Annual and Periodic Variations of Particulates and Selected Gaseous Pollutants in Astana, Kazakhstan: Source Identification via Conditional Bivariate Probability Function

The degradation of air quality remains one of the most pressing environmental issues as exposure to air pollutants is extensively associated with various health problems including respiratory and cardiovascular diseases. The present study aims to (1) reveal the annual and periodic variations of PM2.5, total suspended particles, and selected gaseous pollutants (SO2, CO, NO2, HF) in Astana, Kazakhstan by analyzing 2-year air pollution monitoring data (October 2018–September 2020) divided into two study cycles (October 2018–September 2019 and October 2019–September 2020, respectively); and to (2) identify potential air pollution sources in the region using conditional bivariate probability function (CBPF). Annual concentrations of PM2.5 and other gaseous pollutants were generally high, exceeding World Health Organization air quality guidelines and nationally adopted air quality standards, with heating periods (October–April) characterized, on average, by higher ambient concentrations than non-heating periods. Notably, the concentrations of observed pollutants were higher during the 2018–2019 study cycle than in 2019–2020. Obtained results are useful for subsequent estimation of the burden of respiratory and cardiovascular diseases in the region. The CBPF analysis of PM2.5 data suggested a general contribution of the coal-fired power plants as well as residential heating activities to the air pollution in the city, while a joint contribution of vehicular emissions and power plant activity was identified as the pollution source of SO2. Control measures for PM2.5 and SO2 emissions specifically arising from the coal-fired power plants need to be urgently implemented.


