Improving Infrared Spectroscopy Characterization of Soil Organic Matter with Spectral Subtractions

The method of SOM removal has practical as well as theoretical implications for the interpretation of subtraction spectra. For example, mineral alterations from high temperature ashing can manifest as losses or appearances of peaks and/or as shifted or broadened peaks in the mineral reference spectrum. These spectral artifacts are prone to occur in regions of overlap with organic bands at 1,600-900 cm^-1,²² compromising interpretation of organic bands. Common changes of mineral bands following high temperature ashing (≥ 550 °C) are evident in Figure 2 and include loss of OH peaks at 3,700-3,600 cm^-1, and peak losses and shifts in lattice Si-O and Al-O peaks at 1,050-800 cm^-1. On the other hand, SOM removal using chemical oxidation tends to preserve mineral functional groups and thus risks fewer artifacts (see Discussion). However, this typically comes at a cost of lower SOM removal²⁸, as evidenced for the example soils (Table 2). This restricts interpretation of organic functional group composition in the subtraction spectrum to less than the total SOM pool.

Chemical oxidations can be an attractive alternative to ashing for furnishing mineral backgrounds because they generally preserve mineral structure and thus avoid artifactual absorbance features in reference spectra and thus the resulting subtraction spectrum^14,22. For example, Soil A lost 89% of soil organic carbon (SOC) by NaOCl oxidation compared to 97% by ashing (Table 2) while preserving mineral absorbance features altered by ashing (Figure 2). On the other hand, incomplete and potentially selective removal of SOM means that NaOCl-resistant organic matter will not be characterized^5,29,30, requiring careful interpretation of the subtraction spectrum¹⁴. Quantifying SOM loss with each additional oxidation treatment performed can be used to identify the optimal number of oxidative treatments for SOM removal. Evaluation of the example oxidation method using NaOCl showed that SOM removal plateaued after 3 oxidation steps regardless of whether SOM content of samples was low (<2% C) or high (36% C). With each successive oxidation step, the color of the supernatant is likely to change from various shades of purple/black/orange to lighter brownish-red until the original greenish-yellow color of bleach (i.e., unreacted dilute NaOCl).

Grinding to reduce and homogenize soil particle size can improve spectral quality by reducing light scattering from large particles and inversion of strongly absorbing components³¹. Though a maximum particle size of 250-100 µm has been considered optimal to minimize such artifacts³², recent evaluations suggest that grinding beyond 2000 µm does not necessarily improve spectral quality for chemometric predictions^33,34. However, more finely ground soil samples can still yield enhanced (sharper) absorbance peaks and exhibit decreased absorbance variability³³, which would be expected to benefit spectral subtractions. On the other hand, some studies have found that grinding (e.g., to 200 µm) can induce more heterogeneity in particle size distribution than standard sieving (e.g., 2,000 µm)^35,36. The effect of this on spectral quality is likely to depend on experiment objectives (e.g., interpretation of absorbance bands, chemometric prediction of C fractions) and soil properties (e.g., texture, mineralogy)²⁹. It is recommended that effects of grinding size on subtraction spectra be assessed for a given set of soils and objectives. The treated soil sample used for the mineral reference spectrum is likely to have finer particle size than the corresponding untreated soil due to disruption of aggregates by SOM removal.

The use of neat (undiluted) soil samples is widespread in the use of DRIFT spectroscopy due to greater ease of analysis and higher throughput rate. Sample dilution is traditionally performed for high purity samples using infrared-invisible halide salts (e.g., KBr, NaBr, KI). For DRIFT spectroscopic characterization of the organic component of soils, however, halide salt dilution is less advantageous. This is because dilution of samples tends to improve resolution of absorbances for dominant (mineral) components but not necessarily those representing minor (organic) components^5,37. Nonetheless, there can be benefits to the dilution of soil samples, which can be empirically evaluated for each sample⁵.

