 Methodology
 Open access
 Published:
An automatic peak deconvolution code for Raman spectra of carbonaceous material and a revised geothermometer for intermediate to moderately highgrade metamorphism
Progress in Earth and Planetary Science volumeÂ 11, ArticleÂ number:Â 35 (2024)
Abstract
Carbonaceous material (CM) undergoes progressive changes that reflect its thermal history. These changes are in general irreversible and provide valuable information for understanding diagenetic and metamorphic processes of crustal rocks. Among various approaches to quantify these changes, the R2 ratio, area ratio of specific peaks in CM Raman spectra, is widely used to estimate the maximum temperature of intermediate to moderately highgrade metamorphism. The calculation of the R2 ratio requires peak deconvolution of the original spectrum, and the results depend on the details of how this is carried out. However, a clear protocol for selecting appropriate initial conditions has not been established and obtaining a reliable temperature estimate depends at least in part on the experience and skill of the operator. In this study, we developed a Python code that automatically calculates the R2 ratio from CM Raman spectra. Our code produces R2 ratios that are generally in good agreement with those of Aoya et al. (J Metamorph Geol 28:895â€“914, 2010, https://doi.org/10.1111/j.15251314.2010.00896.x) for the same Raman data, with much less time and effort than was the case in the previous studies. We have confirmed that the code is also applicable to other previous datasets from both contact and regional metamorphic regions. The overall trend of the recalculated data indicates that samples with R2 greater thanâ€‰~â€‰0.7 are not sensitive to the changes in CM maturity and thus should not be used for the calibration of an R2based geothermometer. We propose a modified geothermometer for contact metamorphism that is strictly applicable to samples with R2 from 0.023 to 0.516, with the proviso that a laser with a wavelength of 532Â nm should be used. A slight extrapolation of the newly proposed geothermometer up to R2 of 0.57 provides a temperature estimate that is consistent with the geothermometer of Kaneki and Kouketsu (Island Arc 31:e12467, 2022; https://doi.org/10.1111/iar.12467); the boundary between the two geothermometers corresponds to a temperature of 391Â Â°C.
1 Introduction
The thermal history of crustal rocks provides important insights into the diagenetic and metamorphic processes of the Earthâ€™s interior. Carbonaceous material (CM) derived from organic matter is a common constituent of sediments and metasedimentary rocks. CM undergoes irreversible changes in its organochemical and crystallographic featuresâ€”referred to collectively as its maturityâ€”as temperature increases. Quantification of these features of CM has been widely used as a proxy for the thermal history experienced by the host rock. The maturity of CM has been investigated using numerous different analytical techniques such as Xray diffraction analysis, Raman spectrometry, and elemental analysis (e.g., Buseck and Huang 1985; French 1964; Ganz and Kalkreuth 1987; McKirby and Powell 1974; Seifert 1978; Sweeney and Burnham 1990; Tuinstra and Koenig 1970; van Krevelen 1993). The changes in CM maturity are generally thought to depend on time and temperature alone and be irreversible. However, deformation and in particular brittle deformation may also influence the results (e.g., Nakamura et al. 2015).
Over the past three decades, the use of Raman spectroscopy has dramatically increased in popularity among researchers in the geosciences (see Fig.Â 1 in Henry et al. 2019 and Kouketsu 2023). A Raman spectrum of CM typically exhibits two bands at wavenumbers of 1355Â cm^{â€“1} (disordered band) and 1575Â cm^{â€“1} (graphite band), which are attributed to the disordered and ordered graphitic structures, respectively (Tuinstra and Koenig 1970). Some Raman parameters associated with these two peaks, such as width, position, intensity, and area, have been reported to show close correlation with the maturity of CM (e.g., Beyssac et al. 2002; Kouketsu et al. 2014; Schito et al. 2017; Roberts et al. 1995), as reviewed by Henry et al. (2019). Beyssac et al. (2002) proposed the first empirical geothermometer to estimate the maximum temperature \(T\) during regional metamorphism from the area ratio of specific peaks, i.e., the R2 ratio, as follows:
\(\text{R}2\) was defined as
where \({A}_{\text{D}1}\), \({A}_{\text{D}2}\), and \({A}_{\text{G}}\) are the areas of the D1, D2, and Gbands. Following their pioneering work, various empirical geothermometers based on Raman spectra of CM have been developed for terrestrial rocks (e.g., Aoya et al. 2010; Kouketsu et al. 2014; Lahfid et al. 2010; LÃ¼nsdorf et al. 2017; Rahl et al. 2005). Mori et al. (2015a, 2017) performed thermal modeling of contact metamorphism and concluded that the heating duration required for CM maturation to attain steady state for a given temperature is greater than one hundred years. Throughout the present paper, we assume that such steady state conditions have been achieved for all samples. The idea of estimating \(T\) from CM Raman spectra has been extended to fault rocks (e.g., Hirono et al. 2015; Kaneki et al. 2016; Mukoyoshi et al. 2018) and extraterrestrial rocks (e.g., Busemann et al. 2007; Cody et al. 2008; Homma et al. 2015), although the direct application of the geothermometers to these types of rocks may not be appropriate, due to large differences in the environments where CM matures (e.g., Homma et al. 2015; Kouketsu et al. 2017).
Of the CMbased geothermometers that have been proposed (e.g., Aoya et al. 2010; Beyssac et al. 2002; Kouketsu et al. 2014; Lahfid et al. 2010; LÃ¼nsdorf et al. 2017; Rahl et al. 2005), those of Kouketsu et al. (2014) and Aoya et al. (2010) (hereafter, referred to as K2014 and A2010, respectively) are among the most widely used. Numerous studies have used one of these geothermometers, or a combination, to estimate \(T\) of the samples of interest (e.g., Bonneville et al. 2020; Cavalazzi et al. 2021; HickmanLewis et al. 2020; Mori et al. 2015a, b; Nakamura et al. 2019; Shimura et al. 2021; Yamaoka et al. 2022). The geothermometers of K2014 and A2010 are based on the widths of specific peaks and \(\text{R}2\), respectively, and have distinct but complementary applicable temperature ranges: for K2014 it is 150â€“400Â Â°C and for A2010 it is 340â€“655Â Â°C. Although both approaches require peak deconvolution of the measured Raman spectra, the criteria for setting the initial conditions for nonlinear leastsquares fitting are not clearly presented in the original papers, meaning temperature estimates could vary depending on the skill and experience of the analysts. Kaneki and Kouketsu (2022) (hereafter, referred to as KK2022) addressed this issue by developing a Python code that automatically performs peak deconvolution to calculate the widths of the D1 and D2bands. The main objective of the present study is to develop a similar code to calculate \(\text{R}2\).
