Effect of Hydro-meteorological and Surface Conditions on Variations in the Frequency of Asian Dust Events

Abstract

The effects of hydro-meteorological and surface variables on the frequency of Asian dust events (FAE) were investigated using ground station and satellite-based data. Present weather codes 7, 8, and 9 derived from surface synoptic observations (SYNOP)were used for counting FAE. Surface wind speed (SWS), air temperature (Ta), relative humidity (RH), and precipitation were analyzed as hydro-meteorological variables for FAE. The Normalized Difference Vegetation Index (NDVI), land surface temperature (LST), and snow cover fraction (SCF) were used to consider the effects of surface variables on FAE. The relationships between FAE and hydro-meteorological variables were analyzed using Z-score and empirical orthogonal function (EOF) analysis. Although all variables expressed the change of FAE, the degrees of expression were different. SWS, LST, and Ta (indices applicable when Z-score was < 0) explained about 63.01, 58.00, and 56.17% of the FAE,respectively. For NDVI, precipitation, and RH, Asian dust events occurred with a frequency of about 55.38, 67.37, and 62.87% when the Z-scores were > 0. EOF analysis for the FAE showed the seasonal cycle, change pattern, and surface influences related to dryness condition for the FAE. The intensity of SWS was the main cause for change of FAE, but surface variables such as LST, SCF, and NDVI also were expressed because wet surface conditions suppress FAE. These results demonstrate that not only SWS and precipitation, but also surface variables, are important and useful precursors for monitoring Asian dust events.

1. Introduction

Asian dust events, particularly dust storms in East Asia, affect ocean and human health (Goudie, 2014) and occur frequently from spring to early summer(Lee and Sohn, 2011) over arid and semi-arid regions such as the Gobi Desert, Mongolia, and Inner Mongolia (Kimura et al., 2009). Previous studies (Kurosaki and Mikami, 2003; Kim et al., 2013) have indicated that Asian dust events occur and that their intensities increase in strong and upward wind, low precipitation, and high temperature conditions. With regard to surface wind speed (SWS), higher frequency of dust outbreak is related to the increase in the frequency of strong wind that occurs from March to April (Kurosaki and Mikami, 2003; Ge et al., 2011). In particular, Kimura (2012b) reported that Asian dust events are strongly associated with strong wind speed conditions above 7 m/s. Aside from strong surface wind conditions, effects of other hydro-meteorological and surface conditions on Asian dust events have also been reported (Kimura, 2012a). Vegetation cover can influence the frequency of dust outbreaks (Zou and Zhai, 2004; Lee and Sohn, 2009). Xu et al.(2006)found that vegetation and snow cover are the most significant surface parameters associated with Asian dust events. Regarding the effects of vegetation on Asian dust events, a negative relationship between the Normalized Difference Vegetation Index (NDVI) and Asian dust outbreaks has been reported (Zou and Zhai, 2004). Snow cover alsoplays a role in the occurrence of Asian dust events (Laurent et al., 2005). The impacts of soil cohesion (Kim and Choi, 2015) and reduced soil moisture (SM) (Kim et al., 2013) on Asian dust events have been studied because SM is an important parameter affecting interactions between the surface and the atmosphere (Entin et al., 2000). Evapotranspiration has also been shown to play an important role in the interaction of the surface and the atmosphere, along with SM and precipitation (Boé, 2012; Berg^, 2014). Land surface temperature (LST) is influenced by other surface and hydrologic variables, such as vegetation fraction, albedo, and evapotranspiration (Sandholt et al., 2002). 

Hydro-meteorological and surface variables are associated with Asian dust events directly and/or indirectly, and analyzing the effect of each of these variables is important to understand the frequency of Asian dust events(FAE).Liu et al.(2004) demonstrated the relationships among anomalies of wind, precipitation, SM, and vegetation, and Kim et al.(2013) reported that changes of SM impact FAE. The effects of wind and vegetation on Asian dust events and changes of vegetation influenced by temperature and precipitation have been studied using empirical orthogonal function (EOF) analysis (Lee and Sohn, 2009; Park and Sohn, 2010). As previous studies have mentioned, understanding the anomaly patterns of hydro-meteorological and surface variables is necessary for understandingthe surface and atmospheric conditions over the region of dust origin, and analyzing the principal components of FAE using EOF analysis is also important. Therefore, it is necessary to find the most suitable variables and analyze the effects of these hydro-meteorological and surface variables on FAE. 

