Emission Source Areas of Fine Particulate Matter (PM_2.5) in Ho Chi Minh City, Vietnam

Abstract

This study aims to determine emission source areas of fine particulate matter (PM_2.5) in Ho Chi Minh (HCM) City, Vietnam, using a conditional bivariate probability function (CBPF) and hybrid receptor models, including three-dimensional potential source contribution function (3D-PSCF) and concentration-weighted trajectory (3D-CWT), considering latitudes, longitudes, and height of trajectory segments. Uncertainties of the CBPF and 3D-PSCF/3D-CWT were evaluated based on the 95th confidence intervals and 95% confidence levels, respectively. For the local scale, PM_2.5 in HCM City was primarily emitted from shallow or common ground sources (e.g., vehicle emissions) throughout the year. Regarding non-local source areas, PM_2.5 in HCM City is contributed by those originated from the East Sea (e.g., shipping emissions) and southeastern Vietnam (e.g., Binh Duong and Dong Nai provinces) having several industrial zones with PM_2.5 emission sources, especially in the dry season (December to April of the following year). In the rainy season (May–November), PM_2.5 derived from emission sources in the Mekong Delta (e.g., biomass burning) might be transported to HCM City. However, contribution of the non-local sources to PM_2.5 pollution in HCM City during the rainy season is less important because of PM_2.5 deposition stemmed from the high rainfall amount in this season.

1. Introduction

Vietnam is situated in Southeast Asia, bordered to the west by Laos and Cambodia and to the north by China. The Vietnamese economy is one of the most rapidly rising in Asia. Additionally, Vietnam has become a long-term desirable market because of its average annual growth rate of gross domestic product, steady population growth, and prosperity increase [1]. Most of Vietnam’s area is covered by mountains and hills, with the remainder being flat plains. In addition, the topography of Vietnam diminishes from the northwest to the southeast, causing Vietnamese’s climate to change significantly according to latitude and relief [2]. Particularly, northern Vietnam experiences significant temperature variations among seasons (i.e., spring, summer, fall, and winter), whereas southern Vietnam is mostly warm throughout the year and has two seasons per year (i.e., dry and rainy seasons) [3].

Ho Chi Minh (HCM) City, located in southern Vietnam, is the political, economic, and cultural hub of the country. Due to its high population and traffic density [4], HCM City has faced several environmental problems, including air pollution. The levels of air pollutants in HCM City, such as particulate matter (PM), nitrogen oxide (NO_x), sulphate dioxide (SO₂), carbon monoxide (CO), and ozone (O₃), were reported to noticeably exceed the thresholds for air pollutants introduced by the World Health Organization (WHO) and Vietnamese government [5,6,7]. In particular, considering the global air quality guidelines suggested by WHO, the 24 h (annual) average concentrations of PM_2.5 (PM lower than 2.5 μm), PM₁₀ (PM lower than 10 μm), NO₂, and SO₂ are 15 (5), 45 (15), 25 (10), and 40 μg m⁻³, respectively [8]. For the Vietnamese air quality guidelines, the thresholds for 24 h (annual) average concentrations of PM_2.5, PM₁₀, NO₂, and SO₂ are 50 (25), 150 (50), 100 (40), and 125 (50) μg m⁻³, respectively [9]. Regarding air pollution in HCM City, the 24 h (annual) average concentrations of PM_2.5, PM₁₀, NO₂, and SO₂ were reported as (36.3), 73, 22, and 22 μg m⁻³, respectively [5,6]. These concentrations relatively reached the thresholds issued by the Vietnamese government but were 2–6 times higher than the those introduced by WHO. Thus, exposure to these pollutants (i.e., PM, NO_x, SO₂, CO, and O₃) through inhalation and/or dermal contact can adversely affect human health. Of these pollutants, fine particulate matter, also called PM_2.5, is one of the most frequently monitored pollutants [10] because compositions of PM_2.5 can contain several toxic chemicals, such as trace metals (e.g., arsenic, lead, and mercury), organic compounds (e.g., polycyclic aromatic hydrocarbons and polychlorinated biphenyls), and inorganic compounds (e.g., anions and cations). In addition, PM_2.5 has a small size (i.e., lower than 2.5 μm), thus, it can deeply penetrate the human respiratory system and cause many diseases, such as lung irritation, coughing, asthma, and lung cancer [11].