Introduction
The global degradation in air quality is one of the most critical contemporary environmental issues.Exposure to air pollutants is closely associated with adverse health outcomes, including respiratory and cardiovascular diseases (WHO 2006).These health problems due to ambient pollutants exposure have become a considerable economic burden over time.For example, in 2016, the estimated cost of mortality and morbidity arising from ambient pollutants exposure was 4.8% of the global gross domestic product, which was 5.7 trillion USD (World Bank Group 2020).Particularly, the adverse health impacts of air pollution are associated with exposure to pollutants in particulate matter (PM) and to gaseous pollutants (Morakinyo et al. 2017;Hamanaka and Mutlu 2018;Cheng et al. 2018).
PM can be defined as a combination of solid and liquid suspended air particles with varying size, morphology, and chemical composition.The US Environmental Protection Agency (USEPA) classification of PM is based on aerodynamic diameter (d) of the particles as total suspended particles (TSP) (d < 100 μm), PM 10 (d < 10 μm), coarse PM (PM 2.5-10 , 2.5 μm < d < 10 μm), fine PM (PM 2.5 , d < 2.5 μm), and ultrafine PM (PM 0.1 , d < 0.1 μm) (USEPA 2021).In recent years, the focus of air pollution studies has shifted towards finer fractions of PM due to their ability to penetrate into the lower respiratory system more easily (Farina et al. 2011;Kelly and Fussell 2012), consequently causing more serious adverse human health outcomes (Aztatzi-Aguilar et al. 2018;Farina et al. 2011;Kelly and Fussell 2012;Xue et al. 2019).Similar to PM, human exposure to gaseous pollutants is also associated with numerous health issues, the key adverse outcomes as summarized in Table 1.
PM and other gaseous pollutants may drastically affect urban air quality.Thus, numerous studies have targeted identifying the primary emission sources of air pollutants including fossil fuel combustion, industrial activities, traffic emissions, biomass burning, soil and desert dust, secondary inorganic species, and construction dust (Hopke et al. 2020;Landis et al. 2017;Ramírez et al. 2020;Jan et al. 2020;Zhang et al. 2018).Particularly, fine PM may originate from chemical reactions involving SO 2 and NO x in the atmosphere or be emitted from power plants, industrial sites, and vehicles (USEPA 2021).Fossil fuel combustion is considered a primary source of SO 2 , while CO emissions are associated with biomass burning (Yadav and Devi 2019).Coal-fired power plants have been reported as a major anthropogenic SO x emission source (Lin et al. 2018).Gaseous pollutants such as CO and NO x are linked to vehicle exhaust and incomplete combustion; moreover, NO x contributes to ozone production via photochemical reactions (Atkinson et al. 2018).
The source identification of major ambient pollutants may enhance the understanding of the contribution of particular sources to the levels of air pollution.Consequently, the identification of local and distant air pollution sources through source-receptor modeling has recently received attention (e.g., Jan et al. 2020;Landis et al. 2017;Kim et al. 2018;Ramírez et al. 2020).Receptor models have been applied to source apportionment of PM (e.g., de Miranda et al. 2018;Landis et al. 2017) and volatile organic compounds (VOCs) (e.g., Miller et al. 2002;Song et al. 2008), while the literature on other gaseous pollutants remains more limited (Kumar et al. 2013).The receptor models employing local meteorological data can provide more comprehensive information for source identification.One of these methods is the conditional probability function (CPF) which uses wind direction to identify major pollution sources based on wind and pollutant concentrations (Jeričević et al. 2019;Sooktawee et al. 2020;Uria-Tellaetxe and Carslaw 2014).
CPF is a relatively straightforward approach in providing a dominant source direction in comparison with other tools such as non-parametric regression and trajectory analysis (Kim and Hopke 2004).Moreover, it also enables pinpointing the location of distant point sources and local minor sources that otherwise are challenging to identify.Even in the cases where their contribution at a specific receptor is minor, these distant point sources could have a more significant impact elsewhere; therefore, considering a large spatial area enhances model evaluation opportunities.This is especially relevant in the case of areas with less extensive monitoring systems (e.g., Astana) (Uria-Tellaetxe and Carslaw 2014), which makes CPF a suitable fit for the present study.However, although CPF can provide useful graphical representation, the accuracy of further analysis relies on the judgment of an investigator (Carslaw and Beevers 2013).Moreover, the drawback associated with wind analysis such as CPF lies in the inability to directly coordinate wind speed with distance, leading to possible result misinterpretation (Petit et al. 2017).A more advanced version of CPF is conditional bivariate probability function (CBPF) and uses wind direction and pollutant concentration for source identification while introducing the wind speed as a third variable (Jeričević et al. 2019;Althuwaynee et al. 2020;Owoade et al. 2021;Nguyen et al. 2022).
Kazakhstan has faced a significant degradation of air quality due to rapid population growth, economic development, and resource exploration; and Astana, the capital city of Kazakhstan, remains one of the most polluted cities in the country (Assanov et al. 2021;Kerimray et al. 2018).Major anthropogenic emission sources in Astana include coal-fired combined heating and power plants (henceforth identified as CHP-1 and CHP-2) (Assanov et al. 2021).Given the harsh meteorologic conditions of the region characterized by extreme cold episodes during the winter, these plants are tasked with fulfilling extensive energy demands: in 2020, two power plants combined have produced more than 7 million GCal of heat and coal consumption has increased by 16% from 2015 to 2018 (Serikov 2018).Moreover, residential heating in the city's private districts contributes to air emissions mostly via coal combustion (Kerimray et al. 2018).In Kazakhstan, the emissions from coal-fired power plants drastically exceed permissible levels when compared to most European countries: e.g., by 10-fold for PM 10 , 2.5fold for SO x , and 20% for NO x , vehicle emissions and fossil fuel combustion being major contributors of NO x , CO, and PM emission in the region (Kerimray et al. 2018).
The present study is the first attempt for Astana, Kazakhstan to use CBPF for the source identification of PM 2.5 and gaseous pollutants.Moreover, the identification of the potential emission sources of ambient air pollutants in the region (which includes numerous urban areas alike mostly situated in Russia and Kazakhstan, sharing similar extreme climate characteristics along with comparable pollution characteristics) using receptor modeling techniques has not yet been thoroughly carried out.Given the complexity of the urban environment and the presence of various emission sources, it is essential to investigate the relationship between meteorological factors and pollutants distribution via receptor modeling that has been successfully attempted in certain previous international studies.The objective of the present study is to (1) evaluate spatial and temporal variations of PM 2.5 , TSP, SO 2 , CO, NO 2 , and HF in Astana, Kazakhstan based on the data from the National Hydrometeorological Service of Kazakhstan "Kazhydromet" and the US Embassy in Kazakhstan; and to (2) assess the distribution of pollutant concentration and their relative contribution in the region via CBPF analysis to identify the sources of PM 2.5 and SO 2 .