We first describe the fitting procedures implemented in our code to reproduce \(\text{R}2\) of A2010 for the same dataset. The results obtained are then compared with those reported by A2010 for the validation of the code. Finally, we assess the utility of our proposed approach including discussion of its limitations based on results of applying it to datasets other than those of A2010.
2 Methods
In the development of our code, we reanalyzed the same Raman spectral dataset as that measured and analyzed in A2010. The basic structure of the code is similar to that of Fitting D in KK2022. Note that we developed our code by trial and error to reproduce the analytical results of A2010 as closely as possible for the same dataset. For this purpose, we tried to follow the analytical procedures of A2010 as closely as possible (e.g., fitting function, order of a baseline, number of peaks, etc.).
2.1 Data
We first briefly describe the samples and spectral data of A2010. Readers can refer to A2010 for more detailed information, including their experimental setups. Since the present study focuses on the development, validation, and application of our new code, we will not examine the procedure and methodology used in the original experiments reported in A2010. A total of ten analyzed samples was studied, of which two, K08 and N33, were excluded from the final geothermometer of the present study (Table 1). The detailed reasons for this selection are discussed in Sect.Â 4.3. All samples experienced contact metamorphism, and their independently estimated \(T\) range from 340â€‰Â±â€‰25Â Â°C to 655â€‰Â±â€‰25Â Â°C. A2010 measured at least 50 spectra for each sample and a total of 809 spectra. Most of the spectra show two distinct peaks of the disordered and graphite bands (Fig.Â 1a), except for highly matured CM.
A2010 examined the effect of measurement conditions on the analytical results by varying the magnification of the objective lens (50â€‰Ã—â€‰or 100â€‰Ã—), sample type (thin section or polished chip), and incident angle of laser beam (normal or parallel to the caxis of the crystal structure of graphite). They concluded that these variations in the experimental conditions had a limited effect on the calculated \(\text{R}2\) (Fig.Â 8 in A2010).
2.2 Code description
2.2.1 Fitting function
A2010 employed a Voigtian, \(V\), in their fitting procedures, which is defined as the convolution of a Gaussian, \(G\), and a Lorentzian, \(L\), as follows:
where \(\omega\) is the center position, \(A\) is the area, and \(\sigma\) and \(\gamma\) represent the widths of \(G\) and \(L\), respectively. \(G\) and \(L\) are expressed as
Although we need to determine the values of the arguments of \(V\) as an initial condition for nonlinear leastsquares fitting, it is difficult to constrain \(\sigma\) and \(\gamma\) from a raw Raman spectrum. Some previous studies (e.g., Kouketsu et al. 2014; LÃ¼nsdorf and LÃ¼nsdorf 2016) employed a pseudoVoigtian, \(PV\), for peak deconvolution. One can define a typical \(PV\) as follows:
where \(\eta\) is the mixing ratio of \(G\) and \(L\), and \(\Gamma\) is full width at half maximum (FWHM). \(PV\) in Eq.Â (6) takes its maximum intensity \(I\) at \(x=\omega\) as follows:
Note that the first approximation to all arguments of \(PV\) can be easily determined from a raw Raman spectrum. Since we are interested in the area ratio and the difference in \(A\) between \(V\) and \(PV\) is less than 1% (Fig. S1), we concluded that the usage of \(PV\) instead of \(V\) should have negligible effect on the calculated \(\text{R}2\).
2.2.2 Background correction and normalization
Background correction is required for each raw spectrum prior to peak deconvolution. Although it is obvious that A2010 employed a linear baseline, the details of the algorithm used to describe it are not clearly presented. In our code, we determined the slope and intercept of the linear baseline by a linear leastsquares method using the raw spectral data in the ranges 1100â€“1150Â cm^{â€“1} and 1700â€“1750Â cm^{â€“1} (Fig.Â 1a). However, for samples with a high degree of maturity and thus low effects of fluorescence, a small bulge sometimes appears around 1100Â cm^{â€“1} (attributable to the overlapping of spectra from quartz or calcite grains with the spectra from CM), which makes this procedure inapplicable (Fig. S2a). In this case, it is better to describe the linear baseline using the data in the ranges 1200â€“1250Â cm^{â€“1} and 1700â€“1750Â cm^{â€“1} (Fig. S2b). We defined a â€˜matureâ€™ spectrum as one with the maximum intensity at 1320â€“1380Â cm^{â€“1} (disordered band) less than half that at 1570â€“1600Â cm^{â€“1} (graphite band) after the baseline correction. If a given spectrum satisfies this criterion, the linear baseline is described using the spectral data in the ranges 1200â€“1250Â cm^{â€“1} and 1700â€“1750Â cm^{â€“1}, rather than 1100â€“1150Â cm^{â€“1} and 1700â€“1750Â cm^{â€“1}.
The baselinecorrected spectrum was then normalized such that its maximum intensity in the range 1100â€“1750Â cm^{â€“1} becomes unity (Fig.Â 1b). This normalized baselinecorrected spectrum is referred to as the â€˜measured spectrumâ€™ in the present study, and we used this spectrum for the peak deconvolution.