Based on the emission inventory analysis, PM_2.5 in HCM City has been noted to be mainly emitted from transportation activities, industrial productions, and household emissions [12,13,14]. Additionally, according to directions of backward air trajectories, several regions were reported to be the possible emission source areas of PM_2.5 in HCM City, such as the Mekong Delta at southwestern HCM City [5,15], and some provinces in southeastern Vietnam, including Binh Duong, Dong Nai, and Binh Phuoc provinces [14]. However, to improve the accuracy of source area identification, the backward air trajectories should be associated with PM_2.5 concentrations at the receptor site [16]. Moreover, information on determining the emission source areas of PM_2.5 in HCM City using receptor models, linking air parcels or winds to pollutant concentrations, is limited. Thus, studies on this issue are essential to further understand PM_2.5 pollution and support decision-making related to atmospheric contamination in HCM City, a metropolis of Vietnam.

To identify emission source areas of PM_2.5, conditional bivariate probability function (CBPF) and hybrid receptor models can be used. The CBPF approach combines pollutant concentrations at receptor site, wind direction, and wind speed to determine emission source areas of pollutants [17]. In addition, the CBPF has been widely used to determine local source areas of air pollutants [18,19,20,21] because pollutant dispersion on a local scale (e.g., over a small area) is mainly affected by wind. Added to this, a combination of the CBPF prediction surfaces and their lower/upper 95th confidence intervals should be employed to increase reliability of the emission source areas [22]; however, this approach has not been commonly used in previous studies [18,19,20]. The hybrid receptor models combine pollutant concentrations and geocoordinates of backward air trajectories arriving at the receptor site; therefore, this model type can be used to identify emission source areas of air pollutants at a non-local scale (e.g., over several hundred to thousands of kilometers from the receptor site) [16,23,24]. Of these models, potential source contribution function (PSCF) and concentration-weighted trajectory (CWT) models have been extensively applied because of their relatively straightforward performance [25]. Particularly, in the PSCF/CWT analysis, the study area is divided into numerous grid cells. The value of each is determined based on the endpoints of air trajectories falling inside each cell together with the pollutant concentrations associated with the air trajectories [26]. Generally, grid cells with high PSCF/CWT values are suggestive of emission source areas of air pollutants [27].

Recently, three-dimensional (3D) hybrid receptor models, considering longitudes, latitudes, and heights of trajectory segments, have been introduced and widely applied to improve the accuracy of model outputs [18,23,28,29]. One of the approaches is to consider trajectory segments transporting within the air mixing layer [18,30], since air pollutants are well-mixed and diffused in this layer once they are released [31]. Added to this, air trajectory segments higher than the mixing heights are likely to capture air pollutants originated from other source areas [18]. Therefore, the 3D hybrid receptor models (e.g., 3D-PSCF and 3D-CWT), considering vertical transport of trajectory segments, have been reported to identify emission source areas of air pollutants more effectively. Additionally, uncertainties of the 3D hybrid receptor models should be evaluated to limit the misinterpretation of source areas, which can be stemmed from a small number of trajectory segments [32] and/or trailing effects [16]. In this regard, a weighting function can be used to lower the uncertainty of 3D hybrid receptor models [32] and it has been shown to noticeably affect the output of 3D-PSCF/3D-CWT [18]. However, previous studies mostly used a single set of the weighting functions, which can lead to the omission in identifying emission source areas. Consequently, various sets of the weighting functions should be used to avoid this limitation and provide uncertainty of the 3D hybrid receptor models.

This study aims to identify emission source areas of PM_2.5 in HCM City at local and non-local scales (i.e., inside and outside the city, respectively), utilizing the CBPF, 3D-PSCF, and 3D-CWT models. In addition, variations in the PM_2.5 mass concentrations and their relationships with several meteorological parameters (e.g., air temperature, air pressure, and wind speed) are evaluated. Notably, uncertainties of the CBPF and 3D-PSCF/3D-CWT were evaluated based on the upper/lower 95th confidence intervals and the 95% confidence levels, respectively. The findings of this study add to the understanding of identifying emission source areas of PM_2.5 and to the decision-making related to air pollution reduction in HCM City, a metropolis of Vietnam.

