Source apportionment of PM_2.5 before and after COVID-19 lockdown in an urban-industrial area of the Lisbon metropolitan area, Portugal

Abstract

The lockdowns held due to the COVID-19 pandemic conducted to changes in air quality. This study aimed to understand the variability of PM_2.5 levels and composition in an urban-industrial area of the Lisbon Metropolitan Area and to identify the contribution of the different sources. The composition of PM_2.5 was assessed for 24 elements (by PIXE), secondary inorganic ions and black carbon. The PM_2.5 mean concentration for the period (December 2019 to November 2020) was 13 ± 11 μg.m⁻³. The most abundant species in PM_2.5 were BC (19.9%), SO₄²⁻ (15.4%), NO₃⁻ (11.6%) and NH₄⁺ (5.3%). The impact of the restrictions imposed by the COVID-19 pandemic on the PM levels was found by comparison with the previous six years. The concentrations of all the PM_2.5 components, except Al, Ba, Ca, Si and SO₄²⁻, were significantly higher in the winter/pre-confinement than in post-confinement period. A total of seven sources were identified by Positive Matrix Factorisation (PMF): soil, secondary sulphate, fuel-oil combustion, sea, vehicle non-exhaust, vehicle exhaust, and industry. Sources were greatly influenced by the restrictions imposed by the COVID-19 pandemic, with vehicle exhaust showing the sharpest decrease. Secondary sulphate predominated in summer/post-confinement. PM_2.5 levels and composition also varied with the types of air mass trajectories.

Graphical abstract

1. Introduction

The study of atmospheric particulate matter (PM) is of utmost importance because of the serious health implications it has on humans (Kim et al., 2015; WHO, 2013). It has been demonstrated that the long-term exposure to PM_2.5 was responsible for 417,000 premature deaths in Europe in 2018 (EEA, 2020). PM produces toxic effects according to their physical and chemical properties (Harrison and Yin, 2000). Many PM_2.5-bound organic constituents and metals have been associated with increased hospital admissions (Zanobetti et al., 2009), and namely Fe, Cr, Cu and Mn may result in pulmonary and heart diseases (Cakmak et al., 2014) and polycyclic aromatic hydrocarbons, Cd, Cr and Ni contribute to high carcinogenic risks (Gao et al., 2016).

The year 2020 begun with the outbreak of a new disease, COVID-19, which was declared a pandemic by the World Health Organisation (WHO) on March 11, following which strict measures as quarantines, restrictions of movements, curfews and lockdown of entire cities were implemented worldwide (He et al., 2020; University of Oxford and Blavatnik School of Governement, 2020). The first confirmed case in Portugal was on March 2, on March 16 schools were closed and the state of emergency was decreed on March 18 (Presidência da República, 2020) until May 4, imposing the general duty of home confinement, mandatory adoption of teleworking and closure of non-essential services.

The restriction measures reduced mobility and minimised anthropogenic activities in an unprecedented way around the world, creating a unique opportunity for assessing the effect of anthropogenic activities on air pollutants (Querol et al., 2021) and on the associated health risks (Aix et al., 2022; Cai et al., 2022). Many studies have reported that air quality has improved significantly during the city lockdowns, due to a drastic reduction of primary pollutants (Collivignarelli et al., 2020; He et al., 2020; Rodríguez-Urrego and Rodríguez-Urrego, 2020; Sharma et al., 2020), while increased secondary pollutant levels were reported (Hong et al., 2021; Wang et al., 2021a; Chang et al., 2020; Zheng et al., 2020; Chang et al., 2022; Ma et al., 2022), underlying the need to understand more deeply the formation mechanisms of secondary PM_2.5.

There are some studies that report the comparison of PM_2.5 chemical composition and source contributions in the pre- and post-confinement periods (Hong et al. (2021); Cheng et al. (2022); Cui et al. (2020); Fatima et al. (2022); Li et al. (2021); Zheng et al. (2020); Nguyen et al. (2022); Wang et al. (2021a) and Chang et al. (2020)), but only few are relative to European urban-industrial areas (Massimi et al. (2022); Hicks et al. (2021)) and none to Portugal, to the best of our knowledge.

In Seixal, a highly populated urban-industrial area in the Lisbon Metropolitan Area, comprising heavy industry such as steelworks, and crossed by highways with high density commuting traffic, has suffered from poor air quality in the last decades (Gamelas et al., 2021). Occasional settled dust events, when houses, cars, plants etc. were covered by dust (coarse particles of a granulometric dimension much higher than PM₁₀) have increased the population's concerns regarding the impacts of the air pollution on their health, leading the local Council of Seixal to promote actions, such as the assessment of the chemical composition of the settled dust to determine its sources (Justino et al., 2019) and an air biomonitoring study using lichens in the area (Abecasis et al., 2022). In both studies, the influence of the steel industries was identified due to the content of Fe, Cr and Mn, along with the traffic influence. In fact, although measures have been implemented in many of the steel operations to abate emissions, important amounts of PM are still produced, namely from diffuse emissions (Almeida et al., 2015). Steelworks have been identified as major emission sources of metal(loid)s such as Fe, Mn, Cr, Zn, Pb, As, Cd, Cu, Ni, Se and V (Lage et al., 2016; Taiwo et al., 2014).

In a previous study (Gamelas et al., 2021), where the temporal variability of the concentrations of air pollutants (NO₂, O₃, PM_2.5, PM₁₀ and SO₂) was assessed in Seixal, with an emphasis on the last six years, a significant improvement in air quality regarding PM₁₀ and NO₂ was found due to the COVID-19 lockdown, but O₃ and SO₂ levels did not follow that trend, especially in summer.

The present study aimed to identify, for the first time, emission sources and quantify their contributions to PM_2.5 in Seixal. To achieve this, the concentrations of selected elements, secondary inorganic ions and black carbon in PM_2.5 were analysed and Positive Matrix Factorisation (PMF), a widely used receptor model (Almeida et al., 2020; Belis et al., 2013; Viana et al., 2008), was applied to identify and apportion the emission sources to which the residents of Seixal were exposed to, before and after the COVID-19 confinement, a knowledge that is critical to design effective PM_2.5 mitigation strategies in the near future.

2. Materials and methods

2.1. Study site

PM was sampled at Aldeia de Paio Pires (38.617885°, −9.080055°), located in the municipality of Seixal (Portugal), a suburban and industrial area in the Metropolitan Area of Lisbon (Fig. 1 ). Seixal is one of the most densely populated municipalities in Portugal, with 167,294 inhabitants in 95.5 km² (PORDATA, 2022). The sampling site is at 2–3 km from high density traffic highways (A2-IP7 and A33), and is surrounded by small and medium-sized industries, as well as an industrial park (at 1 km) (Abecasis et al., 2022): a steelworks that manufactures galvanised sheet metal; an Electric Arc Furnace steelworks; a lime factory involving the calcination of limestone in a coke kiln; and a metal waste management and treatment factory. Furthermore, the sampling site is at 3.5 km distance from a shipyard and the ferry terminal that connects the two river banks for commuting (Fig. 1).

