Introduction

Suspended particulate matter (PM) is complex with respect to its mass concentration, particle size distributions, chemical composition and its sources. Anthropogenic activities including industrial effluent, combustion emissions and physical processes such as wind-blown dust and traffic emissions are typically factors dominating the spatial and temporal PM variations (DeGaetano and Doherty 2004). With respect to small-scale spatial variations in urban areas, the largest variations occur in the ultrafine (< 0.1 μm) and the coarse mode (PM2.5–10, re-suspended dust) (Lagudu et al. 2011; Kumar et al. 2012; Wang et al. 2012). Secondary aerosols that contribute to the accumulation mode (0.1–1 μm) show more homogenous spatial distributions (Moon 2001). Despite the limited spatial variations of fine PM within the city, understanding and controlling air pollution in highly polluted areas require multiple-site measurements. During pollution events, it is important to assess the origin of the higher PM concentrations to be able to employ measures to improve air quality (Petit et al. 2017; Waked et al. 2018). Despite the local origin of some PM, the observed PM is also influenced by medium and long-range transport processes. Therefore, for mitigation strategies to be effective, they must be scaled at regional, national and even continental levels (Guerreiro et al. 2014; Guan et al. 2017).

The Moravian-Silesian region, located in the eastern part of the Czech Republic, represents an EU air pollution hot spot. Epidemiological studies from Ostrava, the major city of the region, displayed negative health impacts for the local population, particularly for children, caused by deteriorated air quality (Dostál et al. 2013; Šram et al. 2013; Topinka et al. 2015). The area has been historically burdened with extensive industrial activity in the Upper Silesian basin. The key factors influencing the air quality in the Moravian-Silesian region are (i) exploitation and use of the high quality black coal, (ii) extensive heavy industry, (iii) high density of built-up areas with local heating by solid fuels, (iv) dense transport infrastructure on both sides of the Czech-Polish border, and (v) orography influencing the meteorological conditions with prevailing wind direction from south-west and north-east as well as long-lasting periods of stable conditions (from autumn to spring) with subsequent worsened dispersion conditions. Currently, approximately 760 facilities, large and medium stationary sources of air pollution, are registered in the region (CHMI 2018). Only several dozens of facilities—power plants and enterprise energy generation—and metallurgical production facilities have a substantial effect on overall emissions (TSP, SO2 and NOx); nevertheless, fugitive TSP emissions produced e.g. from landfills, handling of bulk materials and halls with dusty operations are not taken into consideration in the emission inventories. Therefore, the planning and implementation of effective abatement strategies to improve air quality in Ostrava are challenging due to the presence of miscellaneous sources, some of them situated near residential areas.

In order to analyse the spatial and temporal changes in airborne PM regarding to the effective abatement strategy implementation, the application of receptor model is highly effective. Receptor models analyse data matrices of aerosol chemical and physical characteristics of samples collected at a given locality, the receptor, to apportion their sources (Polissar et al. 2001). In the literature, a wide range of statistical models and modelling approaches is available e.g. CMB, PCA, PMF, UNMIX, ME and COPREM, and one of the main differences between the models is the degree of knowledge required about the pollution sources prior to the application of receptor models (Viana et al. 2008). PMF is a powerful tool for source resolution using highly time resolved aerosol composition data and has been used successfully in identifying the sources of airborne particles in many studies (e.g.; Ancelet et al. 2012, 2014; Pancras et al. 2013; Moreno et al. 2013; Hovorka et al. 2015; Hopke 2016). Particulate matter toxicity depends on its chemical and physical characteristics: composition, size, morphology or solubility (Heal et al. 2012). Thus, source apportionment studies focus not only on chemical composition but also on particle number size distribution data (Kim et al. 2004; Zhou et al. 2004, 2005a; Ogulei et al. 2007a, b; Kasumba et al. 2009; Pey et al. 2009; Harrison et al. 2011; Cusack et al. 2013; Beddows et al. 2015; Sowlat et al. 2016; Masiol et al. 2016; Squizzato et al. 2019). By investigating particles in various size ranges, it is possible to more clearly identify and apportion contributions from those sources that contributed more to the particle number than to the particle mass (Vu et al. 2015). The two approaches are often complementary (Zhou et al. 2005b; Beddows et al. 2015).