To illustrate the role of the subtraction factor (SF) in performing spectral subtractions, consider the subtraction spectra obtained for soil A using the spectrum of the NaOCl oxidized soil as the reference (Figure 3). As the SF increases from 0 to 1, the absorbance of bands corresponding to minerals decrease, most notably mineral surface O-H at 3,618 cm^-1 and Si-O at 1,880 and 808 cm^-1. Concurrently, absorbance increases for bands representing organic functional groups such as aliphatic C-H stretch at 2,920 cm^-1 and potentially amide C-N and N-H and/or aromatic C=C at 1,558 cm^-1. With increasing scaling of the SF, however, inversions of mineral absorbances such as Si-O at 2100 – 1780 cm^-1 emerge. As NaOCl oxidation removed 89% of total SOC in a soil with 2.2% C (Tables 1, 2), and given that SOM is estimated to be 58% C³⁸, a theoretical SF of 0.97 can be calculated. However, non-linearity of absorbance results in over subtraction of mineral features using this SF (Figure 3).

A more practical approach is to operationally determine the SF based on removal of a mineral absorbance feature by 'canceling' or zero-ing out a target peak(s). The quartz-like Si-O features at 2100 – 1780 cm^-1 are suitable target mineral absorbances because they arise from fundamental Si-O vibrations that are less susceptible to specular reflection^14,37. For example, in Figure 3, zero-ing of quartz Si-O at 2,100- 1,780 cm^-1 suggests a SF in the range of 0.7 – 0.8 to be more suitable for reducing the dominant mineral absorbances than the a priori SF of 0.97 calculated from the difference in mass (i.e., SOM removed) between the two soil samples.

Potential pitfalls in selecting a SF include inversion and oversubtraction, both evidenced by 'negative peaks' that resemble an upside-down peak or "W" shape¹⁴, or in the case of strong subtraction, a mirror image of the untreated soil spectrum. Inversions may not be possible to avoid altogether since these are common for Si-O at 1,150 – 800 cm^-1, but can be minimized. Inversions should be noted as a region non-grata for spectral interpretation^22,39 because in these regions absorbance can reflect how the subtraction was performed more so than absorbing components, especially for minor constituents such as OM.

Spectral software accompanying modern FTIR spectrometers can be useful to guide and execute subtractions. The utility of such software lies in real-time and iterative adjustment of SFs while viewing the resulting subtraction spectrum. However, the default SF suggested by spectral software should be regarded with caution because the software criteria may not be appropriate for the experiment goals and samples such as soils (e.g., non-linear absorbance, inversion of mineral features). For example, the particular software used to calculate subtraction spectra in Figure 3 (see the Table of Materials) suggested a SF of 1.45, based on linearization of baseline in the resulting subtraction spectrum⁴⁰, which misses the objective of zero-ing out Si-O at 2100-1780 cm^-1 for interpretation of organic absorbances at 1,800-1,000 cm^-1. It is important to note that the SF suggested by software is also influenced by the wavenumber range selected for the subtraction because the software optimizes a subtraction for the defined spectral region. Additional benefits of software include spectral processing transformations (e.g., Kubelka-Munk, power functions) to improve spectral quality (e.g., linear baselines) prior to performing subtractions.

The mineral reference spectrum may contain absorbance features influenced by the method of SOM removal, with consequences for performing and interpreting spectral subtractions. Comparison of differences among untreated and ashed mineral reference spectra (Figure 2) can help identify which of these features may be due to artifacts of SOM removal. In contrast to NaOCl oxidation, ashing produced changes in absorbance bands representing mineral functional groups, notably phyllosilicate surface and interlayer O-H (3,700-3,600 cm^-1), Al-OH (915 cm^-1), and Si-O (796, 521 cm^-1). Quartz-like Si-O bands (2,100 -1,780, 1,159 cm^-1) did not exhibit marked changes in absorbance (intensity or wavenumber) for either treatment.