Fig. 1.

Location of the sampling site: (left) Framework of the study area (black rectangle); (right) A – sampling site, B – ferry terminal, C – shipyard, and whitish rectangle – industrial area.

2.2. Particulate matter sampling

Sampling was carried out in 24 h periods, starting at 10:00 GMT (except for Sundays, when sampling begun at 0:00 GMT), in 128 sampling days, from December 2019 to November 2020, except in the months of April and May, where no sampling was carried out because of the COVID-19 lockdown. Three sampling campaigns are distinguished: winter (from December 18, 2019 to March 18, 2020), summer (from June 15 to August 9, 2020) and autumn (from September 21 to November 12, 2020). Two periods are distinguished: the pre-confinement period, prior to the state of emergency decreed in March 18, 2020 (Presidência da República, 2020) and the post-confinement period, after May 2, 2020 when the state of emergency ended (Assembleia da República, 2021) and when less restrictive regulations were applied.

PM sampling was conducted using two medium volume samplers (MVS6, Leckel, Sven Leckel, Germany) that worked in parallel, equipped with a new sampling head developed by the Institute of Nuclear and Radiological Sciences and Technology, Energy and Safety, N.C.S.R. Demokritos, for the simultaneous collection of PM_2.5 and PM_{2.5 – 10}, at a constant flow rate of 2.3 m³/h. The quality assurance and control of the new heads was described by Faria et al. (2020). One sampler collected PM_2.5 on PTFE filters (SKC) of 47 mm diameter and pore of 2.0 μm and PM_{2.5 – 10} in PTFE filters (Pall) with 25 mm diameter and pore of 0.5 μm. The other sampler collected PM_2.5 on quartz filters (Pall) of 47 mm diameter, and PM_{2.5 – 10} on quartz filters of 25 mm diameter.

PM_2.5 and PM₁₀ concentrations corresponding to the sampling days and also PM₁₀ concentrations for the years 2014–2019, were obtained for comparison from the Paio Pires monitoring station (38.621028°, −9.082056°) of the National Air Quality Monitoring Network, operated by the Portuguese Environmental Agency (APA), through the national information system QualAr (APA, 2022).

2.3. Gravimetric analysis

The filter loads were measured by gravimetry in a clean room (class 10,000). The weight of PTFE filters, before and after sampling, was obtained using a 0.1 μg sensitivity balance (Mettler Toledo UMT5), as the average of three measurements, when observed variations were <0.3 μg. The uncertainties in mass loading assessment were lower than 0.8% for all samples in 47 mm PTFE filters and 0.3% for 25 mm PTFE filters.

The PM concentrations determined gravimetrically show a good correlation with the concentrations obtained from the QualAr database (APA, 2022), for the Paio Pires air quality monitoring station, with r² = 0.74 and 0.89 and gradient of the least squared fitted lines 1.03 and 1.08, for PM_2.5 for PM₁₀, respectively (Fig. S3). The mean PM_2.5 and PM₁₀ concentrations of the Paio Pires monitoring station in the sampling period (10 ± 8 and 22 ± 12 μg.m⁻³, respectively) were not significantly different from the PM concentrations determined by the gravimetric reference method in this work (p-value >0.050).

2.4. Chemical analysis

2.4.1. Elements

The samples collected in PTFE filters were used for elemental analysis by Particle Induced X-Ray Emission (PIXE). PIXE analysis was conducted in the Tandetron Laboratory of ATOMKI, Debrecen, Hungary using the new external millibeam PIXE setup during two measurement campaigns, November 2020 and June 2021. The irradiation of the samples was performed at the left 45^o beamline of the ATOMKI 2MV Tandetron accelerator (Rajta et al., 2018) with a proton beam of 2.5 MeV energy and of 15–20 nA current. The accumulated charge on each sample was 10 mC. During the irradiation the samples were scanned in front of the beam in order to reduce the risk of degradation and to increase the analysed area. In November 2020 the characteristic X-rays were recorded by an ultra-thin window (UTW) Rayspec SDD detector with 30 mm² area and a Canberra type Si(Li) detector with 50 mm² active area. The UTW SDD was equipped with a pair of 4 cm long permanent magnet which protects the detector from the scattered protons, while a 125 mm thick Kapton absorber was used in front of the SiLi detector. In June 2021, a detector cluster containing 4 SDD X-ray detectors were applied. One of the detectors was the above mentioned ultra-thin windowed SDD equipped with the pair of magnets. Three 65 mm² SDD detector with Be window (12.5 μm thick) and a kapton absorber of 125 mm thickness made the cluster complete. In order to improve the analytical performance, spectra recorded by the 3 identical SDD detectors was summarized. This case excellent detection limits were achieved. The irradiation dose was measured by integrating the current on a Faraday cup placed behind the samples. The PIXE measurements were performed in He atmosphere (2 l/min flow rate), which enables the detection of X-rays down to 0.35 keV on the UTW SDD. Detailed description of the setup can be found in Aljboor et al. (2023). PIXE spectra was evaluated with the GUPIXWIN program code (Campbell et al., 2010) handling the samples as ‘thin’ specimen. The elemental composition of PM_2.5 in sampled filters was assessed for a total of 24 elements after blank corrections (Al, As, Ba, Br, Ca, Cl, Cr, Cu, Fe, K, Mg, Mn, Na, Ni, Pb, Rb, S, Se, Si, Sr, Ti, V, Zn, and Zr), although As, Rb, Sr and Zr are not presented since these elements were quantified in <40% of the samples. PIXE spectra by the UTW-SDD provided the concentrations of low-Z elements (Na – Fe) while concentrations of elements with Z > 19 (K—Pb) were calculated from spectra recorded by the SDD cluster or the Si(Li) detectors. Concentrations of elements common in the two spectral ranges (e.g. K, Ca, Ti, Fe) were used to check the goodness of the measurement parameters and evaluation. The accuracy of the PIXE analytical method was evaluated with NIST SRM2783 filter standard and the results for all applied X-ray detectors are presented in Table S1. The obtained concentrations for all detected elements were within uncertainty to the certified or reference values. The fit errors (%) were: Mg, Al, Si, Ca, Fe (1), K (2), S and Ti (4), P (7), Zn (9), Cl (10), Mn (16), Na (18), Cu (28), Ni (39), Cr (57). Analytical uncertainty ranged from 5 to 30%, depending on the elements fit error and concentration (Table S1). The detection limits for the measured elements are provided in Table S2. The Quality Control of the PIXE measurements were done according to best practices: at least one sample per sample holder was re-measured, while PIXE analysis on reference materials was performed on each day of the measurement campaign. In addition, in November 2020 all samples were measured in He flow of 2 l/min and in normal atmospheric conditions. The obtained concentrations were the same within 5% uncertainty.

