Top soil and fresh rock samples from a nearby outcrop were taken from a field site in Rambla Honda, Sierra de los Filabres (37∘7′43′′ N, 2∘22′30′′ W/Southern Spain). The area is located in the Nevado-Filabride complex and contains Devonian-Carboniferous slaty mica schist with graphite and garnets crossed by abundant quartz veins (Puigdefábregas et al., 1996). Carbonates found in the soil sample (0.18 % C) originated from pedogenesis and dust deposition, as the parent rock does not contain carbonates. Soil material was taken from the topsoil (0–5 cm, without sieving crust) under the grass tussock Macrochloa tenacissima to ensure a substantial amount of OC was present. The soil itself was classified as Skeletic Leptosol (colluvic) according to the World Reference Base for Soil Resources (WRB, 2014). Additional soil material was collected from a field near the town of Alboloduy (37∘4′9′′ N, 2∘36′43′′ W), hereafter referred to as AB soil, with similar vegetation and (climatic) conditions. The lithology consists of feldspathic mica schist (IGME, 1979), but without natural graphite and with a much higher CaCO3 content (1.87 % C). The AB soil, classified as Skeletic Leptosol (WRB, 2014), was also sampled from the topsoil, without sieving crust, under the grass tussock Macrochloa tenacissima. The soil samples were dried at 40 ∘C and sieved to ≤2 mm.

Furthermore, three artificial soils were created, resembling a simplified version of a natural soil sample. For artificial soil 1 the organic horizon under deciduous tree species in Tharandter Wald (Saxony, Germany) was collected as OM substitute. Muscovite (American Educational, PN B00657LD62), a primary mineral present in the collected rock and soil samples (IGME, 1975), was taken as mineral component and ground in an agate disc mill. Together with CaCO3 (Merck, Darmstadt, PN 1.02066) and graphite standard material (Merck, Darmstadt PN 1.04206), the components were mixed in a pure quartz matrix (Merck, Darmstadt, PN 1.07536) in the ratio of 10 % Muscovite, 2.4 % OM (=1.0 % C), 2.1 % CaCO3 (= 0.25 % C) and 0.5 % graphite (=0.5 % C). Additionally, the individual components were mixed in quartz in the same concentration. Artificial soil 2 was made without carbonate and Artificial soil 3 without graphite, whereby concentrations of the other components were kept constant (Table 1).

For testing and developing potential quantitative methods, the graphite standard was used as the reference material and added to soil from Rambla Honda (denoted calibration set 1) or with pure quartz (denoted calibration set 2) as a matrix in different quantities from 0.1 % to 4 % (Table 1). All samples were ground in an agate disc mill (Retsch GmbH, Haan, Germany) in order to achieve homogenization.

Total C content of the different mixtures and samples were measured and checked using an elemental CN analyser (Vario EL Cube, Elementar, Langenselbold, Germany). By acidifying samples, using an excess of HCl, the carbonates were removed. After drying at 60 ∘C, the difference between acid and non-acid treatment was measured and denoted as total inorganic C (TIC). TC and TIC values of the used samples are summarized in Table 1. We assumed that with dry combustion at 950 ∘C under pure oxygen atmosphere, all C holding components were decomposed or oxidized and therefore the total C content could be measured. This assumption was later validated by performing TGA temperatures up to 1100 ∘C.

Raman spectra of the soil of calibration set 1, standard (Merck) and graphitic schist were recorded, using a Thermo Scientific DXR Smart Raman Spectrometer, with a 532 nm laser and a power output of 9 mW. Before the measurement, samples were pressed in aluminium cups. The obtained peaks were integrated using Lorentzian profiles fitting in Origin 2019.

