The PM_1.0 samples were simultaneously collected with low-volume air samplers at the roadside area in Daejeon city, central Korea. The sampling site (36°21′45″ N, 127°20′42″ E) is surrounded by densely populated residential and commercial areas. As the study area is located on a roadside with heavy traffic volumes (approximately 8000 vehicles/h) and regular congestion, it represents an area with considerable traffic-related pollution in Daejeon city.

A total of 67 PM_1.0 samples on a 24-h basis were collected in the fall and winter (September 30–December 15) of 2019. A vacuum air sampler (URG, 3000C model) was adopted, and a cyclone of 1 μm cut-point (URG-2000-30EHB) was used to collect PM_1.0 samples on the polycarbonate filter (47 mm, 0.4-μm pore size, Nuclepore). The airflows for the sampler were adjusted to a rate of 16.7 L min⁻¹ at the beginning of sampling. Each filter was weighed in a controlled atmosphere (20 °C and 50% relative humidity) for 24 h before and after exposure to air. These filters were weighed three times in a pre-calibrated and tared microbalance with a readability of 1 μg (XPE26, Mettler-Toledo Ltd.) and controlled electrostatic charges in order to restrain any charge forces from a filter in the balance. The hourly meteorological data (e.g., rainfall, temperature, relative humidity, wind direction, and horizontal wind velocity) were obtained from the Korea Meteorological Administration (KMA).

The PM-bound concentrations of black carbon (BC), ions, and trace elements were determined sequentially. To determine the BC concentration, a multi-wavelength absorption BC instrument (MABI, Australian Nuclear Science and Technology Organization) that functions at seven wavelengths (405, 465, 525, 639, 870, 940, and 1050 nm) was used [35, 36]. The results for the different wavelengths were used to differentiate the different BC size fractions originating from sources such as wood smoke or vehicle exhaust. The BC concentration was evaluated from its absorption value at 639 nm, which the predominant absorber wavelength. The mass concentration of BC converted using the specific mass attenuation given by the manufacturer. When the standard filter was repeatedly measured, the relative standard deviation of the intensities was < 5%.

After measuring the gravimetric mass and multi-wavelength absorption, three-quarters of the PM filter were digested using a microwave digestion system (ETHOS EASY, Milestone). Then, the sample was loaded in a PTFE digestion vessel with 8 mL of HNO₃ (65%, Optima Grade, Thermo Fisher Scientific), and the temperature was increased to 200 °C within 5 min (1200 W) and then maintained for another 20 min. After digestion, the vessels were cooled to room temperature for 1 h, and the remaining solutions were transferred to a centrifuge tube and diluted with deionized water. The centrifuged sample aliquot was introduced into the ICP-MS system (ICAP RQ, Thermo Fisher Scientific) using the kinetic energy discrimination (KED) mode (collision cell with He gas and kinetic energy discrimination). Although the ICP-MS technique is accurate for determining multiple elements in PM samples, interference from molecular overlap or polyatomic ions should also be considered. Recent developments in the ICP-MS technique have removed interfering ions by applying gas reaction/collision in the KED mode [37–40]. Sixteen elements were analyzed: Al, As, Ba, Cd, Cr, Cu, Fe, K, Mg, Mn, Ni, Pb, Se, Ti, V, and Zn. Five working standards were prepared from stock solutions (TraceCERT®, Sigma-Aldrich) in 2% HNO₃. The analytical conditions for ICP-MS are shown in Table S1. The standard calibration curves for the ICP-MS analysis were generally in line with those obtained using the five working standards (r-value > 0.999). Additionally, the final concentrations were corrected using combined reagent and filter blanks. Spiked and recovered ¹¹⁵In concentrations were set as an internal standard for each sample. Recovery of the pretreatment steps and ICP-MS analysis was evaluated and adjusted for the measurement data. The recovery of spiked ¹¹⁵In ranged from 82–105%, with a mean of 93%. For analysis of ionic components (e.g., SO₄²⁻, NO₃⁻, and NH₄⁺) in the filter samples, ion chromatograph (IC, Thermo Scientific Dionex ICS-2100 system) was employed. IC consists of a separation column, guard column, and suppressor using sulfuric acid. The ionic components were extracted from a quarter of the filter sample with ultrapure water using an ultrasonic instrument (Branson 8210, USA) for 30 min.

