Sample preparation and prediction of soil organic matter properties by near infra-red reflectance spectroscopy

Near infra-red reflectance spectroscopy (NIRS) offers the potential for rapid and cost-effective soil analysis. Unfortunately, soil NIRS calibrations have not performed well across soil types, and this is believed to be due to the differences in soil particle size and, or, soil mineralogy. In this study we evaluated the influence of grinding, removal of organic matter, removal of mineral component, and scanning soils through plastic bags, on the prediction of three soil organic. matter properties across 25 soils. The three properties were mineralizable nitrogen, total soil carbon, and total soil nitrogen. Seven soil preparations were made from coarse whole soil (CWS) samples, and their spectra acquired. The preparations were as follows: CWS, finely ground whole soil (FWS), CWS scanned through plastic (PWS), and two methods for the approximation of mineral and soil organic matter (SOM) within the CWS. Mineral soil spectra were derived from combustion or chemical treatment. SOM spectra were approximated by the subtraction of mineral spectra from CWS spectra. The preparations influenced soil NIRS calibrations in different ways for each property, and their performance was compared with the predictive ability derived from calibrations with CWS. FWS, PWS, and mineral preparations did little to enhance the predictions, and were more likely to reduce predictive ability. SOM spectra gave the best predictions of total soil nitrogen and mineralizable nitrogen (20-30% lower in error), while total soil carbon was best predicted from the CWS spectra. The best calibrations for all three properties suggest they can be predicted within at least 8 classes. It was concluded that approximated SOM spectra could improve predictions of some soil properties by reducing soil mineralogical noise in NIRS calibration development.