1. Introduction
Dust storms are a disastrous weather type. They arise in desert areas and release substantial amounts of aerosolized dust into the atmosphere (Engelstaedter et al. 2006; Boucher et al. 2013). This airborne dust can reduce visibility and air quality and cause respiratory and circulatory system diseases (Degobbi et al. 2011). Dust particles can be transported on regional or global scales (Shao and Dong 2006), directly or indirectly transforming atmospheric systems. This transformation further impacts regional and global climate change (Huang et al. 2014), including the hydrologic cycle (Goudie 2009; Huang et al. 2010), marine primary productivity (Mahowald et al. 2005; Han et al. 2008), solar radiation balance, cloud formation (Huang et al. 2014), and glaciers (Calov et al. 2005; Krinner et al. 2006; Bar-Or et al. 2008). Dust activity is a complex process resulting from the interaction of atmospheric circulation, soil characteristics, and the climate (Bryant 2013; Wang et al. 2015). Dust storms are influenced by soil features, large-scale atmospheric circulation, local weather systems, and human activities (Kurosaki and Mikami 2003; García et al. 2004; Liu et al. 2004; Ding et al. 2005; Engelstaedter et al. 2006; Yang et al. 2008; Huebert et al. 2003). Natural factors probably have the greatest impact on dust storms (Liu et al. 2004; Kaskaoutis et al. 2015).
Several studies have indicated that East Asia emits a large quantity of dust into the atmosphere, but no two dust budgets are completely alike. Luo et al. (2003), Miller et al. (2004), and Tanaka and Chiba (2006) estimated that dust emissions from East Asia are 214, 54, and 50 Tg yr−1 respectively, composing 11%, 3.2%, and 4.9% of the total global dust emissions. Observations and dust storm modeling indicate that the main source area of East Asia dust is the desert area of northern China (Wang 2000; Quan et al. 2001; Gallon et al. 2011; Chen and Peng 2012; Tan et al. 2012). Zhang et al. (1997) estimated that 800 Tg of dust is emitted into the atmosphere annually from the desert regions of China. Dust material in northern China is generated by aeolian erosion and uplifted into the middle troposphere by strong surface winds (Jayaratne et al. 2011; Antón et al. 2012). It is then transmitted downwind to South Korea and Japan (Mukai et al. 2004; Chung et al. 2005; Yuan and Zhang 2006) and across the Pacific to North America (Liu and Diamond 2005; Wang et al. 2011; Tan et al. 2012), the French Alps (Grousset et al. 2003), and Arctic regions (Stone et al. 2005). Because of the dramatic impacts produced by dust storms in northern China, concerns about dust storm frequency in China and East Asia have increased. Most attention has focused on dust characteristics (Sun et al. 2001; Yang et al. 2014; Guan et al. 2015) and inducement mechanisms (Kurosaki and Mikami 2003; Li and Zhai 2003; Gong et al. 2006; Wang et al. 2006; Xu et al. 2006; Zhao et al. 2013; Kang et al. 2016).
Northern China contains most of the eastern Asian deserts (Qian et al. 2004) that encompass interior Asia. Peripheral plateaus and mountains in this region act as barriers to the transport of water vapor, making it an area of extreme aridity (Huo et al. 2013; Huang et al. 2016; Wang et al. 2015). Northern China is also recognized as the major source of dust within China (Quan et al. 2001; Zhang et al. 2003; Gallon et al. 2011; Tan et al. 2012; Wang et al. 2005; Wang et al. 2015). Dust researchers in China and surrounding areas have focused on small regional characteristics (Natsagdorj et al. 2003; Liu et al. 2004; J. Zhang et al. 2005; Wang et al. 2006; Tang et al. 2013; Zhao et al. 2013; Guan et al. 2015; Liu et al. 2014) and have given less attention to long-term spatiotemporal characteristics and correlations associated with natural factors (Qian et al. 2002; Liu et al. 2004; Qian et al. 2004; Wang et al. 2005; Hara et al. 2006; Tao et al. 2010). Studies on northern China dust will increase understanding of the mechanism of dust generation and transportation, and point to approaches for controlling and combating dust storms. We selected dust storms, wind speed, temperature, and precipitation datasets from 129 sites during 1960–2007. We also used the Global Inventory Modeling and Mapping Studies Normalized Difference Vegetation Index (GIMMS NDVI) datasets during 2000–07 in northern China. These data were used to analyze the characteristics and main factors controlling dust storms in the study area. The information may enable better dust storm prediction and disaster prevention.
2. Materials and methods
a. Study site
The study area is located in northern China (31°–53°N, 73°–127°E), covering 4.17 × 106 km2 and including 11 provinces (Fig. 1). Based on altitude differences, the entire study area is successively divided into three levels from west to east. The first level is the Qinghai–Tibet Plateau, which is dominated by plateau mountains (mean altitude is 4000–5000 m); the second level is the Mongolian Plateau, Loess Plateau, and Tarim Basin, in the middle part of the study area (mean altitude is 800–3000 m); and the third level includes the North China Plain and Northeast China Plain in the east. Most of the Gobi Desert in China is distributed among the study areas.
Survey map of the study area. (I: Qaidam sandy land; II: Badain Jaran Desert; III: Tengger Desert; IV: Ulan Buh Desert; V: Mu Us sandy land; VI: Kubuqi desert; VII: Otindag sandy land; VIII: Horqin sandy land; IX: Hulun Buir sandy land; X: Kumtag Desert; XI: Taklimakan Desert; XII: Gurbantunggut Desert. A: Beijing; B: Tianjin; C: Hebei province; D: Shanxi province; E: Henan province; F: Inner Mongolia Autonomous Region; G: Shaanxi province; H: Ningxia Hui Autonomous Region; I: Gansu province; J: Qinghai province; K: Xinjiang Uygur Autonomous Region. Designations I–XII have the same meaning in Figs. 2, 4–7, 9, 10, and 12. Solid points represent the meteorological stations selected in this article.)
Citation: Journal of Climate 30, 17; 10.1175/JCLI-D-16-0795.1
b. Source of materials and data processing
Dust storm datasets (dust storm frequency, horizontal visibility, and start and end times of the dust storm) and climatological datasets [precipitation, temperature, average wind speed (AWS), and 10-min average maximum wind direction and speed] during 1960–2007 were all obtained from the Data Service Centre of the China Meteorological Administration (CMA). The 129 meteorological stations selected for this study have high-precision continuous datasets that cover the entire study area (Fig. 1).
Dust storms are defined as events where the measured horizontal visibility is <1000 m based on the standard CMA definition (Central Meteorological Bureau 1979). We selected 2000 LST [all times are local standard time (LST)] as the recording node, if there were several dust storms before 2000 LST, it was counted as 1 day only. If a dust storm event lasted beyond 2000 LST (Central Meteorological Bureau 1979), 2 days were recorded. The dust storm frequency (DSF) represents the number of dust storm days in a year or in a month, and its mean value in all sites can reveal its general standard in the entire study area. The GIMMS NDVI provides a good expression of vegetation spatiotemporal variation and has relatively accurate coverage. Ground vegetation is one of the principal factors that affect dust storm occurrence (Zou and Zhai 2004; R. Zhang et al. 2005; Xu et al. 2006). This study adopted images from January 1982 to December 2006 to provide information on vegetation coverage within the study area. The NDVI dataset was acquired from the Environment and Ecological Science Data Center for West China (http://westdc.westgis.ac.cn/; Tucker et al. 2004, 2005; Pinzon et al. 2005). The spatial resolution of this dataset is 8 km. Using these data, the DSF over the past 48 years (1960–2007) in the study area was statistically analyzed to study the characteristics of dust storm evolution.
