Analysis and classification of individual ambient aerosol particles with field-deployable laser-induced breakdown spectroscopy platform

Abstract

Studying the elemental composition of aerosol particles on a single particle level is of importance when determining the internal and external mixing state of the aerosol population. We present a field-deployable platform for the elemental analysis of ambient single particles using laser-induced breakdown spectroscopy (LIBS). The platform utilizes a wideband spectrometer for simultaneous multi-element detection and size-amplification-aided aerosol charging (SAAC) for efficient particle focusing via a linear electrodynamic quadrupole (LEQ). Carefully designed emission collection system allows minimization of the plasma background emission, which allows us to use short gate delays for increased yield of analyte emission. Performance evaluation with a set of well-defined salt aerosols show excellent capability in determining the relative mass percentage and absolute mass of elements spanning multiple orders of magnitude on a single particle level. Limits of detection for Mg, Na and K were determined to be approximately 2 fg, 40 fg, and 70 fg, respectively. An outdoor aerosol sample was analyzed in the size range of 1–3 µm in diameter, and the particles were classified into distinct categories based on their elemental composition. The maximum analysis speed is about 20 particles per minute and the minimum detectable particle size is, depending on the constituent elements, about 300–800 nm. Emission signals of Al, B, C, Ca, Cu, Fe, K, Mg, Na, Si, and Ti were detected during the measurements of generated or outdoor aerosols.

Graphical abstract

EDITOR:

• Kihong Park

1. Introduction

Ambient air contains a complex mixture of aerosol particles both indoors and outdoors. Aerosols have multiple significant impacts on society: they affect the climate (Ramanathan et al. 2001), cause health issues and premature mortality (Lelieveld et al. 2015), act as disease transmitters (Zhang et al. 2020) and impair visibility (Singh and Dey 2012). Measuring the chemical composition of these particles at a single-particle level is of importance when studying their external and internal mixing state (Riemer et al. 2019; Shen et al. 2019), monitoring for airborne toxic (Lu et al. 2017; Martin et al. 2007) or pathogenic (Li et al. 2023) agents, or determining their sources (Xu et al. 2018). Being able to measure ambient single particle composition is of crucial importance in ice-nucleating particle measurements (Hoose and Möhler 2012; Paramonov et al. 2020).

The elemental and chemical composition of single particles may be resolved using offline or online methods. Offline methods include sensitive tools to analyze the chemical composition up to the spatial level of a single particle (Fletcher et al. 2011). However, microscopy requires sample preparation and transportation and heavy-duty equipment. Obviously, the temporal resolution of aerosol composition is limited with offline techniques, and sample transportation may introduce artifacts to the analysis (Turpin, Huntzicker, and Hering 1994). To analyze single aerosol particle composition online, the most established technique is single-particle mass spectrometry (Riemer et al. 2019; Gemayel et al. 2016; Gard et al. 1997). With it, many organic and inorganic molecules and elements may be detected in real time with portable instruments. However, mass spectrometry requires a good vacuum, expensive equipment, and heavy data analysis.

An alternative for the online elemental analysis of aerosols at a single particle level is laser-induced breakdown spectroscopy (LIBS) (Singh and Thakur 2020; Zhang, Zhang, and Li 2021). LIBS enables the analysis of particles directly from the ambient air without the need for a vacuum or intricate data processing. Thus, multiple systems have been developed to analyze aerosol particles with LIBS in the last decades (Sipich et al. 2022; Heikkilä et al. 2022; Maeng et al. 2017; Hybl et al. 2006; Carranza et al. 2001). As mentioned in our previous paper (Heikkilä et al. 2022), the main challenge in LIBS is to focus the particles into the same volume as the focused pulse laser beam, which may be in the order of a few hundreds of cubic micrometers (Järvinen and Toivonen 2016). To overcome this challenge in the case of ambient aerosol analysis, several approaches have been presented by previous authors. These include using a high laser pulse rate and relying on a probability of hitting a particle every now and then (Carranza et al. 2001), focusing them with sheath air (Park, Cho, and Kwak 2009) and using timed ablation (Maeng et al. 2017; Hybl et al. 2006), or charging and focusing with an electrodynamic balance (Heikkilä et al. 2020). In our approach, we employ a size-amplification aided aerosol charger (SAAC, Heikkilä et al. 2022) to charge the ambient aerosol, followed by focusing them onto a narrow path using a linear electrodynamic quadrupole (LEQ, Hart et al. 2015). This approach enables a flow-through operation for increased particle detection rate while benefiting from precise particle localization of electrodynamic focusing.

