Identification of dust sources and hotspots in East Asia during 1 2000-2015 : implications for numerical modeling and 2 forecasting 3

Xuelei Zhang 1, 3 ; Daniel Q. Tong 2,8,9 ; Guangjian Wu 3 ; Xin Wang 4 ; Aijun Xiu 1,5 ; Yongxiang 4 Han 6 ; Tianli Xu 3,7 ; Shichun Zhang 1 ; Hongmei Zhao 1 5 6 1 Key laboratory of Wetland Ecology and Environment, Northeast Institute of Geography 7 and Agroecology, Chinese Academy of Sciences, Changchun 130102, China 8 2 U.S. NOAA Air Resources Laboratory, College Park, MD 20740, USA 9 3 Key Laboratory of Tibetan Environment Changes and Land Surface Processes, Institute 10 of Tibetan Plateau Research, CAS Center for Excellence and Innovation in Tibetan 11 Plateau Earth System Sciences, Chinese Academy of Sciences, Beijing 100101, China 12 4 College of Atmospheric Sciences, Lanzhou University, Lanzhou 730000, China 13 5 Institute for the Environment, University of North Carolina at Chapel Hill, Chapel Hill, 14 North Carolina, USA 15 6 Key Laboratory for Aerosol-Cloud-Precipitation of China Meteorological Administration, 16 Nanjing University of Science Information &Technology, Nanjing, 210044, China 17 7 University of Chinese Academy of Sciences, Beijing 100049, China 18 8 Cooperative Institute for Climate and Satellites, University of Maryland, College Park, 19 Maryland, MD 20740 20 9 Center for Spatial Information Science and Systems, George Mason University, Fairfax, 21 Virginia, VA 22030 22 23 24 Correspondence to: X. L. Zhang (zhangxuelei@neigae.ac.cn) and G. J. Wu (wugj@itpcas.ac.cn) 25 26 Abstract 27