2. Materials and Methods

2.1. Study Area

Ho Chi Minh (HCM) city has two distinct seasons: the dry season lasting from December to April of the following year and the rainy season covering a period from May to November [4]. The average ambient air temperature and annual rainfall amount of HCM City are approximately 28.4 °C and 2100 mm, respectively [2]. The ambient air temperature of the dry season (28.1–29.3 °C) tends to be higher than that of the rainy season (26.9–27.5 °C) [4]. In addition, the rainfall amount is relatively low in the dry season; however, it rises dramatically in the rainy season and accounts for approximately 80–90% of the total annual rainfall amount, owing to the frequency of rain events in the rainy season [2].

In terms of air pollution, transportation activities have been noted as one of the primary emission sources of air pollutants (e.g., PM_2.5, NO_x, SO₂, CO, and volatile organic compounds (VOCs)) in HCM City [12,14] due to many vehicles (e.g., approximately 8 million) operating in the city [33]. Furthermore, industrial activities, especially textile and food industrial production, have also been one of the main emission sources of air pollutants in HCM City [12,14,34] because fossil fuels, such as diesel oil and coal, are used in the manufacture of these industrial types [12].

2.2. Data Collection

2.2.1. Mass Concentrations of PM_2.5

The hourly mass concentrations of PM_2.5 from 1 January 2019 to 31 December 2020 in this study were obtained from the air pollutant data which are available via the US Environmental Protection Agency (EPA) and the AirNow platform (https://www.airnow.gov/international/, accessed on 15 March 2021), accessed on 15 March 2021. The monitoring site (latitude: 10°47′0.1″ N and longitude: 106°42′2.2″ E) is located at the US Consulate General Ho Chi Minh City in a downtown area of District 1, HCM City (Figure 1). The mass concentrations of PM_2.5 were determined based on the beta-ray attenuation method [35,36], meaning that the relative shift in the intensity of beta ray passing through un-sampled and sampled filter tapes was used to calculate the mass of PM_2.5 depositing on the filters [37]. The dataset of PM_2.5 mass concentrations was firstly checked for invalid values (i.e., QC invalid labeled, missing, negative, and impossibly high values). The invalid data, accounting for 2.99% of the whole dataset, was regarded as null data and filled in using linear interpolation [38]. Consequently, the final dataset had a total number of 17,520 values of the PM_2.5 mass concentrations.

2.2.2. Meteorological Data

Apart from the PM_2.5 mass concentrations, meteorological data in HCM City (i.e., ambient air temperature, wind speed, wind direction, dewpoint temperature, and air pressure) were also gathered to understand relationships between the meteorological parameters and the PM_2.5 mass concentrations in HCM City. The hourly data of these meteorological parameters were downloaded from the National Centers for Environmental Information (NCEI) (https://www.ncei.noaa.gov/maps/hourly/, accessed on 15 March 2021) and the selected station is Tan Son Nhat (ID: 4890009999, latitude: 10°49′2″ N, longitude: 106°40′1″ E) (Figure 1). In addition, hourly data of relative humidity in the study period (1 January 2019–31 December 2020) were calculated based on the downloaded data of the ambient air temperature and dewpoint temperature data following Lawrence [39]. Data on the rainfall amount in HCM City during the study period were also obtained from the Mac Dinh Chi station (latitude: 10°47′2″ N, longitude: 106°42′1″ E) (Figure 1), operated by the Hydrometeorological Observatory, Southern region of Vietnam.