In this paper, we present the Spectroscopy Platform for Ambient Aerosol analysis (SPAA), the first portable instrument capable of analyzing single aerosol particles directly from ambient air at low concentrations with a broadband spectral resolution, which enables simultaneous multi-element analysis. The absence of a lower particle concentration limit enables size separation before analysis, enhancing its versatility in aerosol research. With the SPAA, we conducted test series on different generated dust and salt particles and an outdoor sample comprising size-selected outdoor aerosol. Results from analysis of generated salt aerosols with different proportional fractions of Mg, Na and K and with different particle sizes show excellent resolving power in terms of mass-% and absolute mass amount determination. The resolving power of mass-% exceeded the resolution of bulk X-ray fluorescence (XRF) analysis. The particles of outdoor aerosol could be classified into distinguishable categories based on their elementary composition. During the measurements, emission signals for Al, B, C, Ca, Cu, Fe, K, Mg, N, Na, O, Si, and Ti were detected. All the elements were detected using the same optical setup, allowing for simultaneous detection for any mixture of them.

2. Materials and methods

2.1. Spectroscopy platform for ambient aerosol analysis (SPAA)

The principles of the analysis method are described in more detail in our earlier publication (Heikkilä et al. 2022). In brief, the aerosol is first charged with size amplification aided aerosol charging (SAAC), after which it is focused into a narrow beam of particles with a linear electrodynamic quadrupole (LEQ). The focusing enables to guide the particles through a focus spot of a triggered LIBS pulse, generated by a 355 nm Nd:YAG laser (Litron Nano SG 150-10, Litron Lasers Ltd.) The triggering signal is acquired utilizing a 405 CW laser (RLDE405M-100-5, Roithner LaserTechnik GmbH), combined with a photomultiplier tube (H5773-03, Hamamatsu Photonics K.K.). Both lasers are first brought into the same optical path utilizing a dichroic mirror (69-200, Edmund Optics Ltd.) and then tightly focused on to the particle using a 12.7 mm diameter plano-convex lens with a 50 mm focal length (LA4765-A Thorlabs Inc.).

In comparison to earlier developments, this paper presents a field-deployable platform, wideband spectral analysis, and an enhanced sample rate for aerosol analysis. In addition, the optical chamber and the LEQ dimensions have been revised. Schematics of the SPAA and the revised optical arrangement can be found from Figures 1 and 2, respectively. Online supplementary information (SI) Figures S1 and S2 include photographs of the mobile platform.

Figure 1. The Spectroscopy Platform for Ambient Aerosol analysis (SPAA). The spectroscopy system (10.-12.) is built on a commercially available rack case and the accessory equipment (1.-8.) into an aluminum profile rack. Ambient aerosol flows through a condensation growth tube (2.) into an aerosol charger (8.) and then into the measurement chamber (10.). A virtual impactor (9.) before the chamber concentrates the grown particles and acts as a flow splitter. The lids of the rack case have not been drawn into the figure, but the depth dimension of 1.0 m is with the lids in place.

Figure 2. A schematic figure of the optical arrangement inside the chamber for single particle analysis. The aerosol is focused into the vertical axis utilizing the focusing rods. As they reach the focal spot of the lasers, the photomultiplier tube detects the scattering of the 405 nm CW laser and triggers the 355 nm pulse laser. The pulse laser turns the particle and the surrounding gas into plasma, and its emission is collected with a 19-mm focal length concave mirror. An achromatic triplet lens is utilized to focus the emission into an optical fiber bundle, which guides the emission into a 3-channel USB-spectrometer. Additionally, 2 CMOS cameras and a vertically directed 405 CW laser are utilized to monitor the focusing of the particles and the plasma-particle interaction to achieve optimal pulse laser focusing conditions.

The SPAA is built on a 98 cm x 65 cm x 60 cm (length x width x height) rack case (3RR-11U24-25M, SKB cases) with a removable rack. This allows for installation of the optics inside the case to ensure eye-safe operation on field measurements. As shown in Figure 1, the optics are assembled on an optical table (MBH4560/M, Thorlabs Inc.) mounted at the bottom of the rack. The aerosol is guided into the case from an 8-mm inlet hole at the top. As the power source of the pulse laser is mounted inside the case, active cooling is introduced with a fan from the side of the case. An additional rack for the SAAC, electronic equipment and for the operating laptop was built on top of the case from aluminum profile. Total maximum dimensions of the platform are 1.2 m x 1.0 m x 0.6 m and its weight is about 80 kg.

The optical system revision includes a new emission collection system, more comprehensive plasma-particle monitoring, and tilted viewing angles into the focus spot, rather than keeping everything on a single viewing plane (Figure 2). The atomic emission is collected with a concave UV-enhanced aluminum mirror (f = 19 mm, D = 25.4 mm) and focused into a fiber bundle with a custom-made achromatic triplet lens (Thorlabs LA5370, LA5763 & LC4252). The fiber bundle is then divided into 3 branches leading to a multi-channel USB-spectrometer (Multi-Channel AvaSpec-ULS4096CL-EVO, Avantes B.V.) operating at wavelengths of 200–450 nm, 450–700 nm, and 700–1000 nm (Figure 3). The plasma-particle positioning is monitored utilizing 2 CMOS cameras with different viewing angles to ensure focusing on all dimensions. To enhance the particle focusing efficiency (as calculated in Heikkilä et al. 2022), The LEQ dimensions were downsized to 1.0 mm diameter stainless steel rods separated by 7.0 mm (center-to-center) from adjacent rods.