The spectra of the graphite standard were highly similar to the graphitic schist (Fig. S1 in the Supplement). The D1 (∼1350 cm-1), G (∼1580 cm-1) and D′ (∼1620 cm-1) peaks could be attributed to graphitic C. The ratio between the D1 and the sum of all three peaks are clear indicators for the degree of graphitization (Beysacc et al., 2003; Ferrari, 2007). Ratios of 0.20 for the graphite standard (Merck) and 0.34 for the graphitic schist indicated well-organized carbon (< 0.5, Beysacc et al., 2003) for both samples. Peaks around the 1200 and 1500 cm-1 bands, which would have indicated the presence of pyrogenic / black carbon components (Sadezky et al., 2005; Schmidt et al., 2002), could not be observed in all samples, including the soil sample (blue, Fig. S1).

For transmission-FTIR analyses, 1 mg sample was mixed with 99 mg potassium bromide (KBr; Merck, Darmstadt; 3 sample replicates), finely ground in an agate mortar, and pressed to pellets. The transmission spectra were recorded in a Biorad FTS 135 spectrometer (BIO-RAD company, Cambridge, USA) as 16 co-added scans between wave number (WN) 4000 and 400 cm-1 at a spectral resolution of 1 cm-1. The spectra were corrected against ambient air as background and were converted to absorption units. For DRIFT analyses, ground mixtures of calibration set 1 and 2 were poured into standard cups (three replicates) without any dilution. The DRIFT spectra (16 co-added scans, WN 4000 and 400 cm-1, resolution 4 cm-1) were corrected for ambient air using a background spectrum of a gold target (99 %; Infragold) and were converted to Kubelka–Munk units. All spectra were corrected for CO2 absorption of the ambient air between WN 2400 and 2280 cm-1 and smoothed (boxcar moving average algorithm, factor for transmission spectra: 25, factor for DRIFT spectra: 15) using the software WIN-IR Pro 3.4 (Digilab, MA, USA). For each sample one mean spectrum was calculated from the spectra of three replicate spectra.

The TGAs were conducted on a STA 449 F5 Jupiter analyser (NETZSCH, Hanau, Germany). Therefore, 20–40 mg of sample material was placed in an Al2O3 crucible and heated under a constant heating program from ambient to 1100 ∘C with a ramp of 20 ∘C min-1. First analyses were conducted under an oxygen-rich atmosphere, with an inflow of 250 mL O2 min-1 and 250 mL N2 min-1. Additional tests were done under anoxic conditions whereby the oxygen inflow was cut off between 500 and 850 ∘C. The oxygen inflow was restored and the heating program continued until 1100 ∘C.

As carbonates might interfere in the TGA measurement of graphite and high chloride concentrations damages the equipment, they were removed from the sample using the acid fumigation method of Harris et al. (2001). Briefly, about 40 mg of sample was weighed in a silver foil capsules, moistened to approximately field capacity and put in a desiccator under vacuum conditions with a beaker of 31 % HCl and fumigated for 24 h. Afterwards the sample was dried at 60 ∘C overnight before it was transferred to an Al2O3 crucible for analysis.

Smart combustion denotes the method based on the DIN19539 (GS) standard (DIN Standards Committee Water Practice, 2016), whereby solid C components are separated based on their thermal and oxidizable properties. Smart combustion was conducted with the Soli-TOC cube analyser (Elementar, Langenselbold, Germany). The device is equipped with a non-dispersive infrared detector (NDIR), which measures the degree of infrared light absorbance caused by CO2 concentration in the measuring gas (O2/N2). The NDIR was calibrated with CaCO3 and additionally CaCO3 was used to control and calculate a daily standard for the measured total C content. Depending on the expected carbonate and graphite concentration, 40–90 mg of homogenized sample was placed in the crucible. This was done to make sure that the peak surface fits well to the calibration range without causing unnecessarily large peak areas, which might influence the separation of the peaks/substances during analysis. Following the DIN19539 GS standard/standard gas switching program of the Soli-TOC cube analyser, the sample was first heated to and held at 400 ∘C for 240 s, whereby the “total organic carbon 400 ∘C” (TOC400) was obtained. Subsequently the atmosphere was switched to inert gas (N2) and after an equilibration time of 100 s the sample was heated to 900 ∘C and held for 150 s. C released during this pyrolysing phase is denoted as TIC, mainly consisting of carbonates which do not need oxygen to decompose. After 150 s, the oxygen gas flow was reintroduced, and a third C component, the residual oxidizable C (ROC), was measured. It was hypothesized that this ROC fraction represents graphite.

