use of visibility data, Figure 7 shows a midrange transport route reconstructed for a dust storm originating over the Gobi on 8 April 2001. Visibility measurements for 0000 UTC, 0600 UTC, 1200 UTC and 1800 UTC are shown, and regions with visibility less than 1 km are marked by dark gray shading, whereas the region with visibility <5 km is shown in light gray shading. The evolution and transport of this particular dust plume can be easily observed. The storm that started earlier on 7 April was transported to the southeast-east. Another storm in the Taklamakan started to develop around 0600 UTC on the previous day and the plume was transported to the east on 8 April.
 Overall, our analysis was performed for 1464 visibility fields in March and April of 2001. Two main transport routes determined for this time period were identified: east and southeast. In the Taklamakan (region A), out of eight dust events, three events were local (i.e., no dust was transported outside of the source region); four were transported to the east, and one to the southeast. Out of 27 events occurred in sources B, C, D and E, 15 were transported eastward, seven events to the southeast and then to the east, four to the southeast, and one event was local.
 Midrange transport routes identified here are somewhat different from those reported by Sun et al. . They suggested four routes based on the 1960–1999 data: east (through North Korea, case A), east (through South Korea, case B), southeast (case C), and northeast (case D) (see Figure 3 of Sun et al.). Our east route is similar to the combined routes A and B of Sun et al. We did not distinguish between dust plumes being transported over North or South Korea as separate routes, assuming that there is only one path to the east. Our southeast route out of the Gobi coincides with the southeast path of Sun et al. From the visibility data for spring of 2001, we did not find the route D suggested by Sun et al. However, examining MODIS RGB images for springs of 2000, 2001, and 2002, we found some indications of the presence of the route D out of the Taklamakan (see further discussion in section 3.2.3), as well as several other routes not discussed by Sun et al. It is possible that using a limited number of meteorological stations in mountain regions hampers the reconstruction of midrange transport routes. This implies that Sun et al. climatology, based on ground-based data only, could have some biases. On the other hand, large interannual variability in midrange transport could be the reason for the differences in our and Sun et al. results. Another possible reason is that we included also the blowing dust cases (visibility <5 km) in our midrange transport analysis. Nevertheless, here the MODIS RGB images provided helpful information on transport routes for the considered cases, although reliable discrimination of dust from the surface and clouds on true color images poses a serious problem in using the MODIS RGB data for consistent characterization of dust transport.
3.2.3. Analysis of Time Series of Visibility and Winds for Selected Ground-Based Stations
 Our analysis of visibility time series along with observed and modeled winds had a twofold goal. First, we were interested in finding a time delay between visibility and winds at the starting point of dust events as discussed in section 3.2. Second, we wanted to explore in detail the transport pathways (if any) of dust originating in the Taklamakan. Table 3 lists the stations selected for the time series analysis.
Table 3. Geolocation Data for Selected Stations
 To find a time lag between the wind speed and visibility, we examined the data for 6–10 April for four stations located in different source regions: Mongolian station 44347, Hotan, Ejin and Lindong stations (see Table 3). Observed winds were only available for the Mongolian station, so, for the remaining stations, we used MM5 winds simulated at 10 m with a time step of 20 min. Figure 8 compares the time series of visibility and winds for the four stations considered. A time lag between the wind and visibility is defined here as the time (if any) required for visibility to start decreasing from its background value after a threshold wind speed (assumed 6 m/s) is reached. Unfortunately, we were unable to find a clear time delay. Our explanation is that 1- to 3-hour interval in visibility measurements is probably too large to detect the time delay.
