Investigation of Aerosol Optical Depth ( AOD ) and Ångström Exponent over the Desert Region of Northwestern China Based on Measurements from the China Aerosol Remote Sensing Network ( CARSNET )

Aerosols at ten sites in northwestern China are classified in this study: (1) by using the aerosol optical depth (AOD), the Ångström exponent (α) and the Ångström exponent difference (δα); and (2) by using the total means of AOD440nm and α. The seasonal variations of the AOD and α show that the maximum AODs occur in spring, except at Urumqi and Lanzhou. The seasonal mean α values are lower than 0.80 in all four seasons at Tazhong, Hotan, Hami, Ejina, Dunhuang, Minqin, and Jiuquan, but higher than 0.80 in winter at Urumqi, Lanzhou and Yinchuan. The first classification method shows that coarse mode particles are found at all ten sites, but that fine mode growth only happens at Urumqi, Lanzhou, and Yinchuan. The relationship between AOD440nm and α show that α smaller than 0.80 decrease with increasing AOD440nm at all ten sites. Aerosols are classified into four types (Type I–IV) according to the total mean τ440 ( 440  ) and total mean Ångström exponent ( ) of each site. Aerosols with a τ440 smaller than 440  , but greater than or equal to  (τ440 < 440  ; α ≥  ) are classified as Type I; aerosols with τ440 ≥ 440  and α ≥  are Type II; those with τ440 < 440  and α <  are Type III; and those with τ440 ≥ 440  and α <  are Type IV. The second aerosol classification method shows that Type I and Type III aerosols are the most common at all ten sites. Type II aerosols are the least at Tazhong and Hotan, but are the most common at Urumqi, Lanzhou, and Yinchuan. On the contrary, Type IV aerosols are the most common at Tazhong and Hotan, but are the least common at Urumqi, Lanzhou and Yinchuan.


INTRODUCTION
Atmospheric aerosols play an important role in global and regional climate change (Charlson et al., 1992).Through scatting or absorbing solar radiation, aerosol particles directly affect the earth-atmosphere radiative balance (Ackerman and Toon, 1981).They also serve as cloud condensation nuclei, thereby affecting the size distribution of cloud droplets in an indirect way (Twomey et al., 1984).Despite many aerosol studies, aerosol concentrations and optical properties remain one of the largest sources of uncertainty in current assessments and predictions of global climate change (Hansen et al., 2000).The aerosol optical depth (AOD) and Ångström exponent (α) are two basic optical parameters necessary for climate change research (Breon et al., 2002).There are numerous studies on aerosol optical properties of several regions, including North China, the northeast, the northwest, the Yangtze River Region and the South of China (Xia et al., 2005;Garland et al., 2008;Che et al., 2009a;Pan et al., 2010;Wang et al., 2010a;Che et al., 2011;Liu et al., 2011;Zhao et al., 2013;Che et al., 2014;Tao et al., 2014).
Dust aerosols are the mainly important aerosol particles in East Asia.Dust aerosols from arid and semi-arid regions are transported thousands of kilometers away from their original source regions (Gong et al., 2003).The global dust emission is estimated about 1500 Tg yr -1 (Tegen and Fung, 1995).Nearly 800 Tg yr -1 of the dust emission is emitted into the atmosphere each year, half of which is deposited close to the source and adjacent regions, while the rest is transported to the remote Pacific Ocean (Zhang, 2001).Thus, it is necessary to study the optical properties, and the temporal and spatial variability of dust aerosols in order to accurately estimate the influence of dust aerosols on global and regional climate change (Wang et al., 2006(Wang et al., , 2010b)).
The arid and semi-arid regions in northwestern China are major sources of dust aerosols in East Asia (Zhang et al., 2003).Many researchers have studied the characteristics of dust aerosols in East Asia in recent years.Alfaro et al. (2003) observed aerosol optical characteristics in spring 2002 at the ACE-Asia supersite (Aerosol Characterization Experiments, Zhenbeitai, China) and found that dust optical characteristics measured during dust storms were representative of pure dust emitted from the northwestern high desert sources.Xia et al. (2005) showed that AOD over North China in spring was dominated by contributions from dust over western China.Dust aerosols not only enhance the aerosol loading but also reduce light absorption.Cheng et al. (2006) indicated that high AOD corresponded to dust event occurrence, while the Ångström exponent decreased with increasing AOD to zero or negative values, when very dusty events occurred in the Hunshan Dake desert.Huang et al. (2009) determined that both shortwave and longwave radiative forcing of dust aerosols played an important role in the radiative energy budget, at both the top of the atmosphere and the surface.Xia and Zong (2009) demonstrated that Earth's system was cooled in the shortwave but warmed in the longwave by Taklamakan dust aerosols.These studies are essential to improve understanding of the essential properties and variations of dust aerosols in East Asia.
The objective of this research is to use the ground-based sunphotometer measurements to investigate the detailed aerosol optical properties through aerosol classification by different methods at ten sites in northwestern China.This research will help to validate the satellite observations and improve estimations of the effect of East Asian dust aerosols on global and regional climate change.This paper includes site distribution, instruments and data first.Then, seasonal variation of AOD 440 and the Ångström exponent are analyzed.Aerosol classification by the AOD 670 , Ångström exponent and Ångström exponent difference and aerosol classification by total mean AOD 440 and total mean Ångström exponent are discussed.Finally, the summary and the discussion are conducted.