2.4.2. Water-soluble inorganic ions

Quartz fiber filters were used for the determination of water-soluble inorganic ions (SO₄ ²⁻, NO₃ ⁻ and NH₄ ⁺). The analyses were performed at the Department of Environment and Planning of the University of Aveiro, according to the methodology detailed in Vicente et al. (2018). The uncertainty of the determinations (mean uncertainty ± s.d.) was: SO₄ ²⁻100 ± 23, NO₃ ⁻ 72 ± 47 and NH₄ ⁺ 84 ± 26 ng.m⁻³. Half of the filter was extracted with ultrapure water (< 18.2 MΩ cm) by ultrasonication (Sonorex RK156, Bandelin) for 30 min (3 extractions of 10 min each, with a 5 min stop between them). Filtration through a pre-washed PVDF (Polyvinylidene Fluoride) filter of 13 mm diameter and 0.2 μm pore (WhatmanTM) was then performed and the filtered aqueous extracts were analysed by ion chromatography. For the analysis of anions, an ICS-3000 SP (DIONEX, USA) pump and an injection system Gilson model 401/231, and a column oven CTO-6A (Shimadzu) with a conductivity cell connected to the Shimadzu CDD-6A conductimeter were employed. A Dionex AG14 (4 × 250 mm) pre-column was placed between the injector and the Dionex AS14 (4 × 250 mm) analytical column. A Dionex AMMS II suppressor was used. The system was operated with a 50 μL loop, a reading range of 64 μS/cm and a gain of 0.1. H₂SO₄ (50 mM) and Na₂CO₃ + NaHCO₃ were used as regenerating solution and eluent, respectively, both with a flowrate of 2 mL.min⁻¹. For the analysis of cations, a Dionex DX-100 system was used, with a Gilson 234 injector, a Dionex CS12 column (4 × 250 mm), a Dionex CG12A (4 × 50 mm) pre-column and a suppressor (Dionex CMMS III). The system was operated with a 100 μL loop and a reading range of 100 μS/cm. A regenerating solution of tetrabutylammonium hydroxide (TBA) (100 mM) and an eluent solution of H₂SO₄ (25 mM), diluted in ultrapure water, were used at a flow rate of 1 mL.min⁻¹. Blank PTFE and quartz filters were treated the same way as regular samples and concentrations were corrected by subtracting the filter blank contents.

2.4.3. Black carbon

A Multi-wavelength Absorption Black Carbon Instrument (MABI) developed by ANSTO, was used to measure BC in PM_2.5 sampled in the 47 mm PTFE filters. The light transmission was measured through unexposed filters (I_o) and sampled filters (I) and the BC mass concentrations (in ng.m⁻³) were determined by eq. 1 (Manohar et al., 2021):

$$\mathit{BC}\left({\mathrm{\lambda }}_{639}\right)=\frac{{10}^5.\mathrm{A}}{\mathrm{\epsilon }\left({\mathrm{\lambda }}_{639}\right).\mathrm{V}}ln\left(\frac{I_0}{I}\right)$$ (1)

where ε(λ₆₃₉) is the mass absorption coefficient at λ = 639 nm, with a value of 6.036 m².g⁻¹, A is the sampled filter area (cm²), and V is the sampled volume of air (m³). BC determined at λ = 639 nm (BC(λ₆₃₉)) corresponds to the total BC, including fossil fuels and biomass burning.

The detection limit (DL) at λ = 639 nm was determined from the reported minimum detectable limits (Manohar et al., 2021), expressed in ng.cm⁻², by multiplying by the filter area A and dividing by the volume V of air through the filter (to convert to ng.m⁻³). The fractional error estimates were also taken from Manohar et al. (2021), in terms of tiers below, around and above the DL. The uncertainty of the BC determinations (mean uncertainty ± s.d.) was 400 ± 230 ng.m⁻³.

2.5. Positive Matrix Factorisation

Positive Matrix Factorisation, a widely used receptor model (Paatero, 1999; Paatero et al., 2014), version 5.0 (US-EPA, 2014), was used to identify the main emission sources and estimate their contributions to the PM_2.5 mass. PMF is based on the mass conservation principle:

$$x_{\mathit{ij}}=\sum _{k=1}^p{g}_{\mathit{ik}}{f}_{\mathit{kj}}+e_{\mathit{ij}}i=1,2,\dots m;j=1,2,\dots n$$ (2)

where x_ij is the concentration of the species j in the i^th sample, g_ik is the contribution of the k^th source in the i^th sample, f_kj is the concentration of the species j in the chemical profile of source k, and e_ij is the uncertainty of each individual measurement result. Factor contributions and profiles are derived in the EPA PMF 5.0 model by minimizing an objective function Q, without detailed prior knowledge on sources inventories (Paatero, 1999):

$$Q=\sum _{i=1}^n\sum _{j=1}^m{\left[\frac{x_{\mathit{ij}}-\sum _{k=1}^p{g}_{\mathit{ik}}{f}_{\mathit{kj}}}{u_{\mathit{ij}}}\right]}^2$$ (3)

Data matrix was composed by 128 samples and 24 PM species. For a variable to be included, a threshold of at least 70% of points higher than the detection limit (DL) was set. The model was supplied with the number of factors, the concentrations and uncertainties of the components of PM_2.5. The elements Pb and V were considered “weak” variables in the model; Ba, Cr, Ni, S and Se were considered “bad” variables and were excluded from the analysis (since they did not reach the threshold of at least 70% of points higher than the DL and S would be colinear with SO₄ ²⁻). PM_2.5 mass was used as “total variable”. Values below the DL were replaced by 1/2 DL and their uncertainties were set at 5/6 DL. Missing data were replaced by the median of the measured values and the uncertainties were set at four times the median value (US-EPA, 2014). The uncertainties of concentrations were evaluated as described in US-EPA (2014), using eq. 4:

$$\mathit{Uncertainty}=\sqrt{{\left(\mathit{Error\; fraction}.\mathit{Concentration}\right)}^2+{\left(0.5\mathit{DL}\right)}^2}$$ (4)

In addition to the analytical uncertainty, extra uncertainty was added in order to account for the sampling uncertainty, using the methodology described in Amato et al. (2009).