Study Area
Astana is located in the flat, dry steppe zone (51° 10′ N latitude, 71° 26′ E longitude) 347 m above sea level.The area of the city is 722 km 2 .According to the Köppen climate classification system, Astana has a humid continental climate, characterized by long cold winters and warm dry summers.The average temperature of January, the coldest month, is − 14.2 °C, with the record lowest air temperature of − 51.6 °C (1893).The highest average temperature is registered in July (20.8C), with an extreme summer episode reaching 41.6 °C (1936).The annual average temperature of Astana is 3.5 °C.Astana is also characterized to be windy, with prevailing south and south-west directions.The average wind speed in Astana is 3.9 m/s, along with a common occurrence of winds of more than 20 m/s (Climate data in Astana 2022).
Astana is described by Air Pollution Index data as a city with a "high" level of air pollution and recently is among the two most polluted cities in the country (the other being Almaty, Kazakhstan) (Assanov et al. 2021;Kerimray et al. 2018).The emissions of CHP-1 and CHP-2 (locations shown in Fig. 1) seem to be not effectively controlled.Fly ash collection is achieved via a wet scrubbing method using emulsifiers with an arguable collection efficiency, whereas the collection efficiency of PM 2.5 is not at all specified in the official documents (Assanov et al. 2021).

Data Collection
The data for the present study were obtained from the National Air Quality Monitoring Network of the National Hydrometeorological Service of Kazakhstan "Kazhydromet" from four monitoring stations (Fig. 1; S1, S2, S3, S4).Twoyear monitoring data from October 2018 to September 2019 and October 2019-September 2020 were employed.The data contain 4-h (S1) and 3-h measurements (S2, S3, S4) of TSP, SO 2 , CO, NO 2 , and HF.Furthermore, the air pollution monitoring station at the Embassy of the United States provides hourly measurements of PM 2.5 concentrations (Fig. 1, S5).To compare the values of the variation of pollutants from different periods, normalized values were calculated for the average values, which made variables to be converted to the same scale and thus become comparable.

Methodology for CBPF
The CBF (conditional bivariate function) and CBPF modeling were used to identify potential sources of air pollutants.Mathematical calculation of CBF was performed based on (1): where m Δ is the number of samples in the wind sector Δ with concentration C exceeding or equal to threshold x, and n Δ is the total number of ambient samples in the wind sec- tor Δ .A threshold for the episodes of ambient pollutant concentrations, X, is often taken, e.g., the 75th or 90th percentiles (Uria-Tellaetxe and Carslaw 2014).
CBPF is also based on ambient pollutant concentrations higher than a specific threshold value for a particular wind speed and wind direction sector.CBPF directional information about major emission sources can be compared with the spatial map to determine the consistency with the level of pollutants from the most significant pollution sources measured at air quality monitoring stations.Unlike CBF, CBPF also uses wind speed for the calculation of probability.CBPF is mathematically defined as in (2): where mΔ is the number of samples in the wind sector Δ and wind speed interval Δu with the concentration C exceed- ing or equal to threshold x; nΔ , Δu is the total number of ambient samples in the wind sector Δ and wind speed inter- val Δu (Jeričević et al. 2019;Sooktawee et al. 2020).
Bivariate polar plots give statistical information (e.g., mean values) in a polar pattern i.e., with radial axis (r) and (1) angular axis (θ).In air quality analysis, r and θ represent wind speed and wind direction, respectively.Wind speed and direction data are divided into separate bins, each having a mean concentration distributed into a corresponding bin.In essence, the mean concentration of a particular pollutant is calculated and shaded in the related bin.Surface fitting is done using a Generalized Additive Model (Eq. 3) that can better describe non-linear relationships considering variable interactions in air pollution studies (Sooktawee et al. 2020): where C i is pollutant concentration (µg/m 3 ), 0 is the overall mean of response, s u i , ∨ i is isotropic smooth function of the ith value of wind covariates u i and ∨ i ( u = u.sin(2 ∕ ) and ∨ = u.cos(2 ∕ )) , u is mean with speed (m/s), and i is the ith residual (Uria-Tellaetxe and Carslaw 2014).
In addition to using pollutant point value of concentration exceeding a certain threshold, CBPF can also consider an interval of pollutant concentration that can be defined as in (4): where mΔ , Δu is the number of samples in the wind sec- tor Δ at the wind speed interval Δu (that have a concen- tration C between threshold interval y and x), and nΔ , Δu is the total number of ambient samples at that particular wind direction and wind speed interval.The bivariate CBF model does not consider concentration intervals but only point values greater than a certain threshold.CBPF extends the basic CBF and provides more comprehensive information to identify the most prominent pollution sources as well as the 'hidden' ones based on the contribution to pollutant concentrations (Sooktawee et al. 2020).The programming language 'R' with the Openair package (Carslaw and Ropkins 2012) was used for the source apportionment analysis by CBF and CBPF.