For quality assurance, the NIST standard reference material was used (SRM: NIST, National Institute of Standards and Technology, USA; SRM 2783, air particulate on filter media) for the ICP-MS measurements. The relative errors (against SRM values) of all target elements were < 20%. After repetitive analysis, the relative standard deviations were determined to be < 5% for all the target elements. For the IC analysis, the pooled standard deviation for the peak area of the standard solution was approximately 3%, and the recovery of spiked species was within ± 5%.

The conditional probability function (CPF) was computed using wind direction and wind speed to help understand the characteristics of PM_1.0 and the species concentrations in the study area. A CPF can be used to specifically estimate possible local sources using wind direction data [12, 41–43]. The CPF (probability value from 0 to 1) is defined by the following equation (1):

where m_Δθ indicates the wind frequency blowing from the direction of Δθ on different days, with a concentration higher than the threshold criterion; whereas, n_Δθ indicates the frequency of wind blowing from the direction of Δθ in the overall data. As the current domestic meteorological data can represent wind from 16 directions, Δθ was set as 22.5 in this study. The same daily concentration data were assigned to each hour of a given day to correspond to the hourly wind data. Calm wind (<1.0 m·s⁻¹) was excluded from the analysis; the threshold criterion of the upper 25^th percentile value was used.

The details regarding PMF theory and its applications can be found in several studies [12, 17, 25–28, 30, 44]. In the present study, the EPA PMF (ver. 5.0), which is the most recent version using the multilinear engine 2 (ME-2), was used for source apportionment following the EPA PMF User Guide [45]. The selection of input data is critical for PMF-based source apportionment, and both the concentrations of the chemical components and their error estimates should be provided as PMF input data. The expanded uncertainty in 99 % confidence interval of each analytical value is evaluated from the major uncertainty factors for the analytical process such as reproducibility of the measurements, recovery ratio, and stability of efficiency calibration. Basic equation of PMF model is defined by the following equation (2):

The method has been developed to obtain the unknown matrix, G and F by the iterative treatment of a least square method by the following equation (3):

Here, X(m×n) is the data matrix consisting of the m chemical components analyzed in n samples. G(n×p) is the source contribution to each sample. F(p×m) is the matrix of source profile. E presents the residual matrix of calculation.

In this study, both terms were determined by the following methods: 1) the measured data were used directly with their error estimates at the expanded uncertainty; and 2) data below the detection limit were replaced with a half value of the detection limit for each element with 5/6 of their detection limits as error estimates [42, 44, 45]. The PMF run with the lowest Q representing the global minima was selected for obtaining an optimal solution for the number of factors. The initial solutions were iterated with random seed start and robust modes to minimize random errors and rotational ambiguities. In addition, intra-run residual analysis, bootstrapping (BS), displacement (DISP), and BS–DISP were applied to evaluate the optimal solution [27].

Table 1 summarizes the PM_1.0 and associated species concentration data measured near the roadside in sampling site. According to the summary provided in Table 1, the mean PM_1.0 concentration was 19.9 ± 5.50 μg m⁻³, with a range of 6.86–34.7 μg m⁻³. During the sampling period in 2019, there was no significant trend in the PM_1.0 concentrations for distinctive pollution events. The time series of temperature, rainfall, and wind speed frequency based on a 1-h resolution of the meteorological data during the study period are presented in Fig. S1. As shown in Fig. S1, rainfall was 227 mm (only 5 days), the mean temperature was 9.6 °C, and the frequency of calm wind (defined as < 1.0 m s⁻¹) was 21.6%. Owing to the cumulative effect of stagnant movements of the atmosphere and low rainfall, particulate concentrations in the air increased in the fall and winter in Korea.

