The general term soot is usually used for particulate products of incomplete combustion or pyrolysis of fossil fuels or organic materials (Andreae and Gelencsér, 2006; Pöschl, 2005; Bond and Bergstrom, 2006).

The structure of soot depends on the fuel and on the combustion conditions. Soot produced under oxygen-rich combustion conditions and high temperatures consists of agglomerated primary particles with sizes between 10 and 30 nm (Sadezky et al., 2005). The primary particles consist of onion-like turbostratically ordered graphitic layers. The graphitic domains typically include 3–4 graphene layers and have an extent of about 3 nm with interlayer distances of about 3.5 Å (Sadezky et al., 2005), which is larger than the interlayer distance of an ideal graphitic lattice (3.35 Å) because of the turbostratic arrangement. The atoms in graphitic-like material are bound with π bonds, which form conjugated systems and hence long-range orbitals. As the optical gap is small (≈0.5 eV for disordered graphitic layers; Robertson and O'Reilly, 1987), light from a broad spectral range is absorbed (Bond and Bergstrom, 2006; Chhowalla et al., 1997).

In contrast to this highly ordered type of soot, the molecular structure of BrC, which is formed under oxygen-poor and low-temperature combustion conditions, is similar to that of polycyclic hydrocarbons (PAH) or humic-like substances (HULIS) (Graber and Rudich, 2006; Sun et al., 2007). Due to the smaller expansion of the molecular orbitals there is a larger optical gap between the filled valence band and the unfilled conduction band, which leads to a decreasing absorption efficiency towards the long-wave part of the visible spectrum (Bond, 2001; Kim et al., 2015; Andreae and Gelencsér, 2006).

For the present study, a combustion aerosol standard burner (type miniCAST) was used to produce differently structured soot by varying the combustion conditions. A detailed description of the miniCAST burner is given below. Different fuel-to-air ratios in the flame lead to different compositions and structures of the produced particles. According to Kim et al. (2015) and Mamakos et al. (2013) the particles obtained under oxygen-rich settings are comparable to diesel exhaust aerosols. They show a high EC fraction and form relatively large fractal agglomerates with mobility diameters between 70 and 130 nm, which are composed of small (25–30 nm) spherical primary particles (Kim et al., 2015). Under oxygen-poor combustion conditions, spherical particles with sizes in the range 10–60 nm and a low EC fraction (Kim et al., 2015) are formed. While soot from biomass burning contains also ionic components such as Na+, NH4+, Ca2+, Mg2+, K+, Cl-, NO3- and SO42- which could lead to catalysis effects in the thermal–optical analysis (Ichikawa and Naito, 2017; Wang et al., 2010), these compounds are negligible in soot produced by the CAST burner since only pure propane and air contribute to the combustion.

Raman scattering has been used for the structural investigation of soot for several years (Rosen and Novakov, 1978; Sadezky et al., 2005; Ferrari and Robertson, 2000; Schmid et al., 2011; Knauer et al., 2009; Ivleva et al., 2007a, b). It is sensitive to different C–C bonding types (e.g., graphitic structures) in a material (Sadezky et al., 2005) and depends on the ordering of sp2 sites (Ferrari and Robertson, 2000).

Ideal graphite shows a single peak at a Raman shift of ≈1580 cm-1 (G peak or graphitic peak), which is related to the ideal hexagonal environment in the extended graphene layers (E2g symmetry). For nonideal graphite, an additional peak at ≈1350 cm-1 (D peak or defect peak) appears, which is due to impurities and/or smaller graphene layers and the thus increased number of layer edges. Due to the missing neighbor atoms at these edges, the symmetry of the C–C vibration is reduced to A1g. When the ordering in the material decreases – which is the case in soot – the two formerly sharp peaks broaden and overlap (see also Figs. 5 and 6) as a consequence of more than two overlapping Raman bands. To separate the contributions of the bands, different authors suggest different curve-fitting methods with up to five signals with either Gaussian or Lorentzian shape (Dippel et al., 1999; Jawhari et al., 1995; Sze et al., 2001; Cuesta et al., 1994). A comparison study performed by Sadezky et al. (2005) found that a five-curve fit with four Lorentzian and one Gaussian curves represents most of their spectra best. The curve shapes, band positions and related vibration modes are listed in Table 1.