Pollution Profile for PM 2.5
Table 2 demonstrates the annual (i.e., for 2018-2019 and 2019-2020) and periodic [i.e., during heating (October-April) and non-heating periods (May-September)] variations of the PM 2.5 concentrations.The average annual concentration of PM 2.5 was notably lower in 2019-2020 compared to 2018-2019 (16.5 µg/m 3 and 29.7 µg/m 3 , respectively); however, for both cycles, they exceeded the air quality standards guideline values of World Health Organization (WHO) [5 µg/m 3 for annual mean, 15 µg/m 3 for the 24-h (3) Figure 2a, b presents the diurnal and monthly variations of PM 2.5 concentrations.Notably, the PM 2.5 concentrations display recurring weekly peaks on Thursday (Fig. 2a) and Monday (Fig. 2b), which seems to indicate the accumulation of air pollutants during the working days possibly as a result of increased energy consumption during the former part of the week.
However, in the 2019-2020 study cycle, the PM 2.5 concentration is increasing during the end of the week possibly due to increased traffic load (Ji et al. 2019).The PM 2.5 concentration varied throughout the day with a decrease until 18:00 that is followed by a peak at around 23:00.The decrease in the daytime would be associated with reduced heating requirements and higher atmospheric dispersion characteristic of daytime.The peak concentration of PM 2.5 is observed during the midnight possibly due to temperature drop and pollutants build-up under inversion conditions (Bathmanabhan et al. 2010).The highest concentrations of PM 2.5 were observed during December (Fig. 2a) and March (Fig. 2b).
The wind rose and pollution rose diagrams for PM 2.5 were constructed to identify the link between the meteorological conditions and the PM 2.5 concentrations at the receptor (Fig. 3).The wind direction was mostly from between west and south, with elevated wind speeds registered in south-south-west and south-west directions (Fig. 3a).The prevailing direction of the high concentration of fine PM (Fig. 3b) was also observable: the episodes of very high PM 2.5 (> 300 µg/m 3 ) concentrations were in the east and east-north-east directions.These results show that the wind profile favors the city's ventilation characteristics, indicating that the PM 2.5 sources are mainly located in the downwind direction.

Pollution Profile for TSP, SO 2 , CO, NO 2 , and HF
To understand the underlying meteorological mechanisms of increasing pollutant concentrations in Astana, the annual (study cycles 2018-2019 and 2019-2020) and periodic variations (heating (October-April) and non-heating periods (May-September)) of TSP and gaseous pollutants were analyzed based on the data from S1, S2, S3, and S4 air pollution monitoring stations (Tables 3, 4).For the majority of the air monitoring stations, the wind circulation was governed by southern wind (S1, S3, and S4).However, the prevailing wind for S2 was from the west-north-west direction (Supplementary material, Fig. S1).