 Furthermore, we examined the time series of events (marked by the gray areas in Figure 8) to relate the changes in visibility and wind speed. The results are presented in Figure 9. Figure 9a illustrates the schematics of the expected linkage between winds and visibility for the case of strong dust events that would be observed by the stations located in the source region, whereas Figure 9b shows the schematics for a weak dust event. For the strong event case (Figure 9a), visibility shows a hysteresis behavior as a function of wind speed. Initially, visibility does not respond to the increasing wind speed for some time (branch A). When threshold speed is reached, visibility starts to decrease with some delay following a further increase in the wind speed (branch B). After the large amounts of dust accumulated in the boundary layer, the further increase of wind does not cause decrease of visibility (branch C). When wind starts to decrease, visibility is still saturated and it takes some time to recover back to its background value (branch D). During the weak dust event (e.g., blowing dust), visibility decreases and then increases without any “saturation.” Figures 9c–9f show the hysteresis behavior of visibility found for considered individual stations. The distance between the points is 1 hour for the Chinese stations and 3 hours for the Mongolian station. According to our classification, dust storms (visibility <1 km) were observed at Ejin, Lindong and station 44347, whereas Hotan station reported a blowing dust event (visibility <5 km). It has to be pointed out that the hysteresis behavior discussed above is valid for the stations located in the source region so winds measured locally are responsible for the mobilization of dust. For events that originated upwind from the station, the hysteresis curve can change its direction. For such cases it is possible to have decreased visibility before the local winds reaching the threshold value.
Figure 9c agrees well with the schematics outlined in Figure 9a. In turn, Hotan data show a good agreement with the schematics for blowing dust events (Figure 9b). For this case, visibility decreases from 8 km to 4 km and back to 8 km during the 3-hour time period. The behavior of visibility at Ejin and the Mongolian station follows the hysteresis curve but in the opposite direction. As discussed above, a possible explanation is that dust did not originate locally but was transported from upwind sources. This is supported by the fact that winds were much lower at the beginning of the dust storm compared to the threshold wind required to initiate the dust emission. Thus the initial sharp drop of visibility at Ejin could be caused by transported dust, then long saturation (approximately 8 hours) could be a result of both local and transported dust and, then, slow recovery back to background conditions occurred. This finding was confirmed by the analysis of the wind and the visibility time series for the neighboring stations in the source region. From our time series analysis, we conclude that the hysteresis-like behavior of visibility, identified for the stations located in the source region, can be helpful in constraining the emission, boundary layer mixing and deposition schemes used in dust transport models, although care must be exercised in selecting the stations for such an analysis.
 Somewhat conflicting results were reported by the previous studies regarding dust transport out of the Taklamakan Desert [Sun et al., 2001; Sun, 2002; Iwasaka et al., 2003; Wang et al., 2003]. All studies agree that the Tarim basin has a complex topography and hence a complicated surface wind pattern that drives dust emission. On the basis of a surface wind 1961–1999 climatology, Wang et al.  concluded that winds in the western Taklamakan are blowing predominantly from northwest, winds in the eastern Taklamakan are from the east and, in the southern edge of the basin, prevailing winds are from the west and northwest. Sun et al. , on the basis of the motion of sand dunes, suggested that surface winds in the basin are mainly easterly/northeasterly. In addition, Sun et al. suggested that since the Taklamakan is surrounded by high mountains, easterly winds cannot easily entrain dust out of the desert. Their later study [Sun, 2002] suggested a possible scenario of transport of dust out of the Taklamakan by considering dust accumulation on the southern slope of the Kunlun mountains by surface northeasterlies. Sun  also suggested that under specific meteorological conditions dust can be entrained to elevations above 5 km and transported by a westerly jet. The latter was confirmed by lidar measurements reported by Iwasaka et al. .