We used the interpolation method in GIS software (ArcInfo, previously ARC/INFO) to plot the contours of the dust storm (Fig. 2a), wind speed (Fig. 4), precipitation (Fig. 6), and temperature (Fig. 8). Interpolation is a method of getting new data from some known data points. It is used widely (Vicente-Serrano et al. 2003; Cao et al. 2009; Kourgialas and Karatzas 2015), but still limited. For example, the interpolation result is insufficient to represent the status of the whole county and the interpolation contours are affected by landform. Based on this case, we increase the density of the sites as much as possible to improve the accuracy of the interpolation results. In Fig. 2b we set dust storm frequency as a quantitative index to analyze Pearson correlation between DSF and years in every site using SPSS software, and then used the correlation coefficients as spatial distribution contours. A positive correlation coefficient showed an increasing trend of dust storm and a negative correlation showed a decreasing trend. (We used the same procedure as in Fig. 2a to obtain Figs. 5a, 5b, 7a, and 9a). We used the maximum value composite method to compile the decadal NDVI through the semimonthly NDVI datasets. We also calculated NDVI changes and Pearson correlation coefficient between NDVI and DSF based on correlation analysis using the annual maximum NDVI values, years, and DSF values. The factors controlling dust storms are discussed in the context of the meteorological data and the NDVI data.
(a) Spatiotemporal pattern of mean DSF in different periods, and (b) distribution of DSF trend in northern China during 1960–2007.
Citation: Journal of Climate 30, 17; 10.1175/JCLI-D-16-0795.1
Although single-factor analysis can explain the influence of meteorological factors and NDVI on DSF to some to some extent, it cannot explain or analyze complex relationships between multiple factors. The classification and regression tree (CART) method as a multiple regression model has a number of advantages compared to numerically oriented techniques such as linear and nonlinear regression, logistic regression, artificial neural networks, cluster analysis, factor analysis (principal component analysis), and genetic algorithms (McKenzie and Ryan 1999; Breiman 2001). CART models are easy to build and interpret, and can automatically handle interactions between various data formats (continuous and categorical), but are not affected by the collinearity of variables (Breiman et al. 1984; De’Ath 2002). They have been extensively applied in vegetation (Kandrika ad Roy 2008), ecology (De’Ath 2002), gully erosion modeling (Bou Kheir et al. 2007), soil mapping (Henderson et al. 2005; Bou Kheir et al. 2008, 2010), and medical science (Schröder 2006; Valera et al. 2007; Barlin et al. 2013; Schilling et al. 2016). This paper adopted CART modeling in a simple, realistic, and practical way to analyze the influence of the maximum wind speed, AWS, MAT, MAP, and NDVI all together on dust storms in northern China. Our study set DSF as a dependent variable and meteorological factors and NDVI as independent variables for CART analysis in JMP software (version 10) (Fig. 13).
3. Results and discussion
a. Spatiotemporal variation of dust storms
The highest frequency of dust in northern China occurred in the 1960s. During this period, dust storm high-incidence areas (DSF > 5 days yr−1) were mainly concentrated in northwestern and central regions of the study area, the Mu Us sandy land and its surroundings (Fig. 2a). Dust storms in Xinjiang province can be severe, as most parts are dust storm frequency areas (DSF 3–5 days yr−1), and Taklimakan Desert and its surroundings are high-incidence dust storm areas (DSF > 5 days yr−1) (Fig. 2). From the 1970s to the 1990s, the location of the high-incidence region moved westward, and the scope narrowed and reached a minimum value in the 1990s. Meanwhile, the DSF in most areas during the 1990s was less than 1 day yr−1, with only the eastern part of Badain Jaran Desert being a high-incidence area (DSF > 5 days yr−1) (Fig. 2a). The scope of high-incidence areas (DSF > 5 days yr−1) in Xinjiang extended from the 1960s to the 1970s and peaked in the 1970s, almost covering the whole province, and then began to shrink; from 2000 to 2007 only in part of Taklimakan Desert did DSF remain above 5 days yr−1. Kang et al. (2016) studied dust storms in the Qinghai–Tibet Plateau and observed a similar decreasing trend during the 1960s–1990s. An analogous trend was also observed by Qian et al. (2004) with Inner Mongolia dust from 1980 to 1997, Hara et al. (2006) in a study of the northern China Gobi Desert, and Guan et al. (2015) in a study of the Tengger Desert from the 1960s to 1990s. Thus, both small-scale studies and the large-scale dust storm variation in the study area during the 1960s to 1990s showed a significant downward trend. Since 2000, the scope of the high-incidence dust storm region has significantly expanded, especially in the Otindag sandy land. The DSF of the Otindag sandy land increased since the 1980s and reached a maximum during 2000–07. This region also simultaneously became a high-frequency region together with the Badain Jaran Desert (Fig. 2a). The dust characteristics described above were also observed by Gao et al. (2012) in an interannual study of the Otindag sandy land dust. Overall, during 1960–2007, high-incidence dust storm areas in the study area had essentially the same distribution as deserts, and the average annual DSF was generally more than 1 day yr−1 (Fig. 2a). The spatial distribution of the DSF in a large part of the study area tended to radiate from high-frequency regions to low-frequency regions. High-frequency regions (DSF > 5 days yr−1) include the Taklimakan Desert, Badain Jaran Desert, and the surrounding areas. Low-frequency regions include the marginal zone of the Badain Jaran, the southwestern parts of the Kumtag Desert, the eastern parts of the Mu Us sandy land, and the western parts of the Otindag Hulun Buir sandy land. The general character of DSF in the past 48 years presents a falling tendency (Fig. 2b), but the most significant region was in the Mu Us sandy land and the western margin of the Taklimakan Desert. The sites where DSF increased accounted for only 15.5% of storms overall, and these were mainly distributed in the Qaidam sandy land (r = 0.576, p < 0.01), Otindag sandy land (r = 0.549, p < 0.05), and part of the Taklimakan Desert (r = 0.174).