The model was repeatedly run (100 times for each solution), with different number of factors/sources (n = 5 to 8), and the variation of the Q-values, the scaled residuals, the regression diagnostics (r² and slope), and the physical meaningfulness of the sources were assessed. Finally, the n = 7 solution was adopted, leading to a solution with 95–100% of the scaled residuals located between the range [−3,3] for all the elements. A percentage difference of 2.4% between Q_true and Q_robust was found, in the range of values found in intercomparison studies (Almeida et al., 2020).

The Base Model Displacement Error (DISP) method was then used to explore the rotational ambiguity in the PMF final solution (US-EPA, 2014). It evaluates the largest range of source profile values without a significant rise in the Q value. The factor profile values are adjusted to the maximum allowable level, with the constraint that the difference between the Q values associated with the original and the modified solutions (dQ) is not greater than a value predefined by the model (dQmax). There was no variation in the Q value in DISP and no factor swaps occurred for all the dQ_max values considered, indicating that there was no significant rotational ambiguity. Additionally, the Bootstrap (BS) method was used to assess the uncertainty that originates from random errors in the dataset and partially from rotational ambiguity. BS estimates the random errors on the matrix by altering the original dataset. During BS some rows of the original dataset are deleted while others are repeated. The model then provides a solution for each of the new datasets and the results are compared with the original solution. Base runs were reproduced at least in 86–100% of the factors, considering the correlation value of 0.6, showing that the number of factors (n = 7) is appropriate and the solution is satisfactorily robust (US-EPA, 2014).

2.6. Air mass trajectories

Back-trajectories were simulated using the NOAA/ARL Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT) model version 5.1.0 (Stein et al., 2015) and meteorological data from the Global Data Analysis System (GDAS1). Four days backward trajectories ending at the sampling point, at the start of each sampling day, were calculated using the vertical velocity option and using different starting heights (50, 500 and 1000 m). Back trajectories were clustered using the HYSPLIT module and a cluster number was retained which corresponds to the large variation in Total Spatial Variance percentage (Stunder, 1996).

Air masses were grouped into six clusters of the following classes (Fig. 2 and Fig. S4): M – maritime (clusters 1 and 6); MT - maritime transformed (clusters 3, 4 and 5); C – continental (cluster 2). Four samples could not be clustered.

Fig. 2.

Average backward trajectories of the clusters calculated with the HYSPLIT model (and percentages of sampling days represented by each cluster).

2.7. Meteorological data

Meteorological data were obtained from the Barreiro-Lavradio weather station (38.674444°, −9.0475°) of Instituto Português do Mar e da Atmosfera - IPMA. During the sampling period, the mean temperature was 17.9 ± 4.8 °C (from 8.5 to 30.0 °C) and the mean relative humidity was 78.2 ± 10.9% (from 42.0 to 99.0%), with a total of 31 rainy days (Fig. S1, Supplementary Information).

Fig. S2 shows the wind roses for each sampling campaign. In winter, there was a predominance of winds from North (N), Northeast (NE) and Southwest (SW), and in summer and autumn, from North (N), Northwest (NW) and Southwest (SW). The predominant winds from N and NE make the sampling site downwind of the industrial park where the steelworks are located, but upwind from the A2 highway which is in the SW direction (Fig. 1).

2.8. Statistical analysis

Statistical calculations were carried out using STATISTICA software version 13. The non-parametric Mann–Whitney U test was employed to assess significant differences between two independent groups, at a significance level of 0.050.

3. Results and discussion

3.1. PM mass concentration and chemical composition

3.1.1. PM levels and exceedances

The concentrations of PM and measured constituents of PM_2.5 are summarized in Table S3 and Fig. S5 (in the Supplementary Information), discriminated by sampling campaign. The mean PM_2.5 concentration for the whole sampling period was 13 ± 11 μg.m⁻³, below the PM_2.5 annual limit of 20 μg.m⁻³ of Directive 2008/50/EC, of 21 May 2008, but higher than the WHO annual guideline value of 5 μg.m⁻³ (WHO, 2021). In fact, the WHO daily guideline for PM_2.5 of 15 μg.m⁻³, not to be exceeded >3–4 days per year, has been exceeded 33 times in the 128 sampling days. The PM₁₀ mean concentration (24 ± 13 μg.m⁻³) was below the EU annual limit of 40 μg.m⁻³ and the PM₁₀ daily limit value of 50 μg.m⁻³, which must not be exceeded >35 times a year, was exceeded 4 times in the 128 sampling days. Nevertheless, the WHO annual guideline value of 15 μg.m⁻³ was exceeded (WHO, 2021).

The PM_2.5/PM₁₀ ratio was 0.49 ± 0.15 (ranging 0.21 from to 0.85) in the whole sampling period (with no significant difference between campaigns), meaning that usually there was an almost equal contribution from both the fine and coarse PM fractions. These values are comparable with the average PM_2.5/PM₁₀ ratio observed in 20 European areas (0.60) (Eeftens et al., 2012), and with other studies, namely in the vicinities of steelworks, by Mohiuddin et al. (2014) (0.35–0.54) and Almeida et al. (2015) (0.37–0.63).

In the 33 sampling days when the daily WHO guideline for PM_2.5 (15 μg.m⁻³) was exceeded, the PM_2.5/PM₁₀ ratio was 0.64, significantly higher than in non-exceedance days (p-value = 0.000), suggesting an increase in anthropogenic emissions in these days, since anthropogenic sources are known to contribute mainly to PM_2.5 and natural sources to PM_2.5–10 (Almeida et al., 2012; Fatima et al., 2022).

The sum of the percentage concentrations of the analysed species (Table S3) represented, on average, 71.8% of PM_2.5. The remaining fraction can be attributed to OC and carbonate (which were not measured), oxides of soil elements and water retained in the samples at the weighing conditions. The most abundant species in PM_2.5 were BC (19.9%), SO₄ ²⁻ (15.4%), NO₃ ⁻ (11.6%) (with mean concentrations higher than 1 μg.m⁻³) and NH₄ ⁺ (5.3%), which are mainly associated with anthropogenic sources and secondary production mechanisms (in the case of SO₄ ²⁻, NO₃ ⁻ and NH₄ ⁺) (Calvo et al., 2013). Other major species (above 100 ng.m⁻³) were Na, Cl, K, Ca, Si and Fe. These results were of the same order of magnitude of aerosol data in other urban areas in the Lisbon Metropolitan area (Almeida-Silva et al., 2020) (Table S3). When these figures are compared to Southern Europe urban sites, where BC (EC) only contributes around 8% to PM_2.5 mass (Putaud et al., 2010), it can be concluded that the emission of BC in the study area, probably from the highway traffic and industry, is higher. In fact, the mean BC concentration (2510 ± 2010 ng.m⁻³) was higher than the mean values found in urban background sites of Barcelona (1000 ng.m⁻³) and Milan (1784 ng.m⁻³) (Amato et al., 2016), and also higher than the value reported by Healy et al. (2017) for a site near a highway in Canada (1740 ng.m⁻³), being nevertheless lower than the value in a traffic site in Porto (5006 ng.m⁻³) (Amato et al., 2016).