The aim of this study was to access the spatial and temporal variability of winter air pollution sources including nanoparticles in the districts of Ostrava-Radvanice and Ostrava-Plesná, examples of industry affected urban and suburban residential receptor sites. The source apportionment was performed using highly time-resolved elemental composition of PM0.15–1.15 and particle number size distributions (PNSD, 14–10 μm) measured from February 9 to March 3, 2014. The factors resolved from mass chemical compositions and the PNSDs by PMF were compared and associations were determined. The study complements previous source apportionment studies performed at the sites (Pokorná et al. 2015 and 2016; Leoni et al. 2018; Kozáková et al. 2019). The results from the parallel highly time-resolved chemical and PNSD measurements at these two different site types provided further insights into the spatial and temporal variability of the source contributions to PM at this site.

Methods

Experimental

A twin intensive sampling campaign was performed from February 6 to March 6, 2014 in Ostrava-Radvanice and Ostrava-Plesná as urban and suburban residential receptor sites, respectively (Fig. SI1). The distance between the two sites was approximately 16 km.

Five-minute integrated PNSD was measured with a Scanning Mobility Particle Sizer (14–730 nm, SMPS-3936 L25, TSI Inc. and 14.6–710.5 SMPS-3936 L75, TSI Inc.) and an Aerodynamic Particle Sizer (0.542–20 μm, APS-3321, TSI Inc.).

A standard reference sampler (LVS-3, Sven Leckel Ingenieurbüro GmbH, Berlin, Germany; flow rate of 2.3 m3 h−1) with quartz fibre filters (47 mm; Whatman QMA, GE Healthcare, Maidstone, UK) was used to measure 24-h PM1 concentrations in Plesná. A high-volume sampler (DHA-80, Digitel, Switzerland, flow rate 30 m3 h−1) with quartz fibre filters (150 mm diameter; Whatman QMA, GE Healthcare, Maidstone, UK) was used to determine the 24-h concentrations of PM1 in Radvanice. The samples were analysed for both cations and anions using a Dionex ICS-5000 system (Dionex, Sunnyvale, CA, USA) and for elemental carbon (EC) and organic carbon (OC) using a semi-continuous analyser (Sunset Laboratory Inc., Tigard, OR, USA) operated in an offline mode and using Eusaar2 protocol. PM1 samples from Plesná were also analysed for monosaccharide anhydrides (levoglucosan, mannosan and galactosan) using a Dionex ICS-5000+ system (Dionex, Sunnyvale, CA, USA). Detailed descriptions of the methods and limits of detection (LODs) along with quality assurance and control for the used methods were reported by Kozáková et al. (2019).

Size segregated and time resolved PM were collected with a Davis Rotating-drum Uniform-size-cut Monitor – 3DRUM (DELTA Group UC-Davis), from February 19 to March 3 and were used to provide 1-h PM elemental compositions. The particles were collected on Mylar substrates lightly greased with Apiezon™. The 3DRUM collects particles in 3 size ranges from 150 to 10 μm. The two size range samples from 0.15 to 1.15 μm were analysed for 24 elements using synchrotron X-ray fluorescence (S-XRF) by the Air Quality Research Center, University of California (Leoni et al. 2018). For the source model, data from the two size ranges of 3DRUM were merged to obtain PM0.15–1.15 to replicate the size range of the PNSD.

In Radvanice, 5-min meteorological data—wind speed (WS) and wind direction (WD), temperature (T), relative humidity (RH) and precipitation (P)—were recorded directly at the site during the whole sampling campaign. In Plesná, the meteorological data were available from the nearest representative automatic monitoring stations managed by the Czech Hydrometeorological Institute (Kozáková et al. 2019).

Data analysis

Descriptive statistics

Measured data (online measurement and chemical analysis of filter samples) and modelling results were statistically analysed using descriptive statistics, Pearson correlation and regression analysis to discover underlying patterns and trends regarding the monitoring campaign.

PMF Modelling

PMF (USEPA version 5.0) was applied to PM0.15–1.15 chemical composition and PNSD data (15–10 μm). A 1-h integration time PM0.15–1.15 chemical composition as a sum of two size fractions was used for PMF input along with missing mass (MM) for missing data for OC/EC (Pokorná et al. 2013). Hourly aerosol mass concentrations assigned as total variable were calculated from 5-min integrates of the number size distributions with a particle density of 1.5 g cm−3 (Shen et al. 2002) recorded by an APS and a SMPS. Due to the importance of the local meteorology given by the orography of the region (Pokorná et al. 2015; Černikovský et al. 2016; Leoni et al. 2018; Kozáková et al. 2019), the matrices of the chemical composition from the two sites were modelled separately by PMF.