 Since the Taklamakan is a productive dust source, it is important to understand how much of this dust actually can be transported outside the Taklamakan desert, contributing to midrange and long-range transport of east Asian dust. To examine the findings from previous studies, we analyzed the time series from five stations located in the Taklamakan area (see Figure 10a and Table 3). Two stations are located on the Tibetian Plateau (Kunlun Mountains), two are located at the east edge of the desert, and one is located in the desert. Figures 10b–10d show the visibility time series of individual stations for March and April of 2001. The duration of dust events identified over the Taklamakan (region A) are also shown by horizontal arrows on Figure 10. One can notice in Figure 10 that visibility records at these individual stations are somewhat different from the dust events identified on the basis of the visibility data for all 37 stations located in this source region as discussed in section 3.2.1. For Hotan and Minfeng stations (Figure 10b) located on the windward slopes of the Kunlun Mountains (both stations with elevation higher than 1 km), there is a good agreement between the duration of dust events and the degradation of visibility. This confirms the results of Sun et al.  that dust originating in the Taklamakan desert can be lifted to high altitudes (in this case to 1.5 km). On the other hand, Ruoqiang and Tikanlik stations (Figure 10c) located at the east edge of the Taklamakan show overall marginal agreement, showing some response to the dust events but having the visibility minima about a day after the dust event dissipated. Moreover, visibility at those two stations did not decrease below 3–4 km, in contrast to Hotan and Minfeng stations located high on the Tibetian Plateau where each dust event resulted in visibility close to zero. The results suggest that it is very unlikely for dust to be transported near the surface. A possible explanation is that, once mobilized, dust experiences forced lifting caused by the complex topography and, if dust is transported downwind, it has to be above the boundary layer. This is also confirmed by the time series from Minfeng and Lintai stations (Figure 10d). The former station is located on the Tibetian Plateau, whereas the latter one is located in the north part of the desert but at the relatively low altitude (there are no stations in the Taklamakan located lower than 800 m) (Figure 10a). Despite the fact that Lintai is located much lower than Minfeng, the former station did not experience great visibility degradations during dust events in spring 2001.
Table 4. Comparison Between Surface Visibility Patterns and MODIS RGB Imagery
MODIS RGB Image
Number of Granules
Low (<10 km)
clouds ≤ 20% not overlapping the plume
Low (<10 km)
20% < clouds < 80%, usually partially overlapping the plume
Low (<10 km)
clouds > 80%
Low (<10 km)
High (>10 km)
clouds ≤ 20%
High (>10 km)
20% < clouds < 80%
High (>10 km)
clouds > 80%
High (>10 km)
 Our analysis revealed that out of eight dust events originating in the Taklamakan, three were local, four were transported to the east, and one was transported to the southeast. Of those transported to the east, all were associated with westerly-northwesterly winds. However, as we discussed above, the stations located at the east edge did not show significant response to dust events. Our explanation is that dust was lifted above the boundary layer and then transported downwind to the east. This is also supported by the visibility records at the stations located to the east of the Taklamakan: Magnai (elevation 2947 m) and Chugirty (elevation 924 m) and by MODIS RGB images.
 The single case in spring of 2001 when dust was transported to the southeast seems to be unique. Using the MODIS RGB images for March, April and May of 2000, 2001 and 2002, we searched for other cases of southeast transport of dust from the Taklamakan. The 276 granules (about one per day over the Taklamakan) were analyzed. Our analysis indicated that 31 granules show dust transported to the east of the Taklamakan, five to the northeast, four to the north, four to the northwest, two to the southeast and only one to the south.
 In summary, our analysis of visibility data and satellite MODIS RGB images lead us to the conclusion that dust is transported out of the Taklamakan preferably to the east. Since observational data did not indicate decreased visibility at the surface at the east end of the desert in all eight cases of spring 2001, it is likely that dust was lifted above the boundary layer and then transported to the east. Other routes out of the Taklamakan are also possible but seem to be infrequent, strongly depending on the meteorological conditions and complex topography of the region.
3.3. Satellite Validation of Midrange Transport of Dust
 Dust plumes can be detected by the satellite sensors operating in the UV, visible or IR spectral regions, though different spectral bands have different capabilities in sensing dust over the land and oceans. Here we examine several satellite aerosol products against the surface visibility measurements in terms of the spatiotemporal pattern of dust transported over east Asia to evaluate the potential of satellite aerosol products in constraining the dust distribution. The aerosol products inferred from satellite data are available at different timescales ranging from instantaneous observations to daily and monthly means, as well as at different spatial resolutions. To provide an adequate comparison, the surface data were interpolated to an appropriate spatial grid and averaged over time to match a particular satellite product. All satellite products used here are derived from polar orbiting satellites. Although geostationary satellites can be also helpful for midrange transport studies, no aerosol products retrieved from geostationary satellite sensors over east China are currently available openly to the scientific community.