This study has two main objectives. First, appropriate hydro-meteorological and surface variables that serve as precursors for monitoring the source of Asian dust events and that can be used to improve the capability of forecasting such events using ground and satellite-based data are determined. Z-scores for these hydro-meteorological and surface variables are calculated and compared with the status of each variable. Second, the effect of the hydro-meteorological and surface variables on FAE is determined using EOF analysis, and the meaning of the principal components of the EOF modes are analyzed.

2. Data and method

1) Study area and data

The study area extends from 90°E to 125°E and 30°N to 50°N (Fig. 1). The land cover consists of mixed forest, open shrubland, grassland, cropland, and barren or sparsely vegetated land, following the InternationalGeosphere–Biosphere Programme (IGBP) classification scheme. This study area includes many locations in which Asian dust originates, such as the Gobi Desert, Inner Mongolia, and Manchuria. LST and precipitation are high during summer and snow cover fraction (SCF) increase in winter, because the climate of this region is affected by the Asian monsoon. 

Fig. 1. Study area. The circles represent the ground observation stations that recorded Asian dust events from 2001 to 2014. Circle size indicates the cumulative frequency of Asian dust events during the study period. The background is Moderate Resolution Imaging Spectroradiometer (MODIS) land cover using the International Geosphere-Biosphere Programme (IGBP) classification scheme.

Meteorological variables such as present weather code (WW), SWS, relative humidity (RH), and air temperature (Ta) from surface synoptic observations (SYNOP) were used in this study. These variables were observed at 194 meteorological stations and provided at 3 h intervals. The present weather code includes information about Asian dust events. Present weather codes 7, 8, and 9, which are related to the occurrence of Asian dust at the time of observation at or near ground stations (Kurosaki and Mikami, 2005), were used to analyze the relationship between FAE and the hydro-meteorological and surface variables. However, present weather codes 30-35 were excluded because this study focused on the presence or absence of occurrence of Asian dust events at or near ground stations depending on conditions of the hydrometeorological and surface variables. Table 1 summarizes the meaning of these codes. Meteorological variables observed near the surface such as SWS, Ta, and RH were considered ith WW, but friction velocity and pressure pattern were not considered in this study because this study did not consider dust transport. SWS is known as an important variable causing Asian dust (Liu et al., 2004), and Ta is associated with changes of vegetation (Park and Sohn 2010). RH was considered because moisture in the atmosphere affects Asian dust (Yang et al., 2005). In this study, cumulative values of present weather codes related to Asian dust events with respect to the dust origin were used, and SWS, Ta, andRH were averaged monthly. Fig. 1 shows the distribution of Asian dust event occurrences based on SYNOP data and Moderate Resolution Imaging Spectroradiometer (MODIS) land cover for this study period. The colored circles in the figure indicate the cumulative FAE during the study period of 2001 to 2014.

Table 1. Meanings of code numbers for present weather (WW) according to the World Meteorological Organization (WMO)

Satellite-based hydro-meteorological and surface products were also used because of the spatial limitations of SYNOP data. Although the temporal resolution of satellite-based data is longer than that of ground-based data, satellite-based variables have the advantage of allowing estimation of physical values in areas unobserved by ground meteorological stations. LST, vegetation index (VI), and SCF products retrieved from the MODIS sensor onboard the Terra satellite were used. The MODIS LST (MOD11C3 collection 5) and theMODIS NDVI (MOD13C2 collection 5) products used in this study have a spatial resolution of 0.05° and a temporal resolution of one month. The unit of the MOD11C3 product is degree Kelvin (K). The accuracy of the Terra/MODIS LST during daytime over Northwest China is from -0.91 K to -3.13 K (Li et al., 2014a), and this product underestimates at the desert sites(Wan, 2014). NDVI, one of the widely used vegetation indices retrieved from satellite data, was generally calculated using the reflectance in red and near infrared (NIR) channels, as follows:

\(\mathrm{NDVI}=\frac{\mathrm{RNIR}-\mathrm{R}_{\mathrm{Rad}}}{\mathrm{R}_{\mathrm{NIR}}+\mathrm{R}_{\mathrm{Red}}}\)       (1)

where R_NIR and R_Red are the surface reflectances at the NIR (MODIS Band 2) and Red (MODIS Band 1) channels, respectively.