Introduction
Mineral dust is a major component of atmospheric aerosols and affects various aspects of the earth-climate system (Knippertz et al., 2014), including radiation balance (Huang et al., 2006), precipitation (Creamean et al., 2013;Vinoj et al., 2014) and cloud cover (Solomos et al., 2011;Atkinson et al., 2013), Ocean and terrestrial biogeochemical cycles (Maher et al., 2010), air quality and visibility (Wang et al., 2012), and even human health (Goudie et al., 2014).The relative impacts of dust within the land-atmosphere-ocean system depend on physiochemical characteristics such as particle size distribution, morphology, and mineralogy (Formenti et al., 2011b;Mahowald et al., 2011;Zhang et al., 2015b), which, although subject to modification during transport (Formenti et al., 2011a), are functions of the source from which they are derived (Shao et al., 2011;Perlwitz et al., 2015).Furthermore, whether derived directly from the point of origin or re-entrained along its pathway, dust may become mixed with other natural or/and anthropogenic materials, potentially causing further harm to wildlife, plants, and humans.Knowledge of the specific sources and hotspots (extraordinally active sources) of Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-681, 2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 10 October 2016 c Author(s) 2016.CC-BY 3.0 License.dust emissions at different geographic scales (global, regional and sub-regional) is crucial for improving the prediction accuracy of dust events by numerical models of climate and air quality (Wang et al., 2012;Heinold et al., 2015).Three approaches are frequently used to identify dust sources: frequency statistics, model simulation, and remote sensing.Given the complexity of dust mobilization processes, analyses of storm frequencies have several weaknesses that may result in uncertainties: observer bias is an intrinsic limitation, and the correlation between dust storm frequency and dust sources is not necessarily direct, because both transported dust and locally raised dust can reduce visibility.Numerical modeling is another popular method of identifying dust sources (e.g., Ginoux et al., 2001;Park et al., 2010), but this method is crucially dependent on the parameterization of dust emissions.Different initial inputs will result in variations in source identification.To improve numerical dust models, obtaining more detailed information on dust emission sources is crucial.The newly implemented Saharan source maps, based on sources directly identified by satellites, have been adopted to verify improvements in dust models (Parajuli et al., 2014;Heinold et al., 2015).
Since surface observations are generally sparse in desert regions, it is difficult to locate dust emission sources and subsequent dust trajectories following a dust outbreak.This problem of coarse horizontal-resolution observations can be overcome with the aid of satellite remote sensing.Data from AVHRR (Tegen and Fung, 1995), POLDER (Deuze et al., 2001), TOMS (Prospero et al.,2002), MODIS (Ginoux et al., 2012;Vickery et al., 2013), SeaWiFS (Eckardt and Kuring, 2005), GOES (Wang et al., 2003), MISR (Kalashnikova et al., 2005), SEVIRI (Ashpole and Washington, 2013), OMI (Bryant et al., 2007), AIRS (DeSouza-Machado et al., 2010) and IASI (Klüser et al., 2011) have been successfully used for dust monitoring and source identification.Satellite imaging can also be used for dust modeling and forecasting, now that long-term satellite data on dust storms are available.However, there are constraints to identify dust sources using satellite imaging.
Polar orbiting satellites typically collect only one snapshot per day, which may introduce temporal and spatial biases in source detection; for example, the aerosol index from TOMS or OMI may not be used to determine dust origin sources (Darmenova et al., 2005;Baddock et al., 2009).Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-681, 2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 10 October 2016 c Author(s) 2016.CC-BY 3.0 License.
At the global scale, a low-resolution (2.5°) map of dust sources estimated from the TOMS Aerosol Index has lead to measurable improvements in modeling global distribution of dust emissions (Ginoux et al., 2001), and more recently a synthesis of global-scale high-resolution (0.1°) dust source locations has been developed based on MODIS Deep Blue AOD products which consider potential anthropogenic dust emissions (Ginoux et al., 2012).Engelstaedter and Washington (2007) correlated the TOMS Aerosol Index with wind speed and gustiness, and identified 131 global hotspots of dust emissions during 1984 to 1990.Nine of these were located in East Asia, but due to the coarse resolution their spatial coverage was uncertain and there is limited geomorphological understanding of these sources.It is clear that inherent seasonal and diurnal dust emission variations, along with the spatial heterogeneity of dust sources in global and mesoscale models, are poorly constrained due to inaccurate source alocation and quantification (Uno et al., 2006).Average aerosol optical depth (AOD) over a long time series has been used to detect persistent, regional-scale dust sources (e.g.Ginoux et al., 2012), but accurate and event-specific source identification requires clear delineation of the upwind margin of the plumes.However, it is apparent that the detection of spatially discrete and intermittent sources can be undertaken using moderate resolution polar-orbiting data, and a few studies have focused on detection techniques or compilation of regional hotspots worldwide: for example, in Australia (Baddock et al., 2009), the Middle East (Karimi et al., 2012;Jafari and Malekian, 2015), Africa (Knippertz and Todd, 2010;Ashpole and Washington, 2013;Vickery et al., 2013), North America (Rivera et al., 2010;Lee et al., 2012) and Central Asia (Nobakht et al., 2015).More recently, dry lakes, riverbeds, mines and croplands contributing to dust emissions over eastern East Asia have been identified as hot spots on the basis of high-resolution MODIS images (Zhang et al. 2015a).Nevertheless, an inventory of dust sources and hot spots based on satellite observations at the sub-basin scale is still nonexistent for East Asia as a whole.
Furthermore, it is well known that there is considerable interannual variability in dust emission and transport, and several studies have predicted future variability on the basis of different analyses.Based on the significant negative correlation between surface air Atmos.Chem. Phys. Discuss., doi:10.5194/acp-2016-681, 2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 10 October 2016 c Author(s) 2016.CC-BY 3.0 License.temperature around Lake Baikal and dust storm frequency, Zhu et al. (2008) demonstrated that the future dust storm frequency in spring over the East Asia was anticipated to continuously decrease after the year 2007.Meanwhile, another forecast predicted that sandstorm occurrence in northern China will increase gradually, entering a new, relatively active period (Li and Zhong, 2007).Obviously, large discrepancy and uncertainties still remain in the predications, and there is need to collect more ground-or satellite-based observations to assess the directionality and accuracy of the prediction.
Such assessment can be also be used to re-evaluate and improve our knowledge on regional dust emission and transport by air quality and climate models.
The goals of the research presented in this paper are twofold.First, it will provide a unified, regional and sub-regional dust sources and hot spots inventory for East Asia, to improve numerical modeling of dust emission and transport, and to consider measures to mitigate wind erosion.Second, this study will present annual and seasonal variations of dust events for the period of 2000-2015 to reassess past predictions and improve our knowledge on the regional patterns of dust emission.This paper is structured as follows.Section 2 describes the satellite platforms and observation data, the analysis method and the numerical models.In Section 3, we demonstrate the interannual variations of dust emissions and the distribution of dust hotspots.Relationships with climate indices and implications for dust numerical prediction are then discussed in Section 4. Finally, Section 5 presents our conclusions.(Sayer et al., 2013).Details of the remote sensing data are outlined in Table 1.observed to be attached to a source but could still be attributed to likely emission points based on back trajectory analysis.