The PLSR of the calibration set showed strong relations between the transmission-FTIR spectra from both calibration sets and the graphite concentrations when considering samples with 0.1 %–4 % graphite (Fig. 1a and b). For DRIFT spectra, the quality of these calibrations was at the same level (cal. Set 1: R2=0.97, RMSEP = 0.16; cal. Set 2: R2=0.98, RMSEP = 0.12). For calibration set 1 (based on natural soil) as well as for calibration set 2 (based on quartz), one main component of the PLSR presented most differences in the graphite concentration (Fig. 1c and d). This component showed the highest loading values across the entire range of wave numbers with some exceptions. For calibration set 1, wave numbers with decreased loading values were found at spectral regions 1077, 1031, 1013, 934, 913, 778, 536, 471, and 411 cm-1, which all corresponded to the prominent absorption bands of the original soil used as matrix, comprising functional groups from organic and mineral matter (Hesse et al., 1984; Senesi et al., 2003; Van der Marel and Beutelspacher, 1976). For calibration set 2, the wave numbers with the smallest loadings at 1171, 1084, 796, 778, 694, 506, and 457 cm-1 were specific for quartz (Van der Marel and Beutelspacher, 1976).

The cross-validation of PLSR calibrations 1 and 2 using spectra from the other calibration set yielded a linear relation but overestimated the graphite contents by ca. 2.59 % C for set 1 % and 1.87 % C for set 2 (Fig. 5). The graphite content of 0.50 % for the artificial soil was 3.5 times overestimated by the PLSR calibration using set 1 (predicted: 1.75 % C) and 3 times underestimated by calibration set 2 (predicted: 0.17 % C). The graphite contents in the graphitic schist were estimated to be 1.91 % C by calibration set 1 % C and 3.71 % C by calibration set 2, which was 2 times and 4.5 times higher than the total C content of the graphitic schist (Table 1).

First qualitative TGA results revealed overlapping mass loss peaks of graphite and CaCO3 (Fig. 2). Between 750 and 850 ∘C, the sum of the mass losses of the individual components was smaller than the mass loss of the mixture of these components.

Using the RMSEP values of the two calibration sets of soil and quartz with graphite, the most useful temperature range for modelling graphite content was identified between 680 and 840 ∘C (Fig. S4). According to PLSR, for both models, created with calibration set 1 (R2=1.00, RMSEP = 0.05) and calibration set 2 (R2=1.00, RMSEP = 0.04), predicted graphite content in their respective matrix quite well (Fig. 3). Only cross-validation by predicting graphite content in the other matrix revealed a short coming of the TGA method (Fig. 5). The slope between predicted and actual graphite content is still parallel to the 1:1 line, but applying the model derived from calibration set 1 on set 2 underestimated the graphite content by 1.81 % C (Fig. 5). In contrast, the graphite content in calibration set 1 was underestimated by 1.81 % C using the model calibrated with set 2.

The graphite contents of the artificial soil (0.5 %) and graphitic schist were estimated using the two calibrated models (Fig. 5, inserted detail, and Table 2). The content of artificial soil 1 was overestimated 3.5 times (predicted: 1.70 % C) using the model derived from calibration set 2. Using the model based on calibration set 1, the prediction of graphite content yielded negative values (predicted: -0.19 % C). In the additionally used artificial soils 2 and 3, which did not contain CaCO3 or graphite, the graphite contents were underestimated with the model based on calibration set 1 (based on soil) and overestimated with the model based on calibration set 2 (based on quartz). Independent of the two models, the relative difference between them is in both cases 0.44, which resembles the actual graphite content. The graphite content of the graphitic schist was estimated to be 2.44 % C according to calibration set 2, but the model calibrated with set 1 showed better predictions. Where the total C content of the graphitic rock was 0.84 % C, the model of calibration set 1 estimated a graphite content of 0.64 % C (Table 2).