Among heavy metals, the most abundant were Fe and Zn, contributing to 1.5% of PM_2.5. The levels found of Fe, Cr, Mn and Zn (considered tracers of the steel industry (Calvo et al., 2013)) were in the range of values observed in the vicinity of a major steelworks in the UK (Taiwo et al., 2014), but were lower than the mean values observed in sites under the influence of steel manufacture in Spain (Querol et al., 2007) (Table S3).

3.1.2. Impact of COVID-19 containment measures and seasonality on PM levels and composition

In the sampling period prior to the first state of emergency (winter/pre-confinement campaign), the PM_2.5 and PM₁₀ mean concentrations (17 ± 14 μg.m⁻³ and 30 ± 16 μg.m⁻³, respectively) were significantly higher than in the post-confinement period (9.4 ± 6.1 μg.m⁻³ and 19 ± 10 μg.m⁻³, p-value = 0.000), and no significant difference was found between the two post-confinement campaigns (p-value >0.050). Thus, there was a reduction of 46.1% and 35.9% in PM_2.5 and PM₁₀, respectively, from the pre-pandemic to the pandemic period, a pronounced drop that was certainly due to the confinement measures imposed to halt the COVID-19 pandemic, which led to a reduction in anthropogenic emissions (Chauhan and Singh, 2020). This impact was higher compared to the national level, where there was an average reduction of 18% for PM₁₀, considering 20 monitoring stations dispersed in mainland Portugal (Gama et al., 2021), but it was comparable to the European level, where the largest average reductions in PM₁₀, of about 40% in traffic stations and 30% in urban and suburban stations, were registered in Spain (EEA, 2020). A large span in PM_2.5 reductions was observed worldwide, as for example, from metropolitan areas in Germany (5%) (Balamurugan et al., 2022), Grenoble (22%) (Aix et al., 2022), Lanzhou (Chang et al., 2022) and North China Plain (Li et al., 2021) (∼20%), to Shanghai (Chen et al., 2020a) and Chongqing (Chen et al., 2020b) (30–50%).

Besides the influence of the pandemic, the reduction of PM_2.5 and PM₁₀ may be due in part to the seasonal variability, since PM typically shows higher levels during the cold season, due to the lower mixing layer heights (Granados-Muñoz et al., 2012) and stagnant episodes associated with thermal inversion (Lyamani et al., 2012) that inhibit the dispersion of pollutants in winter. For this reason, a comparison with the previous years should be performed. In fact, the mean PM₁₀ concentrations registered in Paio Pires monitoring station in the previous six years (2014-2019) had a significantly higher value of 27 ± 15 μg.m⁻³ (Gamelas et al., 2021) when compared with the PM₁₀ mean concentrations found for the 128 sampling days in the present study (24 ± 14 μg.m⁻³, p-value = 0.042), thus confirming the impact of the confinement measures on PM levels.

The change in the emission patterns of air pollutants caused by the pandemic was felt not only in the PM mass concentrations, but also in its chemical composition, as depicted by Fig. 3 , which shows that concentrations of all PM_2.5 components, except for Al, Ba, Ca, Si and SO₄ ²⁻, were significantly higher in the winter/pre-confinement than in the post-confinement period (p-value <0.050). These results are in line with other works that have shown significant reductions in PM_2.5 primary components concentrations (e.g. K, Fe, Zn, Ba, Mn, Pb, Cu, As, Ni, Se and Cr, in Cheng et al. (2022); Cr, Fe, Na⁺, Cl⁻, Pb, Ca, Cr, Cu, Fe, Zn and EC, in Li et al. (2021); Zn, Pb, Mn, As, Cu, Ni, V, Cr, Cd and Co, in Wang et al. (2021b); Cd, Se, As, Sr, Ba, Cu, Mn, Pb, K, Zn, Ca, Al, and Mg, in Nguyen et al. (2022)), attributed as an effect of the measures imposed by the COVID-19 lockdown. Soil elements Al and Si, did not vary significantly, since being of natural origin, were less affected by the reduction of anthropogenic activities.

Fig. 3.

Variation of concentration of aerosol components from the pre- to post-confinement period. Significantly different concentrations between the two periods are identified with * (p-value <0.050) and ** (p-value <0.010).

BC concentrations were significantly higher in the winter/pre-confinement (3450 ± 2520 ng.m⁻³) than in the post-confinement period (1860 ± 1200 ng.m⁻³, p-value = 0.000) and no significant difference was found between summer and autumn (post-confinement) campaigns (p-value >0.050). The atmospheric pollutant levels showed different decreasing trends in different sites, in line with different intensity of the COVID-19 mitigation measures put in practice, among other factors, such as meteorology. Nevertheless, reductions in BC concentrations have been consistently registered in lockdown periods relative to pre-pandemic conditions, around the globe, namely in Massachusetts/USA (22–46%) (Hudda et al., 2020), Chongqing/China (39%) (Chen et al., 2020b) and Milan/Italy (Collivignarelli et al., 2020), where the highest decrease (by 71%) was noticed.

The same trend was observed for NO₃ ⁻, for which the most pronounced reduction (76.4%) was observed, from the winter/pre-confinement (2690 ± 3030 ng.m⁻³) to the post-confinement period (635 ± 575 ng.m⁻³, p-value = 0.000), with no significant difference between the two post-confinement campaigns (p-value >0.050). This is in line with the concentrations of its gaseous precursor NO₂ in the studied area, which registered significant reductions in the pandemic period of 2020 relative to the previous six-year averages (44.0% in April, 17.4% in May, 29.0% in June and 31.6% in August) (Gamelas et al., 2021). Less pronounced reductions in NO₃ ⁻ concentrations were generally observed worldwide (namely 17.6%, in Lanzhou (Chang et al., 2022) and 6.7%, in Linfen (Liu et al., 2022)), but there were also cases in which an increase was observed, as the result of enhanced atmospheric oxidation capacity (Ma et al., 2022).