For the PNSD PMF, the time resolutions from 5 min to 1 h were examined and it was determined that a 15-min resolution was optimal as the best compromise to maintain a high time resolution but avoid unwanted noise. Additionally, the input data were handled by merging three consecutive bins. Since the size segregation of SMPS is based on particle electrical mobility while the APS and 3DRUM impactor use particle aerodynamic properties, the mobility diameter was converted into aerodynamic. The uncertainties were calculated according to Vu et al. (2015). The total variable was calculated summing all the bins. The detailed information was reported in Leoni et al. (2018). PMF was also applied to PNSD from Plesná (modelled with and without chemical composition data) to extract those factors governing particle number down to the nanometric scale.

The model was run several times using (i) different number of factors, (ii) different extra modelling uncertainties and (iii) for PNSD data different uncertainty input matrices, different C3 value (Vu et al. 2015) in order to obtain the highest S/N ratio and the Qtrue closest to Qtheoretical, to determine the most physically meaningful result and the best diagnostics.

The polar plots and daily patterns were obtained using the Openair R Package (Carslaw and Ropinks 2012).

Schematic picture synthetizing all methods (measurement techniques, chemical analysis and data analysis) utilized in this study could be found in the Supplementary Information (Fig. SI2).

Results and discussion

Campaign overview

During the sampling campaign, the average PM10 concentrations calculated from the merged SMPS and APS number size distributions in Plesná and Radvanice were 29.2 ± 4.5 μg m−3 and 61.5 ± 8.7 μg m−3, respectively. The PM0.15–1.15 comprised 90% and 48% of PM10 in Plesná and Radvanice, respectively. These PM ratios indicate important differences in the PM size distribution and likely major sources at these locations. The elemental concentrations of almost all elements were lower in Plesná relative to Radvanice (Tables SI1 – 2). On average, the higher arsenic concentrations in Plesná point to the importance of coal combustion there (Tab. SI1). The average particle number concentrations were distributed in four particle modes: nanoparticles (15–30 nm), Aitken mode particles (30–80 nm), accumulation mode particles (80–1 μm) and coarse particles (1–10 μm) are presented in Table 1. The campaign was characterized by prevailing southwesterly and northeasterly winds at both sites due to the topography of the Ostrava region. Wind speed averaged 1.7 ± 0.6 m s−1 in Plesná and 1.2 ± 0.5 m s−1 in Radvanice.

Table 1 Average particle number concentrations (# cm−3) of the sampling campaign at Plesná and Radvanice site
Full size table

Factors resolved with chemical composition

To estimate the optimal number of sources, the PMF model was run several times (units of tens for different model settings) and 3 to 8 factors were tested for each data set. The Q values, the resulting source profiles and the scaled residuals were examined. For each data set, the optimum number of factors was chosen based on an adequate fit of the model to the data as shown by the scaled residual histograms and physically interpretable results. The most stable solutions were found for 6-factor and 4-factor solutions for Plesná and Radvanice, respectively. Extra modelling uncertainties of 13.5% for Plesná and 17.9% for Radvanice were included to encompass errors not considered in the uncertainty assessment. All runs converged, the scale residuals were normally distributed, and no unmapped factors were detected with the bootstrap error estimation. No swaps were observed with the displacement error analysis, indicating that there was limited rotational ambiguity (Table SI3).

The six factors in Plesná were assigned as coal combustion, mixed factor secondary inorganic aerosol (SIA) and biomass burning, residential heating, biomass burning, re-suspended dust and raw iron production. The 4 factors in Radvanice were ascribed to mixed factor SIA and coal combustion, residential heating, raw iron production and sinter/steel production. The factor elemental profiles and the time-series plots of the estimated contributions from each factor to the PM0.15–1.15 mass are shown in Figs. 1 and 3.

