Potential of visible and near-infrared spectroscopy to derive colour groups utilising the Munsell soil colour charts

Abstract

A study was undertaken to investigate the potential of visible (VIS) and near-infrared (NIR) spectroscopy to derive soil colour groups as a new methodology that might assist in soil type classification. A fibre-optic VIS–NIR spectrophotometer (306.5 – 1710.9 nm) was used to measure the light properties in reflectance mode of 342 dry soil samples collected from fields in Belgium and Northern France. The colour of soil samples was measured with the Munsell soil colour charts, dividing soil samples into two groups of 236 and 106 samples belonging to Hue 2.5Y and Hue 10YR, respectively. Factorial Discriminant Analysis (FDA) was performed separately on each Hue data using the first five Principal Components (PCs) of a Principal Component Analysis (PCA) carried out on colour groups defined a priori. Considering the 2.5Y Hue group, soil samples were classified successfully into four colour groups with a correct classification (CC) of 87.3% and 81.8% for the calibration and validation spectra, respectively. Lower accuracy was observed for the 10YR Hue group for classification into three groups, since 79.6% and 75.0% of samples were correctly classified for calibration and validation sample sets, respectively. Establishing four colour groups by combining both Hue groups decreased the classification accuracy considerably. In order to improve the classification accuracy, the first 5 PCs and first 2 PCs of PCA performed on soil spectra and on the total carbon (Tot-C), extractable phosphorous (Ext-P) and texture index (Ti), respectively, were pooled (concatenated) into a signal matrix and analysed newly by the FDA. This led to improvement of the classification results of four Hue 2.5Y groups (CC of 82.1% and 89.1%) and three Hue 10YR groups (CC of 84.1% and 82.4%). However, best concatenation results of four groups for the combination of both Hue groups were obtained when only Tot-C and Ext-P were considered without Ti, obtaining a CC of 84.2% of the validation data sets. This suggests that the VIS-NIR spectroscopy combined with chemometric tools has the potential of identifying different soil colour groups for a large geographical area.