Towards on-the-go field assessment of soil organic carbon using Vis–NIR diffuse reflectance spectroscopy: Developing and testing a novel tractor-driven measuring chamber

• A mobile measuring chamber was constructed for Vis–NIR spectra acquisition using a commercial Vis–NIR-spectrometer.
• The chamber allows spectra acquisition in the field at standardized illumination and geometry.
• Testing on arable fields with heterogeneous SOC contents yielded a standard error for on-the-go SOC prediction of 1.2 g kg−1.
• SOC maps showing small-scale variation can be created at reduced efforts.

Proximal sensing of soil organic carbon (SOC) in the field using diffuse reflectance spectroscopy is still difficult under variable weather conditions. Here, we introduce a tractor-driven measuring chamber for on-the-go visible and near-infrared diffuse reflectance spectroscopy (Vis–NIRS) meeting experimental and precision agriculture demands. A commercial full-range spectrometer operates in a closed dark chamber with artificial light. Sensor view angle, distance to soil, and illumination conditions were optimized. The mobile chamber was placed on drum rollers to flatten the ploughed and tilled soil surface and to minimize disturbances in Vis–NIR spectra by surface roughness. Prior to on-the-go spectra acquisition under field conditions, SOC prediction models for the soils under study were independently calibrated under variable moisture and roughness conditions. Driving at a tractor velocity of 3 km h−1 resulted in measuring spots of approximately 8 cm length and 3 cm width at 0.6 m distance to one another in the direction of movement, delivering geo-referenced SOC concentrations at a sub-m spatial resolution. Gravel on the soil surface resulted in erratic extremes of predicted SOC concentrations, but these could be eliminated as outliers. The system was tested under field conditions on two long-term experiments at two different sites which revealed each a large span of SOC concentrations. On-the-go predicted SOC concentrations and those obtained from conventional plot-wise lab analyses were correlated with coefficients of determination of R2 = 0.65 and a standard error of 1.22 g SOC kg−1. Further improvements, particularly in data processing, will enable a reliable proximal sensing on-the-go for precision agriculture purposes in the near future.