The MODIS SCF (MOD10CM collection 5) product provided by the National Snow and Ice Data Center (NSIDC) was used to analyze the correlation between FAE and SCF. Laurent et al. (2005) assumed that Asian dust could not occur if snow exists. Asian dust commonly occurs after the season of melting snow in East Asia. Therefore, snow cover information is essential for studying Asian dust. The spatial resolution of MOD10CM is 0.05°, and the range of SCF is expressed as from 0 to 100%. The accuracy, determined by comparing with ground and other snow maps, is normlly higher than 93%, based on MOD10_L2 swath snow maps(Hall et al., 2001; Simic et al., 2004; Ault et al., 2006; Hall and Riggs, 2007). The Version 7 Tropical Rainfall Measuring Mission (TRMM) 3B43 product made by merging microwave satellites such as the TRMM Microwave Imager (TMI), Special Sensor Microwave Imager (SSMI), Special Sensor Microwave Imager/Sounder (SSMIS), AMSR-E, Advanced Microwave Sounding Unit (AMSU), Microwave Humidity Sounder (MHS), microwave-adjusted merged geo-infrared (IR), and ground-based measured data to improve accuracy was used for precipitation data (Huffman et al., 2009). TRMM 3B43 is one of the TRMM Multi-sensor Precipitation Analysis (TMPA) products that has high performance in arid regions (Yang and Luo, 2014). Satellite-based real-time precipitation products have limited accuracy because of the low temporal resolution of low earth orbit satellites that cannot monitor continually and the lack of microwave sensors for geostationary satellites; thus, TMPA products were developed. TRMM 3B43 has 0.25° and monthly resolution, and the unit is mm·month-1. Yang and Luo (2014) showed that TRMM 3B43 products perform well in arid regions.

In our study, variables such as SWS, Ta, RH, and precipitation, which can be observed by an automatic weather system (AWS), were defined as hydrometeorological variables. NDVI, LST, and SCF were defined as surface variables because these variables represent surface conditions. The spatial resolution o the different hydro-meteorological and surface variables, such as those of the MOD11C3, MOD13C2, and MOD10CM products, were converted from a spatial resolution of 0.05° to a resolution of 0.25° by taking the simple average to match the resolution of the satellite-based precipitation products. Table 2 summarizes the characteristics of the data used in this study.

Table 2. Ground- and satellite-based data

WW = the present weather

2) Method

(1) Z-score

Asian dust events occur under dry surface conditions during the spring season. Kim et al.(2013)showed that these events occur more frequently when the SM value is lower than the eight-year averaged SM value. In this study, the Z-scores of the hydro-meteorological and surface variables were used to determine the effect of each variable on the variation of FAE and to compare the variability according to the same standard (MuñozDíaz and Rodrigo, 2004). The Z-score can express dryness conditions of variables about normal year that defined as the mean from 2001 to 2014. The Z-score for the variable of interest (X) was computed using the monthly average (\(\bar{X}\)) and the monthly standard deviation (σx) as follows (Mu et al., 2013):

\(Z-\operatorname{score}(X)=\frac{X-\bar{X}}{\sigma_{X}}\)       (2)

whereX is SWS, NDVI, LST, Ta, precipitation, orRH, respectively. Z-score ranges from unlimited negative to unlimited positive values (Mu et al., 2013). The change of FAE depending on the hydro-meteorological and surface conditions was analyzed using this method.

(2) EOF analysis

Asian dust events have diverse characteristics and are affected by various external factors. Thus, understanding the specificity of Asian dust is important for reducing damage caused by Asian dust. This study used EOF analysis to understand the diverse spatial and temporal characteristics of FAE. m × n matrices for FAE, consisting of hydro-meteorological and surface variables, were set using m locations of meteorological stations and n times of Asian dust events at each month in the study period (Björnsson and Venegas, 1997). When such m × n matrices are set in space and time, EOF analysis has the following characteristics (Dommenget and Latif, 2002; Li et al., 2014b). First, the EOF analysis can explain the principal components for each variable. Second, the EOF analysis is useful for describing the variability of spatial patterns. Third, the EOF patterns and time series are independent. Fourth, the EOF patterns do not have physical meaning but are affected by time and space. With consideration to the four characteristics, some meteorological stations in the study area that had discontinuous data were excluded to obtain a more accurate analysis. Each EOF mode for FAE an the hydro-meteorological and surface variables were analyzed, and the spatial patterns, called eigenvectors, were interpolated using the linear interpolation method to examine the spatial patterns.

