Application of mid-infrared spectroscopy for soil analysis in calcareous Argania spinosa forests in Morocco

Abstract

Mid-infrared (MIR) spectroscopy is a promising tool to meet the growing global demand for soil data, as it allows a rapid and inexpensive collection of infrared spectra. Despite an increasing number of soil property predictions based on such datasets, the data availability in arid regions is still sparse. This is particularly problematic as the presence of inorganic carbonates poses additional challenges to model development in arid regions while limiting the applicability of models developed in more humid climates. In this study, the potential application of Fourier transform infrared (FTIR) spectroscopy for soil analysis in degraded Argania spinosa populations is assessed. The underlying objective was twofold. First of all, the models generated may help to monitor soil conditions in an endangered UNESCO biosphere reserve. Secondly, knowledge gaps in arid, calcareous regions of Northern Africa are addressed by creating a sample collection of considerable size. Spectra of 397 soil samples were recorded in addition to conventional laboratory measurements of pH, percolation stability (PS), nitrogen (N), total organic carbon (TOC), and inorganic carbon (iC). Partial least squares (PLS) regression was then used to predict these soil properties based on the MIR records. The models were calibrated, independently validated, and evaluated based on the coefficient of determination (R²), the root mean square error (RMSE), the ratio of performance to deviation (RPD), and the ratio of performance to interquantile range (RPIQ). Promising model validation results were obtained for N (R²: 0.86, RPD: 2.71, RPIQ: 2.45, RMSE: 0.03), TOC (R²: 0.89, RPD: 3.06, RPIQ: 1.72, RMSE: 0.44), and iC (R²: 0.96, RPD: 5.32, RPIQ: 2.01, RMSE: 0.15), but not for the pH (R²: 0.37, RPD: 1.27, RPIQ: 1.51, RMSE: 0.17) or PS (R²: 0.30, RPD: 1.20, RPIQ: 1.31, RMSE: 80.54). Especially the results of the N prediction show that reliable MIR models can be trained in arid environments despite noticeable effects of ubiquitous iC. In contrast, the pH and PS models highlight clear limitations of the technique for surrogate calibrations.