Comparison of Soil pH and Exchangeable Cation Quantification by Various Wet Methods with Near- and Mid-Infrared Spectroscopy Prediction

Abstract

ABSTRACT Rapid soil analyses can be obtained by infrared (IR) spectroscopy, but few studies have evaluated the suitability of different reference wet chemical methods to calibrate models predicting soil fertility. The aim of this study was to assess the accuracy of IR spectral-based models for the determination of soil pH measured in water, potassium chloride (KCl), and calcium chloride (CaCl2), and exchangeable basic cation content measured by the ammonium acetate (NH4OAc, 1 M and 0.02 M variations, pH = 7), Mehlich-III, Ambic-I, citric acid (1%), compulsive exchange, and Bray-II methods. Near- to mid-infrared (NIR, MIR) spectral-based models predicting soil pH and exchangeable cations measures were calibrated by partial least squares regression for a set of soils containing highly and intermediately weathered clay minerals and variable sesquioxide contents. The ratio of root mean square error of prediction (RMESP) to analyte component range was compared to a threshold value for contextual model accuracy assessment. pH measurements in CaCl2 matrices led to better calibrations and predictions of soil pH by IR spectral models compared to pH measurements in water and KCl matrices. Exchangeable calcium (Ca) and magnesium (Mg) concentrations determined by compulsive exchange and the NH4OAc (1 M, pH = 7) extraction methods, respectively, were most accurately predicted by NIR spectral-based models (RMSEP = 1.2–1.4 and 0.37 cmolc kg−1, respectively). The ratio of performance to inter-quartile distance (RPIQ) is recommended as an alternative figure of merit to traditionally reported ratio of performance to deviation (RPD) for IR spectral-based models predicting soil chemical properties. Subsetting soil samples by similarities in physiochemical properties improved the accuracy of soil pH and exchangeable Ca content predictions.