Principal component analysis (PCA) is an effective method that can provide a quantitative simplified description of the principal components by processing the raw data matrix, and has been widely used in the source analysis and classification of air pollutants (Meng et al., 2016). Under the assumptions of the receptor model, it is assumed that at a receptor site, the measured species concentration is a linear combination of the total mass consisting of the sum of contributions from different sources/factors, the formula is as follows. x i j = ∑ k = 1 p g i k f k j + e i j where x _ij refers to the concentration of the j_th species in the i_th sample; g _ik defines the contribution of k_th factor/source to the i_th sample; f _kj is the relative amount of the j_th species in the k_th source. Ideally, e _ij should only include experimental error, but in practice it may also include small unknown and unexplained parts. X = G F + E where X is a data matrix of measured concentrations of different chemical constituents in the aerosol. G is the loading matrix, which serves as the source contribution dataset for each sample. F is a score matrix representing the relative weight of the analyzed species in each source. E is the noise matrix of the residuals.

In order to analyze the origin and transport routes of TSPs, 72-h air mass back trajectories were calculated at a height of 500 m using Hybrid Lagrangian Single-Particle Integrated Trajectory (HYSPLIT) model in the GIS-based software-TrajStat (Draxler and Rolph, 2013). The model was run four times per day at starting times of 00:00, 06:00, 12:00, and 18:00 UTC, with the (National Center for Environmental Prediction) NCEP/NCAR (National Center for Atmospheric Research) meteorological reanalysis dataset generated from the global data assimilation system (GDAS, available at ftp://arlftp.arlhq.noaa.gov/pub/archives/gdas1/).

The potential source contribution function (PSCF) analysis was applied to assess the possible source areas of total suspended particulate matter. According to the results of the HYSPLIT model, the GIS-based TrajStat software (Wang et al., 2009) was used for PSCF analysis. The domain of PSCF was 22–65°N, 35–95°E with a horizontal resolution of 0.5° × 0.5°. The specific principles of these models can be detected in prior studies (Feng et al., 2015; Yao et al., 2016; Zhang J. et al., 2018).

The annual average concentration of TSPs in this study area was 181.4 ± 169.7 μg m⁻³, which was obviously lower than that in urban regions, such as in Urumqi (338.0 μg m⁻³) in 2010–2012 (Zhang X. et al., 2014) and Xi’an (378.7 μg m⁻³) in 2012 (Zhang F. et al., 2015), and higher than the concentrations in remote regions, such as in Nam Co. station (31.9 μg m⁻³) in 2011–2013 (Liu et al., 2017), Lulang site (23.5 μg m⁻³) in 2010 (Zhao et al., 2013) and Qinghai lake (41.4 μg m⁻³) in 2010 (Zhang N. et al., 2014). The results indicated that the air in the source area of Urumqi River had been polluted by human to a certain extent.

The average TSPs concentrations were on the order of 356.2 μg m⁻³ (range: 60.1–795.9 μg m⁻³) in autumn, 239.2 μg m⁻³ (range: 56.8–490.0 μg m⁻³) in spring, 107.0 μg m⁻³ (range: 51.6–280.2 μg m⁻³) in summer, and 64.4 μg m⁻³ in winter (range: 18.8–90.5 μg m⁻³). The average mass concentration in autumn was 5.5 times that in winter, and these typical seasonal patterns could be attributed to a variety of pollution contributions from local and regional areas in different seasons. The high concentration of TSPs in autumn was due to low wind speed and low rainfall, which was not conducive to the removal of pollutants. Second, a road was being built near the TGS, which might cause more road dust and traffic emissions. The average concentration of TSPs was also high in spring, mainly due to the high wind speed, which was easy to produce dust storms. The average concentration of TSPs was lower in summer, one was because there was less human pollution, and the other was because of more precipitation. In summer, rainfall accounted for 60% of the annual precipitation, which was conducive to remove and reduce the mass concentrations of TSPs. At the same time, due to the Covid-19, Urumqi entered into a “wartime-state” on 16 July 2020, which was the strictest mobility restriction that was, absolute lockdown. Human activities had been greatly reduced, and the mass concentration of summer aerosols had been greatly reduced. The reason for the lowest concentration of TSPs in winter was that the Tianshan Mountains were covered by heavy snow and inaccessible. At the same time, it might also be related to the temperature inversion layer. Early researchers have found that there is a strong inversion layer at 1,000–2,400 m above Urumqi in winter, which can prevent pollutants from spreading to high altitudes and trap them in the boundary layer (Zhang et al., 2019). Therefore, although Urumqi had serious air pollution in winter, the atmospheric environment in the source of Urumqi River was still relatively clean. The main reason was that the existence of the temperature inversion layer prevented pollutants from spreading to high altitude areas.