The distribution characteristics of PM_1.0 concentrations (N = 67) observed at the roadside area are presented in Fig. 2; 70.1% of the PM_1.0 concentrations were between 15–25 μg m⁻³ and 92.5% were < 30 μg m⁻³. The results for the relatively steady PM1.0 concentration revealed the absence of a transboundary invasion event with significant influence in the sampling period.

According to previous studies, the PM_1.0 concentration distribution trend is consistent with that of PM_2.5. The mass concentration ratios of PM_1.0 and PM_2.5 were 0.64 [46] and 0.90 [14] in an industrial area, 0.61–0.91 in an urban area [6, 16], 0.80 in a rural area [22], 0.82 at the roadside in a commercial and industrial area [17], and 0.84–0.92 in an underground parking garage [18]. These ratios suggest that submicron particles < 1.0 μm are the major components of PM_2.5, regardless of the characteristics of the area. Currently, there are no legislative criteria to regulate PM_1.0 concentrations in Korea; however, annual PM_2.5 standard was set to < 15 μg m⁻³. It is noteworthy that the field campaign was conducted over three months, and caution is needed in comparison with annual standard. However, even upon comparison with PM_2.5 standard, when considering the PM_1.0/PM_2.5 ratio, it is clear that the PM_1.0 level in the study area was significantly high and needs to be managed.

As a simple means to evaluate the pollution levels of the study area, the PM_1.0 concentration data in this study can be compared to those from comparable studies (see Table 2). The mean PM_1.0 concentrations measured in this study were comparable to those in rural [22] and industrial areas in Korea [46] and the UK [11], residential areas with heavy traffic in Italy [26], urban areas with heavy traffic in Vietnam [16], and industrial areas in Poland [15]. However, the PM_1.0 values in the present study were considerably lower than those of other Asian cities, such as Wuhan (22–128 μg m⁻³: [14]), Harbin (58–429 μg m⁻³: [18]; 11.2–125.3 μg m⁻³: [6]), Beijing (6.40–216.32 μg m⁻³: [17]), and Delhi (31.5–568.8 μg m⁻³: [12]).

Z-statistics were applied to the PM_1.0 datasets to assess the significance of quantitative differences in the mean concentrations with variation in days of the week. Pairwise comparison between PM_1.0 concentrations (through Z-statistics tests) indicated that there was no statistically significant difference between the mean PM_1.0 concentrations on different days of the week.

As shown in Fig. 3, wind direction was not a distinct contributing factor when determining the pollution sources according to the CPF calculations for PM_1.0, BC, SO₄²⁻, and NO₃⁻. Similar probabilities reflected widespread pollution in the study area. However, the comparatively high values in the southeastern region of the study area may reflect the impact of the large residential area combined with a relatively higher traffic volume and heating oil combustion during winter.

Table 1 shows that the BC, SO₄²⁻, and NO₃⁻ concentrations in PM_1.0 were substantially higher than those of the metal species. BC, SO₄²⁻,and NO₃⁻, which are predominant in fine particles with NH₄⁺ in air particulates, are produced as a result of the reaction and transformation between the pollutants emitted from various combustion processes [24]. Thus, it is generally difficult to accurately determine their unique sources. The mean concentrations ± standard deviation (and ranges) of BC, NO₃⁻, and SO₄²⁻ in PM_1.0 were 4.89 ± 4.21 (1.26–8.23), 4.21 ± 1.39 (1.76–9.78), and 4.30 ± 1.10 (1.16–7.03) μg m⁻³, respectively. The BC, NO₃⁻, and SO₄²⁻ concentrations were added and compared with the PM_1.0 mass concentrations; each fraction accounted for approximately 24.6%, 21.2%, and 21.6% of the total content, respectively. Their total amount accounted for a mean of 13.1 μg m⁻³ (65.8%) in PM_1.0 in this area during the study period.