The ratio of the intensities of the D and G peak (ID/IG) can be linked to the crystallite size (La) within the investigated material (Ferrari and Robertson, 2000; Tuinstra and Koenig, 1970). Ferrari and Robertson (2000) propose an increase in La with increasing ID/IG below La≈2 nm. In this regime ID/IG is proportional to the probability to find 6-fold aromatic rings in the cluster (Ferrari and Robertson, 2000). For larger crystallite sizes ID/IG decreases with increase in La (Tuinstra and Koenig, 1970). ID/IG shows a maximum at La≈2 nm but has a broad transition regime between the increasing and decreasing branch. Zickler et al. (2006) compare the ID/IG ratio of Raman spectra of charcoal obtained from spruce wood with the crystallite sizes obtained from X-ray diffraction and confirm the proposal of Ferrari and Robertson experimentally. Based on these findings, several authors (Commodo et al., 2016; Ess et al., 2016; Ivleva et al., 2007a, b) use increasing ID/IG ratios as criteria to detect an increasing degree of ordering in soot.

For the analysis of the filter samples and the heating of the samples a dual-optics EC–OC analyzer (Sunset Instruments Inc.) was used. In the first heating steps, the samples are heated in helium. In the second part of the analysis the samples are exposed to an oxidizing atmosphere consisting of 2 % oxygen and 98 % helium. Reflectance and transmittance signals of laser light (635 nm) are used to monitor the darkening of the sample caused by pyrolysis during heating in the inert atmosphere. In this study, the NIOSH870 protocol was used. Temperature steps and residence times are listed in Table 3.

The NIOSH870 protocol was chosen here to investigate the structural changes of soot during pyrolysis, because of the strong charring occurring in the He mode at the high temperatures. The EUSAAR2 protocol (Cavalli et al., 2010), which is now used widely in the EU (Brown et al., 2017), minimizes charring and is therefore less suitable for the investigation of PC.

BC and BrC (often taken together as light-absorbing carbon, LAC) of the original and the heated samples were analyzed with an extension of the original integrating-sphere technique (described, e.g., by Hitzenberger and Tohno, 2001). In this technique, a sample (e.g., a filter punch) is immersed in a liquid and introduced into the center of an integrating sphere illuminated with diffuse light. In our study, a 15.24 cm (6 in.) integrating sphere (Labsphere, Inc.) coated internally with a nearly ideally diffusely reflective (>99 %) material (Spectraflect^™) was used. Samples were immersed in a mixture of 10 % isopropanol, 40 % H2O and 50 % acetone in PE vials. Enhanced absorption caused by a possible coating effect is reduced: soluble coatings are removed from the particles and the effect of insoluble coatings is reduced because of the low relative refractive index of these coatings compared to the liquid (Hitzenberger and Tohno, 2001). In the extended technique (Wonaschütz et al., 2009) the sphere is illuminated with a halogen light source equipped with three interference filters (450, 550 and 650 nm) and the wavelength-dependent light signal is recorded with a photodiode. The contributions of BC and BrC to the absorption signal are separated in an iterative procedure using calibration curves obtained with a proxy for BC (Elftex 124, Cabot Corp.) and a proxy for BrC (humic acid sodium salt, Acros Organics, no. 68131044).

In this study a confocal Raman microscope (Horiba Jobin Yvon LabRAM 800HR) was used. The Raman microscope was equipped with a 632.8 nm HeNe laser (maximum output <20 mW) and a charge-coupled device (CCD) camera (Peltier cooled at -60 ∘C). The laser beam was focused on the sample with a 20× magnification objective (CF Plan, 20×/0.35, WD 20.5 mm, Nikon). The spectra were calibrated with the Rayleigh line at 0 cm-1 and the silicon peak at 521 cm-1. The laser power was reduced to 10 % for analyzing black samples and 25 % for brown samples to prevent thermal destruction of the material. All instrument settings (grating 300 lines per mm, 3–4 cm-1 resolution; confocal hole 1000 µm; acquisition time 5 s with 30 accumulations) were chosen after a set of test measurements to provide the best signal-to-noise ratio for the analysis. As the samples burned off partially at the 775 ∘C temperature step in the EC–OC analyzer, the Raman spectra were recorded only for the first seven temperature steps.

