Stratospheric particles were sampled on board of the NASA ER-2 aircraft during the SAGE III Ozone loss and validation experiment (SOLVE), which was conducted in January–March 2000 in Kiruna (Sweden). The Multi-Sample Aerosol Collection System (MACS) (Kojima et al., 2004), a thin-plate low-pressure impactor, was used for sampling. The particles were deposited on TEM copper (Cu) grids covered with a formvar film. MACS is designed to collect up to 23 samples per flight. The first sample is not exposed to flow and serves as a blank sample. The MACS was designed to sample and transmit all particles in the diameter range from 20 to 1000 nm to the impactor. Particles larger than approximately 20 nm are collected on the impactor in the pressure range shown in Table 1. In situ measurements of aerosol abundance were obtained with a condensation particle counter (Wilson et al., 1983) simultaneously with the samples. Sampling date, flight characteristics, potential vorticity (PV), pressure altitude, pressure, potential temperature and ambient aerosol number mixing ratio (particles per milligram of air) for the analyzed samples are shown in Table 1. In situ detection of nitrogen oxides on particles (Fahey et al., 2001) was used to determine if samples were collected when the aircraft was in polar stratospheric clouds (PSCs) containing nitrogen. Measurements of N2O provide insight into the residence time of the air parcel in the stratosphere, called age of air (Wilson et al., 2008). These values are also shown in Table 1.

The meteorological conditions of the early winter (November–January) northern hemispheric polar vortex of 1999/2000 are described in detail by Manney and Sabutis (2000). The early winter (November–January) northern hemispheric polar vortex of 1999/2000 had much lower averaged temperatures compared to any previously observed Arctic winter. The vortex was weaker than the early winter polar vortices of the previous years. It was discontinuous in the middle of December, with a large extent in the upper and small extent in the lower stratosphere. During the period of airborne measurement operations, from mid-January on, the vortex evolved to be continuous and stable until mid-March (Greenblatt et al., 2002). Jost et al. (2002) describe anomalous single mixing events occurring during that time at the potential temperature (Θ) range of 350–500 K. These events are probably the result of mixing between deep vortex and extra vortex air.

The stratospheric particle samples (deposited on TEM grids) taken within the polar vortex were packed into single plastic boxes and stored in a desiccator prior to analysis, starting in 2014. Based on the investigation of blank samples, contamination of the samples during the time of storage (e.g., by vapors from the plastic boxes) can be excluded. Furthermore, a change in particle morphology and nanostructure is not expected, since the particles found are either amorphous or show very little ordering. This conclusion is based on the fact that graphitization of carbonaceous material is an irreversible process (Diessel et al., 1978; Itaya, 1981; Pesquera and Velasco, 1988). In any case, it should be kept in mind that other parameters (chemical composition, mixing state) may be changed to a variable extent by aging. In total, 122 samples from 15 sampling days were collected.

A total of 4202 particles (3872 TEM; 330 SEM) from 11 samples were investigated by transmission and scanning electron microscopy. The samples were selected according to meteorological conditions and suitability for electron microscopy (i.e., substrate area covered by particles). Table 2 gives an overview of how many particles were investigated with which method.

The objective of this study is the detailed characterization of refractory stratospheric particles. Similar to Ebert et al. (2016), we have classified all particles that are stable (no visible morphological change) under the high-vacuum conditions and electron beam excitation in the SEM and TEM as refractory.

The size, morphology, mixing state, nanostructure and chemical composition of 60 refractory particles per sample were studied by TEM using a Philips CM20 instrument (FEI, Eindhoven, the Netherlands) operated at 200 kV electron accelerating voltage. The chemical composition of the particles (all elements with an atomic number Z≥5) was determined by EDX using a silicon-drift X-ray detector (Oxford Instruments, Oxfordshire, United Kingdom) and a measurement time of 20 s. Particle size and graphene sheet separation distance were analyzed by the ImageJ software (1.48v; Rasband, 1997–2016). Element distribution images were acquired with a JEOL JEM 2100F (JEOL, Tokyo, Japan) operated in scanning transmission electron microscopy (STEM) mode at an electron acceleration voltage of 200 kV. The instrument is equipped with the same type of EDX detector as the Philips CM20 instrument.