To analyze long-term temporal variation of dust storms from different perspectives—interannual variation, seasonal variation, storm start time, and storm duration time (h)—we plotted a dust storm between-year variation curve (Fig. 3a) and monthly mean values variation curve from 1960s to 2007 (Fig. 3b) and also calculated the frequency of dust storm occurrence at different times of day during different time periods (Fig. 3c) and dust storm duration time (Fig. 3d) during the entire 48 years. The DSF in northern China fluctuated but decreased from 1960 to 2007 (Fig. 3a). The DSF showed a transient increasing trend in the early 1960s, reaching a 48-yr peak in 1966 (5.1 day yr−1). It then decreased to a minimum value in 1997 (0.44 day yr−1), after which the DSF began increasing. Overall, the DSF was greater than the 48-yr mean value in the 1960s, 1970s, and 1980s and was lower than the mean value in the 1990s and 2000–07. Therefore, the 1960s, 1970s, and 1980s were classified as relatively high-frequency periods, and the 1990s and 2000–07 were classified as relatively low-frequency periods. A similar between-year variation was observed by Qian et al. (2002). Seasonal variation of the DSF in the study area during the 48-yr period was high in spring and low in autumn (Fig. 3b). Spring (March–May) is a high-incidence dust storm season (accounting for 65.1% of the annual DSF). April is the most prominent month (DSF in April accounted for 33.4% of the annual DSF). In contrast, the DSF is relatively low in summer and autumn, with the lowest frequency occurring in September (0.4%). This strong unimodal distribution with spring maxima in China was also found by Littmann (1991), Wang et al. (2005), Xu et al. (2006), and Tao et al. (2010) in their studies. Dust storms in the study area are most likely to occur in the daytime, and less likely to occur at night (Fig. 3c). There are two susceptible daily periods (Fig. 3c), with the most significant being 1000–1300 LST (DSF: 6%–13%), followed by 1700–1900 LST (DSF: 5%–6%). The DSF in the other periods is comparatively low (<5%), among which 0200–0400 LST and 2100–2300 LST are two lowest periods (less than 1%; Fig. 3c). Dust storm duration time mostly ranged from 0 to 9 h among which the 3–6-h duration was most common (30.67%). The 1–3-h, <1 h, and 6–9-h durations accounted for 28.17%, 19.61%, and 15.15%, respectively. The 9–12-h and 12–15-h ranges were the least common durations, accounting for 4.51% and 1.39%, respectively (Fig. 3d). This daily variation pattern is consistent with the reports of Natsagdorj et al. (2003) and Wang (2005). The mean duration time and mean annual duration time of dust storms were 3.4 h per event and 11.5 h yr−1, respectively. In summary, dust storms in the study area are most likely to occur at noon and dusk during spring.
Dust storm (a) between-year variation, (b) seasonal variation, (c) diurnal variation, and (d) duration variation.
Citation: Journal of Climate 30, 17; 10.1175/JCLI-D-16-0795.1
Air instability variation caused by thermal instability plays an important role in dust storm formation (Dai 2001; Orlovsky et al. 2005; Xu et al. 2006; Yu et al. 2009; Ratnam et al. 2013; Kaskaoutis et al. 2015). The decreasing trend of dust storms in northern China from 1960 to 2007 may be due to climate warming in Mongolia and cooling in northern China that reduced the meridional temperature gradient, resulting in reduced cyclone frequency in northern China (Qian et al. 2002). In spring, heating increased land surface thawing and soil dryness, which is conducive to the occurrence of dust storm events (Xu et al. 2006). When the polar cold air moved southward and encounter with the warm currents in northern China, a significant pressure and temperature gradient was produced to promote the formation of cyclone activities and dust storms (Natsagdorj et al. 2003; Wang 2005; Zhu et al. 2008). In the summer months, the majority of dust storm events in northern China do not occur except in Taklimakan Desert. Summer is not a season with strong anticyclones, because vegetation and summer rainfall can retard the occurrence of dust storm events. The high frequency of dust storms in Taklimakan Desert may be explained by regional convective action and cyclogenesis (Littmann 1991; Sun et al. 2001). In the daytime, the solar zenith gradually increases, accompanied by strong surface heating. This causes the formation of subaerial local low pressure and reduction of atmospheric stability (Dai 2001; Xu et al. 2006), which is conducive to the occurrence of convection and dust storms. Conversely, at night, strong surface radiation causes temperature to decrease rapidly, and the temperature in the upper part of the boundary layer is higher than that in the subaerial layer. This reduces the upward transport force and produces a stable temperature layer (André and Mahrt 1982), which can weaken vertical air convection and dust activities.
b. Main factors that influence dust storms
In studies of dust storms in northern China, East Asia, and the Taklimakan desert, Kurosaki and Mikami (2003), Gong et al. (2006), and Kim and Kai (2007) reported that surface wind was the main driving force for dust storm generation, transmission, and deposition. Zhu et al. (2008) and Zhao et al. (2013) found that temperature anomalies can affect dust storm generation. Li and Zhai (2003), Wang (2005), and Xu et al. (2006) suggested that surface drying caused by precipitation reduction was a requirement for dust storm promotion. Qian et al. (2002) noted that a climate with reduced precipitation is a prerequisite for dust-forming weather. Kang et al. (2016) studied the effects of vegetation on dust storms and determined the relationship between NDVI and dust storms in the Qinghai–Tibet Plateau. Wind speed, precipitation, temperature, and vegetation growth status (NDVI) are the main factors influencing dust storms.
1) Relationship between dust storms and wind speed
The AWS and DSF in the study area were high in the north and low in the south. Regions with high (low) wind speeds had a high (low) DSF except Xinjiang, and regions with a DSF over 5 days yr−1 were mainly distributed in the southwest and north-central region of the study area that had a mean annual wind speed exceeding 2 m s−1 (Fig. 4). The DSF in most parts of the study area is positively correlated with the wind speed, with the correlation coefficient usually > 0.3 and confidence level > 95% (Figs. 5a,b). From the 1960s through the 1970s, more areas came to be affected by strong winds, including the Otindag sandy land, Qaidam sandy land, and especially northeastern Xinjiang. During these two decades, the wind speed was the strongest, the wind-affected areas were most pronounced, and the DSF was the highest (Figs. 3a, 4, and 5). In the next three decades (1980s, 1990s, and 2000–07), the wind speed had a decreasing trend, low-DSF areas were widely distributed, regions with wind speeds greater than 3 m s−1 were distributed sporadically, and regions where the DSF was > 8 days yr−1 barely existed (Figs. 2a, 4, and 5). The correlation between wind speed and DSF indicates that wind is a direct driving factor for dust storms in the study area. Because maximum wind speed can provide a better explanation for dust storm changes than AWS (Guan et al. 2015), we put the emphases on the maximum wind speed and its direction from 1960 to 2007. In most sites with prevalent northwest wind during dust storm periods, the average maximum wind speed was 12.4 m s−1. In these areas, wind speeds of 10–20 m s−1 accounted for 72.9% of events and this was the principal driving wind speed of dust storms. Compared to AWS, the correlation between maximum wind speed and dust storms is more significant (correlation coefficients between DSF and AWS, maximum wind speed is 0.5 and 0.8, respectively). This indicates that maximum wind speed is more important driving factor.
Spatiotemporal patterns of AWS and DSF during different periods.
Citation: Journal of Climate 30, 17; 10.1175/JCLI-D-16-0795.1
(a) Correlation coefficients between AWS and DSF during 1960–2007. (b) Correlation coefficients between maximum wind speed and DSF during 1971–2007. (c) Between-year variation of AWS and DSF during 1960–2007. (d) Between-year variation of maximum wind speed and DSF in the study area during 1971–2007.