The spectra were recorded in the range from 200 to 2000 cm-1 at four measurement points at three positions for each sample to account for possible variations within a filter sample. The averaged spectra for each filter were analyzed using OriginPro2016, which has built-in functions for baseline correction. Baselines were fitted with a B-spline function based on approximately 20 anchor points. After subtraction of the baseline, the spectra were fitted with five curves (four Lorentzian, one Gaussian; see Table 1). As initial values for the fit, the band positions obtained by Sadezky et al. (2005) and listed above (Table 1) were used. The standard deviations of the mean spectra were used in the fitting software as weighting of the fit.

A 200 kV transmission electron microscope (TEM) Philips CM200 equipped with a Gatan^™ Orius CCD camera was used to analyze the nanostructure of the original and the heated samples. Single filter fibers were separated from the samples and placed onto holey carbon films or between copper grids to facilitate studying freestanding soot particles. High-resolution microscopic images and diffraction patterns of particles deposited at the lateral edges of the fibers were taken. Intensity profiles of the diffraction patterns were obtained by both azimuthal integration along rings and background correction using PASAD-tools (Gammer et al., 2010). For comparison a simulated profile of graphite was calculated with JEMS software (Stadelmann, 2004).

A LAMBDA 750 UV–VIS spectrometer (PerkinElmer, Waltham, Massachusetts, US), equipped with a tungsten and a deuterium lamp, was used to measure the diffuse reflectance of the samples. Reflectance measurements were carried out from 800 to 200 nm with an interval of 1 nm. At 319.2 nm the instrument switches from the tungsten (visible) to the deuterium (UV) lamp. For the measurement the original and heated samples were put into quartz vials without any liquid. The diffusely reflected light from the filter punches was collected with a 60 mm integrating sphere and detected with a photomultiplier. Absorbance spectra where calculated with the Kubelka–Munk equation (Kubelka and Munk, 1931) from the measured reflectance spectra. Although the Kubelka–Munk theory was developed originally for powder samples, its applicability to aerosol filter samples was shown by Aryal et al. (2014).

Absorbance spectra were calculated from the measured reflection signals R∞(λ) using the Kubelka–Munk function: 1K–Mλ=1-R∞(λ)22R∞(λ), where K–M(λ) is proportional to the absorbance, assuming infinite sample thickness. The spectra were corrected for background signals obtained from three filter blanks.

In the thermal–optical analysis most of the carbon of the black sample evolves in the He+O2 phase. Throughout the successive heating steps, the laser reflectance signal remains relatively constant until EC oxidizes (Fig. 2). On the other hand, most of the carbon of the brown sample evolves in the He phase (Fig. 3). The laser reflectance signal decreases during the first three temperature steps, indicating a pyrolysis of the organic material. The signal starts to increase slightly at the last temperature step (870 ∘C) in He and increases rapidly at 625 ∘C after O2 is added, indicating the combustion of initial EC and/or pyrolyzed OC (PC). The laser signal reaches its initial value at the end of the 700 ∘C temperature step; therefore approximately half of the carbon signal in the oxidizing atmosphere is assigned to OC following the normal procedure of the thermal–optical analysis method. Results from EC and OC measurements are summarized in Table 3. Figures 2 and 3 show full thermograms of a black and a brown sample.

Figure 4 shows the change in BC and BrC during the thermal–optical analysis. The changing BC content of the sample is qualitatively consistent with the interpretation of the laser signal in the thermal–optical analysis: BC in the black sample decreases somewhat during the whole analysis cycle, while BrC is below the detection limit for each temperature step.

For the brown sample, BrC decreases continuously during the whole He mode, while BC increases over the He mode up until 870 ∘C (He) and decreases in the He+O2 mode. These findings confirm that BC is built during the thermal–optical analysis of an organic-carbon-containing sample.