The daily concentrations of WSIIs (Cl⁻, NO₃ ⁻, SO₄ ^2–, Na⁺, NH₄ ⁺, K⁺, Mg²⁺, and Ca²⁺) for all samples are listed in Figure 2. Generally, the total WSII average concentration was 9.15 ± 5.53 μg m⁻³, accounting for ∼5.04% of the TSP mass. The order of ionic species abundance was Ca²⁺ >SO₄ ^2– >NO₃ ⁻>Na⁺ >Cl⁻ >NH₄ ⁺ >K⁺ >Mg²⁺. Ca²⁺, SO₄ ^2– and NO₃ ⁻ were the three most prevalent WSIIs, accounting for 81.3% of the total WSIIs. The mass concentrations of the other five ions (Na⁺, NH₄ ⁺, K⁺, Mg²⁺ and Cl⁻) were lower (less than 1.0 μg m ⁻³), accounting for 18.7% of the total ion concentration. Ca²⁺ and TSP had a strong correlation (R = 0.87, n = 75, p < 0.05), indicating that the changes of this ion were closely related to the formation, transmission and removal of TSP (Zhang Q. et al., 2015).

The average seasonal concentrations of WSIIs are presented in Table 1. Similar to TSPs, the concentrations of sum of total ions had noticeable seasonal changes, higher in autumn and spring and low in summer and winter. The seasonal variation of the mass concentration of each ion was also very obvious. The mass concentrations of all ions were highest in autumn (except Na⁺ and Mg²⁺), and lowest in winter (except Ca²⁺ and NH₄ ⁺).

As we all know, Ca²⁺ was an excellent indicator of soil dust (Li et al., 2011; Zhao et al., 2011). When the spring dust storm occurred, the concentration of Ca²⁺ was the highest. But in this study, the Ca2+ concentration in autumn was 1.25 times that in spring. In the autumn, the infrastructure of the Tianshan Glaciological Station was under construction, generating a lot of dust from buildings and roads. Therefore, the calcium ion concentration was higher in autumn. Furthermore, the dust emission was constant in winter and summer, therefore, the seasonal variation of calcium concentration was mainly related to weather conditions, such as rainfall and wind. There was more precipitation in summer and winter, especially in summer, where the precipitation accounted for 60% of the annual precipitation, so the concentration of Ca2+ was low.

SO42^– and NO3^– were also the most significant water-soluble ions in total suspended particles in the source area of Urumqi River. In our study, the annual average concentration of sulfate ions was the second highest, accounting for 28.3% of the total ions. The concentration was highest in autumn, followed by spring and summer, and lowest in winter, because the relatively high SO2 concentration was conducive to the formation of SO42^–. There was a significant correlation between SO₄ ^2– and Ca²⁺ (r = 0.72), indicating that had similar origins, derived from desert and road dust particles and may exist in the form of CaSO₄. Furthermore, the lower ratio (SO₄ ^2–/Ca²⁺ = 0.22) in autumn also illustrated that the sources of SO₄ ^2– and Ca²⁺ were similar, mainly from natural sources, such as sand and dust, etc. (Ming et al., 2007). NO₃ ⁻ was slightly lower than that of SO₄ ^2–, accounting for 10.8% of the total concentrations of the ions. The seasonal variation characteristics of its concentrations were the same as that of SO₄ ^2–, with the highest in autumn and lowest in winter. It is generally believed that NO₃ ⁻ is the reaction product of the emitted NO_x and the NO _x emissions mainly originate from the burning of fossil and biomass fuels. In an alkaline atmosphere, the gaseous HNO₃ formed by the oxidation of NO_x can be adsorbed on the surface of mineral particles and react with it to form nitrate (Clase and Gysels, 1998). Its concentration was the highest in autumn, mainly because there were more vehicles discharged during this period, and more NO_x was produced, which was beneficial to the formation of NO₃ ⁻.