Study Cycle: 2018-2019
The highest annual average concentration of pollutants (except for TSP) was registered in S4 for the study cycle 2018-2019 (Table 3).S4 is situated in the densest traffic zone in the city.The concentrations of TSP and CO were notably high in comparison to other measured pollutants that can indicate higher contribution from exhaust emissions.The TSP pollution in the atmosphere can reduce visibility by scattering and absorbing light (Liu et al. 2014).Although TSP particles that are larger in diameter cannot penetrate deeper into respiratory tract and exhibit lower retention capability, long-term exposure to high TSP concentrations may induce a cytotoxic effect that is related to the heavy metal constituents in the particles (e.g., Cu, Pb, As, Cd) resulting in an increased risk of airway inflammation (Shim et al. 2021).
TSP was the highest in S3 that is also impacted by the traffic, reaching up to 1500 µg/m 3 and with a mean value of 347 µg/m 3 [significantly higher than the nationally adopted maximum allowable concentration for TSP (150 µg/m 3 )] (Assanov et al. 2021).When compared to the most recent available published data in 2016 (Kerimray et al. 2018), the average annual TSP concentration in 2018-2019 was slightly lower in all four stations (400 µg/m 3 in 2016).However, the maximum concentration of TSP in 2016 (1614 µg/m 3 ) was exceeded by that in the 2018-2019 study cycle (1700 µg/m 3 ).
The highest average concentrations of CO varied from 600 to 2800 µg/m 3 with a mean value of 1157 µg/m 3 .The daily maximum allowable concentration of CO according to Kazakhstani standards is 60 µg/m 3 (Kazhydromet 2021), which is far below the values observed in this study cycle.CO stabilizes in the atmosphere and once inhaled, forms carboxyhemoglobin (COHb) in the blood, reducing the uptake of oxygen by the cells (American Lung Society 2020).The population groups experiencing chronic exposure to the high concentration of CO include those working in public transportation as well as traffic wardens and garage/tunnel workers (WHO 2000).A recent study conducted among autoworkers revealed elevated COHb concentration in the blood serum for this occupational group (Bol et al. 2018).Chronic CO poisoning can result in variety of health conditions, ranging from acute coronary syndrome (e.g., myocardial infarction) to detectable brain damage (Lee et al. 2010).High concentration of CO found in the present study indicates a high potential health risk for any exposed population.
Episodes of high NO 2 concentration were also noted during this study cycle.Although NO 2 can originate from natural sources (e.g., lightning, biological decay), it is mostly linked to vehicular emissions and is a primary indicator of traffic related pollution (Atkinson et al. 2018).Persistent exposure to NO 2 has been linked to mortality associated with the respiratory and cardiovascular disease (Atkinson et al. 2018;Huang et al. 2021).NO 2 concentrations varied from 16.6 (S2) to 211 µg/m 3 (S4).Notably, S4 is the only station with the concentration of NO 2 exceeding WHO air quality guideline values (40 µg/m 3 ) (WHO 2006) which could be linked to relatively heavy traffic in that area.The average annual concentration of NO 2 is significantly higher only in S4 (211 µg/m 3 ) compared to the reported annual average NO 2 concentration in 2016 (80 µg/m 3 ) (Kerimray et al. 2019).
The concentration of selected pollutants (i.e., TSP, SO 2 , CO, NO 2 , and HF) in 2018-2019 increased twofold, on average, during heating period compared to the non-heating season (Fig. S2).A reasonable explanation for this trend, as for PM 2.5 , is the increased consumption of fuel during colder periods which couples the impact with unfavorable meteorological conditions.During the heating period, the mean concentrations of TSP reached 416 µg/m 3 (S1), and the mean concentration of CO was 1321 µg/m 3 (S4).During the nonheating period, the highest mean concertation of CO and TSP were 291 µg/m 3 and 914 µg/m 3 , respectively (S4).Interestingly, the concentration of NO 2 during the non-heating period was notably higher in S4 than in the heating period, with an average of 257 µg/m 3 and 179 µg/m 3 , respectively, addressing the highest levels of traffic impact on S4.