3. Results

1) Seasonal characteristics

Fig. 2 shows a histogram of the monthly frequency and monthly mean percentage of Asian dust events. In this study, approximately 64.37% of the events occurred in spring (March–April–May) and the remainder occurred in late winter and early summer (February and June, respectively). This result is consistent with the result of a previous study, in which Asian dust events over NortheastChina (HexiCorridor) occurred during spring about 65.12% (53.64%) of the time from 1954 to 2000 (Wang et al., 2004).

Fig. 2. Histogram of Asian dust events according to month. Gray bars (blue circles) indicate the frequency (percentage) of events.

The seasonal trends of SWS, NDVI, LST, Ta, precipitation, and RH from 2001 and 2014 are shown in Fig. 3. The red (blue)lines and shadings are the mean and standard deviation values, respectively, of the variables in occurrence (non-occurrence) area of Asian dust events within the dust origin area that experienced Asian dust events more than once per year. The green lines are mean values of the variables in non-dust origin area, defined as area in which Asian dust events occurred less than once per year.

Fig. 3. Temporal trends of hydo-meteorological and surface variables from 2001 to 2014 over East Asia. Red lines are averaged temporal trends and orange shading indicates the standard deviation of the averaged temporal trends when Asian dust occurred in areas (main area of dust origin area) that experienced Asian dust events more than one time per year. Blue lines and light blue shading are averaged temporal trends and standard deviation of averaged temporal trends, respectively, when Asian dust did not occur in the main area of the dust origin area. Green lines are averaged temporal trends in the non-dust origin area, defined as the area where Asian dust occurred less than one time per year.

SWS had maximum value in April and minimum value in December. SWS in the dust origin area (red and blue lines) was higher than that in the non-dust origin area (green line), and the difference between SWS in the dust origin area and SWS in the non-dust area was higher in the occurrence area of Asian dust events. These results indicate that strong SWS occurred when Asian dust events occurred. Other variables also expressed differences of value depending on the presence or absence of Asian dust events. NDVI, LST, Ta, and precipitation had maximum values in summer and minimum values in winter. Specifically, NDVI showed a clear difference depending on the presence or absence of Asian dust events. NDVI in the occurrence area of Asian dust events was relatively low compared with NDVI in the non-occurrencearea of Asian dust events, which provides evidence that Asian dust events occurred in non-vegetated area and in area with a low density of vegetation. Thus, growth time of vegetation can affect FAE. Ta is one of the variables related to vegetation growth because vegetation responds to the critical temperature of growth, especially in crop areas. Unfortunately, the difference of Ta between occurrence area and non-occurrence area of Asian dust events was not large, although such a difference existed between dust origin and non-dust origin areas. However, LST did show a difference depending on the presence or absence of Asian dust events. The amplitude of LST in non-dust origin area was lower than that in dust origin area. It was also different between occurrence area and non-occurrence area of Asian dust events. Precipitation, which affected atmospheric and surface dryness conditions, obviously showed differences of value between the dust origin and the non-dust origin areas. Precipitation was lower in the dust origin area than in the non-dust origin area, and its trend was similar to NDVI because precipitation and NDVI are strongly correlated. RH represents moisture conditions in the atmosphere. RH had minimum values in spring, but it increased after May due to precipitation. RH was higher in the non-dust origin area than in the dust origin area, and the differences of RH between the dust origin area and the non-dust origin area were similar to the time series trends of preciitation and NDVI. These trends mean that RH was affected by precipitation.

Overall, SWS and LST were larger in the dust origin (occurrence of Asian dust events) area than in the nondust origin (non-occurrence of Asian dust events) area. On the other hand, NDVI, precipitation, and RH were lower in the dust origin (occurrence of Asian dust events) area than in the non-dust origin (non-occurrence of Asian dust events) area. These results indicate that areas of dust origin and non-dust origin can be separated using hydro-meteorological and surface variables.