3.3.1. Multiday Composite Satellite Product: TOMS Aerosol Index
 Although the TOMS AI posted at the TOMS website are referred to as daily mean data, they are actually calculated as a composite of several days (see section 2). Nevertheless, we compared the daily mean visibility and the TOMS AI. For each day of March and April of 2001, the daily mean visibility fields were produced and compared to the TOMS AI pattern over the study domain. To illustrate, Figure 11 shows the daily mean visibility and TOMS AI for 6 April 2001. For this particular day, the visibility field shows five areas of decreased visibility numbered from one to five in Figure 11. We compare visibility and TOMS AI fields looking for the similar regional pattern. Areas 1 and 2 coincide on both plots so these two cases are classified as agreement. Area 4 on the visibility map does not have a counterpart on the TOMS AI plot and this is classified as disagreement. Areas 3 and 5 show some of the features seen in the visibility fields but not over the same geographical location. So these two cases are classified as conditional agreement. On the basis of this approach, we identified 206 cases of low visibility (<5 km) in March and April of 2001. Of those, 144 cases were in agreement with TOMS AI (70%), 34 cases were in conditional agreement (16%), and disagreement was in 28 cases (14%). The highest number of disagreements and conditional agreements occurred over northeast and southeast China. The average TOMS AI for these areas is 1.3–1.5, in contrast to 1.5–2 over the dust source regions. Most of the disagreement cases occurred when no dust storm was evident and the low visibility over those regions was likely due to industrial pollution. Since the percentage of agreement cases is relatively high (70%), we concluded that the TOMS AI performed well in detecting dust over the areas with the visibility less than 5 km.
3.3.2. Daily Mean Satellite Product: MODIS Aerosol Optical Depth
 Following the procedure outlined above, we compared AOD inferred from MODIS against the visibility fields. The spatial patterns in daily mean values of visibility and AOD were examined and the cases of agreement, disagreement and conditional agreement were identified. Out of 206 cases analyzed, 108 cases were in agreement (52%), 51 cases were in disagreement (25%), 42 cases were in conditional agreement (20%), and the AOD data were not available for the remaining 5 cases (2%). Similar to the TOMS AI most of the disagreement cases occurred in southeastern China when no dust storm was present and the low visibility was likely due to urban pollution. We found that the level of agreement for MODIS AOD (52%) was lower than for the TOMS AI (70%). One possible explanation is the difference in the timescales. Multiday composite products provide a better picture because data from several days are used to fill the gaps in contrast to daily averaged MODIS Level-3 product. However, multiple-day products miss the temporal evolution of dust outbreaks at shorter timescales. Another possible explanation is the problems with AOD retrieved from MODIS over the land as discussed below.
3.3.3. Instantaneous Satellite Products
 We analyzed three different satellite products retrieved from instantaneous observations: MODIS true color images, MODIS AOD over the land, and SeaWiFS AOD over the ocean.
126.96.36.199. MODIS RGB Images
 The MODIS true color granules were analyzed using the following approach. China and Mongolia were divided into three large sectors: west (75°–95°E), central (95°–115°E) and east (115°–135°E). Then, each sector was further subdivided into three subsectors. The areal coverage of each subsector was about the size of a 5-min MODIS granule (see Figure 3). Visibility observations and MODIS granules collocated in time and space were examined to identify the similarities and differences in observed features. The following classification was introduced: agreement, when the image scene agrees with the surface visibility pattern and the cloud coverage is less than 20%, conditional agreement, when only partial agreement is established because of the presence of clouds (cloud coverage is between 20 and 80%), cloud contamination, when the granule is more than 80% covered by clouds, and undetermined, when we could not decide what phenomena we dealt with (Table 4).