2.2.3 Initial conditions and peak deconvolution
To perform the nonlinear leastsquares fitting of the measured spectrum, the initial conditions for each peak should be determined. The approach to setting the initial conditions in our code is similar to that of Fitting D in KK2022. The D1, D2, D3, and Gbands of \(PV\) are assumed to be always present in all the measured spectra. The maximum intensity of the measured spectrum in the range 1320â€“1380Â cm^{â€“1} and the associated wavenumber are employed as the initial values for \(I\) and \(\omega\) of the D1band, respectively. The same analysis is applied to the Gband in the range 1570â€“1600Â cm^{â€“1}. The initial \(\Gamma\) of the D1band is set as the distance between two wavenumbers where the intensities are half the initial \(I\) in the ranges from 1200Â cm^{â€“1} to initial \(\omega\) and from initial \(\omega\) to 1500Â cm^{â€“1}. That of the Gband is set to twice the distance between initial \(\omega\) and the wavenumber at the point where the intensity is half the initial \(I\) in the range from 1500Â cm^{â€“1} to initial \(\omega\). These approaches to determining appropriate values for the initial \(\Gamma\) allow us to analyze Raman spectra of CM with different maturities using a single code, eliminating the subjectivity owing to the selection of the fitting method (a detailed discussion of this point can be found in Sect. 4.3 in KK2022). The D1 and Gbands are initially assumed to be a pure Lorentzian, and thus \(\eta =1\). To determine the initial conditions of the other two bands, we used the residual spectrum after subtracting the initial D1 and Gbands from the measured spectrum (Fig. S3). For the D2band, 90% of the maximum intensity of the residual spectrum in the range 1610â€“1630Â cm^{â€“1} and the associated wavenumber are employed as the initial values for \(I\) and \(\omega\), respectively. The same approach is employed for the D3band in the range 1500â€“1540Â cm^{â€“1}. The initial \(\Gamma\) of the D2band is set to twice the distance between the initial \(\omega\) and the wavenumber where the intensity is half the initial \(I\) in the range from initial \(\omega\) to 1700Â cm^{â€“1}. That of the D3band is set to 130Â cm^{â€“1}, as in Fitting D of KK2022. The D2 and D3bands are initially assumed to be half Gaussian and half Lorentzian, and thus \(\eta =0.5\). We calculated the initial \(A\) values of these four bands using Eq.Â (7).
After determining the initial conditions of the calculated spectrum (Fig.Â 1c), we performed nonlinear leastsquares fitting on the measured spectrum in the spectral range 1100â€“1750Â cm^{â€“1} (Fig.Â 1d). We used scipy.optimize.curve_fit function with the Trust Region Reflective algorithm (SciPy v1.10.1), which is applicable even when the bounds on the model parameters are given. We defined the cost function as half the sum of squares of the residual of the measured spectrum after subtracting the calculated spectrum. The maximum number of iterations was set to 10,000. All values of tolerances during the iterations were set to 10^{â€“8}. \(\Gamma\) and \(A\) of the four bands can take any positive values. The value of \(\eta\) can move from zero to unity. The \(\omega\) values can vary 1300â€“1400Â cm^{â€“1}, 1600â€“1650Â cm^{â€“1}, 1450â€“1550Â cm^{â€“1}, and 1550â€“1600Â cm^{â€“1} for the D1, D2, D3, and Gbands, respectively. The coefficient of determination, \({R}^{2}\), was calculated for each fitting curve:
where \({I}_{i}^{\text{meas}}\) and \({I}_{i}^{\text{calc}}\) are the intensities of the measured and calculated spectra, and \({\overline{I} }^{\text{calc}}\) is the mean intensity of the calculated spectrum in the range 1100â€“1750Â cm^{â€“1}.
For each sample, we calculated the values of the mean, standard deviation (SD), and standard error (SE) of \(\text{R}2\) of the calculated spectra after optimization. As in A2010, we then excluded the data with \(\text{R}2\) deviating by more thanâ€‰Â±â€‰2SD from the mean value. After this treatment, we recalculated the mean, SD, and SE values.
3 Results
3.1 Performance of the code
FigureÂ 2 shows examples of the initial conditions together with the measured spectra for each sample. The initial fitting curves show a good match with the measured data with \({R}^{2}>0.97\). After optimization, the \({R}^{2}\) of the calculated spectra exceed 0.99 regardless of the sample (Fig.Â 3). Since the fitting is performed automatically, the same analytical results are obtained for the same spectral data, irrespective of the analyst. Our code typically analyzes one spectrum in less than one second using an ordinary laptop computer and thus greatly reduces the time and effort required to arrive at a result.
3.2 Comparison with Aoya et al. (2010)
FigureÂ 4 compares \(\text{R}2\) of the present study with those reported by A2010. Error bars indicate the SD. As described in Sect.Â 2.1, A2010 investigates the effect of measurement conditions on the calculated \(\text{R}2\) by varying the sample type, magnification of an objective lens, and incident angle of laser beam. FigureÂ 4a shows the comparisons of \(\text{R}2\) for each measurement condition. Our codegenerated results are generally in good agreement with those of A2010 within the range of SD, regardless of the measurement conditions. When \(\text{R}2\) exceedsâ€‰~â€‰0.6, the D2 and Gbands were not well decomposed for some samples (e.g., K08 in Figs.Â 2 and 3), and the results of the present study showed significantly larger \(\text{R}2\) than those of A2010 with differences exceeding one SD. These trends also hold even when the data measured under different conditions are summed and then analyzed to obtain representative \(\text{R}2\) for each sample (Fig.Â 4b). These results are consistent with the negligible effects on \(\text{R}2\) by varying the measurement conditions as examined by A2010. All data are summarized in Table 1.