2) Frequency changes in Asian dust events depending on hydro-meteorological and surface conditions

Z-scores of indices were computed to find variables related to FAE on the same criterion over the dust origin area because the various hydro-meteorological and surface variables have different units. Fig. 4(a) is histogram of the SWS_Z-score. The proportion of Asian dust events that occurred when the SWS_Z-score was positive was about 63.01%. Skewness represents the asymmetry of the distribution. Skewness is a negative value if the tail at small end of the distribution is more pronounced than the tail at the large end of the distribution. For a reverse distribution, skewness is a positive value. A symmetric distribution has zero skewness. Although the skewness value (0.2294) of the SWS_Z-score was relatively low, the ratio of values higher than 1.0 was large compared o that of other variables. These results indicate that strong SWS affects FAE. Fig. 4(b) shows that 55.38% of Asian dust events occurred when the NDVI_Z-score was negative, and this ratio is similar to the ratio of the Ta_Z-score. The skewness value of the NDVI_Z-score was 0.3541. Fig. 4(c) and 4(d) show histograms of Asian dust events for the LST_Z-score and Ta_Z-score. The skewness of the LST_Z-score was 0.6067, which is a better result than for the NDVI_Z-score and the Ta_Z-score. Fifty-eight percent of Asian dust events occurred when the LST_Z-score was positive. This result is related to snow melting in the area of dust origin because snow changes the LST effects in comparison with normal years. The histogram of the Ta_Z-score was similar to that of the LST_Z-score, because Ta and LST are highly correlated (Table 1) on a monthly time scale. However, LST can explain the occurrence of dust events better than Ta considering the skewness. The skewness of the precipitation_Z-score was 1.4800, and this was the highest value among the variables (Fig. 4(e)). When precipitation was lower compared with normal years, Asian dust events occurred 67.37% of the time. Asian dust events occurred about twice as often when the precipitation was less than the normal value. This result is the highest percentage among the hydrometeorological and surface variables examied in this study. Fig. 4(f)is a histogram of the RH_Z-score. The RH_Z-score showed that Asian dust events occurred more often when the atmosphere was dry. About 25.74% of Asian dust occurred when the RH_Z-score was negative.

Fig. 4. Frequency change in Asian dut events depending on surface conditions: (a) surface wind speed, (b) Normalized Difference Vegetation Index, (c) land surface temperature, (d) air temperature, (e) precipitation, (f) relative humidity

All changes in the hydro-meteorological and surface variables compared with normal years explained FAE. In particular, precipitation affected FAE strongly, and the percentage of Asian dust events was low when precipitation was above 1.0, which indicates wet surface conditions. FAE may increase due to strong SWS because the ratio (30.30%) of -1.0~0.0 in the SWS_Z-score is similar to that (35.73%) of 0.0~1.0 but the percentage (23.98%) of 1.0~2.0 is larger than that (6.41%) of -1.0~-2.0. Surface variables had less effect on FAE than hydro-meteorological variables. However, LST should be one of the variables used as a precursor for monitoring Asian dust events, considering the skewness in the LST_Z-score. Also, it is more useful than NDVI or Ta for explaining Asian dust events.

3) Correlations among frequency of Asian dust events, hydro-meteorological variables, and surface variables

Table 3 is a correlation matrix amon FAE, hydrometeorological variables, and surface variables, including SWS, NDVI, LST, Ta, precipitation, RH, and SCF. FAE had the highest positive correlation with SWS (R: 0.88) on a monthly basis. SWS significantly affects FAE because it can induce strong wind shear. The wind shear leads upward transport of momentum, resulting in lifting of dust particles. RH (R: -0.74) had the most negative correlation with FAE because it reflected the effect of precipitation in the previous year. RH had minimum values in May (Fig. 3) but then increased due to the effect of summer monsoon precipitation in the study area. NDVI, LST, precipitation, and SCF had weak relationships with FAE (R: -0.21, 0.16, -0.19, and -0.20, respectively). NDVI and SCF were variables inhibiting Asian dust events because these variables can increase the threshold of SWS required for Asian dust events to occur (Kurosaki and Mikami, 2004), and precipitation also inhibited dust events by providing wet conditions to the atmosphere and surface. Thus, a decrease of NDVI, precipitation, and SCF can promote Asian dust events. On the other hand, LST had a positive correlation with FAE. The effect of LST may be explained by the dust origin area having drier conditions or/and the SCF decreasing rapidly compared to normal years. These relationships of variables with FAE mean that Asian dust occurs in response to SWS and that other hydro-meteorological and surface conditions also affect FAE.