Furthermore, the artificial soils were used to explore whether changing between oxic and anoxic conditions during thermal analysis could separate between the mass loss peak of CaCO3 and graphite – most important for potential application in soils containing both components. Changing the atmospheric composition resulted in an artificial mass gain when O2/N2 gas was switched to N2 and an artificial mass loss when oxygen was reintroduced (Fig. 4), probably due to changes in pressure, and thereby affecting the mass readings. It should be noted that a constant heating program was used and therefore the measurement time spans 4–5 min for both peaks. Nonetheless using artificial soil 2 (without CaCO3) and artificial soil 3 (without graphite) revealed that qualitative separation between the two mass loss peaks was feasible using changes in oxic conditions. Based on TGA observations on some individual components and simplified artificial soil (Fig. 2), it seemed best to use anoxic conditions from 500 till 850 ∘C as all OM will be oxidized at 500 ∘C and carbonates should be fully decomposed at 850 ∘C (Fig. 2). Furthermore, the mass loss peak had already returned to the baseline after reaching 850 ∘C, indicating that the decomposition of the carbonates was complete (Fig. 4).

First total carbon (TC) measured by the smart combustion method was compared with the TC obtained by dry combustion using the elemental analyser, but hardly any differences were found (Fig. S5). Residual oxidizable C (ROC) values obtained by the smart combustion method were plotted against the added amount of graphite in calibration sets (Fig. 5). The graphite content in calibration set 1 seems to be overestimated by 0.26 % C. This observation can be explained by the fact that the used sample soil for calibration set 1 contains an unknown amount of natural graphite, which explains a constant overestimation. The content of graphitic C of the soils in calibration set 2 is slightly underestimated, especially with higher graphite concentrations (Fig. 5 and Table 2).

The graphite content of the artificial soil 1 was slightly underestimated (Table 2): 0.40 % ROC for the artificial soil 1 % and 0.46 % ROC for the artificial soil 2, compared to the real contents of 0.50 %. Artificial soil 3 revealed 0.00 % ROC as there was also no graphite in this sample. The graphitic schist had 0.79 % ROC, which was very similar to the total C of the rock (Tables 1 and 2). Furthermore, the AB soil spiked with graphite, as an additional sample without natural graphite but with a high CaCO3 content (1.87 % C), showed a similar underestimation as observed with the calibration samples (Fig. 5).

The ROC results for the natural graphitic soil, also used for calibration set 1 and spiked with CaCO3, revealed a downward trend of measured ROC content with increasing amounts of added CaCO3 (from 0.0 % to 2.5 % added C) down to 0.01 % ROC absolute (Fig. 6). This observation was also evident when taking into account the dilution of the sample by the addition of CaCO3 and explained at least parts of the underestimation of graphite in this sample (Fig. 5).

The calibration between infrared spectra and graphite contents of the calibration sets yielded promising results (Fig. 1a and b) and could also be used for a cross-validation (Fig. 5). Although the same substrate materials and similar contents of graphitic C were used in the validation, the graphite contents were systematically over-predicted. Despite the apparent quality of the calibration, this failure could have been caused by the relatively low number of calibration samples. Note that the use of the two calibration data sets, soil and quartz, in a joint PLSR model (R2=0.96 and RMSEP = 0.24; three components) did not improve the calibration nor the prediction accuracy. It cannot be excluded that a higher number of samples for the calibration could improve the PLSR model and the prediction results. Further, Raman spectroscopy might be an alternative approach for quantifying graphite in soil samples (e.g. Sparkes et al., 2013; Jorio and Filho, 2016).

The loading values revealed that the relationships used for the PLSR calibration (Fig. 1c and d) could not be attributed to absorption bands of graphite reported in the literature such as wave numbers 2200, 1587, 1362, and 830 cm-1 (Friedel and Carson, 1971) or 3400, 1729, 1220, 1052 and 870 cm-1 (Tan et al., 2013). These wave-number positions did not match with the absorption bands of the spectra obtained from calibration sets 1 and 2, the graphitic schist and the artificial soil 1 (Fig. 7). The only exception is WN 3400 cm-1, which is mainly caused by O–H of free or adsorbed water and is thus not specific for graphite.