Figure 3. Emission collection optics for single-particle LIBS. The emission signal is collected with a concave mirror (f = 19 mm), focused with an achromatic triplet lens (f = 70 mm) into a fiber bundle, which is used to divide the signal into 3 spectrometers covering a wavelength range of 200–1000 nm. As the emission is collected at the focus spot of the pulse laser, the majority of the plasma emission is not seen by the spectrometer.

The emission collection from the analyte particle is illustrated in Figure 3. As the optical power of the laser exceeds the breakdown value of the gas (in the order of 100 MW), it initiates plasma formation to the focus spot. After the initial breakdown, the plasma starts to absorb the laser irradiance in a cascaded manner, which leads to plasma formation toward the opposite direction of the laser propagation, as illustrated by Thakur and Singh (2007, p. 11). If the particle is at the exact focus spot of the pulse laser and the emission is collected from the same spot, majority of the plasma is not seen by the spectrometer. Thus, the presented arrangement minimizes the background signal of the plasma irradiation, which mainly originates next to the focus spot. Furthermore, this enables to use short gate delay times with minimal interference from the plasma, which is brightest right after the laser pulse. Using short gate delays combined with long integration time is beneficial in capturing most of the emission from different elements within a single measurement, as they have varying optimal emission time windows (Rai and Thakur 2007).

2.2. Spectral data processing

When analyzing initially unknown aerosol particles with LIBS, special consideration needs to be given to the emission signal processing. To reduce the noise from ambient gas emission and the dark noise of the detector, each analyzed spectrum was first subtracted by a background spectrum originating from an average of 30 laser shots without particles present. As each particle only provides one spectrum, conventional multi-spectral averaging methods cannot be directly applied, which leads to a larger noise especially in weak emission peaks. Furthermore, when analyzing initially unknown samples, no prior knowledge about the peak’s wavelengths’ is available.

Considering the boundary conditions in the provided data, we utilized a peak finding algorithm and Gaussian fits to find the peaks and determine their areas and wavelengths in each spectrum. In practice, peaks with prominence of 60 counts were located from each spectrum, and the Gaussian fit was used to define the wavelength and area of the peak after the background and baseline subtraction. The wavelengths of the found peaks were compared to a library of known elements and their corresponding LIBS emission wavelengths. If a wavelength of a peak was within 0.1 nm from a known emission line, its area was used as the emission signal for that element. As most elements divide their emission into multiple energies and our measurement system can operate in a wide range of wavelengths, total peak area per element could be constructed by summing the areas of each peak of the element under inspection or monitoring a single prominent peak with least known interference from other elements. Accepting multiple peaks into the analysis increases the total analyzed peak area, but also increases the noise between subsequent spectra, as that is also multiplied with this approach. A comparison between multipeak and single peak analysis was conducted for several elements and no significant increase in resolution power was obtained when using multiple peaks. As per the results of the comparison, further data analysis was proceeded with only a single emission line per element. The analyzed elements and their wavelengths can be found from SI Table 1. Traditionally in LIBS studies, a Lorentzian fit is used as a fit function, but when processing the data acquired with the SPAA, it became quickly evident that Gaussian fits provide better results in terms of regression analysis parameters and fit stability.

2.3. Laser energy and plasma location optimization

The energy of the pulse laser and the spatial location of plasma with respect to the particle under analysis were investigated to maximize the emission signal. Two different aerosols (Salt solution of MgSO₄, NaCl and KCl; kaolinite dust (Al₂Si₂O₅(OH)₄)) including five analyte elements (Mg, Na, K, Al, and Si) were utilized to optimize the signal on a wide spectral range and with different types of elements.

Figure 4 presents the measurement setup in detail for both salt aerosol and dry aerosol analysis. The same measurement scheme was used in the measurements presented in the following subchapters, excluding the ambient aerosol analysis, in which the sample aerosol was outdoor air. The samples of the salt solution were size-selected by a Differential Mobility Analyzer (DMA 3081 A, TSI Inc.) to a size of 1 µm and the kaolinite aerosol was sampled at a size of 1.25 µm. The size of the aerosol population was monitored utilizing an Aerodynamic Particle Sizer (APS 3321, TSI Inc.) after the size classification, and the aerodynamic sizes were found to be approximately 20% larger than the classification size of the DMA. The geometric standard deviations were approximately 1.2 for both samples. Only a small amount of multiply charged particles were detected, as the original size distributions had smaller median diameters than the classification size and thus the relative abundance of larger particles was low. The salt aerosol consisted of the same mixture as in sample number 1 in Table 1, i.e., the proportional masses of the analyzed elements were Mg: 9.6% Na: 38.5% and K: 51.9% in a dry particle.