Table 4 summarizes the composition of the original brown and black samples regarding their EC, OC, BC and BrC content: the black sample consists mainly of EC or BC with a negligible amount of BrC. On the contrary, the brown sample contains only about 10 % EC or BC and approximately 90 % BrC.

The change in the Raman spectra due to the heating process in the thermal–optical instrument is shown in Figs. 5 and 6. For a better comparison the spectra are normalized to the maximum of the G peak. While the maxima of the D and G peak do not change for the black sample, the spectra of the brown sample show a relative increase in the D peak for samples heated at 870 ∘C in He. The peak does not increase continuously over the whole heating process, which indicates a significant structural change especially at this temperature step. Following the interpretation of Ferrari and Robertson (2000) and its application to soot samples by Commodo et al. (2016), Ess et al. (2016) and Ivleva et al. (2007a, b), this increase in the D peak indicates an increased amount of polyaromatic rings in the sample which can be associated with a higher degree of ordering.

For a better comparison, the spectra of the black and brown original samples and samples heated to 870 ∘C in He are shown in Fig. 7. The D peak of the brown heated sample reaches nearly the height of the D peak of the black original and heated samples.

Figure 8 shows the five-curve fits of these four spectra. All fits were performed on the smoothed and averaged versions of the spectra. It is obvious that there is no significant change in the five peaks between the original and heated black sample. However, the spectra of the brown sample show a significant change at the D4fit peak (≈1200 cm-1) and the D1fit peak (≈1350 cm-1), when the sample is heated to 870 ∘C in He: while the D4fit peak (which is related to C=C double bonds, Sadezky et al., 2005) decreases, the D1fit peak (related to graphene layer edges, Sadezky et al., 2005) increases. This implies that the material in the heated sample contains fewer C=C double bonds and more graphene layer edges compared to the original brown sample.

Possible explanations could be a decomposition and evolution of molecules with C=C bonds (e.g., polyenes) and a coincident fragmentation of preexisting graphene layers or a transformation of molecules with C=C bonds into new small graphene layers. Both scenarios would lead to a higher amount of graphene layer edges (and an increased D1fit peak) and a lower amount of C=C bonds (and a decreased D4fit peak).

TEM images (see below) show that the ordering in the brown sample increases when it is heated to 870 ∘C in He. The UV–VIS spectra (see below) show a decreasing absorption coefficient which requires a growth instead of a fragmentation of the conjugated orbitals in the material. Both findings are more consistent with the second process.

The TEM images of the black and brown samples show entirely different morphologies. The soot in the black sample consists of agglomerates of spherical primary particles with diameters of about 20 nm (Fig. 9). The primary particles show an onion-like graphitic structure as it is also reported by Sadezky et al. (2005) and Kim et al. (2015). From the TEM images of the brown sample an ordered internal structure is not visible. The particles seem to consist of an oily substance (described as “condensed organic species” by Moore et al., 2014) which adheres well to the filter fibers (Fig. 10). However, analysis of the electron diffraction patterns indicates a small degree of ordering (see below).

After heating to 870 ∘C under helium atmosphere, the black sample shows the same graphitic-like structure as the original black sample. However, the brown heated (870 ∘C He) sample seems to have a more ordered internal structure in the form of layers than its original version as indicated by stronger noticeable local fringe-like contrast (Fig. 11).

The electron diffraction patterns (Fig. 12) of both heated and original samples show rings with maxima at 2.8 nm-1 (A), 4.9 nm-1 (B) and 8.4 nm-1 (C). The positions of the rings are very similar to those of simulated graphite with randomly oriented small graphitic domains (cf. inset in Fig. 13). Therefore we conclude that the ring A at 2.8 nm-1 can be related to the layer distance of graphite and the ring B at 4.9 nm-1 to the (100) or (101) planes of graphite. All rings appear in the diffraction patterns of both the black and the brown samples in Fig. 12, but they are broader for the brown original sample. The latter result is demonstrated more clearly in the intensity profiles calculated by azimuthal integration along rings (Fig. 13) and indicates that the brown original sample is less ordered than the black sample. This finding is also consistent with the Raman spectra and the real-space TEM images. Based on the comparison of the positions and profiles of the experimental peaks to the simulated ones, it is concluded that all samples contain small graphitic-like atomic arrangements. The tendency of the maximum A to larger diffraction vectors by thermal treatment of the brown sample refers to a slight reduction of the distance of graphitic layers by heating resembling that of the black one.

