Near-infrared (NIR) and mid-infrared (MIR) spectroscopic techniques for assessing the amount of carbon stock in soils – Critical review and research perspectives

Abstract

This paper is an extensive review of the research that has been undertaken on near-infrared (NIR) and mid-infrared (MIR) spectroscopy applied to soil particularly for determining carbon (C) content. The objective is to determine which acquisition method (NIR, MIR, in the field or in the laboratory) might be recommended for the purpose of C stock measurement with a particular interest in carbon credit trading. For this purpose, an optimal method has to satisfy the dual constraints of low-cost and high throughput analysis. The various methods proposed in the literature are compared. In order to make comparisons as reliable as possible, special attention has been paid to the conditions of data acquisition (sample preparation), and to calibration and validation procedures. In particular, whether the validation has been carried out on fully independent samples or on samples similar to the ones of the calibration set greatly influences the results. Also, for C stock measurement, it is absolutely necessary to measure the bias of the prediction in order to be conclusive about the feasibility of the method. However, only few researchers provide this parameter and we recommend including it as a matter of course in future reports. Finally, although MIR on dried and ground samples is the most accurate method, on-the-go and in-field sensors provide predictions accurate enough to show promise in being a valuable component of technologies that would be used for C-credit purposes. But in order to meet the cost/accuracy trade-off, the main issue using such field sensors is to be able to simultaneously measure the bulk density or, better, to directly measure the volumetric concentration of C in soil. This circumvents the costs of field extraction and laboratory analysis. This is the next great challenge to be met by soil scientists.

Introduction

In order to mitigate climate change due to emissions of greenhouse gases (GHGs), the Kyoto Protocol proposed targets for GHG reductions and set up trading credits that corresponded to verified removal of greenhouse gas from the atmosphere (United Nations Framework Convention on Climate Change²). One of the main options for C mitigation identified by the IPCC is the sequestration of organic C in soils (Metz et al., 2007). Not only can soil C sequestration help to offset GHG emissions, but storing C in soil also provides environmental benefits (Lal et al., 1997). There are two ways to design contracts to purchase C-credits, either practice-based contracts or credit-based contracts, and studies have shown that credit-based contracts are more efficient than practice-based contracts (Mooney et al., 2007). But the same authors stated that (i) the biggest difference between transaction costs under credit-based and practice-based contracts was the cost associated with measuring the amount of C-credit sequestered, and (ii) another factor influencing transaction costs is uncertainty related to the quantity of credits produced, which could affect the payments received by the producers.

This converges with Smith’s view (2002) who also warns: “without quantified knowledge of the precision of the measurement, one runs the risk that sampling designs installed now will fail to detect much of the changes between now and re-measurement in 5 or 10 years. At the same time, measurement cost must be less than the market price of the amount of sequestration detected by the measurement.” This author and others (McCarty and Reeves, 2006, Morgan et al., 2009, Reeves, 2009) converge with unanimity on the fact that the issue of C sequestration assessment is two-fold, i.e., it has to be accurate and low-cost. Reeves (2009) outlined the trade-off of accuracy versus speed and quantity of samples to be analyzed, “the question exists as to whether better overall estimate of C can be made by using large amounts of fair quality data versus much smaller amounts of highly accurate data”. He outlined that this question was related to the inherent heterogeneity of soils. Finally, he suggested that “for large-scale studies, large amounts of more easily acquired surface data be integrated with little amounts of harder to get core/subsurface data to provide an accurate estimate of the total C content within a given area-volume of soil”.

Among the low-cost and easy-to-use alternative techniques, spectroscopic techniques such as Near-Infrared (NIR) and Mid-Infrared (MIR) spectroscopy are very attractive for C content measurement in soil. These techniques have been extensively developed in agriculture in the last 40 years, with a boom starting in the late 80’s, measuring composition of cereals, fruit and vegetables, meat, etc. In some cases, such as cereals, they have replaced classical laboratory analysis in inspection programs carried out by GIPSA (Grain Inspection, Packers and Stockyards Administration)³ or CGC (Canadian Grain Commission) in US and Canada (Williams et al., 1998), respectively. The main difference between both ranges is that absorption in mid-infrared spectroscopy corresponds to fundamental bands of molecular vibrations, whereas absorptions in NIR correspond to overtones and combinations of these fundamental bands (Williams and Norris, 1987). The consequence is that (i) absorption coefficients (also called extinction coefficients) are much smaller in the NIR range, which allows light to better penetrate into the matter, but on the other hand (ii) the NIR spectrum is much more encumbered because of the abundance or combination and overtone bands. Therefore, the specificity of bands is less in the NIR range compared with MIR. Another point is that diffusion of light is much greater in the NIR than in the MIR range. Therefore, NIR spectra will be much more affected by factors which affect the diffusion of light such as: the physical structure (size of aggregates, porosity), but also the presence of water which changes the refractive index and therefore the diffusion of light (Williams and Norris, 1987).