Dust detection algorithm
As advised by Baddock et al. (2009), at the regional scale, dust events can be detected using MODIS AOD or OMI AI products.If either of these two products indicates the presence of dust, then there is the potential for determining dust sources at higher resolution.The MOD/MYD04 aerosol products provide data processed to a common standard that enables comparison from one region to another.The versatile MOD/MYD02 data can be processed simply by using brightness temperature difference to enhance the dust signal.Where cloud is present, or if it is necessary to highlight the dust plume, then one of the methods for employing a dust/non-dust threshold can be used, but it is recommended that event-specific thresholds are calculated manually (as opposed to uniform regional thresholds).The algorithm for the higher resolution technique based on satellite data is illustrated in Figure 1.
The ability to use remotely-sensed data both to detect a dust plume and to identify the location from which it has originated is affected by several factors including the radiative transfer properties of the material emitted, the radiative properties of the ground/ocean surface over which the plume is transported, the size and density of the dust plume, the time of satellite overpass relative to dust emission, the presence or absence of cloud, the  Many previous studies have proposed approaches to discriminate airborne dust over bright land surfaces (Ackerman, 1997;Baddock et al., 2009), including the use of true color images, ultraviolet band absorption (Washington et al., 2003), thermal infrared techniques (Ackerman, 1997), and other more complicated algorithms.Both visible bands (VIS) and infrared bands (IR) (often combined) can be used to discriminate dust.When the dust layer is optically thick, the dust particles cause a negative brightness temperature difference (BTD).It has been shown that the BTD methods are the most consistently reliable technique for dust source identification over bright surfaces (Karimi et al., 2012;Baddock et al., 2016).Thus, IR bands (band 31 and band 32) of the MODIS onboard Aqua and Terra satellites are used to detect dust sources over East Asia in this study.
In order to select a fast and effective algorithm to detect massive dust events with different intensities, we reviewed the mainstream "split windows" algorithms for dust detection (Table 2).Note that some algorithms have not been used in the East Asia region.
After detailed comparisons and evaluation of different algorithms to detect our pre-selected moderate dust event (May 11, 2011) over East Asia, Method 4, which combined visible bands and infrared bands, was selected as being the most suitable algorithm for differentiating dust from land and clouds in this study (Figure 2).This selection is consistent with Baddock et al. (2009)  for visualizing dust, but there are significant problems with precise source identification and determination of dust plume extent.For the majority of events and algorithms, the published or indicative thresholds under-perform and the values vary from event to event.This makes it difficult to suggest appropriate regional scale thresholds and each event was manually adjusted in this study.
While some of this variation is due to factors specific to the algorithms or individual events, other factors such as diurnal and seasonal variations in surface temperature/dust contrast (which affect BTD) will affect all the methods.
Once a dust event was determined, it was numbered and classified as either a local or regional transported dust event.Then, all the dust plumes in this event were recorded and the locations (points or/and polygons) of hot spots for dust emission were noted.to four statistical scores (Anderson and Brode, 2010).Thus, we decided to use it here to calculate three-dimensional trajectories of each transported dust event over a 15-year period from 2000 to 2015.Forward trajectories for all detected dust events were obtained from the National Centers for Environmental Prediction/National Centre for Atmospheric Research reanalysis data (NCEP/NCAR; http://rda.ucar.edu/datasets/ds090.0/).
Trajectory starting locations were determined by central coordinates of identified point or/and polygons.Model vertical velocity used the meteorological model's vertical velocity fields, and the trajectory was simulated at 3 starting heights (500, 1000 and 1500 m above mean sea level) for each transported dust event.

Interannual and seasonal variations of dust events
Following the analysis of MODIS images we identified 462 dust events and 214 long-lived (>48 hours) dust events during 2000-2015 (Table 1 in the appendix).The frequencies and timing of dust emissions have changed markedly over the past 16 years (Figure 3a).  in summer and autumn have been comparable or slightly greater than those in the winter months (i.e.December, January and February).Source regions were mostly located in Inner Mongolia and Northeastern China (named as Manchuria in Table 1 in the appendix), which is consistent with the results of Lim and Chun (2006) and Kim and Lee (2013).The annual frequency of Asian dust events in winter remained stable but with fluctuations (0-4 counts) over the period 2000-2014.Our statistical results based on satellite retrievals show fluctuations corresponding to those in ground-based observations (Yang et al., 2008 and official reported data from CMA), but higher than those in statistical results of transported dust events in satellite retrieves over East Asia (Zhang et al., 2008).
The periodicity of transported dust storms was analyzed using power spectrum analysis from 2000 to 2015 in East Asia which revealed a cyclical period of 3-5 years.This period is consistent with the calculated short cycles (3-4 years) that correspond with the period of El Nino and Southern Oscillation (Littmann, 1991;Wang et al., 2005;Hara et al., 2006;Lee et al., 2015).A long period of 11-12 years corresponding to the sunspot cycle was also identified in the dust storm frequency time series for Asia and Middle East (Wang The variations in dust events appear to be controlled by atmospheric dynamics (temperature, precipitation, wind velocity, Mongolia cyclone, polar vortex and Arctic Oscillation), land surface characteristics (soil moisture, desertification, vegetation) and human activities.Several studies have found that dust storm frequency has decreased over East Asia during the past 60 years (Wang et al., 2005;Ding et al., 2005).Since the late 1970s, both observations and simulations have shown that the magnitudes of dust events (both in terms of dust frequency and dust concentration) have been decreasing over the western part of Northern China (Shao et al., 2013;Guan et al., 2015), and even the Tibetan plateau (Han et al., 2008;Kang et al., 2016), except for the Taklimakan desert (Mao et al., 2011, Tan et al., 2014;Yang et al., 2015).However, an increasing trend was detected in the eastern part of Northern China (Lee et al., 2011;Gao et al., 2012;Tan et al., 2014), and in the whole of Mongolia (Lee et al., 2011;Kurosaki et al., 2011).
In addition to the contraction of the Gobi Desert (Sternberg et al., 2015) and the overgrazing-driven expansion of the Hunsandak Desert and the Horqin Desert in the past 16 years (Gao et al., 2012;Tan et al., 2014), precipitation is also an important factor affecting dust events.Records show that rainfall has increased over the past half century in northwestern China, while rainfall has decreased over Mongolia and Eastern Inner Mongolia since the 1990s (Ding et al., 2005;Kurosaki et al., 2011;Gao et al., 2012).
Furthermore, Wang et al. (2006) reported that dust storm frequency was low in the eastern part of Northern China, where there are high levels of human activity, indicating that intensive land use did not contribute to dust storm occurrence.However, land use change has been found to be a major factor influencing changes in dust storm frequency in Xinjiang and Northeast China (Tan et al., 2014).Thus, what is the true impact of human activity on regional dust emission in the past 16 years?This question will be further investigated at the sub-regional scale with the assistance of high-resolution satellite images in Section 3.3.3.2 Distribution of dust sources at the regional scale In recent years, identifications of dust sources in East Asia have been mainly conducted on global or/and regional scales.Furthermore, different techniques and methods (compilation of literatures, field investigation, geochemical analysis, satellite observation and numerical modeling) have been applied to accurately locate the dust sources (Xuan et al., 2004;Zhang et al., 2006;Kim and Lee et al., 2013) and to quantify the effects of climate change and anthropogenic activities on dust emissions.In contrast to former studies, the sub-regional scale (for dust sources) and fine scale (for hotspots) were adopted in this paper to establish a more comprehensive source map for dust emission.In order to effectively characterize the sub-regional features and simplify modeling comparisons, different division schemes adopted in former studies are depicted in Figure 5a, 5b and 5c (Zhang et al., 2003;Ku and Park, 2011;Kim and Lee, 2013).According to the MODIS image series, we further subdivided the dust emission sources into six distinguishable sub-regions according to their regional characteristics and frequency of dust events (Figure 5d).Among the 214 dust events in China and Mongolia identified by MODIS, most originated in regions S1, S3 and S5 (Table 1 in the supplemental information; Figure 5d).The three main sources (Taklimakan Desert, Gobi, sandy lands in Region 1), in terms of numbers of dust events, accounted for over three quarters of the total dust emission events in East Asia, in contrast to the results of Zhang et al. (2008).
Propelled by cold frontal systems and the Mongolian cyclone, dust from the five regions generally moves eastward, but with some regional differences in transported direction (Figure 6a).