2.2.3. Satellite-Derived Data

The satellite-derived data, including aerosol optical depth (AOD) and incoming shortwave flux in the study period (i.e., 1 January 2019 to 31 December 2020), were also collected to support the result interpretation. In particular, this study used the daily AOD at 550 nm with 1 degree spatial resolution from the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard the Terra satellite [40]. The bounding domain for downloading the AOD values was 85°–125° E and 5°–30° N. Additionally, to know the solar radiation intensity in HCM City, hourly data on incoming shortwave flux for 0.5° × 0.625° grid cells over a domain encompassing 10.5°–11° E and 106.5°–107° N were obtained from the Modern-Era Retrospective Analysis for Research and Applications version 2 (MERRA-2) [41]. Moreover, precipitation data over the study area were obtained from the Integrated Multi-satellite Retrievals for Global Precipitation Measurement (GPM IMERG) final product [42] to support the interpretation of PM_2.5 transport from non-local source areas to HCM City.

2.3. Conditional Bivariate Probability Function (CBPF)

To identify emission source types (e.g., stack and shallow level sources) and emission source areas of PM_2.5 at a local scale (i.e., inside HCM City), the CBPF tool can be used [17]. This is because diffusion of PM_2.5 at the local scale (e.g., over a small area) is mostly affected by wind speed that is associated with the CBPF. The CBPF for the dry and rainy seasons of HCM City were calculated using this equation [17]:

$${ \mathrm{CBPF}}_{∆\theta ,∆u}=\frac{{\mathrm{m}}_{∆\theta ,∆u}|C \ge x}{{\mathrm{n}}_{∆\theta ,∆u}}$$

(1)

where ∆θ and ∆u indicate wind sector and wind speed, respectively. C is the PM_2.5 mass concentration higher than threshold x (i.e., 75th percentiles of the PM_2.5 mass concentrations) connected with wind sector ∆θ and wind speed ∆u. m_∆θ,∆u is the number of times the PM_2.5 mass concentrations in wind sector ∆θ and for wind speed ∆u greater than the threshold x. n_∆θ,∆u represents the total number of samples for the corresponding wind sectors and wind speed intervals. Considering the CBPF uncertainty, this study simultaneously used the prediction and uncertainty CBPF plots, showing the predicted emission source areas and the lower or upper 95% confidence intervals, respectively [22]. The CBPF uncertainty plots were computed using the Generalized Additive Model (GAM) [22,43]. The openair package [22] installed in the RStudio Desktop [44], an open-source software, was used to generate the CBPF prediction and uncertainty plots. Generally, the potential source areas contributing to the higher mass concentrations of PM_2.5 are marked by the high CBPF probabilities. Additionally, areas with similar probabilities between the prediction and uncertainty plots indicate emission source areas having high reliability. On the other hand, the suggested source areas are relatively uncertain due to a lack of data on the corresponding wind speed and wind sector [43].

2.4. Backward Air Trajectory and Trajectory Cluster Analysis

2.4.1. Backward Air Trajectory

The emission source areas of PM_2.5 on the non-local scale (i.e., outside HCM City) can be identified using the 3D-PSCF and/or 3D-CWT models adopting the backward air trajectory [45]. This study used the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model version 5.1.0 [46] coupled with Global Data Assimilation System (GDAS1) meteorological data (https://www.ready.noaa.gov/gdas1.php, accessed on 15 March 2021) to simulate the backward air trajectories arriving at the PM_2.5 monitoring site in HCM City from 1 January 2019 to 31 December 2020. The total run time for backward air trajectory simulation was 96 h (4 days), and the starting heights of the trajectories were set at 50% and 100% of the mixing layer height at the receptor site to acquire high resolution of the hybrid receptor model. A total of 35,040 air trajectories (365 monitoring days × 2 years × 2 starting heights × 24 hrs/day) were used in this study. The uncertainty of air trajectory simulation was also considered by removing trajectory segments exceeded 50% of the mixing layer height [18].

2.4.2. Trajectory Cluster Analysis

To determine the main directions of backward air trajectories, trajectory cluster analysis, performed using TrajStat 1.5.3 [27], was used to group the air trajectories according to their directions and lengths. The Euclidean distance was selected for the clustering and was calculated based on this equation.

$${\mathrm{d}}_{12}= \sqrt{{{\displaystyle \sum }}_{i=1}^n{\left[{\mathrm{X}}_1 \left(i\right) - {\mathrm{X}}_2 \left(i\right)\right]}^2+{\left[{\mathrm{Y}}_1 \left(i\right) - {\mathrm{Y}}_2 \left(i\right)\right]}^2 }$$

(2)

where d₁₂ indicates the Euclidean distance between two backward trajectories. X₁, X₂ and Y₁, Y₂ represent the latitudes and longitudes of these trajectories, respectively. The fluctuation in the total spatial variance (TSV) was computed to identify the number of clusters, and a significant change in the TSV implies an adequate cluster number [47].