Citation: Journal of Climate 30, 17; 10.1175/JCLI-D-16-0795.1
Wind speed is an important factor, but it does not work alone. During the 1970s and 1980s, greater wind speed did not correspond to a larger DSF, and there was a DSF center where the wind speed was about 2.5–3.5 m s−1. The main reason for this was that dust storm occurrence depends not only on the wind conditions but also on source material (dust, sand, etc.) and local terrain. The central region of the study area was surrounded by the Tengger Desert, Ulan Buh Desert, Mu Us sandy land, and Kubuqi Desert. Abundant sand source material together with a favorable terrain in the downwind made the central region become a high-incidence dust storm area. From 2000 to 2007, the wind velocity (including AWS and maximum wind speed) was the lowest ever, but DSF increased instead (the Hexi corridor and the Otindag sandy land became two major growth regions during this decade). This suggested that other existing factors played less obvious roles. Our analysis indicates that high-incidence dust storm areas are not necessarily those with the highest wind speeds; rather, wind speed acts in combination with topography, vegetation coverage, and other factors.
2) Relationship between dust storms and precipitation
Precipitation within the study area gradually decreased from southeast to northwest. Dust storms were mainly distributed in western and northern regions where annual precipitation was less than 450 mm (Fig. 6). Decreased precipitation during 2000–07 corresponds to increased DSF, indicating that precipitation had an inhibiting effect on the occurrence of dust storms. In one of the driest regions (the peripheral area of the Hexi corridor), the annual precipitation is less than 150 mm and the DSF ranges from 0 to 5 days yr−1. In another very dry region (Xinjiang province) DSF remains above 5 days yr−1. However, in the north-central region precipitation is relatively abundant, but DSF is high (DSF is mainly 5–8 days yr−1 but some sites have 8–15 days yr−1) (Fig. 6). The mean annual precipitation in most parts of the study area was negatively correlated with DSF but not significant (correlation coefficients ranged from −0.3 to 0). The significance level in eleven sites was less than 0.05 and three sites had a positive correlation (Fig. 7a). This suggests no obvious spatial correlation between DSF and precipitation. During 1960–2007, precipitation did not exhibit obvious between-decade variation and there was only a slight decreasing trend (Figs. 6 and 7b). However, the DSF had a more obvious downward-fluctuating trend (Fig. 6). In summary, there was no significant relationship between precipitation and DSF in terms of long-term variation. Precipitation did not provide a good explanation for long-term variation in dust storm frequency. Precipitation in Shanxi province, Shaanxi province, and part of Hebei province is positively correlated with DSF (Fig. 7). Because of the rapid spread and wide range of dust storms, DSF observed in these provinces originates from other regions such as the Gobi Desert in the Mongolia Plateau and Loess Plateau (Fig. 1), the situation that weakens correlation between DSF and local precipitation. MAP and DSF had weak correlation from the 1960s to the 1990s, but during 2000–07 the correlation coefficient increased to −0.68, and precipitation became the main control factor of dust storms during this period. Precipitation indirectly affects DSF by controlling vegetation growth and increasing the ground surface humidity, so we made a comparative analysis among simultaneous correlation and two lag correlations (Fig. 8). This analysis indicated that precipitation in previous year showed a more significant negative correlation with DSF compared to concurrent precipitation; furthermore, the previous winter’s precipitation does not affect spring dust storms significantly. Liu et al. (2004) and Lu and Liu (2006) also found a significant correlation between DSF and antecedent precipitation. The precipitation anomaly of the entire year is an important factor affecting the following spring dust storm occurrence (Liu et al. 2004).
Spatiotemporal changes of MAP and DSF during different periods.
Citation: Journal of Climate 30, 17; 10.1175/JCLI-D-16-0795.1
(a) Correlation coefficients between MAP and DSF during 1960–2007; (b) between-year variation of DSF and MAP during 1960–2007.
Citation: Journal of Climate 30, 17; 10.1175/JCLI-D-16-0795.1
Matrix line chart showing lead–lag correlation between precipitation and DSF; r is the correlation coefficient.
Citation: Journal of Climate 30, 17; 10.1175/JCLI-D-16-0795.1
3) Relationship between dust storms and temperature
High-frequency dust storm regions were mainly distributed in the north and northwest regions where the mean annual temperature was less than 10°C and were concentrated in the regions where MAT ranged from 5° to 10°C. However, DSF is relatively low in the southeastern regions where MAT > 10°C and in the southwestern and northeastern regions where the temperature < 5°C (Fig. 9). Therefore, neither high temperatures (MAT > 10°C) nor low temperatures (< 5°C) are conducive to the occurrence of dust storms (Fig. 9). In the southeast regions, areas with relatively high temperatures (MAT > 10°C), low wind speeds (AWS < 3 m s−1; Fig. 5), and abundant rainfall (MAP > 600 mm; Fig. 6) had a higher degree of vegetation coverage and lower DSF (Fig. 12).
Spatiotemporal pattern of MAT and DSF during different time periods.
Citation: Journal of Climate 30, 17; 10.1175/JCLI-D-16-0795.1
The MAT during 1960–2007 fluctuated but generally increased. This is reflected by the decadal temperature anomaly (Fig. 10b). Bounded by 1987, MAT showed two different states: before 1987, the MAT fluctuation range was small and the decadal temperature anomaly was negative, so this was a relatively cold period. Conversely, after 1987 a relatively warm period was ushered in. The relationship between temperature and DSF is complicated (Figs. 10b and 11). DSF was high (low) in relatively cold (warm) period before (after) 1987 (Fig. 10b). The MAT in the study area generally increased but warming was very conspicuous in the northern high-latitude region and high-altitude area of western China (the Qinghai–Tibet Plateau) (Figs. 9 and 10b). Although these two regions presented sites with significantly lower levels than 0.05, the MAT and DSF showed a positive correlation (Fig. 10a). This correlation is especially noticeable in the Otindag sandy land (r = 0.661, p < 0.01; Fig. 10a). There are also lag effects of temperature on DSF (Fig. 11). Concurrent temperature and temperature in the previous winter and the previous year are all correlated with DSF (the correlation coefficients are −0.435, −0.406, and −0.433, respectively) (Fig. 11). Concurrent temperature anomalies mainly alter original temperature gradient and pressure structure in the middle and high latitudes (Qian et al. 2002; Zhu et al. 2008) and then affect the formation of cyclone fronts and dust storms. Antecedent temperatures control the formation of source material to affect DSF. The lower the temperatures during the previous winter, the deeper the frozen soil extends; gradual heating in spring produces more loose material, which is conducive to the occurrence of dust storms (Lu and Liu 2006; Xu et al. 2006).
(a) Correlation coefficients between MAT and DSF in northern China during 1960–2007; (b) interannual variation of temperature anomaly and DSF in northern China during 1960–2007.
Citation: Journal of Climate 30, 17; 10.1175/JCLI-D-16-0795.1
Matrix line chart showing lead–lag correlation between temperature and DSF; r is the correlation coefficient.
Citation: Journal of Climate 30, 17; 10.1175/JCLI-D-16-0795.1
4) Relationship between dust storms and NDVI
The average maximum vegetation coverage value in the study area during 1982–2006 ranged from −1 to 1. Coverage tended to be high in the southeast and lower in the northwest. Long-term average annual NDVI was greater in the southeastern area mainly because there is more precipitation in the southeast. Vegetation dynamics are sensitive to climate change, especially in arid and semiarid regions (Wang et al. 2013). High-frequency dust storm regions were mainly distributed in the northern and northwestern areas with low vegetation coverage, and these areas are generally seen as main sources of dust storms (Sun et al. 2001; Zhang et al. 2003; Wang 2005; Wang et al. 2005; Shao and Dong 2006).