Region 1
The Region 1 sources mainly cover the Taklimakan, Gurbantunggut and Kumtag Basin, where dry river beds and salt lakes lie between the desert and alluvial fans (Washington et al., 2003).However, the suspended dust was always transported along the edge of the basin from northeast to southwest, partially along the Hexi Corridor (Gao and Washington, 2010).As the entry of cold air masses is blocked by the high elevation and the dust-laden atmosphere is poorly ventilated, most local dust events originating in the Taklimakan desert are not easily transported out of the Tarim Basin, and instead are mainly deposited on the windward slopes of the Kunlun Mountains (which could be verified by AOD distribution in Figure 9).
The Taklimakan Desert's role as a strong dust emission source has been identified by frequency statistics (Kurosaki et al., 2003), numerical modeling (Laurent et al., 2005;Tanaka et al., 2006;Wang et al., 2008) and satellite monitoring (Engelstaedter and Washington, 2007;Gao and Washington, 2010;Waggoner and Sokolik, 2010).Once the dust is uplifted by strong winds to elevations exceeding 5000 meters, the dust may be transported eastward, climbing over the Hexi Corridor to begin its long-distance transport over East Asia, even reaching North America and Greenland (Uno et al., 2009).Xuan et al. (2004) suggested that the Taklimakan desert and the Tibetan Plateau are active dust sources only in the late spring and early summer, because the solid frozen crust prevents dust emission.Our results show that the first dust event over the Taklimakan desert was detected in late winter and early spring.It is likely that the surface of the intensely arid desert does not form a solid crust when frozen.

Region 2
Region 2 sources include the Junggar Basin with the Gurbantunggut Desert, the Great Lakes Basin in western Mongolia and Gobi Desert in southwestern Mongolia covering three provinces-Bayankhongor, Govi-Altai and Zavkhan.The Aibi Lake region, Gurbatunggut Desert and agricultural development region in Kelamayi are three major sources for dust emission (Qian et al., 2007).The southeastern side of the Altay Mountains and the Central Gobi-desert together form one of the most frequent dust sources in Mongolia (Tsolmon et al., 2008).Previous studies have shown that the maximum dust emission rates occur in Outer Govi-Altai, in the same place as the maximum aridity, and the greatest occurrence of dusty days occurred in the Gobi Desert and the Great Lakes hollow of west Mongolia (Natsagdorj et al., 2003;Xuan et al., 2004).
MODIS monitoring showed that this region was the source of more than 19% of the dust events originating in East Asia.

Region 3
Region 3 sources include the Gobi and sandy deserts in the Hexi Corridor, and the Alxa Plateau (including Badain Juran Desert, Tengger Desert and Ulan Buh Desert in north-central China).Guaizihu and Minqin have been reported as the two major dust storm centers in the Alxa Plateau based on observational data from meteorological stations from 1961 to 2005 (Yao et al., 2011).Variability of the climate has had less impact on aeolian desertified land expansion than that of human activities over this region (Wang et al., 2013).
MODIS observations during 2000-2015 showed that this region accounted for about 30% of the total number of dust events.Many "hot spots" or "dust plumes" along the China-Mongolia corridor were important point-source contributors to dust emissions from this region during 2000 -2013.These point sources mainly comprised dry lakes, river beds and alluvial fans.Juyan Lake and Guaizihu were the two areas with the most frequent dust emissions in this region.

Region 4
Region 4 sources mainly comprise the Qubqi Desert, Mu Us Sandy Land on the Ordos Plateau and the Loess Plateau.The Loess Plateau is a potential dust emission source because its surface soil consists of fine silt and clay particulates.However, our observations revealed the low frequency of dust storm occurrences on the Loess Plateau (5 dust events), except in its northwest part in Gansu province and Ningxia Hui autonomous region which are adjacent to the regional deserts.Xuan et al. (2004) also found that the vast area of the Loess Plateau was not a strong dust source when compared to other regions, due to the high clay content in the surface soil.Mountain, over the Bohai Sea, and then across the Korean Peninsula to the Sea of Japan and beyond.