3.3 Modified geothermometer
Instead of a linear model proposed by Beyssac et al. (2002) (Eq.Â (1)), A2010 presents a quadratic equation to express the relationship between \(\text{R}2\) and \(T\):
Ideally, \(\text{R}2\) calculated using our code should be suitable to use Eq.Â (9) for estimating \(T\). However, there are differences in methodology of the present study that make this simple approach inappropriate. (1) The present study uses a more limited sample set than A2010 with the data not only for N33 but also for K08 removed when defining the calibration curve. (2) Our codegenerated \(\text{R}2\) were not identical to those of A2010 even for the same spectra (Fig.Â 4 and Table 1). (3) The uncertainties in both \(\text{R}2\) and \(T\) should be taken into consideration when formulating the calibration curve. Due to these differences, we concluded that the previous geothermometer (Eq.Â 9) needs to be modified when employing our code, which is the main issue addressed in this subsection.
Since the calibration data have uncertainties in both \(\text{R}2\) and \(T\) (Table 1), the coefficients of the quadratic function cannot be calculated as a unique solution to a linear equation. Therefore, we performed the Deming regression (Text S1), which is applicable to a regression problem that requires iterative treatments to optimize the model parameters. Samples K08 and N33 were excluded from our regression analysis, owing to their inappropriateness as the calibration samples (see Sect.Â 4.3 for the detailed discussion). Following KK2022, we assume that the errors in \(T\) of the calibration samples can be represented by the SE. We employed SE rather than SD as the uncertainties of each data point, since SE is required to perform the Deming regression (Text S1). Finally, the regression curve obtained is expressed as follows:
The geothermometer of Eq.Â (10) is strictly only applicable to samples with \(\text{R}2\) from 0.023 to 0.516 (Table 1). FigureÂ 5 shows the regression curve (Eq.Â (10)), its 95% prediction interval (Text S2), and the data points used to calculate them (Table 1). Although A2010 defined the uncertainty of Eq.Â (9) as the maximum difference between the estimated and known \(T\) of the calibration samples, we employed the prediction interval for the uncertainty of Eq.Â (10) owing to its clear statistical definition. Therefore, a direct comparison of the uncertainties between Eqs. (9) and (10) is inappropriate. The regression curve of Eq.Â (10) shows a very good fit to the data with \({R}^{2}=0.998\), which is similar to that of A2010 (\({R}^{2}=0.995\)). The optimization of the quadratic function by the iterative treatment affected the results of temperature estimation only modestly, withâ€‰~â€‰1Â Â°C difference in \(T\) between the regression curves after 0th and 10th iterations (Fig. S4). At a given \(\text{R}2\), the calculated \(T\) using Eq.Â (10) differs by a maximum ofâ€‰~â€‰10Â Â°C from the results of using Eq.Â (9) (Fig.Â 6), which is similar to the 95% prediction interval of Eq.Â (10) (~â€‰15Â Â°C). Note that Eq.Â (10) was established using the samples which underwent contact metamorphism, and its direct application to the samples which experienced other processes including regional metamorphism may not be appropriate, although our code may be useful to establish a new R2based geothermometer for these situations.
4 Discussion
4.1 Deviation from Aoya et al. (2010)
\(\text{R}2\) of the K08 sample in the present study was larger than that of A2010 by an amount that exceeds one SD (Fig.Â 4 and Table 1). Let us assume that a measured spectrum is perfectly explained by a calculated spectrum, i.e., \({R}^{2}\) in Eq.Â (8) is strictly unity. It then follows from Eq.Â (2) that
where \({A}_{\text{all}}\) represents the entire area of the measured spectrum in the spectral range 1100â€“1750Â cm^{â€“1} and thus takes a constant value. The Raman spectra of the K08 sample typically show a clear D4band at around 1200Â cm^{â€“1} and a strong â€˜saddleâ€™ intensity at around 1500Â cm^{â€“1} (e.g., Figs.Â 2 and 3). This indicates that the optimized \({A}_{\text{D}1}\) and \({A}_{\text{D}3}\) values may be strongly influenced by which algorithm was employed to perform the peak deconvolution. Therefore, the observed difference in \(\text{R}2\) between these two studies may be qualitatively explained if the optimization algorithm of the present study yields larger \({A}_{\text{D}1}\) and \({A}_{\text{D}3}\) than A2010.
4.2 Applicability of the code to other datasets
Since our code was developed by trial and error to reproduce \(\text{R}2\) of A2010 as closely as possible, its applicability to datasets other than those of A2010 should be examined. We reanalyzed the raw Raman spectral data of the natural samples measured and analyzed by Kouketsu et al. (2019, 2021), Shimura et al. (2021), and Yamaoka et al. (2022). These studies measured the Raman spectra of natural CM under the same experimental conditions as A2010, performed peak deconvolution by manually setting the initial conditions, and calculated \(\text{R}2\). \(\text{R}2\) of the samples to be reanalyzed were reported to range from 0.066â€‰Â±â€‰0.041 to 0.655â€‰Â±â€‰0.021 (Table S1).
The comparison of \(\text{R}2\) obtained in the previous and present studies is shown in Fig.Â 7. Error bars represent the SD. All the data obtained are summarized in Table S1. \(\text{R}2\) of the present study generally agree with those of the previous studies within the margin of SD. However, we note that when \(\text{R}2\) is greater thanâ€‰~â€‰0.6, our code yielded significantly larger \(\text{R}2\) than the previous studies exceeding one SD, something which was also recognized for the samples used in A2010 (Fig.Â 4). These results clearly indicate that our code is applicable to datasets other than those of A2010, although the codegenerated results significantly differ from the previous analysis when \(\text{R}2>\,\sim 0.6\).
4.3 Sensitivity of R2 as a proxy for CM maturity and its implications for use as a geothermometer
K2014 reported that FWHM of the D1band, \({\Gamma }_{\text{D}1}\), linearly decreases with increasing \(T\) while it shows almost no change after reachingâ€‰~â€‰40Â cm^{â€“1}. Therefore, we can consider \({\Gamma }_{\text{D}1}>40 \, {\text{cm}}^{1}\) as a necessary condition that \({\Gamma }_{\text{D}1}\) has its sensitivity to changes in CM maturity. This subsection focuses on the similar sensitivity test for \(\text{R}2\) and discusses its implications for practical temperature estimation.