Figure 4. Measurement setup during the optimization and performance analysis experiments. Depending on the aerosol, one generation branch (Dry or Droplet generation, denoted with dashed line box) was in use at a time. After the generation, the aerosol was classified with a DMA and the particle size and concentration was monitored with an APS after classification.

Table 1. The samples used in the resolving power analysis.

Sample no.	Mass of salt in H2O solution (g/300 ml)			Proportional mass in dry particle
	MgSO4	NaCl	KCl	Mg	Na	K
1	4.99	5.02	5.07	0.120	0.375	0.505
2	0.60	1.51	14.08	0.010	0.079	0.911
3	0.30	10.38	4.59	0.006	0.626	0.369
4	1.20	5.40	8.77	0.022	0.309	0.669

2.4. Multielement aerosol analysis

Several test aerosols were generated to characterize simultaneous multi-element analysis. An approximation of proportional number content (n-%) of measurable elements in single particles was derived utilizing the Boltzmann plot method (Bousquet et al. 2023) and the NIST LIBS database (Kramida, Olsen, and Ralchenko 2019). In the following subsections the tests are described in more detail.

2.4.1. Estimation of number share of elements in generated aerosols

In addition to the bubble generated salt samples, a brush aerosol generator (RGB-1000, Palas GmbH) was utilized to disperse kaolinite, Arizona test dust (ATD, ISO 12103-1, Powder Technology Inc.) and ISO12103 PT1 (ISO 12103-1:2016 A1, Particle Technology Ltd.) test dust. Their spectral footprints were analyzed and compared to each other. As the ATD contains calcium, its emission peaks were used to evaluate the electron temperature of the plasma with the Boltzmann plot method. This temperature was then utilized to acquire data from the NIST LIBS database to calculate initial factors for peak area normalization to estimate the number shares (n-%) of detected elements within single particles. Furthermore, data from the salt particle analysis presented in the next subchapter was used to evaluate optical throughput for the 3 spectrometers. This throughput data was combined with the initial factors acquired from the NIST database for the number share analysis.

2.4.2. Analysis performance with well-defined salt particles

For a more in-depth study of the resolving power considering relative and absolute mass composition of an element per particle, several salt solutions with magnesium sulfate, sodium chloride and potassium chloride were prepared with different relative concentrations. Table 1 includes the concentrations of the salts and calculations of their cations’ (Mg, Na, and K) relative masses in the final aerosol particle. The initial hydration state of MgSO₄ used in the solutions was found to be 4 by measuring the weight loss of a sample after heating it to over 250 °C.

The elements were chosen to provide spectral signal simultaneously into a wide spectral range, as Mg, Na and K have the most prominent peaks at around 280, 590, and 770 nm, respectively. The selection rules and the differences in electron structures of the elements cause some electron transitions to happen more likely than others. Moreover, differences in the plasma temperature and free electron density cause some excited states to be more populated than others, causing the transition probabilities to be dependent on the environment, i.e., the parameters of the plasma. Thus, the atomic emission line intensities within and between the elements may vary many orders of magnitude (Thakur 2007). This variation can be readily observed from Figure 5a. Thus, the solutions were prepared to reach a comparable signal for each of the elements (Mg, Na, K), which leads to Mg having by far the lowest concentration in the particles.

Figure 5. Results from the salt particle analysis. In panel (a), the median peak areas of the LIBS measurements are compared to the original masses of the elements in single particles. In (b), the peak areas have been calibrated using the calibration sample (1.0 µm, sample 4, Table 1) to directly estimate the mass of an element in a single particle. In (a) and (b), the mass in the horizontal axis has been calculated from the original salt solution and the particle size. In panel (c) and (d), the median values of measured fractions of mass in single particles are compared to the original values derived from the salt solutions. Error bars in all the panels indicate the IQR of either the measured or calculated values. The IQR of the original mass of the particles originates from the size distribution measured by the APS. Different colors indicate different elements and different markers indicate different particle sizes. The exceptions are the asterisk and black ‘+’ markers, which indicate the XRF and the calibration measurements, respectively. The particle size has been left out from panel (d) for clarity.

Four particle diameters (0.8, 1.0, 1.25, and 1.5 µm) were used to be able to assess the sensitivity as a function of absolute element mass per particle. Each particle sample-size combination measurement consists of ca. 200 spectra. In addition, each sample underwent bulk X-ray Fluorescence (XRF, M4 Tornado Plus, Bruker Nano GmbH) analysis for comparison. The XRF sample was prepared by drying a few centiliters of the solution on a petri dish and the analysis was conducted on crushed crystals within a 10 mm x 8 mm area. In general, with XRF, limits of detection in the ppm range are achievable, but the sensitivity decreases with lighter elements (Marguí, Queralt, and de Almeida 2022). The values are comparable to the limits achievable with LIBS (Fabre et al. 2018).