2.5. Three-Dimensional Hybrid Receptor Models

To determine emission source areas of PM_2.5 in HCM City at the non-local scale (i.e., outside HCM City), this study used the 3D-PSCF and 3D-CWT models, considering the heights, latitudes, and longitudes of trajectory segments. The 3D-PSCF model was computed following Equation (3) [16].

$$3\mathrm{D}-\mathrm{PSCF}=\frac{m_{ij}}{n_{ij}}$$

(3)

where 3D-PSCF represents the three-dimensional potential source contribution function. m_ij is the number of endpoints falling within cell (i,j) when the PM_2.5 mass concentrations are higher than the threshold values (i.e., 75th percentiles). n_ij denotes the number of times the trajectories were processed as described in Section 2.4.

The 3D-CWT model was also computed to confirm the PM_2.5 emission source areas suggested by the 3D-PSCF and to clarify source areas with high, moderate, and low emissions [16]. The 3D-CWT model was calculated using Equation (4) [16].

$$3\mathrm{D}-\mathrm{CWT}=\frac{1}{{{\displaystyle \sum }}_{l=1}^M{\tau }_{ijl}} \times {{\displaystyle \sum }}_{l=1}^M(C_j \times {\tau }_{ijl})$$

(4)

where M denotes the total number of backward air trajectories, τ_ijl is the endpoint number of trajectory l falling in cell (i,j) and processed as described in Section 2.4. C_j is the PM_2.5 mass concentrations at the receptor site. The 3D-PSCF and 3D-CWT models were performed independently for the dry and wet seasons. In addition, the model outputs were multiplied by weighting functions (W) to reduce uncertainty of the 3D-PSCF and 3D-CWT values stemming from the low values of n_ij and τ_ijl. The bootstrapping technique was applied to generate the W values by assuming the W to follow data distribution referred in previous studies [18,24,27,29,48]. The W were then simulated repeatedly 500 times before being multiplied by the 3D-PSCF and 3D-CWT values. Following this approach, the 3D-PSCF and 3D-CWT values were also generated 500 times, and their upper confidence levels were reported. The W values and their distribution are shown in Equation (5).

$$W=\{\begin{array}{c}1.00 (\mathrm{constant} \mathrm{distribution}), 2\mathrm{s} \le \mathrm{N}\\ 0.50\text{–}0.70 (\min -\max \mathrm{distribution}), \mathrm{s} \le \mathrm{N} < 2\mathrm{s}\\ 0.25\text{–}0.42 (\min -\max \mathrm{distribution}), 0.5\mathrm{s} \le \mathrm{N} < \mathrm{s}\\ 0.05\text{–}0.17 (\min -\max \mathrm{distribution}), \mathrm{N} \le 0.5\mathrm{s}\end{array}$$

(5)

where s is the average number of trajectory endpoints per cell, which are 11 and 30 for the dry and rainy seasons, respectively. N is the total number of trajectory endpoints in each grid cell. The 3D-PSCF and 3D-CWT models were computed for 0.25° × 0.25° grid cells over a domain surrounding 85°–125° E and 5°–25° N using the TrajStat tool [27]. Generally, areas showing the high 3D-PSCF or 3D-CWT values can be considered as possible source areas of PM_2.5 [16].

2.6. Data Visualization and Statistical Analysis

The 3D-PSCF and 3D-CWT values were illustrated using ArcMap software version 10.8 (ESRI, USA). In addition, the OriginPro software version 2020 (Origin Lab, USA) was used to plot the PM_2.5 mass concentrations and values of the meteorological parameters, as well as to conduct the statistical analysis, including the statistical hypothesis test, correlation analysis, and partial least squares regression (PLS). The PLS was used to evaluate the relations between PM_2.5 mass concentrations and meteorological parameters since this approach can effectively deal with multicollinearity problem [49], which is the linear relationships among independent variables (i.e., the meteorological parameters) in the regression analysis causing less reliable results of the regression model [50].