The height of the loading values across broad spectral regions, i.e. across the entire wave-number range in the case of the soil samples, hints at effects of the general optical conditions within the samples. The transmission, i.e. the energy throughput in the sample pellet (transmission FTIR) or the reflectance of the sample surface (DRIFT), seems to be a measure for the amount of graphite added rather than specific graphite signal intensities in the calibration sets. In addition, transmission/reflection characteristics are highly influenced by the mineral composition of a sample. Generally, increasing graphite concentrations caused decreasing transmission over the entire spectral range, which is a characteristic of the mineral composition due to increasing proportions of primary absorption (Kortüm, 1969; Hesse et al., 1984). This assumption is underlined by the fact that calibration was not possible with both calibration sets when using baseline-corrected spectra because baseline correction compensated for the described effect of decreasing transmission over the entire wave-number range with increasing graphite content. In addition, DRIFT measurements of pure, i.e. undiluted, graphite material (not shown) did not reveal any prominent absorption bands. Note that specific graphite absorption bands that have been reported in the literature are only valid for oxidized graphite where C–O and C=O groups have been formed to a certain extent (Tan et al., 2013). While FTIR spectroscopy may be feasible for determining oxidized or impure graphite, it was found to be not useful in our case, where an oxidization or impureness of the used graphite material obviously did not appear. Further, the mentioned potential signal intensities would occur in the same spectral ranges as compared to signals from SOM functional groups, thus hampering a quantification of graphite in soil samples. Consequently, the lack of specific absorption bands resulted in a strong dependency of the calibration and validation quality on the sample matrix, i.e. its main mineral component. This matrix effect was illustrated by the incapability of the PLSR models to predict the graphite contents of the artificial soil 1 or graphitic schist (Fig. 5 and Table 2).

In recent work, the TGA method has been tested and further developed for differentiating between carbonates and OC/OM (e.g. Apesteguia et al., 2018). First analysis of individual components in a quartz matrix revealed that graphite has a similar thermal stability as carbonates (Fig. 2) and overlap in the thermal region where dehydroxylation of various minerals takes place (Földvári, 2011; Fernaíndez et al., 2012). As shown in Figs. S2 and S3 and discussed by other studies, complete removal of carbonates from soil sample by acid fumigation is difficult and might affect further quantitative estimations. For graphite, estimation via TGA becomes challenging as the acid affects the thermal stability of other soil constituents and makes the sample hygroscopic (Agarwal and Bucheli, 2011; Apesteguia et al., 2018). Additionally, sample grinding in an agate disk mill, representing common homogenization process used for small sample amounts (10s of mg), introduce some changes in thermogravimetric patterns for some minerals, e.g. micas, but makes it also more “reactive” (Földvári, 2011). This would mean that mass loss peaks for minerals, like the used muscovite, can appear sharper and at lower temperatures in comparison with non-ground materials.

The best temperature range to relate mass loss to the amount of added graphite was between 680 and 840 ∘C. For calibration set 2 (pure quartz matrix) a lower temperature (range) would also be able to predict the graphite content (Fig. S4), indicating interferences in the soil matrix of calibration set 1. The best temperature range was in line with the observation that the mass loss peak of graphite spans a large range (Fig. 2), most likely a result of the slow oxidation of this pure C. Other studies found that graphite in a (fluid) sand bed already oxidized slowly under oxygen-rich conditions at temperatures below 670 ∘C accelerating at higher temperatures (Hayhurst et al., 1998; Bews et al., 2001).

Validation of the created models from the two calibration sets revealed that interference with other soil components required an individual calibration for every sample set of specific (mineral) composition (Fig. 5). As shown in this study, fresh but ground muscovite dehydroxylates between 600 and 1000 ∘C (Fig. 2), which influenced the total mass loss measured in this temperature range. Other present (minor) minerals, like chlorites (500–860 ∘C) or apatites (200–1400 ∘C), might also increase the bias by influencing mass loss (Földvári, 2011; Tõnsuaadu et al., 2011). This observation could explain why the model of calibration set 1, using the soil spiked with graphite, showed a good predictability of the graphite content in the graphitic schist, as mineral composition is highly similar between these two samples (Fig. 5 and Table 2).