Site Distribution
Sun-photometers (CE318, Cimel Electronique, France) are installed at ten observation sites in northwestern China: Tazhong, Hotan, Hami, Urumqi, Ejina, Dunhuang, Minqin, Jiuquan, Lanzhou, and Yinchuan, as shown in Table 1 and Fig. 1.Tazhong, Hotan, Hami, Urumqi are in the Xinjiang Province, Ejina is in the Inner Mongolia Province, Dunhuang, Minqin, Jiuquan, Lanzhou are in the Gansu Province and Yinchuan is in the Ningxia Province.Among the ten sites, Tazhong and Hotan are located in the Taklamakan Desert, one of the largest sand deserts in the world.The desert covers an area of 270,000 km 2 , and includes of the Tarim Basin, which is 1000 km long and 400 km wide (Huang et al., 2009).It is regarded as one of the largest sources of Asian Aeolian dust aerosol (Mikami et al., 2006).Hami, Ejina, Dunhuang, Minqin, and Jiuquan are all located in arid and semi-arid regions over northwestern China, where aerosols are dominated by dust aerosol particles.However, Urumqi, Lanzhou and Yinchuan are located in the center of cities, where aerosols mainly come from fine particle pollution.The observation times at ten sites are from 2002 to 2012 and the details of the valid data are listed in Table 1.

Instruments and Data
The  and for the validation of satellite aerosol retrievals (Che et al., 2009b).The instrument used by CARSNET is automatic Cimel sun and sky scanning radiometer (Cimel Electronique Cimel-318), the same instrument as the Aerosol Robotic Network (AERONET).The CE-318 sun-photometer has a 1.2° full field-of-view and eight channels: four observation channels at 440 nm, 670 nm, 870 nm and 1020 nm; three 870 nm polarization channels; and a 940 nm water vapor channel (Holben et al., 1998).Measurements at 440 nm, 670 nm, 870 nm and 1020 nm are used to retrieve the AOD, and measurements at 940 nm are used to obtain the total precipitable water content in cm (Holben et al., 1998).The total uncertainty in optical depth is about 0.01-0.02(Eck et al., 1999).The inter-comparison calibration protocol for the CARSNET field instruments were as follows: (a) only raw data collected from 02:00 to 6:00 AM (GMT+8:00) on clean and clear days were used, (b) the AODs at 500 nm on calibration days had to be less than 0.20 and without major fluctuations, (c) the time intervals between the measurements made with two masters and the instruments to be calibrated had to be less than 10 s.The AODs obtained from uncalibrated instruments differed by 4.5% to 15.3% compared with those measured by reference instruments.After the calibration with the master instruments, however, the daily average AODs differed by < 1.5% relative to the master measurements (Che et al., 2009b).According to Holben et al. (1998), yearly calibrations of the field instruments ensured the accuracy of the CARSNET measurements, and therefore, the AODs from the 10 stations in this study were of high quality and reliability.
The valid data at the ten sites are calculated by the ASTPwin software (Cimel Ltd.Co.) for the Level 1.0 AOD (raw result without cloud screening), the Level 1.5 AOD (cloud-screened AOD based on the work of Smirnov et al., (2000)) and the Ångström exponent between 440 and 870 nm.The aerosol optical depth at wavelength λ (AOD λ ≡ τ λ ) represents the extinction of radiation of wavelength λ.The Ångström exponent α represents the slope of the wavelength dependence of the AOD in logarithmic coordinates (Angstrom, 1929): α(λ1, λ2) = -ln(τ λ2 /τ λ1 )/ln(λ2/λ1).Four seasons are defined as: spring (March to May), summer (June to August), autumn (September to November) and winter (December to February) to investigate the seasonal variation of aerosol optical properties.

