Global and Local Calibrations to Predict Chemical and Physical Properties of a National Spatial Dataset of Scottish Soils from Their near Infrared Spectra

Calibrations were developed to predict chemical and physical properties from near infrared spectra of an extensive spatial dataset of Scottish soils. For this purpose we used a spectral library of 1246 soil samples collected throughout Scotland in two campaigns: 546 samples collected in a 10 km grid between 1978 and 1988 (NSIS1); and 700 samples collected between 2007 and 2009 during a re-sample of the same sites but in a 20 km grid (NSIS2). The samples were split into validation (N = 250) and calibration (N = 996) sets, and global partial least squares regression (PLSR) was performed in combination with spectral pre-processing treatments, namely first or second derivative and, optionally, standard normal variate and de-trending or multiplicative scatter correction treatments. For the local model, the calibration set (N = 996) was split into test (N = 121) and library (N = 875). Local calibration was performed using PLSR in batches that iteratively selected from the library between 75 and 425 reference sp...