 This analysis was done for each day of March and April 2001 for the time periods when the satellite passes over China and Mongolia (from about 0300 UTC to 0600 UTC, approximately four granules per day). For different days, the different number of granules was analyzed because the MODIS spatial coverage varies over time. Out of the total 245 MODIS granules analyzed, 48% are in agreement with the surface visibility pattern (the percentage includes agreement in presence of dust and agreement for clear sky scenes). The percentage is smaller compared with the TOMS AI. The reason for low agreement is likely due to the problems in discriminating atmospheric dust from cloud and bright surfaces.
188.8.131.52. MODIS AOD Over the Land
 Examining MODIS RGB images, we found 67 granules that show the presence of dust over the areas with low visibility. Those granules were further examined for retrieved AOD. Figure 12 shows the distribution of the number of granules versus the percentage of pixels for which AOD was retrieved in each granule. We found that there were no granules with more than 50% retrieved pixels, and the majority of granules had only 10–20% retrieved pixels, though the cloud coverage for the most of those granules was low (about 10–20%). This implies that the cloud contamination is not a main factor here. One possible explanation of poor retrievals of AOD over the land is that the MODIS retrieval algorithm does not account correctly for the spectral surface reflection. This problem has already been pointed out by Ichoku et al. . We conclude that at this time, MODIS AOD over the land cannot be used as a standalone tool for detecting and characterizing dust. Moreover, the daily mean AOD, which are calculated from instantaneous AOD by interpolating them to a larger spatial grid, should be used with caution.
184.108.40.206. SeaWiFS AOD Over the East China Sea
 We analyzed 61 SeaWiFS AOD images (one per day) based on HRPT granules (http://seawifs.gsfc.nasa.gov/cgi/hrpt_browse.pl) over the East China Sea. The reason to extend our analysis over the sea was to explore whether SeaWiFS AOD can provide an additional constraint for the midrange transport routes of dust. The procedure consisted of comparing the aerosol optical depth images along the coast with the visibility measurements at the coastal weather stations at similar observational times. Since we worked with granules retrieved over the East China Sea, we were able to validate only those events that reached the east coast of China, 26 events out of total 35. Another limitation was that we had only one granule per day at a specific time and the percentage of the retrieved pixels was usually less than 50% out of the total area of the granule. As a result, we were able to compare directly only 13 dust events. For eight of them, the SeaWiFS granules showed increased AOD over the coastal waters in the expected locations. Therefore we were able to confirm that the midrange transport routes for these eight events based on visibility analysis were correct. Despite the fact that some of the granules agreed with the visibility measurements and confirm the observed transport routes, the results are not very satisfying: The retrieved AOD over the East China Sea was too scattered because of the coverage gaps and probably retrieval problems. In the process of comparing the visibility fields with SeaWiFS AOD, we had a good idea from the visibility fields as to where to look for an increased AOD along the coast. If we had not had the surface measurements, it would have been problematic to do any classification based only on SeaWiFS AOD.
 In summary, we used the RGB images from MODIS, TOMS aerosol index and AOD from MODIS and SeaWiFS to reconstruct the spatiotemporal distribution of dust at different temporal and spatial scales and constrain it against surface visibility measurements. The procedure was prolonged and detailed (we processed and analyzed approximately 1300 satellite images), but it was valuable for determining the ability of the satellite products to characterize the dust distribution. Because of various limitations of satellite data over the land, we believe that satellite aerosol products should be used in conjunction with surface observations to provide a more reliable characterization of dust distribution over the land, as well as for validation of aerosol transport models.
