Best practices for obtaining and processing field visible and near infrared (VNIR) spectra of topsoils

Abstract Diffuse infrared reflectance spectroscopy is considered a promising approach for addressing soil quality, and its use directly in the field might be an achievable challenge. The present work aimed at optimizing the acquisition procedure of visible and near infrared reflectance (VNIR) spectra of topsoils (0–20 cm) in the field, in order to predict usual soil properties. The studied set included 201 samples originating from six fields in different regions of large-scale crop cultivation in France. Spectra were acquired using a portable spectrophotometer. Spectrum acquisition procedures included scanning on the soil surface, on raw or smoothed (cut) cores collected using an auger, and on clods resulting from core crumbling. In addition, spectra were also acquired on air-dried clods, either 2-mm sieved or not (laboratory conditions). Furthermore, 42 mathematic pretreatments were compared (including derivatives, standard normal variate SNV, multiplicative scatter correction MSC, etc.). Identifying the most appropriate scanning and pretreatment procedures was done through four-group cross-validation. Using the most appropriate pretreatment, calcium carbonate content was very well predicted whatever the scanning procedure used (RPD = 6.9–9.1; RPD is the ratio of standard deviation to standard error of cross-validation; for soil properties RPD > 2 denotes accurate predictions); good predictions were achieved for total nitrogen (RPD = 2.5–3.0), organic matter (RPD = 2.1–2.8) and exchangeable potassium (RPD = 2.9–3.2); but available phosphorus was poorly predicted (RPD = 1.6–1.8). Except for available phosphorus, accurate predictions of these properties could therefore be achieved whatever the scanning procedure used, thus in the field. Best predictions were often obtained using spectra acquired on 2-mm sieved air-dried samples (i.e. in laboratory conditions), otherwise using spectra acquired on raw cores. Acquiring spectra on cores, on raw cores especially, was the most appropriate field procedure; it led to predictions comparably accurate to those achieved in the laboratory with 2-mm sieved air-dried samples. Similar prediction accuracy for field and laboratory VNIRS is counterintuitive due to variable field conditions (moisture, temperature, stoniness, etc.). It might result from higher number of replicates in the field than in the laboratory (often inherent to field vs. lab conditions) and/or higher sample density and cohesion, which would improve reflectance signal. For spectra acquired on cores, best calibrations were achieved with MSC and first derivatives for calcium carbonate, total nitrogen and organic matter, but without pretreatment for exchangeable potassium and available phosphorus. Second derivatives always yielded poor results.