2.5. Ambient aerosol analysis and classification

A set of ambient aerosol particles was measured, sampled directly from outdoor air, using a sampling tube through the laboratory window. The sampling site is located in the campus area of Tampere University, in a sub-urban area of Hervanta in Tampere city (61°27'00.0"N 23°51'25.0"E, ∼140 m above sea level). The wind was blowing from the vast forest area from the south. A virtual impactor (VI) with a dp50 value of 1 µm was utilized to concentrate the supermicron particles. After the VI, the aerosol was classified using a DMA (Model 8021, TSI Inc.) set to sample 1.2 µm particles. Due to a high probability of multiple charging in the neutralizer of the DMA, the size distribution was monitored with an APS (3321, TSI Inc.) and verified to stay between 1.0 and 3.0 µm. A total of 313 particles were analyzed and classified into 5 categories based on their elemental composition. The example sample was measured on 2.8.2023 during 15.10–15.50.

3. Results

3.1. Laser energy and plasma location optimization

Figure 6 presents the main results of the pulse laser energy optimization. Considering pulse laser spatial focusing, the best spectral consistency and signal was achieved when the laser was directly focused into the particle, as illustrated in Figure 3. This means, in practice, that the particle is located at the very edge of the plasma toward the direction of beam propagation, as the plasma forms from the focus spot into the opposite direction. Plasma-particle focusing has been studied before with similar results by Järvinen and Toivonen (2016) for Pb analysis. However, with substantially higher energies, other researchers have demonstrated increase in some molecular emission signals if the plasma is focused slightly off the particle and let spread into it (Purohit, Fortes, and Laserna 2017).

Figure 6. Results of the energy optimization measurements for Mg, Na, K, Si, and Al. Each marker represents at least 200 measurements. The energy was optimized in terms of median peak area (a) and signal variance, defined as IQR/Peak area (b).

Figure 6a includes the median peak area as a function of energy for peaks of Mg, Na, K, Si, and Al. Each marker represents at least 200 single particle spectra. The (b) panel includes a measure for the relative variation between single particle spectra, defined as the median area value from panel (a) divided by the interquartile range (IQR) of all the measured peak areas for the energy/element combination. Robust indicators such as median and IQR were chosen instead of mean and standard deviation because multiple charging of the particles cause a few significant outliers to the data. The outliers are still relatively rare, as the initial size distribution has a median in much smaller sizes than the setpoint of the DMA.

As describer earlier, Mg, Na, and K were measured from 1.0 µm particles containing a salt solution sample 1 (Table 1), and Si and Al were measured from dry-generated 1.25 µm kaolinite (Al₂Si₂O₅(OH)₄) particles. Most of them had the highest signal output in the regime close to 4 mJ. The exception was magnesium, which peaked a few millijoules higher. The relative variation of single particle spectra was minimized around 5 mJ for all the elements. To maximize the signal output and minimize the inter-particle signal fluctuations, pulse laser energy of 4.5 mJ were used for the upcoming measurements of the following sections.

As described in Section 2.4, the LIBS emission intensities may vary multiple orders of magnitude for similar analyte masses. This is mostly noticeable by the variation between magnesium and the rest of the elements in Figure 6a: even though the proportional mass of magnesium in a dried particle was less than 10% compared to the other two (Na and K) alkalis, roughly an order of magnitude higher peak areas were measured. Furthermore, the masses of Si and Al in a single particle were multiple times higher than the masses of Na and K, yet similar peak areas were measured for all of them, even when integrating through the counts from all their emission peaks. The anions of the salts (S, Cl) were not observed at all.

3.2. Multielement aerosol analysis

3.2.1. Spectra from generated particles and n-% normalization

Example spectra from generated aerosols and of the background measurement are presented in Figure 7. The x-axis is divided into relevant areas of the spectra, showing ranges of 220–320 nm, 385–405 nm, 580–600 nm, and 750–790 nm. The samples consist of salt sample 4 (Table 1), Arizona test dust (ATD), kaolin dust and ISO test dust. Most of the inspected elements have their emission peaks in the range of spectrometer 1, but sodium and potassium are only visible at spectrometers 2 and 3, respectively.

Figure 7. Example spectra from single particles acquired from generated test aerosols (a)–(b) and an average spectrum of the background measurement. The background is subtracted from each single particle spectrum.

In the background measurement, oxygen and nitrogen of ambient air are clearly visible. Furthermore, even with the subtraction of background from the spectra of the particles, their emission lines are still visible. As the emission collection optics are focused – with the particle – on the very edge of the plasma along the beam propagation direction, the presence of the particle causes the plasma to spread further along the path, causing more emission from the gas to be detected. Some particles are also rich in oxygen, but by comparing the peak areas of oxygen and nitrogen, no visible change was observed for different particle types or for the background. Thus, N and O were not used in the particle spectra analysis.