Figure 3 shows the equivalent concentration ratios of the main ions. The ratio of the sum of cation and anion equivalent concentrations (CE/AE) can better reflect the acidity of air particles. In our samples, the ratios varied from 0.97 to 32.84, with an average value of 4.58, indicating cations dominate the ionic composition during the sampling period and major anions are not detected. Early studies had shown that the lack of anions was mainly due to undetected carbonate when Ca²⁺ was significantly related to ΔC (CE-AE) (Sun et al., 1998; Zhang XY. et al., 2012). In this research, the correlation coefficient between Ca²⁺ and ΔC was 0.9435 (p < 0.01), indicating that HCO₃ ⁻/CO₃ ^2– was the undetected primary anion. Earlier studies also reported similar results (Li et al., 2011; Zhao et al., 2011; Zhang X. et al., 2016; Zheng et al., 2021). This suggested that CO₃ ^2-/HCO₃ ⁻ should probably be the main undetected anion. Ca²⁺ and NH₄ ⁺ are the main alkaline ions for neutralizing acidic particles in the atmosphere. In the paper, the equivalent ratios of Ca²⁺/(NO₃ ⁻+SO₄ ^2–) were higher than 1 in four seasons, especially in winter. However, the ratios of NH₄ ⁺/(NO₃ ⁻+SO₄ ^2–) were less than 1. These illustrated that NH₄ ⁺ was not enough to neutralize NO₃ ⁻ and SO₄ ^2-, while Ca²⁺ was not only formed to combine with NO₃ ⁻ and SO₄ ^2-, but the remaining Ca²⁺ may also combine with CO₃ ^2-/HCO₃ ⁻, which also verified the previous conclusion. Therefore, the dominant inorganic compound in these samples should be CaCO₃, similar to the result found at Mt. Yulong (Zhang X. Y. et al., 2012).

The ratio of NO₃ ⁻/SO₄ ^2- is an important index to distinguish between stationary and mobile sources of sulfur and nitrogen in the atmosphere. In general, when the mass ratio of NO₃ ⁻/SO₄ ²⁻ is greater than 1.0, indicating that the emission of mobile sources is dominant, and vice versa, the stationary sources are dominant (Arimoto et al., 1996; Ye et al., 2017). In this paper, the ratios of NO₃ ⁻/SO₄ ²⁻ varied from 0.00 to 0.82, with an average value of 0.31, which was relatively similar to that found in some regions in China, such as Shiyi Glacier (0.37; Zhang Y. et al., 2016), Mt. Yulong (0.33, Zhang NN. et al., 2012). In semi-urban and remote regions, transportation facilities and vehicular emissions were limited. Therefore, the sources were mostly from stationary sources. The ratios of NO₃ ⁻/SO₄ ^2- had obvious seasonal variations at TGS. The seasonal sequence of the ratio of NO₃ ⁻/SO₄ ^2- was summer (0.38)> autumn (0.35)> spring (0.28) >winter (0.16). In autumn and summer, the high temperature was conducive to the decomposition of organic matter, especially in summer. In addition, there were frequent human activities in autumn, which produced abundant nitrogen. These were important sources of NO₃ ⁻, so the NO₃ ⁻/SO₄ ^2- ratios were relatively high.