We reexamined the raw spectral data of Aoya et al. (2010), Kouketsu et al. (2019, 2021), Nakamura et al. (2019), Shimura et al. (2021), and Yamaoka et al. (2022). \(\text{R}2\) was calculated using our code, while \({\Gamma }_{\text{D}1}\) was determined using the code for Fitting E of KK2022. FigureÂ 8 compares \(\text{R}2\) and \({\Gamma }_{\text{D}1}\) for each sample. FigureÂ 8b shows an enlarged view of the gray area in Fig.Â 8a. Error bars indicate the SD. The large SD of \({\Gamma }_{\text{D}1}\) of highly matured CM with \(\text{R}2\) ofâ€‰~â€‰0.1 were attributed to the disappearance of the D1band (e.g., K04 in Figs.Â 2 and 3). All the data obtained are summarized in Table S2. The results obtained show the same trend regardless of datasets (Fig.Â 8a). Within the region for lowgrade material (upper right part in Fig.Â 8a), \({\Gamma }_{\text{D}1}\) decreases significantly with increasing CM maturity while \(\text{R}2\) shows almost constant values of 0.6â€“0.8. \(\text{R}2\) starts to decrease fromâ€‰~â€‰0.7 toward zero when \({\Gamma }_{\text{D}1}\) decreases toâ€‰~â€‰55Â cm^{â€“1} (Fig.Â 8b). After \({\Gamma }_{\text{D}1}\) reachesâ€‰~â€‰40Â cm^{â€“1} at \(\text{R}2\) ofâ€‰~â€‰0.55, only \(\text{R}2\) shows a systematic decrease with CM maturity. Based on these results, we suggest that \(\text{R}2<\,\sim 0.7\) is a necessary condition for \(\text{R}2\) to be sensitive to changes in CM maturity. In other words, samples with \(\text{R}2\ge \,\sim 0.7\) should not be employed in the calibration of an R2based geothermometer. This is the reason why our regression curve (Eq.Â 10) did not include the K08 sample (\(\text{R}2=0.737\pm 0.029\)), which enables our geothermometer to yield better results around \(\text{R}2=0.516\) than that of A2010 (Eq.Â 9) who included the K08 sample (Fig.Â 6). Since very lowgrade CM sometimes exhibits \(\text{R}2\) ofâ€‰~â€‰0.7 (e.g., dataset of Nakamura et al. 2019 in Fig.Â 8a), the condition \(\text{R}2<\,\sim 0.7\) alone can be misleading for the assessment of CM maturity. It is important to check the consistency of the results obtained with other observations, including the strong influence of fluorescence (e.g., Schito et al. 2017) and the existence of the D4band (e.g., Kouketsu et al. 2014), both of which are absent in the spectra of intermediategrade CM (Fig. S5).
As described in Sect.Â 3.3, our modified geothermometer is determined in the range \(0.023\le \text{R}2\le 0.516\) (Fig.Â 5 and Table 1). However, the geothermometers based on \({\Gamma }_{\text{D}1}\), including that of KK2022, are applicable only when \(\text{R}2\) exceedsâ€‰~â€‰0.55, owing to the insensitivity of \({\Gamma }_{\text{D}1}\) to changes in CM maturity for lower \(\text{R}2\) (Fig.Â 8). Therefore, in a strict sense, a combination of the codebased geothermometers of KK2022 and the present study is incapable of estimating \(T\) of samples with \(\text{R}2\) from 0.516 toâ€‰~â€‰0.55 and \({\Gamma }_{\text{D}1}\) ofâ€‰~â€‰40Â cm^{â€“1}. A straightforward solution to this problem would be to refine our modified geothermometer further by considering additional calibration data of samples whose \(T\) are known, \({\Gamma }_{\text{D}1}\) areâ€‰~â€‰40Â cm^{â€“1}, and \(\text{R}2\) are larger thanâ€‰~â€‰0.55 while smaller thanâ€‰~â€‰0.7. Although the N33 sample almost satisfies this requirement (Table 1), the bimodal distribution in its \(\text{R}2\) histogram (Fig. S6) indicates its inappropriateness as a calibration sample. An alternative approach is to simply extrapolate the applicable range of Eq.Â (10), as implicitly done in some previous studies (e.g., Kaneki and Kouketsu 2022; Kouketsu et al. 2014; Rahl et al. 2005). In this case, we need to consider an appropriate extent for the extrapolation of our geothermometer. One potential clue can be obtained when we try to smoothly connect the geothermometers of KK2022 and the present study on the data trend shown in Fig.Â 8. The \({\Gamma }_{\text{D}1}\)based geothermometer of KK2022 is
By combining Eqs. (10) and (12), we obtain
When the calculated \(\left(\text{R}2,{\Gamma }_{\text{D}1}\right)\) satisfy Eq.Â (13), the two geothermometers yield the same \(T\). For example, if we assume the connecting \(T\) is 391Â Â°C, the corresponding \(\text{R}2\) and \({\Gamma }_{\text{D}1}\) are 0.57 and 41.3Â cm^{â€“1}, respectively (Eqs. (10) and (12)). It is noteworthy that the point \(\left(\text{R}2,{\Gamma }_{\text{D}1}\right)=\left(\text{0.57},\text{41.3} \,{\text{cm}}^{1}\right)\) (indicated by star symbol in Fig.Â 8b) plots well within the data trend, which is drawn by a number of recalculated data both from contact and regional metamorphisms. Therefore, the slight extrapolation of our geothermometer up to \(\text{R}2=0.57\) (within the standard deviation of the lowest\(T\) sample for our calibration curve, N30) provides a consistent switching to the lower\(T\) geothermometer of KK2022, with a boundary corresponding to \(T=391 \, ^\circ \text{C}\). In addition, setting \(\text{R}2=0.57\) as a boundary between the two geothermometers enables an unambiguous classification of all the data in Fig.Â 8 into one of two groups: using our geothermometer (Eq.Â (10)) if \(\text{R}2\le 0.57\) and using KK2022 (Eq.Â (12)) if \(\text{R}2>0.57\).