SO₄ ²⁻ is an exception in the variations presented by Fig. 3, with summer/post-confinement concentration (3000 ± 1570 ng.m⁻³) being significantly higher (p-value = 0.000) than in the winter/pre-confinement (1760 ± 1150 ng.m⁻³) and autumn/post-confinement (1130 ± 770 ng.m⁻³) campaigns (although there was no significant difference between pre- and post-confinement periods, p-value >0.050). The increase in SO₄ ²⁻ concentration in summer is in good agreement with the increase in the concentrations of its precursor SO₂ found in June (39.6%), July (40.6%), August (30.0%) and September (33.0%) (Gamelas et al., 2021), highlighting the impact of the local industries of the study area, which maintained their typical work flow in these months. The increase in SO₄ ²⁻ concentration in summer also confirms that SO₄ ²⁻ tends to increase during the hot season, due to enhanced photochemical activity that promotes the formation of secondary sulphates (NH₄)₂SO₄ (Almeida et al., 2006b). In fact, in summer, SO₄ ²⁻ and NH₄ ⁺ presented a strong correlation (r² = 0.80, Fig. S6 - left).

In winter, NH₄ ⁺ associates with NO₃ ⁻ instead, as evidenced by the very strong correlation in Fig. S6 (right), due to the shifting of the NH₄NO_3(s)↔ HNO_3(g) + NH_3(g) equilibrium to the aerosol phase, under the lower temperatures and higher relative humidity registered in this season (Almeida et al., 2006b), and this also contributes to the higher concentrations of NO₃ ⁻ in PM_2.5 observed in the winter/pre-confinement campaign.

The mass ratio NO₃ ⁻/SO₄ ²⁻ has been used as an indicator of the relative importance of mobile versus stationary sources of nitrogen and sulphur in the atmosphere (Arimoto et al., 1996). The ratio NO₃ ⁻/SO₄ ²⁻ took the value of 1.5 in winter/pre-confinement, 0.18 in summer/post-confinement and 0.64 in autumn/post-confinement campaigns, with a significant difference between pre- and post-confinement (p-value = 0.000). This suggests that in winter/pre-confinement traffic emissions prevailed, while in the post-confinement (especially in summer) stationary sources were predominant instead, due to the decrease in commuting mobility and sustained industrial activity in the area, as found in other works (Chang et al., 2022).

3.2. Source apportionment

To identify emission sources and assess their contribution to the sampled PM_2.5 concentrations, a source apportionment study was conducted using the PMF model. The samples of March 18 and July 20, when Sahara dust events occurred resulting in peak concentrations of Si, Al, Ca and Fe in PM_2.5 (Fig. S7), were excluded from the PMF modelling. The best solution was obtained with seven factors. The relation between the modelled and gravimetric PM_2.5 concentrations (Fig. S8) shows a strong correlation (r² = 0.85) and slope 0.86, meaning that together the sources accounted for 86% of PM_2.5. Fig. 4 shows the profiles of the different factors, Fig. 5 presents the source contributions obtained for each sampling day and Fig. 6 provides the source contributions discriminated by sampling campaign.

Fig. 4.

Source profiles identified by PMF, with concentrations of the species expressed in ng.m⁻³ (bars) and their percentages (red dots) for each source. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 5.

Source contributions for PM_2.5 concentrations (ng.m⁻³) during the sampling period, for the selected PMF solution.

Fig. 6.

(top) Relative source contributions for the different sampling periods, and (bottom) source contributions (in ng.m⁻³) for each sampling campaign.

Factor 1, labelled as “Soil”, is defined by Al, Si and Ti, but also with a contribution of Mg and Fe, all typical soil elements (Calvo et al., 2013). This factor contributed on average to 3.6% of PM_2.5 mass, a value which is relatively low compared to other studies in urban-industrial areas (Silva et al., 2020), but comparable with values found in some coastal urban areas (Manousakas et al., 2017). In fact, it is known that the crustal sources contribute to PM mainly in the coarse mode (>2.5 μm) (Lage et al., 2014; Song and Gao, 2011) and this percentage does not include the contribution of two sampling days when Sahara dust events occurred (Fig. S7).

The contribution of the “Soil” factor was significantly higher in summer/post-confinement/ than in winter/pre-confinement (p-value = 0.003) and autumn/post-confinement (p-value = 0.000) (Fig. 6), although there was no significant difference between the pre- and post-confinement periods (p-value >0.050). In fact, Artíñano et al. (2001) indicated that in the Iberian Peninsula there are factors that favour the suspension of dust, especially in summer, such as the dryness and semi-arid soil associated with the high convective atmospheric dynamics.

Factor 2, which contributed on average to 18.4% of the PM_2.5 mass, has a chemical profile mainly defined by SO₄ ²⁻ and NH₄ ⁺, being labelled as “Secondary sulphate”, since these ions derive from gas-to-particle conversion processes, from SO₂ oxidation and NH₃ neutralisation (Calvo et al., 2013). In addition to the industry installed in the study area, secondary aerosol certainly also derives from vehicle exhaust emissions, since diesel vehicles are important emitters of sulphates (Calvo et al., 2013). The presence of Na and Mg in the profile may indicate a reaction of SO₂ with sea salt (Salvador et al., 2016).

Furthermore, the profile also includes BC and V contents. A secondary sulphate factor influenced by fuel-oil combustion has been observed in other works (Pey et al., 2013; Titos et al., 2014). In fact, secondary aerosols are not emitted directly into the atmosphere by a single source, resulting from atmospheric chemical transformations of gaseous precursors, at shorter or longer timescales, and therefore they share the same marker species with anthropogenic emissions (Viana et al., 2008). High concentrations of sulphate, V and Ni, especially in the summer season, are also commonly reported in Mediterranean environments associated with re-circulation processes (Pey et al., 2013).

“Secondary sulphate” factor was the major contributor in the post-confinement period, accounting for 30.4% of PM_2.5 mass and, especially in summer, this factor constituted 48.0% of the aerosol, presenting a significantly higher contribution than in autumn and winter (p-value = 0.000) (Fig. 6). This seasonal pattern was also found in other studies (Pey et al., 2009; Silva et al., 2020). It is due to the enhanced photochemical reactions in summer, and to the fact that higher vehicle emissions in winter and autumn are associated with the preferential formation of NH₄NO₃ at lower temperatures (Almeida et al., 2013) (Fig. S6).

Factor 3 is mainly characterised by V and K. This factor, which contributed to 11.5% of the PM_2.5 mass, has been labelled as “Fuel-oil combustion”, since V is a marker of combustion of heavy oils (Pacyna and Pacyna, 2001). As described in section 2.8., Ni has been considered a “bad” variable and, thus, it was excluded from PMF modelling in this best solution. Nevertheless, in a previous modelling trial when Ni was included (as a “weak” variable), a very similar PMF solution was attained, in which this factor was characterised by V, Ni and K, in a V/Ni ratio of 4.9. Low V/Ni ratios (∼ 0.5) are typically found in diesel combustion particles, while ratios <2 are associated with the presence of a Ni pollution source (Cesari et al., 2014), and ratios of V/Ni between 4 and 5 and ratios of V/EC < 2 are considered tracers of commercial shipping emissions (Pandolfi et al., 2011; Viana et al., 2009). Since in this work, the V/Ni ratio tentative value is 4.9 and V/BC is 0.03, this supports the contribution of ship emissions in the study area, probably due to the proximity from the ferry terminal that connects the two river banks for commuting. Furthermore, this factor also likely represents a mixed industrial combustion source, namely from the petroleum coke combustion in the kiln of the lime industry located in the industrial area of the study site, since V and Ni are also tracers of this combustion (Querol et al., 2007; Viana et al., 2008).

