Performance of Automated Near‐Infrared Reflectance Spectrometry for Continuous in Situ Mapping of Soil Fertility at Field Scale

Diffuse reflectance of soils in the near‐infrared reflectance (NIR) has been related to many chemical soil properties. Diffuse reflectance spectroscopy may become a part of proximal soil sensing and contribute to bridge the gap of knowledge imposed by the inability of conventional methods to resolve the spatial patterns of soil fertility at field scale. We used a mobile automated NIR spectrophotometer (1100–2300 nm) to map the topsoil of three fields located in an organic farm “on‐the‐go” at a speed of 3 to 6 km h −1 to a depth of 15 cm. Spectral data were related to results from conventional laboratory analysis of soil P, K, Mg, soil organic matter (SOM), N, and pH. Maps and semivariograms of the principal component scores computed from the spectral information showed consistent spatial patterns present at two different scales. The strength of correlation between field spectra and soil chemical parameters was location dependent, which will make it difficult to develop cross‐field calibration models. Stable, semiquantitative predictions were obtained for soil pH, N‐tot, SOM, K‐tot, and Mg‐tot ( R 2 : 0.71, 0.69, 0.61, 0.55, 0.53) when training data from all fields where included. To improve relationships between soil spectra and soil attributes, dual‐wavelength indices (DWIs) were used. Even though DWIs showed better partial least squares regression model results compared with smoothed or differentiated spectra, they did not improve cross‐field model calibration.