4.4 Effect of laser wavelength
Differences in the laser wavelength employed in the Raman system can significantly affect the spectral intensity and subsequent analytical results (e.g., Wang et al. 1990; Ferrari and Robertson 2001). K2014 and KK2022 demonstrated that for the Raman data of a specific sample measured under different experimental conditions (i.e., spectrometer, laser wavelength, pinhole size, and grating), some representative Raman parameters (i.e., \(\Gamma\), \(\omega\), intensity ratio, and area ratio) showed consistent values when the laser wavelength was the same, while the parameters other than \(\Gamma\) differed significantly for different laser wavelengths. These results clearly suggest that the effect of the laser wavelength on the spectral parameters dominates over the other variable experimental conditions. Since these two studies analyzed the data of the sample with \(T=165 \, ^\circ \text{C}\) and \(330 \, ^\circ \text{C}\), which is outside the temperature range of the calibration samples used in our regression analysis, this subsection investigates the dependence of the analytical results using our code on the laser wavelength for samples with higher \(T\).
We measured the Raman spectra of the N30 sample ï»¿(\(T=410 \, \pm \,30 \, ^\circ \text{C}\)) under the same experimental conditions as for A2010, except that we used a laser with a wavelength of 633Â nm instead of the original 532Â nm. Using our code, we calculated the following six Raman parameters: intensity ratio of the D1band to the Gband (i.e., R1 ratio), \(\text{R}2\), \(\Gamma\) of the D1 and Gbands, and \(\omega\) of the D1 and Gbands. The results obtained are compared with those of a 532Â nm laser in Fig.Â 9. Error bars indicate the SD. All the data obtained are summarized in Table S3. Of the six parameters examined, only \(\Gamma\) of the D1 and Gbands were consistent between two laser wavelengths within the margin of SD (Fig.Â 9b), similar to the results reported in K2014 and KK2022. The other four parameters, including \(\text{R}2\), showed significant differences exceeding one SD when the laser wavelength was changed from 532 to 633Â nm (Fig.Â 9a, c). The observed larger \(\text{R}2\) for the longer laser wavelength (Fig.Â 9a) may be attributed to (i) use of the longer laser wavelength resulting in the larger R1 ratio (e.g., Ferrari and Robertson 2001), and (ii) the positive correlation between the R1 and R2 ratios (e.g., Aoya et al. 2010). The mean \(T\) calculated using Eq.Â (10) were 407 and 364Â Â°C for a 532 and 633Â nm laser, respectively, and their difference exceeded 95% prediction intervals. Considering these facts, geothermometers based on \(\text{R}2\) cannot be reliably applied to spectra that were measured using a different laser wavelength from the one used to establish it. Therefore, the usage of our geothermometer (Eq.Â (10)) should be limited to data measured using a 532Â nm laser.
5 Conclusions
In this study, we have developed a Python code that automatically performs peak deconvolution of the Raman spectra of carbonaceous material and calculates the R2 ratio. A close comparison with the results of Aoya et al. (2010) shows very good agreement for the same original data unless \(\text{R}2\) exceedsâ€‰~â€‰0.6. Based on a Deming regression analysis, we propose a modified set of coefficients for the quadratic calibration curve first proposed by Aoya et al. (2010) to model the relationship between \(\text{R}2\) and the maximum metamorphic temperature \(T\). This modified curve is used in our code to calculate the mean \(T\) and its 95% prediction interval, although its direct application to samples experiencing the processes other than the contact metamorphism may not be appropriate. Our code is applicable to datasets other than those of Aoya et al. (2010), as long as the Raman spectral data are obtained using a 532Â nm laser. We document that \(\text{R}2<\,\sim 0.7\) is a necessary condition for \(\text{R}2\) to be sensitive to changes in the sample maturity. For this reason, our modified geothermometer was calibrated without the K08 sample (mean \(\text{R}2\) of 0.737) and should be preferred over the geothermometer of Aoya et al. (2010). Since a combination of the geothermometers of Kaneki and Kouketsu (2022) and the present study is incapable of estimating \(T\) of samples with maturity in a particular range, further refinement or a slight extrapolation of the present geothermometer is required. In the latter case, we propose \(\text{R}2=0.57\) as a possible boundary where users should switch from the geothermometer presented here to the one proposed by Kaneki and Kouketsu (2022).
Availability of data and materials
Readers can download the code developed in the present study from https://doi.org/10.5281/zenodo.10237726. In the development, validation, and application of the code, we reanalyzed the raw Raman spectral data originally measured and analyzed by Aoya et al. (2010), Kouketsu et al. (2019, 2021), Nakamura et al. (2019), Shimura et al. (2021), and Yamaoka et al. (2022). The raw Raman spectral data newly measured in the present study are used only to investigate the dependence of the analytical results on laser wavelength. Requests for access to the new data should be sent to YK. All analytical results obtained are listed in tables either in the main text or in Supporting Information.