It has been well recognized that BC is a marker element of air particulates in combustion processes (e.g., vehicle exhaust, biomass burning, and fossil fuel combustion) regardless of the particle size fraction. BC is likely an important causal agent intrinsically and many carbonaceous species co-emitted from diesel and other vehicles can be adsorbed onto the surface of BC [20]. BC, which is an important contributor to global warming, is also considered an effective additional indicator for estimating human health effects [47]. The light absorption by BC in the filter samples was measured at seven different wavelengths from 405–1050 nm. As the characteristics (e.g., shape and absorption sensitivity) of BC vary with different emission sources, the BC data from each wavelength can be used to distinguish BC source contributions from vehicles and smoke events such as biomass burning [35, 48]). Thus, more information on possible sources of BC could be derived from the difference between the BC values at 405 nm and 1050 nm [49]. The variations in BC concentrations are depicted in Fig. S2, with the differences ranging between 405–1050 nm. As shown in Fig. S2, there was no significant trend in BC concentrations during distinctive pollution events. In addition, there was a distinct spike in the daily BC (405–1050 nm) value. These results imply that the major fraction of BC concentrations originated from the vehicular exhaust in the study area. Further, the CPF result for BC (Fig. 3b) supports the widely distributed traffic volume within the study area.

As a certain portion of SO₂ and NO₂ diffuse to Korea from the Chinese continent with westerly winds, the SO₄²⁻ and NO₃⁻ in air particulates of the study area, which is located in the heartland of the Korean Peninsula, could increase during the fall and winter due to northwesterly winds. Hence, it is surmised that NO₃⁻ and SO₄²⁻ from various regional sources were transported into the study area, which contributed to the formation of secondary aerosols. The CPF results of similarly distributed probabilities with wind direction for NO₃⁻ and SO₄²⁻ (Fig. 3) support this interpretation.

To evaluate the characteristics of the secondary ionic species, the measured NH₄⁺ concentrations were compared with the corresponding NH₄⁺ concentrations calculated based on the stoichiometric ratios of the major compounds, such as ammonium sulfate ((NH₄)₂SO₄), ammonium bisulfate (NH₄HSO₄), and ammonium nitrate (NH₄NO₃). This was carried out under the assumption that NO₃⁻ exists in the form of NH₄NO₃ and that SO₄²⁻ is in the form of either (NH₄)₂SO₄ or NH₄HSO₄ [26, 50]. Sulfate aerosols containing mainly NH₄HSO₄ can be considered as moderately aged aerosols, whereas sulfate aerosols containing mainly (NH₄)₂SO₄ can be viewed as highly aged aerosols [51]. As shown in Fig. S3, the ratios between the calculated NH₄⁺ (assuming (NH₄)₂SO₄ or NH₄HSO₄) and measured NH₄⁺ were 1.2 and 0.84, respectively. This indicates that SO₄²⁻ in PM_1.0 at the study area was well mixed with freshly emitted gas and aged particles. Furthermore, from the correlation analysis between BC and SO₄²⁻ of PM_1.0 measured in a rural area of Korea, Lim et al. [22] stated that BC is likely to mix with SO₄²⁻ sufficiently under the influence of substantial anthropogenic sources. As SO₂ emissions have dramatically decreased in Korea, it could be surmised that the considerable SO₄²⁻ concentrations were affected by long-range transported SO₂. Thus, the compound forms of SO₄²⁻ and the higher correlation (r = 0.81) between BC and SO₄²⁻ in PM_1.0 at the study area suggested that the SO₄⁻² concentrations were influenced by pollutants emitted in the study area and trans-boundary pollutants from outside the study area.