Seasonal Variation of AOD at 440 nm and the Angström Exponent
Fig. 2 illustrates the seasonal variation of the AOD and Ångström exponent at ten sites in northwestern China.The mean AOD values at Tazhong and Hotan in spring and summer (more than 0.50) are higher than those in autumn and winter.Tazhong and Hotan are both located in the Taklamakan Desert, and dust events are very frequent during spring and early summer, causing large aerosol loading in the atmosphere over the region (Xue et al., 2009).In contrast, there are few dust events during autumn and winter.Fine particles (such as black carbon) from anthropogenic Fig. 2. Seasonal box plots of the AOD and Ångstrom exponent at the ten sites (The extreme "" means the maximum and minimum value; the "×" means 99% and 1% percentile value; the"□" means the mean value).
activities add to the aerosol burden (Li et al., 2010), but the anthropogenic aerosols are mainly transported from outside sources.Compared with mineral dust in spring and summer, the anthropogenic loading during autumn and winter is much less.The seasonal variation of AOD at Hami, Ejina, Dunhuang, Minqin, and Jiuquan is similar.The mean AOD values are lower than 0.50 and the maximum values occur in spring at these five rural sites, reflecting the effect of floating or blowing dust events (Wang et al., 2005).However, the maximum mean AODs occur in winter at Urumqi and Lanzhou; in spring at Yinchuan.The higher AODs occur in spring at the three urban sites, corresponding to frequent local or regional dust events in northwestern China.The maximum mean AODs at Urumqi and Lanzhou may be caused by the worse atmospheric diffusion conditions.Cao et al. (2013) pointed out that the pollution emissions are very high in Urumqi and Lanzhou during winter season because of the coal combustion for warm-keeping.The seasonal variation of AOD at Yinchuan is similar to those five rural sites.Because Yinchuan is located between the Gobi desert in the north and the Loess Plateau in the southeast, dust particles are possibly transported when either northerlies or south-easterlies are prevalent (Kim et al., 2004).Eck et al. (2005) showed that an Ångström exponent less than 0.80 indicates coarse mode aerosol dominance.The mean Ångström exponent values at Tazhong and Hotan are less than 0.60 throughout the year (Fig. 2), suggesting that coarse mode aerosol strongly dominates the AOD at the two desert sites.The seasonal variation of the Ångström exponent at Tazhong and Hotan shows characteristically low values in spring and summer and higher values in autumn and winter, indicating that the aerosol particles are larger in spring and summer than in autumn and winter.The mean Ångström exponent values at the rural sites of Hami, Ejina, Dunhuang, Minqin and Jiuquan are lower than 0.80 in four seasons, indicating that coarse mode aerosol dominates the AOD.The seasonal variation of the Ångström exponent at these seven sites is similar to the results of Che et al. (2013) at Tazhong.However, the mean Ångström exponent values are larger than 0.80 year-round at Urumqi, and larger than 0.80 in summer, autumn and winter at Lanzhou and Yinchuan (smaller than 0.80 in spring), suggesting that fine particles mostly contribute to the composition of aerosol at the three urban sites.Kaufman (1993) showed that negative values of the Ångström exponent difference, δα, equal to α(440, 613) -α(440, 1003), indicate the dominance of fine mode aerosols, while positive differences indicate the effect of two separate particle modes.Gobbi et al. (2007) built on this concept and proposed a new aerosol classification method, to track mixtures of pollution aerosol with dust, to distinguish aerosol growth from cloud contamination, and to observe aerosol humidification.The method defines the Ångström exponent difference δα = α(440, 675) -α(675, 870) as a measure of the Ångström exponent curvature with respect to wavelength, λ: dα/dλ.The δα vs. α (440, 870) space is plotted as the framework for analyzing aerosol properties.