Study Cycle: 2019-2020
During the study cycle 2019-2020, the concentration of some measured ambient pollutants was notably lower than in 2018-2019 (Fig. S3).Similarly, CO and TSP had the highest average annual concentration among other pollutants.A large difference in TSP concentrations between the stations  The concentration of pollutants was 35% higher, on average, during heating period than non-heating season.During the heating period, the mean concentration of CO and TSP reached 1232 µg/m 3 (S4) and 191 µg/m 3 (S4), respectively.During the non-heating period, mean concentration of CO was 585 µg/m 3 in S4 with a range of 440-871 µg/m 3 .Similarly, TSP concentration was the highest in S4 during the non-heating period, with a mean of 217 µg/m 3 and ranged from 100 to 400 µg/m 3 .Similar trend of higher concentration during the heating period compared to non-heating season can be attributed to slower movement and increased density of air masses trapping the ambient air pollutants and increased heating from residential area (Manisalidis et al. 2020).

Pollution Profile for SO 2
The concentration of SO 2 during two study periods, in general, did not exceed WHO air quality guidelines.Results also revealed that the SO 2 concentration was higher during the heating period compared to the non-heating period.Moreover, the difference between the two periods was small (Figs.S2, S3).The highest mean annual concentration belonged to 2019-2020 (S1) (1.45 µg/m 3 ) (Table 4).There were episodes of high SO 2 concentrations (> 100 µg/m 3 ) in S1 in November 2019 and February 2020 (Figs.S2, S3).The highest monthly mean concentrations were recorded mostly in February for two study periods in S1, S2, and S3.For S4 highest monthly mean concentration was observed in March (Figs.S2, S3).
The pollution rises (Fig. S8) suggest that potential sources of high SO 2 concentration for S1 were located in the north, north-east, west-south-west, and south direction.A significant source of SO 2 for S2 was located in the south-south-east area of Astana.Moreover, for S3, the highest concentrations of SO 2 were observed in south, north-north-east, east, and east-south-east directions.In addition, south-south-east, south, east-south-east, east, and west were the directions of high SO 2 concentration in S4.

Potential Sources for PM 2.5
Air quality guidelines focus on identifying, quantifying, and mitigating effects from particular emission sources and evaluating control measures to reduce air pollution (Thunis et al. 2019).The bivariate polar plots (Fig. 4) for selected periods on monthly PM 2.5 concentrations and meteorological data could provide valuable information for source identification of the pollutants.A high concentration of PM 2.5 (red shading) was noted at S5 during the winter period (December, January, and February) in the north and east-north-east directions.The potential sources of PM 2.5 are not close to the air monitoring station; however, their location addresses the coal-heated power plants (CHP-1 and CHP-2) and residential heating in the city's private districts, located in that particular direction.Moreover, increased concentration of PM 2.5 during the colder period would also indicate the contribution of power plant emissions.
CBPF plots of PM 2.5 concentrations for the 75th (Fig. 5a) and 95th percentiles (Fig. 5b) were constructed to identify the probability of high PM 2.5 concentration in a particular wind speed and direction sector. Figure 5a indicates the contribution of CHP-2 to the PM 2.5 emission located in the northeastern area of Astana.Results also revealed that the highest concentration of fine particles (95th percentile) was concentrated in the center of the polar coordinate (Fig. 5b), indicating the low wind speed and high concentration at the receptor (i.e., air quality monitoring station) (Sooktawee et al. 2020).The potential sources can also include vehicular fuel combustion (Askariyeh et al. 2020) which may be associated with the heavy traffic in this part of the city.
Although CBPF plots provide valuable information on the spatial distribution of PM 2.5 , they lack data on temporal variation in the concentration of PM 2.5 .Therefore, hourly CBPF plots were constructed in the study to analyze the diurnal variation in the concentration of PM 2.5 (Fig. 6).A higher probability of 95th percentile PM 2.5 concentrations was observed at 8:00, 22:00, and 23:00.One of the possible reasons for increased concentration during the nocturnal period is relatively stable conditions of the atmosphere and reduced boundary layer, decreasing vertical dispersion of the pollutants (Sooktawee et al. 2020).