3. Results and Discussion

3.1. Mass Concentrations of PM_2.5

In general, the mean of PM_2.5 mass concentrations in the dry and rainy seasons of HCM City were 25.2 and 23.9 μg m^–3, respectively. The relatively lower mass concentrations of PM_2.5 in the rainy season is contributed by the higher rainfall amount in this season (Figure S1 in the Supplementary Information), removing PM_2.5 from the atmosphere via wet deposition [51]. In addition, November exhibited the highest mean PM_2.5 mass concentration (35.1 μg m^–3), followed by December (33.4 μg m^–3) and January (32.7 μg m^–3) (Figure 2a). The lower ambient air temperature of these months (27.6±0.4 °C), compared with the others (28.8 ± 1.0 °C) (Figure S2), can lead to the lower air mixing height in these months (i.e., November, December, and January) [51], enhancing the PM_2.5 accumulation and resulting in the greater PM_2.5 mass concentrations in November, December, and January. Moreover, the highest PM_2.5 mass concentration in November would also be contributed by the temperature inversion, which discourages dispersion of air pollutants and contributes to their higher concentrations [52], since the temperature inversion can more frequently occur in November [53]. The lowest mean concentrations of PM_2.5 were found in March (16.8 μg m^–3), followed by August (18.2 μg m^–3) (Figure 2a). These observations are due to the highest wind speeds in March and August (Figure S3) increasing the PM_2.5 dispersion and leading to a decrease in PM_2.5 mass concentrations.

In terms of diurnal variation, PM_2.5 mass concentrations in the dry and rainy seasons of HCM City had a unimodal distribution, with a significant peak in the morning rush hours (i.e., 7:00–9:00 LT) (Figure 2b,c). After this peak, the PM_2.5 mass concentrations were gradually decreased until late afternoon (e.g., 16:00 LT) and slightly increased around evening rush hours (e.g., 17:00–19:00 LT). Then PM_2.5 reached the lowest mean concentrations at midnight (e.g., 23:00 to 1:00 LT) and increased gradually to peak around the morning rush hours (e.g., 7:00–9:00 LT) (Figure 2b,c). The highest PM_2.5 mass concentrations in the morning rush hours of both the dry and rainy seasons are primarily due to the dense automobiles (e.g., bus, car, and motorbike) during this period [54,55]. Transportation activities have also been mentioned as an important emission source of PM_2.5 in HCM City [12,56]. Furthermore, the lower PM_2.5 mass concentration after the morning rush hours until late afternoon (e.g., 16:00 LT) can be due to a decline in vehicle exhaust after rush hours and an expansion of planetary boundary layer, enhancing the diffusion of air pollutants during daytime [51].

Notably, no obvious peaks in the PM_2.5 mass concentrations were observed in the evening rush hours (e.g., 17:00–18:00 LT) (Figure 2b,c) when the significant vehicle density also occurs. This finding is consistent with previous studies on PM_2.5 pollution in HCM City [5] and suggests that the highest PM_2.5 mass concentrations in the morning rush hours might be affected by not only vehicle emissions but also other factors, such as secondary organic aerosol (SOA). The SOA can be formed from the nucleation or condensation of pre-existing particles [57,58] and oxidation of VOCs [59]. The VOCs can be originated from transportation activities [60] and can react with free radicals, such as OH, O₃, and NO₃, to form low-volatility products, subsequently condensing onto particles to form SOA [61,62]. Moreover, the higher solar radiation in the morning (Figure S4) would promote the formation of radicals (i.e., OH, O₃, and NO₃) through photochemical reactions [63], and thus, increase the SVOCs oxidation to form more SOA.