Table 3. Correlaton matrix among the frequency of Asian dust events, hydro-meteorological variables, and surface variables

On a monthly basis, SWS is most closely related to FAE and has a strong negative correlation with RH (R: -0.81). SWS is also correlated with LST and Ta (R: 0.27, 0.16, respectively). In previous studies, changes in SWS were shown to be affected by changes in Ta (Kurosaki and Mikami, 2004). NDVI, which is negatively correlated with FAE, was related to LST, Ta, and precipitation (R: 0.86, 0.92, and 0.90, respectively), and this corresponds with the result of previous studies. NDVIresponded to precipitation, and vegetation activity was affected by Ta (Kawabata et al., 2001; Chuai et al., 2012). LST had a high positive correlation with Ta because the two variables interact with each other(R: 0.99); it was also related to SCF (R:-0.91) because albedo is high in regions where snow is present. Precipitation was highly correlated with surface variables such as NDVI, LST, Ta, and SCF (R:0.90, 0.79, 0.86, and -0.62, respectively). RH, which is a hydro-meteorological variable, was also related to precipitation. It was low during spring in this study area and was positively associated with precipitation (R:0.48).

4) EOF analysis for frequency of Asian dust events

Fig. 5 shows the time series of FAE and the first to third EOF modes of FAE from 2001 to 2014 over the East Asian region. The time series of FAE, shown in Fig. 5(a), shows that FAE was higer in the early 2000s than in the 2010s. An abundance of Asian dust occurred in the 2000s, especially in 2001, 2002, and 2006. On the other hand, Asian dust decreased during 2011-2014. Kurosaki and Mikami (2003) reported that Asian dust was more prevalent in 2000-2002 than in 1993-1999 because of strong surface winds around the North China Plain, Korean Peninsula, and northeastern China.

Fig. 5. Time series for the frequency of Asian dust events and the first to third EOF modes with respect to the frequency of Asian dust events from 2001 to 2014. (a) The frequency of Asian dust events, (b) the first EOF mode of the frequency of Asian dust events, (c) the second EOF mode of the frequency of Asian dust events, and (d) the third EOF mode of the frequency of Asian dust events.

EOF analysis of FAE was performed on a monthly basis to determine the characteristics of FAE. The first EOF mode for FAE (FAE_1st) explained about 37.10% (Fig. 5(b)) of the variation. The second EOF mode (FAE_2nd) explained approximately 17.23% (Fig. 5(c)) of it, and the third EOF mode (FAE_3rd) accounted for about 6.88% of the variance (Fig. 5(d)). To understand the meaning of the representative EOF modes (first to third) for FAE, EOF analysis of hydro-meteorological and surface variables such as SWS, NDVI, LST, Ta, precipitation, RH, and SCF was also performed. FAE_1st was analyzed with the principal component of the hydro-meteoroloical variables. Table 4 is a correlation matrix among FAE, hydro-meteorological variables, and surface variables in the first EOF mode. The correlation between FAE and SWS was highest for the first EOF mode (R: -0.87, Fig. 7(a)). Fig. 5(b) shows the time series of the eigenvalue on FAE, and the values are negative.

Table 4. Correlation matrix among the frequency of Asian dust events, hydro-meteorological variables, and surface variables for the first EOF mode

Fig. 7. Spatial trends for EOF modes of surface wind speed. (a) first EOF mode, (b) second EOF mode, (c) third EOF mode, (d) time series of eigenvalues for first to third EOF modes on surface wind speed from 2001 to 2014.

The eigenvector was also negative for FAE_1st (Fig. 6(a)). After SWS, RH had the second highest influence on FAE_1st (R: 0.49). These results were the same as the correlation results for the real values. The pattern of the eigenvector in FAE_1st was evident in the Gobi Desert, Loess Plateau, Inner Mongolia Plateau, western Mongolia, and Manchuria. In particular, many Asian dust events occurred in the Gobi Desert and western Mongolia. After examining the spatial patterns of FAE1st, the eigenvalue of FAE_1st was analyzed. FAE_1st was highly correlated with FAE (R: 0.94), indicating that FAE_1st was close to the seasonal cycle for FAE. In this study, multiple regression analysis for FAE1st was performed with hydro-meteorological and surface variables to find additional meanings of each first EOF mode with respect to FAE. The dependent variable was FAE, and the argument was the first to third EOF modes for the hydro-meteorological and surface variables. FAE1st was explained by SWS_1st, RH_3rd, and LST_1st. The correlation coefficient was 0.87 when only SWS_1st was considered, but R increased to 0.90 when RH and LST were also considered. This result means that FAE_1st was affected mostly by SWS_1st and that RH and LST also influenced the change of FAE. SWS_1st, which was related to the first EOF mode, was related to the time series of SWS (R: 0.98). In other words, FAE_1st means the seasonal cycle of FAE, and it was affected by SWS, RH, and LST.