To study the variation of annual maximum vegetation coverage from 1982 to 2006, we calculated the Pearson correlation coefficient between the annual maximum vegetation coverage and years into spatial distribution diagram (Fig. 12b). Surface vegetation increased in part of study area from 1982 to 2007 (NDVI increased areas accounted for 45.6% of the total area), and concurrent DSF in most regions showed a decreasing trend except in parts of Qaidam sandy land and Otindag sandy land (Fig. 12b). The extent of surface vegetation in the study area fluctuated (fluctuation range was 0.0307–0.0375) but generally had an increasing trend (Fig. 12d). Uneven dispersion of hydroclimatic conditions combined with climate change, such as global warming observed after 1980, helps to explain the uneven distribution of vegetation cover (Fig. 12b) (Walther et al. 2002; Toreti et al. 2010; Wang et al. 2013). The spatial distribution of the correlation coefficient between NDVI and DSF from 1982 to 2006 is shown in Fig. 12c. The correlation coefficient is mainly concentrated in the −0.3 to 0.3 ranges, indicating that DSF is weakly related to the NDVI. Positive correlations were only seen in the northeastern region where vegetation coverage was high. This positive correlation may be attributed to the fact that dust storms observed in this region mostly originated from remote sources, which can weaken the relationship between the DSF and local NDVI. The areas whose correlation coefficient ranged from −0.3 to 0 accounted for more than half of the total region (Fig. 12c).
(a) Spatiotemporal pattern of NDVI and DSF during 1982–2006. (b) Distribution of NDVI and DSF trend during 1982–2006. (c) Correlation coefficients between NDVI and DSF in northern China during 1982–2006. (d) Between-year variation of NDVI and DSF during 1982–2006.
Citation: Journal of Climate 30, 17; 10.1175/JCLI-D-16-0795.1
5) CART analysis of DSF
This study used an “inverted tree” to analyze correlations between multifactorial combinations and DSF (Fig. 13). Independent variables (the maximum wind speed, AWS, MAT, MAP, and NDVI) included in single factor analysis are all put into model, and were recursively generated in a “arborescence structure” with seven terminal nodes (the minimum cross-validated relative error is 0.74). The number of observations (N) in terminal nodes ranged from 5 to 9, and the mean value of the standard deviation of the target variable (DSF) is presented in the tree chart. All DSF data are first classified into two homogeneous subgroups by maximum wind speed, at a value of 12.4 m s−1 (Fig. 13). Most data were classified in the subgroup that maximum wind speed less than 12.4 m s−1 (N≥12.4 : N<12.4 = 23 : 25), and this subgroup is further divided into two smaller and more homogeneous groups by AWS (the critical value is 3.03). Finally, the remaining data are well separated by MAP and NDVI into three terminal nodes. Meanwhile, another subgroup (maximum wind speed ≥ 12.4 m s−1) is also is divided into two smaller groups, with a critical value is 3.09 (Fig. 13). CART can identify the most decisive variables, namely those that are used for creating the splits near the top of the tree (Bou Kheir et al. 2010). Maximum wind speed strongly determines the first split and is the dominant factor of DSF, followed by AWS and MAT (Fig. 13); thus, the three most predominant factors that can affect DSF are the maximum wind speed, AWS, and MAT. MAP has only a weak effect on DSF and NDVI has the weakest effect.
CART results of DSF analysis using meteorological characteristics (the maximum wind speed, AWS, MAT, and MAP) and NDVI.
Citation: Journal of Climate 30, 17; 10.1175/JCLI-D-16-0795.1
The DSF mean value in seven terminal nodes successively increases from lower left to upper right in Fig. 13, and reaches the maximum at the right end. DSF mean value varies from variable combinations, which can show the different correlations between combinations and DSF (Fig. 13). The highest mean value occurs in the combination that contains the maximum wind speed (≥ 12.4 m s−1) and AWS (≥ 3.09 m s−1). As is stated above, these two factors are the main inducing factors of DSF (Figs. 4, 5). The higher mean values more often appear in the lower MAP/NDVI branch (Fig. 13), indicating that MAP and NDVI have inhibitory effects on the formation of dust storm. Vegetation can protect the topsoil environment and prevent dust storm creation, but it is affected by the primary decisive influence of precipitation. The fractional changes in precipitation often have significant impacts on vegetation growth especially in arid and semiarid areas. The vulnerable terrestrial ecosystems in arid and semiarid areas are susceptible to degradation and desertification due to slight climatic fluctuations (Williams and Albertson 2006). Degraded vegetation causes fine particles in the topsoil to be readily entrained into the atmosphere, with dust storms generated when wind speeds reach a threshold velocity. Thus, precipitation and vegetation can produce synergistic effects, jointly affecting DSF.
The mean value in higher MAT branches was also high (Fig. 13), but this does not imply that temperature only has promoting effects on dust storms. Single factor analysis showed that the effect of temperature on DSF is complex, with positive and negative correlations existing simultaneously (Figs. 10b and 11). In the high-altitude areas and northern high-latitude regions (MAT less than 10°C) temperatures are positively correlated with DSF (Figs. 9 and 10a) because frozen surfaces caused by low temperature can reduce the likelihood of dust being entrained by cyclones (Guan et al. 2015). Heating defrosts frozen surface soil, leading to a large amount of evaporation and more loose particulate matter, and eventually dust storms or severe dust storms are more likely to occur in the presence of strong winds (Gao et al. 2012). Temperature and DSF showed a negative correlation in most regions of the study area; this was especially conspicuous in the midlands (the correlation coefficient in most sites is less than −0.5, p < 0.01; Fig. 10a) because warming weakened cold air force and prolonged the plant growing time (Zhu et al. 2008; Kaskaoutis et al. 2015; Myneni et al. 1997; Zhou et al. 2001; Nemani et al. 2003). Stated thus, moderate warming can reduce gale weather and provide favorable vegetation growth conditions to suppress dust storm occurrence.
4. Conclusions
Dust storms in the study area are mainly distributed in the western and northern regions where AWS is >2 m s−1, MAP is less than 450 mm, vegetation is scanty, and MAT ranges from 5° to 10°C. High-incidence dust storm areas were essentially consistent with the spatial distribution of deserts in northern China and tended to radiate from high-frequency regions to the surrounding areas. The DSF in most regions fluctuated but showed a significant downward trend during 1960–2007 and was prone to occur in spring afternoons. Only parts of the Qaidam sandy land, Otindag sandy land, and the Taklimakan Desert showed an increasing DSF trend. Among them, the Qaidam sandy land showed the greatest increasing trend during the entire 48 years. DSF in the Qaidam sandy land began to rise sharply after the 1980s and gradually become the second high-incidence area. Dust storm duration times in the study area were predominantly 3–6 h, with an average value of 3.4 h.