The contribution of the “Fuel-oil combustion” factor was significantly higher in the pre-confinement period (2050 ± 2740 ng.m⁻³) than in the post-confinement period (987 ± 2180 ng.m⁻³, p-value = 0.003), with no significant difference between winter/pre-confinement/ and autumn/post-confinement (p-value >0.050). In fact, in the confinement and post-confinement periods there was a reduction in the daily number of ferry carriers for commuting between Lisbon and Seixal (Público, 2020), but in autumn this anthropogenic activity had already reached levels similar to the pre-pandemic period.

Factor 4 is labelled as “Sea”, given the high contributions of Na, Cl and Mg, typical tracers of sea salt (Calvo et al., 2013). This factor contributed on average to 11.0% of the PM_2.5 mass, due to the proximity from the Tagus Estuary and the Atlantic Ocean, although the marine aerosol predominantly contributes to the coarse mode (Almeida et al., 2006a; Calvo et al., 2013). The average contribution of this factor (1370 ng.m⁻³) is similar to the sea salt contribution to PM_2.5 found in Porto (1220 ng.m⁻³) (Diapouli et al., 2017).

The average Cl/Na ratio in this factor is 0.98, which is lower than the bulk sea water weight concentration ratio (1.79) (Bowen, 1979). Cl/Na ratios between similar values (1.1) (Almeida et al., 2015) to much lower ones (0.12) (Dall'Osto et al., 2013) have been documented in other works. The discrepancy from the sea water ratio is due to Cl depletion that commonly takes place on urban areas, as a result of the reaction of NaCl with acidic species (HNO₃, SO₂ and H₂SO₄) (Seinfeld and Pandis, 2006). The presence of NO₃ ⁻ and SO₄ ²⁻ in the profile confirms that the sea spray is aged and was subjected to chlorine depletion, with formation of NaNO₃ and Na₂SO₄ (Taiwo et al., 2014)_. The Mg/Na ratio in the factor is 0.12, consistent with the seawater value (0.12) (Seinfeld and Pandis, 2006).

The “Sea” factor presented a significantly higher contribution in winter (p-value = 0.000) and autumn (p-value = 0.011) than in summer (Fig. 6). This can be explained by the higher wave energy in winter and the more intense predominant winds from North (where the Tagus Estuary is located - Fig. 1).

Factor 5 is represented by the elements Zn, Pb, Mn, Cu and Fe. This factor, which accounted for only 2.3% of PM_2.5 mass, is labelled as “Vehicle non-exhaust”, since Zn is a tracer of tyre emissions (Harrison et al., 2012) and Cu, Fe and Zn are markers of brake wear (Grigoratos and Martini, 2015). The elements Pb and Mn are also tracers of brake wear emissions (Lawrence et al., 2013; Song and Gao, 2011). The low contribution of this non-exhaust factor to PM_2.5 mass may be related to the fact that non-exhaust emissions generally contribute more to coarse particles, while the fine mode particles are mainly derived from combustion (Handler et al., 2008).

The mean contribution of the “Vehicle non-exhaust” factor was significantly higher in winter/pre-confinement than in summer/post-confinement (p-value = 0.047) (Fig. 6), but the difference was not significant between the pre- and post-confinement periods (p-value >0.050).

Factor 6 represents “Vehicle exhaust” and secondary nitrate emissions, due to the high percentage of BC, NO₃ ⁻ and NH₄ ⁺ in the factor. In fact, BC can be used as a key tracer of vehicular emissions (Hudda et al., 2020) and NO₃ ⁻ derives from NO_x precursors, mainly emitted by fossil fuel combustions, as vehicle exhaust (Arimoto et al., 1996).

In the literature, the presence of total carbon (TC) in secondary inorganic aerosols (SIA) profiles, namely “secondary nitrates” profile, has been assigned to the aging of SIA and/or the effect of mixing with particles from combustion sources, e.g. vehicle exhaust (Kfoury et al., 2016). Koçak et al. (2015) observed a moderate correlation between NO₃ ⁻ and TC, the absence of correlation between SO₄ ²⁻ and TC, and ascribed the NO₃ ⁻–TC relation to the combustion process.

Furthermore, vehicle exhaust is known to contribute to the enrichment of Cu in aerosol particles (Song and Gao, 2011) and this factor indeed accounts for 35.6% of the species. This factor also has a contribution from Pb, corroborating that Pb is also emitted from motor oil combustion (Handler et al., 2008).

Globally, this factor was the main contributor to PM_2.5 in the study area, accounting for 40.8% of the PM_2.5 mass, considering the whole period, but this factor presented the most pronounced decrease from the winter/pre-confinement (9620 ± 11,880 ng.m⁻³) to the post-confinement period (1910 ± 2640 ng.m⁻³, p-value = 0.000) and its relative contribution to PM_2.5 mass decreased from 56.3 to 20.7% between the two periods (Fig. 6). These results confirm the relevance of traffic and associated secondary aerosol in PM_2.5 in the study area and reveal that the decrease in the PM_2.5 concentrations was due, to a large extent, to the reduction of road traffic. In fact, after the declaration of the state of emergency (on March 18), the volume of traffic during the confinement period decreased drastically to 21% of the volume recorded in January 13, 2020, and then intensified after the end of the state of emergency (May 2). Globally, in the post-confinement period, the volume of traffic in Lisbon Metropolitan Area was on average, 84% of the volume recorded in January 13, 2020 (Apple, 2020).

Factor 7 has high relative contributions of Ca, Fe and Mn, being the main contributor to these elements in PM_2.5. This factor, which represented 12.6% of PM_2.5 concentration in the whole period, is labelled as “Industry”, since Fe and Mn are tracers of iron and steel industry (Calvo et al., 2013), along with Zn and Pb, which also contribute to the factor. Furthermore, Ca, Fe, Al, Mn and Si are primary constituents of the slag produced in the Electric Arc Furnace (EAF) process, where calcium oxide is added as fluxing agent (Proctor et al., 2000). This association of Fe, Mn and Ca in the same source factor attributed to steelworks emissions was also observed in the vicinity of a major steelworks in the UK and the presence of Ca was also attributed to the use of lime as fluxing agent (Taiwo et al., 2014). It should be mentioned that the EAF steelworks, as the one installed in study area, frequently present fugitive emissions from handling of raw materials, slag processing and dust resuspension (Mohiuddin et al., 2014). This factor presents indeed a character of resuspension, given the significant contribution of crustal elements such as Si and Ti, besides Ca. This factor can also be associated with the lime factory that exists in the industrial area of the study site. It should be mentioned that Fe (32.4%), Ca (26.2%) and Mn (4.62%) were also identified as major components of the settled dust collected in January 2019 in the study area, along with Si (14.9%) and Al (9.67%) (Justino et al., 2019).