Fig. 6. Spatial trends for EOF modes of the frequency of Asian dust events: (a) first EOF mode, (b) second EOF mode, (c) third EOF mode, (d) change trend of the frequency of Asian dust events from 2001 to 2014.

Fig. 6(b) shows the eigenvector for FAE_2nd, which has negative values in Inner Mongolia, Manchuria, and the Gobi Desert, but has positive values in the western Gobi Desert. The amplitude of the eigenvalue was higher from 2001 to 2003 than from 2004 to 2014 (Fig. 5(c)). To find the meaning of FAE_2nd, multiple regression analysis was performed, as with FAE_1st. SS_3rd, SWS_2nd, and NDVI_4th were selected as the variables explaining FAE2nd. The correlation coefficient between FAE_2nd and these variables was 0.58. The best variable was SWS_3rd, and the correlation coefficient was 0.42 when only SWS_3rd was considered. This result is important for understanding FAE_2nd because SWS_3rd exhibited trends in which the eigenvalues decreased regularly from 2001 to 2014 (Fig. 7(a-c)). SWS_3rd indicates the changing trend of SWS from 2001 to 2014. Therefore, the temporal trends in FAE from2001 to 2014 were analyzed to find the meaning of FAE_2nd. The temporal trend in FAE was similar to the eigenvector of FAE_2nd in the study area (Fig. 6(d)). The correlation coefficient between the two results was 0.89 in the area of dust origin. In summary, FAE_2nd was the change pattern of FAE from 2001 to 2014, and it was affected by change of SWS from 2001 to 2014 and vegetation conditions.

FAE_3rd was also analyzed using multiple regression analysis, and it was explained by various variables such as LST_1st, SCF_1st, and NDVI_3rd (Fig. 6(c)). The correlation coefficient between FAE and LST_1st, which was the variable selected first, was 0.36. When four variables were considered, the correlation coefficient increased to 0.51. LST_1st and SCF_1st, selcted as the first and the second independent variable, indicated the seasonal change. The selection of LST_1st and SCF_1st means that surface conditions were considered in FAE_3rd. Surface conditions need to be considered to understand changes of FAE because hydrometeorological and surface variables restrict the occurrence of dust. The main spatial pattern of the eigenvector in FAE_3rd showed that the northern region of the study area was influenced by SCF. LST has a negative relationship with SCF because LST increases when snow exists. Therefore, LST_1st and SCF_1st were selected for FAE_3rd. NDVI_3rd was also considered as the third variable for the multiple regression. In summary, FAE_3rd indicated surface influences, such as those by LST, SCF, and NDVI.

4. Discussion

SWS was the factor most related to Asian dust events, as in previous studies (Kurosaki and Mikami, 2003; Ge et al., 2011; Kimura, 2012b). Strong SWS occurred during the same periods as Asian dust events, and SWS was related to FAE_1st (seasonal trend) and FAE_2nd (change of FAE). FAE_1st represented the seasonal cycle, and it was associated with SWS_1st because of the strong cyclonic activity in the spring (Lee and Sohn, 2009). Synoptic-scale cyclonic activity occurs in spring, especially from April to May, and is related to dust storms (Takemi and Seino, 2005). AE_2nd represented the change pattern of FAE over the 14 years, and it was affected by SWS_3rd, SWS_2nd, and NDVI_3rd. SWS_3rd was probably the trend of SWS (Fig. 7(d)). Although the areas where SWS decreased did not exactly coincide with the region where FAE decreased, the SWS trend was associated with the trend of FAE. In the areas of dust origin, such a the Gobi Desert and Manchuria, the SWS trend was negative. Jiang et al. (2010)reported that SWS deceased in North China and along the coastline of China because of global warming. SWS_2nd was related to the Asian monsoon (Lee and Sohn, 2009). Our results reconfirmed that the changes in SWS were related to changes in the pattern of FAE.