Potential Sources for SO 2
In Astana, CHP-1 and CHP-2 exceeded the maximum allowable emissions level for SO 2 in 2019.Moreover, the boiler unit (e.g., No. 7 power boiler) in the power plants is not equipped with a functioning dry desulfurization system designed to minimize SO 2 emissions into the atmosphere (Azhigaliyev 2019).As a result, in Astana, the concentration of SO 2 changed inconsistently with the highest annual mean concentrations observed recently in 2014 (31 µg/m 3 ) and 2016 (40 µg/m 3 ).Although these measurements of SO 2 concentrations were in accordance with Kazakhstani standards on maximum permissible concentration, they exceeded WHO annual limit values (20 µg/m 3 ) (WHO 2006).The maximum concentration measurements were also measured in 2014 (1653 µg/m 3 ) and 2016 (1614 µg/m 3 ) (Kerimray et al. 2018).The CBPF plots were constructed to identify the source of SO 2 emission (Fig. 7).
CBPF analysis suggests that there is a certain probability (> 20%) of SO 2 concentration higher than the 75th percentile (> 1 µg/m 3 ) to occur in the east, east-south-east, and southeast area of Astana (S3 and S4) (Fig. 7).Furthermore, the peak concentrations of SO 2 are also observed in the southwestern direction, showing a previously unidentified significant emission source (Fig. 7a).Further investigation has identified the following potential sources near these locations: a residential district located in the south-south-east direction, and a nearby concrete plant.Moreover, S1, which is located in the busiest traffic network of roads, contributes to the air quality in the region.For S3 and S4, the emission of high SO 2 concentration is related to CHP-2 activity located in the northeastern area and residential heating from private districts in the east, east-south-east and southeastern areas of Astana.

Conclusion
Annual and periodic variations in the concentrations of PM 2.5 , total suspended particles (TSP), and selected gaseous pollutants (SO 2 , CO, NO 2 , HF) for the city of Astana, Kazakhstan for the study periods 2018-2019 and 2019-2020 were analyzed.The annual concentrations of PM 2.5 , CO, and NO 2 drastically exceeded the World Health Organization's air quality guidelines as well as national air quality standards.Moreover, episodes of extremely high concentrations of ambient pollutants were observed in both study periods.The use of conditional bivariate probability function (CBPF) analysis for source apportionment enabled the identification of the potential locations of sources of high emissions in the region and the relationship with meteorological parameters indicated the need for better intervention measures.The CBPF plots of PM 2.5 concentrations for the 75th (> 29.3 µg/ m 3 ) and 95th percentile (> 87.1 µg/m 3 ) revealed the contribution of the coal-fired power plant activities (CHP-2) as well as residential heating to the air pollution in the city.The CBPF analysis for the 75th percentile (> 1 µg/m 3 ) of SO 2 concentration suggested an unidentified pollution source in the southwestern part of Astana and a joint contribution of vehicular emissions along with the activities of combined heating and power plants.Control measures for PM 2.5 and SO 2 emissions arising from two coal-fired combined heating and power plants need to be urgently implemented in Astana.Further, effective environmental control strategies (e.g., switching to natural gas, sulfur removal from coal) should be employed to reduce ambient air pollutants in the region.

Fig. 1
Fig. 1 Study area (Astana, Kazakhstan) along with locations of its air pollution monitoring stations and coal-fired power plants

Fig. 2
Fig. 2 Diurnal and monthly variations of PM 2.5 concentrations for a 2018-2019 and b 2019-2020

Fig. 3
Fig. 3 Diagrams of a wind rose and of b pollution rose for PM 2.5

Fig. 4
Fig. 4 Bivariate polar plots of monthly mean PM 2.5 concentrations

Table 1
Summary of adverse health outcomes arising from exposure to selected gaseous pollutants