The most three predominant factors that can affect DSF are the maximum wind speed, AWS, and MAT. Wind speed has a good corresponding relationship with dust storms during the 1960s to 1990s. Declining wind speed caused by increased temperature is the main factor leading to a significant downward trend of dust storms. Compared to AWS, maximum wind speed (r = 0.8), especially wind speed of 10–20 m s−1, has a more significant positive correlation with DSF and is a more important driving factor. The effect of wind speed on DSF became weak from 2000 to 2007. In the study area, neither high temperatures (MAT > 10°C) nor low temperatures (< 5°C) are conducive to the occurrence of dust storms. The effect of temperature on DSF is complex, with positive and negative correlations existing simultaneously. MAT is positively correlated with DSF in the high-altitude areas and northern high-latitude regions (MAT less than 10°C), as temperature affects the formation of source material (dust, etc.) to affect DSF in this case. MAT is negatively correlated with DSF in most regions of the study area and is especially conspicuous in the midlands; temperatures affect cyclone activity and vegetation growth status to control the generation of dust storms.
Both NDVI and precipitation are negatively correlated with DSF but the effect is weak. Vegetation can protect the topsoil environment and prevent dust storm creation, but is affected by the primary decisive influence of precipitation. Vegetation coverage tended to be high in the southeast and low in the northwest, similar to the distribution of precipitation in some respects. Precipitation and vegetation can produce synergistic effects, jointly affecting DSF. A positive correlation between precipitation/NDVI and dust storms also appears in some regions, which may be attributed to the fact that dust storms observed in many sites originate from remote sources.
Acknowledgments
We thank the editors and reviewers who provided comments on this paper. This work was supported by the National Natural Science Foundation of China (Grant 41671188). We also thank LetPub (www.letpub.com) for its linguistic assistance during the preparation of this manuscript.
REFERENCES
André, J. C., and L. Mahrt, 1982: The nocturnal surface inversion and influence of clear-air radiative cooling. J. Atmos. Sci., 39, 864–878, doi:10.1175/1520-0469(1982)039<0864:TNSIAI>2.0.CO;2.
Antón, M., and Coauthors, 2012: Global and diffuse shortwave irradiance during a strong desert dust episode at Granada (Spain). Atmos. Res., 118, 232–239, doi:10.1016/j.atmosres.2012.07.007.
Barlin, J., and Coauthors, 2013: Classification and regression tree (CART) analysis of endometrial carcinoma: Seeing the forest for the trees. Gynecol. Oncol., 130, 452–456, doi:10.1016/j.ygyno.2013.06.009.
Bar-Or, R., C. Erlick, and H. Gildor, 2008: The role of dust in glacial–interglacial cycles. Quat. Sci. Rev., 27, 201–208, doi:10.1016/j.quascirev.2007.10.015.
Bou Kheir, R., J. Wilson, and Y. Deng, 2007: Use of terrain variables for predictive gully erosion mapping in Lebanon. Earth Surf. Processes Landforms, 32, 1770–1782, doi:10.1002/esp.1501.
Bou Kheir, R., J. Chorowicz, C. Abdallah, and D. Dhont, 2008: Soil and bedrock distribution estimated from gully form and frequency: A GIS-based decision-tree model for Lebanon. Geomorphology, 93, 482–492, doi:10.1016/j.geomorph.2007.03.010.
Bou Kheir, R., M. H. Greve, C. Abdallah, and T. Dalgaard, 2010: Spatial soil zinc content distribution from terrain parameters: A GIS-based decision-tree model in Lebanon. Environ. Pollut., 158, 520–528, doi:10.1016/j.envpol.2009.08.009.
Boucher, O., and Coauthors, 2013: Clouds and aerosols. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 571–657.
Breiman, L., 2001: Decision-tree forests. Mach. Learn., 45, 5–32, doi:10.1023/A:1010933404324.
Breiman, L., J. H. Friedman, R. A. Olshen, and C. J. Stone, 1984: Classification and Regression Trees. Taylor and Francis, 368 pp.
Bryant, R. G., 2013: Recent advances in our understanding of dust source emission processes. Prog. Phys. Geogr., 37, 397–421, doi:10.1177/0309133313479391.
Calov, R., A. Ganopolski, M. Claussen, V. Petoukhov, and R. Greve, 2005: Transient simulation of the last glacial inception. Part I: Glacial inception as a bifurcation in the climate system. Climate Dyn., 24, 545–561, doi:10.1007/s00382-005-0007-6.
Cao, W., J. Hu, and X. Yu, 2009: A study on temperature interpolation methods based on GIS. Proc. Int. Conf. on Geoinformatics, Fairfax, VA, IEEE, doi:10.1109/GEOINFORMATICS.2009.5293422.
Central Meteorological Bureau, 1979: Standard on the Surface Meteorological Observation (in Chinese). China Meteorological Press, 186 pp.
Chen, K.-Y., and Z.-Q. Peng, 2012: Monitoring Mongolia Gobi dust transport using OMI data (in Chinese). Sci. Cold Arid Reg., 4, 446–451, doi:10.3724/SP.J.1226.2012.00446.
Chung, Y.-S., H.-S. Kim, K.-H. Park, D. Jugder, and T. Gao, 2005: Observations of dust-storms in China, Mongolia and associated dust falls in Korea in spring 2003. Water Air Soil Pollut. Focus, 5, 15–35, doi:10.1007/s11267-005-0724-1.
Dai, A., 2001: Global precipitation and thunderstorm frequencies. Part II: Diurnal variations. J. Climate, 14, 1112–1128, doi:10.1175/1520-0442(2001)014<1112:GPATFP>2.0.CO;2.
De’Ath, G., 2002: Multivariate regression trees: A new technique for modeling species–environment relationships. Ecology, 83, 1105–1117, doi:10.1890/0012-9658(2002)083[1105:MRTANT]2.0.CO;2.
Degobbi, C., F. Lopes, R. Carvalho-Oliveira, J. Muñoz, and H. Saldiva, 2011: Correlation of fungi and endotoxin with PM2.5 and meteorological parameters in atmosphere of Sao Paulo Brazil. Atmos. Environ., 45, 2277–2283, doi:10.1016/j.atmosenv.2010.12.005.
Ding, R., J. Li, S. Wang, and F. Ren, 2005: Decadal change of the spring dust storm in northwest China and the associated atmospheric circulation. Geophys. Res. Lett., 32, L02808, doi:10.1029/2004GL021561.
Engelstaedter, S., I. Tegen, and R. Washington, 2006: North African dust emissions and transport. Earth-Sci. Rev., 79, 73–100, doi:10.1016/j.earscirev.2006.06.004.
Gallon, C., M. A. Ranville, C. H. Conaway, W. M. Landing, C. S. Buck, P. L. Morton, and A. R. Flegal, 2011: Asian industrial lead inputs to the North Pacific evidenced by lead concentrations and isotopic compositions in surface waters and aerosols. Environ. Sci. Technol., 45, 9874–9882, doi:10.1021/es2020428.
Gao, T., J. Han, Y. Wang, H. Pei, and S. Lu, 2012: Impacts of climate abnormality on remarkable dust storm increase of the Hunshdak Sandy Lands in northern China during 2001–2008. Meteor. Appl., 19, 265–278, doi:10.1002/met.251.
