Atmospheric pollution by airborne particles significantly contributes to climate change and has been linked to respiratory and cardiovascular diseases and lung cancer (Fuzzi et al., 2015; Kim et al., 2015; WHO, 2013). It has been estimated that in Europe alone more than 500 000 deaths per year can be attributed to PM exposure, and that pollution hot spots of PM are responsible for a loss in life expectancy of up to 36 months (Fuzzi et al., 2015). For EU member states, air quality monitoring – as laid down in the Air Quality Directive 2008/50/EC (European Parliament, 2008, 2015) – is mandatory and comprises quantification of airborne particulate matter (PM) and some of its constituents. The most important regulated metric to monitor particulate air pollution is the mass concentration, or more specifically the total mass per unit volume of air of particulate matter which is small enough to pass through a size-selective inlet with a 50 % efficiency cut-off at 2.5 and 10 µm aerodynamic diameter, commonly referred to as PM2.5 and PM10 respectively. Ambient limit values for PM2.5 and PM10 have been established in Europe (European Parliament, 2008, 2015; FOEN, 2018), the USA (US-EPA, 2016) and other countries worldwide.

Regulatory bodies, air quality networks and atmospheric instrument manufacturers all strive to improve air quality monitoring, yet there is still a lack of metrological traceability in airborne PM measurements. PM mass concentration was established as the default metric of PM based on the assumption that mass measurements are straightforward; they can be performed with a conventional balance. The gravimetric filter-based reference methods for PM10 and PM2.5 are set out in the standards EN 12341:2014 (CEN/TC 264/WG-15, 2014) and EN 14907:2005; however, they fall short in areas such as time resolution and ongoing quality assurance and quality control to control the effects of semi-volatile particles and water absorption by particles, for example (CEN/TC 264/WG-15, 2014; Eisner and Wiener, 2002; Hauck et al., 2004; Zhu et al., 2007). The measurement uncertainties for PM mass concentration in the directive (European Parliament, 2008, 2015) are 25 %, and thus much higher than those for gaseous pollutants (typically 15 %).

Automatic PM monitoring systems were developed in order to avoid these drawbacks and enable time resolutions below 24 h (Schwab et al., 2006; Weingartner et al., 2011; Zhu et al., 2007); however, demonstrating their equivalence to the reference manual gravimetric method is time consuming and expensive (Hauck et al., 2004; Zhu et al., 2007). There are also inconsistencies in the automatic instruments based on different working principles (e.g. light scattering, beta absorption, oscillating microbalance) and the variations of the aerosols used for comparison. Ambient PM is not uniform with respect to chemical composition, particle size and shape. In most cases, PM does not refer to a single pollutant with a distinct chemical signature, but rather to a highly variable mixture of combustion particles, salts, mineral dust, organic substances and other materials (Hueglin et al., 2005; Putaud et al., 2010). Therefore, suitable standard calibration aerosols do not currently exist.

To date, automated PM instruments which are used for regulatory purposes (e.g. at national air quality monitoring stations) are tested for equivalence with the manual gravimetric reference method in monitoring sites using real ambient air (EC-WG, 2010; Hauck et al., 2004). This requires long and expensive testing campaigns at multiple sites during different times of the year in an attempt to include all representative meteorological conditions and the temporal and spatial variations of the ambient air composition. Portable and cost-effective PM monitors, such as the DustTrak (TSI Inc., USA) and Fidas Frog (Palas, Germany), which are mostly employed for industrial or occupational hygiene surveys (Asbach et al., 2018; Davison et al., 2019; Grzyb and Lenart-Boron, 2019), outdoor (Kingham et al., 2006; Viana et al., 2015; Wallace et al., 2011) and indoor (Chowdhury et al., 2013; Manibusan and Mainelis, 2020; Zhou et al., 2016) air quality investigations, process or emissions monitoring (Al-Attabi et al., 2017; Crilley et al., 2012; Grall et al., 2018; McNamara et al., 2011), and aerosol research studies, do not necessarily go through equivalence testing. Instead, they are often calibrated in the laboratory with simple model aerosols, e.g. with dust or salt particles (Hogrefe et al., 2004; Liu et al., 2017; Papapostolou et al., 2017; Schwab et al., 2004) or dried organic particles, such as sucrose and adipic acid (Zhang et al., 2018). Such model aerosols, however, are only partially representative of ambient air since they fail to account for carbonaceous particles and the complex organic matter, which constitute a considerable mass fraction of airborne particulates (Hueglin et al., 2005; Putaud et al., 2010). Light-scattering PM monitors are very sensitive to the aerosol size distribution, refractive index (i.e. chemistry) and humidity, and research findings suggest that a rigorous calibration with “tailored” aerosols, i.e. aerosols representative of the environment of their intended use, is needed (Jayaratne et al., 2020; McNamara et al., 2011).