In order to detect elements present at low abundance, additional 30 refractory particles per sample were analyzed by SEM using a Quanta 200 FEG instrument (FEI, Eindhoven, the Netherlands) equipped with an EDX detector (EDAX, Tilburg, the Netherlands) operated at 15 kV electron accelerating voltage. X-ray spectra were accumulated over 15 min per particle to obtain a low detection limit. These long exposures were not feasible in the TEM due to its higher beam energy and resulting particle evaporation. Element concentrations were obtained from the X-ray count rates by applying a “standard-less” ZAF correction. Detection limits of element / carbon ratios (at % / at %) for the long-time measurements are as follows: O/C = 0.0034; Si/C = 0.0010; S/C = 0.0008; Cr/C = 0.0008; Fe/C = 0.0009; and Ni/C = 0.0011.

The particles were studied by TEM and SEM without coating.

Potential contamination of the samples was checked by investigating blanks (samples transported in the MACS but not exposed to stratospheric air flowing through the impactor orifice) for each sampling day. A few titanium (Ti) and zinc (Zn) oxide particles, as well as few pure C particles were encountered on the blank samples. They look similar to some carbonaceous particles being described as contaminants on TEM grids (Harris et al., 2001). However, these particles are different in both size and morphology compared to the carbonaceous particles observed in the impaction spot of the samples.

In order to verify that the small amounts of Fe, chromium (Cr) and nickel (Ni) detected during long-term SEM measurements are not artifacts from the substrate, five points on each substrate far away from particles were analyzed. These three elements were not detected in the measurements of the clean substrates.

Another artifact can result from scattered radiation within the SEM. This could lead to the detection of chemical elements in the vacuum chamber's housing material. To test this possibility, one sample was measured at larger sample chamber pressures (5×10-3, 200 and 500 Pa) which result in increased scattering of beam electrons in the sample chamber. According to Stokes (2008), scattering varies between 40 and 80 % at 200 Pa and 70 and 98 % at 500 Pa. Fe, Cr, and Ni did not show increased concentrations at higher pressure, but rather their count rates decreased. The small Fe, Cr and Ni concentrations detected in individual particles were not the result of stray radiation.

The elemental composition of the particles was determined by EDX in TEM (n=529) as well as SEM (n=330). Due to the small size of the particles, TEM is the preferred method of analysis. As mentioned above, particles were additionally analyzed by SEM. Both measurements led to small but systematic differences in the ratios of O, Si, S, Mg (magnesium), Fe and Al to C. For example, the median O/C value is 0.236 for TEM and 0.117 for SEM (Figs. S1 and S2 of the Supplement). For all elements TEM-EDX yielded somewhat higher elemental ratios relative to carbon than SEM-EDX. These differences most likely result from differences in the detectors, such as thickness of detector windows, and the different acceleration voltages (15 kV in SEM versus 200 kV in TEM). However, as the differences are small (Fig. S2 of the Supplement), our conclusions are independent of the method used. SEM-EDX data are reported here since the counting time on each particle was much longer in SEM (15 min) than in TEM (20 s), leading to higher precision as well as lower detection limits. The much lower detection limits of SEM-EDX are important for source identification as minor elements may serve as fingerprint for several anthropogenic and natural sources. Thus, the SEM-EDX data are preferred despite the lower number of particles investigated with this technique. In addition, the chemical composition of particles could not be analyzed by TEM-EDX on two samples (labeled as G and K) due to the inappropriate position of the impaction spot on the substrate (too close to the Cu grid, leading to a very high count rate for Cu).

In order to assess the mixing state of the refractory particles, additional image analysis was performed in TEM. For this purpose image series before and after TEM analysis were prepared.