Fig. 1
figure 1

Plesná factor profiles (concentration of the species, bars; percentage of the species, markers; DISP Average, box; DISP Max and DISP Min, whiskers) and time series resolved for PM0.15–1.15 by PMF

Full size image

Plesná

Factor 1. The first factor, local residential coal combustion, was associated with high concentrations of As, MM, Pb, Cu and Br (Rogula-Kozłowska et al. 2012, 2013a, b; Rogula-Kozłowska and Klejnowski 2013). This factor dominated during the monitoring period with an average contribution of 49% to PM0.15–1.15. The high factor contributions with WS up to 2 m s−1 point to local sources primarily situated from the sampling site to the NW/SE (Fig. 2) directions corresponding to the residential area. The time series show a local heating pattern, with a peak in the morning around 8 a.m. and in the evening starting from 5 p.m., with a maximum at 10 p.m. and lasting all night (Fig. SI3). This pattern can be also linked to the stable atmospheric conditions during winter nights, trapping the emissions near the ground. The regression analysis shows moderate relationship between the coal combustion factor and PM1 OC (R2 = 0.40). Surprisingly, this factor does not correspond to the coal combustion factor presented in Kozáková et al. (2019). This difference is likely due to the differences in the monitored periods (from February 6 to March 6, 2014), lack of chemical species (PM1 water-soluble ions and OC/EC) and merged data matrix for Ostrava-Radvanice and Ostrava-Plesná sites.

Fig. 2
figure 2

Polar plots for Plesná PM0.15–1.15 factors

Full size image
Fig. 3
figure 3

Radvanice factor profiles (concentration of the species—bars, percentage of the species—markers; DISP Average—box, DISP Max and DISP Min—whiskers) and time series resolved for PM0.15–1.15 by PMF.

Full size image
Fig. 4
figure 4

Polar plots for the Radvanice PM0.15–1.15 factors

Full size image
Fig. 5
figure 5

PNSD PMF factors for Plesná. The volume size distribution (dashed lines) was re-calculated from the PNSD assuming spherical particles

Full size image
Fig. 6
figure 6

Polar plots for Plesná NSD factors

Full size image

Factor 2. This mixed factor represented mainly by S and K was ascribed to SIA and biomass burning (Leoni et al. 2018). This factor contributed 18% to the mean PM0.15–1.15 mass on average. The polar plot points to the source/s of regional origin from N over E to S (Fig. 2). The factor shows a moderate and strong relationship with PM1 K+, levoglucosan, NH4+ and SO4−2, respectively (R2 = 0.40, R2 = 0.44, R2 = 0.48, R2 = 0.80). In the source apportionment study by Kozáková et al. (2019), secondary sulphate and nitrate of regional origin and biomass burning of local origin contributed to PM1 by 30%, 26% and 24%, respectively. The diverse result is the consequence of the lack of water-soluble ions in the data matrix and a different monitored period since the SIA and biomass burning peaks from 4 to March 6 are missing in the current study (3DRUM data series from February 19 to March 3).

Factor 3. The third factor was associated with high Cl and Br. There were moderate and strong correlations with the combustion factors (Factor 1 r = 0.53 and Factor 4 r = 0.74), and statistically significant correlations with PM1, Cl−, EC and K+ (R2 = 0.73, R2 = 0.80, R2 = 0.62). This factor was assigned as industrial coal combustion of regional origin; however, the residential combustion sources should be also considered due to good correlation with EC 24-PM1, as well as with local residential combustion sources in Plesná. Cl and Br are naturally present in coal as trace elements as well as Hg, As and Se (Finkelman et al. 1993). These elements mostly evaporate during combustion and condense either homogeneously as sub-micron ash or heterogeneously onto already existing fine ash, the former one being more difficult to be captured (Clarke 1993; Vejahati et al. 2010). Zíková et al. (2016) and Li et al. (2019) show the NH4Cl footprint of coal combustion in Beijing as an ageing product of HCl emissions from local combustion appliances and heating plants. The high factor contributions with WS of > 2 m s−1 from the NE point to sources of regional origin (Fig. 2). The three distinct midnight peaks on February 24, 25 and 26 reflect the stagnant atmospheric conditions during winter nights and points to the medium to large combustion source/s of a regional origin. This factor contributed 15% to PM0.15–1.15 and corresponds to the coal combustion factor of regional origin apportioned in Kozáková et al. (2019) with contribution of 18% to PM1.

Factor 4. This factor assigned as local biomass burning contained Zn, Na, MM and K. K also derives from other sources including waste incinerators and cooking (Duan et al. 2004; Wang et al. 2007; Caseiro et al. 2009; Mkoma et al. 2013). The campaign average contribution of this factor to the PM0.15–1.15 mass was 12%. The high factor contributions during periods of low WS (< 2 m s−1) point to local sources situated from NW over N to SE from the sampling site. The factor shows a strong relationship with PM1 levoglucosan and K+, respectively (R2 = 70 and 0.81, respectively). The daily pattern corresponds to the daily pattern of factor coal combustion (Fig. SI3). Kozáková et al. (2019) present a biomass-burning factor with a contribution of 24% to PM1.