The quantitative evaluation of the peak width by the full width at half maximum (FWHM) is displayed for the different samples in Fig. 14. In the case of the brown sample the FWHM of ring B decreases by 30 % as a consequence of heating. This result can be interpreted as an increase in the coherently scattering domain size (Fultz and Howe, 2001) and an increased degree of structural order during heating. This interpretation holds also for the other peaks; for example, the FWHM value of peak A at 2.8 nm-1 changes from 0.54±0.05 to 0.45±0.04 nm-1 by heating up to 870 ∘C (Figs. 13 and 14).

Based on these FWHM values and using the Scherrer equation (Fultz et al., 2001) the size of the coherently scattering domains (in the direction perpendicular to the (002) layers) changes from about 1.6 to 2 nm as a consequence of heating. On the other hand, the FWHM of the black sample does not change and neither does the internal structure of the material (Figs. 13 and 14). Black original and heated samples were found to have coherently scattering domains with sizes of about 2.2 nm from which we conclude a slight difference in the structural ordering between the black and the brown heated sample. The changes in the graphitic domain size of the brown sample during heating confirm the findings of the Raman measurements but the temperature dependence is still different. As shown in Fig. 14 the FWHM of the brown sample decreases already at temperatures below 870 ∘C. This is in contrast to the Raman spectra, where a sudden change in the shape of the curves occurs at 870 ∘C in He. The cause for this difference is not completely clear. A possible explanation might be a reorientation and alignment of existing clusters and layers and therefore an increase in the size of coherently scattering domains at temperatures below 870 ∘C. This would not necessarily change the Raman signal since the structure within the clusters can be kept unchanged and the existing clusters would only rotate, rearrange and align.

Figure 15 shows the wavelength dependence of light absorption by the brown and black original and heated samples. The wavelength dependence of the brown sample changes during the heating process. The absorption spectra for the black heated and original samples, however, do not change significantly, particularly not in one distinct direction. The spectrum of the brown original sample has a strong wavelength dependence, but during the heating process this spectrum changes and becomes more and more similar to the spectra of the black samples.

The wavelength dependence of the absorption was fitted using the Ångström power law: 2ln⁡(I/I0)=K⋅λα, where I0 is the incident intensity, I the reflected intensity, α the Ångström exponent, λ the wavelength and K a constant. Figure 16 shows the absorption spectra for the brown original and heated samples with the spectrum for the black original sample for comparison. The Ångström exponents of the respective curves are given in the figure.

The spectrum of the original brown sample starts with an Ångström exponent of α=-1.79. During the heating process the wavelength dependence decreases and reaches values of α≈-0.63 after the heating step of 870 ∘C in He, which is even lower than the Ångström exponent for the black original sample (α=-0.92).

The low Ångström exponent (-0.35) of the sample heated to 775 ∘C in O2 might suffer from measurement uncertainties. At this heating stage, most of the absorbing material has already been burned off the filter and the K–M absorption signals are already very small.

This decrease in the Ångström exponent for the brown sample indicates an increase in BC, which is consistent with the findings of the IS measurements. At the heating stage of 870 ∘C in He and after this stage, the Ångström exponent is even lower than for the black original sample.

Absorption Ångström exponents depend on both the refractive index and on the size of the absorbing particles (Bohren and Huffman, 1998). Absorption by small particles (in the Rayleigh regime) has a stronger wavelength dependence than that by larger particles in the Mie scattering size range. The observed behavior of the Ångström exponents could be caused by the different sizes and shapes of the particles of the black sample compared to particles in the brown sample. As we saw from the TEM images, the particles of the black sample consist of agglomerates of spherical primary particles with diameters of about 20 nm, whereas the particles in the brown sample are larger than these primary particles. This large size difference is not substantially changed due to heating.