All particles which showed no signs of destruction or morphological change were defined as refractory. Particles which changed under the electron beam were deemed volatile, allowing quantification of the fraction of aerosol which is volatile. In total 3316 particles were analyzed by this method.

To ensure unbiased results, the individual images as well as particles for EDX analysis were randomly taken in inner and outer areas of the impaction spot.

Censored box plots show data taking into account the fraction of values below detection limit. Lower and upper quartiles appear as a box, and minimum and maximum values as whiskers.

The differences in element ratios between samples collected inside and outside the vortex were tested with the generalized Wilcoxon test (Helsel, 2012) applying a significance level of 5 %. Furthermore, the differences in size, projected area diameter and element ratios between the various samples were tested with the Kruskal–Wallis rank sum test (uncensored data) and the generalized Wilcoxon test (censored data). In all individual tests, a significance level of 5 % was applied.

The detection limits for EDX data were calculated from counting statistics (background counts plus 3 times standard deviation of background counts).

All statistical calculations were performed with R (version 3.3.0; R Core Team, 2016) and using the contributed package NADA (version 1.5–6; Lee, 2013).

The size distribution of the refractory carbonaceous particles is indicated in Fig. 3. Approximately 98 % of the particles have an equivalent projected area diameter (Dpa) below 500 nm (range 20–830 nm). The size of the particles slightly increases with time during the campaign; sample J shows the largest median particle sizes.

Besides C, the refractory carbonaceous particles always contain O and Si (Figs. 2, 4 and 5), and in most cases also S. The element Si may at least partly be an artifact of the substrate. The S X-ray peak in EDX spectra originates either from sulfates internally mixed with the carbonaceous particles or from stray radiation. Please note that the heights of the individual peaks in Fig. 2 are not proportional to the element concentrations but give a rough estimate of the element abundance. The elements Cr, Fe, and Ni are often found as minor components (Fig. 4). These three elements exclusively occur within the carbonaceous matrix, and are not abrasion products from ice particles hitting the aerosol inlet as the metallic particles described by Murphy et al. (2004) and Martinsson et al. (2014). Furthermore, none of the samples was collected during the existence of ice particles which could potentially remove material from the impactors' inlet. During collection of samples A, B, E and G, polar stratospheric cloud particles (PSCs) containing oxides of nitrogen, NOy, were abundant. As we found the refractory carbonaceous particles in all samples independent of the occurrence of NOy, they are not artifacts from the removal of material from the inlet system. As the TEM substrates are made of a formvar foil predominantly consisting of C, O, and traces of Si, these three elements may – at least partly – be an artifact of the substrate. However, there are three points of evidence which clearly show that the refractory carbonaceous particles observed are not an artifact of the substrate:

Much higher carbon X-ray count rates were obtained when measuring particles compared to the pure substrate.

The refractory carbonaceous particles have a different morphology than the few carbonaceous particles found on blank samples. In addition, the carbonaceous particles encountered on blank samples are often much larger with a size of several micrometers. Please note that a different foil with much higher O and Si contents was used for sample K. Thus this sample was excluded from Figs. 4, 6 and 8. The TEM grids consist of Cu, leading to strong Cu X-ray peaks in the spectra (Fig. 2). Consequently, Cu is excluded from the further analysis and discussion. Mg is only present in a few particles. S is the major component of the volatile material surrounding the carbonaceous particles. Fe is found as a minor element in the majority (∼ 95 %), Cr in about 87 % and Ni in about 49 % of refractory particles.

The spatial distribution of minor elements within the carbonaceous particles was investigated by element distribution images in STEM (Fig. 5) with a 256×256 pixel resolution as well as by measuring several points on the same particle. With both approaches it is possible to obtain highly resolved information on the spatial distribution of elements within a nanometer-scale particle. C is the most abundant element and is found in the whole particle. The elements O, Si, Cr and Fe only occur in some regions of the particles. The element Al is only detected in few particles. Due to the low number of X-ray counts, the distribution of Mg and Ni is difficult to assess. S seems to occur in the whole particle and is assumed to come from stray radiation of the surrounding sulfates. The heterogeneous element distribution was also observed in multiple point measurements (up to 20 points on one particle).