Factor 5. This factor, re-suspended dust, was associated with the major crustal elements. (Almeida et al. 2005; Rogula-Kozłowska et al. 2012) and on average, contributed 4% to the PM0.15–1.15 mass. The dust source was located in the NE quadrant from the sampling site and became more important when WS was of > 2 m s−1 (Fig. 2). The re-suspended dust factor was not resolved by Kozáková et al. (2019) due to the limited elemental analyses of the PM samples. Therefore, they lacked the tracers of dust emissions.

Factor 6. The last factor was ascribed to raw iron production. Its profile shows a high concentration of metals especially Mn, Fe and Cu (Querol et al. 2007; Zhou et al. 2004; Cohen et al. 2010). The campaign average of the raw iron factor contributions was 2%. Source/s of rather regional origin from the NE and E (Fig. 2), and therefore from across the Polish border. However, the local sources in Ostrava could not be overlooked. The regression analysis shows a weak and strong relationship between the raw iron factor and PM1 Na+, and K+ (R2 = 0.32 and R2 = 0.71, respectively). Since the tracers for raw iron factor were missing in the Kozáková et al. (2019) study, the factor was not resolved.

The difference in the number and origin of the factors resolved by PMF based on PM0.15–1.15 and PM1 chemical composition reflects the different measurement period, analysed variables, sampling integration time and modelling approach (two separate matrices in this study and one common matrix for both sites due to the limited number of 24 h samples in Kozáková et al. 2019).

Radvanice

Factor 1. The first factor was associated with high concentrations of S, As, Br and MM. Thus, this factor represents secondary inorganic aerosol (SIA) and coal combustion. The polar plot points primarily to the residential area of Ostrava-Radvanice district situated to the north of the sampling site (Fig. 4). This mixed factor dominated during the measurement campaign (59%). The highest factor contributions to PM0.15–1.15 occurred for low WS (< 2 m s−1) and at a temperature lower than the campaign average. The factor shows weak and moderate relationships with PM1 EC, SO42−, and OC (R2 = 0.25, R2 = 0.26, R2 = 0.40). The mixed factor secondary inorganic aerosol/coal combustion/biomass burning (SIA/CC/BB) by Leoni et al. (2018) was also resolved based on the chemical composition of PM0.09–1.15. Although the results presented in Leoni et al. (2018) corresponded to a sampling period from February 10 to March 2, 2014, and the present study ran from February 19 to March 3, 2014, the agreement between the two analyses is good.

Factor 2. This factor characterized by Cl, Br and MM was ascribed to local residential heating. However, the industrial coal combustion source/s should be considered since the polar plot shows a high concentration from the north for a WS < 2 m s−1 and from the southwest when the WS was between 2 and 4 m s−1 (Fig. 4). The chemical profile represents coal and wood combustion in residential boilers and high temperature coke combustion in the metallurgical industry (blast furnace and steel production). The three distinct peaks on February 24, 25 and 26 corresponded to the peaks of the industrial coal combustion factor identified in Plesná. The mean residential heating and industry factor contributed 29% to PM0.15–1.15. The regression analysis shows moderate relationships with PM1 Cl− and OC (R2 = 0.46, R2 = 0.47, respectively). This factor was not resolved by Leoni et al. (2018). However, the three main peaks between February 24 and 26 correspond to high contributions of the mixed factor SIA/CC/BB of PM0.09–1.15.

Factor 3 was identified as raw iron production since its profile was primarily associated with Cr, Mn and Fe. The campaign average of the factor contributions to PM0.15–1.15 was 11% that is in agreement with the 10% value reported by Leoni et al. (2018). The high factor contributions from the SW point to local metallurgical sources.