An example of over-subtraction and the risk of unidentified artifacts resulting from the SOM removal method of choice is demonstrated for ashing at 550 °C in Figure 4. Though zero-ing out quartz-like Si-O at 2,100-1,780 cm^-1 is achieved with an SF of 0.76, a prominent W-shaped inversion at 1,100-450 cm^-1 suggests that interpretation of the subtraction spectrum should be limited to > 1,200 cm^-1. As the reference spectrum was obtained by high-temperature ashing, loss of mineral surface OH at 3,700-3,600 cm^-1 (see Figure 2) means that these sharp peaks are an artifact that should not be a target to zero out. Attempts to do so will likely compromise other spectral regions. For example, not recognizing the feature at 3700-3600 cm^-1 as an artifact could motivate the user to increase in the SF to reduce (SF = 1.24) or attempt to zero out (SF = 1.51) this artifact at the expense of over-subtracting the remaining spectral regions, including organic absorbances at 1800 – 1200 cm^-1 (Figure 4).

It is also possible to over-subtract the mineral reference spectrum, such as in an attempt to enhance a specific band without considering effects on other spectral regions (Figure 5). While this may be justified by experimental objectives (see Discussion), the example in Figure 5 of enhancing the aliphatic C-H stretch at 3000 – 2800 cm^-1 'at all costs' renders the remaining portion of the spectra uninterpretable, including the region corresponding to the majority of organic functional groups relevant to SOM characterization at 1,800-1,000 cm^-1. An indicator of over-subtraction is the deepening inversion of spectral features beyond inversion-prone Si-O bands as the SF increases from 0.81 to 1.35 to 1.72. Conversely, the incomplete removal of these target mineral bands means that the SF is likely insufficient, as evidence for the subtraction performed with an SF of 0.32. A SF of 0.81 ideally zero-outs out the target Si-O features at 2,100-1,780 cm^-1 while minimizing inversions.

Comparison of subtraction spectra calculated from reference spectra obtained by high-temperature ashing and chemical oxidation indicate common features enhanced by subtraction at 1,750-1,350 cm^-1 ascribable to organic functional groups, but also exhibit differences due to alteration of minerals and potentially the amount of SOM removed (Figure 6). Note that a similar SF was obtained for subtractions using a mineral reference obtained by ashing (SF = 0.76, Figure 4) versus NaOCl oxidation (SF = 0.81, Figure 5). For a given method of SOM removal, differences are visually evident between the subtraction spectra of high and low organic matter soils that are less visible or absent in the untreated soil spectra (Figure 6), corresponding to aliphatic C-H (stretch) at 2916 cm^-1, and ketone and amide C=O with possible contributions of conjugated C=C at 1,647 cm^-1, amide C-N and N-H at 1,568 cm^-1, aliphatic C-H (bend) at 1,427 cm^-1, and phenol and/or carboxyl C-O at 1,275 cm^-1. It is essential that subtraction spectra be understood as an operational method to assess relative changes in absorbances that may be assigned to organic functional groups with varying certainly depending on sample type and absorbance range (due to potentially overlapping bands)⁴.

Figure 1. Illustration of spectral subtraction to enhance organic bands using a background spectrum of the same soil following SOM removal by chemical oxidation (NaOCl). DRIFT spectra were collected without KBr dilution of the soil sample. The soil is a Mollic Xerofluvent under agricultural use in the Sacramento Valley of California (22 mg C g^-1 soil). Subtraction spectra were calculated with a subtraction factor of 1.0. Please click here to view a larger version of this figure.

Figure 2. Comparison of mineral-enriched background or reference spectra for performing spectral subtractions obtained by different SOM removal methods. Comparisons of treated and untreated soils are shown for a surface horizon (A_p) of a Mollic Xerofluvent formed on mixed alluvium in California, USA (22 mg C g^-1). Spectra correspond to the original soil without treatment (top), followed by soils following SOM removal treatments of chemical oxidation (NaOCl) (middle) and high-temperature combustion (550 °C) (bottom). DRIFT spectra were collected on soil samples without KBr dilution. Grey boxes highlight mineral absorbance bands characteristic of soil spectra, centered at 3620 cm^-1, 1870 cm^-1 and 800 cm^-1. Please click here to view a larger version of this figure.