The abundances of OC, EC, OC/EC and K⁺/EC ratios in the samples are shown in Table 1. The mass concentration range of OC was 0.94–9.00 μg m⁻³, with an average of 3.54 ± 1.82 μg m⁻³. The mass concentration range of EC was 0.08–34.84 μg m⁻³, with an average of 4.42 ± 6.78 μg m⁻³. In autumn, the concentrations of OC and EC were significantly higher than other seasons (Figure 4). For OC, the order of decreasing concentration was: spring (4.50 μg m⁻³) autumn (4.48 μg m⁻³), and summer (3.05 μg m⁻³)>winter (2.42 μg m⁻³). However, the seasonal variation of EC was different, the order was: autumn (11.6 μg m⁻³)>spring (5.74 μg m⁻³)>winter (1.22 μg m⁻³)>summer (1.07 μg m⁻³). The seasonal changes of EC are different from other cities, such as Lanzhou (the highest in winter). The seasonal variation of EC was different from that of cities, such as Lanzhou (the highest in winter) (Tan et al., 2017). There are various sources of EC, including diesel emissions, inefficient combustion of fossil fuels, and biomass combustion. In autumn, there were more human activities. The Tianshan Glacier Station was undergoing infrastructure construction, roads were being built outside the station, and vehicles and people were frequently coming and going. At the same time, the weather conditions were relatively dry, the precipitation decreased significantly (2.8 mm), and the erosion of aerosols was significantly reduced, resulting in higher mass concentrations of EC. Therefore, the differences between OC and EC had a lot to do with their seasonal origin, chemical transformation and meteorology.

The variation range of OC/EC was 0.47–21.43, with the annual average of 3.38 ± 3.68. Early research pointed out that the OC/EC ratio of most urban sites in the world is between 1.0 and 4.0 (Cao et al., 2003). The value of OC/EC in the source area of Urumqi River was between this range, showing that OC and EC in this area were affected by anthropogenic sources. The OC/EC ratio of vehicle emissions is between 0.7 and 2.4, and coal combustion is between 0.3 and 7.6. Larger values of this ratio (between 4.1 and 14.5) can usually be found in biomass combustion emissions (Wang Y. et al., 2016). The value observed at this test point was greater than the value of automobile exhaust emissions and lower than the value of biomass combustion, which was mainly affected by coal combustion. However, the OC/EC value had an obvious seasonal variability, the highest in summer (4.50 ± 2.68) and the lowest in autumn (2.17 ± 5.14). The sources of OC and EC were slightly more complicated and affected by biomass combustion and automobile exhaust respectively in summer and autumn.

In addition, the K⁺/EC ratio was also applied to assess the sources of fossil fuel combustion (0.03–0.09) and biomass combustion (0.2–0.5) (Tao et al., 2016). The K⁺/EC ratio was 0.08 ± 0.08 in our samples, showing that the concentration of EC was more affected by fossil fuel combustion emissions. Compared with the overall data, in summer, the ratio of K⁺/EC was 0.14, which was closer to 0.2. In summer, due to the Covid-19, Urumqi was under absolute blockade, and residents were quarantined at home. During this period, on the one hand, human activities were reduced, and on the other hand, National Highway 216 was also blocked. So, the mass concentration of EC had been greatly reduced. At the same time, precipitation was most abundant in summer, and the removal of aerosol particles by precipitation was another likely reason for the lower EC concentrations during this period. Therefore, the K⁺/EC ratio was the highest in summer. This was consistent with the analysis of the OC/EC ratio. In other seasons, the ratio of K⁺/EC was closer to the ratio of fossil fuel combustion.

Studies have shown that the good correlation between OC and EC can reflect their homology (Turpin et al., 1990). There were moderate correlations between the OC value and the EC value in autumn (R ² = 0.59) and summer (R ² = 0.63), while there was a significant correlation between the two in spring (R ² = 0.86) (Figure 5), indicating that the emission and transmission process of the two were more similar in spring. The higher OC and EC values in spring might be due to the impact of sand and dust storms, and the long-distance air masses from Central Asia carried more carbon components. However, the correlations were relatively low in other seasons compared to spring (R ² = 0.1–0.6); this suggested that there were multiple sources of OC and EC that were not always correlated.