Factor 4. The fourth factor was ascribed to sinter production because of its profile shows high concentration of Na, K and Ca. According to our previous studies (Pokorná et al. 2015), the chemical profile corresponds to sinter production. It does not correspond well to the sinter/steel production factor resolved by Leoni et al. (2018). This factor contributed 1% to PM0.15–1.15 with high contribution from SE with WS of 3 m s−1, and therefore, it likely represents the nearby metallurgical complex (Fig. 4). The sinter production factor showed a significant correlation with the raw iron production factor (Factor 4, r = 0.72) and weak and very strong correlations with PM1 Na+ and K+ (R2 = 0.21 and R2 = 0.83, respectively). The road dust factor identified by Leoni et al. (2018) was not resolved in this study of PM0.15–1.15. The difference in the number of the factors

Resolved from the PM0.15–1.15 and PM0.09–1.15, chemical composition was probably driven by the different measurement periods. Additionally, the Pokorná et al. (2015) source apportionment presented five PM0.15–1.15 factors at the Radvanice site; coal combustion, sinter production—hot phase, traffic, raw iron production and desulfurization slag processing. The 2012 measurement campaign was affected by a long-term smog period. Therefore, the meteorological conditions were not comparable with the present study. However, there was an overall agreement in the main sources influencing the air quality at the receptor site.

Plesná and Radvanice source comparisons

To assess the relationship between the resolved factor contributions at the two sites, the Pearson correlation coefficients (r) were computed. The mixed factor SIA and coal combustion (Factor 1, Radvanice) shows a low correlation with the factors of local residential biomass burning (Factor 4, Plesná; r = 0.16), local residential coal combustion (Factor 1, Plesná; r = 0.21) and industrial coal combustion (Factor 3, Plesná; r = 0.25). Additionally, the mixed factor local residential heating and industrial coal combustion (Factor 2, Radvanice) correlated weakly with industrial coal combustion (Factor 3, Plesná; r = 0.27). These limited pairwise correlations between the factors time series suggest that different sources were contributing to PM0.15–1.15. These results reflect the important effects of the local meteorology given by the region’s orography.

Factors resolved with PNSD and association with chemical composition factors

Plesná

The model was run approximately 100 times to find the most physically meaningful result and the best diagnostics. Due to the low S/N ratio, four variables were set as weak (18.4 nm, 5.1 μm, 6.7 μm and 7.8 μm; midpoint of the size bins) and their uncertainty tripled. Four variables in the coarse size were excluded from the matrix given the high noise level in the data (9.7 μm, 11.9 μm, 14.9 μm, 18.5 μm, midpoint of the size bins). The model was run with different factor numbers (3–9) with 0% extra modelling uncertainty. The most stable solution was found for 6 factors. All runs converged, the scaled residuals were normally distributed, no unmapped factors were detected with bootstrap error estimation and no swaps were observed with displacement error analysis, indicating that the solution was stable (Table S1). The six factors in Plesná were assigned as residential combustion, urban background, traffic/metallurgical industry, new particle formation (NPF)/metallurgical industry, regional pollution and residential heating/re-suspended dust. The non-normalized PNSD (N cm−3) was analysed by the model. For the result presentation, the model outputs were converted to dN/d (log Dp) and dV/d (log Dp). The factor NSD and volume size distributions (VSD) are shown in Fig. 5.

Factor 1. This factor includes particles in the Aitken mode and accumulation mode (Aitk2) with the number mode diameter (NMD) at 75 nm and distinct volume mode diameter (VMD) at 274 nm (small VMD at 2 μm, Fig. 5). The average concentration was 2.5 × 103 particle cm−3 with a maximum of 1.2 × 104 particle cm−3. This factor is the important contributor for PNC (34%) and a minor contributor to volume concentration (11%). The highest concentrations were under an eastern WD at a WS of 2–3 m s−1 (Fig. 6). The diel heating pattern and/or the stagnant atmospheric conditions during winter nights is also obvious (Fig. SI4). There is a moderate correlation with PM0.15–1.15 factors of industrial coal combustion (r = 0.47), local residential biomass burning (r = 0.46), raw iron production (r = 0.44) and local residential coal combustion (r = 0.41). This factor represents urban background also identified in Radvanice by Leoni et al. (2018) (NMD 93 nm) and similar to the London urban background site reported by Vu et al. (2016) (NMD 93 nm). An urban background factor with morning and evening maxima (Fig. SI4), the same factor daily pattern, attributed to road traffic and residential heating, was observed in the studies by Beddows et al. (2015) and by Masiol et al. (2017).