Aerosol Classification by the AOD at 670 nm, Ångström Exponent and Ångström Exponent Difference
In these coordinates, the aerosol particles are further classified by representing their AODs in different colors.A cloud contamination or an increase in coarse aerosols of the AOD larger than 90% will be located at α~δα~0.However, hydration moves in the opposite direction with respect to cloud contamination, leading to a growth in both the size and the fine mode extinction fractions (η).Therefore, it allows for easy identification between fine mode growth and coarse mode particle contamination.Gobbi et al. (2007) performed Mie calculations of the aerosol spectral extinction for fine mode particles (R f ) of 0.05, 0.1, 0.15, 0.2, 0.3 and 0.5 µm, for coarse mode particles (R c ) of 0.75, 1, 2, and 4 µm, and combined them to provide the η values of 1, 10, 30, 50, 70, 90 and 99%.Both the fine and coarse mode particles are assumed to have a lognormal size distribution, respectively.According to Gobbi et al. (2007), the maximum R f and η indeterminations are of ± 25% and ± 10% for refractive index varying between m = 1.33-0.00i(water droplets) and m = 1.53-0.003i(mineral dust) for a given point (α, δα).And the graphical approach for refractive index of m = 1.40-0.001i is robust enough to provide an operational classification of most common aerosol typologies (e.g., Dubovik et al., 2002) on the basis of standard photometric observations.The above aerosol classification method is applied at the ten sites in northwestern China in this study to analyze the aerosol optical properties using the instantaneous observations.Fig. 3 shows simulations of the classification of the aerosol properties at the ten sites as a function of the Ångström exponent α(440, 870) and the difference δα = α(440, 675) -α(675, 870), for bimodal lognormal size distributions with a refractive index of 1.40-0.001i.
The further classification of AODs in different colors is showed as follows: the black hollow circle, 0.15 < AOD ≤ 0.30; the red hollow circle, 0.30 < AOD ≤ 0.40; the green hollow circle, 0.40 < AOD ≤ 0.70; the blue solid dot, 0.70 < AOD ≤ 1.00, the cyan solid dot, 1.00 < AOD ≤ 1.50, the magenta solid dot, 1.50 < AOD ≤ 2.00 and the yellow solid dot, 2.00 < AOD ≤ 3.00.The black solid lines are each for a fixed size of the fine mode, R f , and the dashed blue lines indicate a fixed fraction contribution (η) of the fine mode to the total AOD at 670 nm.In many cases the higher AODs are associated with coarse mode particles (α~δα~0, η < 30%) due to dust events at Tazhong and Hotan: the two sites are located in the Taklamakan Desert, where the aerosols are dominated by dust coarse mode particles.The higher AODs are mainly clustered at the slope of α and δα about 0, indicating that the effects of coarse mode particles at Hami, Ejina, Dunhuang, Minqin, and Jiuquan are prominent.It suggests that the five rural sites are impacted by dust aerosols from northwestern China.Ejina has "typical pollution" with 1.00 > AOD > 0.70 (blue dots) that corresponds to a fine fraction of 99% > η > 70% moving along the black line of 0.15 µm, different from the other four rural sites.It may be associated with cloud contamination (Gobbi et al., 2007).Fine mode growth is evident at Urumqi and Lanzhou, but less so at Yinchuan.The observed data at Urumqi and Lanzhou (AOD > 0.70) are mainly clustered in the fine mode growth wing (α < 1.50, δα < 0).The extension of the two urban sites measures to higher AODs happen perpendicularly to the black line, into larger size of the fine mode and fine fraction mainly between 60% and 90%.At the same time, coarse particles such as mineral dust (α~δα~0, η < 30%), add their signal to the pollution signature in Urumqi and Lanzhou.Hence, it can be inferred that the high extinction at Urumqi and Lanzhou could be linked to a hygroscopic and/or coagulation growth from aging of the fine mode aerosols, and dust storm events.The results are similar to those observed in Beijing and Kanpur, India (Gobbi et al., 2007).However, the effect of the coarse mode particles and the fine mode growth are less evident at Yinchuan, whose mean size is smaller than the other two urban sites.It suggests that anthropogenic activities are frequent at Urumqi and Lanzhou, and infrequent at Yinchuan.