Based on elemental composition, the refractory carbonaceous particles (number n=330) were classified into four groups (Table 3). Only a few particles consist of C, O, and Si only. Many particles contain additional Cr, Fe and Ni (n=131) or Cr and Fe (n=125). Particles only containing additional Fe are rare (n=22). Please note that the element S was not used for particle classification, because this element is found in most spectra, either due to scattering from surrounding volatile particles or because the refractory carbonaceous particles are embedded in sulfates.

Element ratios relative to C (at % / at %) are shown in Fig. 6 for all samples. The median O/C ratio varies between 0.052 and 0.129. The median Si/C ratios vary between 0.003 and 0.012, but may be influenced by the substrate. For all other elements, the respective median ratios are generally very low (< 0.005). Sample K is not shown due to the different substrate used (with lower C and higher Si content). The differences between the various samples are for all element ratios statistically significant on the 5 % level. Most obvious, sample G has lower element / C ratios than the other samples.

The nanostructure was investigated by high-resolution TEM. All particles are either completely amorphous (Fig. 7a, b) or show only very small regions (fewer than 10 graphene sheets) with ordering (Fig. 7c, d). In the latter case, the graphene sheet separation distance was determined (Table 4). As these measurements are very time-consuming, only 23 particles from 5 samples were investigated. The graphene sheet separation distance varies between 0.19 and 0.60 nm, the mean values of different grains between 0.25 and 0.47 nm. This range is slightly larger than typically observed for soot (Vander Wal et al., 2010; Li et al., 2011; Liati et al., 2014; Weinbruch et al., 2016).

All samples, with the exception of one (sample D), were collected inside the polar vortex. Element ratios of the samples inside and outside the vortex are compared in Fig. 8. There are no statistically significant differences on the 5 % level in element ratios and particle size (Fig. 3) between these two cases.

We find that all of the refractory particles are carbonaceous and typically contain minor amounts of Fe, Cr and Ni distributed within the particles. Most of the refractory carbonaceous particles are not included in or coated by sulfate. This is surprising, as the particles were sampled in air having low abundance of N2O and therefore long residence times in the stratosphere (Table 1). Therefore, one would expect that all refractory particles occurring in the polar stratosphere are covered by sulfuric or nitric acid. The low abundance of refractory particles internally mixed with sulfates contradicts expectations based on the models by Mills et al. (2005) as well as the findings of Sheridan et al. (1994) and Murphy et al. (2013), which suggest that most or all stratospheric refractory particles should be embedded in or coated with sulfuric acid. The results of our study can partly be explained by the evaporation of the sulfate component in the electron beam prior to its identification. The mixing state of the refractory carbonaceous particles may also be caused by splattering of volatile material of previously internally mixed refractory/volatile material. However, the reason for most of the refractory carbonaceous particles to be externally mixed remains open.

Refractory carbonaceous particles in the polar stratosphere were identified in several earlier studies (discussed below). Depending on the applied technique different terms were used for such particles. In the present contribution the following nomenclature is used: all particles consisting predominantly of the element carbon are termed carbonaceous; i.e., only the chemical composition is used for classification. The term soot is used for agglomerates of primary particles (20–50 nm size) predominantly consisting of carbon which show a specific, onion-shell-like nanostructure (Buseck et al., 2012). Black carbon (BC) is used for particles strongly absorbing light in a wide spectrum of the visible wavelength (Petzold et al., 2013) with at least 5 m2 g-1 at a wavelength of 550 nm (Bond et al., 2013).