Abbreviations
 A2010:

Aoya et al. (2010)
 CM:

Carbonaceous material
 FWHM:

Full width at half maximum
 KK2022:

Kaneki and Kouketsu (2022)
 K2014:

Kouketsu et al. (2014)
 SD:

Standard deviation
 SE:

Standard error
References
Aoya M, Kouketsu Y, Endo S et al (2010) Extending the applicability of the Raman carbonaceousmaterial geothermometer using data from contact metamorphic rocks. J Metamorph Geol 28:895â€“914. https://doi.org/10.1111/j.15251314.2010.00896.x
Beyssac O, GoffÃ© B, Chopin C, Rouzaud JN (2002) Raman spectra of carbonaceous material in metasediments: a new geothermometer. J Metamorph Geol 20:859â€“871. https://doi.org/10.1046/j.15251314.2002.00408.x
Bonneville S, Delpomdor F, PrÃ©at A et al (2020) Molecular identification of fungi microfossils in a neoproterozoic shale rock. Sci Adv 6:eaax7599. https://doi.org/10.1126/sciadv.aax7599
Buseck PR, Huang BJ (1985) Conversion of carbonaceous material to graphite during metamorphism. Geochim Cosmochim Acta 49:2003â€“2016. https://doi.org/10.1016/00167037(85)900596
Busemann H, Alexander MO, Nittler LR (2007) Characterization of insoluble organic matter in primitive meteorites by micro Raman spectroscopy. Meteorit Planet Sci 42:1387â€“1416. https://doi.org/10.1111/j.19455100.2007.tb00581.x
Cavalazzi B, Lemelle L, Simionovici A et al (2021) Cellular remains in a ~3.42billionyearold subseafloor hydrothermal environment. Sci Adv 7:eabf3963. https://doi.org/10.1126/sciadv.abf3963
Cody GD, Alexander CMO, Yabuta H et al (2008) Organic thermometry for chondritic parent bodies. Earth Planet Sci Lett 272:446â€“455. https://doi.org/10.1016/j.epsl.2008.05.008
Ferrari AC, Robertson J (2001) Resonant Raman spectroscopy of disordered, amorphous, and diamondlike carbon. Phys Rev B 64:075414. https://doi.org/10.1103/PhysRevB.64.075414
French BM (1964) Graphitization of organic material in a progressively metamorphosed Precambrian iron formation. Science 146:917â€“918. https://doi.org/10.1126/science.146.3646.917
Ganz H, Kalkreuth W (1987) Application of infrared spectroscopy to the classification of kerogentypes and the evaluation of source rock and oil shale potentials. Fuel 66:708â€“711. https://doi.org/10.1016/00162361(87)902857
Henry DG, Jarvis I, Gillmore G, Stephenson M (2019) Raman spectroscopy as a tool to determine the thermal maturity of organic matter: application to sedimentary, metamorphic and structural geology. Earth Sci Rev 198:102936. https://doi.org/10.1016/j.earscirev.2019.102936
HickmanLewis K, Cavalazzi B, Sorieul S et al (2020) Metallomics in deep time and the influence of ocean chemistry on the metabolic landscapes of Earthâ€™s earliest ecosystems. Sci Rep 10:4965. https://doi.org/10.1038/s4159802061774w
Hirono T, Maekawa Y, Yabuta H (2015) Investigation of the records of earthquake slip in carbonaceous materials from the T aiwan C helungpu fault by means of infrared and R aman spectroscopies. Geochem Geophys Geosyst 16:1233â€“1253. https://doi.org/10.1002/2014GC005622
Homma Y, Kouketsu Y, Kagi H et al (2015) Raman spectroscopic thermometry of carbonaceous material in chondrites: fourâ€“band fitting analysis and expansion of lower temperature limit. J Mineral Petrol Sci 110:276â€“282. https://doi.org/10.2465/jmps.150713a
Kaneki S, Kouketsu Y (2022) An automatic peak deconvolution method for Raman spectra of terrestrial carbonaceous material for application to the geothermometers of Kouketsu et al. (2014). Island Arc 31(1):e12467. https://doi.org/10.1111/iar.12467
Kaneki S, Hirono T, Mukoyoshi H et al (2016) Organochemical characteristics of carbonaceous materials as indicators of heat recorded on an ancient platesubduction fault. Geochem Geophys Geosyst 17:2855â€“2868. https://doi.org/10.1002/2016GC006368
Kouketsu Y (2023) Fusion of spectroscopy and geology: earthÊ¼s interior environment revealed through light. Jpn Magaz Mineral Petrol Sci 52:gkk.230110a. https://doi.org/10.2465/gkk.230110a
Kouketsu Y, Mizukami T, Mori H et al (2014) A new approach to develop the Raman carbonaceous material geothermometer for lowgrade metamorphism using peak width. Island Arc 23:33â€“50. https://doi.org/10.1111/iar.12057
Kouketsu Y, Shimizu I, Wang Y et al (2017) Raman spectra of carbonaceous materials in a fault zone in the Longmenshan thrust belt, China; comparisons with those of sedimentary and metamorphic rocks. Tectonophysics 699:129â€“145. https://doi.org/10.1016/j.tecto.2017.01.015
Kouketsu Y, Tsai CH, Enami M (2019) Discovery of unusual metamorphic temperatures in the Yuli belt, eastern Taiwan: New interpretation of data by Raman carbonaceous material geothermometry. Geology 47:522â€“526. https://doi.org/10.1130/G45934.1
Kouketsu Y, Sadamoto K, Umeda H et al (2021) Thermal structure in subducted units from continental Moho depths in a palaeo subduction zone, the Asemigawa region of the Sanbagawa metamorphic belt, SW Japan. J Metamorph Geol 39:727â€“749. https://doi.org/10.1111/jmg.12584
Lahfid A, Beyssac O, Deville E et al (2010) Evolution of the Raman spectrum of carbonaceous material in lowgrade metasediments of the Glarus Alps (Switzerland): RSCM in lowgrade metasediments. Terra Nova 22:354â€“360. https://doi.org/10.1111/j.13653121.2010.00956.x
LÃ¼nsdorf NK, LÃ¼nsdorf JO (2016) Evaluating Raman spectra of carbonaceous matter by automated, iterative curvefitting. Int J Coal Geol 160â€“161:51â€“62. https://doi.org/10.1016/j.coal.2016.04.008
LÃ¼nsdorf NK, Dunkl I, Schmidt BC et al (2017) Towards a higher comparability of geothermometric data obtained by raman spectroscopy of carbonaceous material. Part 2: a revised geothermometer. Geostandard Geoanal Res 41:593â€“612. https://doi.org/10.1111/ggr.12178