4. Reconstruction of Long-Range Transport of Asian Dust Plumes
 The reconstruction of long-range transport of dust plumes over the Pacific was based solely on satellite data since visibility is rarely reported over the oceans. Two long-range transport dust episodes were identified and their day-by-day transport was independently reconstructed using two different satellite products: TOMS aerosol index and MODIS AOD. These products are often used for validation of dust transport simulated with the atmospheric general circulation models. Thus it is important to address the differences and similarities in dust spatiotemporal distribution reconstructed from the TOMS aerosol index and MODIS AOD.
Figure 13 shows dust coverage reconstructed from the TOMS AI. The TOMS AI catches the start of the first dust storm on 6 April over the central Gobi. On 7 April, the plume was transported over northeast China to the east. Later on the same day, another dust event started over the central Gobi. On 8 April, the first storm reached far northeast China. On 9 April, it was further transported to north Japan and the second one reached the east coast of China. On 10 April, the front edge of the second storm reached the dust plume of the first one. Overlapping of two plumes resulted in the increased TOMS AI over the entire North Pacific (from China to Alaska) on 11 April. On 12 April, the dust plume reached the coast of North America and was detected with lidars and surface measurements in the United States and Canada.
 In turn, Figure 14 shows the reconstruction of dust transport using instantaneous MODIS AOD higher than 0.4. The structure of the AOD contours is spotty, and the extent of the plume is hard to differentiate. Nevertheless, the MODIS AOD broadly agrees with the TOMS AI. They both capture relatively well the location and the shape of the dust plumes over the ocean. However, MODIS AOD fails to detect the beginning of the dust storms on 6 and 7 April over the source region likely because of the problems in aerosol optical depth retrievals over the land as we discussed above.
 Over the ocean, the reconstructed dust coverage differs especially around 9–10 April. There are a few reasons for this discrepancy. One reason is the fact that we use different variables: aerosol optical depth versus aerosol index. As pointed out by previous studies, the relationship between aerosol index and optical depth depends on several factors. For instance, aerosols at low altitudes have lower TOMS aerosol index than an equivalent depth of aerosol at a higher altitude. Another reason for the differences in the dust coverage is the contamination of clouds. Finally, the temporal resolution issue has to be mentioned again. EP TOMS aerosol index is a 3-day composite, where previous days' retrievals are used to fill the missing information for the gaps, whereas instantaneous values of MODIS AOD were used in the analysis. The latter product often has unretrieved pixels over the specific area. Nonetheless, both satellites show a somewhat similar picture in terms of transport and development of the dust plumes. This helps to assess the transport direction and the propagation speed of dust plumes, though the differences can be critical in evaluating the transport model performance.
 On the basis of our analysis, we believe that the dust cases B14–15, C11, D10, E9, and A5 (see Figure 5 and Table 2) likely contributed to the first long-range transport dust episode, whereas cases C13–14, D11, and E10 contributed to the second one. We were not able to assess the relative contribution of dust originating in the Taklamakan (A5) to the first long-range transport dust episode from the observational data sets used in this study.
 Also, we would like to point out that only two dust outbreaks have been transported across the Pacific Ocean and reached the North American continent, as far as the Great Lakes region, during March and April 2001. Those episodes were called by the community the “perfect dust storm” [Huebert et al., 2003]. This brings up an interesting question. The year of 2001 is considered to be one of the years with highest dust activity during the last decade. Our study confirms this fact by identifying 35 dust events in the source regions only for two spring months. However, only two long-range transport dust episodes occurred. Jaffe et al.  reported one transpacific episode in 1993, one in 1997, two episodes in 1999, and three in 2001, whereas Husar et al.  discussed one long-range episode in 1998. A reasonable question is how to adequately assess the adverse impacts of long-range transport of dust over a wide span of timescales ranging from climate studies to air quality problems, if it turns out that intercontinental transport is relatively infrequent (on average, one to two severe dust episodes per 2 years). It will be important to develop a climatology of long-range transport of Asian dust by combining satellite observations and aerosol transport models to better constrain the magnitude of dust impacts over the large geographical region. On the other hand, east Asian dust undoubtedly poses a serious regional problem, affecting Mongolia, China, Korea, Taiwan and Japan [Chun et al., 2001; He et al., 2003; Kim et al., 2001].