The goal of this study was to develop a standardised laboratory-based calibration procedure for automatic PM-measuring instruments under well-controlled and reproducible experimental conditions. Multi-component model aerosols were generated in order to reproduce the main properties of real ambient air in terms of particle size distribution, chemical composition and number/mass concentration, including semi-volatility and hygroscopicity. The properties of ambient air, of course, may differ dramatically from place to place. Here, the main focus was on simulating aerosols encountered in Europe (Putaud et al., 2010), which are dominated by organic matter, inorganic ions (predominantly sulfate and nitrate, and to a lesser extent ammonium), carbonaceous particles (mostly from fossil fuel combustion rather than biomass burning), mineral dust and water.

Apart from the aerosol generation system (detailed below), the new set-up comprises a flow tube homogeniser and a system for reference gravimetric measurements. The facility is very versatile: the total PM mass concentration of the model aerosols can be adjusted in a range from a few micrograms per cubic metre up to about 500 µgm-3; the percentage fraction of each PM constituent can be tuned to simulate different urban, suburban or rural aerosols; and the aerosol temperature and relative humidity can be adjusted to simulate winter or summer-like environmental conditions. As a proof of concept, three different automated PM monitors, the TEOM 1405 (Thermo Scientific, USA), the DustTrak DRX 8533 (TSI Inc., USA) and the Fidas Frog (Palas, Germany), were compared with the reference gravimetric method under three different environmental scenarios. To our knowledge, this is the very first intercomparison involving the Fidas Frog.

Here, we focused on the calibration of the PM monitors' particle quantification, rather than the particle inlet size selection; i.e. the TEOM 1405 unit was calibrated without its PM sampling inlet. The Fidas Frog and DustTrak DRX 8533, which are optical instruments, do not possess any size-selective inlet. The facility could be, however, extended in the future to calibrate PM monitors together with their sampling inlets, if needed. Finally, the facility for generating ambient-like model aerosols presented in this study is not only relevant for the calibration of PM monitors but can find applications in the performance evaluation and quality assurance of other aerosol instruments meant for monitoring ambient, indoor and workplace air as well as in controlled health studies and in vitro toxicology.

The experimental set-up consists of three distinct parts: (i) the generators of the primary aerosols (dust, salts, soot and aged soot); (ii) a flow tube homogeniser for aerosol mixing, including isokinetic sampling probes; and (iii) a system for reference gravimetric measurements. Each part is described in more detail in the following subsections.

Ion chromatography was performed with a Thermo Scientific Dionex™ICS-1500 Ion Chromatography System for analysis of anions and the ICS-2100 model for cations. The systems consist of a liquid eluent, a high-pressure pump, an automatic sample injector, a guard and separator column, an electrolytic suppressor, and a conductivity cell. Before running a sample, the systems were calibrated using a traceable set of calibration standard solutions, which were prepared in-house. The data produced by the range of calibration standard solutions were used to calculate calibration coefficients, which were used to quantitate the sample ions.

Thermo-optical analysis of carbonaceous particles was performed with an OC–EC Analyzer (Lab OC-EC Aerosol Analyzer, Sunset Laboratory Inc., USA), which classified the carbonaceous material as elemental carbon (EC) and organic carbon (OC). The particles were sampled on quartz fibre filters (Advantec, Tokyo, Japan, QR-100, 47 mm). For the analysis, the EUSAAR2 protocol (Cavalli et al., 2010) was modified by extending the last temperature step (850 ∘C) from 80 s in the original protocol to 120 s in order to ensure the complete evolution of carbon (Ess and Vasilatou, 2019). The charring correction for pyrolysed OC was performed by transmittance. OC, EC and TC (total carbon=sum of OC and EC) masses were calculated by the software based on instrument calibration with sucrose solutions.