Region 6
Region 6 sources are mainly made of the Horqin Sandy Land, the saline and alkaline land around the northeast China plains, the Hunlun Buir desert, and the Moltsog-Els and Ongon-Els sandy lands of eastern Mongolia.In this region, there are also dozens of small sources associated with pasture, dry lakes, river beds, or alluvial fans.
Compared to dust source divisions adopted in three former studies (Figure 5), the dust source area around Lake Balkhash should be excluded from the East Asia region and reassigned to Central Asia (Xin et al., 2015).The role of the Tibetan Plateau has become controversial and will be further discussed in section 3.3.The central part of eastern China (D2 in Figure 5a), which is mainly composed of dry cropland with irrigation, was also excluded as a major dust source in this study, and wind erosion over this region needs further detailed study.Besides the regular eastward (to Korea and Japan) and southeastward (to Taiwan and Hong Kong) transport routes, our trajectory analysis also shows that the dust could even been transported directly northward to the Far East region by the Mongolia cyclone.Two former studies have demonstrated that the dust emitted from East Asia could reach the Arctic Circle via the Japan Sea (Cahill, 2003;Huang et al., 2015).The transported dust potentially affects the climate and ecosystem of the Arctic region, thereby complicating the process of climate change in the Arctic (Di Pierro et al., 2011).

Hotspots of dust emission on a sub-regional scale
The main aim of this section is to detect the dust source hotspots by means of the MODIS images.In some instances, dust plumes may be discernible in the MODIS VIS (e.g. Figure 7); but this was certainly not always the case.Average AOD values over long time series have been used to detect persistent, regional scale dust sources (e.g.Ginoux et al., 2012) but accurate and event-specific source identification requires the clear delineation of the upwind margin of the plumes.Rather than being sourced from large homogeneous areas, much of the global supply of dust comes from hotspots, which are small but relatively consistently active dust-producing areas (Gillette, 1999).Hotspots in the form of dry lakes, river beds, alluvial fans, mines, and croplands can contribute to dust emissions in the arid and semi-arid areas of Africa and Asia (Bullard et al., 2008;Zhang et al., 2015a).Identification of hot spots within the huge areas identified as sources of dust on a regional scale is necessary to improve numerical modeling of dust emission and transportation, to develop effective countermeasures to hinder wind erosion, and to allow precautions to be taken against dust-related health problems.(3) An area with the highest dust emission frequency and magnitude.
Individual or multiple simultaneously active hotspots can be distinguished within the satellite images at the beginning of dust plumes.Based on the identified 214 dust storm events in this study, we also investigated the spatial-temporal distribution and inventory of hotspots over East Asia.Four major regions with several hotspots were discriminated, these were regions S1, S3, S5 and S6 shown Figure 5d.
By analyzing more than 1326 MODIS images, we identified hotspots scattered across the dust source region of East Asia and considered their potential for dust emission on the basis of their land cover.Table 3 summarizes the most viable hotspots and the erodibility features with which they are likely to be associated as inferred from Google Earth, Global Mosaics of the standard MODIS land cover type data product (MCD12Q1) of China and Mongolia (http://glcf.umd.edu/data/lc/,Channan et al., 2014) and visual field observations.
Of these sources, the majority (eighteen) feature dry lakebeds and paleolakes or bare riverbeds and paleochannels; seven feature outwash fans over steep hillsides; and three hotspot sub-regions feature barren sandy land or saline and alkaline land.Note that with the exception of the highlighted sources in Table 3, it is also important that seasonally bare cropland regions (e.g.Northeast China Plain and North China Plain) can contribute, albeit infrequently and weakly, as sources of dust storms.Winds in excess of the thresholds needed to start saltation are presumably exceeded in these areas, but these threshold winds do not evidently occur frequently and there is a limited time when surface soil is free from vegetation and prone to wind erosion (Zhang et al., 2015a).
Due to the significant differences in threshold wind speed and vegetation index between Gobi and sandy lands (Laurent et al., 2005), our results show that hotspots in the Gobi appear to be larger than those in the sandy lands, and dust emissions were more intense over these hotspots (Figure 7).The spatial distribution of 24 identified hotspots along with dust event counts is illustrated in Figure 8.It is obvious that the high-frequency hotspots are located in western China, and the medium-frequency hotspots are located in southern Mongolia and northeastern China.This spatial distribution of dust hotspots is also consistent with the spatial variation of dust events based on ground observations in former studies (see Section 3.1).In order to verify the accuracy and further understand the regional effects of these 24 hotspots, we next mapped the average deep blue AOD at 550 nm over terrestrial East Asia from March 2000 to October 2015 (Figure 9).Besides the two well-known heavy air pollution areas (the southern rim of the Tibetan Plateau, covered with brown cloud; and eastern China, covered with haze), half of the hotspots could be distinguished at the regional scale with a resolution of 1° × 1°.This implies that the significant contribution of dust hotspots to regional air quality cannot be neglected.However, while higher AOD values were observed across the northern and central Tibetan Plateau (white dashed Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-681, 2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 10 October 2016 c Author(s) 2016.CC-BY 3.0 License.rectangle in Figure 9), no obvious hotspots were identified over the Tibetan Plateau.This could be because dust is uplifted and transported from the southern rim of the Qaidam basin under the prevailing winds, subsequently becoming dispersed across the Tibetan plateau because of the 'elevated heat pump' effect as indicated by 3-D satellite observations of aerosol over this region coupled with the assistance of CALIPSO (Gao and Washington, 2010;Jia et al., 2015;Xu et al., 2015).Thus, we concluded that the Tibetan plateau is a receptor region of transported dust but not a significant dust emission source area in East Asia.Higher AOD values were also observed in central northern China (blue dashed rectangle in Figure 9), owing to the anthropological aerosol emissions from city groups in oases within the arid region.The anthropogenically related, high AOD values under the blue dash-dot lines over Northern Indian subcontinent and Eastern China are also represented in Figure 9.The three most important areas, containing several hotspots, are described in more detail in this section.The most frequent hotspots are located at the northeastern corner of the Tarim Basin, precisely over Lop Nur centered at ~90°E, 40°N.The surrounding mountains act as barriers which complicate the circulation pattern over the basin.Lop Nur was a large saline lake (~2000 km 2 ) in the 1930s which dried up in 1962 (Li et al., 2008).
Water diversion projects on the Tarim River, which drains the Tarim Basin, reduced the inflow to such a degree that the lake is now a dry salt lake, largely salt-encrusted and subject to severe wind erosion.Interestingly, the frequency of dust plumes has obviously increased over this region (Label A, B and C) since 2010.Meanwhile, a dramatic expansion of brine-evaporation pools occurred: from 25 km 2 before 2009 to 180 km 2 after 2010.Furthermore, an extensive network of artificial canals with a total length of 145 km was excavated to collect infiltrated brine, which would further accelerate the loss of surface water and development of drought in Lop Nur (Figure 10).However, a more detailed mechanism linking the development of saline lakes and increased dust events over Lop Nur needs to be further investigated in our future works.3) of the Alashan Plateau, along with dust hotspots distributed along the dried lake beds of Juyan and Guaizi Lakes (Figure 7c and 7d).The Juyan Lakes consist of West Juyan Lake (Gaxun Nur) (42.5°N, 100.7°E) and East Juyan Lake (Sogu Nur) (42.3°N, 101°E) at the terminus of the Hei River originating from the north flank of the Qilian Mountains.These lakes dried up in 1961 and in 1994.In these potential emission areas, contrary to expectation, the soils are not sandy desert but instead are semi-lithified deposits of fine-grained mud/silt substrates, according to remote sensing analysis with field validation (Wang et al., 2004;Figure 3 in Yang et al., 2008 andFigure 11d in Taramelli et al., 2012).Protecting grasslands in the lower reaches of the river basin from degradation and rehabilitating the dried-up terminal lake would be highly beneficial in reducing dust plumes in the region.
The third focused area comprises the dried lake beds of the Wulagai Lake group and barren grass or vegetated lands in the Otindag Sandy Land (Label W in Table 3 and    Figure 7a and 7b).In this grassland area, coal mining industries had caused rapid shrinkage and even drying up of the Wulagai Lake group by 2004 (Tao et al., 2015).
The fourth area is distributed over the Horqin sandy land and the saline soils in western Jilin Provence (Label X in Table 3).According to the positions of the origins of dust plumes identified in this study, hotspots contributed much to the dusty weather experienced over Northeastern China, and the dust was transported to Korea, Japan and even to far eastern Russia.
An additional two dry lake beds in Mongolia, at Boon Tsagaan Nuur (45.6°N, 99.1°621 E) and Oroi Nuur (45.1°N, 100.7°E), were also identified as dust hotspots.The central part of Mongolia (Labels I, K, L and U) is fed by fine-grained material from alluvial fans and ephemeral steams and is therefore highly susceptible to wind erosion.
The use of protective farming techniques, afforestation and water conservation in dust emitting basins, along with dust suppression and protection of water resources in mining areas, should be considered to combat dust emissions in the hot spot areas identified in this study.Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-681, 2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 10 October 2016 c Author(s) 2016.CC-BY 3.0 License.