 Complex spatial and temporal dynamics of atmospheric dust in east Asia renders predictions of dust impacts particularly difficult. In this study we utilized several ground-based and satellite data sets to characterize the east Asian dust outbreaks during April–March of 2001, as well as to evaluate the potential of different data sets in dust studies. The main goal was to examine the extent to which the existing routine ground-based observations and satellite data allow us to constrain the spatiotemporal distribution of dust plumes over the land as well as over the oceans at a range of scales. Our major findings are as follows:
 1. We demonstrated that different definitions of the dust event and its duration used by previous studies have led to conflicting results regarding the frequency and spatial patterns of dust outbreaks over east Asia. Here we introduced and consistently applied the following definition: a dust event occurs when two or more meteorological stations located in a particular source region report visibility less or equal than 5 km during at least 6 hours. Dust events are further subdivided into two types: dust storm (visibility < 1 km) and blowing dust (1 km < visibility <5 km). Being tailored for routine visibility observations conducted by meteorological stations, this definition must be taken into account while relating the dust events determined from visibility data to those predicted by the aerosol transport models or detected from satellite data. We believe that a consistent approach for characterizing the dust events that would be widely adopted by the scientific community is urgently needed to achieve a truly integrated approach in dust studies.
 2. Five main dust source regions were identified in the spring of 2001: (A) Taklamakan desert, (B) central Chinese Gobi, (C) central Mongolian Gobi, (D) eastern Chinese Gobi, and (E) eastern Mongolian Gobi. The number of dust events identified in A, B, C, D and E were 8, 10, 7, 7, and 3, respectively.
 3. Routine visibility observations appeared to be a valuable tool in constraining the regions of origin of dust outbreaks and midrange transport. However, meteorological data are of limited use for validation of dust emission schemes mainly because of a limited number of stations in dust source regions and low frequency of observations.
 4. The midrange transport routes were predominantly to the east, with several cases to the southeast-east, and southeast. These are in broad agreement with the climatology of Asian dust midrange transport reported by Chen and Chen  and Sun et al. , though some differences exist in transport routes, especially out of the Taklamakan.
 5. The source, midrange transport, and dissipation of all dust events were analyzed using surface and satellite observations. The TOMS aerosol index showed the best agreement with the visibility fields. MODIS RGB images were useful when no clouds were present. MODIS and SeaWiFS AOD data showed marginal results in terms of detecting the dust events partly because of the high cloud presence during the studied period. Another reason is a poor performance of the MODIS AOD retrieval algorithm over the land. Even for cloud-free cases, MODIS AOD over the land was retrieved only in 20% of the cases. Because of various limitations of satellite aerosol products over the land, we believe that these products should be used in conjunction with surface observations to provide a more reliable characterization of dust distribution, as well as for validation of the aerosol transport models over land.
 6. Only two long-range transport episodes were found during spring of 2001. Long-range transport of these episodes was characterized by utilizing the MODIS aerosol optical depth and TOMS aerosol index over the North Pacific Ocean. The day-by-day dust coverage reconstructed from these products show broad agreement, though various differences were observed that might be critical in evaluating the aerosol transport models. The existing problems in discrimination between dust and clouds in MODIS observations over the ocean may be an important factor [Darmenov and Sokolik, 2004].
 7. Merging surface measurements and satellite data seems to be a promising strategy for the dust study that enables one, at least to some extent, to overcome the intrinsic limitations of individual data sets. Integrated data sets could help to identify the temporal and spatial scales at which the intercomparison between dust transport models and observations would be the most efficient. However, complex dynamics of dust outbreaks and various limitations of ground-based and satellite data pose major challenges in developing an integrated analysis methodology.
 This work was supported by funds from the National Science Foundation and from the Office of Naval Research. We thank Pavel Ya. Groisman for providing meteorological data and helpful comments.