The practical consequence is that NIR requires less sample preparation than MIR and is best fitted for in-field analysis, with lesser specificity requirements, whereas MIR generally shows a better specificity and reproducibility, but requires more sample preparation in order to optimize the sample/light interaction. The two classical modes of MIR analysis of soils are diffuse reflectance on dried/ground soils, as shown in several articles cited by Viscarra Rossel et al. (2006) or Attenuated Total Reflectance (ATR) applied to soil pastes especially when ions, like nitrate, ammonium etc., are assayed (Jahn et al., 2006, Du and Zhou, 2009). Therefore, up to now, MIR has been restricted to laboratory analysis, although its use in the field has been reported (Reeves, 2009).

Although research on NIR/MIR spectroscopy for soil analysis started later than for agricultural products, this research area has been experiencing a boom over the last 10 years, with a current apparent exponential growth (see Fig. 1). More than 210 publications have been produced in the last 15 years, with almost 50% of them in the last 3 years. Numerous investigations have been carried out measuring the properties of soil, in different configurations (i.e., in laboratory, in the field by sampling, or on-the-go with sensors embedded on a tractor). But up to now, bibliographic studies and reviews on this subject have been very scarce, may be because the rapid growth of NIR/MIR studies dedicated to soil is only recent. Viscarra Rossel et al. (2006) compiled a very complete table of the various soil attributes measured by NIR/MIR. Recently, as the result of a first NIR conference dedicated to soil in France, Cécillon et al. (2009) made an extensive review of all the available NIR technologies (proximal sensing-, laboratory analysis, remote sensing) for assessing soil quality and concluded that the most urgent need was for international databases. Kusumo et al. (2008) focused on C issues. Reeves (2009) recently wrote a review emphasizing laboratory versus on-site NIR and MIR analysis of C and pointing out the various advances and bottlenecks of this technique. He ended up with a list of nine questions which are deemed important for the future large-scale implementation of calibration for C content assessment either for C sequestration needs or for improving farming techniques in general. These key points are: 1. The calibration extent (local, regional etc.); 2. The spectrum acquisition procedure (in laboratory, in the field etc.); 3. The calibration transfer issue; 4. The reference measurement accuracy; 5. The trade-off between accuracy versus speed and number of sample analyzed; 6. The need for additional measurements (for instance bulk density) to provide relevant information (for instance Mg C ha⁻¹); 7. Which C form is to be measured with regard to its stability; 8. Can NIR predict sequesterable C from source material? and finally 9. Which is the best chemometric method for calibration development?

His analysis investigates the state of the art to determine “where we are and what needs to be done”. Here we shall be much more focused on a critical analysis of the procedures carried out and results obtained so far for C measurement in soil, in order to address some of these questions.

Soils raise very interesting scientific issues for NIR/MIR spectroscopy. Soil is a very diffusive and absorptive medium. So, soil spectra must be acquired in diffuse reflectance, not in transmission (except of samples diluted in non-absorbing materials like KBr). The incident light is diffusely reflected back when reflecting, diffusing and diffracting on the macro- and micro-optical interfaces of soil particles to create a reflectance spectrum R. The transmitted and reflectance flux can be modelled by the Kubelka–Munk’s (K–M) law which describes light transfer in an absorbing and diffusing medium. But more classically, to get a spectrum in which the intensities are linearly linked to the concentrations of interest, the reflectance spectrum is turned into an absorbance spectrum, by applying a Log(Ro/R) function, where Ro is the reflectance of a diffusing and non-absorbing medium. Reflectance – and consequently absorbance – spectra contain both chemical and physical information linked to scattering and specular reflectance. In soil, this last effect, i.e., the physical effect, mainly related to soil structure, is particularly important because soil structure and particle size can vary greatly and therefore can affect the spectrum (Boonmung and Riley, 2003, Richter et al., 2009, Wu et al., 2009b) and should be removed (Bogrekci and Lee, 2005). In MIR spectroscopy, which has higher absorption coefficients, strong specular reflection effects can be observed, which is another cause of noise. Drying and finely grinding samples or working in ATR mode are two ways to avoid this effect, but this requires sample preparation and makes in-field use of MIR difficult.

In addition to these effects due to soil structure, other physical effects can be encountered when one has to carry out NIR measurements in-field, such as soil temperature variations (Wu et al., 2009a) and soil moisture variations (Boonmung and Riley, 2003, Wu et al., 2009b). These additional external factors alter the spectrum and cause discrepancy in predictions.

Our aim was to investigate the research work on NIR/MIR analysis of soil C content, whatever its form (total C, inorganic C, organic C). This review is based on a very detailed analysis of the methodologies carried out in order to enable us to correctly compare their performance strengths. In particular, factors such as the size and exhaustiveness of the calibration database, the closeness of calibration/validation samples, sample preparation (dry/moist, ground/non ground etc.), the type of instrument, the mathematical process for calibration (Partial Least Squares Regression or PLS, Principal Component Regression or PCR, Multiple Linear Regression or MLR etc.), the spectra pre-treatment etc. have been thoroughly studied to reveal common approaches to the NIR/MIR analysis of soil C. This will help one who reads a publication on NIR/MIR for soil analysis to develop a critical assessment of the performance values presented in this publication.

This analysis will lead us to analyze how far we are from using NIR/MIR techniques as routine measurement techniques, based on a metrological approach and to propose research paths for reaching this goal of having NIR/MIR spectroscopy as a recognized method for assessing C content in soil.