García, J. H., W. W. Li, R. Arimoto, R. Okrasinski, J. Greenlee, J. Walton, C. Schloesslin, and S. Sage, 2004: Characterization and implication of potential fugitive dust sources in the Paso del Norte region. Sci. Total Environ., 325, 95–112, doi:10.1016/j.scitotenv.2003.11.011.
Gong, D.-Y., R. Mao, and Y.-D. Fan, 2006: East Asian dust storm and weather disturbance: Possible links to the Arctic Oscillation. Int. J. Climatol., 26, 1379–1396, doi:10.1002/joc.1324.
Goudie, A. S., 2009: Dust storms: Recent developments. J. Environ. Manage., 90, 89–94, doi:10.1016/j.jenvman.2008.07.007.
Grousset, F. E., G. Paul, A. Bory, and P. E. Biscaye, 2003: Case study of a Chinese dust plume reaching the French Alps. Geophys. Res. Lett., 30, 1277, doi:10.1029/2002GL016833.
Guan, Q., J. Yang, S. Zhao, B. Pan, C. Liu, D. Zhang, and T. Wu, 2015: Climatological analysis of dust storms in the area surrounding the Tengger Desert during 1960–2007. Climate Dyn., 45, 903–913, doi:10.1007/s00382-014-2321-3.
Han, Y., X. Fang, T. Zhao, and S. Kang, 2008: Long range trans-Pacific transport and deposition of Asian dust aerosols. J. Environ. Sci., 20, 424–428, doi:10.1016/S1001-0742(08)62074-4.
Hara, Y., I. Uno, and Z. F. Wang, 2006: Long-term variation of Asian dust and related climate factors. Atmos. Environ., 40, 6730–6740, doi:10.1016/j.atmosenv.2006.05.080.
Henderson, B. L., E. N. Bui, C. J. Moran, and D. A. P. Simon, 2005: Australia-wide predictions of soil properties using decision trees. Geoderma, 124, 383–398, doi:10.1016/j.geoderma.2004.06.007.
Huang, J., P. Minnis, H. Yan, Y. Yi, B. Chen, L. Zhang, and J. K. Ayers, 2010: Dust aerosol effect on semi-arid climate over northwest China detected from A-Train satellite measurements. Atmos. Chem. Phys., 10, 6863–6872, doi:10.5194/acp-10-6863-2010.
Huang, J., T. Wang, W. Wang, Z. Li, and H. Yan, 2014: Climate effects of dust aerosols over East Asian arid and semiarid regions. J. Geophys. Res. Atmos., 119, 11 398–11 416, doi:10.1002/2014JD021796.
Huang, J., H. Yu, X. Guan, G. Wang, and R. Guo, 2016: Accelerated dryland expansion under climate change. Nat. Climate Change, 6, 166–171, doi:10.1038/nclimate2837.
Huebert, B. J., T. Bates, P. B. Russell, G. Shi, Y. J. Kim, K. Kawamura, G. Carmichael, and T. Nakajima, 2003: An overview of ACE-Asia: Strategies for quantifying the relationships between Asian aerosols and their climatic impacts. J. Geophys. Res., 108, 8633, doi:10.1029/2003JD003550.
Huo, Z., X. Dai, S. Feng, S. Kang, and G. Huang, 2013: Effect of climate change on reference evapotranspiration and aridity index in arid region of China. J. Hydrol., 492, 24–34, doi:10.1016/j.jhydrol.2013.04.011.
Jayaratne, E. R., G. R. Johnson, P. McGarry, H. C. Cheung, and L. Morawska, 2011: Characteristics of airborne ultrafine and coarse particles during the Australian dust storm of 23 September 2009. Atmos. Environ., 45, 3996–4001, doi:10.1016/j.atmosenv.2011.04.059.
Kandrika, S., and P. S. Roy, 2008: Land use land cover classification of Orissa using multi-temporal IRS-P6 AWiFS data: A decision tree approach. Int. J. Appl. Earth Obs. Geoinf., 10, 186–193, doi:10.1016/j.jag.2007.10.003.
Kang, L., J. Huang, S. Chen, and X. Wang, 2016: Long-term trends of dust events over Tibetan Plateau during 1961–2010. Atmos. Environ., 125, 188–198, doi:10.1016/j.atmosenv.2015.10.085.
Kaskaoutis, D. G., A. Rashki, E. E. Houssos, A. Mofidi, D. Goto, A. Bartzokas, P. Francois, and M. Legrand, 2015: Meteorological aspects associated with dust storms in the Sistan region, southeastern Iran. Climate Dyn., 45, 407–424, doi:10.1007/s00382-014-2208-3.
Kim, H. S., and K. Kai, 2007: Recent dust outbreaks in the Taklimakan Desert and their relation to surface wind and land surface condition. SOLA, 3, 69–72, doi:10.2151/sola.2007-018.
Kourgialas, N. N., and G. P. Karatzas, 2015: Groundwater contamination risk assessment in Crete, Greece, using numerical tools within a GIS framework. Hydrol. Sci. J., 60, 111–132, doi:10.1080/02626667.2014.885653.
Krinner, G., O. Boucher, and Y. Balkanski, 2006: Ice-free glacial northern Asia due to dust deposition on snow. Climate Dyn., 27, 613–625, doi:10.1007/s00382-006-0159-z.
Kurosaki, Y., and M. Mikami, 2003: Recent frequent dust events and their relation to surface wind in East Asia. Geophys. Res. Lett., 30, 1736, doi:10.1029/2003GL017261.
Li, W., and P. M. Zhai, 2003: Variability in occurrence of China’s spring dust storm and its relationship with atmospheric general circulation (in Chinese). Acta Meteor. Sin., 17, 396–405.
Littmann, T., 1991: Dust storm frequency in Asia: Climatic control and variability. Int. J. Climatol., 11, 393–412, doi:10.1002/joc.3370110405.
Liu, J., and J. Diamond, 2005: China’s environment in a globalizing world. Nature, 435, 1179–1186, doi:10.1038/4351179a.
Liu, Q., Y. Liu, J. Yin, M. Zhang, and T. Zhang, 2014: Chemical characteristics and source apportionment of PM10 during Asian dust storm and non-dust storm days in Beijing. Atmos. Environ., 91, 85–94, doi:10.1016/j.atmosenv.2014.03.057.
Liu, X., Z.-Y. Yin, X. Zhang, and X. Yang, 2004: Analyses of the spring dust storm frequency of northern China in relation to antecedent and concurrent wind, precipitation, vegetation, and soil moisture conditions. J. Geophys. Res., 109, D16210, doi:10.1029/2004JD004615.
Lu, Z., and C. H. Liu, 2006: On the relationship between sandstorm and meteorological factors in China (in Chinese). Meteor. Mon., 32, 35–41.
Luo, C., N. M. Mahowald, and J. D. Corral, 2003: Sensitivity study of meteorological parameters on mineral aerosol mobilization, transport, and distribution. J. Geophys. Res., 108, 4447, doi:10.1029/2003JD003483.
Mahowald, N. M., and Coauthors, 2005: Atmospheric global dust cycle and iron inputs to the ocean. Global Biogeochem. Cycles, 19, 1064–1067, GB4025, doi:10.1029/2004GB002402.