The elemental composition of the model aerosols was characterised by combining a cascade impactor for PM sampling with Total Reflection X-ray Fluorescence Spectroscopy (TXRF, Bruker TStar S4™, Germany) (Osán et al., 2020). A 13-stage low-pressure cascade impactor (Dekati DLPI 10™, Finland) with particle size range from 30 nm to 10 µm was modified to sample at a rate of 10 Lmin-1 on smooth and clean commercial-grade acrylic discs with 30 mm diameter, suitable for TXRF. In TXRF, the incident X-ray beam hits the disc's surface at the total reflection angle. The fluorescence spectrum is detected perpendicular to the surface and is dominated by the contributions from the deposit, i.e. the sampled particles. This allows for the detection of element masses as low as ≈10 to 100 pg and thus short sampling periods. The measured element quantities, combined with the sampled air volume, provide the particle size-selected element mass concentrations in the aerosol. The discs were prepared with a 50 ng yttrium standard for TXRF calibration.

As an example, the TXRF analysis of model aerosol 1 is shown in Fig. 4. The analysis revealed that the mineral dust particles contain primarily the elements Si and Al, and it was assumed that these are present as oxides SiO2 and Al2O3. The mass-based aerodynamic distribution of the SiO2 particles exhibits a maximum in the range 1–2 µm while the Al2O3 particles are larger (≈7µm). Sulfur (i.e. in the form of sulfate ions) appears predominantly in the submicrometre range (aerodynamic diameter of 30 nm–1 µm), but a second weaker mode is visible at ≈4–7 µm, thus simulating the aerodynamic size distribution of sulfates in ambient air (Wall et al., 1988; Zhuang et al., 1999) reasonably well. The coarse-mode arises most probably from internal mixing of sulfate ions with mineral dust particles. Since nitrates and sulfates were generated with the same method, nitrates are expected to exhibit a similar bimodal size distribution, but this could not be experimentally confirmed since nitrogen is difficult to detect with TXRF spectroscopy. Finally, K+ and Cl- ions appear in the micrometre range (>2µm). It is reasonable to expect that Na+ ions also appear in this size range; however, this could not be investigated by TXRF. By comparing the results of ion chromatography with those of TXRF spectroscopy, there is no evidence of insoluble potassium.

The results of the chemical analysis of the model aerosols with ion chromatography, EC/OC analysis and TXRF spectroscopy are summarised in Table 2 and presented graphically in Fig. 5.

Three PM monitors, a TEOM 1405 (Thermo Scientific, USA), a DustTrak DRX 8533 (TSI Inc., USA) and a Fidas Frog (Palas, Germany), were used in this study. The 1405 TEOM takes continuous direct mass measurements of particulates using a tapered element oscillating microbalance and is considered to be one of the most well established automated instruments for monitoring PM mass concentration at air quality monitoring stations. The DustTrak DRX 8533 and the Fidas Frog aerosol monitors are, unlike TEOM, portable and more cost efficient. These do not measure particle mass directly but record instead the particle number concentration and size distribution using optical techniques, from which they calculate the mass concentration using built-in algorithms.

The PM monitors were exposed to three different model aerosols, which were generated in the laboratory with the facility described in Sect. 2. All three model aerosols were ambient-like mixtures; i.e. they contained inorganic salts, elemental carbon (soot), secondary organic matter, mineral dust and water. The aerosol composition was analysed with the methods described in Sect. 3. The chemical composition of the model aerosols and the environmental conditions during each experiment are listed in Table 2 and depicted schematically in Fig. 5. It can be seen that the mass fractions of the different chemical constituents varied in the range ≈30 %–40 % OM, ≈5 %–15 % EC, ≈7 %–15 % nitrate, ≈ 5 %–15 % sulfate, ≈ 2 %–3 % ammonium, ≈ 10 %–20 % mineral dust and ≈10 %–20 % other materials.

The PM10 mass concentration range (20–40 µgm-3) is typical for urban and suburban regions across Europe. The chemical composition is representative of European aerosols containing carbonaceous particles from fossil fuel combustion (rather than biomass burning); secondary organic matter; mineral dust particles; and inorganic ions such as ammonium, sulfate, nitrate and sodium. The temperature and relative humidity of the aerosols were controlled in the range ≈10–20 ∘C and 50 %–70 %, respectively, to simulate different ambient environmental conditions.

