A new mechanistic approach to link soil chemical composition and organic carbon content: decomposing mid-IR spectra with multivariate curve resolution

Soil organic carbon (SOC) is a key player in global carbon cycling and has primary effects on soil quality and functioning. There is a general interest in modeling SOC content and understanding the factors controlling its accumulation and stability. The mid-IR spectra that provide fingerprints of soil chemical composition are well recognized in modeling and predicting SOC contents, which is generally done using different types of empirical multivariate analyses. This work suggests, for the first time, the decomposition of soil mid-IR spectra using nonnegative multivariate curve resolution (MCR) with an alternating least squares (ALS) algorithm [1]. The advantage of the nonnegative MCR-ALS decomposition is that it allows the expression of soil mid-IR absorbance in terms of contributions from chemically meaningful components, following the Beer-Lambert law. Hence, this combination of IR spectroscopy and the nonnegative MCR-ALS decomposition proposes a new analytical approach to decipher soil compositions and elucidate the components controlling soil functions. Potentially, the nonnegative MCR-ALS decomposition can identify chemically individual components or groups of constituents maintaining constant proportions in a series of samples. Based on this decomposition, a simple mechanistic model is developed to link the identified MCR-ALS components with their contribution to the whole SOC content [1]. This approach has been used to examine the SOC of soil samples collected in the north and south of Israel, from different depths and under different land uses. Four components including a carbonate-rich constituent and three others representing clay-organic matter associations were capable of quantitatively describing 99.7% of the variance of soil mid-IR spectra. SOC modeling using these four components suggested a SOC content threshold affecting modeling performance such that SOC content below 1.0 % w w^-1 could be modeled with RMSD of 0.18% w w^-1. The emergence of this threshold is currently related to mechanisms of how different SOC fractions become "mirrored" in mid-IR spectra. This threshold could be useful to distinguish between different types of SOC, i.e., those tending to tightly interact with mineral surfaces and those having weak connections with minerals, if at all. The perspectives in extending the whole approach for a wide range of SOC contents are also discussed.

[1] Borisover, M., Lado, M., & Levy, G. J. (2025). Modeling Soil Organic Carbon Content Using Mid-Infrared Absorbance Spectra and a Nonnegative MCR-ALS Analysis. Soil & Environmental Health, 3(1) 100123.