According to earlier researches (Zhang X. et al., 2014; Li et al., 2016; Xiong et al., 2017), the estimation method of SOC is as follows: SOC = OC - EC ∗ ( OC / EC ) min where (OC/EC) _min is the minimum value of the OC/EC ratio after excluding specific data of obviously changes in the OC/EC ratio caused by rainfall, dust, or other factors. The estimated concentration range of SOC was 0.03–5.62 μg m⁻³, with the average concentration of 2.04 ± 1.05 μg m⁻³. The concentration of SOC was higher in summer and autumn, especially in summer. It also could be seen from Figure 4 that the values of SOC/OC were also higher in summer. Previous studies had shown that the formation of SOC was related to an increase in humidity or temperature (Huang et al., 2012; Niu et al., 2013; Zhang F. et al., 2015). According to the meteorological data collected by the automatic weather station, the relative humidity (75%) and average temperature (13.80°C) in summer were higher than other seasons, which may provide favorable meteorological conditions for the formation of SOC. In winter, the stable atmospheric conditions and higher concentrations of VOCs were conducive to the formation of SOC (Jia et al., 2016). This is consistent with the fact of heating in winter.

Eight kinds of carbon components (OC1-OC4, OP and EC1-EC3) were used for source analysis of carbonaceous components, which was determined following the IMPROVE TOR protocol (Chow et al., 2001). Specifically, OC1 accounts for a relatively high proportion of biomass combustion; OC2 and OC3 may mainly cause by secondary organic reactions and biomass combustion; OC4 and EC1 represent coal combustion and automobile exhaust; OP is mainly related to water-soluble organic carbon; EC2 and EC3 are released at high temperature and may be portion of diesel engine exhaust (Cao et al., 2005).

The average percentage of the eight carbon components in the sample is shown in Figure 6. The carbon component had obvious seasonal variation characteristics. EC2 reached the highest value (70.5%) in autumn, indicating that the impact of diesel engine exhaust during this period was more serious. This was consistent with the fact that construction was underway near the sampling point. OC3, OC4, OP and EC1 were the highest in summer, accounting for 10.9, 7.8, 30.8 and 9.0%, respectively. In winter, their proportions were 8.7, 4.9, 28.0 and 8.8%. The four components were not very different in summer and winter, and they were all affected by secondary reaction, biomass combustion, and coal combustion. summer was slightly higher, mainly because favorable weather conditions such as high relative humidity, high temperature and solar radiation were conducive to form organic carbon.

OC, EC, and ion mass concentrations and associated ratios from previous studies in different locations, including glacial and urban areas, were listed for comparison (Table 2). The OC and EC values (3.54, 4.42 μg m⁻³) were higher than the values (1.49 μg m⁻³ for OC and 0.33 μg m⁻³ for EC) in Qinghai Lake measured by Zhang (Zhang X. et al., 2014), but lower than some cities in China, such as Urumqi (32.0, 10.8 μg m⁻³), Lanzhou (17.7, 9.0 μg m⁻³) and Beijing (15.3, 5.27 μg m⁻³) (Wang M. et al., 2016; Zhang Y. et al., 2018; Turap et al., 2019). Early studies suggested that most urban sites around the world have OC/EC ratios between 1.0 and 4.0 (Cao et al., 2003). In this study, the ratio of OC/EC was lower than that of Qinghai Lake area and close to that of urban area. These comparisons showed that the concentrations of carbon components in the source area of Urumqi River were also affected by anthropogenic activities. The value of K⁺/EC was lower than that of Qinghai Lake, indicating that the burning of fossil fuels in this area had a greater impact on the concentrations of OC and EC. At the same time, we used analysis of variance (ANOVA, SPSS 22) to analyze the OC, EC and OC/EC, K⁺/EC of the study area and the urban areas (Urumqi, Lanzhou and Beijing). The results showed that there was no significant difference between the measured values in the study area and the urban areas (p = 0.60 > 0.05), which also indicated that the OC and EC in the study area had been subject to certain anthropogenic activities.