Aerosol Classification by Total Mean AOD at 440 nm and Total Mean Ångström Exponent
Fig. 4 shows the relationship between AOD at 440 nm (AOD 440 ) and Ångström exponent at ten sites.The vertical and horizontal blue dashed lines represent the total mean AOD and Ångström exponent, respectively.The plots show the mean AOD(τ 440 ) (the red solid circles) with the standard deviations (red error bars) at different Ångström exponents of -0.5-0.0,0.0-0.5, 0.5-1.0,1.0-1.5 and 1.5-2.0.The mean AODs and standard deviations are also listed in Table 2.The relationship between the AOD and Ångström exponent shows that Ångström exponents of smaller than 0.80 decrease with increasing AOD 440 at all ten sites, indicating that coarse particles, e.g., mineral dust, are responsible for large AOD.Aerosols are classified into four types (Type I-IV) according to the total mean τ 440 ( 440  ) and total mean Ångström exponent ( ) of each site, as in Table 3. Type I aerosols are defined as those with τ 440 smaller than the total mean τ 440 , and α larger than the total mean α (i.e., τ 440 < 440  ; α ≥  ).These conditions correspond to relatively small particles with small optical depths, and capture a large range of the basic background aerosols.Type II aerosols are characterized by a τ 440 larger than the total mean τ 440 , and an α larger than the total mean α (τ 440 ≥ 440  ; α ≥  ): i.e., relatively small particles with a large optical depth, representing the emission of anthropogenic aerosol and the corresponding gas-to-particles conversion processes.Type III aerosols have a τ 440 smaller than the total mean τ 440 , and an α smaller than the total mean α, (τ 440 < 440  ; α <  ): i.e. large particles with a small optical depth.This is related to local characteristics.Type IV aerosols have a τ 440 larger than the total mean of τ 440 , and an α smaller than the total mean of α, (τ 440 ≥ 440  ; α <  ): corresponding to large particles with a large optical depth, they are often associated with local dust or dust events.Note that this type classification does not necessarily correspond to species classification of individual aerosols: these types are instead useful for characterizing atmospheric conditions (Aoki and Fujiyoshi, 2003).From the percentage ratios of the four types in Table 3, we see that Type I and Type III aerosols are the most common at all ten sites.They represent the condition of background aerosols and the effect of the local characteristics, respectively.Type IV aerosols are most common at Tazhong (comprising 35.89% of all aerosol found there) and Hotan (32.03%), but Type II aerosols are the least common (Tazhong: 0.87%, Hotan: 2.08%) at all ten sites.This suggests that there are more dust events and fewer anthropogenic activities in and around Tazhong and Hotan than any of the other eight sites.This pattern is similar to those observed in Sapporo, Japan (Aoki and Fujiyoshi, 2003).Type II and Type IV aerosols are medium common at Hami (9.06% and 23.57%), Ejina (10.03% and 20.96%), Dunhuang (5.47% and 27.74%), Minqin (11.44% and 24.14%), and Jiuquan (8.74% and 23.43%).This indicates that the aerosols at the five rural sites are affected by both dust events and anthropogenic activities.This result is similar to those found in Tsukuba (Aoki and Fujiyoshi, 2003).Type IV aerosols are less common and Type II aerosols are more common at Urumqi, Lanzhou, and Yinchuan.This suggests that the effect of the dust events is smaller and anthropogenic activities are more frequent.The result is similar to those calculated in Tokyo, Japan (Aoki and Fujiyoshi, 2003).
Fig. 5 shows the frequency distributions of the four types of aerosols at ten sites.Type I and Type IV aerosols have similar seasonal variations at Tazhong and Hotan.Type I aerosols are the most common in later autumn and winter, but Type IV aerosols are common in spring and summer.It suggests that the atmosphere is clean in later autumn and winter but dust particles are dominant in spring and summer at Tazhong and Hotan due to the contribution of the frequent dust events.Type II aerosols appear infrequently at both Tazhong and Hotan with frequency occurrence less than 15%, which suggests the anthropogenic activities contribute few effect at the two sites.Type I aerosols are common in later autumn and winter at Ejina, Dunhuang, Minqin and Jiuquan, but in summer at Hami.The difference between Hami and the other four sites may be caused by relative large precipitation in June and July (Tu and Kong, 2014).Type IV aerosols are common in spring at the five sites, and the frequency of appearance is about 25-60%.This indicates the influence of dust events is dominant at these sites as well as Tazhong and Hotan.Type II aerosols also appear less than 22% at the five sites except for Hami ~40% in January.This indicates that anthropogenic aerosol emissions and subsequent gas-toparticle processes likely lead to the high mean AOD values in winter at Hami (Tu and Mu, 2010).Type I aerosols are common in summer at three urban sites of Urumqi, Lanzhou, and Yinchuan with the occurrence about 20-55%, which suggests good air qualities at three sites in summer season.Type II aerosols are obviously common (> 48%) during November to January at Urumqi and Lanzhou, which suggests air pollutions from anthropogenic activities (Li et al., 2005;Cao et al., 2013).Type IV aerosols are 20%-50% in spring at three urban sites which are mainly caused by the dust events (Tao et al., 2009).The seasonal variations of Type III aerosols are different, which suggests local characteristics are different at each of ten sites.