McKirdy DM, Powell TG (1974) Metamorphic alteration of carbon isotopic composition in ancient sedimentary organic matter: new evidence from Australia and South Africa. Geol 2:591. https://doi.org/10.1130/00917613(1974)2%3c591:MAOCIC%3e2.0.CO;2
Mori N, Wallis S, Mori H (2015a) Graphitization of carbonaceous material in sedimentary rocks on short geologic timescales: An example from the K inshozan area, central J apan. Island Arc 24:119â€“130. https://doi.org/10.1111/iar.12093
Mori H, Wallis S, Fujimoto K, Shigematsu N (2015b) Recognition of shear heating on a longlived major fault using Raman carbonaceous material thermometry: implications for strength and displacement history of the MTL, SW Japan. Island Arc 24:425â€“446. https://doi.org/10.1111/iar.12129
Mori H, Mori N, Wallis S et al (2017) The importance of heating duration for Raman CM thermometry: evidence from contact metamorphism around the Great Whin Sill intrusion, UK. J Metamorph Geol 35:165â€“180. https://doi.org/10.1111/jmg.12225
Mukoyoshi H, Kaneki S, Hirono T (2018) Slip parameters on major thrusts at a convergent plate boundary: regional heterogeneity of potential slip distance at the shallow portion of the subducting plate. Earth Planets Space 70:36. https://doi.org/10.1186/s406230180810z
Nakamura Y, Oohashi K, Toyoshima T et al (2015) Straininduced amorphization of graphite in fault zones of the Hidaka metamorphic belt, Hokkaido, Japan. J Struct Geol 72:142â€“161. https://doi.org/10.1016/j.jsg.2014.10.012
Nakamura Y, Hara H, Kagi H (2019) Natural and experimental structural evolution of dispersed organic matter in mudstones: The Shimanto accretionary complex, southwest Japan. Island Arc 28:e12318. https://doi.org/10.1111/iar.12318
Rahl J, Anderson K, Brandon M, Fassoulas C (2005) Raman spectroscopic carbonaceous material thermometry of lowgrade metamorphic rocks: calibration and application to tectonic exhumation in Crete, Greece. Earth Planet Sci Lett 240:339â€“354. https://doi.org/10.1016/j.epsl.2005.09.055
Roberts S, Tricker PM, Marshall JEA (1995) Raman spectroscopy of chitinozoans as a maturation indicator. Org Geochem 23:223â€“228. https://doi.org/10.1016/01466380(94)00126L
Schito A, Romano C, Corrado S et al (2017) Diagenetic thermal evolution of organic matter by Raman spectroscopy. Org Geochem 106:57â€“67. https://doi.org/10.1016/j.orggeochem.2016.12.006
Seifert WK (1978) Steranes and terpaues in kerogen pyrolysis for correlation of oils and source rocks. Geochim Cosmochim Acta 42:473â€“484. https://doi.org/10.1016/00167037(78)901977
Shimura Y, Tokiwa T, Mori H et al (2021) Deformation characteristics and peak temperatures of the Sanbagawa Metamorphic and Shimanto Accretionary complexes on the central Kii Peninsula, SW Japan. J Asian Earth Sci 215:104791. https://doi.org/10.1016/j.jseaes.2021.104791
Sweeney JJ, Burnham AK (1990) Evaluation of a simple model of vitrinite reflectance based on chemical kinetics. Am Assoc Petrol Geol Bull 74:1559â€“1570
Tuinstra F, Koenig JL (1970) Raman spectrum of graphite. J Chem Phys 53:1126â€“1130. https://doi.org/10.1063/1.1674108
van Krevelen DW (1993) Coal, 3rd edn. Elsevier
Wada H, Suzuki K (1983) Carbon isotopic thermometry calibrated by dolomitecalcite solvus temperatures. Geochim Cosmochim Acta 47:697â€“706. https://doi.org/10.1016/00167037(83)901047
Wang Y, Alsmeyer DC, McCreery RL (1990) Raman spectroscopy of carbon materials: structural basis of observed spectra. Chem Mater 2:557â€“563. https://doi.org/10.1021/cm00011a018
Yamaoka K, Wallis SR, Miyake A, Kouketsu Y (2022) Recognition of detrital carbonaceous material in the Ryoke metamorphic belt by using Raman thermometry: implications for thermal structure and detrital origin. Lithosphere 2022:3899340. https://doi.org/10.2113/2022/3899340
Acknowledgements
We thank Seira Katagiri, Kazuki Matsuyama, and Shunsuke Ogino of Nagoya University for their kind help with assessing the utility of the code. SK also thanks Yukitoshi Fukahata of Kyoto University for helpful comments on the Akaike Information Criterion and Hiroyuki Noda of Kyoto University for fruitful discussion on the preliminarily results. The web page created by Toshihide Ihara and Shin Shigemitsu and maintained by Organic Primary Standards Group, AIST, greatly helped our regression analysis (https://unit.aist.go.jp/mcml/rgorgp/uncertainty_lecture/index.html). We thank the associate editor Madhusoodhan SatishKumar for careful handling of the manuscript. Comments from two anonymous reviewers helped us improve the paper.
Funding
The present study was supported by JSPS KAKENHI Grant nos. JP20J01284 and JP24K17151 to SK, JSPS KAKENHI Grant no. JP21H05202 to SW, and JSPS KAKENHI Grant nos. JP20J12701 and JP22H05314 to YS.
Author information
Authors and Affiliations
Contributions
SK, YK, and MA proposed the initial concept for the project based on work initiated by SW. YK, MA, YN, YS, and KY provided the raw Raman spectral data published in the previous papers. YK performed additional measurements using a 633Â nm laser. SK developed, validated, and applied the code. SK wrote an early version of the first draft, which was subsequently refined by SW. All authors discussed the results, contributed to the writing of the manuscript, and approved its final version.
Corresponding author
Ethics declarations
Competing interests
The authors declare that they have no competing interests that could have appeared to influence the work reported in this paper.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Kaneki, S., Kouketsu, Y., Aoya, M. et al. An automatic peak deconvolution code for Raman spectra of carbonaceous material and a revised geothermometer for intermediate to moderately highgrade metamorphism. Prog Earth Planet Sci 11, 35 (2024). https://doi.org/10.1186/s40645024006378
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s40645024006378