The concentration of Ca²⁺ was the biggest in TSP samples, which was higher than some urban areas and was more affected by dust storms. The concentrations of SO₄ ^2- and NO₃ ⁻ were lower than that in urban areas and similar to those in remote areas, indicating that they were mainly from natural sources. The ratio of NO₃ ⁻/SO₄ ^2- was lower than that in urban areas and similar to those in glacier areas, indicating that nitrogen and sulfur in aerosols mainly came from stationary sources. Then we used the analysis of variance method to analyze the water-soluble ions (NO₃ ⁻, SO₄ ^2-, Na⁺, NH₄ ⁺, Ca²⁺) and ion ratio (NO₃ ⁻/SO₄ ^2-) in the glacier areas. The results showed that there was no significant difference between the data in the glacier region (p = 0.51 > 0.05). Therefore, although the source region of Urumqi River has been polluted to a certain extent, natural sources still play a dominant role.

This study used principal component analysis (PCA, SPSS version 22.0) to analyze the potential origin of chemical components. Varimax rotation is used to make the loading pattern of the factor clearer, that is, the loading of some obvious variables in a factor tends to 1, while for other non-obvious variables, and the corresponding loading tends to 0. Variables are considered indicators of source categories (factors) only when their respective loadings >0.5 (Callén et al., 2009; Xu et al., 2017; Zhang Y. et al., 2018). Factors with eigenvalues >1.0 remain in the results. The results were shown in Table 3.

In autumn, four factors were appraised by principal component analysis, which explicated 90% of the total variance. Factor 1 had high loadings for K⁺, Mg²⁺, Ca²⁺, NO₃ ⁻, OC3, OP, OC4, and EC1, EC2 (all above 0.7), which indicated that dust (construction dust and mineral dust), biomass combustion, fossil fuel combustion, and automobile exhaust emissions might account for 59%. Factor 2 was principally influenced by OC2, OC3 and Cl⁻, NH₄ ⁺, explaining 21% of the variance. The primary emissions generated during the combustion process and the products of agriculture and animal husbandry might have a bearing on this factor. The main load of SO₄ ^2- in Factor 3 indicated that it was related to the burning of fossil fuels. In winter, a total of 3 principal components were extracted to explain 81% of the total variance. The load of Cl⁻, NO₃ ⁻, SO₄ ²⁻, K⁺, and OC2, OC3, OC4, OP, EC1 in Factor 1 was more than 0.7. The research showed that the combined effects of secondary reactions and the burning of biological and fossil fuels caused 50% of pollution. Ca²⁺, Mg²⁺ and EC2 in Factor 2 explained 22% of the total variance, mainly from dust and diesel-vehicle exhaust. In spring, four factors explained 94% of the total variance. Factor 1 had high loadings for Cl⁻, SO₄ ²⁻, Na⁺, K⁺, Mg²⁺, Ca²⁺, and OC4, OP, EC1, EC2, EC3, indicating the sources of dust, evaporated rock salt, coal combustion and vehicle exhaust. Factor 2 had high loadings for OC2, indicating the contribution of secondary organic carbon (SOC), accounting for 14% of the total variance. In summer, 4 factors were identified, which accounted for 84% of the total variance. Factor 1 was composed of SO₄ ^2-, K⁺, Mg²⁺, Ca²⁺, and OC2, OC3, OC4, OP, EC1, EC2, indicating the contribution of sand dust, secondary organic carbon, biomass combustion, coal combustion and automobile exhaust. Factor 2 might cause by Salt Lake evaporation under high temperature conditions because of high contents of Na⁺ and Cl⁻. The loadings of NO₃ ⁻ and NH₄ ⁺ in Factor 3 were above 0.5, indicating the impact of anthropogenic pollution. The high loading of OC1 (0.80) in Factor 4 indicated the source of biomass combustion.

In summary, the main pollution sources in this mountainous glacier area are sand, dust, rock salt, fossil fuel combustion, biomass combustion, and secondary reactions, which can account for more than 70% of pollution in each season.