Soot in the stratosphere was previously identified by scanning and transmission electron microscopy in accordance with the nomenclature outlined above (Pueschel et al., 1992 (diameter = 0.2–0.3 µm), 1997 (diameter ≤ 1 µm); Sheridan et al., 1994 (diameter ∼ 0.3 µm); Blake and Kato, 1995 (diameter ≤ 0.5 µm); Strawa et al., 1999 (diameter = 0.3–0.4 µm); Ebert et al., 2016 (diameter ≤ 0.5 µm)). Carbonaceous particles (diameter ∼ 0.1 µm) which might be soot although they were not unequivocally identified as such, due to the lack of high-resolution images, were found by Chuan and Woods (1984). Testa et al. (1990) found seven poorly graphitized carbon particles in their samples and regarded them as artifacts of carbon films from the TEM grids. As they did not provide images of these particles, they cannot be compared to our findings. However, we can exclude that our particles are substrate contaminants, since such particles did not occur on the blank samples. In addition, the refractory carbonaceous particles were only observed within the impaction spot and not on the clean substrate.

Carbonaceous particles partly containing heterogeneous metallic inclusions were found by Chen et al. (1998) (diameter = 0.1–2 µm). Some of these particles were called soot without providing a more precise description. Thus, it is not clear if they are similar to particles we identified as soot. According to these authors, the “soot particles” most likely stem from aircraft exhaust, as the samples were – at least partly – collected in the exhaust of an aircraft in the lower stratosphere.

Mixed carbon–sulfur particles were observed by Nguyen et al. (2008) (diameter ≤ 1 µm) at 10 km altitude between 50∘ N and 30∘ S. These particles were assumed to have formed from condensed organic matter. The differences between these particles and those found in the current study might result from differences in sampling altitudes and regions. Therefore we cannot totally exclude the particles to be different, taking into account that the particles might have evolved from condensed organic matter. However, we do not know if secondary organic particles become refractory as a result of atmospheric processes.

Stratospheric carbonaceous particles were also often detected by means of mass spectrometry (MS). For example, Murphy et al. (1998, 2013) identified carbonaceous particles, with a lower abundance in the stratosphere compared to the upper troposphere (Murphy et al., 2013; diameter = 0.3–0.8 µm; tropics and midlatitudes). The same group (Murphy et al., 2007) reported the presence of a small fraction of carbonaceous particles (diameter ≤ 300 nm) within the stratosphere. As some of these particles contained potassium, they were assumed to originate from biomass burning. Particles from rocket and space shuttle exhaust (collected in the stratosphere) were investigated by Cziczo et al. (2002) with the same instrument. In solid-fuel rocket exhaust (Athena rocket and space shuttle boosters) the most frequent observed signals stem from different aluminum oxide species (often with minor amounts of Fe). About 17 % of the particles (Athena rocket) were classified as carbonaceous as compared to 1 % from space shuttle exhaust. In principle, the carbonaceous particles found by MS could be similar to the carbonaceous particles of the present study. Since MS does not yield images of the particle morphology or information on the nanostructure, no definite conclusion can be drawn.

Stratospheric carbonaceous particles were also identified by optical measurements. For example, a single-particle soot photometer (SP2) was applied by Baumgardner et al. (2004) (diameter = 0.15–1 µm) on board of an aircraft in order to identify ozone loss processes in the polar vortex. In total, 60 % of the light absorbing particles incandesce at temperatures above 3500 K and are, thus, interpreted as BC. According to Baumgardner et al. (2004) these particles originate from tropospheric sources rather than aircraft emissions. The same technique was applied by Schwarz et al. (2006) (diameter = 0.15–0.7 µm) to identify BC in midlatitudes from the boundary layer to the lower stratosphere. Only ≤ 1 % of the particles were classified as BC and no potential source was specified.

Local enhancements of carbonaceous material at altitudes around 25 km were also deduced from simultaneous measurements of a spectrometer on board a satellite as well as from radiance and particle counter data (Renard et al., 2008) (diameter = 0.35–2 µm) obtained on a stratospheric balloon. This material was thought to be injected into the stratosphere by the pyro-convective effect (i.e., fire-started or fire-added convection). Some of the particles with submicron size were supposed to originate from vaporized interplanetary material (Renard et al., 2008). Due to the lack of information on morphology, chemistry and microstructure of the particles, a direct comparison of the carbonaceous material deduced from optical measurements with the particles encountered in the present study is not possible.