Factor 2. This includes accumulation mode particles (Acc1) and is the major contributor to PNC (30%) and to volume concentration (38%). The size distribution has a predominant mode, with NMD at 144 nm and VMD at 379 nm (Fig. 5). The average concentration was 2.6 × 103 particle cm−3 and peaks up at 6.7 × 103 particle cm−3. The daily variability reflects the heating pattern with morning and afternoon peaks. However, this pattern was also linked to the stagnant atmospheric conditions during winter nights (Fig. SI4). The highest concentrations were observed with a low WS (< 2 m s−1) mainly from the northeast (Fig. 6). In comparison with the factors resolved based on the PM0.15–1.15 chemical composition in Plesná, this factor has a moderate correlation with local residential coal combustion (r = 0.50), local residential biomass burning (r = 0.48) and industrial coal combustion (r = 0.44). This factor represents the residential combustion sources of local origin. In Radvanice, a coal combustion factor with NMD at 160 nm and VMD at 500 nm contributed by 14% to PNC and by 42% to PVC (Leoni et al. 2018).

Factor 3 contributed 12% to PNC and 1% to PVC, having a prominent mode at 39 nm (Aitk1, Fig. 5). It shows an average concentration of 7.6 × 103 particle cm−3 with a peak at 9.7 × 103 particle cm−3. The polar plot shows high PNC from the north-northeast to south-southeast for W > 2 m s−1 (Fig. 6). There was no significant correlation with the factors resolved based on the PM chemical composition due to the size range of the impactor. These particles are likely to be from traffic and industrial origins given the prevailing wind direction and the industrial source identification by Leoni et al. (2018) with NMD at 45 nm in Radvanice. Other studies reported a high concentration of ultrafine particles emitted by steel and iron production (Weitkamp et al. 2005; Cheng et al. 2008; Riffault et al. 2015).

Factor 4. This factor includes particles in the nucleation mode (Nano) with a mode at 21 nm (Fig. 5) and a contribution of 8% to PNC. The average particle concentration was 6.0 × 102 particle cm−3 with a maximum of 3.7 × 103 particle cm−3. The factor shows a diel pattern with the maxima between 1 PM and 2 PM (Fig. SI4) pointing to photochemical nucleation by the highest solar radiation and particle growth recorded on 26.2, beginning between 1 PM and 2 PM (Fig. SI5). The highest PNC was recorded with WS > 3 m s−1 from the north-northeast and south-southeast. According to the polar plots, new particle formation (NPF) events were most probably accompanied by industrial emissions of local and regional origin, from transboundary transport of pollution from the Polish part of Silesia (Fig. 6). The metallurgical industry has been identified as a source of nanoparticles (Dall'Osto et al. 2008; Riffault et al. 2015; Leoni et al., 2016 and 2018); however, no statistically significant correlation with the factors resolved based on the PM chemical composition was found.

Factor 5. It is characterized by accumulation mode particles (Reg) with two NMD (144 and 379 nm) and VMD (524 and 724 nm). This factor is a major contributor to volume concentrations (33%) and a minor contributor to PNC (7%) with an average concentration of 4.9 × 102 particle cm−3 and a maximum at 2.3 × 103 particle cm−3. The factor shows the dial course linked to the heating pattern and/or the stagnant atmospheric conditions during winter nights. The highest PNC occurred for northeasterly winds with a WS of > 2 m s−1. The factor correlates moderately with the PM0.15–1.15 factors of industrial coal combustion (r = 0.65), local residential biomass burning (r = 0.47) and secondary inorganic aerosol along with biomass burning.

The last factor, Factor 6 is characterized by super-micrometre mode particles (Coarse) with appreciable VMD at 3 μm (small VMD at 306 nm, Fig. 5). It contributes 16% to PVC and 2% to PNC. The factor average concentration was 1.4 × 102 particle cm−3 with a peak value of 8.8 × 102 particle cm−3. The polar plots showed the highest PNC for northeasterly winds of > 2 m s−1 (Fig. 6). The factor does not have a clear diel pattern (Fig. SI4). The factor shows moderate correlation with raw iron production (r = 0.53). However, the expected correlation with the re-suspended dust factor is very weak (r = 0.16), probably due to the upper cut point of the impactor being 1.15 μm. In comparison with Radvanice, Leoni et al. (2018) identified a mixed industrial coarse particle and road dust factor.

Radvanice

Since the source apportionment of aerosol particles based on PNSD in Radvanice is the subject of the study by Leoni et al. (2018), only the PNSD sources apportionment in Plesná is fully presented in this study. Briefly, Leoni identified six sources. Mixed source industrial and fresh road traffic, industrial ultrafine particles and urban background contributed most to the PNC (26%, 28%, 24%, respectively). The highest contribution to the particle volume concentration (PVC) was coal combustion along with mixed source industrial coarse particles and road dust (42% and 20%).