McKenzie, N. J., and P. J. Ryan, 1999: Spatial prediction of soil properties using environmental correlation. Geoderma, 89, 67–94, doi:10.1016/S0016-7061(98)00137-2.
Miller, R. L., I. Tegen, and J. Perlwitz, 2004: Surface radiative forcing by soil dust aerosols and the hydrologic cycle. J. Geophys. Res., 109, D04203, doi:10.1029/2003JD004085.
Mukai, M., T. Nakajima, and T. Takemura, 2004: A study of long-term trends in mineral dust aerosol distributions in Asia using a general circulation model. J. Geophys. Res., 109, D19204, doi:10.1029/2003JD004270.
Myneni, R. B., C. D. Keeling, C. J. Tucker, G. Asrar, and R. R. Nemani, 1997: Increased plant growth in the northern high latitudes from 1981 to 1999. Nature, 386, 698–702, doi:10.1038/386698a0.
Natsagdorj, L., D. Jugdea, and Y. S. Chung, 2003: Analysis of dust storms observed in Mongolia during 1937–1999. Atmos. Environ., 37, 1401–1411, doi:10.1016/S1352-2310(02)01023-3.
Nemani, R., C. D. Keeling, H. Hashimoto, W. M. Jolly, S. C. Piper, C. J. Tucker, R. B. Myneni, and S. W. Running, 2003: Climate-driven increases in global terrestrial net primary production from 1982 to 1999. Science, 300, 1560–1563, doi:10.1126/science.1082750.
Orlovsky, L., N. Orlovsky, and A. Durdyev, 2005: Dust storms in Turkmenistan. J. Arid Environ., 60, 83–97, doi:10.1016/j.jaridenv.2004.02.008.
Qian, W., L. Quan, and S. Shi, 2002: Variations of the dust storm in China and its climatic control. J. Climate, 15, 1216–1229, doi:10.1175/1520-0442(2002)015<1216:VOTDSI>2.0.CO;2.
Qian, W., X. Tang, and L. Quan, 2004: Regional characteristics of dust storms in China. Atmos. Environ., 38, 4895–4907, doi:10.1016/j.atmosenv.2004.05.038.
Quan, L., S. Shi, Y. Zhu, and W. Qian, 2001: Temporal–spatial distribution characteristics and causes of dust-day in China. Acta Geogr. Sin., 4, 477–485, doi:10.11821/xb200104011.
Pinzon, J., M. E. Brown, and C. J. Tucker, 2005: Satellite time series correction of orbital drift artifacts using empirical mode decomposition. Hilbert–Huang Transform: Introduction and Applications, N. Huang, Ed., World Scientific, 167–186.
Ratnam, M. V., Y. D. Santhi, M. Rajeevan, and S. V. B. Rao, 2013: Diurnal variability of stability indices observed using radiosonde observations over a tropical station: Comparison with microwave radiometer measurements. Atmos. Res., 124, 21–33, doi:10.1016/j.atmosres.2012.12.007.
Schilling, C., D. Mortimer, K. Dalziel, E. Heeley, J. Chalmers, and P. Clarke, 2016: Using classification and regression trees (CART) to identify prescribing thresholds for cardiovascular disease. Pharmacol. Econ., 34, 195–205, doi:10.1007/s40273-015-0342-3.
Schröder, W., 2006: GIS, geostatistics, metadata banking, and tree-based models for data analysis and mapping in environmental monitoring and epidemiology. Int. J. Med. Microbiol., 296, 23–36, doi:10.1016/j.ijmm.2006.02.015.
Shao, Y., and C. H. Dong, 2006: A review on East Asian dust storm climate, modelling and monitoring. Global Planet. Change, 52, 1–22, doi:10.1016/j.gloplacha.2006.02.011.
Stone, R., G. Anderson, and E. Andrews, 2005: Asian dust signature at Barrow: Observed and simulated incursions and impact of Asian dust over northern Alaska. Proc. Workshop on Remote Sensing of Atmospheric Aerosols, Tucson, AZ, IEEE, 74–79.
Sun, J., M. Zhang, and T. Liu, 2001: Spatial and temporal characteristics of dust storms in China and its surrounding regions, 1960–1999: Relations to source area and climate. J. Geophys. Res., 106, 10 325–10 333, doi:10.1029/2000JD900665.
Tan, S.-C., G.-Y. Shi, and W. Hong, 2012: Long-range transport of spring dust storms in Inner Mongolia and impact on the China seas. Atmos. Environ., 46, 299–308, doi:10.1016/j.atmosenv.2011.09.058.
Tanaka, T. Y., and M. Chiba, 2006: A numerical study of the contributions of dust source regions to the global dust budget. Global Planet. Change, 52, 88–104, doi:10.1016/j.gloplacha.2006.02.002.
Tang, Y., G. Han, Q. Wu, and Z. Xu, 2013: Use of rare earth element patterns to trace the provenance of the atmospheric dust near Beijing, China. Environ. Earth Sci., 68, 871–879, doi:10.1007/s12665-012-1791-z.
Tao, G., Zhang, X., and Wulan, 2010: A seasonal forecast scheme for spring dust storm predictions in northern China. Meteor. Appl., 17, 433–441, doi:10.1002/met.175.
Toreti, A., F. Desiato, G. Fioravanti, and W. Perconti, 2010: Seasonal temperatures over Italy and their relationship with low-frequency atmospheric circulation patterns. Climatic Change, 99, 211–227, doi:10.1007/s10584-009-9640-0.
Tucker, C. J., J. E. Pinzon, and M. E. Brown, 2004: Global Inventory Modeling and Mapping Studies, NA94apr15b.n11-VIg, version 2.0. University of Maryland Global Land Cover Facility, accessed 15 October 2016. [Available online at http://staff.glcf.umd.edu/sns/branch/htdocs.sns/data/gimms/.]
Tucker, C. J., J. E. Pinzon, M. E. Brown, D. Slayback, E. W. Pak, R. Mahoney, E. Vermote, and N. El Saleous, 2005: An extended AVHRR 8-km NDVI data set compatible with MODIS and SPOT vegetation NDVI data. Int. J. Remote Sens., 26, 4485–5598, doi:10.1080/01431160500168686.
Valera, V. A., B. A. Walter, N. Yokoyama, Y. Koyama, T. Iiai, and H. Okamoto, 2007: Prognostic groups in colorectal carcinoma patients based on tumor cell proliferation and classification and regression tree (CART) survival analysis. Ann. Surg. Oncol., 14, 34–40, doi:10.1245/s10434-006-9145-2.
Vicente-Serrano, S. M., M. A. Saz-Sánchez, and J. M. Cuadrat, 2003: Comparative analysis of interpolation methods in the middle Ebro valley (Spain): Application to annual precipitation and temperature. Climate Res., 24, 161–180, doi:10.3354/cr024161.
Walther, G. R., and Coauthors, 2002: Ecological responses to recent climate change. Nature, 416, 389–395, doi:10.1038/416389a.
Wang, H., X. Jia, K. Li, and Y. Li, 2015: Horizontal wind erosion flux and potential dust emission in arid and semiarid regions of China: A major source area for East Asia dust storms. Catena, 133, 373–