A higher correlation coefficient suggests a strong relationship between the pollution sources. As shown in Fig. 4, strong relationships between PM_1.0 and the major species (e.g., BC, SO₄²⁻, NO₃⁻, and NH₄⁺) were observed with correlation coefficients between 0.52–0.96. In particular, the correlation coefficients between PM_1.0 and BC, NO₃⁻, and SO₄²⁻ were 0.96, 0.82, and 0.84, respectively. Hence, it may be concluded that the PM_1.0 concentrations in the study area were strongly ruled by secondary ionic species and BC concentrations. According to a statistical summary of the data in Table 1, the concentrations of major species (e.g., BC, SO₄²⁻,NO₃⁻, and NH₄⁺) were apportioned by 79.7 ± 6.3% (66.1–98.2%) of the PM_1.0 concentration.

However, the concentrations of the 16 analyzed elements (Al, As, Ba, Cd, Cr, Cu, Fe, K, Mg, Mn, Ni, Pb, Se, Ti, V, and Zn) were approximately 5.5 ± 3.8% (1.6–26.5%), according to the results of the quantitative elemental analysis for airborne PM_1.0 samples by ICP-MS. Although the elements constitute only approximately 5% of the total particulate mass, they can play a critical role in the identification of diverse source processes. As large amounts of C, ionic components, and major crustal elements are commonly bound to each other in most sources, each source type can be specifically assigned to its tracer [3]. The concentrations of metals associated with crustal sources (such as Al, Fe, and K) were much higher than those of any other toxic metals (As, Cd, Cr, Cu, Ni, Pb, Se, and Zn). The mean concentrations (± standard deviation) of As, Cd, Cr, Cu, Ni, Pb, Se, and Zn in PM_1.0 were averaged as 1.91 (± 0.56), 2.51 (± 0.74), 19.0 (± 6.22), 18.8 (± 16.6), 14.8 (± 13.1), 21.6 (± 9.05), 1.62 (± 0.49), and 41.1 (± 117) ng m⁻³, respectively. Based on a simple comparison between the metal concentrations and their magnitudes, the datasets can be arbitrarily grouped into three different categories: 1) < 10¹ ng m⁻³: As, Cd, Se, V, and Ti; 2) 10¹–10² ng m⁻³: Ba, Cr, Cu, Mn, Ni, Pb, and Zn; and 3) > 10² ng m⁻³: Al, Fe, and K. These distribution characteristics of quantitatively measured elemental concentrations are comparable with those observed in residential areas with heavy traffic [26].

Although various constituents of particles (e.g., C, ionic components, and major crustal elements) are produced from and are related to diverse source activities, these sources can be distinctly classified by the relative composition of specific trace elements [3]. In this study, the PMF analysis from four to ten factors along with the input of data for PM_1.0 and 20 species were tested. Subsequently, the possible sources of PM_1.0 were classified into six categories that provided physically meaningful source profiles and contributions. The number of factors was determined based on the Q-value. The initial solutions were iterated with random seed start and robust modes to minimize random errors and rotational ambiguities. The residuals could be examined to evaluate the number of factors that corresponded to the optimized Q-value. If the scaled residuals followed a symmetrical distribution with a range of −3–3, then the number of factors could be considered appropriate. The reliability of PMF modeling can be estimated by correlation analysis between the measured and predicted mass concentrations. According to the modeling results optimized in this study, the determination coefficient between the observed and predicted PM_1.0 concentrations was 0.95, and six sources explained 99% of the measured PM_1.0 concentration. The source profiles are shown in Fig. 5, where the bars and circles represent the absolute number of factors generated in a source and the percentage of species for each source, respectively. To estimate possible regions of local source, the CPF plots for the source contributions from PMF were presented in Fig. S4.

The relative contributions of these categories (with marker species) decreased in the following order: secondary aerosols (SO₄²⁻, NO₃⁻, and NH₄⁺), vehicle exhaust (BC, V, Cr, and Ti) [52,53], re-suspended soil-road dust (Ti, Al, Mg, K, Fe, Ba, and Zn) [27, 41, 42], fossil fuel combustion (Se, Cd, As, Ni, and Cr) [15, 54,55], biomass burning (K and Pb) [56–59], and industrial activities (Fe, Ni, Cu, and Al) [60–62].