For the purpose of this study, the Pearson correlation coefficients (r) between the Radvanice PNSD and PM0.15–1.15 resolved factors were computed. Mixed source industrial and fresh road traffic nucleation (Factor 1, Radvanice; Leoni et al. 2018) shows moderate to strong correlation with sinter (Factor 4) and raw iron production (Factor 3) r = 0.47 and 0 = 0.74, respectively. Industrial ultrafine particles (Factor 2, Radvanice; Leoni et al. 2018) strongly correlated (r = 0.74) with sinter production (Factor 4). Coal combustion (Factor 4, Radvanice; Leoni et al. 2018) shows moderate correlations with mixed sources residential heating and industrial coal combustion (Factor 2, r = 0.64), and secondary inorganic and coal combustion (Factor 1, r = 0.57). Regional pollution (Factor 5, Radvanice; Leoni et al. 2018) correlated moderately with mixed sources residential heating and industrial coal combustion (Factor 2, r = 0.53). Mixed source industrial coarse particles and road dust (Factor 6, Radvanice; Leoni et al. 2018) shows moderate correlations with raw iron and steel production (Factor 3 and 4) r = 0.49 and r = 0.64, respectively.

Plesná and Radvanice source comparisons

To assess the relationship between the resolved PNSD factors at the two sites, the Pearson correlation coefficients (r) were computed to compare the shared variance. The factor coal combustion (Factor 4, Radvanice; Leoni et al. 2018) shows moderate correlation with the Plesná factors of residential combustion of local origin (Factor 2) r = 0.68 and regional pollution (Factor 5) r = 0.52. Additionally, the factor regional pollution (Factor 5, Radvanice; Leoni et al. 2018) correlated moderately with the factor regional pollution in Plesná (Factor 5) r = 0.53.

Conclusions

This study accessed the spatial and temporal variability of the winter air particulate pollution sources in an urban settlement influenced by metallurgical industry using atmospheric aerosol chemical composition and PNSD. The modelling results confirmed the relevance and importance of the complementary datasets enabling the identification of sources contributing to both particle mass and particle number. While in Radvanice, the PNSD and PM0.15–1.15 sources show moderate to strong correlation, in Plesná, the two mixed PNSD sources show weak correlation (traffic and industry—Aitken mode at 39 nm and nanoparticles with a mode at 21 nm) with the PM0.15–1.15 sources.

The study results elucidated the importance of the local meteorology given by the region’s orography. Similar sources according to the factor chemical profiles resulted from different physical sources. In Plesná, residential combustion sources of local origin followed by regional combustion sources primarily contributed to PM0.15–1.15 (Tab. SI4). Surprisingly, the influence of the local industrial emissions was not very important at the site. In Radvanice, local residential combustion and the metallurgical industry were the most important sources (Tab. SI5), which agrees with our previous studies (Pokorná et al. 2015, 2016). Aitken and accumulation mode particles emitted by the urban background (residential heating, industry and traffic) along with local residential combustion sources were the major contributors to PNC in Plesná (Tab. SI4). Additionally, NPF events contributing 8% to PNC were probably mixed with local industrial nanoparticles similar to what was reported in the previous study by Leoni et al. (2018). In contrast, at the Radvanice site, the local industrial sources were the major contributor to PNC (Leoni et al. 2018). However, for a NE prevailing wind with speeds of > 2 m s−1, air pollution transboundary transport and regional sources were evident at both receptor sites in agreement with the results presented by Kozáková et al. (2019). Last but, not least, since the measurement campaign was conducted in winter and the sampling sites are moderately influenced by the traffic density, the contribution of road traffic sources to PM and PNSD seems to be reasonable. However, the measurement of additional road traffic tracers and/or the annual measurement will enable more precise source apportionment of road traffic at both sites.

To conclude, the presented results are representative for winter seasons by the same meteorological conditions, since the meteorology plays a crucial role in the air pollution not only in the Ostrava city, but also in the whole Moravian-Silesian region. Therefore, the long-term complex measurement (highly time- and size-resolved chemical composition and PNSD) at least at three measurement sites well distributed in the city will enable precise source apportionment, and consequent planning and implementation of effective abatement strategies to improve air quality.

The methodology presented in this study can be replicated at any receptor to elucidate the spatial and temporal variability of the air particulate pollution sources contributing to particle mass as well as particle number concentrations.