Clustering trajectories of different seasons during the sampling time are displayed in Figure 7. Most of the air masses transported from the west and northwest were in good agreement with the meteorological data and mainly affected by the westerly circulation. In spring, the trajectories were classified into three clusters, and most short-range trajectories from the arid areas around the Tianshan Mountains accounted for 96.7% of the total. The air mass in Cluster 1 originating from the Gurbantunggut Desert in the northern part of the sampling site accounted for 50.0%, and the air mass in Cluster 2 originated from the Sary Ishikotrau Desert in Kazakhstan and accounted for 46.7%, which may bring more sand and dust particles, while carrying more carbon components, especially EC. Thus, the concentrations of Ca²⁺, Mg²⁺, OC, and EC were higher. The air mass from Central Asia and Siberia (Cluster 3) had the lowest frequency of long-range aerosol transport, accounted for 3.3% of the total trajectories. In summer, the air masses in Cluster 2 and Cluster 4 originating from the arid regions of Kazakhstan increased gradually, accounting for 27.5% of the total trajectories, resulting in the input of some anthropogenic pollution, which may cause the concentration of NO₃ ⁻ and SO₄ ²⁻ to be higher than in winter. The air masses in Cluster 3 originating from the Gurbantunggut Desert north of the sampling site decreased to 25.0%, while the air mass in Cluster 1 arising from the Tianshan Mountains in the west of the sampling point increased, accounting for 47.5%. The air masses (Cluster 1, 46.9%) in autumn originated from the Gurbantunggut Desert and then passed through the Shihezi urban area, which may cause the input of desert dust and anthropogenic pollution. Therefore, the concentrations of SO₄ ²⁻, NO₃ ⁻, and Ca²⁺ were very high. The air mass originating from Lake Issyk-Kul in Kyrgyzstan (Cluster 3) accounted for 18.8% of the total air masses, which also brought more Na⁺ and Cl⁻. The air mass in Cluster 2 originating from Uzbekistan and passing through the arid regions of Tajikistan, Kyrgyzstan, and Kazakhstan accounted for 34.3% of the total trajectories. In winter, the proportion of short-range trajectories originating from the Tianshan Mountains west of the sampling point continued to increase, accounting for 56.7% of the total trajectories. The number of trajectories in Cluster 2 originating from the arid regions of Kazakhstan accounted for 25.0%, of which the trajectories from the Caspian Sea accounted for 18.3%, passing through the arid regions of Uzbekistan, Kazakhstan and Kyrgyzstan, arrived at the sampling point from west.

Overall, the air masses from northwest and west were the main factors that affected the air quality in the source area of Urumqi River. The area was mainly influenced by the short-distance trajectories in the arid regions surrounding the Tianshan Mountains and anthropogenic pollution in the surrounding urban areas (such as Urumqi). In spring, the air masses mainly came from the Gurbantunggut Desert and the urban of Urumqi, bringing more desert dust particles and anthropogenic pollution. The air masses in autumn mainly derived from the Gurbantungut Desert, passing through Shihezi City and other areas, which may cause a certain degree of dust particles and human pollution input. The air masses in summer and winter mainly originated from the Tianshan area west of the sampling site, which was relatively clean. During this period, the research area was also affected by the long-distance transmission from Central Asia, especially in autumn. A large number of fossil fuels and biomass combustion emissions in Central Asia could affect the aerosol concentration in the source area of Urumqi River in the short term through long-distance transmission (Zhang et al., 2019).

The PSCF results are shown in Figure 8. These are the spatial distributions of potential aerosol sources combined with backward trajectories and mass concentrations. The potential source regions with high PSCF values in spring were distributed in the Gurbantunggut Desert, the urban area of Urumqi, and some places in the arid regions of Kazakhstan, which showing no difference with the cluster analysis results of the backward trajectories. The potential sources of atmospheric aerosols in summer and autumn were mainly distributed in the Tianshan area west of the sampling site. However, the mass distributions of aerosols were different. In summer, it mainly concentrated around the sampling points. In autumn, it mainly focused on the vicinity of the sampling site, the Gurbantunggut Desert and Shihezi City. In winter, most of the potential source regions with high PSCF values were located in the surrounding areas of the TGS.

In conclusion, local sources significantly contributed to aerosols. Additionally, the long-range transportation in some western regions, including some parts of Central Asia and Siberia, was also one of the sources of aerosols.