SUMMARY AND DISCUSSION
The maximum mean AOD values occur in spring at Tazhong, Hotan, Hami, Ejina, Dunhuang, Minqin, Jiuquan, Fig. 4. Relationship between the AOD at 440 nm and the Ångström exponent at ten sites.Yu et al., Aerosol and Air Quality Research, 15: 2024-2036, 20152032 and Yinchuan, reflecting the effect of dust events, floating dust and blowing dust events occurring in arid and semi-arid regions over northwestern China.However, the mean AODs are higher in spring and winter at Urumqi and in spring, autumn and winter at Lanzhou, which reflects the hybrid effect of the dust events and anthropogenic activities.
Table 2.The mean and standard deviation (St.dev.) of AOD at different Ångström exponents of -0.5-0.0,0.0-0.5, 0.5-1.0,1.0-1.5 and 1.5-2.0 at the ten study sites.The mean Ångström exponent values are lower than 0.80 at Tazhong, Hotan, Hami, Ejina, Dunhuang, Minqin and Jiuquan, which suggests coarse mode aerosols dominate the AOD at these seven sites.However, the mean Ångström exponent values are larger than 0.80 year-round at Urumqi, larger than 0.80 in summer, autumn and winter at Lanzhou and Yinchuan, which indicates fine particles are the primary contributors to aerosols at the three urban sites.
Aerosol classification by the AOD at 670 nm, Ångström exponent and Ångström exponent difference show that the higher AODs are mainly clustered at the range of α and δα about 0 at all ten sites, which indicates the presence of coarse mode particles.Only at Urumqi, Lanzhou, and Yinchuan, is fine particles growth found.
Aerosol classification by total mean AOD at 440 nm and total mean Ångström exponent shows Type I aerosols are dominant in later autumn and winter seasons at rural sites of Tazhong, Hotan, Ejina, Dunhuang, Minqin and Jiuquan, but in summer at urban sites of Urumqi, Lanzhou, and Yinchuan.Type II aerosols are mostly less than 20% at rural sites but larger than 48% at Urumqi and Lanzou in later autumn and winter.Type III aerosols show different seasonal variation characteristics due to the different local characteristics.Type IV aerosols are dominant at all ten sites in spring season because of the effect of dust events.
Combining two kinds of aerosol classification methods, we can easily understand the aerosol characteristics at the ten sites of Northwestern China.From the second aerosol classification method, we can know the dominant aerosol types at each site.Type IV aerosols are common at Tazhong, Hotan, Hami, Ejina, Dunhuang, Minqin and Jiuquan, which is associated with many high AOD cases caused by coarse mode particles (α~δα~0, η < 30%).However, Type II aerosols are not common at above sites, which is associated with few fine mode growth because of less frequent anthropogenic activities.At three urban sites of Urumqi, Lanzhou and Yinchuan, Type II aerosols are common and the fine particles growth phenomena are obvious due to the frequent anthropogenic activities.Moreover, Type IV aerosols are also common especially in spring season caused by coarse mode particles.

Fig. 5 .
Fig. 5. Monthly frequency of appearance τ 440 and α for each type (Type I-Type IV) at ten sites.

Table 1 .
Details for each site.

Table 3 .
Percentage of each aerosol type at the ten study sites.