The PM_2.5 mass concentrations observed in this study were compared with those in previous studies on PM_2.5 pollution in Vietnam and other countries in Southeast Asia. Generally, the mass concentrations of PM_2.5 in HCM City were approximately 3 times lower than those in Ha Noi City, the capital of Vietnam [64,65], and were comparable with those in some other cities of Southeast Asia, such as Bangkok in Thailand [66,67] Bandung in Indonesia [68] and Kuala Lumpur in Malaysia [69]. More information about this comparison is shown in Table S1. In addition, the relationships between several meteorological parameters and the PM_2.5 mass concentrations were evaluated using the PLS approach and correlation analysis. Briefly, wind speed showed the significantly negative relations with the PM_2.5 mass concentrations in the dry and rainy seasons (Figures S5 and S6). The relative humidity and ambient air temperature also had the remarkably negative relations to the PM_2.5 mass concentrations, especially in the rainy season (Figures S5 and S6). More discussion on this issue is presented in Text S1.

3.2. Local Emission Source Areas of PM_2.5

The CBPF plots for identifying emission source areas of PM_2.5 at the local scale (i.e., inside HCM City) are shown in Figure 3. Three same-scale plots were simultaneously considered, including the prediction plot illustrating the predicted areas, the lower and upper uncertainty plots presenting the lower and upper limits of the 95% confidence intervals, respectively. As shown in Figure 3, the CBPF prediction plots of the dry and rainy seasons had the highest probabilities associated with relatively low wind speed (i.e., <3 m/s), suggesting shallow or common ground sources of PM_2.5 (e.g., vehicle and household emissions) [17] as low wind speed tends to disperse PM_2.5 over a small area [20]. This result is supported by previous studies also noting that particulate matters (i.e., PM₁₀ and PM_2.5) in HCM City have been primarily emitted from on-road emission sources, such as exhaustion from motorcycles (e.g., cars, buses, and trucks) [12,70].

Taking in account the CBPF uncertainty, the lower and upper uncertainty plots of the dry season showed similar areas to the prediction plot (Figure 3a), indicating that the PM_2.5 emission sources of the dry season (i.e., shallow or common ground sources) were real and not an artifact due to a lack of PM_2.5 mass concentrations observed in the corresponding wind speeds and wind sectors [17,22]. For the rainy season, the upper uncertainty plot of this season highlighted the PM_2.5 sources at the southeast and northwest associated with high wind speed (i.e., >3 m/s) (Figure 3b). These sources might be stack emissions [17] from the industrial zones located in the southeast and northwest of HCM City, such as the Linh Trung, Binh Chieu, and Tan Tao industrial zones [12]. Industrial activities in these zones (e.g., textile and food manufacturing) [12] emit a large amount of PM_2.5 since their production uses fuel (e.g., diesel oil, coal, and wood), one of the major sources of PM_2.5 [60]. In addition, PM_2.5 originated from these industrial zones is transported to the monitoring site by the seasonal wind (Figure S7). However, the high CBPF probabilities of the southeast and northwest areas were only shown in the upper uncertainty CBPF plot (Figure 3b); therefore, PM_2.5 emission sources in these areas might be uncertain due to a lack of observation data in the corresponding wind sectors and wind speeds [43]. Thus, further studies are essential to confirm the PM_2.5 emission sources in southeast and northwest of HCM City in the rainy season.

3.3. Non-Local Emission Source Areas of PM_2.5

The emission source areas of PM_2.5 at a non-local scale (i.e., outside HCM City) were identified using the 3D-PSCF and 3D-CWT models considering longitudes, latitudes, and heights of trajectory segments (Figure 4). In general, areas showing the higher 3D-PSCF and the 3D-CWT values are possibly regarded as emission source areas of PM_2.5 at the receptor site [71]. In other words, PM_2.5 originated from these areas can be transported to the receptor site by air masses. Regarding the dry season, an increase in PM_2.5 mass concentrations at the receptor site in HCM City are contributed by PM_2.5 emitted from the East Sea (i.e., from shipping emissions) and the southeastern Vietnam (Figure 4), covering some provinces with several industrial zones, such as Dong Nai and Binh Duong provinces [2] (Figure S8). The productions emitting a large of amount of PM_2.5 in such industrial zones of these provinces are textile and food industries, metal, plastic, and paper productions, wood processing, and construction [2]. In addition, results from the trajectory cluster analysis revealed that in the dry season air trajectories arriving at the receptor site in HCM City mostly came from the southeastern Vietnam (i.e., 70% of the total air masses) (Figure 5). Added to this, the mean of PM_2.5 mass concentrations corresponding to each trajectory cluster showed the higher values when air parcels came from the southeastern Vietnam (Table S2). These observations emphasized the contribution of PM_2.5 emission sources in the southeastern Vietnam (e.g., industrial zones in some provinces, such as Binh Duong and Dong Nai) to the elevation of PM_2.5 mass concentrations in HCM City.