The first source was estimated to be a secondary pollutant, as the resolved factors accounted for 56%, 42%, and 43% of the total NO₃⁻, SO₄²⁻, and NH₄⁺ concentrations, respectively. As secondary aerosols are produced as a result of the reaction and transformation between pollutants emitted in the air, it is generally difficult to allocate their sources accurately. Thus, it was surmised that NO₃⁻, SO₄²⁻, and NH₄⁺ from various regional sources were transported into the study area and contributed to secondary aerosol formation. Secondary pollutants in this study area showed the highest contribution among all sources, with a mean of 10.7 μg m⁻³ (54.1%). The quantitative NO₃⁻, SO₄²⁻, and NH₄⁺ concentrations in the first factor agreed with the previous results of source apportionment for the PM₁₀ concentrations in this study area [42]. This finding reveals that secondary pollutants (e.g., NO₃⁻, SO₄²⁻, and NH₄⁺) were major contributors to air PM regardless of the particle size. The second source was estimated to be vehicle exhaust because it is dominated by elements such as BC (47%), V (38%), Cr (33%), and Ti (32%). Carbonaceous species, such as EC, OC, and BC, have been widely used as representative markers for vehicular sources [52,53]. Vehicle exhaust constituted approximately 21% of PM_1.0 with a mean of 4.2 μg m⁻³. The relatively higher contribution of vehicle exhaust to PM_1.0 in the study area reflected the on-site characteristics of the roadside area. Furthermore, because the study was conducted in the fall and winter, the contribution of vehicle exhaust emissions could be explained by the stimulated mechanism of transforming vehicle exhaust into PM at low temperatures [25]. The third source was typical soil dust. This was elucidated by the high percentages of major crustal elements, such as Ti (43%), Al (41%), Mg (39%), K (26%), and Fe (27%). Due to the relatively higher loadings of Ba (29%) and Zn (24%), this source was also estimated to be re-suspended road dust by mobile transportation from paved or non-paved roads. Further, Ba is added to the lubricating oil of diesel vehicles to prevent smoke and engine abrasion [41 and references therein). Zn is a well-known road dust marker element which is generated by tires and brake wear of mobile vehicles [27]. The fourth source was assumed to be fossil fuel combustion, such as coal and heavy oil, reflected by the high loadings of Se (51%), Cd (48%), As (40%), Ni (29%), and Cr (28%). Cd is known to occur at high temperatures during fuel combustion [54]. Ni and V are widely used as pair markers for the combustion of heating fuel, whereas Se is a representative marker species for oil-fired power plants and coal combustion [55]. The fifth source was surmised to be biomass burning due to the high percentages of K (47%) and Pb (42%); a large proportion of K could be explained by biomass burning [56–58]. The relatively high percentage of Pb in the fifth source profile could be explained that the emission of particulate matter from illegal incineration in the plain ground increased with biomass burning activities in the fall season [59]. The sixth source was assigned to industrial activities such as metal smelting [60, 61], refuse incineration [62], and mechanical abrasion [61], reflected by the high loading of Fe (38%), Ni (34%), Cu (28%), and Al (27%).

Estimated mean source contributions (SCE ± standard deviation; relative SCE) for the following are shown in Fig. 6: secondary aerosols (10.7 ± 3.1 μg m⁻³; 54.1%), vehicle exhaust (4.2 ± 1.4 μg m⁻³; 21.2%), re-suspended soil-road dust (1.9 ± 0.8 μg m⁻³; 9.6%), fossil fuel combustion (1.8 ± 0.7 μg m⁻³; 9.1%), biomass burning (0.7 ± 0.3 μg m⁻³; 3.3%), and industrial activities (0.5 ± 0.3 μg m⁻³; 2.6%). As shown Fig. S5, there was no significant variation in the estimated source contribution by distinctive pollution events. The relatively increasing trend in contributions of fossil fuel combustion source in winter seasons was reflected increasing consumption of the heating fuel.