While in the winter/pre-confinement period, this factor only represented 5.8% of the PM_2.5 mass, in the post-confinement period, this factor was one of the main contributors with 21.3% (Fig. 6). In relative terms, this was a consequence of the reduction of “Vehicle exhaust”, but the absolute contribution of the “Industry” factor was also significantly higher in the post-confinement (1960 ± 2400 ng.m⁻³) than in the pre-confinement (991 ± 1439 ng.m⁻³, p-value = 0.000), probably due to increased resuspension in the dry season and also corroborating the fact that the industries (namely, steelworks) of the study area did not stop to operate during the confinement period (Abecasis et al., 2022).

Globally, these results are in line with other works, where PMF source apportionment generally revealed decreased contributions of vehicle emissions (Cui et al., 2020; Cheng et al., 2022; Wang et al., 2021a) and in some cases, also increased contributions of secondary inorganic aerosol (Chang et al., 2022; Hong et al., 2021; Massimi et al., 2022).

3.3. Influence of air masses

Since Seixal is at a distance of 12 km from the Atlantic coast, it is influenced by the maritime aerosol, as found for the Lisbon Metropolitan Area (Almeida et al., 2013), with the air mass trajectories of maritime (24%) and maritime transformed (60%) origin predominating in the sampling period (Fig. 2). Lower PM_2.5 concentrations were determined in maritime clusters (8 ± 5 μg.m⁻³), followed by maritime-transformed clusters (12 ± 10 μg.m⁻³), while significantly higher levels were observed for the continental cluster (23 ± 15 μg.m⁻³, p-value = 0.000). This is expected, since maritime air mass trajectories are usually associated with the transport of cleaner air masses with better dispersion of pollutants (Almeida et al., 2013). Crustal elements Al, Si, Ca, Fe, as well as BC, SO₄ ²⁻, NO₃ ⁻ and NH₄ ⁺, Mn and Zn, registered significantly higher concentrations (p-value <0.050) in continental and maritime transformed air masses, while as expected, Na, Cl and Mg predominate in maritime air masses (Fig. S9).

The study area was occasionally affected by episodes of dust transport from the North of Africa, which is a phenomenon observed in Portugal and Mediterranean countries (Almeida et al., 2008; Diapouli et al., 2017). Inspecting the time series of the elemental concentrations in the sampling period, peak concentrations were observed in March 18 and July 20 (Fig. S7), for Si (2600 and 2440 ng.m⁻³), Al (1190 and 1110 ng.m⁻³), Ca (1070 and 1030 ng.m⁻³) and Fe (740 and 775 ng.m⁻³). In these two days, PM_2.5 concentrations 22 and 20 μg.m⁻³, and PM₁₀ concentrations 42 and 33 μg.m⁻³ were registered, much higher than the mean concentrations of the whole sampling period (13 and 24 μg.m⁻³, respectively). Backward trajectories by the HYSPLIT model, ending in the sampling site in March 18 and July 20, confirmed the long range transport of air masses from the North of Africa or Southern Europe (Fig. S10), which are usually enriched in soil elements, namely Fe (Almeida et al., 2008). These two days were excluded from PMF modelling and thus do not affect the following conclusions.

The source contributions were calculated for the main types of air mass trajectories. The results presented in Fig. 7 indicate that the different air mass types have influence on the contribution of the emission sources at the receptor. In fact, for maritime and maritime transformed samples, the mean contributions of “Fuel-oil combustion” and “Vehicle exhaust” factors were significantly lower (p-value = 0.000) compared with the samples affected by continental air masses, and as expected, the marine factor contributed significantly more (p-value = 0.000). Maritime transformed air masses present higher contributions than maritime air masses, from “Soil”, “Secondary sulphate”, “Vehicle non-exhaust” and “Industry” sources (due to the final recirculation through the continent).

Fig. 7.

Source contributions for PM_2.5 concentrations discriminated by air mass trajectory type.

4. Conclusions

This study investigated the PM_2.5 concentrations, composition (including 24 elements, black carbon and secondary inorganic ions) and sources in an urban-industrial area of the Lisbon Metropolitan Area (Seixal), from December 2019 to November 2020, including periods before and after the COVID-19 lockdown.

Source apportionment by PMF yielded seven sources contributing to PM_2.5 in the study area: soil, secondary sulphate, fuel-oil combustion, sea, vehicle non-exhaust, vehicle exhaust (the main contributor, considering the whole period), and industry. It was concluded that PM_2.5 concentrations, chemical composition and sources were greatly influenced by the restrictions imposed by the COVID-19 pandemic, with vehicle exhaust presenting the most pronounced decrease. The predominance of secondary sulphate was registered in summer/post-confinement. Besides the restrictions imposed by COVID-19 and the seasonality, the source contributions were affected by the types of air mass trajectories.

The COVID-19 pandemic context offered a unique opportunity to examine the effects of human-related activities on air quality in the study area, helping to define targeted mitigation measures to improve air quality in one of the most air polluted areas of Portugal. This study highlights the need for the priority implementation of traffic reduction measures, through better planning of the metropolitan area, creation of an efficient and high interconnectivity public transport network, incentives to the population for the use of public transport and electric mobility, and stimulation of teleworking, in order to reduce commuting movements. Industry measures are also needed to mitigate fugitive emissions. Secondary aerosol and its underlying factors deserve a special attention and need further research in the study area.

Author statement

The roles of each author in the submitted manuscript are the following:

C.A.G., A.S. and S.B. performed the filter sampling.

Z.K. performed the chemical analysis by PIXE.

A.V. and C.A. performed the ions chemical analysis.

C.A.G. performed the black carbon analysis.

C.A.G. performed the formal analysis, data curation and analysis, including statistical and modelling.

C.A.G. and N.C. wrote the original draft.

S.M.A., C.A. and N.C. reviewed the manuscript.

S.M.A. was responsible for the conceptualization, scientific supervision, review of the manuscript, funding acquisition, coordination of the project.

All authors have read and agreed to the published version of the manuscript.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary material

Data availability

Data will be made available on request.