FAE is also closely related to surface conditions, such as the change of FAE and the EOF analysis for primary modes on hydro-meteorological and surface variables. In FAE_2nd, SWS_2nd and NDVI_3rd influenced FAE, and the surface dryness conditions for Asian dust events was represented in FAE_3rd. In the multiple regression analyses, these variables explained that vegetation affected FAE (Mao et al., 2013; Kang et al., 2014). Mao et al. (2013) showed that vegetation changes in spring were linked to Asian dust events was represented in FAE3rd. In the multiple regression analyses, these variables explained that vegetation affected FAE (Mao et al., 2013; Kang et al., 2014). Mao et al. (2013 showed that vegetation changes in spring were linked to Asian dust. The low NDVI compared to the value in normal years was associated with weakness of soil cohesion due to a paucity of vegetative roots. The relationship between NDVI and Asian dust events has been reported to be negatively correlated (Zou and Zhai, 2004).In addition, leaf conditions acquired from NDVI led to improvement of dust simulation (Kang et l., 2014). When the NDVIZ-score was negative, the percentage of Asian dust events indicated by NDVI was similar to that indicated by SM. Kim et al. (2013) showed that Asian dust events occurred about 10% more often when the SM anomaly was negative using only AMSR-E data. Moreover, FAE increased when the surface conditions become drier than the previous SM values (Kim et al., 2013). Similarly, 55.75% of Asian dust events occurred when the SMZ-score that was computed using AMSR-E SM data in this study had negative values (figure not shown). NDVI is highly correlated with SM, and the percentage of Asian dust events according to the increase or decrease of NDVI was similar to that of SM, which is influenced by rainfall (He et al., 2012). Sugimoto et al. (2010) confirmed that vegetation reduced dust emission notably. Therefore, NDVI is a useful variable for analyzing FAE using auxiliary data.

In addition to NDVI, LST and snow cover are important surface variables for analyzing FAE. Approximately 90% of Asian dust events occurred under the conditions of LST < 74.8 K and SCF <4.4%.These results indicate that most Asian dust events occurred when LST was lower than about 0°C and snow did not exist. Kurosaki and Mikami (2004) showed that strong wind velocity was needed to cause Asian dust events when SCF was large and that SCF had a positive relationship with wind velocity related to Asian dust events. Lee and Sohn (2009) reported that snow coverage was a reliable variable for Asian dust events. Also, improved snow cover and SM data have contributed to the analysis of Asian dust levels (Sekiyama et al., 2011). In summary, LST and snow cover data are useful as precursors for monitoring Asian dust outbreaks.

Although precipitation was the best variable for explaining FAE using the Z-score, the correlation between FAE and precipitation and the priority to select precipitation as a variable of multiple regression in the EOF analysis were low because this study did not consider the time lag and anomaly of precipitation. However, NDVI, which is influenced by precipitation, was selected as the priority variable related to FAE_3rd, and it had high correlation with precipitation (R: 0.90). RH was related to FAE seasonally, but the relationship between the change trend of FAE and RH was weak. Measuring RH spatially using satellite remote sensing is technically difficult, and the spatial resolution of satellite-based surface data such as NDVI, LST, and SCF is better than that of precipitation data. Therefore, we recommed using surface data to monitor FAE in areas of dust origin.

5. Conclusions

In this research, ground and satellite-based hydrometeorological and surface variables were used to analyze FAE. The Z-scores of SWS, NDVI, LST, Ta, precipitation, and RH, excluding SCF, were computed and compared for normal years. The change in FAE was determined to be strongly related to changes in hydro-meteorological and surface variables. Not only hydro-meteorological variables, Such as SWS_Z-score and precipitation_Z-score, but also surface variables, such as LST_Z-score and NDVIZ-score, explained changes in FAE. Although surface variables are not a main cause, such as wind, generating Asian dust events, these were useful auxiliary data for explaining FAE.

The first to third EOF modes of FAE represented the seasonal characteristics, the change trends induced by SWS, and the surface influence related to dryness condition for FAE, respectively. FAE_1st corresponded with the area for FAE. FAE_2nd was related to the change trend of FAE, which was related to the trend of SWS. The surface variables were related to FAE_3rd. These results indicated that changes in FAE were affected by surface conditions. Therefore, monitoring surface conditions is important for understanding change in FAE. Especially, LST expressed change of FAE in the Z-score results and EOF analysis. The results of this investigation are importnt for improvement of dust models and developments in dust detection algorithms based on satellite observations.