The Boltzmann plot method was utilized to determine the apparent plasma temperature. For this, singly ionized calcium emission lines of ATD sample were used, as the method requires two or more lines with a sufficient (a couple of eV) spread of upper level energies. With the method, an average plasma temperature of approximately 11600 K (corresponding to 1.0 eV/k_B in atomic units) was derived from the results. This temperature was approximated to be valid for all the other elements as well. With the temperature, the NIST LIBS database was utilized to acquire reference spectra for all the elements and the same spectral analysis was conducted to all of them as was conducted to the measurement spectra. These peak areas were then used as normalization factors to conduct an approximation of the relative number concentration of elements within single particles. Furthermore, correction factors for different spectrometers were added to the approximation from the results of the next section. A few example compositions normalized with the method are presented in Figure 8 for all the four example aerosols. In the figure, each spectrum number in the horizontal axis represents a single particle spectrum. As can be seen, the salt aerosol, kaolinite, and ISO Test dust yield relatively homogenous composition, but the ATD has more heterogeneity between the particles. A similar observation of heterogeneous distribution of minerals in single ATD particles has also been noted by Vlasenko et al. (2005).

Figure 8. Example compositions calculated for the test aerosols using the Boltzmann method and NIST LIBS database normalization. Each stack of bars represents one particle, and the colors (patterns) indicate different elements. The height of a segment indicates the relative amount of an element in a particle.

3.2.2. Analysis performance with well-defined analyte particles

Well-defined salt solutions with different compositions of Mg, Na and K were aerosolized and size-selected with multiple particle sizes. The particles were analyzed with the SPAA and compared to the original solute masses and bulk material XRF analysis. Figure 5 includes the results from all the particles over all the sizes (0.8 µm − 1.5 µm, n = 3465). Each marker includes spectra from at least 200 measurements, excluding the XRF measurements, which were acquired from a bulk sample.

In Figure 5a, the measured median values of peak areas are compared to the masses of elements calculated from the original salt solution and of the particle size. The error bars on the y-axis include the interquartile range (IQR) of the peak areas for the particle size/sample combination and on the x-axis the IQR of the masses of the analyte elements in single particles. The masses of elements and their IQRs in single aerosol particles are derived from the original salt concentrations, the median particle size of the DMA and the deviation of the particle diameter measured by the APS. Particle size of 1.0 µm of sample 4 was chosen as a calibration sample, from which a simple linear calibration factor ${\chi }_{\alpha }$ was derived for each element $\alpha ,$ relating the peak area $I_{\alpha }$ to the analyte mass $m_{\alpha }$ as ${\chi }_{\alpha }=I_{\alpha }/m_{\alpha }.$ The calibration sample is marked in all the panels (a)–(d) for all the elements.

Using the calibration sample, estimations of analyte mass/particle could be calculated for the other samples. These estimations are shown in Figure 5b. The vertical error bars present the IQR of the measured masses and the horizontal error bars are again the IQR of the particle mass in the aerosol population. As can be seen, the linear calibration predicts the masses quite well over a wide range of masses. Furthermore, most of the variance in the mass measurement may be interpreted to originate from the variance in the aerosol distribution, as the error bars have almost identical lengths in both axes. Other significant source of variance in the measured mass is caused by the fluctuation of the pulse laser energy, which causes variation in the electron temperature and in the spread of the emitting analyte gas. Different electron temperatures directly affect the intensity of emission lines as predicted by the Boltzmann distribution and the spreading of the emitting gas causes a decrease in optical throughput, as the image size exceeds the spectrometer slit width.

The same calibration sample (sample 4, particle size 1.0 µm) was also used to relate the relative fractions of peak areas to the relative fractions of analyte elements with a linear constant defined with $\frac{I_{\alpha }}{\Sigma I}={\eta }_{\alpha }\frac{m_{\alpha }}{\Sigma m}.$ Using the constant ${\eta }_{\alpha },$ the measured mass fractions are shown as a function of the actual mass fractions of the solutions in Figure 5c. These values were also divided with the actual mass fraction and plotted on a linear scale in panel (d). Furthermore, the results from the XRF analysis are shown in panels (c) and (d). As can be seen from the panel (d), the LIBS measurements mostly stay within 25% error in determining the mass fractions of known elements in the measured particles, even in the < 1% concentration range for magnesium. The XRF measurements have over 50% error in concentrations below 15%, but for the > 30% concentrations, both measurements yield good results.

An approximation of the limit of detection (LOD) was evaluated for elementary mass as the smallest mass that produced a signal for over 75% of the single particle spectra. These values were found to be about 2 fg, 40 fg and 70 fg for Mg, Na and K, respectively. However, for K, no values were measured below this definition and thus the LOD is probably smaller. Furthermore, the densities of the particles were calculated by comparing the particle size values measured by the APS and the classification size of the DMA and were found to be approximately 1.3 times smaller than excepted from the salt compositions. Thus, the particles probably still contained some water or hollow cavities, as has been reported before by Cheng, Blanchard, and Cipriano (1988). Thus, the LOD values are most likely somewhat smaller than presented, but we have adopted a conservative approach and report them without correction for this effect.