Roth et al. (2012) suggested that the use of anoxic conditions/a pyrolysing phase during measurement might be useful way to differentiate between wood and black C. According to our gas switching experiment with the TGA, it is at least a useful approach to differentiate between graphite and CaCO3 (Fig. 4). Due to the artificial mass gain/loss induced by switching the gases during the measurement, exact temperature ranges for developing a quantitative method could not be established. As no universally accepted temperature limits for the quantification of TOC, TIC or other carbonaceous substances exist, the best temperature ranges for switching between oxic and anoxic conditions are difficult to define (Pallasser et al., 2013; Ussiri et al., 2014). For instance, according to the DIN19539 standard, TOC is defined as the oxidizable C at a maximum of 400 ∘C. Others showed that 1/5 or even 1/3 of the TOC is not oxidized at 400 ∘C (Pallasser et al., 2013; Schiedung et al., 2017). For the artificial soil in our study a temperature limit of 400 ∘C seems to be too low to oxidize all OM, as indicated by the TGA in Fig. 2, and therefore the pyrolysing phase was set to 500 ∘C. To obtain a clear peak for the graphite oxidation, it is important that the other substances, i.e. CaCO3, are already decomposed. In the case of the artificial soil, it was found that at 850 ∘C all CaCO3 was decomposed and a clear peak for graphite was formed upon re-establishing the oxic conditions (Fig. 4). With higher CaCO3 levels or dolomitic carbonates, a higher temperature might be needed to create a clear separation between the substances (Földvári, 2011).

TGA seemed to be a good method to identify different organic components of samples and thus can be used as a complemental technique to other methods for (organic) C content estimation. For high graphite content with negligible amounts of dehydroxylating minerals and/or decomposing carbonates, TGA might be a useful method to quantify graphite.

With the TGA method it was already shown that qualitative differentiation between carbonates and graphite was possible by changing between oxic and anoxic conditions during heating of the sample (Fig. 4). Using the Soli-TOC device, a direct measurement of the released C could be achieved during the heating/gas changing program, which correspond very closely to the amount of (added) graphite (Fig. 5 and Table 2). The fact that the Soli-TOC device measured almost the same TC values as the elemental analyser (Fig. S5) supported the idea that a direct comparison between the ROC fraction and (added) graphite content is possible.

As shown by Hayhurst and Parmar (1998), very small impurities in the graphite can cause a small part of the graphite to pyrolyse during anoxic conditions at higher temperatures. Graphitic C of lesser graphitization might therefore result in a larger loss of graphitic C during pyrolysis and a greater underestimation of the graphitic C content. Taking a closer look at the measurements of the artificial soil reveals that a small part of the graphite began to oxidize under anoxic conditions (Fig. 8). The measurement of graphite in quartz, as done for calibration set 2, showed that about 6 % of the total C was lost during the pyrolysis phase, while for the graphitic schist this loss was 2 % (Fig. 8c), resulting in a small underestimation of the graphitic C content. Bews et al. (2001) suggested that at temperatures higher than 700 ∘C, radicals like HO2 and OH might act as reactants with the pure C. Furthermore, in the method comparison study for recovering different black C types, Roth et al. (2012) suggested a (relatively) strong catalytic effect of oxides on black C oxidation, which was most predominant in soils. These ideas are also supported by our observation that artificial soil 2 (without CaCO3) measured higher ROC values (0.06 % more C absolute) than artificial soil 1 (with CaCO3, Table 2). Also in the carbonate-rich AB soil the added graphite was underestimated by 7 % (Fig. 5). Furthermore, graphitic C was underestimated with increasing CaCO3 content (Fig. 6). The 7 % underestimation by the AB soil, which contains 1.87 % C–CaCO3, coincided with ROC underestimation of the calibration soil with 2 %C–CaCO3 addition.