The results of the comparison between the automated PM monitors and the reference gravimetric method are shown in Fig. 6. For the automated PM monitors, which measure continuously and with high time resolution, each data point corresponds to the arithmetic average over a 30 min measurement period. The reference method delivers only one data point, i.e. the average PM10 mass concentration over the whole measurement period, which is illustrated in the graph as a straight solid line and summarised in Table 2. It must be noted that the operating temperature of the TEOM 1405 monitor was set as low as possible, i.e. to 30 ∘C, to minimise losses due to (semi)volatile material (Meyer et al., 2000). For the DustTrak and Fidas Frog the default factory settings were used.

Figure 6a presents the results of the TEOM 1405, Fidas Frog and the reference gravimetric method for model aerosol 1. The results of the DustTrak 8533 are not reported because of a technical problem (obstruction of the aerosol inlet) which compromised the measurement accuracy. The TEOM 1405 seems to agree well with the reference method in the beginning but indicates a decrease of about 15 % in mass concentration at the end of the 4 h measurement. Particle number concentration measurements of the primary aerosols before and after the experiment revealed that the number concentration of the fresh soot particles decreased by about 60 % during the measurement period, whereas the number concentration of the dust, salt and aged soot particles remained largely constant. The reason was a defect in the valve regulating the flow of the fresh soot particles into the homogeniser. The decrease in the aerosol mass concentration recorded by the TEOM is therefore real and can be attributed predominantly to the decreasing number and mass concentration of the uncoated soot particles. Since the concentration of the model aerosol decreased during measurement, the best way to assess the performance of the TEOM 1405 with respect to the reference method is to calculate the 4 h average mass concentration. This amounts to 41.6 µgm-3 (see Table 3), only 3.7 % lower than the reference measurement (43.2 µgm-3).

The fresh soot particles consist mainly of EC and have a geometric mean mobility diameter of about 120 nm, i.e. below the cut-off limit of the Fidas Frog. Indeed, experiments with miniCAST soot showed that the Fidas Frog and DustTrak 8533 failed to detect soot particles of this size. This explains why the Fidas Frog reported a constant mass concentration over the whole measurement period. In Table 3, it can be seen that the Fidas Frog reported an average PM10 mass concentration of 38.8 µgm-3, i.e. -4.4µgm-3 with respect to the reference method. This deviation agrees well with the EC mass concentration of 5.0 µgm-3 (Table 2), as determined with EC/OC analysis. Note that the cut-off curve of optical instruments depends on the refractive index of the particles: the Fidas Frog fails to detect fresh soot particles below ≈200nm but detects a considerable mass fraction of the coated soot and salt particles despite their small size.

The results obtained with model aerosol 2 are displayed in Fig. 6b. Here, the concentration of the aerosol remained constant throughout the measurement period. The Fidas Frog and TEOM 1405 monitors underestimate the mass concentration by 29 % and 14 %, respectively, compared to the reference method, while the DustTrak 8533 overestimates the mass concentration by 50 %. The larger deviation between the TEOM 1405 and the reference method compared to model aerosol 1 results from the winter-like environmental conditions; the temperature of model aerosol 2 was set to 12 ∘C, the relative humidity was set to 70 % and the nitrate content was relatively high (about 15 %) as shown in Table 2. Since the aerosol stream sampled by the TEOM 1405 is heated to 30 ∘C, a fraction of the (semi)volatile components (e.g. nitrate, secondary organic aerosol and water) evolves into the gas phase and is therefore not collected on the filter. These results are in agreement with previous studies reporting that TEOM monitors set at a lower temperature than the standard configuration (50 ∘C) still could lose semivolatile materials (Lee et al., 2005), especially in cooler months (Sofowote et al., 2014; Su et al., 2018).