3.3. Ambient aerosol analysis and classification

A total of 313 spectra were acquired during the 40-min sampling period. A total of 13% of shots were misses or included undetectable elements, and the rest could be classified and underwent the same n-% normalization routine as described in Section 3.2. With the sample data, it became evident that a few elemental footprints are constantly repeated. Thus, these spectra were classified into categories presented in Figures 9 and 10. The categories are “Salt particles” (Na, Mg, Ca), “Biological particles” (C, trace amount of K and Ca), “Dust particles” (Si, Al, Fe) and “Carbonaceous particles” (only C). Thorough criteria for classification into a certain group are presented in Figure 11. The criteria were chosen to automatically categorize the particles while avoiding single particles being classified into multiple groups and maintaining compositional homogeneity within the groups. The categories are based on the elemental compositions of different particle types presented in the literature: The “Salt particles”—category is based on Seinfeld and Pandis (2016, p. 354), presenting the composition of sea salt particles. The” Biological particles”—category is based on measurements by Heldal, Norland, and Tumyr (1985) and Saari et al. (2016) of individual bacterial cells. The “Dust particles”—category is based on the abundance of silicon in the particles accompanied by other crustal elements, as presented by Usher, Michel, and Grassian (2003).

Figure 9. Normalized example compositions from the outdoors aerosol measurements. As marked into the figure, the particles are divided into four main categories due to their spectral footprints. Each bar stack represents a single particle, and the colors (patterns) indicate measured elements.

Figure 10. Example single particle spectra from the four categories of the outdoor aerosol measurement. Due to wideband operation, multiple elements could be resolved simultaneously from each particle, which enables the particle classification.

Figure 11. The relative fractions of the particle categories in the outdoor aerosol sample. Most particles were classified into the Salt category (28%), followed by Biological and Dust (20% each). 7% Included only carbon and 11% were unclassified. The unclassified particles consisted mainly of Na-rich (6%) and Ca-rich (4%) particles.

In Figure 9, the first few spectra from each category are shown as a similar n-% bar plot as in Figure 8, utilizing the normalization method described in chapter 3.2. The Boltzmann plot method was applied for calcium-rich particles and yielded the same temperature of 11600 K (1.0 eV/k_B) as in the earlier analysis for generated aerosols. A single example spectrum from each group is presented in Figure 10.

An interesting finding about the spectra of carbon-containing particles is the prominence of carbon’s emission line at 229.7 nm (Figures 10b and c). It originates from a doubly ionized state and should thus appear at much higher plasma temperatures than calculated with the Boltzmann plot method. The discrepancy between the observed and excepted emission is likely due to the plasma not being in LTE conditions, as the emission collection period (gate width) is long (10 µs), and the gate delay is short (100 ns). Within the gate width, the plasma undergoes a transition through high temperatures and electron densities into lower ones. Thus, the peaks of carbon from different ionized states appear at different times, but as they are integrated into a single spectrum, they seem to conflict with the apparent temperature acquired using the Ca emission lines. However, carbon was the only element showing such evident discrepancy between the apparent temperature and the emission line profile acquired from NIST LIBS database. For the normalization of C, we decided to use a different temperature (21,000 K, i.e., 1.85 eV/k_B) for reference spectrum acquisition to better match the perceived and expected line profile.

When comparing the biological and carbonaceous spectra in Figure 9, the difference seems to be minimal. However, from Figure 10, the difference is obvious. This is due to the emission intensity of carbon is still, with the elevated temperature, expected to be at least an order of magnitude smaller than for the others. Thus, even with clearly visible peaks in the raw data spectrum, the n-% of Mg, K and Ca are normalized to be small in the biological particle group.

Figure 11 illustrates the proportional fractions of the particle types defined earlier. The most abundant types were Salt (28%), Dust (20%) and Biological (20%) particles. 8% of the particles were only C. 11% of the particles were unclassified, of which particles with Na-rich and Ca-rich composition consisted of 10% in total. 13% of the spectra were classified as missed, meaning that no known elements were detected. It should be noted that the aerosol was size separated prior to the analysis with a virtual impactor with a cut point at 1 µm and a DMA sampling at 1 µm. Thus, the sample is not representative of the total aerosol, but gives insight to the local aerosol population in the 1 µm–3 µm size range, as measured with the APS. Furthermore, as the analysis is dependent on successful charging of the aerosols, which is dependent on the condensation growth with water, some bias might be caused by the growth efficiency. For example, hydrophilic aerosols such as salts grow relatively easily, and hydrophobic aerosols such as pure DEHS particles don’t grow at all (Asbach et al. 2017). However, few atmospheric particles are as hydrophobic as the DEHS particles.