In the present study, only carbonaceous particles and sulfates were observed similar to previous findings (Pueschel et al., 1992; Blake and Kato, 1995; Strawa et al., 1999; Nguyen et al., 2008). There are, however, several previous publications which describe the presence of a variety of other refractory particle groups in addition to carbonaceous particles. These additional particle groups include metallic particles (Chuan and Woods, 1984; Sheridan et al., 1994; Chen et al., 1998; Baumgardner et al., 2004; Ebert et al., 2016), meteoritic particles (Murphy et al., 1998, 2007, 2013; Renard et al., 2008, Ebert et al., 2016), silicates (Testa et al., 1990; Ebert et al., 2016), crustal-type particles (Sheridan et al., 1994; Chen et al., 1998), and Ca-bearing particles (Della Corte et al., 2013; Ebert et al., 2016).

In summary, the sole occurrence of refractory carbonaceous particles and sulfates in stratospheric samples was reported in previous literature but seems to be uncommon. The median number mixing ratio (1.1 mg air-1) of carbonaceous particles is smaller by an order of magnitude than the abundance of non-volatile particles reported by, e.g., Weigel et al. (2014) for measurements in the winter stratospheric polar vortex in 2003, 2010 and 2011. The method described by Weigel et al. involves exposure of particles to a temperature > 250 ∘C and determination (with a CPC) of the number of particles that did not evaporate to sizes below the detection limit of the CPC. They concluded that up to 80 % of the particles present were non-volatile by this criterion. Following our definition only a few percent of the SOLVE particles are non-volatile in the electron microscope. This discrepancy may be caused by the different definitions of a non-volatile particle.

The most likely sources of refractory carbonaceous particles found in the current study include aircraft emissions, extraterrestrial sources, rocket exhaust and explosive volcanic eruptions as these sources emit material directly into the stratosphere. In addition, biomass burning can be considered as a possible source, as it was shown by several authors that large fires have sufficient energy to inject particles into the lower stratosphere (Fromm et al., 2000, 2006; Siebert et al., 2000; Fromm and Servranckx, 2003; Jost et al., 2004; Siddaway et al., 2011). Domestic wood burning, coal combustion, diesel engines and ship exhaust are not expected to significantly contribute to the stratospheric particle load. However, as these sources emit large amounts of carbonaceous material on a global scale (Bond et al., 2004; Gaffney and Marley, 2009; Corbett and Koehler, 2003; Lauer et al., 2007), they will be discussed shortly. For example, Thornberry et al. (2010) show that tropical tropospheric particles contain high amounts of thermostable carbonaceous material, and it is possible that these particles become mixed within stratospheric air. Although the vertical exchange may be less effective than the direct injection processes, however, it is conceivable that fractions of carbonaceous aerosol material released in the troposphere are vertically transported into the stratosphere by processes such as tropical convection (and lifted further via the Brewer–Dobson circulation), cyclogeneses, warm conveyor belts, tropopause folds and/or isentropic transport.

Most particle groups discussed in the following were collected close to their emission source. We are aware of the fact that particles collected in the polar stratosphere may in principle change their properties during their atmospheric lifetime. However, ordering of carbonaceous material is an irreversible process leading always to a higher degree of ordering (Diessel et al., 1978; Itaya, 1981; Pesquera and Velasco, 1988). As most of the particles analyzed show no or only very little ordering, it is assumed that the particles did not change their nanostructure during their atmospheric lifetime. On the other hand, several electron microscopy studies describe soot particles in the stratosphere (Pueschel et al., 1992, 1997; Sheridan et al., 1994; Strawa et al., 1999; Ebert et al., 2016). Thus, it can be expected that soot particles – once injected into the stratosphere – do not change their typical nanostructure under stratospheric conditions.