When the soil contains more thermally resistant OM, which is not oxidizable at 400 ∘C and can be 1/4 of the OM (Schiedung et al., 2017), the question arises whether this fraction is pyrolysed during heating under anoxic conditions or whether it is taken as part of the ROC fraction when oxygen is again available. The TGA method showed that not all OM has been oxidized at 400 ∘C (Fig. 2). Taking a closer look at the smart combustion measurement of the artificial soil 2 and particularly its fresh OM component (Fig. 8) revealed that a small peak formed upon heating the sample above 400 ∘C. This peak represents only a few percent of the total OC, but clearly indicates that OM was oxidized incompletely at 400 ∘C. Although this study focuses on the ROC component, this point might be important when considering the TOC400 and TIC fractions of the smart combustion method.

When a sample contains other forms of thermally resistant OM or even black C, which are not pyrolysed during the anoxic phase, this C component is likely to end up in the graphitic C fraction with the smart combustion method. Especially as most temperature boundaries are empirically derived (Pallasser et al., 2013; Ussiri et al., 2014), a pre-test with continuous heating under oxic conditions is therefore recommended to assess the number of C-containing substances present in the sample by the occurrence of peaks. Further studies should focus on temperature boundaries of different substances in relation to their properties and see, for instance, how graphitic C can be distinguished from other thermally stable C components.

Although the thermal boundaries for the different C fractions given in the DIN19539 (GS) standard are debatable (Ussiri et al., 2014; Schiedung et al., 2017), we showed that the ROC fraction corresponded closely to the graphitic C content. Through the smart combustion method, graphitic C could be differentiated from the other C components in the soil matrix and quantified satisfactorily, as indicated by the offset in the calibration with graphite estimation in the artificial/spiked samples.

Comparing the ability of the examined methods on predicting graphite content, it becomes clear that FTIR overestimated, TGA was highly variable, and smart combustion was most accurate in predicting the graphite content (Fig. 5). An interesting observation was the similar predicted graphite content in calibration set 1 by both the FTIR and TGA methods, especially as FTIR is based on spectral properties and TGA on the thermal stability of the graphite. It has previously been suggested to combine FTIR and TGA systems to rapidly characterize the soil OM (Demyan et al., 2013). Oxidation of graphite upon heating could result in specific infrared absorption bands (Tan et al., 2013), which nevertheless would still be superimposed by SOM-specific bands in natural soil samples. As discussed by Demyan et al. (2013), not only the available oxygen but also the heating rate has an important effect on the charring of OM and thereby on the thermal and spectral properties of the studied material.

As Raman spectroscopy is suitable for distinguishing graphitic C and determining its degree of graphitization, it seems to be a promising method. Nonetheless, to use Raman spectroscopy for quantification of substances in a soil matrix, further studies should first focus on standardization of sample preparation, as it has a large influence on the measured intensities and baseline determination and thereby the direct quantification of components (Beysacc and Lazzeri, 2012; Sparkes et al., 2013).

Because we focused in this study on the ROC component, which significantly correlated with the graphite content, a consideration of other components mentioned in the DIN19539 standard was beyond the scope of this study. Nonetheless, we found indications that the thermal boundaries defined in the DIN19539 standard are not ideal for differentiating between soil OM and inorganic C (Fig. 8). As most carbonates start to decompose at temperatures of 550 ∘C (Földvári, 2011), it might be more suitable to increase the temperature level from 400 to 500 ∘C for the TOC component. Only when black C, which might oxidize between 375 and 540 ∘C (Roth et al., 2012), is present in the sample might this lead to an overestimation of the TOC content. Using TGA simultaneously with differential scanning calorimetry, CO2/H2O flux measurements (i.e. evolved gas analysis, Fernández et al., 2012) or the Rock-Eval method focusing on hydrocarbon, CO2 and CO release (Behar et al., 2001) could improve the development of a more standardized method applicable to soils using combustion elemental analysers. The overlap between the thermal properties of different C components emphasizes the need to consider the potential as well as possible interferences with the considered methodology before applying a fast and standardized analytic method.