The classification categories are based on prior knowledge of aerosol particle types (Seinfeld and Pandis 2016; Usher, Michel, and Grassian 2003; Heldal, Norland, and Tumyr 1985; Saari et al. 2016), but have not been verified with, for example, microscopy imaging. It should be noted that this is the first study of wideband LIBS for ambient outdoors aerosol classification, thus there might be discrepancies in the classification. However, there are several arguments that might justify these categories. For the salt particles, utilizing the HYSPLIT data from NOAA archive (Stein et al. 2015) to trace back the trajectory of the local outdoors airmass, it was noted that it had traveled above the Baltic Sea within a few hours before arriving to the measurement location. As the sea is a common source of salt aerosols, it could act as an obvious source for the abundance of the category. The dust particles are easily classified due to the abundance of Si, Al, and Fe (Usher, Michel, and Grassian 2003), and during summertime, the source could be either local or remote. Biological particles are expected to be present during summer, as the biological activity in the surrounding nature is high. Furthermore, previous studies have presented the presence of K and Ca in bacteria. The authors acknowledge this as the most uncertain classification, as particulate carbon combined with trace elements might originate from other sources as well, such as biomass or fossil fuel burning or lubricant oils. However, the abundance of Ca and K combined with the lack of other metallic compounds in the particles would suggest that they are of biological origin (Park and Kim 2005; Yatkin and Bayram 2007; Jahn et al. 2021).

4. Conclusions

In this study, the Spectroscopy Platform for Ambient Aerosol analysis (SPAA) was presented. Its capability for simultaneous multi-element analysis of single aerosol particles sampled directly from the surrounding air was demonstrated and its analytical performance was evaluated with test aerosols and a sample of outdoors aerosol. The transportability was confirmed with field campaigns with results being published later. The sampling rate was found to be in the order of 10 particles/min and the optimum concentration in the order of 1 particle/ccm. The maximum concentration for successful particle analysis is approximately 10 particles/cm3. Above this, the aerosol must be diluted. The limit is due to the electric interaction between particles interfering successful focusing, leading to missed laser pulses. This phenomenon is discussed and demonstrated in more detail in Heikkilä et al. (2022). When comparing the sampling speed of ca. 10⁴ analyzed particles/day to other methods gathered by Riemer et al. (2019), the SPAA is faster than usual microscopic single particle methods (10¹–10⁴ particles/day) but slower than many single particle mass spectrometers (10²–10⁶ particles/day). Thus, an important future research area would be to further increase the analysis speed to achieve better sampling statistics, comparable to those of the single-particle mass spectrometers.

The liner electrodynamic quadrupole (LEQ) focusing and the optical system for conducting single-particle LIBS was revised from our previous work. With the revised LEQ, the particles are focused more efficiently without compromising the optical throughput, as the focusing rods are smaller and their relative distance is downsized. The new optical chamber enables for accurate monitoring of the focusing of the trigger laser, the plasma, and the particles from all dimensions. This is of great practical importance to be able to line the lasers and the emission collection system in the most efficient manner. Furthermore, the emission collection system enables to use short gate delays in the LIBS analysis, which was experimentally verified to increase the throughput of the elementary radiation. As the emission is collected from a perpendicular angle of the plasma at its very edge, only a minimal amount of the plasma irradiation ends up to the spectrometer, thus it does not shade and absorb the emission from the analyte particle.

The SPAA showed excellent performance in measuring the proportional fractions of the generated salt aerosol, exceeding the resolution power of the laboratory XRF analysis. Furthermore, with a simple linear calibration, it could resolve the mass content in single particles in a decent manner, spanning over multiple orders of magnitude in elementary mass. The detection limits for Mg, Na, and K were also evaluated experimentally, and were found to be about 2 fg, 40 fg, and 70 fg, respectively. The wideband spectral resolution enables simultaneous observation of multiple elements on a single-particle basis. Thus, the particles of the outdoor aerosol sample could be divided into distinct categories based on their elementary composition. The sample consisted of 313 particles sampled during a 40-min period.

This research paves the way for multiple directions of the aerosol-LIBS instrument development and LIBS research in general. The optical chamber enables the simultaneous addition of other spectroscopic methods to the system, such as fluorescence and Raman (Saari et al. 2016; Sivaprakasam, Hart, and Eversole 2017). These methods could enable the analysis of the molecular structure and the biological origin of the particles. Furthermore, adding research effort to minimizing the particle losses from the condensation growth assisted aerosol charger (SAAC) system and pre-focusing the aerosol flow into the LEQ chamber could dramatically increase the sample rate from even smaller concentrations. Moreover, the unique nature of the emission collection enables the use of short or even zero gate delays, which is not yet a widely studied region of LIBS analysis.

Acknowledgements

This work made use of Tampere Microscopy Center facilities at Tampere University. CAD-figures of Thorlabs components were reproduced with the permission of Thorlabs, Inc.

Data availability statement

All the reported data is displayed in this article.

Disclosure statement

The authors report that there are no competing interests to declare.

Additional information

Funding

This work was supported by the Maj and Tor Nessling Foundation and the Research Council of Finland flagships ACCC (decision No.’s 337551, 357903) and PREIN (346511).