Optimising Sample Preparation and near Infrared Spectra Measurements of Soil Samples to Calibrate Organic Carbon and Total Nitrogen Content

Abstract

There is increasing attention in soil science to using near infrared spectroscopy as a fast and cheap method to analyse various soil properties. However, soil chemical components such as carbon and nitrogen are of low concentrations and analysis is exacerbated by a heterogeneous sample matrix. Thus, for soil analysis, the optimisation of sample preparation and measurement protocols is a prerequisite to establish NIR spectroscopy as a soil analytical method. Several methodological aspects, such as sample-grinding and drying, have been identified as important factors, but results on the best methods still remain inconsistent. The objective of this study was to comprehensively investigate the impact of these factors on a common sample set in order to be able to give recommendations for a standard measurement protocol. The examinations were performed on a set of agricultural soils covering a broad range of soil types, texture classes and carbon and nitrogen content. The number of spectra replicates and size of the sample cups (scanning area 6.8 cm² and 19.6 cm²) had negligible impact on NIR calibration accuracy. Drying (oven-dried vs air-dried) and grinding (grinding vs crushing and sieving) significantly decreased the calibration error for organic carbon when ground and oven-dried samples were used, whereas these influences were smaller for N (grinding) or had hardly any affect on calibrations (drying). Overall, lowest root mean square error of cross-validation (RMSEC)values were reached for ground, oven-dried samples measured in a large cup (scanning area = 19.6 cm²), with 0.391% for C and 0.028% for N. A factorial analysis on the RMSECV for drying, ring cup and grinding, showed the biggest influence on RMSECV by grinding for C with 35% and for N with 28% of the sum of squares accounted for. The effect of varying the laboratory temperature (20°C, 24°C and 28°C) was small when oven-dried samples were used but might have caused a considerable bias in predictions of air-dried samples. In addition, only oven-dried soil samples showed satisfactory reproducibility of predicted values within nine months when NIR measurements were repeated. Since calibrations of different soil parameters seemed to be affected differently by sample preparation and measurement parameters, this study can be considered as a first step towards a standardised measurement protocol.