Figure 3. Example of increasing the subtraction factor (SF) at 0.1 intervals from 0.1 (minimal subtraction) to 1.0 (equal subtraction of original soil spectrum and reference spectrum). The SF weights the degree to which the reference spectrum (soil A after heating at 550 °C for 3 h) is subtracted from the original (untreated) soil spectrum (soil A). As the subtraction factor decreases, absorbance bands corresponding to minerals increase (3,618 cm^-1, 1,880 cm^-1, 808 cm^-1), and bands representing organics decrease (2,920 cm^-1, 1,558 cm^-1, 1,240 cm^-1). DRIFT spectra were collected on soil samples without KBr dilution. Please click here to view a larger version of this figure.

Figure 4. Examples of over-subtraction and the risk of unidentified artifacts resulting from the SOM removal method (ashing at 550 °C). Even with zero-ing out quartz-like Si-O at 2,100-1,780 cm^-1 (gray box) (SF = 0.76), a prominent W-shaped inversion at 1100 – 450 cm^-1 (arrows) suggests interpretation of the subtraction spectrum should be limited to >1200 cm^-1. Since the reference spectrum was obtained by high-temperature ashing (550 °C), loss of mineral surface OH at 3,700 – 3600 cm^-1 (gray box) (see Figure 2) indicates that this peak is a likely an artifact. Not recognizing the feature at 3,700-3,600 cm^-1 as an artifact can mislead the user to increase the SF to reduce (SF = 1.24) or zero out (SF = 1.51) this spectral feature at the expense of oversubtracting the remaining spectral regions, including organic absorbances at 1,800-1,200 cm^-1. Additional artifacts from ashing include a feature at 2300 cm^-1 (arrow). DRIFT spectra were collected on soil samples without KBr dilution. Please click here to view a larger version of this figure.

Figure 5. Examples of ideal and non-ideal subtraction spectra using a mineral reference obtained by NaOCl oxidation. Insufficient subtraction (low SF = 0.32) means that prominent mineral absorbances (grey boxes) are still present. Increasing the SF to zero out quartz-like SiO at 2,100-1,780 cm^-1 (SF = 0.81) strikes a balance among the criteria of removing this mineral absorbance as a proxy for mineral overlap at 1,800-1,000 cm^-1, minimizing inversions of Si-O <1,100 cm^-1, and maintaining a linear baseline. In an attempt to further enhance the aliphatic C-H-stretch at 3,000-2,800 cm^-1, the SF may be increased (to 1.35 or 1.72), but this results in oversubtraction of varying degrees. Spectra were collected on neat (no KBr dilution) samples by DRIFT spectroscopy. Please click here to view a larger version of this figure.

Figure 6. Paired comparison of DRIFT spectra of high (top) and low (bottom) organic matter soils showcasing the potential of spectral subtractions to improve interpretation of organic absorbances compared to the original soil spectra. Differences in absorbances corresponding to mineral (grey box) and organic (dashed line) absorbance bands are evident between high and low SOM soils and/or among the three set of spectra: spectra of the original (untreated) soils, and spectra obtained by subtracting a mineral reference from the untreated soil spectrum, using a mineral-enriched soil for the mineral reference spectrum obtained by thermal combustion ('ashing') or chemical oxidation. Organic absorbances that differ between high and low SOM samples may be ascribed to aliphatic C-H (2,916 cm^-1), ketone and/or amide C=O with possible contributions of conjugated C=C (1,647 cm^-1), amide C-N and N-H and/or possible aromatic C=C (1,568 cm^-1), aliphatic C-H (1,427 cm^-1), and carboxyl and/or phenol C-O (1,275 cm^-1). Soils are from the A_p horizon of Entisols formed from mixed alluvium in the Sacramento Valley (California, USA). Spectra were collected on neat (no KBr dilution) samples. Please click here to view a larger version of this figure.

Table 1. Properties of soils used to demonstrate methods for spectral subtraction to improve characterization of SOM using DRIFT spectroscopy. Soils are of similar mineralogy but exhibit differences in organic matter concentration. Soils were sampled from the A_p horizon of Entisols formed on mixed alluvium in the Sacramento Valley of California, USA.

Table 2. Example of varying removal rates of SOM by combustion versus chemical oxidation. Soil samples treated to remove organic matter are used to acquire a mineral-enriched reference spectrum for performing spectral subtractions to improve characterization of SOM functional group composition using DRIFT spectroscopy.