For the rainy season, the 3D-PSCF and 3D-CWT results also suggested the East Sea and southeastern Vietnam as possible source areas of PM_2.5 at the receptor site in HCM City (Figure 4). However, these sources might have less effect on the PM_2.5 mass concentrations in HCM City in the rainy season because air parcels arriving from the southeast contributed to approximately 26% of the total air trajectories reaching the monitoring site in the rainy season (Figure 5b). In addition, the 3D-PSCF and 3D-CWT also highlighted the Mekong Delta, located southwest of HCM City (Figure S8), as the potential source area of PM_2.5 in HCM City in the rainy season (Figure 4). The air masses passing by the Mekong Delta accounted for over 70% of the total air trajectories reaching the receptor site in this season (Figure 5b). Additionally, AOD values over the Mekong Delta were high in the rainy season (Figure S9), which could be because of biomass-burning activities (e.g., burning straw and crop residuals) after harvesting the summer–autumn rice crops (i.e., around June–September) [5]. These findings indicate that PM_2.5 originated from the Mekong Delta can transported to HCM City by air masses. However, it should be noted that during the rainy season, transport of PM_2.5 (i.e., middle- or long-range transport) from outside (i.e., the Mekong Delta and southeastern Vietnam) to HCM City is affected by wet deposition of PM_2.5 caused by the high rainfall amount over southern Vietnam, including HCM City and its surrounding regions (Figure S10). Thus, the contributions of non-local emission sources to the PM_2.5 pollution in HCM City during the rainy season can be less important compared with those of the local emission sources (e.g., vehicle emissions).

4. Conclusions

In this study, the local and non-local source areas of PM_2.5 in HCM City, Vietnam, were identified using the CBPF and 3D-PSCF/3D-CWT models. Uncertainties of these models were evaluated using the upper/lower 95th confidence intervals of the CBPF prediction surfaces and the 95% upper confidence levels, computed from the dataset of 3D-PSCF/3D-CWT generated using the bootstrap technique. Additionally, the mass concentrations of PM_2.5 and their relationships with meteorological parameters (e.g., air temperature, air pressure, and wind speed) were evaluated. The results showed that the PM_2.5 mass concentrations in the dry season were higher than those in the rainy season because of the deposition of PM_2.5 caused by the high rainfall amount during rainy season. For the diurnal variation, the PM_2.5 mass concentrations in HCM City followed a unimodal distribution, with a significant peak in the morning rush hours (i.e., 7:00–9:00 LT) due to a contribution of SOA, which can be formed in the morning of HCM City.

Regarding the local source areas, the CBPF revealed that PM_2.5 in HCM City was mainly emitted from shallow or common ground sources (e.g., vehicle and household emissions) throughout the year. Additionally, PM_2.5 emitted in the northwest and southeast of HCM City might be transported to the receptor site in the rainy season. However, these emission source areas have uncertainty due to a lack of data, thus, further studies are essential to confirm our findings.

In terms of the non-local source areas, PM_2.5 originated in the East Sea (i.e., from shipping emissions) and southeastern Vietnam (e.g., from several industrial zones in the region) can transported to HCM City, especially in the dry season. Additionally, in the rainy season, the Mekong Delta, located southwest of HCM City, might be the emission source area of PM_2.5 in this city. However, the contributions of non-local sources to the PM_2.5 pollution in HCM City might be less important in the rainy season because PM_2.5 can be removed from the air during its transport as a result of wet deposition under the effect of the large rainfall amount and great rain event frequency during the rainy season. Results from this study can add more information on the identification of emission source areas of air pollutants as well as support decision-making related to the reduction in PM_2.5 pollution in Vietnam.