Implications for dust modeling and forecasting
The above discussion revealed that East Asia is a complex, inhomogeneous dust production region comprising various types of individual sources, each having distinct properties and different strengths.Dust models combine dust source information with predictions of atmospheric dynamics to forecast the occurrence of dust events.The key component of a dust forecast model is its treatment of dust sources and emissions, however, heterogeneous dust source maps are presented in different models in East Asia (Figure 11).If the dust source map is inaccurate or ambiguous, the emission flux and spatial-temporal distribution of the modeled regional dust distribution will be erroneous.
The fact that global dust emissions are controlled by a few very productive sources provides useful information to design efficient mitigation strategies (Engelstaedter and Washington, 2007;Haustein et al., 2015).The concept of preferential dust sources (Ginoux et al., 2001;Bullard et al., 2011) acts to nudge models towards the observed dust emission patterns by relaxing the threshold emissions and, in essence, removes the surface crusting issue from the modeling process.Nevertheless, none of the current model emission schemes is able to reproduce the spatial distribution of the major dust sources correctly (Haustein et al., 2015).
Traditionally, models have used the bare ground categories of land cover maps to locate dust sources.However, new model representations of dust sources are based on topographic, hydrologic, and geomorphologic considerations; alternatively they are derived directly from satellite data, for example considering the surface bareness, topographical depression features and soil freezing and thawing.The latter approach includes surface reflectance, frequency of high aerosol values, and ultraviolet-visible albedos.Waggoner and Sokolik (2010) suggested that soil with a high silt-to-clay differential or ratio will have a higher albedo and act as a preferential source for potential dust emission.These new representations provide a much more refined view of global and regional dust source regions.As mentioned above, we divided the dust source functions/maps presented earlier into eight categories: 1) topography (Ginoux et al., 2001); 2) topography combined with soil bareness (Kim et al., 2013); 3) surface roughness (Laurent et al., 2005;Koven and Fung, 2008); 4) geomorphological classification (Bullard et al., 2011); 5) back-tracking by satellite observation (Schepanski et al., 2009); 6) a threshold-frequency method using ground or satellite observations (Park et al., 2010;Schepanski et al., 2012); 7) the self-organizing map (SOM) neural network method (Lory et al., 2016); and 8) a dynamic physical-mechanism method (Kok et al., 2014;Zhang et al., 2015c).However, more detailed inter-comparisons between the above eight methods need to be conducted over the East Asia region in future works to improve the modeling and forecasting of dust events, in particular the weak, small-scale dust emissions.to their behavior as dust sources, based on current understanding of the geomorphological controls on dust emissions.The authors also pointed that this scheme can be applied to map potential modern-day dust sources in four major dust source regions (the Chihuahuan Desert, the Lake Eyre Basin, the western Sahara and the Taklimakan), primarily using remote-sensing imagery to classify surfaces, and thus is suitable for global application.However, while detailed geomorphological mapping has been achieved for some regions, at present there is no standardized methodology or dataset for global scale coverage.Furthermore, the method has only been specifically validated for the Chihuahuan Desert in North America and the Lake Eyre Basin in Australia (Bullard et al., 2011;Baddock et al., 2016).Comparing plots of dust sources over the Taklimakan in Figure 2 of Bullard et al. (2011) with Figures 8 and 9 in this paper, we observe that the geomorphological method performs poorly over East Asia and needs further careful validation.
As mentioned in Section 3.1, the frequency of dust events can increase in other seasons besides the spring, and thus the regional dust storm forecasting system should be operational for the whole year rather than intermittently (for example, the Asian dust forecasts from the Korea Meteorological Administration are only provided from March to May (http://web.kma.go.kr/eng/weather/asiandust/forecastchart.jsp).Based on demands for forecasting hazards related to poor air quality, traffic visibility and resident health, a framework for sub-regional forecasting of the onset and development of dust plumes and their effects on downwind regions should be urgently established in local meteorological organizations, especially in the regions containing several of the hotspots identified in Table 3.