The large positive deviation of the DustTrak 8533 by a factor of about 1.5 is not surprising. Previous studies have found that different DustTrak models over-recorded PM values by a factor of 1.2–3 (Chung et al., 2001; Grzyb and Lenart-Boron, 2019; Heal et al., 2000; Kingham et al., 2006; Liu et al., 2017; McNamara et al., 2011; Wallace et al., 2011; Yanosky et al., 2002) depending on the aerosol properties. It has been suggested that the “over-estimation is a simple calibration issue in which differences between the optical properties of the manufacturer's factory calibration PM (Arizona Road Dust) and the PM under study explained the uniform relative errors recorded” (Kingham et al., 2006). The results are nevertheless puzzling. Considering that the device fails completely to detect fresh soot and underestimates the amount of aged soot, we would have rather expected to observe a negative deviation with respect to the reference method. In any case, the large range of the positive systematic bias (factor of 1.2–3) highlights the need for source-specific calibration procedures against a reference method.

In the case of Fidas Frog, if the reading of the monitor (21.0 µgm-3; Table 3) is corrected for the undetected mass of fresh soot (3.8 µgm-3; Table 2), then the Fidas Frog still underestimates the mass concentration by ≈15% with respect to the reference method.

The results obtained in the case of model aerosol 3 are illustrated in Fig. 6c. With an average PM10 mass concentration of 19.2 µgm-3, the TEOM 1405 exhibits an excellent agreement with the reference method (19.3 µgm-3; see Table 2). The DustTrak 8533 overestimates the mass concentration by approx. 33 % and thus performs slightly better than in the case of model aerosol 2. Fidas Frog underestimates the mass concentration by about 23 %, or ≈15% after correction for the undetected mass of fresh soot, in agreement with the findings of the experiment with model aerosol 2. As mentioned above, PM monitors based on light scattering, such as the Fidas Frog and the DustTrak, measure particle number concentration and convert this into mass concentration by using a size-dependent particle density function. This function is integrated into the software of the instrument. Deviations may occur if the built-in functions differ substantially from the real density function of the aerosol. Hygroscopic growth of aerosol particles can also lead to considerable measurement artefacts especially when low-cost PM sensors are used (Di Antonio et al., 2018; Crilley et al., 2018). More experiments with ambient-like model aerosols under low and high relative humidity would be needed to define a comprehensive set of calibration factors for these instruments.

In this study, we present the first steps towards the generation of ambient-like model aerosols in the laboratory. A custom-made facility (PALMA) for the stable and reproducible generation of such model aerosols was developed, which presents the following advantages:

The model aerosols are complex, consisting of elemental carbon (fresh soot), soot coated with SOA (aged soot), inorganic ions (such as ammonium, sulfate and nitrate) and mineral dust particles.

As a proof of concept, three different automated PM monitors, the TEOM 1405 (Thermo Scientific, USA), the DustTrak DRX 8533 (TSI Inc., USA) and the Fidas Frog (Palas, Germany), were compared with the reference gravimetric method under three different environmental scenarios. The TEOM 1405, operated at 30 ∘C, agreed very well with the reference gravimetric method in the case of summertime aerosols (21 ∘C) but showed a negative deviation in PM10 mass concentration of ≈15% when the model aerosol was conditioned at 12 ∘C due to losses of semi-volatile material. The Fidas Frog underestimated the PM10 mass concentration by ≈10 %–30 %, whereas the DustTrak 8533 overestimated the PM10 mass concentration by ≈30 %–50 % depending on the aerosol chemical composition and environmental conditions.

Currently, one limitation of the facility is that the model aerosols cannot be conditioned to temperatures lower than 10 ∘C, but this could be improved by thermally insulating the homogeniser (e.g. with black nitrile foam insulation). Moreover, the composition of the model aerosols could be further refined by adding more components, such as metallic particles with the use of a spark-discharge generator, bioaerosols (e.g. with a Sparging Liquid Aerosol Generator – SLAG, CH Technologies, USA) and particles from biomass burning. This last step could pose challenges since the mass output is usually not very stable over time and the physicochemical properties of the aerosol depend heavily on the combustion material, as well as the stove design.

To conclude, the facility presented in this study can be used to generate ambient-like model aerosols for quality assurance testing, intercomparisons of different instruments, and performance evaluation and calibration with respect to PM mass concentration. The same facility could also be used for other PM measurements such as number concentration and absorption properties (e.g. those related to black carbon). The aerosol facility also provides excellent opportunities for basic aerosol research and aerosol health-related studies.