Conclusions
In this paper, the use of MODIS products has achieved significant improvements in quantifying atmospheric dust: first in the assessment of the temporal variability of dust events, and then in identifying dust sources and hotspots for the period 2000-2015.The combined use of MODIS L1B data and the BTD algorithm provided an effective method of dust discrimination and hotspot validation.Furthermore, the dust emission source locations were divided into six distinct sub-regions according to their geographical characteristics and frequency of dust events, thereby aiding future regional modeling studies.Comparing dust source maps and functions in current dust models with those applied to the whole East Asia has also revealed that heterogeneous distribution of dust sources is still one of the major factors that affects the prediction accuracy of dust events in East Asia.
Having systematically analysed the satellite data, we highlight the following key findings and implications.
1) A relatively detailed inventory of dust events in East Asia during the past 16 years has been established based on satellite observations.A slightly increasing tendency was observed during the study period; however, the seasonal variability was significant, and the frequency of dust events in spring sharply decreased while the frequency slightly increased in summer and autumn.
2) It is clear that the Tibetan Plateau is not an important dust source region in East Asia, and the contribution of this region has been overestimated in some previous studies.hotspots.Further studies need to develop our understanding of the relationships between small-scale changes in land cover, population and industrial growth, and changes in the redistribution of hot spots.The hotspots identified on the basis of MODIS L1B data provide small-scale information about dust emissions and can be used to improve both our understanding of regional-to global-scale dust cycles and numerical modeling of dust emission and transport.
4) Current dust models commonly use semi-empirical dust source functions to help parameterize spatial variability of dust emissions.A high-resolution (1 km) dust source database for East Asia is still lacking.The dynamic, physically-based approach used in dust models reduces the need to use an empirical source function in global dust cycle simulations; however, down-scaling this scheme to the mesoscale needs to be further verified.This also emphasizes the importance of reliable and higher resolution soil texture data.
There is still work to do with respect to establishing high-resolution gridded datasets of the dust sources and hotspots identified in this study, and the real test of our findings will only come when these data are implemented and compared with other predefined sources within a dust-cycle model.In this context, we note that the human activities, especially the unreasonable exploitation of water resources, are another key factor which may cause dry lake beds to become dust hotspots and exacerbate the regional dust emissions.We will quantify this human contribution to regional dust emission and climate forcing in a forthcoming study using a regional climate model.

Acknowledgement
This research was funded by NSFC (Grant No. 41571063, 41205108 and 21407148).

2. 1
Data MODerate Resolution Imaging Spectroradiometer (MODIS) collects observations in 36 spectral bands with wavelengths from 0.41 to 14.4 µm and nadir spatial resolutions of 1 km, 0.5 km, and 0.25 km.It is currently operating onboard the NASA Earth Observing System (EOS) Terra and Aqua satellites, launched in December 1999 and May 2002, respectively.The higher temporal and spectral resolutions of MODIS improve its dust identification capability over those of previous-generation earth-observing systems.Daily MODIS Level 1B (L1B) 1 km data (MOD021KM=Terra, and MYD021KM=Aqua) used in Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-681,2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 10 October 2016 c Author(s) 2016.CC-BY 3.0 License.this work have been processed to convert the sensor's on-orbit responses in digital numbers to radiometrically calibrated and geo-located data products (v5.06 processing for Terra and v5.07 for Aqua).Data were obtained from the National Snow and Ice Data Center (NSIDC; http://nsidc.org/)and the Level 1 and Atmosphere Archive and Distribution System (LAADS; http://ladsweb.nascom.nasa.gov/).Daily MODIS Level 2 Aerosol data are available as a 10 × 10 km resolution (at nadir) pixel array.There are two MODIS Aerosol data product file types: MOD04_L2, containing data collected from the Terra platform and MYD04_L2, containing data collected from the Aqua platform.Here we only use the MYD04 Aqua product because Deep Blue (see below) retrievals are not yet available for MOD04 Terra data.Improvements in the surface reflectivity retrieval and algorithm mean that Collection 6 MODIS Deep Blue aerosol products, both absolute AOD and its spectral variation, have changed since Collection 5.1 horizontal and vertical plume trajectory, and the sensor characteristics and radiative transfer model used to detect dust.Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-681,2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 10 October 2016 c Author(s) 2016.CC-BY 3.0 License.

Figure 1
Figure1Flowchart of dust detection algorithms used for satellite data in sub-regional scale

Figure 2 .
Figure 2. Comparison and evaluation of different algorithms applied to detect a pre-selected moderate dust event (May 11, 2011) over East Asia.(a) True color; (b) Retrieved AOD; (C) BTD method of Ackerman; (d) BTD method of Roskovensky and Liou.

Figure 4 .
Figure 4. Review of potential source areas over East Asia in the published literature (updated from Formenti et al., 2011b)

Figure 5 .
Figure 5.Comparison of our optimized division of dust source regions with previous partition schemes in regional dust modeling over East Asia.
Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-681,2016   Manuscript under review for journal Atmos.Chem.Phys.Published: 10 October 2016 c Author(s) 2016.CC-BY 3.0 License.3.2.5 Region 5 Region 5 sources mainly include the Gobi Desert around Ulaan-nuur Lake in Central-Eastern Mongolia, the Gobi Desert around Dornogov province of Southeastern Mongolia (which includes the Doolodyn Gobi-desert, the Ooshiyn Gobi-desert and Dalay-Els Sandy Land) and the Otingdag Sandy Land of China.MODIS monitoring shows that this region is the principal contributor (42%) to long-range dust transport to the North Pacific.Dust from Region 5 can follow two main transport pathways (Figure 6a): eastward, rising over the Da Xinganling mountain ranges, to the North Pacific; or southeastward, climbing over Yishan Mountain and Taihang Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-681,2016   Manuscript under review for journal Atmos.Chem.Phys.Published: 10 October 2016 c Author(s) 2016.CC-BY 3.0 License.

Figure 6 .
Figure 6.Dust sources and transport pathways identified by MODIS and HYSPLIT model.(a) The purple solid arrow indicates the uplift of transported dust.(b) Three major trajectories of transported dust reaching the Arctic.

Figure 8 .
Figure 8. Spatial distribution of 24 identified hotspots with labels from Table 3 along with their dust event counts, in East Asia (Grey shades represent deserts, yellow shades represent semi-deserts and the stripped lines represent the distribution of losses).

Figure 9 .
Figure 9. Time-averaged map of deep blue aerosol optical depth at 0.55 um over East Asia from March 2000 to December 2015 (Bold letter labels corresponded to the identified hotspots in Figure 8; the

Figure 10 .
Figure 10.Historical MODIS images showing the development of a saline lake factory with marked artificial canals for infiltrated brine in Lop Nur, Taklimakan desert.
Meanwhile, northeast China, one of the major dust sources, has been overlooked or underestimated by most previous modeling studies, yet dust from this region can be transported to the Far East.Besides the regular eastward (to Korea and Japan) and southeastward trajectories (to Taiwan and Hong Kong), our trajectory analysis has also shown that the dust can even be transported northwards to the Far East region by the Mongolia cyclone.The transported dust potentially affects the climate and ecosystems of the Arctic region, therefore complicating the impact of climate change in the Arctic.3)Twenty-four dust hotspots were identified in East Asia, of which the high-frequency hotspots were located in western China, and the medium-frequency hotspots were located over southern Mongolia and northeastern China.Anthropogenic activities appear to be the dominant causes of dust emissions from the medium-frequency Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-681,2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 10 October 2016 c Author(s) 2016.CC-BY 3.0 License.
The project was also supported by the Open Research Fund of Key Laboratory of Tibetan Environmental Changes and Land Surface Processes, Chinese Academy of Sciences (Grant No. TEL201504).We thank the National Snow and Ice Data Center (NSIDC) and the Level 1 and Atmosphere Archive and Distribution System (LAADS) for the provision of satellite data.The authors gratefully acknowledge the NOAA Air Resources Laboratory (ARL) for providing the HYSPLIT transport and dispersion model used in this publication.Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-681,2016   Manuscript under review for journal Atmos.Chem.Phys.Published: 10 October 2016 c Author(s) 2016.CC-BY 3.0 License.

Table 1
Detailed spatial and temporal information of remote sensing data used for dust detection in this study

Table 2
Summary of dust detection algorithms applied to MODIS L1B data.

Table 3 .
Details of 24 identified hotspots for dust emission with erodibility features over East Asia, based on MODIS data, during 2000-2015.Index of typical events with identified dust plumes are listed in Sheet 2 of the appendix.542 * Table 4．Comparisons of dust source functions/maps in different dust numerical models covering East Asia Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2016-681,2016 Manuscript under review for journal Atmos.Chem.Phys.Published: 10 October 2016 c Author(s) 2016.CC-BY 3.0 License.